2.1 Microglia diversity in development, health and disease

The processes of proliferation and repopulation of microglia occur from within the microglia compartment exclusively in the resting CNS. Broadly, this also occurs following injury, provided the blood–brain barrier (BBB) remains intact [[45]]. Immediately adjacent to the brain parenchyma, ‘border-associated’ or ‘CNS-associated’ macrophages derived from bone marrow occupy a physically distinct niche in choroid plexus, meningeal and perivascular spaces [[55]]. During embryogenesis, these cells and other bone marrow-derived macrophages originate from the fetal liver's haematopoietic stem cell (HSC) pool [[47, 56]]. However, circulating bone marrow-derived progenitors can also contribute to the repopulation of the parenchyma following CNS damage if BBB integrity is compromised [[57, 58]]. Following lethal irradiation and bone marrow (BM) transplantation, donor-derived peripheral monocytes alone can reconstitute this compartment in chimeric mice [[59]]. Similarly, microglia dominate the microenvironment in the early glioma transformation and infiltration phases. However, malignant progression is associated with hypoxia, necrosis, endothelial cell tight junction breakdown, chemotaxis, and accumulation of peripheral monocyte-derived macrophages [[20]]. In parallel, the immune activation signatures characteristic of low-grade glioma (LGG) are replaced by immunosuppressive expression profiles [[30, 44]]. For example, in glioma, microglia-derived TAMs exhibit pro-inflammatory signatures, whereas most macrophages display immunosuppressive signatures [[44, 60]]. Additionally, the extent of microglia infiltration correlates with mutation profile and glioma progression. Namely, IDH-wildtype glioblastomas demonstrate increased macrophage infiltration, whereas IDH mutant gliomas, in particular grade II gliomas, enrich for microglia infiltration [[60, 61]]. Mutations at the NF1 gene are also associated with increased macrophage infiltration and total TAM numbers [[62]]. However, whether these observations represent an intrinsic difference based on myeloid cell origin is unclear. It also remains to be determined if such effects could result from differential enrichment of macrophages in the BBB-deficient tumour core and microglia in the invasive periphery.

2.2 Gender differences and myeloid-derived suppressor cells in glioblastoma

Sex differences in microglia have been observed in the developing and adult brains [[63]]. For example, microglia from females expressed higher levels of interferon regulatory factor 3 [[64, 65]], whereas microglia from males exhibited elevated MHC II expression [[66-68]]. Differences in expression signatures were preserved in microglia from females transplanted into male mice, pointing to a lasting impact of early hormonal influences. Significant sex differences have also been identified in myeloid-derived suppressor cells (MDSCs). MDSCs represent a specific TAM subgroup derived from immature circulating monocytes and distinguished by the following features: (a) a downregulation of MHC Class II expression [[69]], and (b) an upregulation of transcription factors hypoxia-inducible factor 1a (HIF-1α) and signal transducer activator of transcription 3 (STAT3), compared to their parent cells [[70-72]]. MDSCs potently suppress B cell, NK cell and especially T cell activation across various cancers, including glioblastoma [[73-75]]. Selective enrichment of MMDSCs within GBM tumours has been demonstrated in male mice and male patients. In contrast, a separate granulocytic or polymorphonuclear myeloid-derived suppressor cell (GMDSC or PMN-MDSC) population is enriched in the bloodstream of GBM-bearing females [[74]]. Since host immunosurveillance is key to regulating tumour initiation and progression, sexual dimorphism in microglia, MDSC and TAM compartments may contribute to the established finding of a higher GBM incidence and poorer prognosis in men.

2.3 Microglia and macrophage ontogeny and cell surface markers

It is unclear to what extent the role of yolk sac-derived, brain-resident microglia is interchangeable with bone marrow-derived macrophages for tumour progression. However, microglia-specific biology is important functionally. For example, microglia exhibit distinct electrophysiology compared to macrophages, mediated through potassium channel expression [[76, 77]]. Spalt-like transcription factor 1 (SAll1), Pu.1 and interferon regulator factor 8 (Irf8) are selectively expressed by microglia [[48, 78, 79]]. Microglia-specific surface antigens common to mice and humans include a transmembrane protein of unknown significance, TMEM119, and the purinergic receptor P2YR12 [[80, 81]]. A recent meta-analysis of five previously published murine transcriptional datasets additionally identified solute carrier family two members (Slc2a5) and fc receptor-like molecule (Fcrls) to be consistently upregulated in microglia. In contrast, elastin microfibril interfacer 2 (Emilin2), guanine deaminase (GDA), haptoglobin (Hp) and selectin L (Sell) were enriched in macrophages/monocytes in RCAS and GL261 mouse models of glioblastoma [[49]].

