Introduction:

Clear-cell Renal-cell carcinoma (ccRCC) is a highly metabolic disease with multiple mutated genes fundamentally involved in the regulation of cellular metabolic processes. A dysregulation of metabolic pathways involved in oxygen, energy and/or nutrient sensing plays a key feature in RCC carcinogenesis. Previous studies have demonstrated that the tumor-immune microenvironment modulates therapeutic responses and significantly influences clinical outcome. Metabolic profiles could shape tumor microenvironment by determining nutrient sufficiency and engagement of metabolic pathways, thus influencing the activity of immune cells and final treatment response. Identifying key metabolites in ccRCC associated with improved response to immunotherapeutic or systemic therapy provides a unique opportunity for the development of more effective therapeutic strategies. With the use of T-MIRTH (Transcriptomics-Metabolite Imputation via Rank-Transformation and Harmonization), we imputed levels of metabolites from publicly available RNA sequencing data in 7 advanced ccRCC clinical trials. Multiple metabolites were found significantly associate with drug response and patients’ survival outcome.

Methods:

We developed T-MIRTH which could impute metabolite abundances from RNA sequencing data by modeling metabolite-RNA covariation across datasets with paired metabolomics and transcriptomics data. To evaluate T-MIRTH’s performance, we used three ccRCC datasets and compared the imputed metabolite values by T-MIRTH with the true values by calculating Spearman's correlation coefficient rho. Metabolites that are significantly predicted and have spearman’s rho > 0.3 in at least 2 datasets are defined as reproducibly well-predicted.

For 262 reproducibly well-predicted metabolites, we then tested the association between imputed levels of individual metabolites by T-MIRTH and overall survival (OS) or progression free survival (PFS) in 7 published clinical trials of immunotherapeutic vs systemic agents in advanced ccRCC. Multivariate Cox proportional-hazards model (adjusted by age and sex) was performed on immuno-therapy and sunitinib arms separately. Regression results across multiple clinical trials were then aggregated by the random effects meta-analysis model.

Results:

With the use of 3 ccRCC datasets with paired metabolomics and transcriptomics data, we demonstrated T-MIRTH accurately impute metabolite levels by RNA sequencing data. 65% metabolites in RC18 were significantly predicted and positively correlated with their ground truth values. Combining imputation results in 3 datasets, we found 262 reproducibly well-predicted metabolites. To study the association between metabolic profiles and clinical outcomes, we imputed levels of metabolites from RNA-seq data in 7 advanced ccRCC clinical trials. Meta-analysis results across trials show that there are 19 metabolites significantly associated with better survival in immunotherapy arms (p <0.05, HR <1). In sunitinib arms, 29 metabolites are significantly associated with better survival, while 2 metabolites (Beta-alanine, guanosine 3'−monophosphate) are significantly predictive of poorer survival. Prognostic metabolites are enriched in amino acids and lipids. Metabolic pathway enrichment analysis shows that prognostic metabolites are enriched in Urea cycle (arginine and proline metabolism) and Lysine metabolism.

Conclusion:

As T-MIRTH imputed metabolomic profiles predict response to therapy, these prognostic biomarker metabolites could also be potential drug targets in combination therapies to enhance the drug response of immune-therapy drugs or sunitinib in advanced ccRCC. Feasible ways of targeting these metabolites include stimulating or inhibiting upstream enzymes of them. The development of drug combinations that can simultaneously target tumor cells, immune cells and the metabolites in the microenvironment may represent a solution to increase therapeutic response.

Funding: N/A