Glioma is the most common primary cancer of the central nervous system, and around 50% of patients present with the most aggressive form of the disease, glioblastoma. Conventional therapies, including surgery, radiotherapy, and pharmacotherapy (typically chemotherapy with temozolomide), have not resulted in major improvements in the survival outcomes of patients with glioblastoma. Reasons for this lack of progress include invasive tumour growth in an essential organ, which limits the utility of local therapy, as well as the protection of tumour cells by the blood–brain barrier, their intrinsic resistance to the induction of cell death, and lack of dependence on single, targetable oncogenic pathways, all of which impose challenges for systemic therapy. Furthermore, the unique immune environment of the central nervous system needs to be considered when pursuing immune-based therapeutic approaches for glioblastoma. Nevertheless, a range of different immunotherapies are currently being actively investigated in patients with this disease, spurred on by advances in immuno-oncology for other tumour types. Herein, we examine the current state of immunotherapy for gliomas, notably glioblastoma, the implications for combining the current standard-of-care treatment modalities with immunotherapies, potential biomarkers of response, and future directions for glioblastoma immuno-oncology.
The current standard of care for patients with glioblastoma includes surgery, temozolomide chemotherapy, radiotherapy, and corticosteroids, all of which have immunosuppressive effects; we must be cognizant of this complexity when developing immunotherapies.
Evidence for immunostimulatory effects of these treatments in the clinic, including abscopal effects, induction of immunogenic cell death, and depletion of regulatory T cells by temozolomide, remains limited.
Vaccination has been considered one of the most promising approaches to improving the outcomes of patients with glioblastoma, although negative results from several phase II and phase III trials challenge the current concept of vaccination as a single-modality immunotherapy.
Oncolytic viruses might exert pro-inflammatory responses that could potentially be exploited in future combined modality immunotherapy studies, whereas the future of chimeric antigen receptor (CAR) T cell therapy for glioblastoma depends on the identification of stably expressed and sufficiently tumour-specific antigens.
Immune-checkpoint inhibitors have promising therapeutic activity in preclinical glioblastoma models, whereas the results emerging from clinical trials in patients with recurrent glioblastoma are disappointing; larger studies are underway in the frontline treatment setting.
Future immune-based strategies are focused on combinations of different immune-checkpoint inhibitors with diverse treatment modalities that reverse local immunosuppression in the microenvironment, converting a ‘cold’ tumour into a ‘hot’ tumour.
Gliomas are intrinsic brain tumours thought to originate from neuroglial progenitor cells on the basis of their localization, morphological similarities to nonmalignant neuroglial cells, and generation of experimental gliomas by targeted manipulation of certain brain cells1,2,3. This disease entity encompasses a diverse range of central nervous system (CNS) cancers, including astrocytomas, oligodendrogliomas, and ependymomas — historical designations that reflect the putative tissue of origin4. Some of these tumours typically occur in childhood, such as pilocytic astrocytoma or subgroups of ependymomas, whereas others, including glioblastoma, have a peak incidence in the seventh decade of life5. Selected subtypes of gliomas can be cured using surgery alone, but the largest group, diffuse gliomas of adulthood, is generally resistant to all current standard-of-care therapeutic interventions (surgery, radiotherapy, and systemic chemotherapy); glioblastoma, the most aggressive variant, is invariably lethal. The overall annual incidence of gliomas in the USA is ~6 cases per 100,000 individuals, with glioblastoma accounting for ~50% of cases, and the disease has a male predominance5.
According to the revised 2016 WHO classification of CNS tumours4, diffuse oligodendroglial and astrocytic gliomas of adulthood are graded from grades II to IV on the basis of histological features that are known to correlate with the natural disease course. These tumours are further classified based on the absence or presence of mutations in IDH1 (which encodes isocitrate dehydrogenase [NADP] cytoplasmic) or IDH2 (which encodes isocitrate dehydrogenase [NADP], mitochondrial) and the absence or presence of 1p and 19q (1p/19q) chromosomal co-deletion4,6. This classification results in the categorization of adulthood gliomas into three main groups: IDH-mutant, 1p/19q-co-deleted tumours with a predominantly oligodendroglial morphology and a favourable prognosis; IDH-mutant, non-1p/19q-co-deleted tumours, which usually have an astrocytoma morphology and intermediate survival outcomes; and IDH-wild-type tumours that are mostly glioblastomas (WHO grade IV), exhibit gain of chromosome 7 and loss of chromosome 10, and have an unfavourable prognosis3. In fact, most glioblastomas are IDH wild type, including the histologically defined subtypes giant-cell glioblastoma, gliosarcoma, and epithelioid glioblastoma. By contrast, IDH-mutant glioblastomas are now considered a distinct disease entity related to IDH-mutant WHO II and III gliomas, and most of the following discussion relates mainly to classical IDH-wild-type glioblastoma.
In 2017, the European Association for Neuro-Oncology published updated recommendations for the diagnosis and treatment of adult diffuse gliomas, including glioblastoma7. According to these guidelines, the goal of surgery should be gross total resection, when feasible; whether macroscopically incomplete resections confer a survival benefit relative to biopsy sampling alone remains controversial8. Moreover, whether macroscopically resectable tumours share a less aggressive biology than those for which gross total resection cannot be achieved, and hence have an intrinsically better prognosis, continues to be debated9. For several decades, radiotherapy at a dose of up to 60 Gy, administered to a region consisting of the tumour plus a margin of nonmalignant tissue in 30 fractions of 1.8–2.0 Gy, has been part of the standard of care for patients with glioblastoma. This approach resulted in a doubling of median survival in early studies10,11. Patients with baseline characteristics indicating a less favourable prognosis, specifically a Karnofsky performance status (KPS) of ≤60, or an age of >70 years, are now typically treated with hypofractionated radiotherapy with a similar biologically effective dose of 40 Gy administered in 15 fractions12. For newly diagnosed adult patients aged ≤70 years who have a favourable general and neurological performance status, for example, defined as KPS of ≥70, the addition of concomitant and maintenance temozolomide chemotherapy to conventional radiotherapy (TMZ/RT → TMZ) became the standard of care in 2005 after the publication of results demonstrating an improvement in 2-year overall survival from 10.4% with radiotherapy alone to 26.5% with chemoradiotherapy13. In 2017, a clinical trial of a similar treatment regimen essentially confirmed that elderly patients also derive benefit from chemotherapy when added to hypofractionated radiotherapy14. TMZ provides greater benefit to patients with tumours that have methylation of the 6-O-methylguanine-DNA methyltransferase (MGMT) gene14,15 because this epigenetic alteration results in decreased expression of MGMT, which is a DNA-repair protein that limits the cytotoxic activity of TMZ. Numerous failed efforts at further improving the clinical outcomes achieved with the TMZ/RT → TMZ regimen involved the addition of various antiangiogenic compounds, notably bevacizumab (an anti-VEGF antibody), but also the αv integrin antagonist cilengitide16,17,18. Tumour-treating fields, also known as alternating electric field therapy, is a novel approach to the treatment of glioblastoma and other cancers. The addition of this modality to the TMZ/RT → TMZ regimen (during the TMZ maintenance period) resulted in improved progression-free survival (PFS) and overall survival in patients with newly diagnosed glioblastoma in an open-label phase III trial19. However, whether and how this treatment should be integrated into the current standard of care remains controversial: this treatment had no effect at recurrence20, no specific imaging responses or pathological findings upon progression and reoperation were reported, and thus the mode of action in vivo remains obscure. Moreover, the intense care provided by health care professionals for patients enrolled into the experimental arm of this trial has led to concerns that the prolonged survival was not entirely attributable to the specific intervention.
