The TCGA database (https://portal.gdc.cancer.gov/) was utilized to collect the transcriptomic profiling of RNA-seq data of KIRC patients. Our study included 72 normal and 541 KIRC samples. Subsequently, we gathered the corresponding clinical data for these patients from TCGA. Samples lacking corresponding clinical characteristics or possessing a survival time of less than zero were eliminated, resulting in the inclusion of 532 KIRC samples for the subsequent analysis in this research.

From GeneCards, 496 ARGs were retrieved by relevance score > 0.4 ( Supplementary Table S1 ). The study utilized differentiation analysis with the “limma” package in R software (log2|fold change (FC)| >1 and adjusted P< 0.05) to identify differentially expressed genes (DEGs) in KIRC, comparing all gene expressions in normal and tumor samples using the Wilcoxon test. Afterward, by interacting with ARGs, differentially expressed ARGs were gathered.

The 532 patients with KIRC were separated into training (n = 266) and test (n = 266) groups at random. Table 1 provides a breakdown of the demographic characteristics for the training, test, and entire sets. The training and test sets did not show any statistically significant differences in their clinicopathological variables

To create an anoikis-associated signature, we first identified potential predictive genes in the training set based on differentially expressed ARGs using the univariate regression method. A machine learning technique called LASSO was employed to reduce overfitting. Furthermore, multivariate Cox regression was employed to create a new anoikis-related signature. A risk score formula according to the signature we built was established as follows:

where Coef (i) and exp(i) denote the expression value and the regression coefficient of eachprognosis related anoikis gene for each patient. The study categorized the cases into high-risk and low-risk groups based on the threshold of the median risk score.

We performed univariate and multivariate Cox regression analyses to examine the independence of the risk model through the “survival” package. Clinicopathological characteristics (age, gender, grade, stage) and risk score were employed to generate a nomogram using the “rms” package, and all variables were computed and examined to determine the survival probability of ccRCC patients at 1, 3, and 5 years. To assess the consistency of the predicted and actual survival outcomes, calibration curves were employed. Moreover, receiver operating characteristic (ROC) curves were employed to confirm the prognostic precision of the risk score and clinicopathologic factors via the “pROC” package.

The “limma” package was employed to determine differentially expressed genes in the low-risk and high-risk groups. The analysis employed a cutoff of log2| FC | >1 and adjusted P< 0.05. Furthermore, the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) enrichment analyses were performed using the “clusterProfiler” package.

Six enrichment analysis algorithms, including the TIMER, MCPcounter, xCell, CIBERSORT, QuanTIseq, and EPIC algorithms, were utilized to evaluate cellular elements or cellular immune responses. The “GSVA” package was utilized to determine the immune function scores for all cases. Additionally, we contrasted immune checkpoints expression levels of the low-risk and high-risk groups. In addition, we assessed the scores of tumor immune dysfunction and exclusion (TIDE) for each ccRCC sample by utilizing the TIDE database (http://tide.dfci.harvard.edu/login/) for predicting the possibility of an immunotherapeutic response.

We utilized the TCGA dataset to obtain the mutation data of KIRC patients. The data comprising somatic variations were then evaluated and summarized via the “maftools” package.

Samples of both ccRCC tissues and adjacent normal tissues were provided by the Department of Urology at Ningbo Yinzhou No.2 Hospital. It is necessary to respect the privacy of the individuals involved in the study. The Institutional Research Ethics Committee of Ningbo Yinzhou No.2 Hospital has approved the study (batch number: 2023-KY-003).

To conduct the western blot analysis, RIPA buffer (Solarbio) that contained a protease inhibitor (1% PMSF, Solarbio) was used to lyse the ccRCC and adjacent normal tissues. 30μg of protein lysates were then isolated utilizing 10% SDS-PAGE gels and shifted onto a Millipore PVDF 0.45µm membrane. After the membranes were placed in TBST buffer with 5% non-fat milk to block, specific primary antibodies were added and left to incubate overnight at a temperature of 4°C. In this research, the primary antibodies utilized were: KIF18A (rabbit polyclonal, 1:1000; ab72417, Abcam), BID (rabbit polyclonal, 1:1000; 10988-1-AP, Proteintech), CHEK2 (rabbit polyclonal, 1:1000; 13954-1-AP, Proteintech), CEACAM4 (rabbit polyclonal, 1:1000; A10055, ABclonal), and β-actin (rabbit polyclonal, 1:50,000; AC026, ABclonal). Following extensive washing with TBST buffer, the membranes were then subjected to secondary antibodies (goat anti-rabbit IgG, 1:5000, AS014, ABclonal) at room temperature for another hour. The presence of stains was identified using an advanced chemical reaction known as enhanced chemiluminescence (ECL) reagent from the company Beyotime. The resulting protein bands were then examined using a gel imaging system (Tanon, China).

