Glial and myeloid heterogeneity in the brain tumour microenvironment

Abstract

Brain cancers carry bleak prognoses, with therapeutic advances helping only a minority of patients over the past decade. The brain tumour microenvironment (TME) is highly immunosuppressive and differs from that of other malignancies as a result of the glial, neural and immune cell populations that constitute it. Until recently, the study of the brain TME was limited by the lack of methods to de-convolute this complex system at the single-cell level. However, novel technical approaches have begun to reveal the immunosuppressive and tumour-promoting properties of distinct glial and myeloid cell populations in the TME, identifying new therapeutic opportunities. Here, we discuss the immune modulatory functions of microglia, monocyte-derived macrophages and astrocytes in brain metastases and glioma, highlighting their disease-associated heterogeneity and drawing from the insights gained by studying these malignancies and other neurological disorders. Lastly, we consider potential approaches for the therapeutic modulation of the brain TME.

Malignant brain tumours make up a diverse group that can be coarsely divided into primary tumours, originating in the brain, and brain metastases (BrMs). Prognosis for most patients is poor, with a median survival of less than 2 years for patients with glioblastoma (GBM) or most types of BrMs. In the past decade, immune-checkpoint blockade therapies modulating T cell responses have improved survival in patients with melanoma and lung cancer, even when metastasized to the brain^1-3. Valuable insights have been gained from the study of T cell biology in brain tumours^4-6, but less is known about other cell populations contributing to immune suppression in the brain tumour microenvironment (TME)^7-12. The modulation of these other cell populations in the brain TME could improve the efficacy of immunotherapy against brain malignancies.

The brain TME is highly heterogeneous both in its evolution from early to late disease and its architecture, with differences detected across tumour type, between individuals with the same diagnosis, non-neoplastic cell types and cell states, and among individual tumour cell clones^10,13-15. Fibroblasts, pericytes, endothelial cells, glial cells and leukocytes interact with one another and with tumour cells to perpetuate brain tumour growth¹³. Dendritic cells^13,16,17, neutrophils^10,18, natural killer (NK) cells^7,19,20 (BOX 1), macrophages^10,11,21-24 and astrocytes^12,25 have been implicated in the modulation of the TME to impact T cell responses in brain cancer. Tumour-associated macrophages (TAMs), comprised of monocyte-derived macrophages (MDMs) and microglia, can represent up to 50% of the cells in the brain TME^10,26-29. While TAMs appear to be more abundant in primary brain tumours than BrMs^7,10, various TAM-targeting approaches in animal models lead to primary and metastatic brain tumour regression^11,23,30. Brain tumour-associated microglia and MDMs have distinct origins³¹ and incompletely overlapping functions²³. Their differential investigation has been hampered by technical difficulties such as the lack of differentiating markers³², the perturbation of their phenotype and function by isolation methods^33-35, and the permeabilization of the brain vasculature with irradiation, which enables monocytes to engraft into the brains of healthy adult mice^21,33,36. In addition, astrocytes are the most abundant glial cells in the central nervous system (CNS), and while they have emerged alongside microglia as key orchestrators of neuroinflammatory responses³⁷, their heterogeneity has been largely understudied in brain malignancies.

Box 1 ∣. Neutrophils, dendritic cells and natural killer cells in the brain tumour microenvironment.

In addition to brain tumour-associated macrophages and astrocytes, other cells participate in the inflammatory response in the brain tumour microenvironment (TME) (reviewed in REFS^38,226) (FIG. 1a).

Neutrophils

Neutrophils are the most numerous granulocytes and are early responders to injury and infection. Tumour-associated neutrophils are observed in isocitrate dehydrogenase wild-type (IDH^wt) and IDH mutant (IDH^mut) glioma^10,227, and brain metastases (BrMs)¹⁰ and are particularly numerous in BrMs¹⁰. BrM-associated neutrophils display phenotypic changes, including expression of the adenosine receptor and the receptor tyrosine kinase MET¹⁰, associated with neutrophil immune suppressive activity in other contexts^228,229. Neutrophils were recently found to amplify necrosis and promote tumour progression in a xenograft model of glioblastoma (GBM)²³⁰. Little is known about brain tumour-associated neutrophil subsets.

Dendritic cells

Dendritic cells (DCs) are myeloid cells capable of internalizing, processing and presenting antigens to tumour-reactive T cells in lymph nodes and numerous tissues²³¹, including the brain²³². At least five DC subsets have been identified in human GBM: conventional DC1 (cDC1), cDC2, migratory DCs, pre-DCs and plasmacytoid DCs (pDCs)^7,16. Moreover, cDC1s, cDC2s and pDCs were also identified in BrMs⁷. Mechanisms such as pDC-mediated expansion of regulatory T cells²³³ and T cell exhaustion via the immune-checkpoint protein PDL1 (REF.²³⁴) have yet to be studied in brain cancer, but numerous cell-contact and paracrine mechanisms could participate in brain malignancies²³¹.

Natural killer cells

Natural killer (NK) cells are innate lymphocytes capable of cytokine secretion and direct cytolytic activity against tumour cells when engaged by the proper combination of activating ligands²³⁵. Single-cell proteomic analysis classified brain tumour-associated NK cells into immature, highly cytotoxic and intraepithelial group 1 innate lymphoid-like cells⁷. NK cells from GBM appear predominantly immature, while NK cells from low-grade glioma and BrMs appear cytotoxic⁷. Early pre-clinical evidence demonstrates the potential for NK cell-mediated anti-tumour activity in the brain²³⁶. Therefore, the functional investigation of individual disease-associated NK cell subsets will be valuable.

Recent technological advances have enabled in-depth multi-omic studies of MDMs, microglia and astrocytes at single-cell resolution (BOX 2), revealing multiple cell subsets and activation states in development, health, and neurodegenerative and neuroinflammatory diseases. In this Review, we explore the diversity of microglia, MDMs and astrocytes in the TME of glioma and BrMs. A better understanding of cell heterogeneity within the TME could provide novel insights into the pathogenesis of brain malignancies and guide the development of efficacious immunotherapeutic approaches.

Box 2 ∣. Models to study brain tumour-associated astrocytes and macrophages.

Brain tumour researchers are daunted by a history of repeated clinical trial failures, particularly in malignant glioma, which call into question numerous aspects of drug development, including the pre-clinical models used to identify and evaluate therapeutic targets. The lack of in vitro models recapitulating the complexity of the brain environment has made in vivo models a critical component for the investigation of brain malignancies. These include orthotopically implanted, patient-derived or syngeneic tumours, genetically engineered mouse models²³⁷, and numerous models of metastasis (the majority of which are in vivo selected and implanted in various locations: the brain, other organs or intra-arterial¹⁴). As observed in several cancer models, the complex genetic landscape of human glioma is not fully recapitulated in genetically engineered mouse models²³⁷, and major differences in the evolution of xenografted brain tumours at the genomic level exist^237,238. Indeed, a recent comparison of the immune infiltrate in syngeneic orthotopic glioma models and human tumours concluded that the spontaneous 005 model, followed by the carcinogen-induced models GL261 and then CT2A, recapitulated some but not all aspects of the glioblastoma tumour microenvironment (TME)²²³. Humanized mouse models may more faithfully recapitulate tumour immunology²³⁹ as recently demonstrated for T cell suppression in glioblastoma⁶, but the humanized components are limited to haematopoietic cell compartments, and their widespread adoption has been limited by technical limitations and costs.

Once a pre-clinical mod el is chosen, numerous tools exist to characterize tumour-associated macrophages, astrocytes and other TME components²⁴⁰. Rare cell populations can be investigated using multi-omic and multidimensional profiling of single cells. Genomic^241,242, epigenomic²⁴³, transcriptomic^{6,27,29,50,51}, proteomic^7,22,23 and metabolomic²⁴⁴ profiles of individual cells can be obtained and compared with available datasets, while retaining information about the spatial organization of the TME. Single-cell transcriptomics are now well established for the identification of cell types, cell subsets and cell activation states in the TME^245,246. Going further, the prediction of molecular interactions and the computational inference of regulatory networks and cell state trajectories²⁴⁷ based on epigenetic and gene expression data may enable the unbiased detection of biological pathways of interest and candidate therapeutic targets²⁴⁸. Finally, novel approaches for in situ transcriptomics^249-252 allow the investigation of tumour architecture and spatial heterogeneity. Hence, efforts to integrate high-dimensional datasets with spatial genomic, proteomic and metabolomic approaches are needed to maximize our understanding of the brain TME^103,253. As compartment-specific, high-resolution molecular data become available, researchers may welcome the use of databases, such as Single Cell Portal at the Broad Institute⁵⁰, RecuR⁵⁵, GBMseq²⁷, Glioma Microglia RNA Expression⁸⁷, Brain TIME¹⁰, Adult Astrocyte RNA-Seq Explorer¹⁵⁰, Myeloid Single-cell RNA-Seq⁵⁹, DropViz²⁵⁴, Brain Immune Atlas¹⁶ and UCSC Cell Browser⁴³, that facilitate independent validation across multiple datasets.

