1 Introduction

Renal cancer is one of the most common malignant tumours of the urinary system, and the incidence rate of renal cancer continues to increase in the United States [1]. KIRC accounts for 90% of all renal cell cancers and is the most common pathological type [2]. The aetiology of renal cancer remains unclear, and current research suggests that smoking and obesity are risk factors for renal cancer [3]. Early renal cell carcinoma is often detected during physical examinations, and approximately 60% of patients have no clinical symptoms [4]. Renal cell carcinoma is not sensitive to radiotherapy or chemotherapy, and surgical treatment is an important means to treat renal cancer. However, for advanced renal cancer patients, a combination of targeted drugs is necessary [5]. The European Association of Urology guidelines recommend targeting agents as first-line and later-line treatments for metastatic renal cell carcinoma [6]. Some studies have shown that the targeting agents can significantly improve the overall survival of patients with advanced metastatic renal cell carcinoma [7, 8]. Immunotherapy is a recent research hotspot in the treatment of renal cell carcinoma, and it has become an important method to treat renal cell carcinoma. Compared with sunitinib, lenvatinib combined with pembrolizumab can significantly improve the progression-free survival, overall survival (OS), and objective response rates in patients with advanced renal cell carcinoma in the CLEAR trial [9]. Santoni M et al. reported that the combination of pembrolizumab and TKI inhibitors such as lenvatinib or acitinib significantly improved OS in advanced renal cell carcinoma patients [10]. Although targeted therapy and immunotherapy have achieved encouraging results in advanced renal cell carcinoma patients, the prognosis of patients with metastatic KIRC remains poor [11]. Therefore, we must identify additional therapeutic targets and develop more drugs to treat KIRC.

Ferroptosis is a special form of cell death triggered by iron-dependent phospholipid peroxidation and is characterized by distinct morphological, biochemical, and genetic features [12, 13]. The main cause of tissue damage caused by lipid peroxidation may be the ferroptosis pathway [14]. Ferroptosis is closely related to many biological processes [13] and may also regulate the tumour immune microenvironment, which has enormous research value in cancer treatment [12, 15]. CD44 is a non-kinase cell surface transmembrane glycoprotein encoded by human chromosome 11 [16] and is a ferroptosis suppressor gene. CD44 is significantly upregulated in many types of tumours and considered a molecular marker for cancer stem cells [17]. Silencing CD44 can reduce the glycolytic phenotype of cancer cells, which affects the glucose uptake, ATP production, and lactate production and produces an inhibitory effect on tumour growth [18]. CD44 is considered a potential target for various tumour treatments [19, 20], with treatment strategies including CD44 neutralizing antibodies, tumour delivery of shRNAs, ectodomain mimics, and aptamers [21]. CD44 is also a prognostic factor for the survival of renal clear cell carcinoma patients [22], but its potential mechanism in KIRC remains unclear. Iron-sulfur clusters serve as crucial cofactors for numerous enzymes in eukaryotic cells. GLRX5 is involved in iron–sulfur [Fe–S] cluster biogenesis, and defects in this complex process can lead to various diseases [23, 24]. GLRX5 is a type of ferroptosis suppressor gene that participates in the development of tumours through the ferroptosis pathway, and is a potential target for tumour treatment [25]. Currently, there is very little research on the role of GLRX5 in renal cell carcinoma.

Once renal cell carcinoma invades local organs or spreads to distant areas and enters the advanced stage, treatment of the tumour becomes notably difficult, and the prognosis is poor [26]. Therefore, the treatment of advanced renal cell carcinoma faces enormous challenges [27]. The purpose of this study is to explore the potential role of ferroptosis-related genes in KIRC, and identify new therapeutic targets and immunotherapy markers to achieve precise treatment and prognostic evaluation of KIRC patients.