Microglia have also been isolated from macrophages using specific fluorescent antibody cell sorting (FACS) marker combinations, such as gating for the CD11b⁺, CD45-low, CX3 motif chemokine receptor 1 (CX3CR1⁺) microglia population [[38]]. A CD11b⁺/CD206low/CD163⁻ human microglia population has been distinguished from perivascular macrophages (CD11b⁺/CD206high/CD163⁺) [[20, 60, 61, 82, 83]]. Moreover, monocyte-derived TAMs show a high expression of CD163/CD206/CD9 [[20, 43, 60, 61, 82-85]] and an upregulation of genes related to growth factor activation (EREG, ligand for EGFR, IFITM2) and folate synthesis (FPR3) and in line with this, the S100A gene family having a range of inflammatory functions [[86]]. However, caveats apply to marker combinations across species. For example, CX3CR1 expression is restricted to microglia from mice but is common to microglia and monocyte-derived macrophages from human microglia [[61, 87]]. Furthermore, lymphocyte antigen 6 (Ly6C) is expressed in mice but not humans [[88]].

A healthy human consensus ‘core microglia’ gene expression signature would include TMEM119, P2RY12, and CX3CR1. Other reported markers include P2RY13 and selectin p ligand (SELPLG) [[43, 67, 68, 85]], colony-stimulating factor (CSF1R), triggering receptor expressed on myeloid cells-2 (TREM2), c-c chemokine receptor 5 (CCR5) and Myristoylated alanine-rich C-kinase substrate (MARCKS) [[20, 43, 61, 80, 83-85]] (Table 2).

Conversely, genes enriched in bone marrow-derived macrophages (BMDMs) include transforming growth factor beta-induced (TGFB)I, C-type lectin domain family 12 member A (CLEC12A), interferon-induced transmembrane protein 2 (IFITM2), formyl peptide receptor 3 (FPR3), S100A11, kynureninase (KYNU), integrin subunit alpha 4 (ITGA4) [[89]], monocyte derivation surface markers CD184 and CD354 [[20]]. There is significant variability across data sets [[84, 90]], and a (non-exhaustive) summary of differential key human glioma-associated macrophages and microglia marker expression is provided in Table 1.

4.1 Microlgia and macrophages retain a signature of origin and acquire niche-specific spatially-differentiated markers

Unsupervised clustering of single-cell transcriptomic profiles distinguishes brain-resident microglia and circulating monocytes/macrophages, and spatial transcriptomics confirms the enrichment of these subpopulations in distinct niches within each glioblastoma sample [[20, 44, 61, 82, 110]] (Table 2). In particular, glioma-associated microglia are enriched at the invading tumour margin [[20, 60, 61, 82, 83, 109]], whereas monocyte-derived macrophages are located preferentially at the tumour core (Fig. 1A,B). Here, dense cellular packing and poor vascularity result in low oxygen tension, HIF stabilisation and hypoxic/angiogenic TAM states associated with angiogenic growth factor secretion [[89, 111, 112]].

Although signatures of cell origin are retained, peripheral macrophages are nevertheless found to upregulate microglia-associated markers such us TMEM119 and P2RY12 on exposure to the brain microenvironment. Conversely, microglia can acquire monocyte/macrophage profiles in response to blood–brain barrier breakdown for example CD206 and CD163 [[16, 44]].