Brain tumour-associated macrophages

The prevalence and plasticity of TAMs makes them attractive therapeutic targets for the modulation of tumour-specific immunity in the brain^38-40. However, the proper investigation of TAM heterogeneity and its implications for brain tumour pathology should be based on the differentiation of microglia and MDMs — cell subsets with different origins and functions that have been usually commingled in TME studies (BOX 3).

Box 3 ∣. Ontogeny and functions of brain TAMs.

Tumour-associated macrophages (TAMs), consisting of brain-infiltrating monocyte-derived macrophages (MDMs) and microglia are frequently grouped together in brain tumour microenvironment (TME) studies^{42,94,188,205,207} despite their distinct origins and development³¹. However, microglia derive from embryonic yolk sac erythro-myeloid progenitors that seed the central nervous system during development and self-renew in the adult brain^{21,31,255-257}. By contrast, monocytes originate from the postnatal bone marrow myeloid lineage and renew from haematopoietic stem cells²⁵⁸. While monocytes are reported to acquire a more immunosuppressive phenotype than microglia^21,43, both cell types exert immune suppressive or stimulatory functions depending on the context^8,44,259,260.

Despite similarities between monocyte-derived macrophages (MDMs) and microglia, their distinct ontogenies can lead to different roles in the brain TME²³. MDMs traffic to the TME, where they shift their transcriptional programmes in response to the local microenvironment^261,262. If microglia are depleted, monocytes re-populate the microglial niche and acquire a microglia-like gene expression signature²⁶². Nevertheless, core differences exist in ontogeny-specific transcription factors and the spatial distribution of brain tumour-associated microglia and brain tumour-associated MDMs^21,43. In addition, functional differences exist between tumour-associated microglia and MDMs. For example, minimizing MDM infiltration via deletion of the gene encoding CC-chemokine receptor 2 (Ccr2) in mice was not sufficient to prevent brain metastasis (BrM) formation, whereas depletion of microglia ameliorated T cell suppression in the BrM microenvironment²³. Multimodal single-cell analyses will likely enable further studies clarifying the individual functions of glioma-associated MDMs and glioma-associated microglia.

While recent studies employing single-cell analysis^7,8,16,22 and fate-mapping tools in mice²¹ distinguish tumour-associated microglia from MDMs, earlier methods, including irradiation and labelling of individual canonical markers (including CD45, CD11b, AIF1, CD68 and CX₃C-chemokine receptor 1 (CX₃CR1)), are now considered less reliable²¹. It was recently reported that the novel antigen CD49d is upregulated in glioblastoma-associated MDMs and downregulated in glioblastoma-associated microglia in mice and humans²¹ but usually numerous antigens or gene signatures ought to be analysed to confirm the identity of brain TAMs¹⁰⁰.

Novel methods (BOXES 2,3) have uncovered the broad diversity of TAM activation states in the TME, which extend beyond the dichotomous model of pro-inflammatory or anti-inflammatory phenotypes^10,41-44 (TABLE 1). In glioma and BrMs, TAMs outnumber other leukocyte populations such as T cells, NK cells and neutrophils. TAM composition in gliomas is influenced by mutations in the genes encoding isocitrate dehydrogenase (IDH1 or IDH2; hereafter referred to as IDH), which lead to the production of the oncometabolite 2-hydroxyglutarate and are highly prevalent in grade 2–3 gliomas^7,10,45,46 (FIG. 1a). Specifically, microglia comprise the majority of CD45⁺ cells in human IDH mutant (IDH^mut) gliomas, which contain smaller numbers of MDMs and few lymphocytes or neutrophils⁴⁵. However, human IDH wild-type (IDH^wt) gliomas harbour fewer microglia than IDH^mut gliomas but greater numbers of microglia than BrMs^7,10,24 (FIG. 1a). Human IDH^wt gliomas and BrMs contain a large number of MDMs and a moderate number of neutrophils^7,10. In addition, BrMs contain larger numbers of lymphocytes in the TME than do IDH^wt gliomas¹⁰. Furthermore, immunostaining studies established that glioma-associated microglia show an even distribution throughout the brain parenchyma¹⁰ and do not populate the perivascular space like MDMs. Mouse models of lung cancer-to-BrMs suggest that microglia make up 80% or more of the TAM compartment in BrMs⁴⁷ but their prevalence in human BrMs is much lower¹⁰. Of note, in both BrMs^7,10 and GBM from patients²⁷, microglia are more abundant at the tumour margin compared to the tumour core. These findings illustrate how tumour origin and genomic and metabolic aberrancies shape the composition of the TAM compartment (FIG. 1b,c).

Table 1 ∣.

Microglia-derived and monocyte-derived TAM subsets identified by single-cell technologies

Technique	Reported subsets	Disease or Model	Functional or prognostic evaluation
Monocyte-derived TAMs (MDMs)
scRNA-seq and CITE-seq16	Transitory/differentiating; phagocytic/lipid metabolic; hypoxic/glycolytic; SEPP1low/microglia-like; interferon-induced	ND and recurrent GBM; GL261 (GBM model)	No
	SEPP1high/anti-inflammatory	Recurrent GBM, GL261 (GBM model)
CyTOF7	MDM2	IDHwt glioma, melanoma BrM, carcinoma BrM	Positive correlation with OS in WHO grade II and III glioma
	MDM3	IDHwt and IDHmut glioma, melanoma BrM, carcinoma BrM
	MDM4	IDHwt glioma	No
scRNA-seq95	Microglia-like; glioma-associated; actively differentiating; interferon-induced	U87 (GBM model)	No
scRNA-seq and CyTOF8	CD73hi ICI-resistant, hypoxic, immunosuppressive	GBM	Negative correlation with survival (TCGA GBM); Cd73 KO in GL261 mice improved survival in combination with ICI
CyTOF223	Macrophages-monocytes; microglia-like; SIGLECF+ macrophages	GL261 (GBM model)	No
scRNA-seq47	MDM; radiation-induced MDM	H2030 (lung cancer BrM model)	No
CyTOF and CITE-seq23	Ly6Chigh BrM-enriched; Ly6Clow microglia-like; Ly6Chigh MDM	E0771 (breast cancer BrM model)	No
Microglia-derived TAMs
scRNA-seq16	Phagocytic/lipid metabolic; IFNγ-induced	ND and recurrent GBM, GL261 GBM model	No
	Hypoxic	Recurrent GBM and GL261 GBM model
scRNA-seq95	Homeostatic; glioma-associated; cycling/proliferating; undefined; glioma-associated; MHC II high	U87 (GBM model)	No
scRNA-seq and CyTOF22	Glioma-associated; ageing-associated and glioma-associated; homeostatic	IDHwt GBM	No
CyTOF223	Two undefined	GL261 (GBM model)	No
scRNA-seq47	TAM microglia; radiation-induced TAM microglia	H2030 (lung cancer BrM model)	No
CyTOF and CITE-seq23	Homeostatic; primed; TNF+ inflammatory state; undefined inflammatory state; migratory; migratory and inflammatory; S100+ LGALS1+, LGALS3+	E0771 (breast cancer BrM model)	No
	VISTA+ PDL1+ BrM-promoting CNS myeloid cells		Targetable with anti-VISTA and anti-PDL1
scRNA-seq100	Homeostatic; transcriptionally active; MHC I and MHC II high; proliferative	GL261 (GBM model)	No

Studies where TAM compartments were not further subclustered beyond all tumour-associated MDMs or microglia are not shown, although their single-cell data may be a useful resource^7,43,100. Further populations of myeloid cells that are not shown include immature/less differentiated monocytes, intermediate macrophages, mature macrophages and border-associated macrophages^23,100,223. BrM, brain metastasis; CITE-seq, cellular indexing of transcriptomes and epitopes by sequencing; CyTOF, cytometry by time of flight; GBM, glioblastoma; ICI, immune-checkpoint inhibitor; IDH^mut, isocitrate dehydrogenase mutant; IDH^wt, isocitrate dehydrogenase wild type; KO, knockout; MDMs, monocyte-derived macrophages; MHC, major histocompatibility complex; ND, newly diagnosed; OS, overall survival; scRNA-seq, single-cell RNA sequencing; TAMs, tumour-associated macrophages; TCGA, The Cancer Genome Atlas.

