The immune landscape of common CNS malignancies: implications for immunotherapy

Abstract

Immunotherapy has enabled remarkable therapeutic responses across cancers of various lineages, albeit with some notable exceptions such as glioblastoma. Several previous misconceptions, which have impaired progress in the past, including the presence and role of the blood–brain barrier and a lack of lymphatic drainage, have been refuted. Nonetheless, a subset of patients with brain metastases but, paradoxically, not the vast majority of those with gliomas are able to respond to immune-checkpoint inhibitors. Immune profiling of samples obtained from patients with central nervous system malignancies using techniques such as mass cytometry and single-cell sequencing along with experimental data from genetically engineered mouse models have revealed fundamental differences in immune composition and immunobiology that not only explain the differences in responsiveness to these agents but also lay the foundations for immunotherapeutic strategies that are applicable to gliomas. Herein, we review the emerging data on the differences in immune cell composition, function and interactions within central nervous system tumours and provide guidance on the development of novel immunotherapies for these historically difficult-to-treat cancers.

Gliomas are the most common primary tumours of the central nervous system (CNS) and have historically been stratified into grades 1–4 based on the histological features defined by the WHO¹. This classification system has transitioned to one based on molecular features² to align with the availability of targeted therapeutic approaches and has been shown to have a higher level of prognostic accuracy^2-4. A key molecular determinant of the prognosis of patients with gliomas is the presence of mutant forms of isocitrate dehydrogenase 1 (IDH1), which is associated with a more favourable prognosis and is frequently expressed in low-grade gliomas but is only detected in approximately 5–10% of WHO grade 4 diffuse astrocytomas. Clinically relevant molecular subsets of gliomas based on the IDH mutation and 1p/19q codeletion status can also be inferred from DNA methylation profiles⁵.

Glioblastoma, IDH-wild-type is the most common form of malignant glioma, comprising 57% of all CNS gliomas. These glioblastomas typically arise spontaneously and can be further subtyped into either proneural, mesenchymal or classical based on transcriptional signatures⁶. For example, data obtained using RNA deconvolution algorithms such as CIBERSORT indicate that the mesenchymal subtype is characterized by reduced resting CD4⁺ T cell numbers with augmented populations of immunosuppressive M2 macrophages. A prior analysis of data provided by The Cancer Genome Atlas revealed that the mesenchymal subtype is paradoxically enriched for genes associated with both antitumour pro-inflammatory responses and immunosuppression⁷, suggesting that this subtype is generally more immunologically responsive, although this hypothesis has never been formally tested or retrospectively evaluated. The classical subtype is associated with higher levels of EGFR expression, indicating a modest selection bias in clinical trials in which the expression of EGFR variant III (EGFRvIII) is used as a selection or stratification criterion. Notably, these subtypes are in a continuum and can transition to another subtype following disease recurrence and should therefore not be used for trial enrolment and/or stratification purposes; although they have been used to retrospectively evaluate responses to bevacizumab⁸.

Glioblastoma, IDH-wild-type, WHO grade 4, afflicts 12,000 patients annually in the USA and has a 5-year relative survival rate of 6.8%⁹. Even with standard-of-care treatment, consisting of maximal safe surgical resection, radiotherapy and temozolomide, the median overall survival (OS) duration is approximately 16 months¹⁰. Glioblastomas with an O6-methylguanine-DNA methyltransferase (MGMT) methylation are more likely to respond to temozolomide¹¹. Patients with newly diagnosed MGMT-unmethylated glioblastomas do not usually benefit from temozolomide; therefore, many oncologists advocate for treatment with investigative agents in this subset of patients (NCT02977780)^12-17. The vast majority of preclinical research and clinical trials have been focused on glioblastoma. This focus in part reflects the rapid, unrelenting disease progression and unmet medical needs of patients with glioblastoma but also the higher incidence relative to that of other gliomas. Alongside the newly emerging molecular reclassification of gliomas, the comprehensive characterization of the unique immune biology of CNS gliomas will fundamentally reshape both immunotherapy treatment strategies and the understanding of how to identify the most suitable patients for each approach. In this Perspective, we discuss the underlying reasons why immunotherapies that have shown marked responses in patients with other types of malignancies thus far have not provided similar levels of benefit to patients with glioma.

CNS malignancies versus brain metastases

Biomarkers of response to immune-checkpoint inhibition.

Immunotherapies, such as immune-checkpoint inhibitors (ICIs), have had a profound positive effect on the survival and quality of life outcomes of patients with certain forms of cancer. This is especially the case for those with melanomas or non-small-cell lung cancers (NSCLCs), including those with CNS metastases^18,19, for which several of these agents have been approved by the FDA. By contrast, large-cohort phase III trials of anti-PD-1 antibodies in patients with glioblastoma have yielded only negative results²⁰. In CheckMate-498, approximately 550 patients with primary MGMT-unmethylated glioblastoma were randomly assigned to either nivolumab or temozolomide plus radiotherapy with a primary endpoint of OS; although nivolumab was well tolerated, the primary endpoint was not met²¹. Similarly, CheckMate-548 was designed to explore the efficacy of nivolumab in addition to standard-of-care temozolomide and radiotherapy in patients with MGMT-methylated glioblastomas but failed to show an improvement in progression-free survival (PFS) with the addition of nivolumab^22,23. This trial was ceased owing to futility following the randomization of a total of 693 patients²⁴. The concurrent use of a chemotherapy such as temozolomide with another form of immunotherapy (an EGFRvIII-targeted peptide vaccine) can induce lymphopenia and inhibit T cell activation, although its use has been shown to promote antitumour immunity in a phase II trial²⁵. The sustained use of temozolomide might also be associated with hypermutation^26-28 and downregulation of the mismatch repair (MMR) pathway²⁹, thereby potentially enhancing tumour immunogenicity. The lack of additive or synergistic effects with the use of nivolumab in addition to standard-of-care temozolomide in CheckMate-548 might therefore have been a function of time and/or reflect that the induced alterations are not immunogenic. We highlight that important immunological differences exist in how hypermutation arises in patients with glioblastoma relative to the same effects in those with smoking-induced NSCLCs or UV-induced melanomas. Hypermutations in glioblastoma usually arise late after extended use of chemotherapy, whereas hypermutated NSCLCs typically arise prior to treatment. Tumour mutational burden (TMB)-high tumours with a large percentage of clonal mutations might be more responsive to ICIs relative to TMB-high tumours induced by chemotherapy, which are typically subclonal^30,31.

Clinical data from glioblastoma and most types of solid tumours indicate that ICIs only provide benefit to a subset of patients and/or when administered in a specific treatment setting³². For example, in a randomized phase II multicentre study, 35 patients with recurrent surgically resectable glioblastomas who were randomized to receive neoadjuvant pembrolizumab with continued adjuvant therapy following surgery had a median OS duration of 14 months, which was a notable improvement relative to those receiving only adjuvant pembrolizumab following surgery (median OS 7.6 months)³². Neoadjuvant PD-1 blockade was associated with the tumour-specific upregulation of T cell and interferon signalling-related genes in the tumour, which was not seen in patients that received adjuvant therapy only³². Others have pointed out that intergroup imbalances in certain patient characteristics, such as MGMT-promoter methylation, should theoretically have favoured the adjuvant group³³. We acknowledge that this study has several limitations, such as the small number of patients enrolled and the use of OS as only a secondary endpoint, yet the findings prompt speculation as to whether surgery might facilitate an antitumour immune response. A possible hypothesis is that the neoadjuvant setting enables sufficient antigenic stimulation during the initial priming event, which then becomes at least partially unencumbered with the first dose of an anti-PD-1 antibody. Thereafter, the trauma of surgery might trigger a stimulator of interferon genes (STING)-mediated type 1 interferon response³⁴ and also remove other redundant mechanisms of tumour-mediated immunosuppression³⁵. The cyclic GMP–AMP synthase (cGAS)–STING signalling pathway is activated by cytosolic DNA from various sources. Binding with double-stranded DNA activates the DNA sensor cGAS to generate the second messenger cyclic GMP–AMP (cGAMP). cGAMP then binds to the endoplasmic reticulum-localized adaptor protein STING, which triggers the phosphorylation of IRF3 via TBK1. The phosphorylated IRF3 then translocates to the nucleus to trigger the transcription of genes encoding inflammation-related proteins³⁶. For patients in the adjuvant pembrolizumab group (who received pembrolizumab after surgery only), the activation of cGAS–STING signalling after surgery is unlikely to promote the development of protective immunological memory owing to a lack of tumour-associated antigens³⁷. If this was the case, we would speculate that the additional adjuvant doses of anti-PD-1 antibodies post-surgery might not be necessary for the neoadjuvant strategy to be effective, although this remains to be tested. An alternative explanation as to why neoadjuvant anti-PD-1 antibodies might promote antitumour immunity is that ICIs in the presence of tumour-associated antigens (before surgery) induce IFNγ-dependent responses in the tumour and facilitate the early systemic activation following surgical resection that leads to the reduced immunosuppression that is then amplified by adjuvant treatment. Data from murine models are consistent with both theories³⁸.