Single-cell profiling has confirmed and elaborated the core human microglia homeostatic gene expression signature – Namely, purinergic receptors P2RY12/13 and TMEM119/SELPG ‘sensosome’ genes previously involved in microglia process extension and sensing were detected via single-cell profiling. Additionally, cytokines/chemokines TGF-beta1, CX3CR1, CSF1R and CCR5 mediating cell recruitment and maintenance were also found in these studies [[83, 113]]. Based on microglia in the ‘normal’ brain (neurosurgical access cortex), Sankowski et al. performed gene ontology (GO) enrichment analysis based on cluster-enriched gene sets identifying clusters 2 and 3 expressing high levels of core microglia genes CX3CR1 and TMEM119. In contrast, clusters 2, 6 and 7 were characterised by the strong expression of major histocompatibility complex II (MHC-II) and antiviral immunity genes, such as HLA-DRA, CD74 and IFI44L [[83]]. C2 and C3 accounted for > 50% of microglia in their own sample set and previously published normal human datasets [[85]]. Clusters C6 and C7 expressed integrin receptor protein binding and metabolism genes, osteopontin (SPP1), apolipoprotein E (APOE) and low-density lipoprotein (LDL), and low levels of CX3CR1. ‘Pro-inflammatory’ clusters C1, C5, C8 and C9 expressed chemokines and cytokines including CCL2 and IL1beta. In comparison, glioma microglia were found to downregulate the microglia core signature in favour of the disease-associated microglia (DAM) signature, which is typical of ageing microglia and microglia in neurodegenerative disease [[113, 114]]. Microglia in pro-inflammatory clusters were characterised by upregulation of interferon signalling (IFI27, IFITM3), lipo and apo – lipoprotein processing (lipoprotein lipase (LPL), APOE), angiogenesis (VEGFA), immunosuppression and anti-inflammatory action (TREM 1/2) inflammatory, metabolic, interferon- and hypoxia-associated modules (HLA-DR, SPP1, TREM2, CD163, APOE, lipopolysaccharide (LPL, IFI27 and IFITM3)) [[108]]. The term ‘inflamm-ageing’ has been used to describe parallel responses in macrophage populations [[115]]. Microglia and macrophages enrich the invasive margins and the tumour core, respectively, retain a signature of origin and upregulate ‘disease-associated microglia’ and tumour-niche-specific markers.

4.2 TAM cell activation states in glioblastoma echo TAM subtypes identified in other cancers

Even before the widespread application of single-cell approaches, it was apparent that the M0/M1/M2 classification could not adequately capture the spectrum of in vivo functional states common to myeloid cells across diverse healthy and diseased tissues [[116]]. Accordingly, in human glioblastoma, individual TAMs typically express markers classically associated with resting ‘M0’ and activated ‘M1’ and ‘M2’ states in combination [[61, 103, 117]]. Indeed, of nine myeloid cell subtypes distinguished by unsupervised clustering of 83 479 glioblastoma TAM cells, none showed a clear enrichment for reference-based cell type gene meta modules corresponding to M1 or M2 states, including M2a/b/c/d subtypes [[60]]. Direct comparisons of glioma-derived and control microglia in mice likewise identify distinct tumour-associated states associated with unique marker expression profiles, falling outside the M1/M2 spectrum [[118]].

A recent classification has incorporated scRNA-seq and CITE-Seq profiling to identify six distinct TAM cell clusters [[20]]. Cluster 1 included ‘transitory monocyte-derived TAMs’ expressing monocyte genes EREG, S100A6 and lysozyme (LYZ) at high levels, with corresponding downregulation of mature macrophage markers C1QA and IGF1 [[20]]. Cluster 2 exhibited a phagocytic and lipid metabolism signature, including upregulation of Glycoprotein non-metastatic B (GPNMB), galectin 3 (LGALS3) and Fatty Acid Binding Protein 5 (FABP5). Cluster 3 exhibited a hypoxic and glycolytic gene expression profile, enriching for Bcl-2 interacting protein 3 (BNIP3), ADAM metallopeptidase domain 8 (ADAM8), macrophage inhibitory factor (MIF) and Solute Carrier Family (SLC) 2A1. Cluster 4 expressed microglia signature genes, such as CX3CR1, Bridging Integrator 1 (BIN1) and scinderin (SCIN). In contrast, cluster 5 downregulated these genes in favour of an anti-inflammatory signature, including selenoprotein P (SEPP1), SLC40A1, folate receptor beta (FOLR2), mannose receptor c-type 1 (MRC1) and ribonuclease 1 (RNASE1) expression. Cluster 6 was associated with interferon-gamma signatures, including CXCL10, CXCL11, and guanylate-binding protein 1 (GBP1). Using Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq) and extensive antibody panel broad protein validation, a ‘transitionary’ monocyte-to-macrophage subset was captured using CD36 and CD64. Monocyte-derived (Mo) TAMs captured included CD44/CD49d/CD9 versus CD69/CD151/TMEM119 for microglia-derived TAMs. SEPP1-high monocyte-derived cluster TAMs expressed folate receptor-β, CD206 and CD209; hypoxic Mo-TAMs CD184 and CD354; and phagocytic/lipid Mo-TAMs CD71 and CD72.