Fig. 1 ∣. Microglia, MDM and astrocyte diversity in brain tumours.

a ∣ Pie charts represent the composition of leukocytes in isocitrate dehydrogenase wild-type (IDH^wt) and IDH mutant (IDH^mut) glioma and brain metastases as determined by flow cytometry¹⁰ and cytometry by time of flight (CyTOF)⁷. Leukocyte composition varies by tumour type^7,10. The pie charts represent general trends from multiple tumours. b ∣ Non-malignant cell types in the tumour microenvironment (TME) have the potential for diverse activation states during both homeostasis (left panel) and cancer (right panel). Environmental cues and differential recruitment to the TME are the major drivers of heterogeneity in tumour-associated macrophages (TAMs; comprised of monocyte-derived macrophages (MDMs) and microglia). The plasticity of TME-infiltrating MDMs results in their acquisition of microglia-like transcriptional programmes. Astrocytes are highly heterogeneous, which is amplified upon activation. c ∣ Organizational structure for TAM nomenclature. Notable subsets are listed from recent reports^16,23. Intermediate states represent subsets that exist on a continuum between undifferentiated monocytes and macrophages or homeostatic microglia and tumour-associated microglia. GBM, glioblastoma; NK, natural killer.

The intratumoural heterogeneity of GBM plays a role in its resistance to therapy^48,49, while also shaping the inflammatory TAM infiltrate^27,50. For instance, four transcriptional GBM cell states (neural progenitor-like, oligodendrocyte progenitor-like, astrocyte-like and mesenchymal-like) are detected in different tumour areas^50,51 and other epigenetic states have also been described^52-54. A transcriptional mesenchymal-like signature is correlated with an enrichment in TAMs exhibiting anti-inflammatory transcriptional signatures and worse clinical outcomes⁵⁵. Moreover, non-mesenchymal-like GBMs that transition to mesenchymal-like GBMs harbour increased immunosuppressive TAMs at recurrence. Thus, while the regulation and plasticity of different GBM cell states is still unclear, tumour cell diversity has strong (and likely reciprocal) effects on the immune microenvironment. In this context, future studies should leverage the power of new in situ transcriptomic techniques to determine how the transcriptional and epigenetic heterogeneity of GBM cells controls TAM subsets in different tumour regions.

Microglia

Physiological functions and heterogeneity.

Microglia arise from yolk sac haematopoietic progenitors³¹, which self-renew in situ via mechanisms dependent on colony-stimulating factor 1 receptor (CSF1R) signalling^31,56,57. Microglia mature prior to the differentiation of other glia or neurons³³ and thus they are crucial for normal CNS development and homeostasis as they prune synapses through a mechanism that involves complement proteins⁵⁸. While studies in healthy adult mice did not detect extensive spatial heterogeneity^59,60, human microglia display region-specific transcriptional programmes, for example, associated with grey and white matter^22,61,62 (FIG. 1b). In addition, microglial transcriptional programmes diversify as humans age^22,63,64. Indeed, the comparison of microglia from ten species identified greater microglial diversity in humans than in all other species analysed, including macaque, marmoset, sheep, hamster and mice⁶⁵. This higher microglial diversity was attributed to evolutionary biological complexity rather than to microbial exposure for two reasons: (1) all individual human samples shared the transcriptional signatures of most microglial subtypes and (2) microglial diversity was the same between wild mice and mice housed in specific pathogen-free facilities⁶⁵. However, the roles of most of these microglial subsets in health and disease are still unknown.

Neurological disorders are associated with changes in microglial activation states related to immune cell recruitment, cell debris removal and cytokine production but typically not antigen presentation³³. Several microglial subsets have been identified based on their transcriptional signatures detected in neurological diseases such as Alzheimer disease and multiple sclerosis^66-70. In the context of neurologic disease, at least two microglial subsets have been identified in mice: (1) microglia associated with neurodegeneration driven by TREM2 receptor–apolipoprotein E (APOE) signalling and depicting the increased expression of genes believed to perpetuate disease (Apoe, Axl, Clec7a, Csf1 and Ccl2)⁷⁰; and (2) homeostatic microglia, featuring upregulation of the phagocytosis and lipid metabolism pathways thought to ameliorate disease through clearance of pathogenic cell debris⁶⁷. The roles of these microglial subsets in the pathogenesis of neurological diseases are still under investigation as are the location-specific cues responsible for inducing disease-associated transcriptional and epigenetic changes^34,60,71.

Evidence for microglia–brain tumour interactions.

In brain malignancies, tumour-associated microglia display substantial transcriptional heterogeneity, potentially associated with specific functions and therapeutic targets¹⁰ (FIG. 1b,c). For example, it was recently reported that microglial neuropilin 1 (NRP1), which acts as a receptor for placental growth factor, semaphorin 3A, vascular endothelial growth factor A (VEGFA) and tuftsin, could be a therapeutic target in GBM^72-76 (FIG. 2). NRP1 amplifies transforming growth factor-β (TGFβ) signalling⁷⁵, driving the expression of anti-inflammatory genes that limit glioma-specific immunity^77,78. The administration of the selective NRP1 inhibitor EG00229 modulated microglial gene expression, increased glioma-specific CD8⁺ T cell immunity and extended survival in a mouse model of GBM⁷⁹. Moreover, elevated NRP1 expression is associated with poor prognosis in patients with GBM, suggesting that the administration of NRP1 inhibitors in combination with antibodies targeting the immune-checkpoint protein PD1 may successfully activate GBM-specific T cells⁸⁰.

Fig. 2 ∣. Mechanisms of cell communication involved in tumour establishment, invasion and immune escape.

Multiple signalling pathways are activated in brain tumour microenvironment astrocytes and tumour-associated macrophages, but their contribution to immunosuppression as well as the dominance of a specific pathway within specific subpopulations (not shown) are unknown. Astrocytes (part a) upregulate genes encoding immunosuppressive (IL-10, transforming growth factor-β (TGFβ) and PDL1) and pro-oncogenic (CHI3L1) factors in the glioma tumour microenvironment, while nuclear factor-κB (NF-κB) signalling is concomitantly induced by tumour-derived RANKL. STAT3 is a central regulator of astrocyte reactivity programmes and characterizes a brain metastasis-promoting astrocyte subset. Other subsets promote tumour growth through paracrine factors such as brain-derived neurotrophic factor (BDNF) or TNF through cell contact-dependent (CX43) or cell contact-independent mechanisms. Microglia (part b) and monocyte-derived macrophages (MDMs) (part c) can have both pro-inflammatory and anti-inflammatory functions. In microglia (part b), tumour sensing (SIGLEC-H and GPR34) and homeostatic (CX₃C-chemokine receptor 1 (CX₃CR1)) programmes are disrupted, while TLR2 and NRP1 signalling respectively increase tumour invasion and immunosuppression. In monocyte-derived macrophages (part c), an anti-inflammatory phenotype supported by aryl hydrocarbon receptor (AHR) activation suppresses T cell responses (part d) via the catabolism of ATP. T cells supply tumour-associated macrophages with a CSF1R-independent survival signal by inducing IGF1 via IL-4. Anti-tumour T cell responses are suppressed by cytokines such as IL-10 and TGFβ, checkpoint proteins such as PDL1 and VISTA, and adenosine signalling. Overall T cell exhaustion is evidenced by increased immune-checkpoint expression and the inhibitory receptors A2aR and CD161 are broadly present in glioblastoma-infiltrating T cells. ARGl, arginase 1; cGAMP, cyclic GMP–AMP; EGF, epidermal growth factor; ER, oestrogen receptor; HIF1α, hypoxia-inducible factor 1α; IGF1, insulin-like growth factor 1; MIF, macrophage inhibitory factor; MMPs, matrix metalloproteinases; NRP1, neuropilin 1; p-PDGFRβ, phosphorylated platelet-derived growth factor receptor-β; p-STAT3, phosphorylated signal transducer and activator of transcription 3; TLR, Toll-like receptor; VEGFA, vascular endothelial growth factor A.

Several glioma-derived factors have been shown to induce pro-tumorigenic functions in brain tumour-associated microglia. Toll-like receptor 2 (TLR2) is upregulated in glioma-associated microglia and has been shown to support tumour progression and invasion⁸¹. Glioma cells express the endogenous TLR2 ligand versican, which upregulates the microglial expression of matrix metalloproteinase 14 (MMP14), linked to increased tumour invasiveness and growth⁸² (FIG. 2). Therefore, the therapeutic inhibition of TLR2 using O-Vanillin and other antagonists has been proposed as an approach to reduce microglia-dependent glioma invasiveness⁸³. Conversely, systemic TLR9 stimulation via type B⁸⁴ and C⁸⁵ CpG oligodeoxynucleotides inhibited glioma growth^84,86 and lung cancer and melanoma BrM formation⁸⁵ in mice. While the precise mechanisms involved were not determined, TLR9 stimulation upregulated the phagocytic machinery in microglia, and contact between microglia and tumour cells was required for tumour cell death⁸⁵. Further work in organotypic glioma cultures detected enhanced microglial phagocytosis following TLR3 and TLR9 co-activation⁸⁶. Thus, innate sensors represent candidate targets to augment tumour-associated microglial phagocytic and cytotoxic programmes.