Retrospective longitudinal profiling of samples from 222 patients with glioma indicate that standard-of-care alkylating agents promote the emergence of a hypermutator phenotype, albeit at different rates across the glioma subtypes, and that this phenotype is not associated with differences in OS³⁹. From this perspective, these data suggest that patients with disease recurrence might be more likely to respond to ICIs and that stratification criteria, such as TMB, could be considered as a means of selecting patients who are most likely to respond. In June 2020, the FDA granted pan-cancer approval of pembrolizumab for patients with TMB-high (≥10 mutations per megabase (mut/mb)) solid tumours, including CNS gliomas⁴⁰. However, data from two compelling analyses strongly suggest that a TMB-high status will not be an appropriate biomarker in this setting. In a study by Gromeier et al.⁴¹, a retrospective genomic analysis of tumour tissue samples from patients with recurrent WHO grade 4 glioblastoma acquired prior to receiving immunotherapy found that a very low TMB (≤1.3 mut/mb) is associated with a longer OS after recombinant polio virotherapy or an ICI. Specifically, the median OS durations of patients with glioblastoma with a low versus a high TMB receiving the virotherapy or an ICI were 17.1 months versus 10.7 months and 15 months versus 10 months, respectively⁴¹. Furthermore, transcriptomic analyses revealed an inverse relationship between TMB and the enrichment of inflammatory gene signatures in patients with recurrent disease but not in those with newly diagnosed glioblastomas, implying that the relationship between TMB and tumour-intrinsic inflammation evolves upon disease recurrence. In a second retrospective analysis by McGrail et al.⁴², samples from a total of 1,551 patients were analysed, including from 207 patients with gliomas based on cancer cell lineage who received ICIs owing to a TMB-high status (≥10 mut/mb). In cancer types such as melanoma, NSCLC and bladder cancers, in which CD8⁺ T cell counts are positively correlated with neoantigen load, TMB-high tumours (≥10 mut/mb) were found to have a 39.8% overall response rate (ORR) to ICIs, which was significantly higher than that observed in TMB-low tumours (≤10 mut/mb) (OR 4.1)⁴². By contrast, in cancer types such as breast cancer, prostate cancer and gliomas, in which CD8⁺ T cell counts and neoantigen load are unrelated to responsiveness to ICIs, ORRs <20% were observed in patients with TMB-high tumours, which were substantially lower than those of patients with TMB-low tumours. These findings again emphasize the fundamental differences in the immune biology of different cancer lineages.

Other biomarkers of response to ICIs include T cell infiltration, PD-L1 expression, DNA mismatch repair defects (dMMR) and microsatellite instability. Comprehensive genomic profiling as well as a series of studies conducted by our group indicate that these factors or biomarkers, which are typically associated with a favourable prognosis, are not commonly detected in glioblastomas. More specifically, <1% of glioblastomas have microsatellite instability⁴³, <10% have a high TMB, PD-L1 expression or dMMR, and only about 30% have any appreciable level of T cell infiltration^44-46. The increased number of neoantigens in hypermutated cancers with dMMR deficiencies is thought to be a key contributor to therapeutic responses to anti-PD-1 antibodies. Data from December 2020 demonstrate that tumour cells that lack the dMMR gene Mlh1 markedly accumulate cytosolic DNA and produce interferon in a cGAS–STING-dependent manner, which renders tumours harbouring this alteration slowly progressive and highly sensitive to ICIs in mouse models^47,48. Unfortunately, this deficiency is not typically detected in gliomas when assessed in large omics datasets such as The Cancer Genome Atlas. By contrast, CNS brain metastases from tumours originally located outside of the CNS have much higher levels of several of these response biomarkers⁴⁹ and, in some cases (such as NSCLC brain metastases), these metastases are preferentially enriched with one or more of these biomarkers relative to the primary tumour⁵⁰. In contrast to clinical trials in which patients with gliomas received ICIs, which revealed a lack of therapeutic benefit for the vast majority of patients, those with untreated brain metastases from either melanoma or NSCLC who received pembrolizumab had an ORR of 33%, of which almost all responses were durable⁵¹. Furthermore, 57% of patients with melanoma and CNS metastases who received nivolumab plus ipilimumab had evidence of intracranial clinical benefit, including a complete response rate of 26% and a partial response rate of 30%¹⁸. These data indicate that a lack of response to ICIs is not solely a function of a tumour residing within the CNS but also reflects differences in immune biology between cancer lineages. Differences in the extent of de novo T cell infiltration between brain metastases and gliomas are another point to consider. Patients with brain metastases receiving ICIs typically have tumours located outside of the brain before the onset of CNS metastases and are therefore more likely to have adequate T cell priming, which can be subsequently boosted by the administration of an ICI. Thus, for systemic cancers, immunological memory responses that are more robust than the de novo immune responses occasionally seen in patients with glioma might exist owing to T cell priming before the administration of an ICI, although such responses alone might be insufficient to prevent disease progression. Many malignancies, and especially glioblastomas, have substantial redundancy in terms of the possible tumour-mediated mechanisms of immunosuppression, including those beyond the upregulation of immune checkpoints and/or ligands, which also contribute to the refractory and recalcitrant nature of these tumours³³. Alternatively, the immune cell populations of most glioblastomas might simply be refractory to reinvigoration⁵².

In another analysis from a clinical trial in which patients with glioblastoma received neoadjuvant pembrolizumab or nivolumab, Zhao et al.⁵³ investigated samples from 66 patients with glioblastoma, including 17 long-term responders, and found that non-responders had tumours that were enriched for mutations in PTEN and immunosuppressive gene expression signatures whereas responders had enrichment for alterations in genes involved in the MAPK pathway (such as PTPN11 and BRAF)⁵⁴. In our opinion, given the genetic heterogeneity of glioblastoma and the complexities involved in generating an antitumour immune response, a single parameter is unlikely to be a robust predictor of responsiveness but might be an essential component of a more complex, integrated biomarker signature or some form of poly-ligand profiling⁵⁵.

Differences in tumour immune composition.

Gaining more detailed knowledge of the similarities and differences among the immune cell populations of gliomas and brain metastases and how their interactions contribute to the brain tumour microenvironment (TME) is an important area of current research and is relevant to understanding the differences in the responses to ICIs as well as for the development of more suitable immunotherapies for patients with gliomas. Studies published in June 2020 (REFS^56,57) provide new insights into the immune cell landscape of CNS tumours. Through the use of fluorescence-activated cell sorting followed by RNA sequencing or high-dimensional single-cell profiling, these authors^56,57 collectively revealed differences in the immune composition of gliomas (depending on their IDH mutation status) and brain metastases, with a focus on the heterogeneity and abundance of tissue-resident and peripherally infiltrating leucocytes. These data demonstrate a correlation between cancer lineage and immune landscape that might be sufficient to provide a clear distinction between gliomas and brain metastases and suggest that the TME is shaped by the diseases themselves and not by the surrounding brain tissue. IDH-mutant gliomas were found to be highly enriched with tumour-resident microglia, albeit with a virtual absence of lymphocytes. By contrast, IDH-wild-type gliomas and brain metastases from tumours of various primary origins were both highly infiltrated with peripheral monocyte-derived macrophages (MDMs). Brain metastasis also had substantial levels of T cell infiltration, including of both activated and exhausted T cells, and neutrophils. Gliomas, especially those of an IDH-mutant subtype, harboured a substantially lower number of infiltrating T cells and these T cells were not sufficiently activated. Regulatory T cells were more frequent in brain metastases relative to gliomas, with the lowest frequency of such cells observed in patients with the IDH-mutant subtype. These differences in both the abundance of T cells and their activation status are likely one of the contributors to the differences in responses to ICIs seen between patients with brain metastases and those with gliomas.

