A better understanding of the morphology, diversity and function of GAMs is needed to improve their rationale targeting. Sankowski et al. were among the first to characterize the states of microglia in adult human patients who have undergone brain surgery to resect epileptic foci (n = 10), gliomas (n = 4) or brain metastases (n = 1) (26). They implemented scRNA-seq and time-of-flight mass cytometry (CYTOF) to identify regional and age-associated heterogeneity in microglia phenotypes including a) higher abundance of activated microglia in the white matter compared to the gray matter; b) common features expressed by all subsets of microglia (e.g., CX3CR1, TMEM119, CSF1R, P2RY12, SELPLG, MARCKS) versus discriminating features that identify specialized functions e.g. in antigen presentation (HLA-DRA, CD74), chemotaxis/inflammation (CCL2, TNF, IL1B), or hypoxia-response/angiogenesis (HIF1A, VEGFA); and c) downregulation of microglia homeostatic genes countered by upregulation of metabolic (APOE, LPL), inflammatory (SPP1) and interferon-induced (IFI27, IFITM3) genes in aging and in GB tumors (26). Batchu et al. profiled the immune landscapes of IDH1 WT GB versus IDH1 mutant astrocytoma (27). They identified 7 GAM subtypes present in both, including interferon-primed (IFN-), immunoregulatory (reg-), cytokine-enriched (inflam-), lipid-associated (LA-), pro-angiogenic (angio-), tissue resident-like (RTM-), and proliferating (prolif-) GAM subsets. Ligand-receptor maps showed that all 7 GAM subpopulations strongly interacted with T cells in IDH1 WT GB while only two subsets (inflam-GAMs and LA-GAMs) did in IDH1 mutant tumors. GAM-T cell crosstalk correlated with signatures of T cell exhaustion (PD-1, LAG-3, TIM-3, CTLA-4, TIGIT) and of inhibited T cell migration, highlighting different immunological ecosystems in these two advanced glioma types and a more immunosuppressive environment in IDH1 WT GB. Among the pathways engaged in all 7 GAM subpopulations in GB were those involving SPP1, MIF, GALECTIN, COMPLEMENT and PTN (Pleiotrophin), unraveling these pathways as potential immunosuppressive and pro-tumoral effectors (27). These phenotypes were confirmed in glioma mouse models, particularly a glioma-associated shift to phagocytic and DC-like states. These are evidenced by upregulation of CD11c, genes associated with migration (Cxcl13, Cx3Cr1, Csf1r), actin cytoskeleton regulation (Fscn1, Coro1a), cholesterol homeostasis (Abca1, Abcg1) and endocytosis (Apoe, Lrp1) (28). In vivo two-photon microscopy on open cranial window of immunocompetent Cx3cr1^GFP/WT; Ccr2^RFP/WT reporter mice revealed that resident microglia (that strongly expressed GFP) accumulate in clusters at the tumor periphery and are stationary while newly infiltrated MDM are highly mobile (29). Using the GL261 model and scRNA-seq of myeloid cells (CD11b⁺), Ochocka and colleagues found that in healthy brains, microglia represented the vast majority of myeloid cells (90%) while BAMs constituted only ~6% (19). In tumor-bearing mice, microglia (Tmem119⁺) remained dominant, constituting ~ 65% of all myeloid cells, but were displaced from the tumor core by infiltrating MDMs (Gal-3⁺) (19). These results were corroborated by Banerjee and colleagues who reported the spatiotemporal distribution of microglia and MDMs at early (14–15 days post-injection [dpi]), intermediate (24–25 dpi), and terminal (28–36 dpi) phases of glioma using Cx3cr1^CreER/+: R26^tdT/+ mice and two syngeneic mouse glioma models (GL261 and CT-2A). Their results showed that microglia accumulated inside the tumor at the early phase of tumorigenesis but