Tumour-derived extracellular membrane particles are also important microglial modulators in the TME. It was recently reported that the microglial uptake of fluorescently labelled membrane material from mouse glioma cells in vivo triggers functional modifications, including the upregulation of the immune-checkpoint protein PDL1 and multiple MMP-encoding genes, concomitant with the downregulation of pathways involved in tumour sensing such as SIGLEC-H and the G protein-coupled receptor GPR34 (REFS^87,88) (FIG. 2). Indeed, the analysis of human glioma samples detected the downregulation of two-thirds of the known microglial sensome⁸⁷, the set of receptors that sense the local microenvironment⁸⁹. This downregulation of sensome-encoding genes is more pronounced at the tumour core^27,87 and in microglia that have taken up GBM extracellular membrane particles⁸⁷. Further investigations of the cargo of tumour-derived extracellular membrane particles are needed to identify additional molecules (for example, proteins, microRNAs and metabolites) involved in the modulation of microglial responses in the TME and potentially at distant sites.

Recent work reported preliminary comparisons of microglia in glioma and BrMs^7,10. As mentioned previously, microglia comprise a smaller proportion of the TAM compartment in BrMs than in glioma¹⁰ (FIG. 1a). In addition, three findings support a more activated microglial phenotype in IDH^wt glioma and BrM than in IDH^mut glioma: (1) microglia from IDH^wt gliomas and BrMs express increased levels of CD14 and the scavenger receptor CD64; (2) microglia from IDH^wt gliomas and BrMs show an amoeboid morphology, associated with activation; and (3) microglia from IDH^mut gliomas display a ramified (branched) morphology associated with homeostatic microglia⁷. Further observations were made through bulk RNA sequencing of microglia and MDMs in human BrMs and gliomas. First, BrM-associated microglia express gene sets associated with type I interferon signalling and nuclear factor-κB (NF-κB) signalling, whereas glioma-associated microglia lack expression of these gene sets. Second, BrM-associated microglia distinctly showed increased expression of CXC-chemokine ligand 8 (CXCL8, also known as IL-8), a known neutrophil attractant and, accordingly, neutrophils more robustly infiltrated BrMs than gliomas¹⁰. Taken together, these findings suggest complex tumour-specific functional interactions between microglia and tumour cells in the TME.

Microglial heterogeneity in brain tumours.

Single-cell studies have made considerable progress in revealing TAM transcriptional diversity, paving the way for the analysis of the functional roles of TAM subpopulations in BrMs and other CNS malignancies. An analysis of microglial heterogeneity in human IDH^wt GBM and matched controls combined single-cell RNA sequencing (scRNA-seq) with mass cytometry, identifying nine microglial clusters in published scRNA-seq datasets and surgical controls²² (TABLE 1). Gene ontology analyses detected gene expression signatures suggestive of diverging functions in different subsets: antigen processing and presentation, production of pro-inflammatory cytokines, responses to type I interferons, and chemotaxis²². GBMs contained all microglial clusters identified in control tissue (albeit with different frequencies) and at least two additional unique transcriptional clusters; one such tumour-associated microglial cluster was also detected in an independent IDH^wt GBM dataset²⁷. GBM-specific microglia clusters had downregulated microglial core gene transcripts and upregulated metabolic, inflammatory and interferon-related transcripts²². Interestingly, mass cytometry studies detected preserved moderate-level expression of the protein products of microglial core genes, suggesting that post-transcriptional regulatory mechanisms play an important role in the control of microglial responses in the TME²². Collectively, these findings identify novel microglial subsets associated with GBM²²; the study of their functions and regulatory mechanisms may guide new approaches for their therapeutic modulation.

In a mouse model of non-small-cell lung cancer BrM, the diversity of BrM-associated microglia was lower than that of MDMs, yet scRNA-seq identified four clusters that had upregulated pro-inflammatory genes, including Il1b (encoding IL-1β), Ifnb1 (encoding IFNβ1), Il12 (encoding IL-12) and Tnf (encoding TNF), along with downregulated homeostatic genes⁴⁷ (TABLE 1). In addition, radiotherapy amplified inflammatory responses and stress-associated programmes in microglia⁴⁷. While distinct gene signatures were detected during the establishment of metastasis and response to radiation, this study did not investigate the roles of these transcriptional programmes in functional experiments⁴⁷.

Another study combined cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) and cytometry by time of flight (CyTOF) to define seven clusters of BrM-associated microglia in mice and humans²³. Intriguingly, CNS-resident myeloid cells promote BrM growth²³. By contrast, depletion of CC-chemokine receptor 2 (CCR2)⁺ tumour-associated MDMs did not reduce tumour growth as demonstrated in mouse models of glioma³². A subset of CNS-resident myeloid cells, which consists primarily of microglia, secrete CXCL10 and are enriched for the T cell-suppressive molecules VISTA and PDL1. Blockade of CXCL10 or VISTA in combination with anti-PDL1 treatment led to the control of BrM in mice²³ (TABLE 1; FIGS 1c,2). This study demonstrates the importance of tumour-associated microglia in brain malignancies and the complex role of CXCL10, which promotes T cell recruitment⁹⁰ but also immune-suppression through the microglial expression of VISTA and PDL1 (REF.²³). Moreover, this study described a key difference between CCR2⁺ macrophages in BrM and glioma, where CCR2⁺ macrophage depletion impairs tumour growth⁹. Taken together, these studies exemplify how single-cell technologies may identify novel therapeutic targets in unique TAM subsets across tumour types.

Monocyte-derived macrophages

Recruitment and education.

Monocytes do not infiltrate the healthy brain parenchyma but, in the context of inflammatory conditions such as cancer, monocytes are recruited to the CNS through mechanisms mediated by the CCR2 (REF.¹⁶) ligands CCL2 (REF.⁹¹) and CXCL12 (REF.⁹²). This process is facilitated by the breakdown of the blood–brain barrier⁹³. In malignant gliomas, MDMs represent over a third of the absolute tumour mass and their abundance correlates with glioma grade⁹⁴. Hence, it is not surprising that some of the earliest TAM-targeting drugs tested clinically in brain cancer were chemokine receptor blockers such as maraviroc and plerixafor (TABLE 2). More recently, high IL-33 expression was associated with increased tumour-associated MDM infiltration, possibly as a result of the control of microglial chemokine production⁹⁵. The roles of these pathways in the control of microglia and MDMs in the brain TME should be further investigated.

Table 2 ∣.

Clinical trials targeting MDMs, microglia and astrocytes in brain cancer

Targeted pathways	Therapeutic agent and preclinical ref.a	Trial IDb and/ or ref.	Date initiated	Phase	Tumour type	Result	Designed to target TAMs or astrocytes?
Microglia, MDM
MIF–CXCR4 and CXCL12–CXCR4	Plerixafor + bevacizumab	NCT01339039	2011	1	HGG (R)	Terminated, low accrual	Yes
	Chemoradiation followed by plerixafor	NCT01977677 c	2013	1	GBM (ND)	Completed, safe
	RT followed by plerixafor + TMZ	NCT03746080 (REF.224)c	2018	2	GBM (ND)	Recruiting
PDL1–PD1 and PDL2–PD1	Several8,178,180,d	NCT03047473	2017	2	GBM (ND)	Accrual met, ongoing	No
		NCT02968940	2016	2	IDHmt glioma (R)	Completed, results not published
		NCT03174197	2017	1/2	GBM (ND)	Active, not recruiting
		NCT02550249 (REFS179,225)	2015	2	GBM (ND, R)	Completed, possible benefit
		NCT02617589 (REF.178)	2015	3	GBM (ND)	Completed, did not meet primary endpoint (OS)
CCL5–CCR5	Leronlimab	NCT04504942	2020	2	Advanced cancer (R; stable BrM)	Recruiting	Yes
CD39–CD73–AMP–adenosine	Oleclumab8 (with durvalumab + paclitaxel + carboplatin)	NCT03616886	2018	1/2	Triple-negative breast cancer, advanced (R, stable BrM)	Recruiting	No
CSF1–CSFR1	Pexidartinib (also known as PLX3397)	NCT01349036 (REF.204)c	2011	1	GBM (R)	Terminated by sponsor	Yes
	Pexidartinib + RT/TMZ	NCT01790503 c	2013	1b/2	GBM (ND)	Completed, safe but no evidence of efficacy
AXL kinase	Bemcentinib188 (pre- or post-surgery)	NCT03965494	2019	1	GBM (R)	Recruiting	Yes
SIRPα–CD47	BI 765063 +/− anti-PD1	NCT03990233	2019	2	Advanced solid tumours (stable BrM allowed)	Recruiting	Yes
MDMs
IDO1	Epacadostat197 (with RT + bevacizumab +/− anti-PD1)	NCT03532295	2018	2	GBM (R)	Recruiting	No
	Linrodostat (with anti-PD1 + RT +/− TMZ)	NCT04047706	2019	1	GBM (ND)	Recruiting
	Indoximod + TMZ	NCT02052648	2014	1/2	HGG (R)	Completed, results not published
	Indoximod + RT + chemotherapy	NCT02502708 c	2015	1	Paediatric HGG (R)	Completed, safe
	Indoximod + RT + chemotherapy	NCT04049669 c	2019	2	Paediatric HGG (R)	Recruiting
	PF-06840003	NCT02764151 (REF.198)	2016	1	Glioma (R)	Completed, safe but sponsor not pursuing
Brain tumour-associated astrocytes in contact with BrM
Connexin 43	Meclofenamate175	NCT02429570	2015	1	BrM (R)	Accrual met, ongoing	Yes
STAT3-high tumour-associated astrocytes
STAT3	WP1066	NCT01904123	2013	1	GBM (R) or melanoma BrM (R)	Recruiting	No
		NCT04334863	2020	1	Paediatric medulloblastoma or HGG	Recruiting	No
	Silibinin12	NA	2016	1	Lung cancer BrM (R)	Completed, 75% overall response rate	No

