Single-cell profiling of human gliomas reveals macrophage ontogeny as a basis for regional differences in macrophage activation in the tumor microenvironment

Abstract

Background: Tumor-associated macrophages (TAMs) are abundant in gliomas and immunosuppressive TAMs are a barrier to emerging immunotherapies. It is unknown to what extent macrophages derived from peripheral blood adopt the phenotype of brain-resident microglia in pre-treatment gliomas. The relative proportions of blood-derived macrophages and microglia have been poorly quantified in clinical samples due to a paucity of markers that distinguish these cell types in malignant tissue.

Results: We perform single-cell RNA-sequencing of human gliomas and identify phenotypic differences in TAMs of distinct lineages. We isolate TAMs from patient biopsies and compare them with macrophages from non-malignant human tissue, glioma atlases, and murine glioma models. We present a novel signature that distinguishes TAMs by ontogeny in human gliomas. Blood-derived TAMs upregulate immunosuppressive cytokines and show an altered metabolism compared to microglial TAMs. They are also enriched in perivascular and necrotic regions. The gene signature of blood-derived TAMs, but not microglial TAMs, correlates with significantly inferior survival in low-grade glioma. Surprisingly, TAMs frequently co-express canonical pro-inflammatory (M1) and alternatively activated (M2) genes in individual cells.

Conclusions: We conclude that blood-derived TAMs significantly infiltrate pre-treatment gliomas, to a degree that varies by glioma subtype and tumor compartment. Blood-derived TAMs do not universally conform to the phenotype of microglia, but preferentially express immunosuppressive cytokines and show an altered metabolism. Our results argue against status quo therapeutic strategies that target TAMs indiscriminately and in favor of strategies that specifically target immunosuppressive blood-derived TAMs.

Keywords: Glioma; Immunotherapy; Macrophage; Single-cell sequencing.

Fig. 1

ScRNA-seq of neoplastic and immune cells from human primary gliomas. a Both whole-tumor and CD11b-purified single-cell suspensions, derived from glioma biopsies, were subjected to scRNA-seq (top) allowing for quantification of markers in single cells from both populations (bottom). b t-distributed stochastic neighbor embedding plot of cells from whole-tumor and CD11b-purified scRNA-seq, colored by the presence of somatic mutations that are clonal in exome sequencing (top) or by the expression of canonical marker genes (bottom), measured in counts per million (CPM). c Hierarchical clustering of cells (columns), grouped by their expression of canonical marker genes (rows)

Fig. 2

Analysis of published data identifies markers of ontogeny. a The intersection of: (1) genes that are differentially expressed between blood-derived and microglial TAMs in mouse (outer circle); (2) their homologues; (3) genes expressed (mean CPM > 1) in human BMDM/microglia from non-malignant tissue; and (4) TAMs from human gliomas (n = 16 patients). b Distributions of the log2 ratios (human BMDMs over microglia) (y-axis) for the differentially expressed murine-TAM homologues from (a)

Fig. 3

A gene signature to separate TAMs by ontogeny in mouse and human gliomas. a PCA of human TAMs in the space of genes that are ontogeny-specific in murine gliomas. The density curves of a Gaussian mixture model are in gray. b Consensus clustering of TAMs in the space of genes that are ontogeny-specific in murine gliomas. PCA-based cluster assignments from (a) are indicated by color. c Heatmap of the average expression (z-score) of indicated genes in windows of ten cells, sorted according to their PC1 score. d Log2 ratios of gene expression in murine blood-derived TAMs over murine microglial TAMs, averaged over the mouse models of Bowman et al. * = adjusted p value < 0.05 in both mouse models. Error bars indicate standard error of the mean. e Pearson correlation coefficients, computed via RNA-seq of LGGs and GBMs from TCGA (n = 558 cases). Genes are ordered by hierarchical clustering, boxes indicate a dendogram cut obtaining two clusters. f Top left: Flow cytometric analysis of TAMs gated on live CD11b + myeloid cells from a primary GBM (SF10941) stained for P2RY12 and CX3CR1. Top right: Flow cytometric analysis of TAMs gated on live CD11b + myeloid cells from a primary GBM (SF10941) stained for P2RY12 and HLA-DR. Bottom: Flow cytometric analysis of TAMs gated on live CD11b + myeloid cells from a primary GBM (SF11425) stained for P2RY12 and CD49D (encoded by ITGA4). g Gene expression from the Ivy Glioblastoma Atlas Project. Each column annotates expression in RNA‐seq of an anatomically defined tumor compartment. h In situ hybridization for BIN1 and TGFBI in anatomically annotated regions (indicated by color) for two primary GBMs

Fig. 4

Markers of ontogeny from human TAMs also separate brain-derived perivascular macrophages from microglia in scRNA-seq of mouse and human non-malignant cortex. A PCA of human TAMs (orange/pink, n = 1416 cells), human microglia from non-malignant cortex (purple, n = 17 cells), murine microglia from non-malignant cortex (blue, n = 33 cells), and murine perivascular macrophages from non-malignant cortex (red, n = 65 cells). PCA was performed in the space of 87 genes that are differentially expressed between murine-TAM lineages and robustly measured across all datasets (mean CPM > 1 in all datasets)

Fig. 5

Infiltration of blood-derived TAMs varies by glioma molecular subtype and correlates with inferior survival. a Z-scores of averages over blood-derived (top) and microglial (bottom)-TAM signature genes, compared across glioma subtypes (n = 371 cases, 117 oligodendrogliomas [OLIGs], 110 astrocytomas [ASTROs], 144 GBMs). CLS classical, MES mesenchymal, PN proneural. Significance was assessed via Tukey’s range test. NS indicates the test is not significant at p = 0.05. b Kaplan–Mayer survival curves, based on LGG TCGA RNA-seq for which survival information is available (n = 363 cases). Gene expression was averaged over the blood-derived and microglial signature genes, respectively, to assign a signature score to each case. The median signature score was used to divide cases into high-expressing and low-expressing cohorts. All comparisons were adjusted for age and gender, using cox proportional-hazards regression. HR hazard ratio

Fig. 6

TAMs simultaneously co-express canonical M1 and M2 markers in individual cells. a–c Distributions of canonical M1 and M2 marker genes, in cells expressing IL10, compared across scRNA-seq platforms. d Flow cytometric analysis of tumor-infiltrating CD206 + CD86+ TAMs gated on live CD11b + myeloid cells. e Representative flow cytometric analysis of tumor-infiltrating CD204+ TLR2+ cells gated on CD11b+, CD49D+ live macrophages (left) and CD11b+, P2RY12+ live microglia (right). f Quantification of flow cytometric analysis. Cells positive for indicated markers are given by circles (n = 3 patients). The fraction of cells positive for each individual marker is given by the histogram on the left, the fraction of cells positive for each marker combination is given on top of each panel