Expression profiling of single cells and patient cohorts identifies multiple immunosuppressive pathways and an altered NK cell phenotype in glioblastoma

Abstract

Glioblastoma (GBM) is an aggressive cancer with a very poor prognosis. Generally viewed as weakly immunogenic, GBM responds poorly to current immunotherapies. To understand this problem more clearly we used a combination of natural killer (NK) cell functional assays together with gene and protein expression profiling to define the NK cell response to GBM and explore immunosuppression in the GBM microenvironment. In addition, we used transcriptome data from patient cohorts to classify GBM according to immunological profiles. We show that glioma stem-like cells, a source of post-treatment tumour recurrence, express multiple immunomodulatory cell surface molecules and are targeted in preference to normal neural progenitor cells by natural killer (NK) cells ex vivo. In contrast, GBM-infiltrating NK cells express reduced levels of activation receptors within the tumour microenvironment, with hallmarks of transforming growth factor (TGF)-β-mediated inhibition. This NK cell inhibition is accompanied by expression of multiple immune checkpoint molecules on T cells. Single-cell transcriptomics demonstrated that both tumour and haematopoietic-derived cells in GBM express multiple, diverse mediators of immune evasion. Despite this, immunome analysis across a patient cohort identifies a spectrum of immunological activity in GBM, with active immunity marked by co-expression of immune effector molecules and feedback inhibitory mechanisms. Our data show that GBM is recognized by the immune system but that anti-tumour immunity is restrained by multiple immunosuppressive pathways, some of which operate in the healthy brain. The presence of immune activity in a subset of patients suggests that these patients will more probably benefit from combination immunotherapies directed against multiple immunosuppressive pathways.

Keywords: NK cells; glioblastoma; immune-inhibition.

Figure 1

Natural killer (NK) cell‐mediated killing of glioma stem‐like cells. (a) NK cell cytotoxicity: cell tracker violet‐labelled glioblastoma (GB) stem‐like cell (GSC) lines (targets) were co‐cultured with unlabelled, interleukin (IL)‐15 activated NK cells (effectors) for 5 h at effector : target (E : T) ratios as shown. Co‐cultures were then stained with a live/dead discriminator. The panel on the left shows identification of effector and target cells in the co‐culture (for gating purposes) and the panels to the right show death of the labelled target cells at the different E : T ratios. The zero hour control is included as background cell death of the GSC cells. The three graphs summarize data obtained using three GSC lines (GBM1, GBM4 and GBM20) and three different NK cell donors (coloured dots), with standard deviation from the mean. (b) NK cell specificity: cytotoxicity assays of IL‐15‐activated NK cells co‐cultured with a 1 : 1 mix of the GSC line (indicated) and neural progenitor cells (NP). The GSC and NP lines were labelled with different cell tracker dyes, allowing their fate in the assay to be determined separately. The flow cytometry plots show the percentage of dead GSC (here GBM4) and NP cells after zero and 5 h co‐culture with NK cells. The graphs summarize these data for assays containing the three GSC lines using NK cells from four separate donors (coloured dots), with standard deviation from the mean.

Figure 2

The cell surface phenotype of glioblastoma (GBM)‐infiltrating lymphocytes. (a) Expression of programmed cell death 1 (PD‐1) on CD3⁺ T cells in GBM patient tumour (GBM), patient blood (PB) and control blood from healthy donors (CB). Each dot represents a single patient sample (n is the number of GBM patient samples analysed); the bar indicates the mean ± standard deviation. The patient‐derived tumour (GBM) and blood (PB) samples were analysed using a paired t‐test; *P < 0·05, **P < 0·01; ***P < 0·001; n.s. = not significant. (b) Expression of NK cell surface molecules (gating on CD45⁺, NKp46⁺, CD3^neg cells) in GBM patient tumour, patient blood and control blood from healthy donors as in (a). (c) Representative histograms of CD9 expression on PB and GBM‐derived natural killer (NK) cells, grouped data as in (a). Statisical analysis was performed using a paired t‐test. *P < 0·05, **P < 0·01.

Figure 3

The repertoire of immunosuppressive molecules expressed in glioblastoma (GBM). (a) Expression of selected cell surface antigens on GBM stem‐like cell (GSC) lines; the data show expression by GBM20. A summary of expression across the four GSC lines is provided in Supporting information, Table S1. (b) Single‐cell (sc) RNAseq data [18] were clustered, revealing immune and tumour groups marked by protein tyrosine phosphatase receptor type C (PTPRC) and SRY‐box transcription factor 9 (SOX9/EGFR co‐expression, respectively. Expression of marker genes for cell lineages and those encoding immunomodulatory molecules are indicated. Expression is scored according to the values and key shown.

Figure 4

A spectrum of immune activity in glioblastoma (GBM). (a) Classification of GBM tumours [from 154 patients in the The Cancer Genome Atlas (TCGA) data set] into consensus immunome clusters (CIC) using the nearest centroid classification. The number of patients in each CIC is indicated in brackets. The cell signatures used to derive the CIC [14] are shown. (b) Mutational load in GBM CIC2 (red) and CIC4 (green) expressed as mutations per megabase. (c) Expression of granzyme B (GZMB) and interferon gamma (IFN‐γ) in CIC2 (red) and CIC4 (green). (d) Expression of negative regulators of immunity in CIC2 (red) and CIC4 (green). (e) Expression of cytokines and enzymes associated with immunosuppressive activity. (b–e) Data from CIC2 and CIC4 were compared using the Mann–Whitney test; n.s. = not significant, *P < 0·05, **P < 0·01, ***P < 0·001, ****P < 0·0001.