All patients with glioblastoma eventually have disease relapse. The management approaches used at tumour progression or recurrence are typically more individualized, accounting for patient-specific and disease-specific factors that include the radiological pattern of relapse, time since diagnosis, previous treatment, and, above all, general and neurological function. Favourable prognostic factors at glioblastoma progression include younger age and MGMT-promoter methylation as well as a long PFS duration with the first-line treatment21,22. The therapeutic options for patients with disease relapse include repeat surgery or radiotherapy, but mainly consist of pharmacological treatment with alkylating agent chemotherapy or bevacizumab depending on local availability7. From the time of first progression or recurrence, median overall survival durations in the range of 6–9 months have been achieved in large series of patients or in clinical trials23, but are shorter at the population level. Notably, a substantial proportion of patients (up to 50%) do not receive any second-line anticancer therapy23,24. In fact, evidence that any therapeutic intervention administered in the recurrent setting has a major effect on survival is lacking. For instance, in the EORTC 26101 trial25, combining bevacizumab with lomustine did not improve overall survival outcomes. More importantly, the median PFS duration with lomustine alone — the current standard-of-care systemic therapy for recurrent glioblastoma — was only 1.5 months, but overall survival was 8.6 months, indicating that the natural course of disease might be more relevant at recurrence than the current standard-of-care treatment options. Accordingly, novel approaches to the treatment of recurrent glioblastoma are urgently needed. Most recurrent tumours have previously been exposed to the genotoxic stress of irradiation and/or chemotherapy and are, therefore, predicted to have a higher mutational load and to be more immunogenic than untreated tumours. Herein, we review the theoretical framework driving the development of immunotherapy for glioblastoma, describe the strengths and weaknesses of the immunotherapy approaches that have been tested in the clinic to date, and discuss future directions in this promising area of therapy.
The CNS — an immune privileged system?
The concept that the CNS is immune privileged was based on initial experimental data reported >50 years ago by the group of Peter Medawar, which showed that foreign cells implanted into the brains of rodents became successfully engrafted, whereas the same cells were eradicated by the host immune system when they were placed in peripheral tissues26,27,28. Indeed, until 2015, the brain was thought to lack dedicated lymphatic channels, which was speculated to limit the presentation of antigens originating in the brain to immune cells. In addition, microglial cells were broadly considered to be the major antigen-presenting cells in the brain tumour microenvironment, and these cells have been hypothesized to skew T cells away from a cytotoxic phenotype29,30. However, more recent data have refined our understanding of immunological mechanisms that are active in the CNS. For example, we now know that the CNS is subject to active immunosurveillance and vigorous immune responses31 (Fig. 1). In retrospect, indications of immunological activity in the CNS have long existed. Indeed, follow-up experiments by Medawar’s group revealed that engraftment of foreign cells in the brains of rodents was prevented by vaccination of the animals against the same foreign cells before cell implantation26,27,28. Furthermore, in 2015, Louveau et al.32 defined a novel route of lymphatic egress from the brain along distinct channels that run parallel to dural venous sinuses. Thus, most antigen-presenting cells exiting the brain are likely to travel to the deep cervical lymph nodes, where they can prime T and B lymphocytes32. These findings are corroborated by inflammatory conditions, such as multiple sclerosis and cerebral brain abscesses, which demonstrate that immunogens present in the brain are capable of generating robust immune responses31,33. Taken together, these findings support the notion that, while the brain is an immunologically distinct site, the immune microenvironment offers adequate opportunities to implement immunotherapy for the treatment of brain tumours.
Unique mechanisms of immunosuppression
Glioblastoma has been extensively studied as a paradigm for cancer-associated immunosuppression34. Glioblastoma rarely metastasizes to extracranial sites, although circulating tumour cells have been detected in patients with glioblastoma35,36. Many assume, therefore, that glioma cells are either poorly suited to survive outside the brain or that, given the rapid progression of the tumour, limited time is available for tumour cells to colonize extracranial sites37. Nonetheless, systemic immunosuppression has been demonstrated, as evidenced by impaired cellular immunity in patients with glioblastoma and murine glioblastoma models38,39,40. Indeed, glioblastomas have a paucity of infiltrating T cells and harbour a relatively low number of somatic mutations compared with other tumour types41, and these tumours profoundly affect the immune system both locally and systemically, although no evidence indicates an increased incidence of infections typically associated with immunosuppression in patients with glioblastoma42. The mechanisms underlying these immunological effects are incompletely defined, but seem to involve both tumour-intrinsic factors and host responses to tumour antigens originating from the CNS. For example, in one study, investigators found that when melanoma cells expressing a model neoantigen were implanted in the brains of mice, neoantigen-specific T cells were actively deleted43. Furthermore, the T cells that escaped deletion failed to produce pro-inflammatory cytokines or execute cytotoxic functions, even after encountering the cognate neoantigen in peripheral lymphoid organs. By contrast, immune function was mostly preserved in mice with comparably advanced flank or lung tumours43. Brain-tumour-related immune suppression was hypothesized to be mediated by TGFβ, which was found to be secreted by microglia within the tumour microenvironment and was present at elevated levels in the serum of mice bearing brain tumours compared with mice with flank tumours or no brain tumours. Pharmacological inhibition of TGFβ signalling partially reversed the immune suppression but did not prolong the survival of mice with brain tumours43. In addition to tissue resident myeloid cells (microglia), evidence suggests that migrating myeloid cells have an important role in glioblastoma-associated immunosuppression. For example, Bloch et al.39 reported that macrophages isolated from the peripheral blood of patients with glioblastoma express increased levels of programmed cell death 1 ligand 1 (PD-L1), a ligand that activates the programmed cell death protein 1 (PD-1) immune-checkpoint receptor that restricts the activity of cytotoxic T cells. Correspondingly, these macrophages suppressed the activation of co-cultured T cells derived from the same patients39. Furthermore, Fecci et al.44 found that glioblastomas induce broad sequestration of T cells out of the circulation into the bone marrow through downregulation of sphingosine 1-phosphate receptor 1 (S1P1) in T cells.
Tumour-intrinsic factors involved in glioblastoma-associated immunosuppression include the induction of signalling pathways that are known to suppress immune responses. Wainwright et al.45 have demonstrated that glioblastoma cells express indoleamine 2,3-deoxygenase (IDO) enzymes, which catalyse the rate-limiting step in the catabolism of tryptophan to kynurenine — a pathway that is involved in T cell immune tolerance and, therefore, immunosuppression. Moreover, Heimberger et al.46 characterized a role of signal transducer and activator of transcription 3 (STAT3) signalling in glioblastoma cells in suppressing immune cell activity. STAT3 expression from human glioblastoma specimens seems to be driven by IL-10 in both tumour cells and immune cells. Indeed, soluble factors, notably TGFβ43,47, IL-10 (Refs48,49), and prostaglandins50, were the first immunosuppressive mediators identified in patients with glioblastoma, although subsequent studies have also indicated that direct interactions between glioblastoma cells and immune cells are involved in immune-escape mechanisms. Such interactions result in suppression of natural killer (NK) cell activity, which is mediated by atypical HLA molecules (including HLA-E51 and HLA-G52), direct induction of apoptosis in immune cells via tumour necrosis factor receptor superfamily member 6 (TNFRSF6, commonly known as FAS)–FAS ligand (FASL) interactions53, or triggering of inhibitory T cell checkpoints by PD-1 ligands54.
