51 Discovery of Exhausted CD8+ T-cell States Associated with Clinical Outcomes in Renal Cell Carcinoma

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Background Therapeutic resistance to immune checkpoint inhibitors (ICIs) and antiangiogenic therapies remains a significant clinical challenge in advanced renal cell carcinoma (RCC). The prognostic value of CD8+ T-cell infiltration—a central player in cancer immunity—is often equivocal across RCC studies, supporting the need for a systematic characterization of response-associated T cell states. Here, we used a machine learning-based approach that integrated bulk RNA-sequencing (RNA-seq) data from multiple ICI-based clinical trials with an in-house single-cell RNA-sequencing (scRNA-seq) dataset to identify CD8+ T-cell states associated with progression-free survival (PFS), providing novel prognostic insights for advanced RCC. Methods We collected bulk RNA-seq data of baseline tumor samples and patient outcomes from four major clinical trials of advanced RCC with at least one treatment arm receiving ICI monotherapy or ICI combined with tyrosine kinase inhibitors (ICI+TKI): HCRN GU16-260, JAVELIN Renal 101, COSMIC-313, and CheckMate 9ER. An in-house scRNA-seq dataset of tumor-infiltrating CD8+ T cells from 70 advanced RCC patients was used as a cellular reference. Our analysis focused on the CD8+ exhausted T cells (Tex), which generally overlap with tumor-specific T-cell population. Using non-negative matrix factorization (NMF), we classified the Tex into four functional states. To link the scRNA-seq-derived T-cell states with the PFS and bulk RNA-seq data, we applied Scissor (Sun et al., 2022), a machine learning model that identifies phenotype-associated cell subsets through cell-sample correlations. For the bulk cohort meta-analysis, shared genes were normalized, scaled, and batch-corrected using ComBat (Johnson et al., 2007). Statistical significance was established via a permutation test of the PFS data, and calculating empirical p-values based on the concordance index (C-index). Results Our final bulk dataset comprised 1,527 RNA-seq tumor samples from four clinical trials of RCC, including 1,004 tumor samples from patients receiving ICI or ICI+TKI regimens. We first applied Scissor to each treatment arm of each trial individually, and found that worse PFS-associated Tex cells exhibited elevated expression of tissue-resident markers, such as ZNF683 and ITGAE, consistently across all ICI and ICI+TKI arms. These markers represent one specific NMF program defined in Tex, which we have termed the resident-memory-like exhausted state (Tex-rm). Notably, the Tex-rm population was enriched among worse PFS-associated cells in all ICI-based treatment arms (ratio of observed to expected [Ro/e] = 1.37-2.33) but not in any TKI monotherapy arms. Subsequently, we combined all ICI-based treatment arms and applied Scissor to the batch-corrected bulk dataset. This again identified Tex-rm as associated with worse PFS. Such a correlation was supported by a permutation test with p-value < 0.001. Conclusions Through machine learning-based integration of bulk and single-cell RNA-seq datasets, we found that baseline tumor infiltration by exhausted T cells with a resident memory phenotype (Tex-rm) is significantly associated with worse PFS in advanced RCC patients receiving ICI monotherapy or ICI+TKI combination therapy. This finding highlights the Tex-rm state as a potential predictive biomarker for ICI efficacy. DOD CDMRP Funding no