Metabolic heterogeneity and survival outcomes in papillary renal cell carcinoma: insights from multi-datasets and machine learning analyses.

Abstract

Background: Renal cell carcinoma is characterized by immune and metabolic alterations. These metabolic reprogramming processes enhance tumor cell proliferation and infiltration. The purpose of this study was to investigate the characteristics of metabolism-related molecules and to identify potential prognostic biomarkers in kidney renal papillary renal cell carcinoma (KIRP).

Methods: We conducted a comprehensive analysis of metabolism-related genes using weighted gene co-expression network analysis and differential expression analysis. Subsequently, we constructed a metabolism-related signature (MRS) by integrating 90 machine learning algorithms. Based on Cox regression analyses, we developed a predictive nomogram. Functional enrichment analysis, genomic variant analysis, chemotherapy response evaluation, and immune cell infiltration profiling were then performed among the MRS subtypes. Finally, the MRS was further examined at the single-cell level, and quantitative PCR and immunohistochemical staining were conducted to validate the key genes.

Results: We identified 16 differentially expressed metabolic genes. The random survival forest (RSF) emerged as the optimal machine learning model in the TCGA-KIRP and GSE2748 cohorts. The MRS demonstrated robust predictive performance, with an AUC of 0.989 for 5-year survival predictions. The risk score was significantly correlated with T stage and pathological stage and was identified as an independent prognostic factor. Patients in the high-risk group exhibited higher tumor mutation burdens and derived greater benefits from sunitinib, pazopanib, lenvatinib, and temsirolimus. A four-genes nomogram was then constructed to predict overall survival. PYCR1, INMT, and KIF20A were highly expressed in KIRP according to scRNA-seq analysis and were validated in vitro.

Conclusion: This study revealed the heterogeneity of metabolic molecules in KIRP and established a prognostic machine learning model that enhances risk stratification and may optimize chemotherapy strategies in the management of KIRP.