Overall, glioblastoma appears to be a highly immunosuppressive tumour. The exact mechanism of immune escape is unknown, although myeloid cells are probably key mediators of glioblastoma-associated immunosuppression. Notably, resident microglia and macrophages outnumber infiltrating T cells in these tumours55, and this paucity of T cells in the tumour microenvironment is in striking contrast to findings in other tumour types, such as melanoma or lung cancer56. The results of some studies have suggested that the glioblastoma-associated myeloid cells are immunosuppressive, with an M2-like phenotype57, whereas other animal studies have suggested that the ingression of myeloid cells, such as dendritic cells (DCs), from the periphery is required to elicit an immune response58. Whether or not these tumours are intrinsically non-immunogenic, or whether T cells are actively excluded, remains to be defined. Regardless, the myeloid compartment will probably be a key target for future glioblastoma immunotherapy.
Strategies focused on exploiting myeloid cells to foster antitumour immune responses include reducing or neutralizing the biological activity of immunosuppressive molecules (such as TGFβ) and administering adjuvants such as granulocyte–macrophage colony-stimulating factor (GM-CSF) or Toll-like receptor (TLR) agonists. Notably, the current focus of efforts to facilitate general antitumour immune responses in patients with glioblastoma is centred in the field of immune-checkpoint inhibition, but immune-checkpoint inhibitors might need to be combined with therapies targeted at the myeloid compartment59.
Of note, the profound immunosuppressive properties of glioblastomas have also influenced the design of most immunotherapy trials to focus on patients with only a limited tumour burden — that is, after gross total resection — and to minimize or omit the use of standard corticosteroid therapy and chemoradiotherapy, which can have immunosuppressive effects. However, these strategies might limit the availability of tumour antigens that is necessary for an effective anticancer immune response.
Current state of glioblastoma vaccines
The field of innovative immunotherapeutic approaches to treat glioblastoma is rapidly expanding. On the basis of the apparent failure of glioblastoma to metastasize outside the CNS, efforts to induce active immune surveillance against glioma cells in the brain by strengthening the adaptive arm of the immune system, predominantly by vaccination (Fig. 2), have been pursued as a promising path forward. Despite the relative paucity of data on active immunotherapies in humans with glioblastoma, three vaccination approaches have reached phase III clinical development, and numerous others are at earlier stages of clinical testing. Rindopepimut (also known as CDX-110 or PEPvIII) is a peptide vaccine that mimics and thus targets EGFR variant III (EGFRvIII), which is a constitutively active mutant form of EGFR that is expressed exclusively on glioblastoma cells in 25–30% of patients with this disease60. This agent enables a simple vaccination approach based on just one immunogenic peptide rather than on the generation of a vaccine from patient-derived immune and tumour cells. The advantage of targeting the EGFRvIII neoantigen relates to the exclusivity of neoantigen expression on tumour cells, which limits the risk of ‘on-target, off-tumour’ toxicities. The major disadvantage, however, is that EGFRvIII is heterogeneously expressed on glioblastoma cells in vivo60, thus creating the potential for outgrowth of tumour cells that lack this antigen. Furthermore, expression of EGFRvIII is unstable throughout the course of disease61,62, opening further avenues for immune escape of tumour cells that downregulate this protein. Three uncontrolled phase II studies of rindopepimut vaccination in favourably selected patients, that is, those with gross total resection and no evidence of progression after completion of concomitant chemoradiotherapy, have provided evidence of improved median survival of 24 months compared with historical controls63,64,65. These findings led to the initiation of ACT IV (Ref.66), an international phase III trial, in a large cohort of patients with newly diagnosed EGFRvIII-positive glioblastoma (Table 1). Of 745 patients enrolled, 405 fulfilled the predefined criteria for minimal residual disease (MRD), defined as the presence of <2 cm2 of contrast-enhancing tumour tissue after surgery and chemoradiotherapy66. All patients were randomly assigned (1:1) to receive either rindopepimut or keyhole limpet haemocyanin (KLH; used in rindopepimut as the carrier protein for the 13-amino acid EGFRvIII peptide epitope), together with maintenance TMZ; however, the primary end point was prolonged overall survival in the MRD subgroup, specifically66. The trial was terminated early, after a pre-planned interim analysis revealed the futility of treatment. At the final analysis, no significant difference in overall survival between the treatment arms was detected in the MRD population (median 20.1 months in the rindopepimut group versus 20.0 months in the control group; HR 1.01, 95% CI 0.79–1.30; P = 0.93)66. Interestingly, the hazard ratios seemed to be more favourable for those patients with “significant residual disease”(REF.66) (>2 cm2; HR 0.79, 95% CI 0.61–1.02; P = 0.066), although this difference was statistically significant only among patients treated in the USA (HR 0.70, 95% CI 0.51–0.96; P = 0.027), raising the possibility that this result occurred by chance. Nevertheless, this trend for potential benefit in patients with bulky residual disease rather than only MRD supports the possibility that a certain level of tumour tissue needs to be present to drive immune responses. Other important observations from that trial included the failure of a strong humoral immune response to translate into clinical benefit and the spontaneous loss of antigen expression, even in the control arm, as observed in patients undergoing repeat surgery66. Thus, EGFRvIII might not be a good target for immunotherapy owing to a lack of stable expression even in the absence of EGFRvIII-targeted treatment.
While ACT IV was being conducted, a non-comparative phase II trial, known as ReACT67, was performed to explore the combination of rindopepimut with bevacizumab in a smaller cohort of patients with recurrent EGFRvIII-positive glioblastoma (n = 72). The results of this trial suggested favourable outcomes relative to those obtained with bevacizumab plus KLH only67 (Table 1). This evidence of the potential activity of rindopepimut combined with bevacizumab, taken together with the negative outcome of ACT IV, lends support for further trials of combined immunotherapy and anti-angiogenic therapy, specifically with inhibitors of the VEGF pathway68,69.
IDH1 peptide vaccines have also been investigated in clinical trials, further emphasizing the 2016 WHO classification of IDH-mutant gliomas as a major subtype of this disease4. Preclinically, Schumacher et al.70 demonstrated CD4+ T helper 1 (TH1) cell-mediated immune responses in a transgenic mouse model with humanized MHC class I and MHC class II molecules loaded with 15-mer mutant peptides derived from IDH1-R132H. Subsequently, two ongoing phase I clinical trials have been initiated: NOA-16 and RESIST (NCT02454643 and NCT02193347, respectively; Supplementary Table 1). An IDH1-R132H-mutated peptide vaccine is being utilized in both trials, although in NOA-16, the peptides are being administered by subcutaneous injections concurrently with topical imiquimod (a TLR7 agonist that can activate myeloid cells71), whereas the vaccine is being delivered in combination with the adjuvants GM-CSF and Montanide ISA-51 in the RESIST trial. These trials might provide a new avenue for the treatment of patients with IDH-mutant gliomas.
