Co-expression network-based analysis of gene programs contributing to immune checkpoint inhibitor (ICI) resistance in renal cell carcinoma (RCC).

Abstract

4530 Background: There is an unmet need to characterize molecular drivers of resistance to ICIs in the RCC tumor microenvironment, especially in cell-type-specific contexts. To address this, we identified coherent gene programs in each cell type using scRNA-seq data to uncover an immune cell phenotype associated with ICI resistance. Methods: We performed per-cell-type weighted coexpression network analysis on scRNA-seq data of 443,337 cells from 70 RCC tumor samples, mostly from patients (pts) who received ICI (S. Kashima, ASCO, 2024), to construct mutually co-expressed gene sets (modules) and their cell-level expression. We compared module expression scores between responders (R, complete or partial response) and non-responders (NR, progressive disease) with the Wilcoxon rank-sum test plus FDR correction, and ran tests with permuted labels to rule out statistical artifacts. For validation, we performed this analysis on another scRNA-seq cohort from a phase II trial (HCRN GU16-260; NCT03117309), and further, we used CoxPH survival analysis on module signature scores computed by GSVA to assess the prognostic role of module expression in bulk RNA-seq samples from the IMmotion 150, CheckMate 009/010/025, and Javelin 101 trials, after stratifying them into immune-high (top 50%) and immune-low (bottom 50%) groups according to CIBERSORTx-inferred total immune infiltration. Results: Analysis of scRNA-seq (two independent datasets) from ICI-treated RCC tumors revealed a module of robustly co-expressed ribosomal and translation-associated genes that was significantly upregulated in NR in immune cells (but not tumor or stromal cells), including macrophages, CD4+ and CD8+ T, NK, and B cells (p-values ranging from 1e-20 to 1e-200). Analysis of an independent scRNA-seq cohort (HCRN GU16-260) also yielded a module of mutually co-expressed ribosomal proteins only in immune cells that was upregulated in NR (immune p < 1e-20, non-immune p > .05). Finally, validation in large-scale bulk RNA-seq data from clinical trials shows that for patients receiving any ICI-based therapy (nivolumab, avelumab + axitinib, or atezolizumab +/- bevacizumab), module signature scores were significantly associated with worse PFS only in the CoxPH analysis of immune-high samples (HR = 1.65, 95% CI, 1.18–2.3, p < .005), while immune-low samples showed no effect (HR = 0.93, 95% CI 0.71–1.25, p > .5), concordant with findings from the scRNA-seq analysis. Those receiving a TKI only (sunitinib) exhibited no association, regardless of immune infiltration. Conclusions: Though individual ribosomal proteins have been found to be prognostic for RCC, a cell-type-specific module-level analysis elucidated a link between a coherent translation program in immune cells and resistance and poor PFS for pts receiving ICI specifically, contributing to a model of ICI-resistant molecular phenotypes in RCC.