Clear cell renal cell carcinoma (ccRCC) accounts for approximately 70 to 80% of all RCC pathological subtypes. Despite advances in targeted therapy, survival rates for metastatic kidney cancer remain low, with two-year survival below 20%, and about one-third of patients presenting with metastatic disease at the time of diagnosis.
The Warburg effect describes a metabolic hallmark of cancer in which tumor cells preferentially convert glucose to lactate through glycolysis even in the presence of oxygen. This aerobic glycolysis is particularly pronounced in kidney cancer, where altered glucose metabolism drives cell proliferation, invasion, and metastasis. In ccRCC, approximately 90% of sporadic cases harbor VHL gene mutations that activate HIF-alpha and downstream pro-tumorigenic pathways including VEGF and PDGF.
Multiple glycolysis-related genes (GRGs) have been reported to play distinct functional roles in ccRCC progression. Hexokinase 2 (HK2) promotes cell proliferation and invasion, while FBP1, PLOD2, VCAN, and CD44 have been linked to epithelial-mesenchymal transition. Despite growing evidence connecting GRGs to tumor behavior, no prognostic model for ccRCC based on GRGs had been established prior to this study.
RNA-sequencing data and clinical information for 539 ccRCC tumor samples and 72 normal tissues were obtained from The Cancer Genome Atlas (TCGA). Differentially expressed GRGs were identified using the Limma R package with cutoffs of absolute log2 fold change greater than 1 and FDR less than 0.05. An external validation cohort of 99 ccRCC samples from the E-MTAB-1980 ArrayExpress dataset was used for independent testing.
Gene selection proceeded through three statistical steps: univariate Cox regression identified 40 candidate genes significantly associated with OS; LASSO regression was applied to reduce dimensionality and prevent overfitting by shrinking coefficients of non-informative genes to zero; multivariate Cox regression then finalized a 10-gene panel. A risk score for each patient was calculated as the sum of each gene's expression value multiplied by its regression coefficient.
Validation was performed across three independent sets: the external ArrayExpress cohort, and two internal splits of the TCGA data designated test 1 and test 2. Prognostic performance was assessed using Kaplan-Meier curves with log-rank test, time-dependent AUC curves, and univariate/multivariate Cox regression to confirm independence from clinical covariates including TNM stage, grade, and age.
The final 10-gene signature consists of ANKZF1, CD44, CHST6, HS6ST2, IDUA, KIF20A, NDST3, PLOD2, VCAN, and FBP1. These genes span diverse molecular functions including cell adhesion, extracellular matrix remodeling, enzyme activity, and glycolytic pathway regulation. Several, including FBP1 and PLOD2, have established roles in cancer metabolism and invasion.
Patients were classified into high-risk and low-risk groups using the median risk score of the TCGA training cohort as the threshold. High-risk patients showed significantly lower overall survival compared to low-risk patients (p = 5.548 x 10-13) in the training set. This stratification was consistently reproduced across all three validation cohorts.
Time-dependent AUC values for the signature exceeded 0.70 across all validation datasets for 1-year, 3-year, and 5-year OS prediction. This performance threshold is considered clinically meaningful for prognostic biomarkers and was maintained in the external ArrayExpress cohort, which used a different platform and patient population than TCGA.
Univariate Cox regression confirmed that the 10-GRG risk score was significantly associated with OS across all validation cohorts. Multivariate Cox regression models that included clinical variables (TNM stage, tumor grade, age, sex, and histological subtype) consistently showed the risk score to be an independent prognostic factor in both TCGA and ArrayExpress cohorts (all p less than 0.05).
A prognostic nomogram was constructed integrating the risk score with six clinical factors to predict 1-year, 3-year, and 5-year OS probabilities. The nomogram achieved strong discriminatory performance as measured by the C-index and AUC, with calibration curves showing close agreement between predicted and observed survival across both training and external validation datasets.
Protein expression levels of the 10 hub GRGs were validated in the Human Protein Atlas (HPA) dataset using immunohistochemical staining data from ccRCC tissue samples. Kaplan-Meier Plotter analyses further confirmed that individual gene expression levels were associated with survival, supporting the biological relevance of each component of the signature.
The study assessed the relationship between the glycolytic risk signature and tumor immune microenvironment using CIBERSORT deconvolution with the LM22 gene signature across 1,000 permutations. Among 21 tumor-infiltrating immune cell (TIIC) types evaluated, 9 were significantly associated with risk group classification.
High-risk patients showed differential infiltration patterns in multiple immune cell types, suggesting that glycolytic reprogramming in ccRCC may co-occur with specific patterns of immune microenvironment composition. This relationship between metabolic phenotype and immune cell infiltration could have implications for immunotherapy response prediction.
The association between glycolysis-related risk score and immune infiltration adds another dimension to the signature's clinical utility. Beyond predicting overall survival, it may help identify patients who would benefit from immune checkpoint therapy based on their tumor's metabolic and immunological profile.
This study establishes a validated 10-gene glycolysis-related prognostic signature for ccRCC that reliably stratifies patients by overall survival risk across multiple independent cohorts. The signature was confirmed as an independent prognostic factor after adjustment for standard clinical variables including tumor stage and grade.
The integration of the signature into a nomogram with clinical factors provides a practical, interpretable decision-support tool for clinicians. The nomogram allows individualized survival probability estimation at 1, 3, and 5 years without the need for complex genomic assays.
Future research should examine whether the glycolytic risk score can guide treatment selection, particularly for immunotherapy, given its association with immune cell infiltration. Prospective validation in diverse clinical cohorts and testing against existing prognostic systems such as IMDC criteria would further establish its clinical utility.