Expression of telomerase reverse transcriptase (TERT) is increased in many cancers, including gliomas; indeed, TERT promoter mutations are the genetic alterations most frequently detected in glioblastomas72,73. Capitalizing on these findings, vaccine therapy predicated on the transfection of DCs to overexpress TERT and thus present TERT antigens has been tested in patients with pancreatic adenocarcinoma, with encouraging results reported in one patient74. However, no clinical trials have yet been conducted to investigate TERT-peptide-based vaccine therapy in patients with glioblastoma.
Vaccine strategies using more than one peptide have also been developed. In comparison with rindopepimut, such approaches are more complex and clinically less advanced. IMA-950 is a multipeptide vaccine that is based on the administration of 11 tumour-associated peptides (nine HLA-A*02 peptides, an elongated HLA class I peptide, and one HLA class II peptide) and the synthetic hepatitis B virus marker peptide IMA-HBV-001. Results of a phase I trial confirmed the tolerability of this vaccine, although two instances of dose-limiting toxicity potentially related to the vaccine (fatigue and anaphylaxis) were reported; the survival data were not remarkable75.
ICT-107 is another multipeptide vaccine, which was specifically designed for the treatment of glioblastoma76. This vaccine consists of patient-derived DCs incubated ex vivo with six peptides from proteins selected based on their over-representation in the gene-expression profiles of glioblastoma cells compared with nonmalignant tissues: melanoma-associated antigen 1 (MAGEA1), HER2, interferon-inducible protein AIM2, i-dopachrome tautomerase (DCT), melanocyte protein (PMEL), and IL-13 receptor subunit-α2 (IL-13Rα2). Preclinical proof of concept for this approach was difficult to obtain because similar peptides might not have the same immunogenicity between humans and mice and because mouse glioma models might not overexpress the same genes as human tumours. The vaccine was generated within a time frame from post-surgery apheresis to chemoradiotherapy completion and was intended to be administered together with maintenance TMZ. This strategy was chosen because the generation of the vaccine required time, although one might assume that an earlier initiation of vaccination might have benefits associated with the immunostimulatory effects of radiotherapy. The phase I study of ICT-107 enrolled 17 patients with newly diagnosed glioblastoma and 3 patients with recurrent glioblastoma. Safety was confirmed, with a median follow-up duration of 40.1 months, while the median PFS duration was 16.9 months and median overall survival duration was 38.4 months in the patients with newly diagnosed disease76. The results of a randomized phase II trial suggest that this agent has some therapeutic activity, at least in HLA-A2-positive patients77. In this trial77, 124 patients were randomly assigned (2:1) to receive ICT-107 or control treatment with autologous DCs that had not been exposed to the glioblastoma-associated antigens; 77 patients were HLA-A2-positive. Overall survival was not significantly improved (Table 1), but outcomes tended to favour the experimental therapy in HLA-A2-positive patients, potentially because four of six peptides were predicted to be presented in an HLA-A2-dependent manner. However, the trial was underpowered to derive meaningful conclusions from further subgroup analyses. No safety concerns were reported. A phase III trial of this vaccine, referred to as STING (NCT02546102), was opened for patient accrual in 2016 but was suspended owing to a lack of funding on 21 June 2017. The ICT-107 programme seems not to have been pursued further.
DCVax-L is a third therapeutic vaccine that has reached the phase III stage of clinical testing in patients with glioblastoma (NCT00045968; Supplementary Table 1). With the DCVax-L approach, whole tumour lysate from a patient’s resected glioblastoma is used to pulse analogous DCs and serves as a source for the entire spectrum of tumour antigens78. DCVax-L probably has the longest history of any glioblastoma vaccine, with the approach confirmed preclinically in a rat model in which vaccination induced a substantial increase in tumour T cell infiltration79. On the other hand, this vaccine is also the most logistically challenging to generate because it requires the collection of tumour samples from each individual patient, which are then processed and used to stimulate autologous DCs80. This strategy has the theoretical advantage of being personalized but carries the limitation of using autoantigens to which tolerance has already been established. The phase I study of this agent included 12 patients with newly diagnosed or recurrent glioblastoma, and the results suggested that a low tumour burden and low levels of TGFβ2 expression will help select patients who are more likely to benefit from this treatment. The pivotal DCVax-L trial was initiated in December 2006 (NCT00045968; Supplementary Table 1), but patient enrolment has been put on hold for unidentified reasons and outcomes data have not yet been made available.
Currently, a plethora of clinical trials are ongoing in order to evaluate different antigens for vaccine therapy against glioblastoma. At the time of this Review, 20 phase I, 16 phase II, and 2 phase III trials are ongoing (Supplementary Table 1). The two phase III trials, involving DCVax-L (as discussed above; NCT00045968) or an individual proteomics-based approach (NCT01759810), reflect the ongoing paradigm shift in medicine towards a personalized medicine approach. This trend is also seen in the ongoing phase II studies that involve vaccines generated using autologous tumour lysates (NCT01204684, NCT01635283, NCT02772094, NCT03018288, NCT02709616, and NCT02808364). The wide spectrum of phase II studies also illustrates the importance of the myeloid compartment in future immunotherapy approaches to the treatment of glioblastoma. For example, multiple trials are including the use of myeloid-activating adjuvants, such as polyriboinosinic–polyribocytidylic acid (poly(I:C); NCT02358187, NCT00766753, NCT01204684, and NCT02078648). Finally, the ongoing phase I trials reflect similar trends, with the use of myeloid-modulating adjuvants and personalized approaches to tumour antigens.
In general, trials of DC-based vaccine approaches are time and resource intensive, particularly if autologous tumour tissue is required for generation of the vaccine, as is this case for DCVax-L. Even with a defined, ‘off-the-shelf’ peptide antigen cocktail, such as the one used in ICT-107, patients still need to undergo apheresis after surgery in order to harvest DCs to be expanded ex vivo and need to wait for weeks until the vaccine has been generated. Furthermore, for vaccine therapies to be implemented for broader use, such as a part of standard-of-care treatment, challenges include logistical and regulatory requirements involved with transporting biological materials to and from a site of vaccine generation, meeting Good Manufacturing Practice standards, and vaccine quality control.
Nevertheless, the strong induction of a humoral response observed in patients involved in the ACT IV trial66, although of no apparent clinical benefit, illustrates that inducing immune responses through vaccination is, in principle, feasible. Furthermore, data demonstrating the induction of antitumour T cell responses have also been generated75. Thus, the experiences that have been gathered with glioblastoma vaccination therapy to date indicate that many vaccines are biologically active, but the degree of immune stimulation is generally insufficient to translate into clinical benefit; combinatorial approaches might provide superior results.
Oncolytic viral therapies
Initially, virus-based anticancer therapies were considered a treatment strategy separate from immunotherapy and were predicated on selective viral replication in and subsequent destruction of cancer cells. However, the antitumour immune responses induced as a result of oncolytic viral infections have blurred this distinction81 (Fig. 2). Viruses can activate the immune system through pathogen-associated molecular patterns and pattern recognition receptors. Furthermore, viruses often activate macrophages through receptors, such as TLRs82. As a secondary effect, activated myeloid cells can improve the infiltration of T cells into tumours to promote an inflamed microenvironment. As a result, viral therapies are a very interesting approach to overcoming the immunosuppression of glioblastoma. While initial strategies used replication-incompetent viruses to avoid complications of encephalitis83, contemporary oncolytic viral treatment approaches are increasingly utilizing replication-competent viruses, such as retroviruses, adenoviruses, herpes simplex viruses (HSVs), polioviruses, and measles viruses84,85 (Table 2; Supplementary Table 2).