TCGA and GTEx (https://gtexportal.org/) were utilized to obtain the mRNA expression levels of 541 tumor samples and 100 normal samples of ARGs in the signature. In addition, we accessed TCGA to acquire mRNA expression data for the signature ARGs across 72 paired ccRCC and adjacent normal tissue samples for paired analysis. Finally, five matched pairs of ccRCC and paracancer tissues from our center were used to evaluate protein levels of the signature ARGs.

To compare the differential functions of the two groups, we employed the Wilcoxon rank-sum test. The survival disparities of the two cohorts were evaluated utilizing the Kaplan-Meier survival curve analysis. The pairwise differences between groups were evaluated utilizing the Student’s t-test. All analyses were executed on the R software (version 4.1.3) and a statistical significance threshold of P< 0.05 was applied.

The procedure flow for our research is displayed in Figure 1 . After evaluating data of gene expression from KIRC samples and normal controls in TCGA, we collected 4912 DEGs ( Figure 2A ). The set of DEGs was compared with 496 ARGs obtained from GeneCards. As a result of this comparison, 133 ARGs were determined as being both expressed differentially and shared between the two sets ( Figure 2B ).

532 patients diagnosed with KIRC were assigned randomly into two groups: a test set (n=266) and a training set (n=266), as presented in Table 1 . We employed univariate Cox method to identify 26 significant predictive ARGs from the training set ( Figure 3A ). We then performed LASSO-penalized regression to exclude the overfit gene of signature ( Figures 3B, C ). Finally, the signature of four ARGs (KIF18A, BID, CHEK2, and CEACAM4) was identified employing multivariate cox analysis. Moreover, the risk model was created utilizing the subsequent formula: risk factor = KIF18A ×0.5799 + BID × 0.6975 + CHEK2 × 0.5761 + CEACAM4 × 0.5914. The subjects were categorized into two groups, namely high- and low-risk, based on the median value of the risk score.

As displayed in Figures 4A–C , the scatter dot plot distinguished the risk score and clinical status of KIRC samples. Based on the heatmap, the KIF18A, CHEK2, BID, and CEACAM4 expressions were considerably greater in the high-risk group in both cohorts. The Kaplan-Meier survival curves indicated that in both cohorts, patients in the high-risk group had poor clinical outcomes ( Figures 4D–F ). Moreover, in the entire cohort, Kaplan-Meier survival analysis revealed that patients in the low-risk group presented considerably better progression-free survival ( Figure 4G ). The outcomes demonstrated that the prognostic signature based on ARGs can offer a dependable evaluation of the prognosis for KIRC.

The univariate and multivariate cox regression analyses were used to assess the independence of the risk model. Univariate Cox regression analysis revealed that age (hazard ratio (HR) = 1.032, P< 0.001), grade (HR = 2.309, P< 0.001), stage (HR = 1.897, P< 0.001), and risk score (HR = 1.080, P< 0.001) were all highly associated with overall survival (OS) rate in KIRC ( Figure 5A ). After conducting a multivariate Cox analysis, it was discovered that the risk score remained statistically significant, indicating that the prognostic signature proposed using ARGs could be independent of other clinical factors (as depicted in Figure 5B ). To capitalize on the signature’s prognosis value, a novel nomogram was developed by integrating clinical factors and a risk model for precisely predicting the 1-, 3-, and 5-year survival possibility of KIRC ( Figure 5C ). The calibration plots for predicting survival rates at 1, 3, and 5 years demonstrated that the nomogram possessed considerable accuracy in predicting the prognosis (as indicated in Figure 5D ). Moreover, the prognostic model based on ARGs demonstrated strong predictive ability for 1-, 3-, and 5-year overall survival (OS) based on the high AUC values in the time-dependent ROC analysis ( Figure 5E ). The ROC curve analysis implied that the ARGs prognostic signature had an AUC of 0.726, indicating a stable and reliable predictive ability ( Figure 5F ). The findings suggested that a risk score calculated using ARGs was found to be an independent predictor of prognosis, accurately evaluating the survival probability of KIRC patients relative to clinicopathological characteristics.

