The transcriptome profile and corresponding clinical information for ccRCC samples were downloaded from The Cancer Genome Atlas (TCGA-KIRC, http://protal.gdc.cancer.gov/). Furthermore, we obtained the validation cohort with follow-up information from the ArrayExpress database (E-MTAB-1980, https://www.ebi.ac.uk/arrayexpress/). For the data format, the fragments per kilobase million (FPKM) were transformed into transcripts per million (TPM). Further, the ccRCC and normal kidney single cell sequencing data were downloaded from the GEO database (GSE159115). We used Seurat v4 to process single cell data and merge them. Afterward, we annotated the different clusters according to the marker genes reported in the previous studies. Also, we collected preoperative coagulation data from the case system for ccRCC patients from the First Affiliated Hospital of Yangzhou University, and all protocols met the requirements of the ethics committee of Yangzhou University.

In order to classify ccRCC patients for personalized treatment, we performed molecular subtyping based the CRGs using the ConsensusClusterPlus R package with the follow setings (maxK=7, reps=100, pItem=0.8, pFeature=1, distance=“manhattan”, clusterAlg=“pam”) (18). Meanwhile, unsupervised clustering and the corresponding representative data were generated using the ggplot2 R package. To explore the immune cell infiltrations between two clusters, the single sample gene set enrichment analysis (ssGSEA) was performed and each type of immune cell was calculated according to the score. After that, the effect of immunotherapy between different clusters were evaluated using the Cancer Immune Atlas (TCIA) database. The potential molecular enrichment between the clusters was annotated with the ClueGO plugin from the Cytoscape software. We then used the pRRophetic R package to assess of the sensitivity between different clusters to clinical drugs in advanced ccRCC (19).

First, we retrieved the CRGs from the MsigDB database (HALLMARK_COAGULATION and MALLMARK_COMPLEMENT, http://software.broadinstitute.org/gsea/msigdb), and the detailed information on the 281 CRGs were shown in the Supplementary Table S1 (20). Then, in the R environment, the limma R package was condemned to screen out the differentially expressed CRGs (DECRGs) between ccRCC and normal samples, based on the set cutoff criteria of P<0.05 in TCGA and single-cell sequencing cohorts (21). Univariate Cox regression was used to identify the prognostic CRGs. Genes with the P value less than.05 in Cox regression were identified as prognostic genes for further LASSO regression analysis.

The human RCC (786-O, 769-P, A498, ACHN, Caki-1, OS-RC-2, RCC4) and normal kidney cell lines (293) were cultured in Dulbecco’s modified eagle’s medium or RPMI 1640 with 10% fetal bovine serum and 1% Penicillin/Streptomycin. The cell lines were placed in a 37°C aseptic incubator with 5% CO₂,and the fluid was changed every 2-3 days. The cell precipitation was collected, and the total RNA was extracted by TRIzol Reagent (Invitrogen). The patient samples were collected from the First Affiliated Hospital of Yangzhou University, and all the procedures were approved by the Ethics Committee. Then, cDNA synthesis was reverse transcribed using the Takara reagent kit. Then we performed quantitative PCR through SYBR green SuperMix and calculated the results using 2-ΔΔCT method (22). The primers used in this study can be found in Supplementary Table S2 .

To explore the prognostic value of the CRGs in ccRCC patients, we performed the LASSO regression of overall survival (OS) with a maxit=1000, using the glmnet R package. We divided the ccRCC samples in TCGA into training and validation cohorts according to the ratio of 3:2 and used the samples of E-MTAB-1980 as the external validation cohort. Then we calculated the risk score of each patient using the following formula: risk score = coffiCRG1 × CRG1 expression + coffiCRG2 × CRG2 expression + · ···· + coffiCRGn × CRGn expression. Afterwards, we used the timeROC R package to draw the patient’s 1-, 3-, and 5-year ROC curve to evaluate the prognostic value of the signature. At the same time, we used the survival R package and Kaplan-Meier ‘s method to compare the prognostic differences between high- and low-risk patients. Finally, through the univariate and multivariate regression analyses, we established a prognostic nomogram integrating the risk score and independent clinical parameters.