In May 2016, the recombinant oncolytic poliovirus PVSRIPO received breakthrough therapy designation from the FDA on the basis of the findings of an ongoing phase I study in patients with recurrent glioblastoma (NCT01491893; Supplementary Table 2). PVSRIPO is a genetically engineered form of the oral poliovirus Sabin type 1, in which the internal ribosome entry site is replaced with that of human rhinovirus type 2 in order to eliminate neurovirulence. PVSRIPO infects and replicates within cells that express the poliovirus receptor, which is an onco-fetal cell adhesion molecule that is often expressed in glioblastoma, thus exploiting the natural affinity of the virus for CNS cells. The virus is administered by convection-enhanced delivery via a catheter inserted directly into the tumour. Preliminary data presented at the 2015 ASCO annual meeting indicated that in 24 patients, 42% of whom had received prior bevacizumab, the 2-year overall survival was 24%, and the median overall survival across all doses of PVSRIPO was 12.5 months; updated results reported in 2016 revealed that 3 patients (13%) remained alive at 3 years86. No full safety or preliminary efficacy data are currently available in the public domain.
Data from an ongoing phase I clinical trial (Supplementary Table 2) are also available for vocimagene amiretrorepvec (Toca 511), which is a non-lytic, replicating retrovirus derived from the Moloney murine leukaemia virus that has been engineered to encode a modified yeast cytosine deaminase and preferentially infects tumour cells87. Although the virus infects both normal and tumour cells, the tumour cells lack typical viral defence mechanisms that prevent viral DNA integration into their genome88. The virus was injected into tissues lining the resection cavity in 45 patients undergoing surgery for recurrent or progressive high-grade glioma, followed 6 weeks later by intravenous administration of Toca FC, an extended-release formula of the prodrug 5-fluorocytosine. In virus-infected cells, this prodrug is converted to the antimetabolite 5-fluorouracil by the exogenous cytosine deaminase, which is not otherwise expressed in human cells, thereby providing some degree of selectivity for tumour cells. The median overall survival duration of patients treated with vocimagene amiretrorepvec and Toca FC was 13.6 months, which was interpreted as being superior to historical controls88. The proposed mechanism of improved efficacy is thought to be not only a direct tumoricidal effect but also a virally induced generalized antitumour immune response. Later-phase studies of vocimagene amiretrorepvec are currently underway, including a randomized phase II trial in patients with recurrent high-grade glioma.
Adenoviruses have also been extensively studied in the context of anticancer therapy (Table 2; Supplementary Table 2) because they are common respiratory viruses with relatively easy mechanisms of in vitro manipulation and genetic engineering89; however, challenges have included the identification of target receptors for the virus. Adenovirus 5-Delta 24RGD (DNX-2401) is one oncolytic, conditionally replicative adenovirus that achieves tumour cell targeting through a 24-base deletion of the transforming protein E1A and insertion of an Arg–Gly–Asp (RGD) motif onto a viral capsid protein for improved targeting towards αv integrins90. DNX-2401 has been investigated in a phase I trial in combination with TMZ and is currently under investigation in early phase clinical studies in combination with the anti-PD-1 antibody pembrolizumab (NCT02798406) or IFNγ (NCT02197169; Supplementary Table 2). A phase I trial of a different conditionally replicating adenovirus, named ONYX-015, has been performed after resection of recurrent glioma91 (Table 2). Tumour cell targeting was achieved by modifying the adenovirus protein E1B, which normally blocks p53-induced host cell apoptosis in order to sustain viral replication: ONYX-015 was designed with an attenuated E1B protein that cannot bind to p53, rendering the virus capable of replicating in p53-deficient tumour cells only92. Notably, a maximum tolerated dose was not defined; ONYX-015 was well tolerated at doses of up to 1010 plaque-forming units91, indicative of the excellent safety of many oncolytic viral therapies. The median overall survival duration, however, was only 6.2 months. This lack of efficacy might be linked to efforts to optimize the safety of this virus: ONYX-015 might have been safe because it was highly attenuated, but for the same reason, it might have been less effective93.
Adenoviruses have also been modified to serve as tumoricidal gene delivery vectors, most notably aglatimagene besadenovec (AdV-tk). This replication-incompetent adenovirus was transfected to express the HSV thymidine kinase (HSV-TK) gene, which converts the prodrug ganciclovir (GCV) into a toxic nucleotide analogue that can kill replicating tumour cells94. This approach, termed gene-mediated cytotoxic immunotherapy, was found to be safe in the phase I clinical trial BrTK01 (Ref.95; Table 2). Subsequently, two phase II trials, BrTK02 (Ref.94) and HGG-01 (REF.96), have been conducted using intratumoural AdV-tk administration and valacyclovir or intra-arterial AdV-tk administration and GCV, respectively. Together, the results of these two trials demonstrated favourable PFS and overall survival outcomes associated with AdV-tk-based therapy (Table 2).
Oncolytic viral therapies based on the measles virus are another approach to the treatment of glioblastoma and are supported by promising preclinical data. Specifically, a measles virus engineered to produce carcinoembryonic antigen (MV-CEA), which serves as a serum marker for in vivo expression of the viral genome, caused the regression of flank tumours and markedly improved the survival of immunocompetent glioblastoma-bearing mice97. These findings led to a phase I clinical trial to investigate MV-CEA in patients with recurrent glioblastoma (NCT00390299; Table 2), although the trial has been suspended for unidentified reasons. No other trials of a measles virus have been initiated in patients with glioblastoma.
HSVs are also well understood, and the use of HSV strains genetically engineered to target cancer cells is another promising investigational approach to glioblastoma therapy98. Generations of HSV constructs have been developed by selectively attenuating genes to ensure that the virus predominantly targets replicating cells in the CNS99. These manipulations include thymidine kinase deletion, γ134.5 dual knockout, viral ribonucleotide reductase disruption, and lacZ gene insertions into the viral ribonucleotide reductase gene promoter; extensive preclinical work has demonstrated both the anti-glioma effects and low neurotoxicity of such viruses99. Trials of several modified HSV constructs, including G207 (NCT00028158, NCT00157703, and NCT02457845), HSV-1716 (NCT02031965), G47Δ, and M032 (NCT02062827), have been conducted or are ongoing in patients with glioblastoma (Table 2; Supplementary Table 2). The future direction of oncolytic viral therapies seems to be focused on combinations with immunotherapy strategies, such as those mentioned for DNX-2401, in the hope of exploiting the potentially durable anticancer immune responses initiated by the viral infection to elicit prolonged clinical responses.