In the past, macrophages have been classified as either tumour supportive (M2-like) or tumour suppressive (M1-like); this classification has been extensively described elsewhere^58,59. However, transcriptional analyses of brain-resident microglia and monocyte-derived macrophages published in 2020 (REFS^56,57) have uncovered complex and multifaceted phenotypes across the various brain malignancies that do not reflect the classical subdivision into M2-like or M1-like phenotypes. For example, a bulk profiling alignment analysis of the innate immune cells derived directly from human glioblastoma samples demonstrated that the vast majority of these cells do not align with the classical phenotypes⁶⁰. Rather, the different myeloid populations have a distinct transcriptional signature that largely depends on the origins of the tumour (glioma versus brain metastases) and the glioma subtype (IDH-wild-type versus IDH-mutant) than on the local TME. Additional studies^61,62 have focused exclusively on myeloid cell populations, further exploring the heterogeneity and plasticity of microglia and macrophages in the glioma TME. Muller et al.⁶¹ used single-cell RNA sequencing to compare myeloid cells obtained from glioma biopsy samples with those obtained from non-malignant brain tissues and identified an MDM-specific gene signature, suggesting that these cells had extravasated through the blood–brain barrier and infiltrated the tumour. Furthermore, these investigators demonstrated that MDMs have elevated levels of markers of immunosuppression, such as IL-10 and TGFβ2, relative to glioma-resident microglia and therefore imply the potential therapeutic benefit of specifically targeting MDMs, including preventing their migration into the brain tumour, rather than aiming to non-selectively target all myeloid subpopulations. By contrast, Sankowski et al.⁶² identified microglia as potential therapeutic targets through combined assessments of gene and protein expression using single-cell RNA sequencing and cytometry by time of flight (CyTOF) to identify distinct microglial states in brain tissue samples from patients who underwent surgery for non-malignant conditions, such as epilepsy, and from those with glioma. In total, 14 distinguishable microglia subsets were identified, of which 2 were almost exclusive to glioma. In general, the subsets were determined by their location, patient age and brain tumour pathology. Furthermore, different subsets of microglia were associated with distinct functional gene expression signatures, such as antigen processing and presentation, regulation of leucocyte chemotaxis, cytokine production, hypoxia, and phagocytosis, indicating several diverse functional states. The glioma-specific cluster showed enrichment for genes associated with functional signatures such as hypoxia and antigen processing via MHC class I.

Data reported as a meeting abstract⁶³ describing the single-cell transcriptional profiles of immune cells from both primary and recurrent IDH-mutant and IDH-wild-type gliomas support these findings. An unbiased clustering analysis revealed 62 distinct transcriptional immune cell states, including 17 microglia subsets. Immune-related gene ontology analysis identified, in line with data from Sankowski et al.⁶², distinct functional signatures, including considerable differences in antigen presentation, phagocytosis, angiogenesis, hypoxia, and both pro-inflammatory and anti-inflammatory functions. The observation that certain microglial cell clusters have gene expression signatures associated with antigen presentation and cell lysis suggests roles as ‘professional’ antigen-presenting cells and as direct mediators of antitumour activity, although these findings will require functional validation. Overall, and although not currently available in a peer-reviewed format, the single-cell transcriptomics study⁶³ identified associations between distinct immune cell states and IDH-wild-type status, implicating the enrichment of distinct immune cell functions such as antigen presentation. Future functional studies are needed to validate these observations. Taken together, these studies demonstrate not only the important role of myeloid cell populations in the glioma TME but also point to their diversity and, accordingly, to the importance of future subset-specific therapeutic interventions.

Differences in immune reactivity.

Immune cell composition differs not only between cancers of specific lineages but also according to spatial location. Using the Ivy Glioblastoma Atlas, we found differences in the spatial distribution of the various immune cell populations when comparing the immune cell compositions of glioblastomas from seven different anatomical locations (FIG. 1). T cell-specific (CD3, CD4, CD8) and B cell-specific (CD19, CD20) marker expression is generally low at the leading edge and in the tumour-infiltrating and tumour areas, including the necrotic zones, but is enriched in the vascular areas of glioblastomas. Assessing the presence of myeloid cell populations using the myeloid marker CD33 or the more specific microglia and macrophage marker CD68 revealed a relatively uniform distribution throughout the TME. When examining the distribution of more-specific markers, a clear regional enrichment of microglia-specific markers (CX3CR1, Tmem119, P2RY12) in areas of the leading edge and the infiltrating tumour is apparent, whereas MDMs are most abundant in hyperplastic blood vessels and areas of microvascular proliferation. Furthermore, the scavenger receptor CD163, which is often associated with immunosuppressive myeloid cells, is much more prominent in vascular areas, indicating that MDMs rather than microglia are the main contributor to the heavily immunosuppressive glioblastoma TME. Nonetheless, the use of CX3CR1 as a microglia-specific marker can be confounded by the upregulation of this marker in MDMs under certain conditions⁶⁴. Many chemokines are equally distributed throughout the glioblastoma TME, albeit with noteworthy exceptions such as CCL2 (MCP1), CCL3, CCL7, CCL8, CCL17, CXCL10 and CSF-1/2 — all of which are enriched in the peripheral zone (FIG. 2). Interestingly, the expression of cGAS and STING is increased in the vascular-related areas (hyperplastic blood vessels and microvascular proliferation) and these are the only areas in which elevated T cell numbers could be detected. In addition to differences in the anatomical locations of various immune cell populations in the TME, immune cell interactions are also likely to differ between cancers of different lineages. For example, MDMs with the ability to present antigens have been shown to exist in close proximity to T cells in brain metastases but not in gliomas⁶⁵. We have recently shown that an antigen-presenting cell (APC)–T cell interaction is a requirement for a fully functional antitumour immune response in the CNS⁶⁶. Thus far, to the best of our knowledge, APC–T cell interactions have not been investigated as a biomarker of responsiveness to ICIs. Spatial data, which provide this crucial information on the location and interactions between immune cell types, are likely to be informative of responsiveness to immunotherapy but are not typically provided by deconstructive strategies, such as flow cytometry, that involve dividing the various TME components into single-cell suspensions.

Fig. 1 ∣. Analysis of immune cell populations in various anatomical locations within glioblastomas.

RNA sequencing data from the Ivy Glioblastoma Atlas project was analysed based on differences in the anatomical locations of the primary tumour. Expression of various immune cell lineage markers according to spatial location. CT, cellular tumour; HPV, hyperplastic blood vessels in cellular tumour; IT, infiltrating tumour; LE, leading edge; MDM, myeloid-derived monocyte/macrophage; MP, microvascular proliferation; PS, pseudopalisading cells around necrosis; PZ, perinecrotic zone.

Fig. 2 ∣. Analysis of chemokine expression in various anatomical locations within glioblastomas.

RNA sequencing data from the Ivy Glioblastoma Atlas project was analysed based on differences in the anatomical locations of the primary tumour. Expression of molecules involved in immune cell attraction to the tumour. CT, cellular tumour; HPV, hyperplastic blood vessels in cellular tumour; IT, infiltrating tumour; LE, leading edge; MP, microvascular proliferation; PS, pseudopalisading cells around necrosis; PZ, perinecrotic zone.