BrM, brain metastasis; GBM, glioblastoma; HGG, high-grade glioma; IDH^mut, isocitrate dehydrogenase mutant; IDO1, indoleamine 2,3-dioxygenase 1; MDMs, monocyte-derived macrophages; NA, not available; ND, newly diagnosed; R, recurrent; RT, radiation; TAMs, tumour-associated macrophages; TMZ, temozolomide; +/− indicates treatment with or without the drug that follows this abbreviation.

^aDrug class can be determined by conventional nomenclature: drugs ending in ‘-mab’ are monoclonal antibodies; those ending in ‘-ib’ are small-molecule inhibitors. Other drugs not following this convention are small-molecule inhibitors, with the exception of BI765063, which is a monoclonal antibody. Preclinical references are given where available.

^bNCT Trial IDs can be accessed at https://clinicaltrials.gov/.

^cClinical trials that are linked (phase I and II of the same agent).

^dMore recent trials have been listed in the table; however, there are more than 60 other GBM trials and 35 other BrM trials targeting the PD1–PDL1 axis^1,2.

In the brain TME, monocytes assume activation states that result from an adaptation to both the CNS and the TME (FIG. 1b,c; TABLE 1). Among them, RNA expression signatures containing genes encoding anti-inflammatory molecules, such as arginase 1 (ARG1), CD39, TGFβ, IL-10, IL-1 receptor antagonist (IL1RN), CD155, macrophage scavenger receptor 1 (MSR1) and PDL1 (REFS^{10,21,27,43,96,97}), are of particular interest for therapeutic immunomodulation (FIG. 2). TME-infiltrating monocytes also have increased expression of major histocompatibility complex (MHC) class II molecules, proteins involved in the modification of the extracellular matrix and proangiogenic factors such as VEGFA. Taken together, these findings suggest that brain tumours control not only the recruitment but also the phenotype of MDMs in the TME.

MDM heterogeneity.

Numerous studies analysed MDMs in bulk following the isolation of defined cell populations using CD45, CD11b or lineage tracing in mouse models in combination with canonical markers to distinguish MDM from microglia^{10,21,24,41,42,44,96,98}. Unbiased single-cell technologies have recently identified additional MDM subsets (TABLE 1). Notably, the acquisition of a microglia-like signature in both the glioma and BrM TME suggests an adaptation of MDM to the brain microenvironment regardless of tumour presence (TABLE 1; BOX 3), similar to previous findings^35,99.

Most recently, single-cell transcriptomic and proteomic approaches were used to comprehensively profile MDM subsets in human GBM and the mouse model GL261 (REF.¹⁶). In addition to undifferentiated as well as transitory monocytes, the authors identified six monocyte-derived TAM clusters with distinct signatures, including phagocytosis/lipid metabolism, hypoxia/glycolysis, microglia-like (selenoprotein P (SEPP1)^low), anti-inflammatory (SEPP1^high) and IFNγ-induced MDMs¹⁶ (TABLE 1). A specific MDM subset featuring hypoxia-related signalling was exclusively present in recurrent GBM, suggesting that this signalling pathway may enable therapeutic resistance¹⁶. BrM-associated MDMs could also be classified into different subsets and had upregulated antigen presentation pathways as well as responses to a hypoxic environment characterized by increased hypoxia-inducible factor 1α (HIF1α) and VEGFA⁴⁷. Heterogeneity was also detected along the macrophage differentiation axis, and sex-specific transcriptional differences in the expression of MHC class II genes and the gene encoding macrophage inhibitory factor (MIF)¹⁰⁰ exist. Overall, the detection of diverse MDM subpopulations and tumour type-specific MDM education hint at specific roles performed by different MDM subsets in the TME, which require further functional investigation.

MDSCs and their overlap with MDMs.

Myeloid-derived suppressor cells (MDSCs), derived from monocytic or granulocytic cells, are detected in the TME of many cancers, including GBM¹⁰¹. MDSCs were previously distinguished from MDMs and microglia based on the expression of CD14, CD45 and CD11b, though highly specific markers are still lacking^41,101. The ability of MDSCs to suppress GBM-specific CD8⁺ T cell responses via ARG1, IL-10, TGFβ and other molecules^101,102 has raised interest in this TAM population. Indeed, a granulocyte–macrophage colony-stimulating factor (GM-CSF)-induced subset of glioma-associated IL-4Rα⁺ MDSCs suppressed T cells via ARG1 in mice³⁰. Thus, the investigation of brain tumour-associated MDSCs at single-cell resolution is likely to reveal heterogeneous cell states and shed light on therapeutically relevant mechanisms driving specific MDSC subsets, as recently reported in fibrosarcoma¹⁰³.

Mechanisms of MDM-mediated immune suppression.

MDMs participate in multiple immunosuppressive mechanisms that limit tumour-specific T cell responses (FIG. 2). The aryl hydrocarbon receptor (AHR), a ligand-activated transcription factor expressed by TAMs, T cells and GBM cells, is activated by numerous ligands, including GBM-derived kynurenine and other oncometabolites, through IDO1, TDO2 or IL-4I1 (REFS^104,105). AHR activation in monocytes boosts their recruitment to the brain and upregulates immunosuppressive gene expression in MDMs via the induction of the transcription factor KLF4, the suppressor of cytokine signalling 2 (SOCS2)-dependent downregulation of NF-κB signalling⁴⁴ and the expression of the ectoenzyme CD39, which cooperates with CD73 to catalyse the production of the T cell-suppressive metabolite adenosine¹⁰⁶. In addition, AHR activation is involved in the induction of forkhead box P3 (FOXP3)-negative type 1 regulatory T cells^107,108, which limit effector T cell responses and are enriched in peripheral blood mononuclear cell and brain samples from patients with GBM¹⁰⁹. These findings suggest that AHR signalling in TAMs and potentially in T cells contributes to the suppression of brain tumour-specific immunity.

Further mechanisms of MDM-mediated suppression of antitumour immunity have been proposed and demonstrated to varying degrees in brain tumour-associated MDMs, other TAMs or MDSCs. These mechanisms include the expression of PDL1 (REFS^{7,97,100,110-113}) and PDL2 (REF.⁹⁷), soluble factors such as IL-10 (REFS^41,111), VEGFA¹¹⁴ and TGFβ^97,100,114, catabolic enzymes such as ARG1 (REFS^30,41), and indoleamine 2,3-dioxygenase 1 (IDO1)⁸, chemokines^9,44 and IL-33 (REF.⁹⁵) (FIG. 2). While multiple mechanisms are common to both human and mouse cells, increased ARG1 expression was detected in mouse but not human glioma-associated MDMs¹⁶. Most of these mechanisms have not been definitively investigated in loss-of-function studies but inferred based on gene expression analyses and functional studies in tumours outside the CNS^115-117.

Roles of astrocytes in brain tumours

Astrocytes are the most abundant glial cells in the CNS, with remarkable functional diversity in health and disease. During homeostasis, astrocytes regulate synaptogenesis and neurotransmission, provide essential metabolic support to neurons, and modulate blood flow as well as the blood–brain barrier¹¹⁸. These homeostatic functions are disrupted in disease^119,120. In addition, astrocytes secrete multiple immunomodulatory molecules such as cytokines, chemokines and metabolites with important effects in inflammation and neurodegeneration^121-128.

