Enrichment of glioblastoma tumor microenvironment in a GZMK+ T cell population.

2065 Background: Glioblastoma (GBM) bears a survival estimate below 10% at 5 years, despite standard chemoradiation treatment. Trials evaluating immune checkpoint blockade (ICB) in gliomas have been disappointing so far, dealing with the unique immunosuppressive tumor microenvironment (TME) of the brain. Nevertheless, we have clinical evidence of ICB efficacy for brain metastases (BrM) of immunosensitive tumors. Even if both GBMs and BrM share the same organ-specific microenvironment, it is clear that disease-specific immune characteristics instruct differential response to ICB. Methods: By using advanced deep learning algorithms to combine public and in-house-generated single-cell RNA sequencing profiles, we analyzed immune infiltrate characteristics of 602949 CD45+ cells from 76 GBM, 30 BrM from different types of extracranial tumors (13 NSCLC, 12 melanoma, 2 ovarian, 1 breast, 1 colorectal and 1 renal carcinoma) and 3 normal brain samples. The analysis was restricted to newly diagnosed GBM, excluding recurrent and pre-treated cases. CD8+ cells underwent segregation into clusters based on log2 normalized expression levels of CD3E and CD8A (with expression values greater than 0). The data integration utilized DESC, a deep learning model founded on autoencoders, and subsequent examination was conducted within the Scanpy framework. Cluster analysis was executed employing the Leiden algorithm. Differentially expressed genes (DEGs) were pinpointed across clusters using the Wilcoxon test, with False Discovery Rate (FDR) correction carried out through the Benjamini-Hochberg procedure. Results: From CD45+CD3+ cells, we isolated CD8+ T cells and identified 7 informative CD8+ T cells clusters. Based on DEGs, we could appreciate T cells clusters that were also recapitulated in other extracranial malignancies. The most abundant subset was composed of GZMK+ effector T cells (Cluster 0: GZMK, GZMM, CD44, KLRG1, IL7R, GZMA, CXCR3), followed by putative tumor-specific tissue-resident memory cells (Cluster 1: GZMB, LAG3, CTLA4, TIGIT, CXCL13, ITGAE, ENTPD1, PDCD1, CD27, HLA-DRB1), stem-cell memory-like T cells (Cluster 3: IL7R, CCR7, JUNB, KLF2, TCF7) and a less represented cluster characterized by high expression of genes belonging to the HSP family (Cluster 4: HSPA1, HSPH1, HSPE1, HSPD1, IFN). Interestingly, while the majority of the identified clusters was equally represented in the 3 different tissues, the GZMK+ effector T cells cluster (Cluster 0) was found to be more represented in GBM compared to BrM and normal brain tissue. Conclusions: A GZMK+ effector T cells subpopulation, specifically enriched in GBM, has been poorly characterized in its function and spatial localization. Our data highlight the importance of studying the composition of the immune infiltrate in GBM with multi-omics approaches in order to uncover deep mechanisms of resistance to ICB and eventually provide hints on potential therapeutic targets.