were driven out of the tumor at later times by MDMs, demonstrating competition between these two macrophage sub-populations (30) ( Figure 2 ). This was further observed by De Leo et al. using another murine glioma model, namely SB28, where MDMs constituted >75% of all F4/80⁺ brain/glioma macrophages and in which neutralization of MDMs with anti-CD49d antibodies led to a surge in intra-tumoral microglia frequencies (31). MDM are not prevalent in tumors of patients with newly diagnosed GB, accounting for <20% of all macrophages, but constitute ~50% of all GAMs in recurrent GB (31), pointing to distinct TMEs in recurrence. Antunes and colleagues extended GAM profiling to identify differences between primary and recurrent tumors from human and mouse cohorts (18). Their data identified SALL1, TMEM119, P2RY12 as markers of microglia-derived GAMs that dominated the TME of primary tumors and TGFBI, CLEC12A and FXYD5 as markers of MDMs that outnumbered microglia in recurrence and were enriched at pimonidazole (PIMO)-positive hypoxic tumor regions ( Figure 2 ). Both microglia and MDMs presented significantly higher pro-angiogenic activity compared to normal brain microglia from control mice, as revealed using culture on chicken chorioallantoic membrane. In addition, they were incapable of inducing allogeneic proliferation of CD4⁺ or CD8⁺ T lymphocytes ex vivo, potentially due to an immunosuppressed state. Yeo et al. further detailed the immune changes occurring with GB progression particularly in an EGFR-driven GB mouse model (32). Their results showed that early GB tumors were mainly composed of pro-inflammatory microglia with expression of complement (C1qb, C1qa, C1qc) and lipoprotein catabolic enzymes (Apoe, Ctsd) genes, in contrast to late GB tumors that were populated by immunosuppressive pro-tumoral macrophages. It is important to note that not all GAMs are deleterious. Using scRNA-seq in two orthotopic glioma models (GL261 and CT-2A) and in a genetically engineered mouse glioma model (EGFRvIII+/TP53-/PTEN-), Kim and colleagues identified CD169⁺ (SIGLEC1) MDMs, induced by IFN-γ, as a beneficial GAM subset contributing to antitumor immunity (33). They showed that blockade of CD169 on bone marrow-derived macrophages (BMDMs) in vitro blunted their phagocytic and immunostimulatory capacities (33).

The TME of GB is lymphocyte poor, consistent with the reported resistance of GB patients to immune checkpoint blockade (34–36). In contrast, brain metastases (BrM) originating from various non-CNS cancers, such as melanoma, lung or breast cancer exhibit higher proportions of recruited T and B lymphocytes (37, 38). A recent analysis of >800 GB patients tumors identified 3 TME subtypes (TME low, med, or high) (39). Notably, TME high MES class tumors are enriched in T cell activation and exhaustion markers, regulatory T cells, and expression of immune checkpoints and associated with a trend towards improved survival in clinical trials testing neoadjuvant/adjuvant anti-PD-1 treatment or oncolytic virus PSVRIPO therapy (39). Interrogation of the Glioma Longitudinal AnalySiS (GLASS) cohort further revealed TME transitions with GB progression (39). Collectively, the fine mapping of the GB TME has revealed the heterogeneity of GAM phenotypes according to the glioma type, stage and treatment and have unraveled subsets of potential clinical impact. We posit that a 3D holistic view of the entire tumor (40) will localize the identified GAM subsets in space, define the molecular basis of their heterogeneity and ultimately lead to new approaches to overcome therapeutic failures in GB (41–44).

