This study developed a new tool called the Glycosyltransferase Risk Score (GTRS) to predict the prognosis of patients with clear cell renal cell carcinoma (ccRCC), the most common type of kidney cancer. The GTRS is based on a set of 13 genes that control how sugar molecules are added to proteins and lipids, a process called glycosylation.
Glycosylation plays a crucial but often overlooked role in cancer. When cells add sugars to proteins in abnormal ways, it can change how proteins function, how cancer cells interact with the immune system, and how aggressively the tumor grows. The researchers identified a specific set of glycosylation-related genes that, together, can reliably predict whether a kidney cancer patient is at high or low risk of poor outcomes.
Published in the International Journal of Molecular Sciences in 2025, this study used data from three independent patient cohorts and 117 different machine learning algorithm combinations to ensure the final model is robust and reproducible.
Glycosyltransferases are enzymes that attach sugar molecules to proteins and fats. These sugar additions, called glycosylation, affect virtually every aspect of protein function, including how proteins fold, how stable they are, how they interact with other molecules, and how cells recognize each other.
In cancer, abnormal glycosylation patterns are common and can have profound consequences. For example, altered glycosylation on tumor cells can help them evade the immune system by disguising surface proteins that would otherwise be recognized as foreign. Glycosylation changes can also promote tumor invasion and the formation of new blood vessels that feed tumor growth.
Despite their importance, glycosylation-related genes have received relatively little attention in kidney cancer research. The human genome contains 337 known glycosyltransferase genes, and this study systematically evaluated which of them are altered in ccRCC and which have the strongest relationship to patient survival.
The researchers began with all 337 glycosyltransferase genes and narrowed them down systematically. First, they identified 55 genes that were differentially expressed in ccRCC tumors compared to normal kidney tissue. From these, 16 genes were found to be individually associated with patient prognosis.
To find the best combination of these 16 genes for a prognostic model, the team evaluated 117 different algorithm combinations using 8 machine learning methods. The approach that performed best across all three validation datasets was StepCox[forward] plus Ridge regression, ultimately selecting 13 genes for the final GTRS model. The C-index of this model was 0.753, indicating good ability to rank patients by their actual survival outcomes.
The model was validated on three independent datasets: the TCGA cohort (437 patients), the CPTAC cohort (103 patients), and the EMTAB1980 cohort (101 patients). Achieving an AUC above 0.75 across all three cohorts confirms that the model generalizes beyond the training data and is not simply overfitted to one dataset.
Among the 13 genes in the GTRS, two stand out as particularly important. TYMP (Thymidine Phosphorylase) is an oncogene in this context, meaning it promotes cancer behavior. Laboratory experiments confirmed that TYMP promotes tumor cell proliferation (uncontrolled growth) and invasion (spread into surrounding tissue), and patients with high TYMP expression have a worse prognosis.
GCNT4 (Glucosaminyl N-Acetyl Transferase 4) acts as a tumor suppressor. Experiments showed that GCNT4 inhibits cell proliferation and migration, and patients with high GCNT4 expression tend to have better outcomes. The contrasting roles of these two genes illustrate how glycosylation can influence cancer behavior in both pro-tumor and anti-tumor directions.
Understanding the specific functions of TYMP and GCNT4 also opens potential therapeutic avenues. If TYMP is driving tumor growth, drugs that inhibit TYMP activity might slow the cancer. If GCNT4 is suppressed in aggressive tumors, finding ways to restore its expression could potentially be beneficial. These ideas require further research but represent interesting directions for future investigation.
Beyond survival prediction, the study examined what the tumor microenvironment looks like in patients classified as high-risk by the GTRS. High-risk patients showed a more immunosuppressive tumor environment, with higher levels of regulatory T cells (Tregs), M0 macrophages (an immature, undifferentiated immune cell type), and myeloid-derived suppressor cells (MDSCs). All of these cell types are associated with suppressing anti-cancer immune responses.
High-risk patients also showed elevated expression of immune checkpoint molecules including PDCD1 (PD-1), CTLA4, and LAG3. These molecules act as brakes on the immune system and are the targets of checkpoint inhibitor drugs used in kidney cancer treatment, such as nivolumab and ipilimumab.
High-risk patients also had higher tumor mutational burden (TMB), meaning their tumors had accumulated more mutations. High TMB is sometimes associated with better response to immunotherapy, but in the context of an already immunosuppressive microenvironment, more mutations simply means more genomic instability driving aggressive behavior. The combination of high TMB and immunosuppressive cells is characteristic of particularly difficult-to-treat tumors.
The GTRS is designed to be calculated from standard gene expression data, which can be obtained from tumor tissue using RNA sequencing. As RNA sequencing becomes more common in clinical settings, tools like the GTRS could be incorporated into routine genomic profiling of kidney cancer tumors, alongside other molecular tests already used to guide treatment.
A high GTRS score could prompt oncologists to consider more aggressive initial treatment or closer monitoring for early signs of recurrence. The associated immune checkpoint marker findings suggest that high-risk patients might be particularly good candidates for immunotherapy with checkpoint inhibitors, although this hypothesis would need to be tested in dedicated clinical trials.
The validation of GTRS across three independent cohorts from different countries and institutions is an important foundation for clinical translation. However, prospective studies, meaning following patients forward in time while applying the GTRS at diagnosis, will be needed to formally establish whether using the score improves clinical decision-making and patient outcomes.