Although the microglia TAM subpopulations in the CNS are transcriptionally unique [[22, 41]], there is a strong correlation between glioblastoma myeloid cell subtypes and pan-cancer TAM subtypes [[22]]. This is an encouraging finding since incorporating data and clustering from glioblastoma studies into the wider landscape of TAM states across diverse cancers could enable us to generalise across tumours and accelerate treatment advances. Across a range of cancers, a series of consensus TAM subtypes have been identified by multiple groups: interferon-primed TAMs (IFN-TAMs), immune regulatory TAMs (Reg-TAMs), inflammatory cytokine-enriched TAMs (Inflam-TAMs), lipid-associated TAMs (LA-TAMs), pro-angiogenic TAMs (Angio-TAMs), RTM-like TAMs (RTM-TAMs), and proliferating TAMs (Prolif-TAMs) [[22, 41]]. Table 2 focuses on emerging glioblastoma myeloid cell functional states, ontogeny, and their distribution within this working classification. Sex-based differences have also been linked to better male immunotherapy responses [[67, 119]]. Seeking commonalities across pan-cancer datasets may be key to understanding this compartment and developing tissue-agnostic immunomodulatory treatments with selectivity for the TAM compartment over the broader myeloid population. This will be an important consideration to ensure safe application in patients.

Perhaps most promising in functional application, recent work from the Bernstein group has collapsed myeloid cell identity in glioblastoma into two dimensions, echoing a conceptually appealing and widely adopted classification of glioblastoma malignant cell identity [[44, 120]]. This classification distinguishes two inflammatory and two immunosuppressive myeloid cell subtypes in glioblastoma. The pan-inflammatory myeloid signature includes CCL3 and CCL4 expression, and the population is further subdivided according to the expression of ‘CXCR4 inflammatory’ (CXCR4, CXCL12, CX3CR1) and ‘IL-1B inflammatory’ (IL1B, IL1A, CC2, TNF, OSM and CXCL8) signatures. The pan-immunosuppressive signature includes expression of CD163, and ‘C1Q immunosuppressive’ (C1QA, C1QB, C1QC, CD16, C3, C2, VSIG4) and ‘Scavenger immunosuppressive’ (MRC1, MSR1, LYVE1, COLEC12, and STAB1) are distinguished. Although the cell of origin is not explicit within this classification, microglia and macrophages are shown to express predominantly inflammatory and immunosuppressive signatures, respectively, in keeping with previous reports (Table 2, Fig. 1B).

4.3 Malignant cell genetics and expression profile contribute to mutual glioma cell/TAM cell interactions mediating TAM chemotaxis and polarisation

Myeloid cell profiles differ between IDH mutant and IDH mutant WHO grade 4 gliomas at the bulk tumour level, with the former enriching for microglia in inflammatory states and the latter for macrophages in suppressive states. Interestingly, however, myeloid composition may be better predicted by histopathological tumour grade rather than by mutation profile [[44]]. Neurofibromatosis type 1 (NF1) mutation predicts the extent of TAM infiltration [[62]], and malignant cell transcription profiles may play a key role. Irrespective of mutation profile, glioblastoma cells converge and depend on conserved neural stem cell and glial progenitor transcriptional identities normally responsible for expansion and proliferation in embryonic development [[107, 120-123]]. These identities exist in dynamic equilibrium with each other, adapting to tissue and treatment context [[124]]. However, an additional ‘mesenchymal’ or ‘injury response’ signature can also be discerned, and this has no direct developmental correlation [[120, 125]].