Spatial profiling of human GB showed distinct immune landscapes in the tumor peri-vascular (PVZ) versus peri-necrotic (PNZ) zones, which correlated with patient survival (15), highlighting the impact of metabolic gradients in shaping anti-tumor immunity. Disrupting the metabolic symbiosis between the tumor and its TME is thus a promising area for therapeutic intervention. In GB, metabolic adaptations can partly explain the impressive plasticity of GAMs. Current research in this area is focused on characterizing metabolic effectors that impact GAM diversity and functions in immunosuppression, tumor progression and clinical prognosis (45, 46). A notable example is the enzyme arginase 1 (Arg 1) expressed in GAMs that hydrolyzes L-arginine into urea and L-ornithine. Arg 1 mediates immune evasion by depleting L-arginine and depriving effector T cells and natural killer (NK) cells that need this nutrient for their proliferation (47). In GB, Arg 1 inhibition using pegylated arginine deiminase (ADI-PEG20) improved radiotherapy efficacy in immunocompetent orthotopic mouse models and promoted GAM differentiation into a pro-inflammatory anti-tumoral phenotype (48). The immunomodulatory role of Arg 1 is thought to be mediated by nitric oxide (NO), which limits mitochondrial metabolism by inhibiting the activity of acotinase 2 (A2) and pyruvate dehydrogenase (PDH), driving macrophages into a glycolytic state (49). De Leo et al. showed that ER stress-induced activation of protein kinase R (PKR)-like endoplasmic reticulum kinase (PERK) promoted enhanced glycolysis in MDMs through upregulation of the glucose transporter GLUT1. They showed that MDM-immunosuppressive function was partly mediated by IL-10, which was induced by lactylation downstream of lactate intracellular accumulation (31). To survive in the GB hypoxic and acidic TME, GAMs also engage the arginine-ornithine-polyamine axis to produce the highly basic polyamine that buffers their intracellular pH. Blockade of polyamine biosynthesis proved efficient in GAM depletion and reprogramming leading to improved OS in glioma-bearing mice (50). In parallel, GAMs upregulate the expression of Na/H1 exchange protein (NHE1) to export excess protons to the extracellular space. Hasan et al. demonstrated that selective deletion of NHE1 in Cx3cr1⁺ myeloid cells led to an immunogenic TME particularly in response to TMZ (51). GAMs surrounding the peri-necrotic hypoxic regions of the tumor express the creatine enzyme arginine-glycine amidinotransferase and produce creatine from arginine. Using isotopic tracing, Rashidi et al. showed that secreted creatine is taken up by tumor cells expressing the creatine transporter SLC6A8 and that pharmacological inhibition of this transporter slowed tumor growth (52). Tryptophan metabolism to kynurenine provides an additional axis governing the immunosuppressive GB TME. Kynurenine produced by tumor cells activates the aryl hydrocarbon receptor (AHR) in GAMs. AHR induces CCR2 expression, which furthers GAM recruitment, and upregulates CD39 that promotes adenosine-dependent T cell dysfunction (53). Tryptophan metabolism also results in the accumulation of low concentrations (<100 nM) of quinolate (QA), nduces the activation of the N-methyl-D-aspartate (NMDA) receptor and the Foxo1/PPARγ pathway that induces the differentiation of highly immunosuppressive macrophages (54). Pires‐Afonso et al. uncovered aconitate decarboxylase ACOD1 (Irg1) as an additional metabolic effector involved in this process. Compared to control mice, Acod1-deficient mice exhibited GAMs with increased immune reactivity and had reduced tumor load (55). Collectively, these studies allow better understanding of how GAM spatial distribution according to metabolite gradients affects their phenotypic and functional heterogeneity. Metabolic effectors thus provide new therapeutic strategies and perspectives in the treatment of GB.

Several examples illustrate the tight dialogue between GAMs and GB phenotypes. On one hand, glioma programs e.g. regulated epigenetically by SETD2, SOX10, CLOCK and KDM6 instruct GAM landscapes. Liu et al. showed that deletion or nonsense mutations in the histone methyltransferase gene SETD2 were associated with increased infiltration of GAMs mediated by TGFβ signaling (56). Wu et al. demonstrated that loss of the chromatin regulator SOX10, particularly in the receptor tyrosine kinase I (RTK-I) glioma tumor subtype, resulted in NPC- to MES-like transition, associated with GAM recruitment (57), and Xuan et al. linked glioma CLOCK expression to microglia intra-tumoral infiltration through the Olfactomedin-like 3 (OLFML3)/HIF1α/Legumain (LGMN)/P-selectin glycoprotein ligand-1 (PSGL-1) axis (58).