2 Materials and methods

2.1 Data collection and preprocessing

Clinical data and gene transcription information for KIRC patients were downloaded from TCGA (https://portal.gdc.cancer.gov/). This study included RNA-seq profiles from 614 KIRC samples in the HT-SEQ-TPM dataset and processed clinical data from 537 patients via the R package “XML”, including age, sex, T stage, N stage, M stage, pathological grading, survival time, and survival status. Kaplan–Meier survival analysis was performed on prognostic information from 493 samples (excluding 3 samples that lacked survival status and 41 samples with survival times less than 30 days).

The tumour mutation burden (TMB) was calculated based on simple nucleotide variation and defined as the number of mutations per megabase. Simple nucleotide variation data from 375 patients with KIRC were collected from TCGA, and the mutational conditions in KIRC were evaluated via the R package “maftools”. Waterfall plots were used to show the genetic mutation of patients via the R package “ComplexHeatmap” [28].

The gene expression profiles of the GSE53757 dataset, including 72 normal and 72 tumour samples, were used to validate the immunological features between the two clusters; these data were downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/geo/).

In total, 304 ferroptosis-related genes were acquired from FerrDb, including 3 ferroptosis marker genes, 165 ferroptosis suppressor genes, and 152 ferroptosis driver genes (http://www.zhounan.org/ferrdb/current/).

We compared the expression of CD44 and GLRX5 proteins between KIRC and normal kidney tissue by analyzing immunohistochemical images from the Human Protein Atlas (HPA) data resource (https://www.proteinatlas.org/).

2.2 Identification of differentially expressed overlapping ferroptosis-related genes

The expression profiling data (HT-Seq-TPM) were compared to identify the differentially expressed genes (DEGs) (normal vs. cancer; Cluster 1 vs. Cluster 2) using the R package “limma” and Wilcoxon rank-sum test. The threshold values were |log2FoldChange|> 1 and adjusted P value < 0.05.

Overlapping analyses were performed for the differentially expressed genes with the recognized ferroptosis-related genes to screen the ferroptosis-related differentially expressed genes (FR-DEGs). The overlap of DEGs and FRGs was visualized using a Venn diagram generated via online websites (https://bioinfogp.cnb.csic.es/tools/venny/).

Univariate Cox regression analyses and subsequent multivariate Cox regression analyses were used to investigate the relationships of the FR-DEGs with the overall survival. Subsequently, least absolute shrinkage selection operator (LASSO) regression analyses were performed to screen suitable FR-DEGs as biomarkers [29]. The minimum lambda of the lasso model was selected via tenfold cross-validation.

2.3 Consensus clustering based on CD44 and GLRX5

To evaluate the tumour microenvironment of KIRC, we used the R software package “estimate” to calculate the scores of the tumour samples, including the stromal score, immune score, ESTIMATE score, and tumour purity [30]. After the correlations between prognosis-related genes and the scores had been calculated, CD44 and GLRX5 were selected because they had the highest correlation with the scores. The TCGA-KIRC samples were divided into two clusters using the R package “ConsensusClusterPlus” [31]. Clusters were verified via principal component analysis (PCA). The gene expression of CD44 and GLRX5 in different clusters was visualized via the R package “pheatmap” (the samples were subjected to log2(TPM + 1) transformation). Finally, a correlation analysis was conducted on the clinical information of the two clusters.

2.4 Survival analysis

The predictive effect of clustering was illustrated using Kaplan–Meier survival curves, which were generated using the R package “survminer”. A nomogram based on the risk score model, age, clinical stage, T stage and clustering was constructed using the R package “rms”.

2.5 Gene set variation analysis (GSVA) of ferroptosis-related genes

The KEGG pathway was analysed to explore the differences in biological processes between the two clusters using the R package “GSVA” [32].

2.6 Tumour immune infiltration analysis

CIBERSORT was used to calculate the proportions of 22 immune cells in each sample with KIRC [33]. The sum of 22 immune cell type fractions in each sample was 1. An ssGSEA analysis was performed using R packages “limma”, “GSVA”, and “GSEABase” to calculate the extent of infiltration of 23 immune cells in each KIRC tumour sample.