Antibodies that reduce the activity of endogenous, negative regulatory pathways limiting T cell activation are arguably the most important advance in cancer therapy made in the past decade, with major improvements in the outcomes of patients with some difficult-to-treat cancers, such as advanced-stage melanoma or non-small-cell lung cancer100. To date, the most notable examples have been antibodies that block the inhibitory immune-checkpoint proteins cytotoxic T lymphocyte antigen 4 (CTLA-4) and PD-1, which are expressed on T cells, or PD-L1, which is expressed on certain subsets of immune cells and is aberrantly expressed on tumour cells of various histologies. Indeed, PD-L1 has emerged as a biomarker of sensitivity to immune-checkpoint inhibition with anti-PD-1 antibodies in the context of many solid tumours101,102,103. This ligand is also expressed in a subset of glioblastomas ranging from ~2% to ~88%, although the extent of this expression remains the subject of debate104,105. Moreover, higher PD-L1 expression in glioblastoma has been correlated with poorer patient prognoses in some studies105, which might be related to increased suppression of antitumour immunity, although this relationship was not detected in other studies104. Despite these contrasting findings, immune-checkpoint inhibitors have garnered considerable interest from the glioblastoma community, considering the profound immunosuppression that is characteristic of this disease. Preclinical studies have demonstrated the impressive activity of immune-checkpoint inhibitors as monotherapies or in combination with radiotherapy in mouse models of glioma106,107,108. Notably, however, caution in interpreting these data is prudent because orthotopic implantation of glioma cell lines with a high mutational load might not enable the accurate prediction of responses in patients with spontaneously arising glioblastomas109. Indeed, glioblastomas typically have a relatively low mutational load and a paucity of T cell infiltration compared with other tumour types41. Furthermore, the use of orthotopically implanted tumour models, typically involving the GL-261 and SMA-560 cell lines, might introduce variables such as violation of the blood–brain barrier.
The anti-PD-1 antibody nivolumab is the immune-checkpoint inhibitor for which clinical development has advanced furthest in patients with glioblastoma. Safety studies in this setting have revealed only grade 1 or 2 toxicities from nivolumab monotherapy, and the toxicities of combined treatment with nivolumab and the anti-CTLA-4 antibody ipilimumab were similar to those observed in patients with other tumour types110. In the ongoing phase III CheckMate 143 trial (NCT02017717; Supplementary Table 3), the efficacy of nivolumab is being compared with that of bevacizumab in patients with glioblastoma across different lines of treatment. Data from this study have not yet been published in a peer-reviewed journal, but preliminary data reported at the 2017 World Federation of Neuro-Oncology Societies meeting revealed that the primary end point of the trial was not met: the median overall survival of patients with recurrent disease was 9.8 months with nivolumab versus 10.0 months with bevacizumab7,111. In addition, this trial contained exploratory phase I cohorts that included combined nivolumab and ipilimumab treatment arms, in which high rates of serious adverse events were observed (adverse events leading to discontinuation occurred in 50% of patients across two nivolumab and ipilimumab groups112); thus, this combination strategy is not being pursued further in the phase III stage of this trial. The incidence and types of adverse events associated with nivolumab monotherapy seem to be similar in patients with glioblastoma and in those with other tumour types7,111,112, although the combination of this agent with TMZ/RT → TMZ was associated with an acceptable safety profile in exploratory cohorts comprising patients with newly diagnosed glioblastoma113. In two additional ongoing studies, CheckMate 498 and CheckMate 548 (NCT02617589 and NCT02667587, respectively; Supplementary Table 3), investigators are exploring nivolumab as an alternative to TMZ (both in combination with radiotherapy) in patients with MGMT-promoter-unmethylated tumours and as an addition to the standard TMZ/RT → TMZ regimen in patients with MGMT-promoter-methylated tumours, respectively.
Case reports have suggested that anti-PD-1 therapy can be effective for patients with glioblastoma. First, nivolumab was reported to result in long-term disease control in an adult patient with recurrent glioblastoma114. Second, a case report published in 2016 recounted the impressive and durable responses to nivolumab in two siblings with germ-line, biallelic DNA-repair defects and recurrent paediatric glioblastoma115. Finally, another adult patient with widely disseminated glioblastoma showed a response to pembrolizumab116. In the latter two case reports, the patients had tumours with a high mutation burden, which is a known predictor of response to immune-checkpoint inhibitors. Notably, in 2017, pembrolizumab was approved for patients with microsatellite instability-high or mismatch repair-deficient solid tumours, independent of histology117; however, only a small fraction of all patients with glioblastomas have mutations affecting the DNA mismatch repair machinery118.
Chimeric antigen receptor T cell therapy
In addition to vaccines, oncolytic viruses, and immune-checkpoint inhibitors, another interesting immunotherapy approach leverages genetically modified T cells (Table 3; Supplementary Table 4). T cells can be engineered to express chimeric antigen receptors (CARs), which consist of the antigen recognition domains of antibodies linked to T cell activation domains derived from the T cell receptor CD3 ζ-chain (CD3ζ) and co-stimulatory receptors (such as CD28 and/or TNFRSF9 (commonly known as 4-1BB))119. The antigen recognition domains impart CAR T cells with specificity for tumour-associated antigens, and these agents have shown promise in the treatment of glioblastoma. Furthermore, the appeal of this approach lies in the capacity of CAR T cells to recognize antigens that are not presented in the context of MHC molecules, as is typically required for adaptive immune responses. Additionally, the CAR T cells can be engineered to have an activated phenotype (for example, through co-expression of OX40 or 4-1BB)120,121. These multiple features can help CAR T cells overcome some of the immunosuppressive effects of a tumour microenvironment.
Brown et al.122 reported a case study in which a patient with recurrent glioblastoma underwent repeat resection of three of five intracranial lesions followed by an infusion of CAR T cells targeting IL-13Rα2, which is often overexpressed on glioblastoma cells. The findings of a previous phase I study of a prior version of the CAR T cell targeting IL-13Rα2 supported the safety of this approach, with a low risk of common complications of CAR T cell therapy, such as cytokine-release syndrome123,124. The patient reported in the follow-up study by Brown et al.122 had leptomeningeal disease and, therefore, the CAR T cells were administered intracranially via two routes: initial weekly infusions of CAR T cells into the resection cavity for 6 weeks, followed by intrathecal delivery into the ventricular system after the appearance of new lesions for a further ten infusions. A dramatic radiographical response was observed, with shrinkage of all lesions by 77–100%122. Nevertheless, 7.5 months after the initiation of adoptive cell therapy, the patient ultimately had disease recurrence122. While these results are encouraging, the recurrence indicates that perhaps the tumour underwent immunoediting and ultimately selected for cells that were negative for IL-13Rα2; indeed, preliminary results from tissue analyses support this hypothesis122. Another notable feature of this study is that the CAR T cells were administered into the cerebrospinal fluid (CSF) rather than intravenously or directly into the tumour, suggesting that different ways to administer immunotherapies should be considered125.
In 2017, data from the first ten patients of a first-in-human clinical trial of CAR T cells directed at EGFRvIII were published126 (NCT02209376; Supplementary Table 4). These ten patients with recurrent EGFRvIII-positive glioblastoma received a single intravenous infusion of autologous anti-EGFRvIII CAR T cells126. Notably, treatment was safe, with no incidence of cytokine-release syndrome or neurotoxicity126, which are common and life-threatening adverse effects of anti-CD19 CAR T cell therapy in patients with B cell malignancies124. The overall survival of the patients did not seem to be affected by CAR T cell therapy, with only one patient having disease stabilization (although lasting >18 months)126. Nevertheless, tumour infiltration of the CAR T cells was detected126. Furthermore, evidence of T cell activity against tumour cells expressing EGFRvIII was observed in tumour samples from seven patients who underwent surgical resections after CAR T cell infusion, with samples from five of the seven having decreased tumoural EGFRvIII expression and concomitant upregulation of immunosuppressive factors, including IDO1 and PD-L1, and increased numbers of regulatory T (Treg) cells in the tumour microenvironment126. Together, the findings of these early clinical studies suggest that glioblastomas can activate various adaptive responses to subvert anticancer immune responses and reinstate an immunosuppressive milieu; these escape mechanisms will need to be overcome if we are to improve the effectiveness of immunotherapy for this disease.