Serial mouse glioma transplants in an immune-competent host expand mesenchymal malignant cell (MES-like) and TAM fractions, mimicking that seen in recurrent human disease [[126]]. This process, termed epigenetic immunoediting, was shown to reflect the outgrowth of those malignant cell clones which had succeeded in recruiting immunosuppressive TAMs and escaping adaptive immune clearance through the derepression of myeloid master regulator Irf8.

The resulting signature from epigenetic immunoediting includes the expression of chemokines and interferons responsible for TAM chemotaxis. Conversely, TAMs have been shown to drive glioblastoma cells to a MES-like state in vivo through the STAT3 activation downstream of receptor tyrosine kinase AXL and oncostatin M receptor (OSMR), targeted by Amphiregulin (AREG), Heparin-binding EGF-like growth factor (HBEGF) and OSM ligands expressed on macrophages. Accordingly, macrophage depletion therapy reduced MES-like malignant fraction [[127]].

New spatial transcriptomic approaches promise to add detail to the landscape of interactions between TAMs, tumour cells, and other components of the tumour microenvironment. Examples of established interactions include the osteopontin pathway (e.g., SPP1-CD44), which promotes TAM invasion, pro-tumour polarisation, and stem cell maintenance around blood vessels [[128, 129]]. Likewise, PTPRZ1 glioma cell/PTN myeloid cell interactions play a key role [[130]]. TAMs interact with the extracellular matrix (ECM) via tenascin C (TNC) through TAM CD74 receptor binding to MIF, coatomer-associated protein subunit alpha (COPA), or the Amyloid Beta Precursor Protein (APP) ligand [[60]]. These signalling networks may be key to establishing and sustaining the MES-like glioblastoma phenotype.

Tumour-associated macrophage interactions with the adaptive immune system are likely critical to function. For example, the NK cell receptor (KLRB1) encoding CD161 has been identified as an inhibitory receptor in myeloid cells, mediating reduced T-cell cytotoxicity and cytokine secretion [[131]]. Additional interactions of key potential significance have yet to be functionally validated. For example, Darmanis et al. have identified immune checkpoint ligand-receptor interactions between tumour and myeloid cells. These include but not limited to the programmed cell death Protein 1 (PD1) (CD274 [PDL1], programmed cell death one ligand 2 (PDCD1LG2) [PDL2]) and Cytotoxic T-Lymphocyte Antigen 4 (CTLA4) (CD80 and CD86), as well as the genes inducible T-cell costimulator ligand (ICOSLG) (ligand of ICOS receptor), CD276 (B7-H3), TNF receptor superfamily member 14 (TNFRSF14) (ligand of BTLA), and LGALS9 (ligand of TIM3) [[108]].

6.1 TAM compartment depletion

Tumour-associated macrophage depletion can be achieved by preventing recruitment and retention. Besides CSF1/CSF1R signalling, a blockade of several pathways mediating TAM chemotaxis was attempted. Kynurenine produced by glioma stem cells activates the aryl hydrocarbon receptor on macrophages, resulting in CCR2 upregulation, TAM recruitment and expression of CD39 and CD8⁺ T cell dysfunction [[11, 147]]. Inhibition of the CCL2/CCR2 axis reduces tumour infiltration by MDSCs, increases interferon production and stimulates T-cell responses [[148, 149]]. However, mixed effects of targeting this axis on tumour progression have been reported [[150-152]]. Likewise, VEGF/VEGFR [[144, 153]] and CXCL12/CXCR4 [[154, 155]] signalling axes have been targeted to reduce TAM recruitment. Intracellular targets, including p38 MAPK [[156]] and RNA demethylase ALKBH5 [[147, 157]], have also been targeted to reduce TAM recruitment.