Astrocytes communicate with microglia, monocytes and T cells among other cell types, playing important roles in infection, neurodegeneration and autoimmunity^{37,122,129-135}. In addition, astrocytes have been reported to both inhibit¹³⁶ and promote^137-141 brain tumour growth in mice in addition to human in vitro systems but less is known about astrocyte subsets and their potential roles in the TME and brain tumour-directed immunity.

Potential astrocyte subpopulations

Astrocytes exhibit substantial heterogeneity, which is further amplified by region-specific transcriptional signatures in health and disease^142-146. Specifically, trophic signals induce regional transcriptional programmes during development that remain stable into adulthood^147-149. In addition, astrocytes acquire specialized functions as a result of neural circuit-specific modulation^150-153 (FIG. 2). Indeed, often, the maintenance of a non-activated state in astrocytes depends on signals produced by local neurons^154,155. The role of these region-specific astrocyte subsets on tumour development and their regulation by tumour cells is poorly characterized.

Trauma, infection, stroke, neurodegeneration, inflammatory disease and malignancy alter astrocyte phenotype and function. Bulk transcriptional analyses identified Janus kinase (JAK)–signal transducer and activator of transcription 3 (STAT3) signalling as a master activator of astrocytic neurotoxic and anti-inflammatory or reparative responses^156,157. Moreover, NF-κB activation is reported to potentiate the neurotoxic functions of STAT3-positive astrocytes in some neurological diseases such as multiple sclerosis¹⁵⁸. Conversely, the activation of p38 MAPK¹⁵⁹, the deubiquitinating enzyme A20 (also known as TNFAIP3)¹⁶⁰, otubain 1 (REF.¹⁶¹), SOCS1 (REF.¹⁶²), SOCS2 (REFS^122,163) and SOCS3 (REFS^162,164) decreases astrocyte neurotoxic and pro-inflammatory activities^122,159-166 (FIG. 2). Collectively, these findings identify signalling pathways involved in the control of astrocyte responses, although the specific astrocyte subsets involved are still mostly unknown.

Single-cell analyses (BOX 2) are beginning to identify functionally relevant astrocyte subsets in neurological diseases^127,128,167, including brain cancers^25,168. Indeed, the heterogeneity of astrocyte functional responses suggest that astrocyte subsets specialized in promoting CNS repair and limiting inflammation promote tumour growth and the generation of an immunosuppressive TME in GBM and BrMs. While tumour-associated astrocyte subsets have not been defined using single-cell analyses, changes in peritumoural astrocytes have been described with traditional immunohistochemistry^12,168-170 and flow cytometry¹⁴². In pioneering studies, reporter mouse strains and flow cytometry antibody screens were used to define five astrocyte subsets conserved across different brain regions, with region-specific variations in their relative abundance¹⁴². Interestingly, the transcriptional programme of malignant glioma cells was closely related to that of a synapse-promoting astrocyte subset¹⁴², setting the stage for the study of astrocyte heterogeneity and function in CNS malignancies.

Astrocyte subsets in brain tumours

Several studies have recently analysed the roles of astrocyte subsets in CNS malignancies and their potential as therapeutic targets^{12,25,168-170}. As with macrophages, some reports describe subsets without identification of a clear functional role. For example, one study identified a subset of astrocytes that expresses phosphorylated platelet-derived growth factor receptor-β (p-PDGFRβ) and is located at the borders of brain micrometastases of HER2-amplified breast cancer¹⁷⁰. The authors reasoned that, as p-PDGFRβ participates in the recovery from stroke¹⁷¹ and penetrating brain injury¹⁷², p-PDGFRβ may also participate in the early stages of breast cancer BrM. While the exact role of this astrocyte subset remains unknown, it could be depleted by the inhibition of p-PDGFRβ with the tyrosine kinase inhibitors pazopanib¹⁷⁰ or crenolanib¹⁷³.

A second study focused on the role of astrocyte subsets in regulating blood–tumour barrier permeability. First, the authors performed RNA microarray gene expression analysis on metastases in mice displaying varying degrees of blood–tumour barrier breakdown¹⁶⁸. This study detected the upregulation of sphingosine 1 phosphate receptor 3 (S1PR3) expression in astrocytes in mice bearing breast cancer BrMs and in patients with BrMs. Interestingly, S1PR3 inhibition restored blood–tumour barrier integrity in mice. Conversely, S1PR3 activation in cultured human astrocytes induced IL-6 and CCL2 production, and S1PR3 expression in human BrMs was predictive of blood–tumour barrier permeability as determined by contrast enhancement¹⁶⁸. Therefore, an S1PR3-induced astrocyte state may be responsible for the disruption of vascular integrity, affect recruitment of leukocytes from the circulation and play a role in the establishment of BrMs.

Additionally, a recent study investigated the role of oestrogen receptor (ER)-expressing BrM-associated astrocytes to provide paracrine signals promoting tumour cell invasion¹⁶⁹. Prior studies reported an increased risk of BrM in premenopausal women with HER2⁺ breast cancer or triple-negative breast cancer (TNBC) (in which cancer cells lack ER, HER2 and the progesterone receptor) compared with postmenopausal women with either cancer type¹⁷⁴. Initial studies confirmed the expression of ERα and ERβ in astrocytes, which produced brain-derived neurotrophic factor (BDNF), a ligand for epidermal growth factor receptor (EGFR) and tropomyosin-related kinase-β (TRKβ). EGFR and TRKβ signalling induced the migration and invasion of TNBC BrM cells, which could be inhibited by depletion of oestrogen or inhibition of TRKβ and BDNF with small molecules¹⁶⁹. These studies suggest that BrM-associated astrocytes could be manipulated with hormonal therapies.

In a fourth study, BrM cells induced a subset of tumour-associated astrocytes to produce pro-inflammatory cytokines through cell contact and gap junctions formed by connexin 43 (CX43)¹⁷⁵. In particular, the translocation of cyclic GMP–AMP (cGAMP) into the astrocyte cytosol via CX43 channels activates stimulator of interferon genes (STING), inducing the production of IFNα and TNF, which promoted metastatic growth¹⁷⁵. Astrocyte–tumour cell communication through CX43 also plays a role in GBM, although further molecular mechanisms mediated by CX43 are still unknown¹⁷⁶.

Two recent studies^12,25 investigated the role of brain tumour-associated astrocytes in shaping the neuroinflammatory milieu. In the first, bulk astrocytes isolated from GBM and epilepsy tissue samples were analysed by bulk RNA sequencing²⁵. Gene set enrichment analysis detected increased IFNγ and JAK–STAT signalling in GBM-associated astrocytes; the expression of pro-inflammatory, neurotoxic molecules such as CD37, PDL1, chitinase 3-like protein 1 (CHI3L1), and complement components C1S and C1R was increased in purified astrocytes from tumour specimens²⁵. The gene expression profiles of GBM-associated astrocytes differed from those of progenitor, mature, neurotoxic or neuroprotective astrocytes. Microglia were found to be important regulators of astrocyte responses in GBM as described in other neurological disorders^37,124,158. Indeed, astrocytes isolated from tumours secreted anti-inflammatory cytokines such as IL-10 and TGFβ but only in the presence of microglia or macrophages²⁵. Interestingly, the JAK inhibitor ruxolitinib boosted astrocyte secretion of pro-inflammatory cytokines such as IFNγ and TNF, suggesting that the modulation of astrocyte responses can abrogate immunosuppressive mechanisms operating in the TME²⁵.

A second study described an anti-inflammatory astrocyte subset driven by STAT3 in the BrM TME in patient samples and mouse models of BrM¹². STAT3⁺ astrocytes from the margins of the human and mouse BrM TME secreted MIF, modulated TAM responses and suppressed tumour-specific T cells. These studies raise several important points for consideration for further studies. First, STAT3 is a therapeutic target in a subset of BrM-associated and potentially GBM-associated astrocytes. As STAT3 is linked to both neurodegeneration-associated and BrM-associated astrocytes, single-cell studies are needed to characterize STAT3⁺ astrocytes associated with brain malignancies to investigate further defining and function-driving gene programmes for targeting. In addition, the location of STAT3⁺ astrocytes at the margins of BrMs calls for further studies of the spatial heterogeneity and cell–cell interactions in the TME. Although numerous differences exist regarding the TME of primary brain tumours and BrMs (FIG. 1a), the activation of JAK–STAT signalling in astrocytes associated with malignancies of different ontogenies highlights the potential to identify common therapeutic targets.

Taken together, these studies highlight the need for a better characterization of astrocyte subsets in the TME to identify novel mechanisms of disease pathogenesis and candidate targets for therapeutic intervention.

Targeting TAMs and astrocytes

The therapeutic targeting of TAMs and astrocytes for the treatment of glioma and BrM has been attempted in many clinical trials. Thus far, these trials have shown modest responses, and multi-target approaches will likely be required to shift the neuroinflammatory environment of the TME (TABLE 2).

