Development and Validation of a Novel Plasma Metabolomic Signature for the Detection of Renal Cell Carcinoma

Abstract

Background and objective: Early diagnosis is critical for improving survival in renal cell carcinoma (RCC); yet, effective laboratory tests remain lacking. We aimed to characterise metabolic reprogramming in RCC and develop an artificial intelligence (AI)-enabled plasma metabolic model for RCC detection.

Methods: In this multicentre diagnostic model development and validation study, plasma samples from RCC patients and healthy controls (HCs) were collected across five hospitals between December 2019 and October 2023. Eligible patients had pathologically confirmed RCC without prior treatment; HCs were recruited from routine physical examination. Participants with a history of malignancy were excluded. Untargeted plasma metabolomics was conducted to identify candidate metabolites via a support vector machine, further confirmed by a high-resolution targeted metabolic analysis. An AI-aided diagnostic model, Renal Cell Carcinoma Artificial Intelligence Detector (RCAID), was developed using selected metabolites and validated in six independent validation cohorts. Multiomic analyses were performed to elucidate the underlying metabolic mechanisms.

Key findings and limitations: The study enrolled 1680 participants, comprising 920 RCC patients and 760 HCs. Among RCC cases, 744 (81%) had clear cell RCC and 633 (69%) had stage I disease. Seven key plasma metabolites, including 2-hydroxyphenylacetic acid, azelaic acid, N,N-dimethylglycine, N-acetyl-L-aspartic acid, N-epsilon-acetyl-L-lysine, proline, and (Z,Z)-4-oxo-2,5-hetpadienedioic acid, were identified and used to develop the RCAID model, which demonstrated an area under the receiver operating characteristic curve (AUROC) of 0.988 in the training cohort (n = 503). The model exhibited excellent diagnostic performance, with AUROCs of 0.977, 0.911, 0.945, and 0.972 in the internal (n = 202), external (n = 158), multicentre (n = 346), and temporal (n = 123) validation cohorts, respectively. Additionally, RCAID achieved an AUROC of 0.940 in the late-stage RCC (n = 179) and 0.932 in the non-clear cell RCC (n = 169) validation cohorts. Multiomic analyses further revealed six RCAID-associated dysregulated metabolic pathways in RCC.

Conclusions and clinical implications: This study identified metabolic alterations in RCC and developed a promising AI-based plasma metabolic model with potential clinical application for RCC diagnosis.

Keywords: Artificial intelligence; Early diagnosis; Machine learning; Mass spectrometry; Metabolomics; Plasma biomarker; Renal cell carcinoma.