2.7 Enrichment analysis

GO and KEGG analyses were conducted to understand the functions of the hub genes using the R package “clusterProfiler” [34, 35].

2.8 Drug sensitivity assessment

The R package “oncoPredict” includes accurate and convenient drug response prediction methods, which makes it easy to predict sensitive drugs in cancer patients [36]. Drug sensitivity analysis was performed on the two clusters using this R package.

2.9 Weighted gene coexpression network analysis

We performed WGCNA to identify highly correlated gene modules in the two clusters. Overlapping DEGs in the two clusters were subjected to WGCNA using the R package “WGCNA” [37]. In total, 542 samples with 618 overlapping differentially expressed genes were used as an expression matrix for further analysis. We obtained four modules and calculated their correlations with the stromal score, immune score, ESTIMATE score, tumour purity, cluster, sex, age, and pathological grade. Ultimately, we obtained 12 genes according to the calculation of module membership (MM) and gene significance (GS).

2.10 PPI network construction

To explore the interactions of the hub genes at the protein level, we used the STRING database (https://cn.string-db.org/) to construct a PPI network and set the medium confidence (0.400) interactive score. Subsequently, Cytoscape (version 3.9.1) was used for visual display.

2.11 Construction and validation of the nomogram

A nomogram was constructed using the “rms” R package after the selection of variables. The univariate Cox regression analysis and multivariate Cox regression analysis were performed to identify the independent prognostic factors of KIRC patients. C-index and calibration plots were generated to assess the predictive performance of the established nomogram model. A decision curve analysis (DCA) was performed using the rmda package to compare the benefit of the nomogram and TNM system in OS prediction.

2.12 Statistical analysis

All statistical analyses were performed in the R software (version 4.3.1). Adobe Illustrator (CC 2021) was used to piece the figure panels together. Wilcoxon rank-sum test was used for the box plot analyses. Survival curves were constructed using Kaplan–Meier method (log-rank test) [38]. Correlation analysis was performed (genes–genes, genes–ESTIMATE scores, and gene–ssGSEA results) using the Spearman’s coefficient. A p-value < 0.05 was considered statistically significant.

3 Results

3.1 Identification of differentially expressed overlapping immune-related genes

We first identified the differentially expressed genes in the TCGA-KIRC cohort, and the results showed that 3515 genes were differentially expressed, including 1135 downregulated genes and 2380 upregulated genes (Fig. 1A). Then, we obtained 304 ferroptosis-related genes from the FerrDb database and screened 63 differentially expressed the overlapping ferroptosis-related genes for further analysis (Fig. 1B). To further screen for genes related to the prognosis, we first applied 63 genes to screen for candidate prognostic genes and identified 24 genes, via the univariate Cox proportional hazard regression analysis (Fig. 1C). Then, we conducted a multivariate Cox regression analysis (using the "both" regression method) and identified 12 genes (Fig. 1D). We subsequently applied these 12 genes to perform a LASSO regression and identified 9 genes with a minimum lambda value of 0.02928 (Fig. 1E). Based on the results of the LASSO regression, a risk score model was built with the following formula: Risk score = 0.0017625956*CD44 + 0.0045543002*CD82 + 0.0234827079*ETV4 − 0.0111784895*GLRX5 − 0.0005476334*HMOX1 + 0.0238423527*KIF20A -0.1067884492*MYCN + 0.0404748206*PLA2G6 + 0.0007843920*TF. The KIRC patients in the TCGA cohort were divided into high-risk and low-risk groups according to the median of the risk score (Supplemental Fig. 1A).