Until now, most studies have focused on the role of immune responses within the tumour. Based on a series of observations made over several years, our research group has begun to focus on a more detailed evaluation of antitumour immune responses at the tumour–CNS interface, which likely differ from those occurring in the tumour mass itself. This discrepancy might misinform the scientific community regarding biomarkers of potential response and lead to key pathways and mechanisms that are important for antitumor immune surveillance and eradication being overlooked. In cancers of the CNS, the adjacent brain tissue might also be mechanically or molecularly triggered, thereby upregulating the expression of fractalkines, which could then act as CNS-specific chemokines for CX3CR1⁺ cells such as microglia, macrophages and monocytes. In a preliminary analysis of different tumour regions spanning from the adjacent brain interface to the necrotic core, investigators used multiplexed staining to demonstrate the existence of more APC–T cell cluster interactions in brain metastases relative to gliomas and especially at the tumour–brain interface⁶⁵. These data indicate that lineage-specific interactions with the CNS are able to influence immune reactivity.

Enrichment of immunosuppressive mechanisms.

An extensive portfolio of immunotherapeutic modulators are currently available such as inhibitors targeting PD-1 or PD-L1, CTLA4, LAG3, TIM3, or various different steps of the adenosine signalling pathway; however, whether such specific immune mechanisms are operational in gliomas, which are known to have several heterogeneous and redundant mechanisms of immune suppression, remains unclear. In many clinical trials involving patients with glioma, the justification for testing a given agent has been extrapolated from other malignancies without sufficient consideration as to whether the target is frequently expressed in this patient population. A study with results published in September 2020 aimed to prioritize the various available immunotherapies by profiling the immune cell landscape of gliomas of various grades and molecular subtypes⁶⁷. Immune profiling was conducted on T cells and monocytes and/or macrophages isolated from the blood and TME of patients with glioma and compared to samples from individuals without glioma. These data showed that some immune checkpoints, such as TIM3 and LAG3, are not widely expressed and agents targeting these checkpoints are therefore not expected to be therapeutically beneficial for the majority of patients with gliomas. The most frequently expressed candidate that emerged across all grades, histologies and immune subsets was the immunosuppressive adenosine (A2a) receptor followed by PD-1 (TABLE 1).

Table 1 ∣.

Summary of immune cell features and location between gliomas and brain metastases

Immune cell population	CNS malignancy of enrichment	Immunotherapeutic target prioritization in gliomas	TME localization
			Gliomas	Solid tumour CNS metastases
MDMs	IDH-wild-type gliomas	A2aR > CD39 > B7-Hx > PD-L1, CD73 (REFS67,69), PD-1 (REFS123,124), CSF1R79, CD47–SIRPα86-88, LILRB family86,88,125-128, PD-L1 (REFS129,130), pSTAT3 (REFS103-105,110), STING91, VISTA69	CD163+ gradient, necrotic core, perivascular	CD163+ gradient
Microglia	IDH-mutant glioma	PD-1 (REF.124), CD47–SIRPα88,131, pSTAT3 (REFS103,105,110), PD-L1, VISTA132	Distributed throughout the tumour57	Confined to tumour border and mostly absent from the tumour core57
Dendritic cells	Metastases	CD47–SIRPα133-135, PD-L1 (REFS136-138), pSTAT3 (REFS106,107), LILRB4 (REF.139)	Minimal	Stroma, brain–tumour interface
NK and NK T cells	Equivalent	KIR140,141, CD94/NKG2A142,143, TIGIT144-147, CD96 (REFS144,148,149), LAG3 (REF.150), TIM3 (REFS151-153), PD-1 (REFS154-156), A2aR157	Nanostring indicates a diffuse pattern	Nanostring indicates a diffuse pattern
T cells	Metastases	A2aR > PD-1 > CD39 > TIGIT > LAG3 > CTLA4 > BTLA > CD160 > CD73 > KIR > TIM3 (REF.67), CD96 (REF.158), pSTAT3 (REFS67,107-109)	Brain–tumour interface, perivascular	Brain–tumour interface, stroma
Neutrophils	Metastases	Unknown	Nanostring indicates a diffuse pattern	Nanostring indicates a diffuse pattern

A2aR, adenosine receptor 2A; B7-Hx, members of the B7 homologue family, including but not limited to B7-H1–4; CNS, central nervous system; CSF1R, macrophage colony-stimulating factor 1 receptor; LAG3, lymphocyte activation gene 3 protein; KIR, killer cell immunoglobulin-like receptor; LILRB, leucocyte immunoglobulin-like receptor; MDMs, myeloid-derived monocytes/macrophages; NK, natural killer; NKG2A, NKG2A/NKG2-B type II integral membrane protein; pSTAT3, phosphorylated signal transducer and activator of transcription 3; SIRPα, tyrosine-protein phosphatase non-receptor type substrate 1; STING, stimulator of interferon genes; TIGIT, T cell immunoreceptor with Ig and ITIM domains; TIM3, T cell immunoglobulin and mucin domain-containing protein 3; TME, tumour microenvironment; VISTA, V-type immunoglobulin domain-containing suppressor of T cell activation.

The availability of adenosine, the main physiological ligand of the A2a receptor, is regulated by two ectonucleotidases (CD39 and CD73) that can be expressed on a variety of cells, including on different immune cells, such as T cells and macrophages, as well as on cancer cells. In gliomas, the expression of CD39 and CD73 is induced by cellular stress and hypoxia, resulting in the conversion of extracellular ATP to adenosine, which can then bind with the A2a receptor. Similar to CD39 and CD73, A2a receptors are expressed on a variety of immune cells, such as macrophages, T cells, natural killer (NK) cells and dendritic cells, and these receptors induce a variety of immunosuppressive signalling pathways upon ligand activation⁶⁸. Indeed, A2a receptor signalling might be particularly relevant to gliomas given that pan-cancer profiling using CyTOF analysis demonstrates that CD73 is more commonly expressed in the immune cells of patients with gliomas⁶⁹. Immune cell CD73 expression is upregulated by 2-hydroxyglutarate in patients with IDH1-mutant gliomas⁶⁷. Furthermore, data from multiple mouse glioma models, including those that express CD73, indicate that adenosine receptor inhibitors have a modest therapeutic effect and that the addition of other adenosine pathway inhibitors did not further enhance this effect⁶⁷. Although inhibitors of the A2a receptor could recover the immunological effector functions of T cells exposed to standard adenosine receptor agonists, irreversible exhaustion limited the effectiveness of inhibitors of the adenosine pathway when the T cells were co-cultured with glioma cells⁶⁷.

Immune cell exhaustion might not be reversible in glioblastoma.

The immune exhaustion phenotype of tumour-infiltrating T cells from patients with glioma might be exceptionally refractory to restoration with ICIs as has been previously documented⁵². Glioblastomas elicit a particularly severe T cell exhaustion signature that is characterized by the prominent upregulation of multiple immune checkpoints that are generally refractory to further immune modulation⁵². Several different states of immune cell dysfunction have been described, whereby T cell exhaustion in particular, alongside T cell anergy, is frequently linked with tumour-induced immunosuppression. T cell exhaustion is caused by the upregulation of multiple immune checkpoints under repeated antigen exposure as opposed to anergy, which is induced by sustained low levels of antigens and insufficient co-stimulation⁷⁰. Analysing in parallel both glioma-infiltrating and peripheral blood lymphocytes isolated from patients, T cell exhaustion characterized by the upregulation of multiple immune checkpoints and T cell unresponsiveness was found to be more prominent in glioblastoma-infiltrating T cells⁵². Elsewhere, CD8⁺ glioblastoma-infiltrating T cells have been shown to have an increased expression of PD-1 and CTLA4 compared with peripheral blood CD8⁺ T cells⁷¹. PD-1⁺ T cells have also been characterized as being more terminally differentiated as reflected by an Eomes^hiT-bet^lo phenotype, which is inversely correlated with proliferation. The combination of antibodies targeting CTLA4 and PD-1 can induce CD8⁺ T cell proliferation in patients with low percentages of Eomes^hiT-bet^lo CD8⁺ T cells in the glioblastoma TME but not in those with high percentages of such cells among the glioblastoma-infiltrating T cell population. The former group was found to be more likely to respond to anti-PD-1 antibodies ex vivo, leading investigators to conclude that the differentiation status of CD8⁺ glioblastoma-infiltrating lymphocytes determines their ability to be reinvigorated by ICIs⁷¹. Whether the various clinically approved immunomodulatory agents are capable of reversing T cell exhaustion remains unclear; therefore, investigating immune cell function in this context might enable the selection of agents that are most likely to be clinically relevant.