All statistical analyses were conducted using R 4.1.3 and Prism GraphPad 9.3. Continuous variables were compared by using Student’s t test, the Mann–Whitney test, or the Wilcoxon rank-sum test. Meanwhile, cumulative survival analyses were performed using the Kaplan–Meier method, and the survival differences were analyzed using the log-rank test by survival R package. The predictive value of signature was evaluated using ROC cures. An AUC value greater than 0.75 is considered well, and a value greater than 0.60 is considered acceptable. The univariate and multivariate Cox regression were used to assess the correlation of the signature and clinical parameters with overall survival. Among all the results, P < 0.05 was considered to be statistically significant.

First, the workflow for the whole study is shown in Figure 1 . In our clinical work, we have found that patients with high stage or grade tend to be accompanied by more pronounced coagulation system dysfunction, which seems to be an interesting direction ( Supplementary Figure S1 ). Based on the clinical founding and gene set of HALLMARKS, we obtained a total of 279 expression matrix of CRGs in the TCGA database. These CRGs were analyzed by KEGG Tree, and the functions of these CRGs were divided into a total of five broad categories, including apoptosis, coagulation, and important pathways such as complement and cytokines ( Figure 2A ). Subsequently, we performed a cluster analysis of the ccRCC single cell data (GSE159115) and annotated the different Clusters ( Figure 2B ), and the Dot plot of the relevant features is also shown in Figure 2C . Subsequently, we identified copy number variants in epithelial cells by the SCEVAN method and classified normal epithelial cells and malignant epithelial cells. By performing differential expression analysis between the two types of epithelial cells, we obtained a total of 527 differentially expressed genes with p <0.05. Meanwhile, the CRGs in TCGA data were analyzed by Limma package, and a total of 223 DECRGs were obtained with p <0.05 ( Figure 2D ). The DEGs from the single cell sequencing data were subsequently intersected with the DECRGs from TCGA to obtain a total of 16 hub genes ( Figure 2E ). We then performed correlation and regulatory pathway analysis by GeneMANIA on 16 hub genes, which are mainly involved in transcriptional regulatory functions in response to stress as well as coagulation-related mechanisms ( Figure 2F ).

To further investigate whether CRGs-based treatment can be individualized for ccRCC patients. We performed consistent clustering of ccRCC patients in TCGA based on 279 CRGs and initially classified patients into 4 clusters based on the decay of CDF values ( Figure 3A, B ). By performing survival analysis, the four Clusters showed obvious two survival states, so we merged Cluster A and Cluster C into Cluster 1 and Cluster B and Cluster D into Cluster 2. The two Clusters showed obvious survival differences between them ( Figure 3C ). It could also be seen by principal component analysis that when the TCGA samples are divided into 2 Clusters, the samples can be clearly separated, while with 4 Clusters, the boundaries of sample separation are not obvious ( Figure 3D ). The subsequent construction of a heatmap also demonstrated the existence of significant changes in gene expression profiles between these two Clusters ( Figure 3E ).

To explore potential biological functions and features between two clusters, we identified DECRGs between clusters by limma R package and performed functional enrichment using ClueGO plugin. We found that patients in Cluster 1 was significantly associated with Platelet Activation, Blood Coagulation, and Smooth muscle cell migration, while patients in Cluster 2 were significantly correlated with Complement and coagulation cascade and Negative regulation of low-density lipoprotein ( Figure 4A ). Subsequently, we explored the sensitivity of two clusters to drugs commonly used in clinically advanced ccRCC by pRRophetic. Compared to Cluster 2, patients in Cluster 1 were more sensitive to axitinib, pazopanib, Voninostat and sorafenib, which may be closely related to their enriched pathways ( Figure 4B ). Since ccRCC is a strongly immunogenic tumor, we then compared the immune cell infiltration of the two types of Clusters. Eosinophil and plasma cell infiltration was more pronounced in Cluster 1 patients, whereas T-cell infiltration was more pronounced in Cluster 2 patients, suggesting that there may be different sensitivities to immunotherapy between the two Clusters ( Figure 4C ). Also, we explored the responsiveness to immunotherapy in the TCIA database. Cluster 2 patients were more sensitive to CTLA4 or PD-1 and the combination of both ( Figure 4D ).