Fig. 1
figure 1

Identification of key ferroptosis-related genes involved in immune infiltration of KIRC. A Volcano plot of differentially expressed genes in TCGA-KIRC dataset. B Venn diagram of overlapping genes. C Univariate Cox regression analyses of the 63 overlapping ferroptosis-related genes. D Multivariate Cox regression analyses of the 24 genes obtained from results of univariate Cox regression analyses. E LASSO regression analyses to screen the 12 genes obtained from results of multivariate Cox regression analyses. F Association between the 9 ferroptosis-related genes and results of ESTIMATE

Full size image

To explore the potential role of FRGs in the tumour immunity of KIRC, we calculated four scores for each sample using ESTIMATE, and the correlation between the expression of the 9 FRGs and the results of ESTIMATE was evaluated (Fig. 1F). Considering that they had the strongest correlation with immune scores, CD44 and GLRX5 were included in the subsequent analysis.

3.2 Consensus clustering of KIRC samples based on the expression of CD44 and GLRX5

We investigated the protein expression levels of CD44 and GLRX5 in renal cancer through the HPA database (Fig. 2A). Immunohistochemistry revealed increased protein expression of CD44 and decreased protein expression of GLRX5 in tumour tissues. A comparison of the expression of CD44 and GLRX5 in tumour and normal tissues revealed that tumour tissues had lower GLRX5 expression and higher CD44 expression than normal tissues (Fig. 2B). Then, we performed consensus clustering on 542 tumour samples based on the expression of CD44 and GLRX5. The results showed that k = 2 was the best parameter for dividing KIRC samples into two clusters (Supplemental Fig. 1B). The heatmap showed a difference between the two clusters, and we found that Cluster 1 had low expression of CD44 and high expression of GLRX5, whereas Cluster 2 had low expression of GLRX5 and high expression of CD44 (Fig. 2C). We also detected a weak negative correlation (r = − 0.14, p = 4.02e−4) between CD44 and GLRX5 (Fig. 2D). To further understand the characteristics of the two clusters, we conducted PCA, which clearly classified the two clusters (Fig. 2E). The two clusters had different clinical characteristics, and there were statistically significant differences in gender, pathological grade, clinical stage, and T stage (Fig. 2F). Then, we generated a survival curve and found that Cluster 1 had a higher survival probability than Cluster 2 (Fig. 2G). These results indicate that TCGA-KIRC can be classified based on the CD44 and GLRX5 expressions.

Fig. 2
figure 2

Clustering of TCGA-KIRC patients based on CD44 and GLRX5, and clinical features of two clusters. A Comparison of immunohistochemistry images of CD44 and GLRX5 between tumor and normal kidney tissues based on the Human Protein Atlas. B Comparison of expression of CD44 and GLRX5 between tumor and normal tissues. C TCGA-KIRC patients were divided into two clusters according to CD44 and GLRX5. Heatmap expression of CD44 and GLRX5 between two clusters. D Association between CD44 and GLRX5 expression. E Principal component analysis of two clusters. F Comparison of clinical features between two clusters. G Kaplan–Meier survival curves of two clusters. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ns, not significant

Full size image

3.3 Nomogram development and different characteristics of biological behaviours between the two clusters

To explore the clinical value of this novel clustering method, we developed a nomogram to predict the OS of KIRC patients based on the clustering and clinical features. First, we screened variables with p < 0.05 via univariate Cox regression, including age, T, N, M, risk score, and stage. Then, these six variables were incorporated into a multivariate Cox regression model, and the stepwise backward regression method was used to determine the optimal model with the minimum Akaike information criterion (AIC). The model constructed with age, M stage, risk score, and T stage achieved a minimum AIC of 1016.37. To explore the clinical value of the clusters, the variable of clustering was included. Based on the above results, a nomogram was constructed that incorporated the risk score model, age, clustering, T stage, and M stage, which provided an effective predictive tool with a C-index of 0.814 (Fig. 3A). As shown in the figure, the overall survival probabilities at 1, 3, and 5 years were 0.974, 0.910, and 0.826, respectively. The calibration curves (Fig. 3B) show that our predicted values are close to the actual values. The decision curve analysis shows that the nomogram model performed better than the TNM system (Fig. 3C).

