4529 Background: There is a major unmet need for novel biomarkers in advanced renal cell carcinoma (RCC). A promising candidate biomarker is the classification of advanced RCC into 7 distinct molecular subtypes identified in an exploratory analysis of the IMmotion151 trial (Motzer et al., Cancer Cell 2020). These transcriptomic clusters were predictive of treatment outcomes with atezolizumab + bevacizumab vs sunitinib. However, a recent validation study reported that while these clusters were able to be replicated in an external cohort, they failed to stratify survival between avelumab + axitinib vs sunitinib (Saliby et al., Cancer Cell 2024). Given these mixed results, there is a need for further validation of the IMmotion151 molecular clusters. We used a multi-institutional cohort from the Oncology Research Information Exchange Network (ORIEN), an alliance of cancer centers, to replicate these clusters and assess if they were prognostic for 5-year overall survival (OS) in patients with advanced RCC. Methods: Patients with RCC metastatic to lymph nodes or a distant site were identified in the ORIEN database. Included patients had mRNA sequencing data available from a metastatic tumor sample. We used methods similar to those used in the IMmotion151 exploratory analysis to attempt to replicate the molecular clusters. Using an unsupervised clustering algorithm called non-negative matrix factorization (NMF) on the top 10% most variable genes in each sample, we derived 7 distinct molecular clusters. To compare our clusters with those identified in IMmotion151, we constructed a heatmap of expression (using z-scores) for pre-specified gene sets across each cluster. We then conducted Kaplan-Meier analysis to assess if our clusters were able to stratify 5-year OS in our patient cohort for a significance level of p ≤ 0.05. OS was measured from the time of diagnosis with metastatic RCC. Results: 155 eligible patients were included in our analysis. We successfully replicated clusters 2 (angiogenic), 3 (complement/Ω-oxidation), 4 (T-effector/proliferative), and 6 (stromal/proliferative). Cluster 7 (snoRNA) was not possible to replicate as our transcriptomic data only included mRNA. Kaplan-Meier analysis did not show any statistically significant difference (p = 0.85) in 5-year OS between our 7 clusters. Cluster 2 had a median OS of 47.5 months, cluster 3 had a median OS of 43.6 months, median OS was not reached for cluster 6, and there were insufficient patients in cluster 4 to be included in the survival analysis. Conclusions: While our results show that the IMmotion151 molecular subtypes are robust and replicable even in smaller datasets, the lack of any association between subtypes and survival outcomes casts further doubt on the clinical utility of these biomarkers. Our results may also be influenced by our smaller sample size compared to the IMmotion151 analysis (n = 823).
Prakash et al. (2026) studied this question.