Data selection and processing
The ccRCC sequencing data (HTseq-FPKM) and the latest corresponding clinical information (Supplementary Table 1) were downloaded from The Cancer Genome Atlas (TCGA) database (https://cancergenome.nih.gov/), including 539 ccRCC samples and 72 normal controls. Meanwhile, we downloaded the transcriptome profile and corresponding survival data from the International Cancer Genome Consortium (ICGC) database (http://daco.icgc.org/) as the validation cohort. Then, we distinguished between lncRNA and mRNA using the human GTF annotation file. The sequencing data processing was conducted via limma and sva packages in R (v 4.0.3).
Cell lines and clinical specimens.
The 293, HK-2, 786-O, 769-P, ACHN, A498, OS-RC-2, and Caki-1 cells were obtained from the American Type Culture Collection (ATCC). All cells were maintained in RPMI-1640 (Corning, USA) or high-glucose DMEM (Gibco, USA) with 10% fetal bovine serum (BI, Israel) and 1% Penicillin and Streptomycin (Gibco, USA) at 37°C and 5% CO2. 10 paired ccRCC and adjacent normal tissues were obtained from ccRCC patients undergoing surgical resection at Peking University First Hospital. The detailed information of 10 paired tissue specimens were shown in the Supplementary Table 2. The Ethics Committee approved this study of PUFH, and all patients signed informed consent. All procedures were performed according to the World Medical Association Declaration of Helsinki.
Real-time quantitative PCR (qPCR)
Total RNA of 10 paired clinical samples and six cell lines were extracted by Takara kit according to the manufacturer's protocol. Then, the RNA was reversed to cDNA in a 20ul reaction system. The quantitation of all gene transcripts was done by qPCR using SYBR Premix ExTaq kit, and TUBA was used as a normalized control. The primer sequences were listed in the supplementary Table 3. Each reaction was performed four times, and the 2^-△△CT method was used to calculate the relative mRNA expression level.
Identification of the prognostic ferroptosis-related differentially expressed LncRNAs
According to the former studies(15-17), we obtained 259 FRGs, and the list was shown (Supplementary Table 4). After that, we screened the ferroptosis-related LncRNAs with the filter (Correlation >0.5, p-value < 0.01), including 2854 LncRNAs. The limma package was conducted to determine the differentially expressed LncRNAs (DELncRNAs) between the ccRCC patients and normal controls, including 1333 LncRNAs(18). We intersected the DELncRNAs and ferroptosis-related LncRNAs to obtain the ferroptosis-related DELncRNAs. Univariate Cox regression was performed in the ferroptosis-related DELncRNAs with overall survival time. Those with p-value < 0.01 were considered prognostic ferroptosis-related DELncRNAs.
Molecular subtyping in ccRCC based on the prognostic ferroptosis-related DELncRNAs
After obtaining the prognostic ferroptosis-related DELncRNAs (FRDELncRNAs), we performed consensus clustering to identify the molecular subtypes of ccRCC by using the "ConsensusClusterplus" R package(19). We selected 80% of the prognostic ferroptosis-related DELncRNAs resampling 100 times and determined clusterings of specified cluster counts (k). Following this, the pairwise consensus values were calculated and stored in a symmetrical consensus matrix for each k. The k, at which there is no appreciable increase, was determined by the cumulative distribution function (CDF) plot and delta area plot. The alteration of Immune infiltration between different clusters was estimated using the Cibersoft method.
Potential biological functions enrichment
To gain insights into the cellular functions directly regulated by FRGs transcriptional control, we compared the list of FRGs to the biological pathways annotated by the Kyoto Encyclopedia of Genes and Genomes (KEGG)(20). Afterward, according to the two clusters, Gene Set Variation Analysis (GSVA) was conducted using the package of the same name in R software v.4.0.3 to investigate the enrichment of HALLMARK pathways with the h.all.v7.4.symbols.gmt gene set from the Molecular Signature Database(21).
Potential therapeutic targets analysis
Since targeted drugs are commonly used to treat advanced kidney cancer, we used the R package "pRRophetic" to estimate drug response as determined by the half-maximal inhibitory concentration (IC50) for each kidney cancer patient on the Genomics of Drug Sensitivity in Cancer (GDSC) website(22). Further, based on the TCIA database, the Immunophenoscore (IPS) was calculated depending on immune therapy data(23).
Construction of the prognostic predictive risk signature
Firstly, the TCGA cohort patients were randomly divided into the training set and internal validation set. Meanwhile, the patient in the ICGC cohort was used as the external validation cohort. Based on the prognostic ferroptosis-related DELncRNAs, we constructed the least absolute shrinkage and selection operator (LASSO) Cox regression using the "glmnet" R package. We calculated each patient's riskscore using the regression coefficient score of the individual LncRNAs and their expression value. Besides, we defined the formula for calculating the prognostic risk score as follow: Risk score = coef(Lnc1)*Exp(Lnc1) + coef(Lnc2)* Exp (Lnc2) + … + coef(Lncn)* Exp (Lncn). Where "coef" represented the coefficient score estimated by LASSO Cox regression, and "Exp" represented the expression value of the individual LncRNAs. The detailed information from the signature was shown (Table 1). Then, we classified the ccRCC patients into the high- and low-risk groups, according to the median risk score of the training group as the cut-off(24).
Validation of the prognostic risk signature
We conducted the Kaplan-Meir and receiver operating characteristic (ROC) curve analyses to assess the prognostic risk signature's validity. According to the calculated median risk score, all samples in each group were divided into high- and low-risk groups, and the survminer and timeROC packages were performed to validate the predictive accuracy in the training and validation sets. The area under the curve (AUC) values corresponding to 1-, 3-, and 5-years were calculated. The time-dependent ROC curve was used to validate the predictive performance of the signature. The AUC value of 0.75 or higher was considered the significant predictive value, and the value of 0.60 or higher was regarded as acceptable for prediction. Furthermore, univariate and multivariate Cox regression was conducted to explore if the ferroptosis-signature(FerroSig) could serve as the independent factors.
Construction and validation of the nomogram
To better predict the prognosis of patients with renal clear cell carcinoma, we established the predictive nomogram based on clinical parameters and prognostic signature(25). Briefly, we first performed univariate and multivariate Cox regression analyses to identify clinical parameters and riskscore that could be used as independent risk factors. Subsequently, the significant factors were used to construct the predicted nomogram. We then evaluated the nomogram effect using calibration curves and time-dependent ROC curves. The AUC value of 0.75 or higher was considered a significant predictive value, and the value of 0.60 or higher was regarded as acceptable for prediction.