Enhancer-driven gene regulatory networks reveal transcription factors governing T cell adaptation and differentiation in the tumor microenvironment.

Tumor infiltrating lymphocytes (TIL) with a CD8+ T cell tissue-resident memory (Trm) phenotype are associated with improved patient outcomes in solid malignancies. To define programs governing the formation of Trm-like TIL, we performed paired single-cell RNA sequencing and single-cell ATAC sequencing of T cell receptor (TCR)-matched CD8+ T cells in models of infection and cancer. Enhancer-driven regulons assembled from multiomic profiling data revealed epigenetic and transcriptional programs regulating the formation of Trm-like TIL in relation to canonical exhausted and memory T cell states. The transcriptional regulator KLF2 repressed the formation of CD69+CD103+ Trm-like TIL and limited anti-tumor activity. Conversely, sustained expression of the transcription factor BATF enhanced formation of CD69+CD103+ TIL, contingent upon downregulation of KLF2. Transforming growth factor β (TGF-β) signaling and CD103 expression were necessary for Trm-like TIL formation, but BATF overexpression was sufficient to drive formation of CD69+CD103+ TIL in TGFBR2-silenced cells. These findings reveal mechanisms of Trm-like TIL differentiation and provide a framework for considering tissue residency in the context of CD8+ T cell heterogeneity in the tumor microenvironment. Antigen specific P14 CD8 T cells were single cell sorted from 11 conditions spanning: naive, acute infection (LCMV Armstrong) at early and late time points from lymphoid and non-lymphoid tissues, chronic infection (LCMV CL13) from early and late time points in lymphoid tissue, and tumor models (B-ALL, PDAC, melanoma, and TNBC). Single cells were sorted for Viable donor (CD45.1 or CD45.1.2) CD8+Va2+ and processed for the 10X Genomics multiomic single-nuclei RNAseq + ATACseq kit (10x Genomics). Libraries were sequenced using either the Illumnia NovaSeq6000 or NextSeq 2000 and processed using CellRanger-Arc (v2.0.0 for NovaSeq6000 samples; or v2.0.2 for NextSeq2000 samples). Downstream analysis of filtered matrix files followed standard Seurat v5 and Signac multiomic GEX+ATAC Weighted Nearest Neighbors (WNN) pipelines.