Based on the 16 hub DECRGs, we wanted to know if these genes could be used as markers for the diagnosis of ccRCC. We screened 9 and 7 CRGs using LASSO and SVM-REF regression analyses, respectively ( Figure 5A, B ). We then selected the intersecting genes of the two methods as diagnostic signature, including PSMB9, SPARC, PLG, APOC1, and FDX1 ( Figure 5C ). In the HPA database, we observed the protein expression of these five genes by immunohistochemical data. Compared to normal kidney tissues, PSMB9, SPARC and APOC1 expression levels were upregulated in ccRCC, while FDX1 and PLG expression levels were downregulated in ccRCC ( Figure 5D ). Afterwards, we evaluated the predictive diagnostic value of the diagnostic signature in the TCGA cohort, and the results showed a good predictive value in ccRCC ( Figure 5E ).

To further investigate the prognostic value of CRGs, we conducted a univariate Cox regression of 279 CRGs in relation to OS. A total of 99 prognostic CRGs with P < 0.01 were identified. We then build the prognostic signature by LASSO regression ( Figure 6A, B ). We then calculated the riskScore by using the formula in the material part of the method. Based on the calculated median riskScore cut-off, patients in training and validation cohort were divided into the high- and low-risk groups. Furthermore, the Kaplan–Meier log-rank test and the time-dependent ROC curve were used to evaluate the predictive ability and accuracy of the prognostic signature. The outcome of the Kaplan–Meier log-rank test showed that the high-risk group had a significantly worse OS compared with the low-risk group in the TCGA training set ( Figure 6C ), TCGA validation set ( Figure S2A ), and E-MTAB validation set ( Figure S2C ). Meanwhile, the time-dependent ROC curve proved the 1-year, 3-year, and 5-year predictive accuracy of the signature for OS ( Figures 6D , S2B, S2D ). In addition, the risk score distribution, survival status, and expression of CRGs from the signatures are shown in the TCGA training cohort, TCGA validation cohort, and E-MTAB validation cohort ( Figures 6E , S2E, S2F ). Also, as shown in Figure S3 , we compared our signature with previously published signatures in the TCGA dataset, and the results showed that our signature had better predictive performance, especially for 5-year survival (23, 24).

To predict the prognosis of ccRCC patients more accurately, we identified independent risk factors affecting OS by univariate and multivariate regression analysis ( Table 1 ). The outcomes showed that the Stage, Grade, Age, and Riskscore could be the independent factors for OS of ccRCC patients ( Table 1 ). Combining the calculated riskScore and independent clinical parameters, we established the nomogram with a C-index 0.773 ( Figure 7A ). Then, we performed the calibration curves to verify the predictive efficacy of Nomogram for 1-year, 3-year and 5-year OS ( Figure 7B ). We then confirmed the prognostic value of this Nomogram over the TNM Staging system or the Grade system for ccRCC patients by using multiple ROC curves ( Figure 7C ).

Since the FDX1 was shown to be important in both diagnostic and prognostic signature, we further explored the potential value of FDX1 in ccRCC plus GTEx normal kidney mRNA expression data ( Figure 8A ). We then compared the expression levels of FDX1 in ccRCC and normal kdiney tissues in 8 GEO cohorts and found that the FDX1 were significantly upregulated in normal tissues ( Figure 8B ). Subsequently, based on the Timer database, we found that FDX1 copy number alterations significantly affected the level of infiltration of several major immune cells ( Figure 8C ). We verified FDX1 expression levels in our own clinical samples and cell lines, and consistently with the results in the online database, FDX1 was significantly downregulated in ccRCC ( Figure 8D ).