On the other hand, myeloid-derived signals shape glioma states. For instance, Rao et al. described a differential requirement of CSF-1R signaling in PDGFB-induced NPC-like versus in HRAS-driven gliomas (59) and the IL-1β pathway was shown to promote PDGFB-driven GB, through induction of monocyte chemoattractant protein (MCP)1-mediated recruitment of MDMs (60). Zhang et al. demonstrated that TREM1 expression on CD163⁺ GAMs is central in GB TME remodeling (61). This was corroborated by Dong et al. who reported that hypoxia-inducible factor (HIF)1α-dependent induction of TREM1 on GAMs promoted MES-like state transition via TGFβ secretion (62). Goswami et al. showed that lysine-specific demethylase 6B (KDM6B) promoted the expression of key immunosuppressive factors in GAMs, including signal-regulatory protein α (Sirpa), suppressor of cytokine signaling 3 (Socs3), and v-Maf avian musculoaponeurotic fibrosarcoma oncogene homolog B (Mafb) (63), and that myeloid-specific ablation of KDM6B reprogrammed GAMs into anti-tumoral effectors, improving the efficacy of anti-PD1 treatment in glioma-bearing mice (34). Altogether, these studies illustrate the complex mechanisms underlying GSC-myeloid cell reciprocal adaptations to intrinsic and extrinsic stresses in the tumor, which drive clonal selection, ITH and therapeutic resistance.

Besides their effects on lymphocytes, both PD-1 and PD-L1 have been demonstrated to modulate macrophage survival, proliferation and activation. PD-L1 regulates macrophage activation and proliferation (64) and PD-1 expression by macrophages inhibits their anti-tumoral phagocytic activity (65–67). High-dimensional analysis of the intra-tumoral immune remodeling elicited by anti-PD-1/CTLA-4, revealed important contributions of the myeloid compartment in therapeutic resistance (68) through GAM reprogramming (69, 70). Despite ICI approval in over 14 cancer indications, and their efficacy in durable responses in several advanced solid tumors (71), the three phase III clinical trials in GB had disappointing results. CheckMate 143 compared nivolumab to bevacizumab in recurrent GB at first recurrence and found comparable median OS between the two treatments (34). CheckMate 498 tested nivolumab in combination with RT in newly diagnosed GB with unmethylated MGMT promoter. The study did not meet the primary endpoint of improved OS and found that nivolumab + RT demonstrated shorter OS than TMZ + RT (36). CheckMate 548 reported similarly disappointing results of the evaluation of nivolumab in combination with TMZ and RT in newly diagnosed GB with methylated MGMT promoter (35). The U.S. Food and Drug Administration (FDA) has granted approval of the anti-PD-1 pembrolizumab (Keytruda^®) for certain adult and pediatric patients with advanced solid tumors that have high microsatellite instability (MSI-H), DNA mismatch repair deficiency (dMMR), or high tumor mutational burden (TMB-H) in a tumor agnostic manner. This approval was based on results of phase 2 trials i.e., KEYNOTE-158 (NCT02628067), KEYNOTE-164 (NCT02460198), and KEYNOTE-051 (NCT02332668) that showed the efficacy of pembrolizumab in this indication in 504 patients across over 30 cancer types. A phase I trial (NCT02359565) is currently evaluating the safety and preliminary efficacy of pembrolizumab in children and young adult patients with recurrent, progressive, or refractory high-grade gliomas, diffuse intrinsic pontine gliomas, hypermutated brain tumors, ependymoma or medulloblastoma. Ongoing trials evaluating ICIs in GB are testing safety and efficacy of nivolumab in combination with the anti-CTLA-4 antibody ipilimumab, administered either intravenously (i.v.) (NCT02311920) or intra-tumorally (i.t.) (NCT02017717), or in combination with the anti-LAG3 antibodies urelumab (NCT02658981) or relatlimab (NCT06325683). Other clinical trials are testing anti-PD-L1 antibodies, i.e., a) durvalumab in combination with the anti-CTLA-4 antibody tremelimumab in recurrent GB (NCT02794883), b) atezolizumab in combination with RT (NCT05423210), RT plus TMZ (NCT03174197) or with the anti-TIGIT antibody tiragolumab (NCT06328036) and c) avelumab in combination with standard-of-care (NCT03047473). The latter reported their findings in (72). In this phase II monocentric study, 30 patients with newly diagnosed GB were treated with avelumab at 10 mg/kg (i.v.) concurrent with TMZ. No new safety signals were induced by the combination, however addition of avelumab to standard-of-care did not show apparent benefit in improving OS. Besides ICI combinations, immune checkpoint blockade may also increase the efficacy of other targeted therapies including other immunotherapy modalities. Several ongoing studies are testing such ICI combinations e.g. of atezolizumab with the tyrosine kinase inhibitor (TKI) cabozantinib in recurrent GB (NCT05039281), the anti-PD-1 ezabenlimab with a HER2-CAR-NK (NK-92/5.28.z) in recurrent HER2-positive GB (NCT03383978), avelumab with a T cell vaccine against VEGFR2 (VXM01) (NCT03750071) or pembrolizumab with tumor treating fields (TTF) therapy (NCT03405792) or with the oncolytic virus lerapolturev (formerly PVSRIPO) (NCT04479241).