To evaluate the prognostic significance of the ARGs signature in predicting overall survival (OS) in KIRC patients based on clinicopathological characteristics, the ccRCC patients were stratified into high- and low-risk groups among various clinical and pathological features. Across all the stratifications, the low-risk group consistently exhibited considerably higher rates of OS in comparison to the high-risk group, as illustrated in Figure 6 . The study demonstrated that the anoikis-related signature for OS can provide an accurate prediction for the prognosis of KIRC, irrespective of the clinicopathological characteristic.

KEGG pathway and GO enrichment analyses were conducted to investigate the biological functions and pathways of genes that were differentially expressed between the high-risk and low-risk groups. GO analysis demonstrated enrichment of immune-related molecular processes such as humoral immune response, complement activation, and regulation of B cell activation ( Figure 7A ). Meanwhile, KEGG pathway studies revealed that DEGs were abundant in the primary immunodeficiency, complement and coagulation cascades, cytokine-cytokine receptor interaction, IL-17 signaling pathway, and chemokine signaling pathway ( Figure 7B ). These findings showed that the function of the ARGs in the tumorigenesis of ccRCC may be regulated by immune-related signaling pathways.

Figure 8A depicted an immune response heatmap based on several enrichment analysis algorithms. The relationship of immune cell subsets and associated functions was examined using ssGSEA, and the results revealed considerable differences in IFN response, MHC, HLA, T cell functions, checkpoint, cytolytic activities, APC, CCR, promoting inflammation, and para-inflammation between the low-risk group and high-risk group ( Figure 8B ). Additionally, our findings revealed a substantial difference in the gene expression of immune checkpoints such as TNFRSF9, CD244, IDO2, and PDCD1 between the two groups ( Figure 9A ). Because the immune infiltration landscape for individuals with ccRCC differs noticeably, the efficacy of immunotherapy in patients with high-risk and low-risk conditions was further examined. TIDE results revealed that patients with a high-risk score had a greater TIDE score, indicating that the low-risk group reacted better to immunotherapy in comparison to high-risk group ( Figure 9B ).

First, the somatic variation driver genes of the low-risk and high-risk groups were compared. The top 15 driver genes with the highest mutation frequency were shown using a waterfall diagram. Accordingly, it was discovered that the prevalence of these gene mutations was generally greater in the high-risk group compared to the low-risk group ( Figures 10A, B ). Subsequently, the tumor mutation burden (TMB) was examined and the TMB of the high-risk group was substantially greater than the low-risk group, as shown by the result (P< 0.001) ( Figure 10C ). We found a highly significant positive relationship between the risk score and TMB (Spearman coefficient: r = 0.24, P< 0.001) ( Figure 10D ). Furthermore, we examined the TMB impact on prognosis. Based on the results, individuals in the high-TMB group exhibited a less favorable prognosis compared to those in the low-TMB group (P = 0.002) ( Figure 10E ). TMB and risk score were assessed for their impact on prognostic stratification due to their synergistic effect. As revealed by the outcomes, in both low- and high-TMB groups, the survival difference of risk score subtypes were significant. (P< 0.001) ( Figure 10F ). The findings may present new strategies for exploring targeted therapy and immunotherapy from the perspectives of gene mutation and anoikis.

Analysis of TCGA and GTEx data revealed significantly lower expression of the four ARGs in normal kidney tissues compared to ccRCC samples ( Figures 11A–D ). Furthermore, paired analysis showed higher expression of these four genes in renal cancer tissues versus matched normal adjacent tissues ( Figures 11E–H ). Western blot validation in five ccRCC and paired paracancer tissues confirmed the mRNA results, with increased protein levels of the four ARGs in ccRCC compared to paracancer tissues ( Figure 11I ).