Irreversible immune cell exhaustion and dysfunction might not solely be a function of antitumour T cell responses. Using an experimental approach that enabled head-to-head comparisons of biomarker expression at the single-cell level in NK cells obtained from peripheral blood samples and from ex vivo tumour specimens from patients with glioblastoma, Shaim et al.⁷² found that NK cells have an altered immune phenotype that correlates positively with a reduced NK cell cytolytic function. Glioblastoma stem cells (GSCs), which are both the cell of origin and the source of disease recurrence, are susceptible to NK cell-mediated killing in vitro. However, these cells can also evade NK cell recognition through the activation of TGFβ signalling. If the TGFβ receptor or downstream signalling pathways are inhibited before NK cells become exposed to GSCs, the NK cells then become capable of cytotoxicity. Notably, the application of a TGFβ inhibitor is unable to restore tumour eradication functions. In a patient-derived xenograft model of glioblastoma, GSC-induced NK cell dysfunction was completely prevented by the direct inhibition of integrin or TGFβ signalling or by CRISPR gene editing of the TGFβ receptor 2 (TGFβR2) on allogeneic NK cells, resulting in effective tumour control. Taken together, these data indicate that the inhibition of the αv integrin–TGFβ axis could overcome a major obstacle to the clinical implementation of effective NK cell immunotherapy for patients with glioblastoma, although the inhibition of TGFβ signalling alone is insufficient in the setting of pre-existing immune dysfunction⁷². In summary, these data indicate that multiple immune cell types are refractory to immunomodulatory strategies in the context of glioblastoma.

Can cellular immunotherapies overcome immune exhaustion?

Theoretically, immunotherapy generated outside of the immunosuppressive TME, or even ex vivo, could overcome the barrier of immune exhaustion. Nonetheless, the success of chimeric antigen receptor (CAR) T cell-based strategies in CNS cancers has been limited thus far by tumour heterogeneity and the insufficient distribution of antigens across the TME. Similar to trials involving an EGFRvIII peptide vaccine, in which disease recurrence was associated with the loss of antigen expression^25,73, clinical trials of peripherally infused EGFRvIII-directed CAR T cells also revealed adaptive resistance with a loss of tumour antigen expression at recurrence⁷⁴. The loss of EGFRvIII expression might be a function of the natural history of recurrent glioblastoma⁷⁵, although similar observations have also been made with CAR T cells targeting IL-13Rα2 (REF.⁷⁶). A strategy that might overcome the emergence of escape variants is to try to target the CAR to ubiquitous and homogenously expressed targets such as EGFR⁷⁷, although even with tuning based on EGFR density, lingering concerns remain regarding cross-reactivity and toxic effects on non-malignant tissues. CAR T cells are presumed to be refractory to immune exhaustion, although emerging data from mouse models indicate that the CAR, upon encountering the tumour cell, strips the antigen from the tumour via trogocytosis and then expresses the tumour-targeting antigen on the CAR surface and thereby undergoes fratricide and exhaustion⁷⁸. Strategies designed to inhibit or avoid CAR T cell trogocytosis are currently being developed, although other additional challenges remain such as the lack of a uniform distribution of either native or CAR T cell populations across the different glioma regions (these cells are usually found in small isolated areas at the tumour–brain interface or in perivascular regions), heterogeneous antigen expression and immune escape, a lack of glioma-specific immunogenic target antigens, chronic use of immunosuppressive steroids, and the optimal treatment sequencing relative to standard-of-care therapies.

Developing new immunotherapeutic approaches

Targeting innate immunity.

Another factor contributing to the irreversible immune cell exhaustion and dysfunction seen in immune cells of the glioblastoma TME is the predominance of immunosuppressive myeloid cells⁶⁹, which can express immunotherapeutic targets that differ from those expressed on T cells (FIG. 3). Despite being the primary target population modulated by ICIs in other malignancies, T cell infiltration in the glioblastoma TME is rare. Even if T cells become reactivated and/or are induced to infiltrate the TME by immunotherapy, they are confronted by the predominance of immunosuppressive myeloid cells, which are likely to render the T cells non-functional. In many instances, the various myeloid cell populations of the glioblastoma TME are unable to fulfil their role of engulfing the tumour cells through phagocytosis followed by antigen processing and presentation to T cells. To induce a fully functional antitumour immune response, both the innate and adaptive immune systems need to work in a coordinated fashion. As such, combination therapies that target and/or modulate both the innate and adaptive immune systems are likely to be required to successfully treat patients with glioblastomas. Most scientific effort has been directed to T cell modulation, although emerging efforts are now being directed towards modulating innate immune cell dysfunction.

Fig. 3 ∣. Frequencies of the major immune cell populations and potential immunotherapeutic targets in gliomas and brain metastases.

The immunological landscape of cancers of the central nervous system largely differs between gliomas (depending on isocitrate dehydrogenase (IDH) mutation status) and brain metastases. The presence and relative frequency of immunotherapy targets or potential targets is therefore an important point to consider when seeking to select a therapeutic approach. CSF1R, macrophage colony-stimulating factor 1 receptor; DC, dendritic cell; LILRB, leucocyte immunoglobulin-like receptor; MDM, monocyte-derived macrophage; pSTAT3, phosphorylated signal transducer and activator of transcription 3; SIRPα, signal regulatory protein-α (also known as tyrosine-protein phosphatase non-receptor type substrate 1); STING, stimulator of interferon genes; TME, tumour microenvironment.

Preclinical studies targeting tumour-associated macrophages and microglia using an inhibitor of the colony-stimulating factor 1 receptor (CSF1R), a receptor whose activation is necessary for macrophage proliferation and survival (TABLE 1), demonstrate that CSF1R inhibition leads to macrophage polarization towards an inflammatory antitumour (or M1-like) phenotype and delays tumour progression⁷⁹. In an initial study, the therapeutic effect of CSF1R inhibition was transient and >50% of the mice developed resistance driven by an increase in insulin-like growth factor 1 (IGF1) secretion by macrophages and upregulation of IGF1R expression on tumour cells, thereby leading to the activation of the PI3K signalling pathway and enhanced glioma cell survival and invasion. In a follow-up study from the same research group, the inhibition of IGF1R or PI3K, together with CSF1R, led to a substantial increase in survival⁸⁰. Similar improvements in survival could be observed when CSF1R inhibition was combined with radiotherapy. Even when eliciting only a transient antitumour effect, radiotherapy has also been shown to induce transcriptional alterations in the myeloid population, resulting in the progressive accumulation of tumour-promoting microglia and macrophages in mouse models of glioma⁸¹. CSF1R inhibition reversed the radiation-induced alterations in myeloid cells, thereby enhancing the efficacy of radiotherapy⁸¹. In another preclinical study, CSF1R inhibition was demonstrated to modulate the immunosuppressive myeloid cell populations induced by dendritic cell vaccination and, in combination with PD-1 blockade, to generate long-term survivors (50% survival at 80 days) in a mouse model of glioma⁸². Nonetheless, a clinical trial of a CSF1R inhibitor as a monotherapy in patients with glioblastoma did not produce any indication of clinical efficacy⁸³. Combinatorial regimens including CSF1R inhibitors remain a consideration, although specific strategies will require preclinical prioritization prior to implementation in patients.