CSF-1R blockade is an attractive therapeutic option aimed at inhibiting the recruitment and pro-tumoral reprogramming of GAMs (73). In preclinical models of GB, CSF-1R inhibition had opposite effects on tumor growth depending on the tumor driver oncogene. In a PDGFB-driven glioma model, CSF-1R blockade inhibited tumor growth and improved mouse survival by re-programming GAMs rather than causing their depletion (59, 74). In contrast, in a RAS-driven model, CSF-1R depletion accelerated tumor growth (75). Two CSF-1R inhibitors have been evaluated in patients with GB in early phase trials, namely BLZ945 (in combination with anti-PD-1 - NCT02829723) and Pexidartinib (PLX-3397), either as a monotherapy (NCT01349036) (76) or in combination with RT plus TMZ (NCT01790503) ( Figure 3 ). The results of the phase II study testing PLX3397 as a monotherapy in recurrent GB showed that it crossed the BBB and attenuated CD14^dim/CD16⁺ monocytes in plasma but did not show efficacy in the 37-patient study cohort (76). Similarly, no improvement was attained by combining this inhibitor with RT plus TMZ (NCT01790503). In a phase I-II study assessing BLZ945 safety and preliminary efficacy as a monotherapy or in combination with the anti-PD-1 antibody spartalizumab (PDR001) in 146 patients with advanced/metastatic solid tumors including in 29 patients with GB, limited efficacy was reported particularly in non-MES GB patients (77), as determined by a correlation between PDGFRA gene expression and best percent change in tumor size. This was associated with decreased non-classical and intermediate monocytes in peripheral blood mononuclear cells (PBMCs) after treatment at multiple doses of BLZ945 and a downregulation of a macrophage geneset in the TME in tumor biopsy samples. Together, these results suggest that different GB subtypes might present differential responsiveness to CSF-1R inhibition.

Preclinical studies have shown that low dose chemotherapy depletes myeloid-derived suppressor cells (MDSC) (78–80). An early phase 0/I trial evaluated the effect of neoadjuvant treatment with low-dose capecitabine as a strategy to deplete MDSC in patients with recurrent GB, followed by capecitabine in combination with bevacizumab (NCT02669173). In 11 patients evaluated, the treatment was tolerated and led to a reduction in circulating MDSC countered by enhanced intra-tumoral infiltration of cytotoxic lymphocytes (81). An alternative approach is to restrict monocyte recruitment and to abrogate MDM-mediated tumor revascularization post RT (82, 83). Two clinical trials in patients with GB evaluated blocking CXCR4-stromal cell-derived factor-1 (SDF-1/CXCL12 interaction using the CXCR4 antagonist plerixafor (AMD3100) (NCT03746080, NCT01977677) (84). Plerixafor was also tested in combination with bevacizumab (NCT01339039) or with hematopoietic stem cell transplantation to improve chemotherapy tolerance (NCT00669669) (85). Promising results from these trials have been reported (84, 85) that warrant further testing of CXCR4 blockade in larger trials.