Fig. 3
figure 3

Construction of the nomogram and comparison of immune characteristics between two clusters. A The nomogram was constructed based on the risk score model, age, M-stage, T-stage and the clustering. B Calibration curves of 1, 3, and 5 years overall survival. C Decision curve analysis of the nomogram and the TNM system in predicting the OS of KIRC. D KEGG pathway analysis of GSVA in two clusters. E Comparison of stromal score, immune score, ESTIMATE score, between two clusters. F Comparison of proportion of immune cells between two clusters. G Comparison of expression of immune cells between two clusters. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ns, not significant

Full size image

GSVA was performed to explore the functional differences between the two clusters. The results revealed that Cluster 1 was mainly enriched in metabolism-related pathways such as arginine and proline metabolism, alanine aspartate and glutamate metabolism, glycine serine and threonine metabolism, tryptophan metabolism and fatty acid metabolism. Cluster 2 was enriched in many significant immunity-related pathways such as the vascular endothelial growth factor (VEGF) signaling pathway, MAPK signaling pathway, antigen processing and presentation, allograft rejection and B-cell receptor signaling pathway (Fig. 3D). Moreover, ESTIMATE, CIBERSORT, and ssGSEA were performed to identify differences in immunological function. Compared with Cluster 1, Cluster 2 had higher stromal, immune, and ESTIMATE scores (Fig. 3E). The CIBERSORT results revealed that Cluster 2 had a higher proportion of plasma cells, CD8 T cells, CD4 memory activated T cells, follicular helper T cells, regulatory T cells (Tregs), gamma delta T cells, Macrophages M0, Macrophages M2 and Neutrophils (Fig. 3F). The ssGSEA results revealed that most immune cell subtypes were highly expressed in Cluster 2, except for eosinophil, neutrophil and CD56dim natural killer cells (Fig. 3G). These results indicate that Cluster 2 tended to have a greater level of immune infiltration than Cluster 1.

3.4 Drug sensitivity evaluation and comparison of genetic mutations

To explore the sensitivity of the two clusters to drugs, we conducted an oncoPredict analysis and obtained several sensitive drugs. Cluster 2 showed more sensitivity to drugs such as KU-55933, staurosporine, BMS-536924 and YK-4-279 than Cluster 1 (Fig. 4A). In addition, we identified targets for several common immunomodulatory drugs [39]. We compared the expression of these immune regulatory targets between the two clusters and found that the expression of most immune regulatory targets (PDCD1, CTLA4, LAG3, PDCD1LG2, TIGIT, and CD86) was significantly higher in Cluster 2 (Fig. 4B, C). These results indicate that the patients in Cluster 2 might exhibit a better response to immunotherapy than those in Cluster 1.

Fig. 4
figure 4

Exploration of sensitive drugs and comparison of mutational landscapes between two clusters. A Comparison of sensitive drugs between two clusters. B, C Comparison of immunomodulatory targets between two clusters. D Mutational landscape of Cluster 1. E Mutational landscape of Cluster 2. F Comparison of tumor mutation burden between two clusters. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ns, not significant

Full size image

Genetic mutations may significantly impact the efficacy of immunotherapy. Therefore, we compared the mutational conditions between the two clusters and found no statistically significant difference in TMB (Fig. 4D–F).

3.5 Identification of genes that are highly correlated with the clusters and immunity

First, we obtained 618 DEGs (463 upregulated and 155 downregulated) between the two clusters (Fig. 5A). Then, we performed WGCNA to identify the gene modules that were highly correlated with the clusters and immunity (Fig. 5B, C). Afterwards, we performed a correlation analysis between modules and traits, as shown in the heatmap (Fig. 5D). The blue module, which was most relevant to the clusters (R = 0.49, P = 2e−30) and immunity (R = 0.77, P = 1e−93), was selected. Ultimately, we obtained 12 hub genes (C1QC, CD96, CXCR6, DOCK2, FCGR1B, FYB1, LAIR1, PLEK, PTPN22, RHOH, SLAMF8 and TNFSF13B) from the blue module based on MM > 0.8 and GS > 0.4 for further analysis (Fig. 5E).