The SIRPα–CD47 pathway, which acts as a phagocytosis checkpoint in macrophages and other innate immune cells, is another immunotherapy target that is applicable to patients with gliomas (TABLE 1)⁸⁴. SIRPα is expressed by a variety of myeloid cells such as macrophages, microglia, dendritic cells and neutrophils. The ligand, CD47, is an anti-phagocytic ‘don’t eat me’ signal that is widely expressed on non-malignant cells. CD47 is upregulated in many types of cancer, including glioma, and enables cancer cells to evade antitumour immunity⁸⁵ by preventing phagocytosis⁸⁶. Preclinical studies indicate that anti-CD47 antibodies induce phagocytic activity in macrophages and microglia, thereby prolonging the survival durations of various mouse models^87,88. Furthermore, the combination of an anti-CD47 antibody with either radiotherapy or temozolomide (the current standard of care for glioblastoma) decreased tumour size and extended the survival duration of preclinical glioma models⁸⁹. The combination of an anti-CD47 antibody plus temozolomide is reported to activate the cGAS–STING signalling pathway, which acts as a bridge between the innate and adaptive immune responses. This combination triggered tumour cell phagocytosis, leading to enhanced antigen cross-presentation and stimulation of the cGAS–STING pathway in APCs and resulting in improved T cell priming measured as significant increases in CD8⁺ T cells with an effector or central memory phenotype. Because the combination treatment also induced the expression of PD-1 and PD-L1, an anti-PD-1 antibody was added to the regimen, resulting in a further increase in survival duration and leading to >50% of mice surviving beyond 80 days⁸⁵. STING agonists are particularly compelling therapeutics for gliomas owing to their ability to induce T cell infiltration into otherwise immunologically ‘cold’ tumours by promoting the inflammation and activation of the usually immunosuppressive tumour stroma^36,90-94. STING is a widely expressed sensor of cellular stress that is activated by the presence of DNA in the cytoplasm and can act as a bridge between the innate and adaptive immune systems, both by triggering interferon release and through the cis-activation of myeloid cells. Our group is currently preparing the first clinical trial of a novel STING agonist in patients with glioblastoma. We propose that the immunosuppressive STAT3 and the immune-activating cGAS–STING signalling pathways are antagonistic transcriptional immune modulators within APCs (FIG. 4; therefore, preclinical studies attempting to determine the immunological and therapeutic effects of targeting these pathways in CNS cancers using combination therapies are currently ongoing.

Fig. 4 ∣. Competing transcription factors regulate the activation of innate immune cells.

Tumour cell phagocytosis is triggered through interactions between calreticulin and Prolow-density lipoprotein receptor-related protein 1 (LRP1), both of which are antagonized by CD47-SIRPα, which are both therapeutic targets for modulating the innate immune system in patients with cancer. Within the phagosome, tumour degradation results in the cytoplasmic accumulation of double-stranded DNA (dsDNA), which then activates cyclic GMP–AMP synthase (cGAS)–stimulator of interferon genes (STING) signalling. Phosphorylated IRF3 and NF-κB translocate into the nuclear compartment and drive the transcriptional regulation of pro-inflammatory responses such as the release of interferons and TNF. Antigens processed in the phagosome are loaded onto MHCs, which subseguently promote the activation of antitumour T cells, which have direct cytotoxic effects on tumour cells (such as by the release of perforin and/or granzymes). By contrast, the signal transducer and activator of transcription 3 (STAT3) signalling pathway is activated by IL-6, which is released from reactive astrocytes. Upon phosphorylation, STAT3 dimerizes and then translocates into the nuclear compartment, where it regulates the transcription of immunosuppressive cytokines such as IL-10 and TGFβ, inhibits the production of pro-inflammatory cytokines and antigen presentation, and induces PD-L1 expression. Red crosses indicate processes that can potentially be inhibited by STAT3. IRF3, interferon regulatory factor 3; SIRPα, signal regulatory protein-α (also known as tyrosine-protein phosphatase non-receptor type substrate 1); TAP, transporter associated with antigen processing; TCR, T cell receptor.

The importance of preclinical validation.

An appropriate immunotherapy target should be specific to the tumour and have a high frequency of expression. If this is not the case, then a companion biomarker needs to be considered in order to identify the most appropriate patient population. Without a robust biomarker strategy, the subsequent clinical trial is at risk of being negative owing to the presence of a non-responsive and/or non-applicable patient population, thereby diluting the signal of response. In other malignancies, many of the immune targets are consistently expressed; thus, sufficient numbers of unselected patients are likely to respond. However, in glioblastoma, most of the target biomarkers are restricted to a minority of patients. Target and/or antigen expression might also be heterogeneous within the malignancy, which can make defining a cut-off point challenging. However, the use of biomarker stratification or enrichment does not overcome these challenges when no actionable target is present or, owing to either the natural course of the disease or clonotypic selection, the malignancy rapidly acquires an antigen-negative cellular population that gives rise to disease recurrence. The best-studied examples of the key features of a biomarker are provided by the experience with immunotherapies targeting the tumour-specific antigen EGFRvIII^73,74,95. The ability to reproducibly generate and maintain a robust antitumour immune response is another critical component of a successful immunotherapy. The strategy should activate an antitumour immune response, support the infiltration of immune cells into the tumour and sustain immune effector function within the TME, or at least not impair these immune functions. Peripheral blood monitoring is typically used, although these assessments have not correlated with the outcomes of patients with glioblastoma receiving immunotherapies in clinical trials because such methods fail to account for the infiltration of the tumour and the maintenance of effector responses in an overwhelmingly immunosuppressive TME. This weakness is especially the case for T cell-mediated responses because such cells can potentially be sequestered into the bone marrow⁹⁶. Furthermore, the immunotherapy should be capable of adequate blood–brain barrier penetration and/or demonstrate adequate effector function in the glioma microenvironment.

Selecting a target that is consistently expressed at a high level and confirming an operational role in the pathology of a certain disease is an important first step before considering a potential therapeutic strategy, although data from emerging studies show that therapeutic strategies addressing these criteria alone are not sufficient. Ott et al.⁶⁷ used CD4⁺ and CD8⁺ T cells and CD11b⁺ myeloid cells isolated from the blood of individuals without cancer and from the blood and tumours of patients with newly diagnosed and recurrent gliomas to profile the expression of potential targets of available therapeutics. Mouse models of glioma were then used to assess the efficacy of agents targeting the most frequently expressed targets. Immune effector function was then analysed in human in vitro co-culture experiments in the setting of glioma-induced immune suppression. Despite adenosine signalling being the most frequently upregulated pathway in patients with glioma, the inhibition of this pathway with several inhibitors targeting various steps of this axis in different mouse models had only modest therapeutic effect. In vivo experiments revealed that, although adenosine receptor inhibitors could promote the recovery of immune effector functions in T cells after the engagement of this pathway, the recovery of function remains impaired in the presence of gliomas, indicating that irreversible immune exhaustion most likely limits the effectiveness of inhibitors of the adenosine signalling pathway⁶⁷. These data also indicate that, among currently available immunotherapies, only a few would be applicable to a substantial number of patients with glioblastomas and, more importantly, underline the necessity of adequate preclinical testing of assays designed to quantify the restoration of immune dysfunction in the context of the intended use, which should be incorporated into preclinical vetting strategies.

Antigen presentation in the TME might be essential for tumour eradication.

T cell lymphopenia is a well-documented feature of high-grade gliomas. In both patients and mouse models of glioma, T cells are sequestered in large numbers in the bone marrow when tumours either develop or are introduced into the CNS⁹⁶. T cell bone marrow sequestration is accompanied by the tumour-imposed loss of S1P1 from the T cell surface and is reversible when S1P1 internalization is blocked, which is a potential therapeutic strategy. The CNS is drained by lymphatic vessels, which have the potential to prime T cell responses; nonetheless, CD8⁺ T cell-mediated antitumour immune responses to glioblastoma-associated antigens in the TME are typically very limited. VEGFC expression induced by an adeno-associated viral vector injected into the cisterna magna has been shown to promote CD8⁺ T cell priming in the deep cervical lymph nodes, with migration of effector T cells into the tumour and protective immunological memory responses in mouse models of glioblastoma⁹⁷. This strategy has been shown to synergize with immune-checkpoint inhibition. Notably, these observations are supported by Cloughesy et al.³², who identified a correlation between high levels of VEGFC expression and T cell infiltration in the neoadjuvant group of a small phase II trial involving patients with resectable glioblastomas. The cellular origins of this effect and whether it is induced directly by anti-PD-1 antibodies are currently unknown and remain an area for future investigation. Brain metastases from NSCLC and melanoma are at least partially responsive to ICIs despite no appreciable elevation of VEGFC expression in the primary cancers. This idea seems provocative, although the analysis of data from bulk RNA sequencing and single-cell RNA sequencing suggests that the small amounts of VEGFC produced do not originate from macrophages but rather from other non-immune cells such endothelial cells or pericytes³².