The PD1–PDL1 axis

Antibodies blocking T cell inhibitory checkpoints induce remarkable clinical responses in numerous malignancies. In particular, antibodies blocking PDL1 and PD1 are efficacious in patients with metastatic cancer (TABLE 2), but their efficacy for the treatment of brain malignancies varies widely^1-3,177. Anti-PD1 antibodies induce the regression of melanoma and lung cancer BrMs with similar degrees of success to those achieved with metastases to other organs, but patients with more rapidly growing tumours respond less often^1-3. Renal cell carcinoma BrMs show positive responses in only 12% of patients, whereas 25% of patients with renal cell carcinoma metastases respond elsewhere¹⁷⁷. As with systemic metastases, BrMs respond more often when additional T cell checkpoints, such as CTLA4, are targeted in addition to the PD1–PDL1 axis^1,2. In the context of GBM, standard checkpoint blocking antibody treatments induce positive responses in roughly 8% of patients¹⁷⁸. The timing of PD1 blockade may be relevant as a recent clinical trial in recurrent GBM reported a significant extension of median survival when anti-PD1 treatment was initiated prior to tumour resection¹⁷⁹ (TABLE 2). Potential mechanisms involved in resistance to checkpoint blockade therapy in the brain include bioavailability, intratumoural heterogeneity and a relative paucity of tumour neoantigens.

PDL1 is expressed in GBM and BrM tumour cells, but numerous researchers have also reported its expression on microglia and MDMs^87,110,180. PDL1 expression in tumour cells and TAMs is heterogeneous¹⁸¹ and may be induced in TAM subsets by local paracrine signals such as tumour-derived prostaglandin E2 (REF.¹⁸²), CXCL10 (REF.²³) or IL-6 (REF.¹⁸³). In a mouse model of hypermutated GBM, myeloid PDL1 expression negatively correlated with the clinical response to anti-PD1 antibody treatment, suggesting that mechanisms other than canonical PD1-mediated T cell suppression may be at play⁹⁷. Indeed, PDL1 also interacts with CD80 (also known as B7-1), which is expressed on CD4⁺ and CD8⁺ T cells¹⁸⁴. Furthermore, PDL1-expressing MDMs increased CD4⁺ FoxP3⁺ regulatory T cell numbers in vitro⁹⁷, further supporting a role for PDL1 as an important mechanism of MDM-mediated immunosuppression.

PDL1 expression in TAMs from non-brain metastases predicts response to antibodies against PDL1 or PD1 in TNBC¹⁸⁵ and non-small-cell lung cancer¹⁸⁶ but correlative data on PDL1 expression in BrM TAMs are currently lacking. The heterogeneity of PDL1⁺ and PDL1⁻ TAM subsets in BrM and GBM also remains unknown. Nevertheless, ongoing investigations are focused on targeting the PD1 axis in combination with microglial signalling pathways such as the receptor tyrosine kinases AXL, TYRO3 and MER^187-189. The AXL inhibitor bemcentinib (also known as BGB324) is now being tested in a phase I trial for recurrent GBM as a monotherapy¹⁹⁰ (TABLE 2), but dual inhibition of PD1 and AXL has been shown to extend survival in preclinical models of melanoma BrMs and GBM^187,188.

CD39 and CD73 ectoenzymes

CD39 cooperates with CD73 to produce the immunosuppressive metabolite adenosine¹⁰⁶. Mass cytometry in combination with scRNA-seq was recently used to analyse various patient tumours after anti-CTLA4 and/or anti-PD1 treatment, resulting in the identification of a CD73^high population of TAMs⁸ depicting a transcriptional signature closer to that of MDMs than microglia. Indeed, the combination of CD73 targeting with CTLA4 and/or PD1 blockade improved survival in GBM mouse models⁸.

Pharmacological inhibitors and blocking antibodies targeting CD73, CD39 and adenosine receptors have been developed to interfere with the immunosuppressive mechanisms driven by CD39 and CD73 (REF.¹⁰⁶). However, extensive blockade of this axis did not show efficacy in mouse models of GBM, dampening enthusiasm for clinical trials targeting CD73, CD39 or adenosine in GBM¹⁹¹. Thus far, the only active clinical trial in this area is focused on the effects of CD73-specific antibodies in combination with anti-PDL1 therapy in patients with metastatic TNBC (TABLE 2). However, unless clinically stable, BrMs are an exclusion criterion. Two studies have documented the association between soluble CD73 and immunotherapy-refractory disease, which is sometimes associated with systemic metastases¹⁹². However, there are no reports of the specific upregulation of CD73 and CD39 in BrMs, with one study reporting CD73 expression only in endothelial cells and not in melanoma BrM-associated leukocytes¹⁹³. While there may be a rationale for targeting the CD39–CD73 axis for systemic anti-cancer immunity, and not all types of BrMs may possess the same biology, currently the evidence for targeting this axis for brain tumours remains weak.

IDO1

GBM cells¹⁹⁴ and associated TAMs¹⁰ express IDO1, the rate-limiting enzyme metabolizing tryptophan into kynurenine⁸. IDO1 limits T cell responses through multiple mechanisms^44-195, including the expansion of regulatory T cells¹⁹⁶ and the activation of AHR in TAMs and other cell types⁴⁴ (FIG. 2). IDO1 expression by non-malignant cells was sufficient to drive tumour progression in mouse models of GBM, supporting therapeutic interventions based on IDO1 inhibition in combination with radiation, cytotoxic chemotherapy and immune-checkpoint blockade¹⁹⁷. While initial clinical trials with IDO1 inhibitor monotherapy in GBM did not show efficacy¹⁹⁸ (TABLE 2), two ongoing phase II trials are investigating the combination of IDO1 inhibitors with anti-PD1 antibodies, radiation or other systemic therapies^199,200 (TABLE 2). Moreover, a recent report described IL-4I1 as the main tryptophan-catabolizing enzyme that leads to AHR activation in glioma and numerous other malignancies¹⁰⁵. Thus, the future of IDO1 inhibition in GBM is uncertain as tryptophan-catabolizing enzymes other than IDO1 may have compensatory effects and IDO1 inhibitors may not adequately suppress AHR agonist production in GBM. An alternative approach may be the blockade of AHR, which mediates the immunosuppressive activities of the products of IDO1, tryptophan 2,3-dioxygenase (TDO) and IL-4I1^44,201,202. Moreover, while tryptophan metabolism has been targeted in GBM and IDO1 inhibition has been tested in metastatic cancer in general, the effect of IDO1 inhibition on BrMs has not been evaluated.

The CSF1–CSF1R axis

The disruption of colony-stimulating factor 1 (CSF1) signalling via CSF1R can lead to the depletion of certain subsets of macrophages^16,56,57. Consequently, CSF1R inhibition has been explored for the therapeutic modulation of the GBM TME (TABLE 2). CSF1R inhibition reduced tumour growth in a mouse model of the proneural subtype of GBM^11,203. Interestingly, instead of the expected depletion of GBM TAMs, it was their phenotypic reprogramming that diminished pro-tumour functions and increased sensitivity to chemotherapy^11,203. A more recent study reported that CSF1R inhibition interferes with the maturation of tumour-associated MDMs in a mouse model of GBM that resembles the mesenchymal cellular state¹⁶. Unfortunately, a previous clinical trial failed to show efficacy of the CSF1R inhibitor pexidartinib to improve survival in GBM^50,204 (TABLE 2), even when pharmacodynamic studies confirmed adequate drug concentrations in tumours and microglia depletion²⁰⁴. While the investigators hypothesized that the relative proportion of GBM subtypes could have led to treatment resistance, no biomarker or correlative studies were performed to demonstrate this mechanism of resistance²⁰⁴.

Resistance to CSF1R inhibition may develop as a result of the local production of other macrophage or microglial survival factors⁹⁸. Indeed, resistance to CSF1R inhibition in a mouse model of proneural GBM developed through the production of insulin-like growth factor 1 (IGF1) by CD45^low CD11b⁺ TAMs⁹⁸. Despite strong evidence showing resistance to CSF1R blockade as a monotherapy, CSF1R inhibition could provide greater efficacy in combination with other treatments such as radiotherapy²⁴. For example, resistance to radiation is characterized by the progressive influx of MDMs, which is prevented by CSF1R inhibition^16,24. In experimental melanoma BrM models, the depletion of brain TAMs with pexidartinib prevented metastasis formation, which was attributed to a reduction in TAM-derived MMP3. However, MMP inhibitors led to a modest decrease in metastases, suggesting that TAMs contribute to metastatic colonization only partially through MMPs²⁰⁵. Thus, future studies should investigate the combination of CSF1R blockade with additional therapies or its use in specific patient subsets.