Active ablation of the TAM compartment also presents treatment possibilities. The bisphosphonate clodronate delivered in a liposomal suspension [[158]] and trabectedin [[142]] are toxic to macrophages. The scavenger receptor MARCO has been shown to define immune-suppressive TAMs in glioblastoma and other cancers, and monoclonal antibodies targeting the MARCO⁺ TAM fraction in carcinomas can enhance tumour immunogenicity and slow growth [[159, 160]]. Likewise, TAM expression of the scavenger receptor TREM2 is a marker of poor prognosis. GBMs express the highest levels of TREM2 across a panel of cancers [[161]], and a candidate therapeutic monoclonal antibody, PY314, against TREM2, is available. Nevertheless, effectively delivering large molecules like monoclonal antibodies to the brain compartment remains challenging. Chimeric antigen receptor (CAR) T cells engineered to recognise TAM-specific surface antigens have also been used to ablate TAM recruitment [[162]].

6.2 TAM compartment re-education

Tumour-associated macrophage re-education has been attempted using diverse strategies in preclinical and clinical settings [[104]]. Induction of ‘immunogenic’ cancer cell death by conventional chemotherapeutics, such as doxorubicin, can drive TAM activation [[163]]. Likewise, existing clinically available immunotherapy can re-educate TAMs, for example, PD-L1 and PD-1 checkpoint inhibitors [[10, 164, 165]] and potentiate the T-cell cytotoxic effector response for this re-education. Checkpoint inhibitors have, therefore, served as the basis for several combination re-education therapy strategies. For example, the conversion of ATP to adenosine by CD73 is associated with the induction of immunosuppressive TAM states. CD73 shRNA knockdown confers survival benefits in combination with checkpoint inhibitors in a preclinical model [[166]]. An important caveat is the sex-based differences that may implicate success with checkpoint inhibition and TAM re-education. Differences in T-cell exhaustion in males versus females in glioblastoma [[167]] may warrant careful selection of candidates for immune therapies [[75, 168, 169]].

The PI3K/mTOR pathway has been targeted to prevent TAM accumulation and to achieve TAM re-education [[24, 170]]. Secretion of galectin-9, associated especially with glioblastoma-bearing PTEN mutations, drives immunosuppressive polarisation through binding of the Tim-3 receptor, a key immune checkpoint receptor expressed on TAMs. Checkpoint receptor blockade can impact macrophage polarisation and cytotoxic T-cell responses, reducing tumour growth in glioblastoma xenografts [[171]]. Chloroquine, galactan, and Toll-like receptor agonists have been reported to induce inflammatory myeloid states. At the same time, macrophage activating factors and antibodies directed at CD40 and IL-1a have also been the subject of trials to achieve macrophage re-education in solid tumours [[172]].

Cyclic GMP-AMP synthase (cGAS) and Stimulator of Interferon Genes (STING) are potent drivers of the innate immune response to pathogens and cancer alike [[173]]. Mechanistically, cGAS binds pathological cytoplasmic double-strand DNA associated with tumours and viruses and catalyses cyclic GMP-AMP (cGAMP) production. In response to this, cGAMP and related pathogen-associated cyclic dinucleotides bind STING, resulting in the recruitment and activation of Tank-binding kinase I (TBK1). TBK1, in turn, phosphorylates interferon regulatory factor 3 (IRF3), which translocates to the nucleus and drives type I interferon (IFN) transcription [[174, 175]]. The cGAS/STING/interferon axis is typically epigenetically silenced in glioma cells, perhaps enabling malignant transformation [[176]]. Therefore, targeting this axis in the tumour microenvironment, especially TAMs, may be key. The cGAS-STING/Interferon axis may also be a useful target in TAM-directed gene therapy, aiming to repolarise the TAM populations of the wider tumour microenvironment by forcing the expression of interferon in these cells, for example [[177]].

Oncolytic virus infection can simultaneously induce immunogenic malignant cell death and pro-inflammatory TAM phenotypes in glioblastoma and other cancers [[178]]. Therefore, brain-penetrant oncolytic viruses with appropriate tumour selectivity represent a compelling immunotherapy option [[179]]. However, treatment efficacy will depend on addressing the ability of activated TAMs to restrict virus propagation through the tumour [[109, 180]].