Data are beginning to emerge on the role of the innate immune system in the TME and in antitumour immune responses. Conventional theory holds that antigen presentation and immune activation are triggered in the peripheral lymph nodes. The activated T cells subsequently exit the lymph nodes and travel to the TME, guided by a chemokine gradient, to effect an antitumour immune response. If CX3CR1⁺ cells are absent from the TME, T cell-mediated immune surveillance of the tumour is lost^98,99. After encountering CX3CR1⁺ cells, T cells are capable of undergoing proliferation — a hallmark of productive engagement with an APC. This line of investigation might ultimately lead to the development of novel therapeutic agents that enhance the trafficking and functions of specific innate immune cells in the TME and thus support T cell infiltration and antitumour activity against tumours with minimal levels of pre-existing T cell infiltration. Radiation is known to trigger very potent antitumour immunity in immunogenic tumours in mouse models though the STING–cGAS pathway¹⁰⁰. In preclinical models of glioma, we found that the combination of whole-brain radiotherapy and STAT3 inhibition triggers interactions between CD103⁺ dendritic cells and T cell interactions in the glioma microenvironment, including antigen presentation and T cell activation⁶⁶. This combination triggered extensive clustering of APCs, including dendritic cells and T cells, in the local TME covering large parts of the whole tumour area. This clustering is an early event during immune activation and during the initial antigen presentation events, typically occurring before immune-checkpoint upregulation¹⁰¹, thus allowing T cells to exert their effector responses, including the erradication of the tumour.

Tumour-mediated immunosuppression is especially problematic when STAT3 is activated because this transcription factor renders a variety of immune cells dysfunctional and/or immunosuppressed, including tumour-infiltrating dendritic cells, T cells, microglia and macrophages (TABLE 1)^102-110. Whole-brain radiotherapy triggers the activation and tumour infiltration of dendritic cells¹¹¹; therefore, because STAT3 blockade reverses the immunosuppressive glioma TME, de novo antigen presentation and T cell activation can occur unencumbered in this scenario⁶⁶. STAT3 is not only a key regulator of immune cell activity, thereby promoting tumour growth in brain cancer, but its activation in cancer cells and the tumour-supportive stroma has also been shown to be a key contributor to gliomagenesis by inhibiting apoptosis, inducing tumour cell proliferation, promoting invasion and inducing angiogenesis^112-115.

Data from the past few years indicate an interesting role for STAT3 in a subset of reactive tumour-associated astrocytes^116,117. Priego et al.¹¹⁷ explored a metastasis-promoting programme driven by STAT3 signalling in a subpopulation of reactive astrocytes surrounding metastatic lesions and demonstrated the modulation of immunosurveillance in experimental models. In another study involving human ex vivo organotypic slice models of glioblastoma¹¹⁶, a distinct anti-inflammatory tumour-associated reactive astrocyte population characterized by JAK–STAT3 pathway activation and PD-L1 expression was identified. This astrocyte subpopulation contributed substantially to the evolution and maintenance of the immunosuppressive glioma TME via crosstalk with other cell types such as microglia, resulting in the release of anti-inflammatory cytokines and transcriptional changes in myeloid cells towards an anti-inflammatory state. These findings illustrate how non-immune cells can alter the immune TME by influencing immune cell interactions and again demonstrates the complexity of the TME alongside other components that could be therapeutically modulated.

The clinical relevance of elucidating immune cell interactions as a biomarker of therapeutic response is now emerging. For example, a single-cell analysis of samples from a clinical trial in which patients with ovarian cancer received a PARP inhibitor plus pembrolizumab demonstrated prominent interactions between exhausted CD8⁺ T cells and PD-L1⁺ macrophages and PD-L1+ tumour cells as mechanistic determinants of response. The spatial analysis of samples from two near-complete responders revealed differential clustering of exhausted CD8⁺ T cells with PD-L1⁺ macrophages whereas exhausted CD8⁺ T cells interacting with cancer cells harbouring genomic amplification of both PDL1 and PDL2 were detected in non-responders¹¹⁸.

Window-of-opportunity clinical trials: an important and necessary step.

Owing to various documented immunosuppressive effects, such as bone marrow T cell sequestration, a lack of adequate effector T cell infiltration into the TME and impaired immune effector function within the tumour, verifying a glioma-specific immune effector response remains a paramount consideration prior to implementing later-stage clinical trials because an interpreted ‘positive signal of response’ could fail to account for study biases such as above average performance status, more-intensive medical management, and certain prognostic variables. Window-of-opportunity clinical trials involve treating the patient prior to surgical resection, thus enabling the assessment of drug concentrations (including in the infiltrating non-enhancing tumour margin), target engagement, the presence and dispersal of an immune effector response in the TME, and the interrogation of possible alternative mechanisms of action. Immune effector function within the glioblastoma TME has been studied in such a manner in patients receiving pembrolizumab in an open-label, single-centre, single-arm phase II trial in which 15 patients with recurrent, operable glioblastoma received up to two doses of pembrolizumab before surgery and every 3 weeks afterwards until disease progression or unacceptable toxicities¹¹⁹. Similar to the neoadjuvant cohort that demonstrated therapeutic responses reported by Cloughesy et al.³², response durations exceeded 37 months in two patients (one with IDH1-mutant and one with IDH-wild-type disease) with a median OS duration of 20 months and an estimated 1-year OS of 63% for the entire cohort — outcomes that exceed those achieved with standard-of-care therapies. Notably, samples from these recurrent tumours contained few infiltrating T cells; nonetheless, the glioma TME was markedly enriched for PD-1⁺ CD68⁺ macrophages that also expressed other immune-checkpoint ligands, wherein T cells expressed few markers of immunological activity¹¹⁹. These results might reflect a suboptimal sampling time for the detection of effector T cells or, alternatively, that pembrolizumab alone cannot induce an effector immunological response in most patients with glioblastomas, probably owing to a scarcity of T cells within the TME and a preponderance of CD68⁺ macrophages. Notably, CyTOF analysis of the CD68⁺ macrophage population demonstrated widespread PD-1 expression, which might be an alternative mechanism of antitumour activity.

As well as providing an opportunity to potentially reconfigure biomarker design, window-of-opportunity studies can also clarify whether additional immunomodulatory strategies might be required to optimize a therapeutic response. Despite data from several other trials suggesting a therapeutic response in patients vaccinated (in various manners) against cytomegalovirus (CMV) antigens^120,121, we found, in a window-of-opportunity study¹²², a number of factors that limit the effectiveness of adoptive immunotherapies. Highly functional autologous polyclonal CMV pp65-specific T cells from patients with glioblastoma were numerically expanded under good manufacturing practice-compliant conditions and administered after 3 weeks of lymphodepleting dose-dense temozolomide (100 mg/m²) treatment¹²². Despite an overall increase in the number of CMV-specific T cells in the peripheral circulation with each administered cycle, these CMV-specific T cells lacked the expression of biomarkers associated with the ability to generate cytotoxic effector responses. We postulated that the effector CMV-specific T cells were trafficking to the TME and were thus absent from the peripheral circulation. However, the overall frequency of CMV-specific CD8⁺ T cells in the glioma TME was low and immune profiling revealed that these T cells also lacked effector function. CMV-specific T cells were also only found in the perivascular regions of the tumours and were not dispersed throughout the tumour parenchyma. Cumulatively, these data indicate that the CMV-specific T cells generated in this study lack sufficient anti-glioblastoma activity. Furthermore, we found that heterogeneous CMV pp65 antigen expression might be a major limitation, even if sufficient anti-CMV T cell responses are generated.