Overall, evidence from clinical trials of CAR T cells suggests that the engineered cells can infiltrate glioblastomas, become activated, and, in one patient, eradicate a considerable amount of malignant tissue122. However, in many solid tumour studies, CAR T cells alone had insufficient antitumour activity127. While the exact reasons for this observation are unknown, one could postulate that targeting one antigen in a highly heterogeneous tumour might not be sufficient to eradicate all cancer cells. Indeed, to date, the best results with CAR T cells have been achieved in patients with cancers that are highly clonal, such as leukaemias and lymphomas, resulting in approvals in these settings. Nevertheless, in most patients, CAR T cells will probably need to be administered in combination with other therapies, or CAR T cells targeting multiple different antigens will be required. Moreover, additional immunosuppressive pressures within the tumour microenvironment are likely to inhibit the anticancer activity of CAR T cells128. Thus, approaches to improving the efficacy of CAR T cells might involve promoting antigen spreading, combination therapies (for example, with immune-checkpoint inhibitors129), or targeting of immunosuppressive myeloid cells with agents such as IDO130 or macrophage colony-stimulating factor 1 receptor (CSF1R)131 inhibitors.
Immunotherapy and the current standard of care
With immunotherapy now firmly established in the management of a variety of malignancies and having demonstrated some promise in the treatment of glioblastoma, integrating immunotherapy into the current standard-of-care TMZ/RT → TMZ regimen is a critical next step to improving outcomes of patients with glioblastoma. The paradigm of administering systemic chemotherapy concurrently with radiotherapy demands particularly careful consideration, as this combination might have the unintended consequence of weakening the immune system by suppressing or permanently ablating critical immune cell populations42,132. Myelosuppression is commonly observed with chemotherapy in patients with various cancers, and the degree of immunosuppression is likely magnified in patients with brain tumours42,132: multiple aforementioned studies have found that glioblastoma can profoundly affect the peripheral immune system at baseline43,44.
Indeed, the use of hyperfractionated radiation and systemic chemotherapy with additional corticosteroids to prevent or control toxicities seems to cripple the immune system. Grossman et al.42 found that hyperfractionated radiation correlated with a marked depletion of CD4+ T cells in patients with glioblastoma, suggesting severe immunosuppression. These investigators also found that the degree of radiation-induced immunosuppression was a negative prognostic indicator of survival outcomes. Subsequent studies by the same group in patients with pancreatic cancer receiving only radiation revealed that lymphopenia is less severe in patients receiving stereotactic body radiotherapy than in those treated with conventional hyperfractionated radiotherapy133. The researchers hypothesized that the greater extent of lymphopenia in the latter group results from direct exposure of a larger number of circulating immune cells to radiation owing to the less-conformal delivery of radiotherapy133, supporting their previous findings in modelling studies relating to glioblastoma therapy134. These findings suggest that the use of shorter courses of radiation should be considered, particularly when immunotherapy is planned. TMZ can also cause lymphopenia and results in changes to immune cell populations that can even persist in patients with recovery of normal blood counts. For example, TMZ can negatively affect the number of memory T cells in a permanent fashion132. Finally, corticosteroids that are commonly used in the treatment of patients with glioblastoma can adversely affect the effectiveness of immunotherapies135. In fact, a corticosteroid-related gene-expression signature has been associated with unfavourable survival in patients with glioblastoma135,136. While many confounding factors might have affected the results of these studies, valid concerns exist and the effects of corticosteroids on immunotherapy should be studied further. Taken together, deleterious effects of glioblastomas on the immune system might be compounded by conventional therapies that are immunosuppressive and can ultimately hinder an effective immunotherapy strategy. Therefore, considering strategies including stereotactic radiosurgery (SRS) or local chemotherapy and the judicious use of corticosteroids might enable improved antitumour immune responses to be achieved (as discussed in a following section of this Review).
Biomarkers for glioblastoma immunotherapy
The lack of validated biomarkers is one of the current challenges in treating patients with glioblastoma. MRI remains the gold standard method for determining the disease burden, despite the well-documented challenges associated with accurately assessing disease status using MRI, including pseudoprogression and pseudoresponses137,138. Unlike for some systemic cancers, repeat tissue sampling, either through biopsy sampling or surgical resection, is fraught with a substantial risk of procedure-related morbidities in patients with glioblastoma. These challenges are compounded in the setting of immunotherapy, with reports of initial enlargement of tumour volume owing to immune infiltration (pseudoprogression), true progression before a delayed response, and contrast enhancement and associated oedema in patients with tumour types that are responsive to immune-checkpoint inhibition139,140 — including glioblastoma114,141.
Sporadic biomarkers have been reported to correlate with response to immunotherapy in glioblastoma. Many vaccine therapies using DCs have been and are currently being investigated (Table 1; Supplementary Table 1), but few biomarkers have been identified as prognostic indicators for immunotherapy. Everson et al.142 reported that increased responsiveness of patient-derived peripheral blood CD8+ T cells to immunostimulatory cytokines after DC vaccination was associated with prolonged overall survival and identified patients who survived for >2 years. IDO pathway flux, specifically the kynurenine:tryptophan ratio, has also been reported to be prognostic in patients with recurrent glioblastoma undergoing surgery and HSPPC-96 vaccine treatment143. In the rapidly expanding field of immune-checkpoint inhibitors, PD-L1 has emerged as a biomarker for response to anti-PD-1 or anti-PD-L1 therapy in multiple solid tumour settings, as discussed previously102,103,144. On the other hand, biomarkers for a response to antagonists of the immune-checkpoint protein CTLA-4, as well as PD-1–PD-L1 blockade, include a high mutational burden of tumours — although this association has been reported for non-CNS tumours, such as melanoma145, case reports have demonstrated a similar trend in patients with glioblastoma, as mentioned previously115. However, anti-CTLA-4 treatment will probably require the development of biomarkers that are different from those associated with PD-1–PD-L1 blockade because the effects of CTLA-4 inhibition are biologically distinct and abrogate global immunosuppression via effects on CD4+ effector T cells and Treg cell activity101. In general, the mutational burden of each individual glioblastoma might influence the patient’s response to immunotherapy. This point is highlighted by genomic characterizations of breast and ovarian cancers that have shown how deletions affecting the genes encoding members of the apolipoprotein B mRNA editing enzyme catalytic polypeptide (APOBEC) family of cytosine deaminases lead to hypermutated tumours that can prompt stronger immune system activation146. No studies of APOBEC mutations in glioblastoma have been reported, although one could assume that higher mutational loads will translate into a greater number of tumour neoantigens that are amenable to immune responses and thus facilitate immunotherapy147.
Given the surgical challenges associated with lesions located within the brain, liquid biopsy148 might be an attractive approach to analysing the genetic make-up of brain tumours. However, tumour-derived DNA and circulating tumour cells are rarely detectable in the blood of patients with CNS neoplasms149,150, prompting several groups to explore the potential of CSF as a reservoir for tumour-associated biomarkers149,151,152,153,154,155. Burgeoning data suggest that tumour-derived DNA in the CSF can be evaluated qualitatively and quantitatively and that the levels of this DNA are correlated with disease burden154,155. In a proof-of-concept study, Wang et al.151 demonstrated that whole-exome sequencing can be performed directly on DNA in CSF samples and can be used to explore the genetic landscape of brain tumours without the need for invasive procedures. This approach could potentially be a powerful asset in determining appropriate treatments. For instance, metabolomic analyses of these samples can identify the presence of IDH1 mutations156. Metabolic biomarkers might therefore have utility as surrogate indicators of responses to therapy and of disease progression in IDH-mutant tumours. One application of liquid biopsy to immunotherapy involves the ability to detect EGFRvIII mutations that could inform decisions relating to the use of vaccines or CAR T cell therapies152. Therefore, analysing CSF DNA has the potential to help select appropriate patients for immunotherapy and can be combined with conventional imaging studies to understand changes in disease burden and disease characteristics after therapy has been initiated; incorporation of CSF-based biomarkers should be an important component of future clinical trials.
