Integrated multi-Omic profiling reveals macrophage-driven prognostic signatures in clear cell renal cell carcinoma through machine learning optimization

Wei Cao¹Wenyuan Zhuang^1,2Kai Xu^1*

• ¹The Third Affiliated Hospital of Nanjing Medical University, Changzhou, China
• ²The Affiliated Changzhou Second People's Hospital of Nanjing Medical University, changzhou, China

Objective: Clear cell renal cell carcinoma (KIRC) is the most prevalent and aggressive form of kidney cancer, with limited survival despite advances in combination immunotherapy. Tumor-associated macrophages (TAMs) critically shape the tumor microenvironment (TME) and influence treatment resistance. We aimed to delineate TAM heterogeneity, identify prognostic macrophage signatures, and characterize the immune-metabolic programs underpinning KIRC progression. Methods: We integrated single-cell RNA sequencing (scRNA-seq) data from ten KIRC tumors with high-dimensional weighted gene co-expression network analysis (hdWGCNA) and twenty machine-learning models. Five macrophage subpopulations were defined by canonical markers and validated spatially. A macrophage-centric prognostic signature was trained using a random survival forest model and validated in another independent cohort. We further interrogated mutational landscapes, immune-stromal infiltration (xCell), pathway activation (ssGSEA), and clinical correlations. Results: scRNA-seq identified five transcriptionally distinct macrophage (Mac) subsets, including three lipid-associated Mac populations (LA-Mac: ALOX5AP+LA-Mac, HERPUD1+LA-Mac, and PRDX1+LA-Mac), an FCN1+inflammatory Mac subset (FCN1+Inflam-Mac), and an oxidative phosphorylation-enriched subset (OxP-Mac), distinct from canonical M1/M2 signatures. hdWGCNA revealed ten co-expression modules, among which the Mac-M2 module demonstrated the highest macrophage specificity and was preferentially enriched in tumor tissues. Based on seven hub genes from the Mac-M2 module, RSF model was constructed, achieving robust prognostic performance and effectively stratifying patients into high-and low-risk groups (log-rank p < 0.0001) in both the TCGA and CPTAC cohorts. Deconvolution analysis and gene set scoring of macrophage subtypes consistently identified PRDX1+LA-Mac as the predominant and prognostically relevant pathological subtype enriched in high-risk patients across both TCGA and CPTAC cohorts. Moreover, high-risk patients in TCGA exhibited elevated tumor mutation burden, increased pro-inflammatory M1 macrophages, Th2 polarization, metabolic dysregulation, and enhanced EMT signatures, all correlating with poorer survival. Conclusion: This multi-omics study illuminates the transcriptional and functional heterogeneity of TAMs in KIRC and establishes a macrophage-derived prognostic signature with translational potential. Our findings underscore the dual roles of macrophage polarization in mediating immune suppression and metabolic adaptation, offering novel targets for clinical diagnosis and treatment of KIRC.