Window-of-opportunity trials are increasingly being used to assess the effects of immunotherapies on the glioma TME. For example, a glioma lysate is being administered before surgery in patients with WHO grade 2 gliomas in order to assess the extent of CXCL10 induction and the presence of vaccine-specific T cells (NCT02549833). T cell receptor clonality, T cell density and gene expression signatures will be assessed in patients with recurrent glioblastomas receiving pembrolizumab in combination with a dendritic cell tumour lysate vaccine (NCT04201873). The effects of anti-GITR and anti-PD-1 antibodies, both with and without stereotactic radiosurgery, are also being assessed in small groups of patients with recurrent glioblastomas (eight per arm; NCT04225039). A trial using microdialysis for the real-time immune monitoring of cytokines in patients with recurrent glioblastoma receiving nivolumab plus the anti-LAG3 antibody BMS-986016 provides a novel strategy enabling the direct longitudinal measurement of immune responses in the TME (NCT03493932). In addition to defining T cell responses, trials exploring alternative immunotherapeutic approaches are also using this strategy. A clinical trial in which NK cells are derived from human placental CD34⁺ cells aims to evaluate the immune phenotype, effector function and distribution of these cells in the glioblastoma TME after either intravenous or direct intratumoral administration (NCT04489420) to clarify whether either strategy merits testing in a larger cohort of patients.

A hypothetical alternative mechanism of action of ICIs.

Conventionally, the therapeutic activity of ICIs has been assumed to be mediated by effector T cells. PD-1 is commonly expressed on innate immune cells, including macrophages, where its engagement limits the phagocytic activity of these cells¹²³. Owing to discrepancies between signals of clinical response in patients with glioblastoma receiving ICIs^32,119 and an immune cell biomarker analysis demonstrating a paucity of effector responses in the TME during the window-of-opportunity component¹¹⁹, we conducted further modelling studies in a genetically engineered mouse model that develops spontaneously arising high-grade gliomas after induction of PDGFB and STAT3 expression in nestin-positive cells¹¹³ in which we also ablated Cd8 to determine how CD8⁺ T cells affect the progression of malignant glioma¹²⁴. We observed a survival benefit in immunocompetent mice with endogenously arising intracranial glioblastomas after the intravenous administration of anti-PD-1 antibodies. Furthermore, the therapeutic effect of anti-PD-1 antibodies persisted even after Cd8 ablation¹²⁴. CD11b⁺ and Iba1⁺ monocytes and macrophages were enriched in the glioma TME of the Cd8^−/− mice. The macrophages and microglia present in the TME assumed a pro-inflammatory M1 response signature in response to an anti-PD-1 antibody, resulting in the elimination of PD-1-positive macrophages and microglia. Anti-PD-1 antibodies inhibited the proliferation of microglia and induced their apoptosis through antibody-dependent cellular cytotoxicity, which was demonstrated by the ability of fluorescently labelled anti-PD-1 antibodies to directly access the glioma TME.

Cumulatively, the available data indicate that the therapeutic effects of anti-PD-1 antibodies in glioblastoma might be mediated by the innate immune system rather than by CD8⁺ T cells¹²⁴. The paucity of CD8⁺ T cells in patients with glioblastoma has long been considered the reason for the failure of ICIs. However, the population of PD-1⁺ macrophages identified in the window-of-opportunity clinical trial involving patients with glioblastoma¹¹⁹ was frequent and could be amenable to anti-PD-1 antibodies. These data have two major implications: (1) biomarkers of response to ICIs in patients with glioblastoma might not necessarily reflect the function of CD8⁺ T cells and (2) combination treatment strategies that involve innate immune cell modulation might need to be reprioritized based on the unique ability of anti-PD-1 antibodies to induce M1 macrophage polarization. Ultimately, any promising preclinical findings will need to be validated in human patients to ascertain whether an activated or polarized microglia or macrophage signature is reflective of responsiveness to ICIs.

The therapeutic effects of anti-PD-1 antibodies are likely to be mediated by several different immune cell types. In malignancies that are typically enriched with T cell infiltration, anti-PD-1 antibodies likely exert most of their therapeutic activity through direct T cell-ligand interactions. By contrast, in malignancies such as glioblastoma that are largely devoid of infiltrating T cells, anti-PD-1 antibodies might have therapeutic effects through the activation of alternative immune cell populations such as macrophages and/or microglia. This putative alternative mechanism of action also provides an explanation as to why conventional biomarkers (such as TMB and a hypermutated phenotype, which are surrogates for T cell responses and are used to predict clinical responses in patients with other cancers) have not enabled enrichment for patients with glioblastoma who are more likely to respond in clinical trials²⁸. Emerging transcriptomic findings indicate the existence of a subgroup of microglia that possibly have lytic functions⁶³; these data provide evidence for an enticing potential role of ICIs in modulating innate immunity, including microglia. Whether microglia can function similarly to NK cells at the junction between innate and adaptive immunity awaits future investigations.

Future directions

Over the next few years, multiplex data will likely reveal the differences in immune cell interactomes that are currently being inferred using connectomic analysis from receptor–ligand protein expression datasets. These studies will also most likely identify novel interactions involving CNS-resident cellular populations, such as astrocytes, neurons and microglia, and various infiltrating immune cell populations both within the context of the adjacent non-malignant brain tissues and in pathological states such as cancer. Differences in various immune cell interactions based on cancer lineage are already emerging that might be biomarkers of sensitivity to immunotherapy. For example, if the apparent dendritic cell–T cell cluster interaction observed at the interface of CNS brain metastases are activating naive T cells that are not yet rendered exhausted by chronic stimulation, then these interactions might be a biomarker of response to ICIs.

Unbiased ‘omics’ profiling is now clarifying the diversity of the various immune cell lineages within the context of CNS malignancies, although the inferred functional status will require validation, most likely at the single-cell level. To date, our understanding of immune cell lineage functions is largely derived from bulk analysis in which the majority of subsets are generally inferred as the presumed source of specific immune activities. The failure of innate immune cells to align to a polarized phenotype has long been questioned, although now, when placed in the context of new findings regarding immune cell heterogeneity^56,57,63, this observation might not be totally refuted but rather needs to be reconfigured based on lineage subtypes.

Current immunotherapies have generally focused on the effector arm of cancer immunity, although this might not be the optimal path forward in certain malignancies given the rarity of T cell and B cell populations in these types of malignancies given that their functional activities are unable to be restored either by ICIs or by inhibitors of the adenosine signalling pathway. New findings, such as microglia possibly having lytic functions⁶³, open a new therapeutic trajectory focused on inducing the activity of this subtype rather than the conventional skew from the M2 to the M1 paradigm. These newly identified subtypes will likely also reveal new immunological mechanisms of tumour clearance that act as a bridge between innate and adaptive immunity. Whether these lytic microglia can exert direct antitumour effects independent of antigens and/or MHCs, similar to NK cells, is unknown. Investigating the balance between activating and inhibitory receptors that are associated with this population will be an intriguing area of investigation.

Conclusions

Cumulatively, the scientific evidence is becoming increasingly clear that the efficacy of immunotherapies and the validity of their associated biomarkers is not generalizable across all cancer lineages, especially those residing in the CNS. In the near term, the field has developed an emerging focus on the modulation of the innate immune system with strategies such as anti-CD47 antibodies, STING agonists and CSF1 inhibitors. At the interface with the pharmaceutical industry, appropriate vetting steps should be considered prior to implementing a clinical trial for patients with immunotherapy-refractory cancers, including testing the ability of an agent to restore immunological function in the context of the intended indication. The inclusion of window-of-opportunity analyses as a method of verifying target engagement, a sufficient level of immune cell trafficking and the maintenance of effector activity would minimize the risk of false-positive findings arising from unaccounted-for biases. Finally, the reluctance of funding agencies to support ‘characterization’ studies as opposed to their sole focus on ‘hypothesis-generated’ research has hindered progress in a field that should be grounded in the truth of human biology.