Taken the massive GAM infiltration in the GB TME, arming monocytes and macrophages with therapeutic agents, e.g., cytokines such as IFNα, or an engineered chimeric antigen receptor (CAR), is a promising therapeutic approach. An ongoing phase I-IIa clinical trial (NCT03866109) is currently recruiting patients with unmethylated MGMTp GB to evaluate the safety and preliminary efficacy of Temefron, an autologous CD34⁺-enriched HSPCs genetically modified to express human IFNα2 under the Tie2 promoter. This is based on the high tumor-homing property of Tie2⁺ monocytes and promising preclinical data by De Palma et al., showing improved anti-tumor immunity and reduced tumor angiogenesis using this approach (99). Contrary to limited intra-tumoral accumulation of CAR-T cells, the significant infiltration of GAMs in the TME also supports the rationale for CAR-macrophages (CAR-M) development. An active trial in HER2-overexpressing solid tumors will be testing anti-HER2 CAR-M (CT-0508) alone or in combination with pembrolizumab (NCT04660929) ( Figure 3 ).

As discussed above, GAMs rewire their metabolism to survive the harsh metabolic tumor environment that is hypoxic, acidic and nutrient-depleted. Through metabolic adaptations, GAMs further TME immunosuppression e.g., by depriving effector T cells of tryptophan via the activity of Indoleamine 2,3-Dioxygenase 1 (IDO1). IDO1 blockade has been evaluated in several clinical trials in patients with solid tumors with contrasting results. Four IDO1 inhibitors have been evaluated in GB in early phase studies, namely epacadostat, BMS-986205, indoximod and PF-06840003 ( Table 1 , Figure 3 ). Epacadostat (100) and BMS-986205 were tested in combination with nivolumab (NCT02327078; NCT04047706). Epacadostat was also evaluated in combination with retifanlimab (anti-PD-1), RT plus bevacizumab (NCT03532295). Indoximod was tested in combination with TMZ (NCT02052648) and PF-06840003 as a single agent (NCT02764151). However, the disappointing results (101) from a randomized phase III clinical trial (ECHO301) evaluating epacadostat in combination with pembrolizumab in patients with metastatic melanoma of (NCT02752074) has dampened enthusiasm for IDO1 inhibitor development (102). Such failure could stem from compensatory mechanisms by the additional tryptophan metabolism enzymes IDO2 and tryptophan 2,3- dioxygenase (TDO).

Blockade of the phagocytosis inhibitor CD47 with the antibody magrolimab has demonstrated futility with an increased risk of death in patients with acute myeloid leukemia (AML) (phase III trial ENHANCE III - NCT05079230). The FDA has requested a full clinical hold on all magrolimab studies in AML and myelodysplastic syndromes (MDS), and Gilead has since February 2024 paused all ELEVATE studies testing magrolimab in solid tumors. A phase I study testing this agent in children and adults with recurrent or progressive malignant brain tumors (NCT05169944) has also been suspended. CD47 is overexpressed on cancer cells and provides a “don’t eat me” signal by binding to signal regulatory protein α (SIRPα) on macrophages leading to inhibition of phagocytosis. Antibodies targeting CD47 release this inhibitory signal and bind to the Fc receptor on macrophages providing a concurrent “eat me signal” needed for the macrophage killing activity. Since CD47 is ubiquitously expressed on normal human cells, the reported safety issues with magrolimab might be related to on-target toxicity. Alternative approaches are currently testing anti-CD47 agents with mutated Fc (ALX148) or SIRPα decoys (e.g., TTI-622) fused to IgG4 Fc with a weaker killing activity. Bi-specific antibodies that dually recognize a tumor-associated antigen (e.g., CD19, CD20, PD-L1, EGFR) and CD47 can enhance specificity and limit toxicity, and are being evaluated in hematologic and solid tumors, in preclinical (103–105) and clinical (NCT03804996; NCT04806035) studies but testing in brain malignancies is not yet reported.