# Angiogenesis-Informed Preoperative CT Radiogenomics Predicts Overall Survival in Clear Cell Renal Cell Carcinoma: Development and External Validation

## TL;DR

This study combines CT scans and genetic data to predict survival in kidney cancer patients and identifies new treatment targets.

## Contribution

A novel radiogenomics model integrating imaging and transcriptomic data to predict survival and identify anti-angiogenic targets in ccRCC.

## Key findings

- An XGBoost-based radiogenomics model achieved high accuracy in predicting overall survival in ccRCC patients.
- PDLIM1 protein expression was elevated in ccRCC tissues and suppressed endothelial tube formation.
- Five angiogenesis-related genes were identified as key prognostic markers for ccRCC.

## Abstract

Clear cell renal cell carcinoma (ccRCC) is a highly vascular tumor, and angiogenesis plays a central role in its progression and response to therapy. However, reliable tools that combine imaging and molecular information to predict patient survival and reveal actionable targets are still needed. In this study, we integrated bulk and single-cell transcriptomic datasets to identify angiogenesis-related genes associated with overall survival, and we linked these risk patterns to radiomics features extracted from preoperative contrast-enhanced CT scans. Using multiple machine learning approaches, we developed a radiogenomics model and found that an XGBoost-based prediction model achieved the best and consistent performance across internal and external validation cohorts. We further confirmed elevated PDLIM1 protein expression in ccRCC tissues and demonstrated that PDLIM1 can suppress endothelial tube formation in functional assays. Overall, our angiogenesis-related radiogenomics model enables effective risk stratification for ccRCC patients and highlights potential therapeutic targets for anti-angiogenic treatment.

Background/Objectives: We aimed to identify angiogenesis-related prognostic biomarkers and develop a radiogenomics model to predict overall survival (OS) in clear cell renal cell carcinoma (ccRCC), supporting risk stratification and potential therapeutic target discovery. Methods: Bulk transcriptomes from The Cancer Genome Atlas Kidney Renal Clear Cell Carcinoma cohort (TCGA-KIRC), seven Gene Expression Omnibus (GEO) microarrays, and a single-cell RNA sequencing (scRNA-seq) dataset were integrated to identify angiogenesis-related prognostic genes. Preoperative contrast-enhanced computed tomography (CT) images from The Cancer Imaging Archive Kidney Renal Clear Cell Carcinoma collection (TCIA-KIRC) were used for radiomics feature extraction, and a radiogenomics signature was constructed by linking radiomic features with transcriptomic risk patterns. Nine machine learning models were trained to predict OS; the best model was further evaluated in an independent external retrospective cohort. PDLIM1 (PDZ and LIM domain protein 1) was validated at the protein level, and conditioned medium from stable ccRCC cell lines was applied to human umbilical vein endothelial cells (HUVECs) for Matrigel tube formation assays. Results: Five angiogenesis-related hub genes (PDLIM1, EMCN, ARPC1B, PLAT, and TIMP1) were identified. The extreme gradient boosting (XGBoost)-based radiogenomics model showed the best performance, with time-dependent concordance index (C-index) values of 0.880, 0.816, and 0.789 at 1, 3, and 5 years in the training set and 0.864, 0.758, and 0.736 in the internal validation set, respectively. In the external validation cohort, C-index values were 0.800, 0.726, and 0.703 at 1, 3, and 5 years. PDLIM1 protein was upregulated in ccRCC versus normal tissues. Functionally, PDLIM1 overexpression suppressed, whereas PDLIM1 knockdown promoted tube formation. Conclusions: This study developed and validated an angiogenesis-related radiogenomics model that accurately predicts OS in ccRCC patients and provides potential therapeutic targets for anti-angiogenic therapy.