Metabolomic Profiles of Metastatic Renal Cell Carcinoma

Poster #: 28
Session/Time: A
Author: Liam Spiers
Mentor: Dean Troyer, MD
Co-Investigator(s): 1. Aditya Chakraborty, Leroy T. Canoles Jr. Cancer Research Center, 2. Brian Main, Department of Biomedical & Translational Sciences, Leroy T. Canoles Jr. Cancer Research Center 3. Liang Li, Chemistry, University of Alberta 4. Ya-Chun Chan, Chemistry, University of Alberta 5. Xiaohang Wang, Chemistry, University of Alberta 6. Nico Verbeeck, Aspect Analytics, Genk, Belgium 7. Mike Williams, MD, Department of Urology, Urology of Virginia
Research Type: Clinical Research

Abstract

Introduction: Patients who are status-post radical or partial nephrectomy to treat renal cell carcinoma (RCC) undergo clinical surveillance regimes to investigate the presence of recurrent or metastatic disease. The clinical problem we're addressing is the frequency and intensity of surveillance after surgical resection of RCC. The aim of this study was to identify biomarkers which were statistically demonstrated to predict the development of metastatic disease. Identification of such biomarkers would create a diagnostic tool to aid clinicians in the management of patients on RCC surveillance.

Methods: We conducted a retrospective case control study of RCC patients, treated with surgical resection. Tumor samples were classified based on patients who either did or did not develop metastatic disease within 3 years. Freshly obtained RCC samples were collected, frozen, processed, and sent for metabolomic analysis using liquid chromatography-mass spectrometry (LC-MS). Three different types of chemical groups were analyzed for each sample: Carboxyl, Amine-Phenol, and Hydroxyl. Metabolites were identified using a chemical library for crossreference. LC-MS results were then analyzed with machine learning approaches and statistical models. Unsupervised machine learning (UMAP, Non-negative Matrix Factorization) was used for exploratory analysis of the data. Supervised machine learning, using both linear and nonlinear Support Vector Machines was used to construct classification models to distinguish between cases and controls, showing good classification performance with an area under the Precision-Recall curve (AUPRC) of 0.92 for the best-performing model. Repeated crossvalidation was used to avoid overfitting and gain representative classification performance of the model.

Results: Statistical analysis yielded metabolites which were predominantly involved in differentiating metastatic from control samples. Definitively identified metabolites included the following: 4-Hydroxybutanoic acid, 3-Aminosalicylic acid, Indoxyl sulfate, Norcotinine, 3,4Dihydroxyphenylpropanoic acid, Glutamyl-Glutamic acid, and N-ethylglycine.

Conclusion: This study identifies certain metabolites which predict RCC metastasis. Metabolomics is a promising method of biochemical analysis for predicting outcomes such as RCC metastasis. Directions of future studies would include samples obtained from multiple centers to increase the sample size of identified metabolites. More research is necessary to strengthen support for using this diagnostic tool with certainty.