Fig. 5
figure 5

Identification of module genes associated with clustering and immune score in WGCNA. A Volcano plot of differential analysis between two clusters. B Determination of the soft thresholding power. C Dendrogram of differentially expressed genes clustered based on a dissimilarity measure (1-TOM). D Correlations of gene modules with clinical traits. E Scatter plot of module eigengenes in the blue module

Full size image

3.6 Functional enrichment and immune infiltration of the hub genes

To explore the role of the hub genes in the blue module, we performed GO and KEGG pathway enrichment analyses. The results (Fig. 6A) revealed that these genes were highly enriched in the negative regulation of the immune response (GO–BP), external side of the plasma membrane (GO–CC) and immune receptor activity (GO–MF). The KEGG pathway enrichment analysis showed that the 12 hub genes were enriched in the staphylococcus aureus infection, Chagas disease, osteoclast differentiation and systemic lupus erythematosus (Fig. 6B). We also explored the correlation between the 12 genes and immune infiltration (Fig. 6C, D). To further explore the interactions of these 12 genes, we constructed a PPI network using the Cytoscape software and obtained 10 nodes and 13 edges (Fig. 6E). Overall, the genes in the blue module were significantly correlated with immunity.

Fig. 6
figure 6

Enrichment analysis and immune infiltration analysis of 12 hub genes. A The GO analysis of hub genes. B The KEGG analysis of hub genes. C Correlation between hub genes and results of ESTIMATE. D Correlation between hub genes and expression of immune cells. E PPI network of hub genes

Full size image

3.7 Verification of clustering and the effects of CD44 and GLRX5

First, we compared the expression of CD44 and GLRX5 between normal and tumour samples in GSE53757 (Fig. 7A) and performed consensus clustering (Fig. 7B). Next, the correlation between CD44 and GLRX5 was calculated (Fig. 7C), and immune infiltration analysis (ESTIMATE, CIBERSORT, and ssGSEA) was performed on the GSE53757 dataset (Fig. 7D–F). Finally, the expression of the immune targets was compared (Fig. 7G, H), and a drug sensitivity analysis was performed between the two clusters (Fig. 7I). The results indicated that tumour immune infiltration was more active in Cluster 2 than in Cluster 1 in the GEO cohort, which was consistent with the results in the TCGA cohort.

Fig. 7
figure 7

Verification of biological characteristics between two clusters. A Comparison of expression of CD44 and GLRX5 between tumor and normal tissues in GSE53757. B GSE53757 samples were divided into two clusters according to CD44 and GLRX5. Heatmap expression of CD44 and GLRX5 between two clusters. C Association between CD44 and GLRX5 expression in GSE53757. D Comparison of stromal score, immune score, ESTIMATE score, between two clusters in GSE53757. E Comparison of proportion of immune cells between two clusters in GSE53757. F Comparison of expression of immune cells between two clusters in GSE53757. G, H Comparison of immunomodulatory targets between two clusters. I Comparison of sensitive drugs between two clusters in GSE53757. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, ns, not significant

Full size image

4 Discussion

In this study, we identified the key ferroptosis-related genes CD44 and GLRX5, which are involved in KIRC tumour immunity. The TCGA-KIRC samples were divided into two clusters according to the CD44 and GLRX5 expressions. GSVA was performed to understand the functional pathways between the two clusters. Immune infiltration analysis between the two clusters was conducted using ESTIMATE, CIBERSORT, and ssGSEA. The results showed that immune infiltration was more active in Cluster 2 than in Cluster 1. Next, we investigated the drug sensitivity, tumour mutation burden and expression of immune targets of the two clusters. WGCNA was used to screen for hub genes that were highly correlated with KIRC tumour immunity and the clusters. The biological processes associated with the hub genes were analyzed using GO and KEGG, the correlation between infiltrating immune cells and hub genes were analyzed, and the PPI analysis of hub genes was performed. In summary, we identified the key ferroptosis-related genes CD44 and GLRX5 involved in KIRC tumour immunity, identified two subgroups of KIRC and constructed a nomogram to predict the overall survival of KIRC patients.

