The CD8⁺ T cell landscape of human brain metastases

Abstract

Despite improved outcomes with checkpoint blockade immunotherapy, patients with brain metastases have the worst prognosis among patients with metastatic cancer. Immune checkpoint blockade agents target inhibitory receptors, such as PD-1, on exhausted CD8+ T cells to restore their anti-cancer function. Many patients, however, either do not respond or progress after an initial response to immune checkpoint blockade, and distant intracranial failure is common despite excellent options for local treatment of brain metastasis. To develop more effective therapeutic strategies for the treatment of brain metastases, an understanding of the phenotype of brain metastasis-infiltrating CD8+ T cells is essential. Here we performed a detailed characterization of the CD8+ T cells contained in brain metastases. Brain metastases were densely infiltrated by CD8+ T cells; blood contamination of tumor samples was rare. Compared to patient-matched circulating cells, brain metastasis-infiltrating CD8+ T cells had a distinct phenotype characterized by more frequent expression of PD-1, with subpopulations defined by expression of additional co-inhibitory molecules and the residence marker CD69. Single cell RNA-sequencing identified four phenotypic subpopulations within brain metastasis-infiltrating PD-1+ CD8+ T cells. Two of these populations — a terminally-differentiated and a dividing population — were characterized by high expression of co-inhibitory molecules and lacked expression of progenitor markers such as TCF-1. There was significant T cell receptor (TCR) overlap between the terminally-differentiated and dividing populations, suggesting that the dividing cells give rise to the terminally-differentiated cells. There was minimal TCR overlap between these two populations and other brain metastasis-infiltrating PD-1+ CD8+ T cells. T cell clones from brain metastasis-infiltrating CD8+ T cells were rare in circulation, particularly clones from the terminally-differentiated and dividing populations. We systematically identified bystander CD8+ T cells specific for microbial antigens; these cells infiltrated brain metastases and expressed genes shared with exhausted progenitor CD8+ T cells, such as TCF7 and IL7R. We performed spatial transcriptomics on brain metastases and used a novel method to obtain TCR sequences from spatial transcriptomics data. These data revealed distinct niches within the TME defined by their gene expression patterns and cytokine profiles. Terminally-differentiated CD8+ T cells preferentially occupied niches within the tumor parenchyma. Together, our results show that antigen-specificity restricts the spatial localization, phenotypic states, and differentiation pathways available to CD8+ T cells within the brain metastasis TME.

Analysis of RNA processing directly from spatial transcriptomics data reveals previously unknown regulation

“…To show that we could identify spatial patterns in tissues without a priori anatomy annotations, we used this same methodology to identify spatial patterns in human tumor data (Sudmeier et al, 2021). We identified the calcium-binding protein S100A9 as the 14th-most-significant ReadZS spatial pattern by Moran's I in a brain metastasis from breast cancer carcinoma.…”

“…Human tumor data was downloaded from (Sudmeier et al, 2021) from GEO accession GSE179373. Data from p20218-s002_L2 (patient 27, lung adenocarcinoma) and p20218-s001_L1 (patient 26, breast carcinoma) was used in this study.…”

“…This includes two sections of the sagittal anterior mouse brain (V1_Mouse_Brain_Sagittal_Anterior (sagittal-anterior mouse brain) and V1_Mouse_Brain_Sagittal_Anterior_Section_2 (sagittal-anterior mouse brain (Section 2))), two sections of the sagittal posterior mouse brain (V1_Mouse_Brain_Sagittal_Posterior (sagittal-posterior mouse brain) and V1_Mouse_Brain_Sagittal_Posterior_Section_2 (sagittal-posterior mouse brain (Section 2))), and one section of the mouse kidney (V1_Mouse_Kidney (mouse kidney)). A bash script to download all of the mouse data is available here: https://github.com/juliaolivieri/visium_analysis/blob/main/bash_scripts/download_mouse_data.sh Human tumor data was downloaded from (Sudmeier et al, 2021) from GEO accession GSE179373. Data from p20218-s002_L2 (patient 27, lung adenocarcinoma) and p20218-s001_L1 (patient 26, breast carcinoma) was used in this study.…”