A minimal gene set characterizes TIL specific for diverse tumor antigens across different cancer types

Abstract

Identifying tumor-specific T cell clones that mediate immunotherapy responses remains challenging. Mutation-associated neoantigen (MANA) -specific CD8+ tumor-infiltrating lymphocytes (TIL) have been shown to express high levels of CXCL13 and CD39 (ENTPD1), and low IL-7 receptor (IL7R) levels in many cancer types, but their collective relevance to T cell functionality has not been established. Here we present an integrative tool to identify MANA-specific TIL using weighted expression levels of these three genes in lung cancer and melanoma single-cell RNAseq datasets. Our three-gene “MANAscore” algorithm outperforms other RNAseq-based algorithms in identifying validated neoantigen-specific CD8+ clones, and accurately identifies TILs that recognize other classes of tumor antigens, including cancer testis antigens, endogenous retroviruses and viral oncogenes. Most of these TIL are characterized by a tissue resident memory gene expression program. Putative tumor-reactive cells (pTRC) identified via MANAscore in anti-PD-1-treated lung tumors had higher expression of checkpoint and cytotoxicity-related genes relative to putative non-tumor-reactive cells. pTRC in pathologically responding tumors showed distinguished gene expression patterns and trajectories. Collectively, we show that MANAscore is a robust tool that can greatly enrich candidate tumor-specific T cells and be used to understand the functional programming of tumor-reactive TIL.

Although individual genes that distinguish tumor-reactive CD8+ T cells from bystander T cells in tumors have been described, a functionally meaningful integrative signature has not been established. Here authors show that mutation-associated neoantigen-specific CD8+ tumor-infiltrating lymphocytes can be recognized by MANAscore, an algorithm that uses weighted expression levels of CXCL13, ENTPD1 and IL7R in single-cell RNAseq datasets of lung cancer and melanoma patients as input.