CD44 participates in regulating the tumour microenvironment and plays an important role in tumour progression and metastasis, which makes it a potential target for tumour therapy [40]. A study shows that SNX5 can inhibit the expression of CD44, which traffics CD44 and inhibits the proliferation, migration, and invasion of KIRC cells by regulating CD44 [41]. Our research revealed that CD44 was expressed at high levels and GLRX5 was expressed at low levels in KIRC. We discovered a weak negative correlation between CD44 and GLRX5(r = -0.14, p = 4.02e-4). In the ESTIMATE analysis, their correlation with the immune score was exactly opposite. Therefore, CD44 and GLRX5 may have opposite functions in the immune infiltration of KIRC patients. GLRX5 is a ferroptosis-inhibiting gene, and the activation of ferroptosis can lead to nonapoptotic forms of cancer cell death [25, 42]. Inhibiting GLRX5 can increase the level of intracellular free iron and promote ferroptosis in drug-resistant cancer cells to overcome head and neck cancer chemoresistance [43]. However, there is currently limited research on the potential molecular mechanisms of GLRX5 in renal clear cell carcinoma. In our study, we explored the potential mechanisms and clinical value of CD44 and GLRX5 in KIRC tumour immunity.

Previous studies have shown that the FRGs CD44 and GLRX5 are closely related to the prognosis of KIRC patients [22, 44], which is consistent with our results. Unlike previous studies, we identified CD44 and GLRX5 as the key ferroptosis-related genes in KIRC tumour immunity and identified the KIRC ferroptosis-related subgroups based on CD44 and GLRX5. The TCGA-KIRC samples were divided into two clusters. The expression of CD44 and GLRX5 in Cluster 1 was similar to that in normal samples, which suggests that tumour cells in Cluster 1 may exhibit characteristics of normal mature cells. Survival analysis revealed that the patients in Cluster 1 had earlier clinical stages and a higher probability of survival, which suggests that the biological processes of tumour cells in Cluster 1 may be closer to those of normal cell types. Based on this novel clustering, we constructed a nomogram that included the risk score model, age, T stage, M stage and clustering. Both C-index and calibration plots indicate the effectiveness of our model in evaluating the OS of KIRC patients. The decision curve analysis shows that the nomogram model performed better than the TNM system in predicting the OS of KIRC patients.

The results of GSVA revealed that several immune-related pathways were significantly enriched in Cluster 2, such as the VEGF signaling pathway, MAPK signaling pathway, and antigen processing and presentation. The VEGF pathway promotes angiogenesis, which provides oxygen and nutrients for renal tumours and contributes to its growth and spread [45]. In clinical practice, various VEGF inhibitors have been recommended for the treatment of KIRC [5]. Similarly, the MAPK signaling pathway is widely used in tumour pathway research and is activated by various factors such as EGFR and the Ras protein family. The MAPK pathway promotes the proliferation of tumour cells and can affect the sensitivity of KIRC patients to targeted drugs [46]. The GSVA results indicate that Cluster 2 might have a stronger immune response in the tumour microenvironment.