While these TAM re-education strategies collectively offer considerable promise, brain cancer is rarely the first choice for treatment development. This reflects the limitations in brain penetration and prediction for antibodies, viral vectors and other large molecules. A further concern relates to specificity and control of treatment response, especially because brain inflammation is very poorly tolerated within the fixed volume of the skull. Therefore, new approaches to achieving efficient and specific delivery to the brain tumour TAM compartment may prove key to delivering on this promise [[181]].

6.3 Gene editing of macrophages

Genetically engineered macrophages (GEMs) can repolarise the tumour microenvironment by expressing pro-inflammatory cytokines like IL-21 or by CRISPR knockout at anti-inflammatory loci like IL-10 or PD-L1 [[182]]. Proof-of-principle for this process is typically based on monocytes harvested from peripheral blood. However, editing TAMs in situ may ultimately prove possible and desirable, especially as gene delivery to microglia and brain TAMs improves [[183]]. CAR Macrophages (CAR-M) represent a special case engineered to recognise tumour-specific surface antigens. As for CAR-T cells, the engineered receptors incorporate antigen binding, transmembrane and intracellular signalling domains [[184]]. Although CAR-Ms are typically generated by in vitro differentiation or ex vivo modification of peripheral blood monocytes, CAR-Ms can also be generated in situ by delivering CAR transgenes to TAMs in the wall of the tumour cavity [[185]]. CAR-M can phagocytose tumour cells expressing the target antigen and drive adaptive anti-tumor responses by releasing pro-inflammatory cytokines and recruitment of cytotoxic effectors, including CD8⁺ T cells [[186]]. Compared to CAR-T or CAR- Natural Killer (NK) cells, CAR-M may infiltrate glioblastoma more effectively [[187]], exploiting the same chemotaxis mechanisms responsible for TAM accumulation. However, optimal surface antigens for CAR-M targeting remain to be determined. Glioma stem cells typically express CD133 and/or other ‘neural stem cell’ surface markers [[181, 188]], and the mutant constitutively active EGFRviii receptor is present in many tumours. However, targeting single surface epitopes typically results in the selection and outgrowth of resistant/downregulated clones without a survival benefit [[189]]. Intravenous infusion is associated with CAR-M accumulation in liver tissue, impairing treatment effect [[190]]. CAR-M may not persist for any prolonged duration following infusion and does not increase effectively in vivo. Therefore, effective treatment may depend on repeated CAR-M dosing. A few clinical trials are underway, but the extent of possible CAR-M toxicity remains unclear. Additionally, some challenges need to be solved for CAR-M treatment against glioblastoma. Firstly, the CAR-M may not exist in patients' bodies for long, even after repeated infusions, which may further limit the effectiveness of the treatment. The second is related to the delivery methods. Lastly, due to the heterogeneity of glioblastoma tissue, the absence of an ideal tumour-specific antigen represents a bottleneck for CAR-M application, which is also the plight of CAR-T therapy against glioblastoma.

6.4 Blocking signalling in the TAM compartment

Tumour-associated macrophage pathway blockade can target specific cell functions and individual interactions mediated through the TAM cell ‘surfactome’ and ‘secretome’. TAMs can promote malignant cell proliferation and invasion in glioma by producing growth factors and signalling moieties, including EGF, TGF-beta, and stress-inducible protein 1 [[27, 191, 192]]. For example, blocking a key TAM-expressed ‘don't eat me’ signal, such as Clever1, has been associated with reactivating anti-tumour cytotoxic effector cell responses [[193]]. Likewise, SIRP1a expressed on TAMs recognises the CD47 antigen, a ‘don't eat me’ signal, upregulated in cancer cells, inhibiting phagocytosis [[194, 195]]. CD47 antibody blockade, in combination with alkylating chemotherapy, has proved effective in murine preclinical models [[196]]. Blockade of TAM pro-angiogenic activity has been most extensively studied in the context of VEGF/VEGFR signalling. While VEGF-directed monotherapy using bevacizumab has been associated with minimal survival benefit in newly diagnosed glioblastoma [[197]], combined targeting of VEGF and TIE2 angiogenic growth factor signalling pathways is a promising strategy for targeting the angiogenic TAM state for treatment benefit [[198-200]].