The SIRPα-CD47 axis

Phagocytic programmes in TAMs bridge innate and adaptive anti-tumour immunity via the clearance of malignant cells and the presentation of tumour-associated antigens. The phagocytic state of TAMs is controlled by numerous pro-phagocytic signals (such as calreticulin, SLAMF7, opsonizing antibodies and phosphatidyl serine) and anti-phagocytic signals (such as IgG₄ antibodies, MHC class I, PDL1 and CD47)²⁰⁶. CD47, which is frequently upregulated in primary brain cancer cells²⁰⁷, binds the receptor signal-regulatory protein-α (SIRPα), inhibiting the phosphorylation of myosin II required for phagocytosis²⁰⁸. CD47 expression is widespread in primary brain cancers^207,209 with IDH^mut gliomas expressing lower levels of CD47 (REF.²¹⁰). Many types of glioma, including those in children and GBM, upregulate CD47, which triggers signalling through SIRPα in TAMs²⁰⁹. Several studies in mice have demonstrated efficacy of anti-CD47 antibodies against primary brain tumours^{207,209,211,212}, which in particular generated enthusiasm for the potential of this treatment in four paediatric tumour types²⁰⁷. Anti-CD47 antibodies enhance phagocytosis in MDMs^207,211 and microglia^86,209 and can synergize with anti-PD1 antibodies²¹¹, temozolomide²¹¹, and TLR3 and TLR9 activation⁸⁶ in mouse GBM models. The majority of studies identify microglia as the main targets of anti-CD47 antibodies in primary brain cancer^86,209. Although anti-CD47 antibodies are being tested in advanced solid tumours, including in patients with stable BrMs²¹³ (TABLE 2), the targeting of the CD47–SIRPα axis has not been investigated in primary brain tumour clinical trials.

STAT3

STAT3⁺ astrocytes can promote BrMs via immune suppression, which is reversed by STAT3 inhibitors in mouse models¹². The natural polyphenolic flavonoid STAT3 inhibitor derived from milk thistle seeds, silibinin, induced the regression of BrMs in the majority of patients in a phase I clinical trial¹² (TABLE 2). Silibinin administration was safe and it appeared to selectively target BrMs, leaving metastases outside the brain unaffected¹². This selective inhibition of BrMs argues that the beneficial effects of STAT3 inhibition result from the modulation of the brain TME and not the inhibition of STAT3 signalling in tumour cells¹². STAT3 activation may also characterize a subset of astrocytes that cooperate with microglia to maintain an immune suppressive TME in GBM²⁵. STAT3 inhibitors have been effective in mouse models of melanoma BrM, although they are thought to act directly on tumour cells and cells in the TME other than astrocytes²¹⁴. A phase I clinical trial using the STAT3 inhibitor WP1066 as a monotherapy in recurrent GBM or melanoma BrM is currently recruiting patients²¹⁵ (TABLE 2).

CX43

Gap junction formation via CX43 is inhibited by the anti-inflammatory drug meclofenamate, which is used for the treatment of migraine¹⁷⁵. A phase I clinical trial evaluating the use of meclofenamate in patients with BrMs is progressing²¹⁶. Enrolment has been completed, but the results have not been published (TABLE 2). However, it should be noted that, although meclofenamate was selected for BrM treatment because of its effects on gap junctions, its anti-tumour effects may also involve the inhibition of cyclo-oxygenase activity²¹⁷ or tumour cell–tumour cell connections²¹⁸.

Conclusion

The brain TME includes heterogeneous populations of tumour cells, astrocytes, TAMs and other immune cells with sometimes opposing roles in tumour initiation, progression and response to therapy^219,220. Thus, we believe it is most appropriate to study the TME of brain malignancies at the level of single cells and not as bulk tissue. Technological advances now enable the interrogation of thousands of individual brain tumour^29,50 and TME cells^7,16,22,23 (BOX 2). Several TAM subsets have been recently associated to brain malignancies^7,16,22 (TABLE 1) but our knowledge on tumour-associated astrocytes is still limited. Determining which subpopulations of TAMs or tumour-associated astrocytes are functionally important for brain malignancies remains challenging as these studies require the mechanistic investigation of candidate cell subsets defined by dozens of gene, protein or metabolic changes. In addition, the crosstalk between astrocytes, microglia, MDMs and other cells in the brain TME plays a central role in the control of tumour-specific immunity. Ideally, TAMs and tumour-associated astrocyte subsets should be studied in parallel with tumour cell heterogeneity to identify candidate interactions of therapeutic interest as was recently done in GBM-associated macrophages²²¹. We recently combined rabies virus-based tracing of cell interactions with scRNA-seq to develop a novel approach (named RABID-seq) to study CNS cell interactions, including the mechanisms mediating them and their effects on interacting cells in vivo²²². RABID-seq offers a unique opportunity to study TAM and astrocyte subsets and their interactions with brain tumour cells in the TME. TAM and tumour-associated astrocyte heterogeneity and interactions must be studied separately in BrMs, GBM and other brain malignancies because the tumour type has a major influence on the TME^7,10. The study of TAM–astrocyte–tumour cell interactions will have important implications for our understanding of the pathogenesis of brain malignancies and the development of novel therapeutic approaches.

Glossary

Brain tumour microenvironment

(TME). The dynamic, heterogeneous mixture of extracellular matrix and malignant and non-malignant cells, including neurons, glia, leukocytes and endothelial cells, that make up the gross brain tumour mass. Half or more of the cells within the brain tumour microenvironment, in particular in glioma, are non-neoplastic.

Cell types

The stable identities of cells, primarily defined by intrinsic gene expression programmes, which contain permanent traits, including the potential to express certain genes and exert certain functions in response to external signals.

Cell states

Within cell types, states are defined by the current gene expression, metabolic or functional conditions, which are a product of external signals from infection, inflammation, ageing or other disease. The duration of each cell state may be a function of ongoing signalling.

Microglia

CNS-resident phagocytic cells primarily derived from the yolk sac with multiple roles in the development of disease, including synapse, surveillance and cell debris engulfment, cytokine production, and cross-talk with astrocytes and oligodendrocytes.

Multi-omic studies

Approaches that include the quantification and characterization of data from various molecular domains such as the genome, transcriptome, proteome, metabolome and epigenome.

Isocitrate dehydrogenase

(IDH1 or IDH2). Enzymes in the tricarboxylic acid cycle that catalyse the decarboxylation of isocitrate, producing alpha-ketoglutarate and carbon dioxide. Some mutations in IDH lead to the production of 2-hydroxyglutarate, a metabolite whose accumulation leads to genomic instability of glial cells through DNA methylation.

Mesenchymal-like

One of the four transcriptional meta-signatures of GBM defined by bulk RNA sequencing featuring loss of neurofibromin 1 (NF1) expression and increased expression of mesenchymal-related, hypoxia-response, stress and glycolytic genes. It was associated with increased tumour-associated macrophage infiltration and a more immunosuppressive tumour-associated macrophage phenotype, suggesting its relevance for tumour–tumour microenvironment cross-talk.

Extracellular membrane particles

Subcellular particles shed from malignant or non-malignant cells that are composed of a lipid bilayer with variable size and cargo, including proteins, metabolites and nucleic acids. Communication with neighbouring and sometimes distant cells is possible through their contents.

Mass cytometry

A cytometric technique that quantifies protein expression at the single-cell level by combining the use of rare earth metal isotope-conjugated antibodies and cytometry by time of flight (CyTOF).

Cellular indexing of transcriptomes and epitopes by sequencing

(CITE-seq). Combining DNA barcode-conjugated antibody labelling with single-cell RNA sequencing to enable immunophenotyping and unbiased transcriptome analysis.

Blood–brain barrier

A variety of features specific to brain vasculature that limit the influx of chemicals or immune cells in addition to the active efflux of chemical molecules, thereby limiting neurotoxicity.

Myeloid-derived suppressor cells

(MDSCs). A heterogeneous group of immature myeloid cells that contain monocyte-like and granulocyte-like subsets, which inhibit anti-tumour T cell responses through ARG1, IL-10 and TGFβ.

Ectoenzyme

An extracellular enzyme that can be membrane bound or secreted.

Blood–tumour barrier

Heterogeneous impairment of the blood–brain barrier and the interface between the blood and tumour due to tissue destruction by tumour growth and neoangiogenesis, resulting in altered but not necessarily efficient penetration of drug molecules and immune molecules.

Ontogenies

The origins and development of specific cell types.

Proneural subtype of GBM

One of four subtypes defined by bulk RNA sequencing of GBM through the Cancer Genome Atlas (TCGA), characterized by elevated expression of PDGFRA and CDK4. Originally, it was thought that each tumour represents a distinct subtype, but single-cell RNAseq of GBM revealed that all four subtypes can be detected within each tumour with varying proportions. More recently, the proneural subtype has been hypothesized to coincide with the oligodendrocyte-like and neural progenitor-like cellular states identified by single-cell RNA sequencing.