ESTIMATE analysis revealed that Cluster 2 had higher stromal, immunity, and ESTIMATE scores than Cluster 1, which indicates that Cluster 2 had more active tumour immune infiltration. CD8 T cells played a significant role in anti-cancer immunotherapy, and high levels of activated CD8 T cells are associated with a good prognosis [47]. However, in renal clear cell carcinoma, infiltration by CD8 T cells has been associated with a worse prognosis [48]. Exhaustive CD8 T cells refer to the gradual loss of the ability of CD8 lymphocytes to secrete pro-inflammatory cytokines and enhance their cytotoxic function after long-term exposure to chronic infections or sustained antigen stimulation [49, 50]. With advancing disease stage, CD8 T cells are progressively dysfunctional and have a more terminally exhausted phenotype in the tumour microenvironment of KIRC [51]. Our results revealed a significant increase in proportion of CD8 T cells and a poorer prognosis in Cluster 2, which may indicate that Cluster 2 represented an advanced stage of renal clear cell carcinoma. Myeloid cells exhibit both anti-tumour and immunosuppressive effects and play complex roles in the tumour microenvironment [52]. Macrophages are commonly divided into two phenotypes: M1 (anti-tumoral) and M2 (pro-tumoral), but this division is controversial. A new classification of tumour-associated macrophages in KIRC was recently proposed [53]. M0 macrophages typically exhibit a resting state and require stimulation to acquire the M1 or M2 phenotype [54]. In Cluster 2, the proportion of M2 macrophages in Cluster 2 was higher than the proportion of M1 macrophages, which may be due to greater activation promoting M0 macrophage polarization to the M2 phenotype in Cluster 2. A study revealed that M2 macrophages played a crucial regulatory role in the malignant progression of KIRC [55]. The patients in Cluster 2 had a greater proportion of M2 macrophages and a worse prognosis, which is consistent with previous findings. The immune infiltration analysis results revealed different patterns of immune cell abundance between the two clusters, which may be due to the impact of CD44 and GLRX5 on the TME through ferroptosis resistance.

Immune checkpoint receptors such as PD-1, CTLA4, and LAG-3 play crucial roles in regulating the immune responses and maintaining self-tolerance, and they help cancer cells evade immune surveillance [56]. Immune checkpoint inhibitors (ICIs) (anti-PD-1 drugs such as pembrolizumab and nivolumab; anti-CTLA-4 drugs such as Ipilimumab) aim to block these interactions and disrupt the immune escape ability of cancer cells [57]. The US Food and Drug Administration has approved different types of ICIs for cancer treatment [58], and some ICIs have successfully improved the survival rates for patients with advanced renal cancer [59, 60]. We found that the expression of PDCD1, CTLA4, and LAG3 in Cluster 2 significantly increased, which suggests that Cluster 2 patients were more likely to benefit from the ICI treatment. WGCNA was performed to identify 12 genes (such as C1QC and CRCX6) that were most relevant to the clustering and immune scores. Knockdown of C1QC inhibits the proliferation, migration, and invasion of KIRC cells and is negatively correlated with the clinical prognosis of KIRC [61]. The high expression of CRCX6 is a high-risk factor for KIRC patients, which leads to significant reductions in OS and RFS [62]. The increased expression of C1QC and CRCX6 suggests that the patients in Cluster 2 might have greater risk of tumour recurrence and progression.

Although our study identified crucial ferroptosis-related genes and subgroups of KIRC and revealed the relationship between CD44/GLRX5 and tumour immunity, several limitations remain. First, the sample sizes of the RNA-seq data from the TCGA-KIRC and GSE53757 cohorts were limited. Second, we could not verify the survival differences between the two clusters because of the lack of clinical features of patients in GSE53757. Finally, this study was a retrospective study. Further exploration of the specific molecular mechanisms and downstream pathways of CD44/GLRX5 through basic experiments is necessary.

5 Conclusion

This study identified key ferroptosis-related genes CD44 and GLRX5 in KIRC tumour immunity and explored the subgroups of KIRC. Cluster 1 had low expression of CD44 and high expression of GLRX5, whereas Cluster 2 had low expression of GLRX5 and high expression of CD44. Cluster 2 may represent the advanced stage of renal clear cell carcinoma with a poor prognosis and more active immune infiltration, which may lead to a more positive response to immunotherapy. This study helps lay a foundation for further studies on the role of CD44 and GLRX5 in the carcinogenesis and progression of KIRC.