The concept of cold versus hot tumours
Impressive responses to immunotherapy have been observed for a range of different tumour types, although certain tumours seem to be insensitive to the current approaches; the concept of ‘cold’ versus ‘hot’ tumours has been used to explain this observation. With the reporting of negative results from the CheckMate 143 trial of nivolumab after first relapse in patients with glioblastoma7,111, many are starting to believe that glioblastoma will fall into the category of cold tumours. The reasons that tumours are unresponsive to immunotherapy are likely multifactorial and include a highly immunosuppressive tumour milieu, defects in tumour antigen presentation, and features of the physical microenvironment such as hypoxia and necrosis.
Multiple findings support the hypothesis that glioblastoma is a cold tumour. Glioblastomas are known to have relatively few tumour-infiltrating lymphocytes (TILs) compared with other tumour types, suggesting that they are quiescent tumours in terms of immune reactivity157. Moreover, the TILs that are present have been shown to highly express markers of exhaustion, such as the inhibitory immune-checkpoint proteins T cell immunoglobulin mucin receptor 3 (TIM3; also known as hepatitis A virus cellular receptor 2) and lymphocyte activation gene 3 protein (LAG3)158. Interestingly, instead of TILs, high numbers of myeloid cells, such as microglia and macrophages, are present in the glioblastoma microenvironment and probably have predominantly immunosuppressive activities41. Furthermore, defects in the antigen-presentation machinery can make the tumour cold with respect to T cell-dependent immune responses. For example, β2-microglobulin (β2M) is an essential subunit of the MHC class I antigen-presenting complex, and studies have demonstrated that antitumour immune responses against tumour cells containing β2M mutations are attenuated159. Notably, mutations in β2M have been identified as an acquired resistance mechanism in patients with melanoma treated with pembrolizumab or ipilimumab159,160. Aberrations in the IFNγ–JAK–STAT pathway, which promotes antigen presentation and drives the expression of immune-checkpoint molecules including PD-L1, seem to confer both intrinsic and acquired resistance to immune-checkpoint inhibitors159,161. These mechanisms have been mainly shown in non-CNS tumours, although downregulation of MHC expression, defects in β2M, and mutations in JAK and STAT proteins are not uncommon in glioblastomas and suggest similar defects in antigen presentation162,163.
Finally, physical aspects of the glioblastoma microenvironment are thought to play important parts in attenuating antitumour immune responses. For example, one of the hallmark features of glioblastoma is necrosis, which seems to have a key role in immunosuppression. This necrosis often stems from the disordered tumour vasculature of glioblastoma, leading to areas of hypoxia. Eli et al.164 showed that tumour necrosis results in increased extracellular concentrations of potassium, which can inactivate tumour-infiltrating T cells and might, therefore, limit the activity of immunotherapy in patients with glioblastoma.
Novel and combinatorial approaches
The initial results with immunotherapy in patients with glioblastoma have been disappointing; however, combination therapies are being actively investigated in the hope of transforming glioblastomas into hot tumours. Given the rapidly expanding list of immunological targets implicated in the disease, as well as the large number of immunotherapeutic agents at various stages of development, the number of possible therapeutic combinations is prohibitively large for random testing. Instead, rationally designed approaches are critical to the development of effective treatment strategies. For example, Koyama et al.165 found that resistance to anti-PD-1 antibodies in a mouse model of lung cancer was correlated with the upregulation of TIM3. In addition, Kim et al.166 detected TIM3 expression in seven of eight human glioblastoma specimens and showed that the number of exhausted tumour-infiltrating T cells (positive for both PD-1 and TIM3) increased over time in a mouse model of this disease; durable tumour control could be achieved when mice were treated with both anti-PD-1 and anti-TIM3 antibodies.
Activation of multiple arms of the immune system is another potential approach to overcome immunosuppression. Lymphocytes are critical in mounting an antitumour immune response, although macrophages, NK cells, and suppressor cells (such as myeloid-derived suppressor cells (MDSCs) and Treg cells) could also be targeted in order to enhance antitumour immunity. The immune microenvironment of glioblastoma is characterized by an abundance of M2-polarized myeloid cells and Treg cells and a paucity of NK cells57,167,168,169. Strategies for activating the antitumour functions of macrophages, DCs, and microglial cells by polarizing them to the M1-like phenotype and inhibiting MDSCs include antibodies to CSF1R and IDO inhibitors, both of which are being investigated in glioblastoma models45,170 and in patients with this disease (NCT02052648). IDO inhibitors also have effects on Treg cell accumulation45,170, and antibodies that target glucocorticoid-induced TNFR-related protein (GITR; also known as TNFRSF18) provide another avenue for targeting this cell type, with supporting evidence coming from mouse glioblastoma models171. Finally, targeting NK cells with agents including the IL-15 superagonist ALT-803 has also shown promise in preclinical glioblastoma models172.
Preclinical data have demonstrated that other localized treatment modalities can synergize with immunotherapy108,166,171,173. The use of local therapies to increase the availability of tumour antigens and immunotherapy to drive an antitumour immune response provides the rationale for this combination approach. Currently, this paradigm is mostly predicated on combining immunotherapy with SRS108,166,171,173, as well as local chemotherapy, oncolytic viral therapy, and laser ablation (NCT02311582, NCT01811992, NCT01205334, NCT02197169, NCT02798406, and NCT02576665); however, many other modalities are being actively investigated, such as nanoparticle therapies and various therapeutic devices (NCT03020017, NCT00734682, NCT02340156, and NCT02766699).
The use of SRS has gained popularity across multiple tumour types, owing to the potential to promote antigen release, limit the systemic immunosuppression associated with radiotherapy, and thereby aid in the initiation of an antitumour immune response174. SRS has been shown to synergize with anti-PD-1 antibody therapy in orthotopic mouse glioblastoma models, resulting in increased numbers of activated CD8+ T cells expressing IFNγ and decreased numbers of tumour-infiltrating Treg cells compared with those observed with either treatment alone108. A durable survival benefit was achieved in mice with the combination approach, and this benefit was abrogated when CD8+ T cells were depleted108. Moreover, evidence of immunological memory was provided by a lack of tumour cell engraftment upon rechallenge of mice previously treated with SRS and anti-PD-1 therapy108. This paradigm has subsequently been expanded in preclinical models using focused radiation in combination with targeting of other immune-checkpoint molecules, such as CTLA-4, TIM3, GITR, and 4-1BB108,166,171,173. Currently, ongoing clinical studies are evaluating the use of SRS with anti-PD-1 antibodies in newly diagnosed and recurrent glioblastoma (NCT02648633 and NCT02866747; Supplementary Table 3), with case reports relating to other tumour types supporting the viability of this strategy175.
Other groups are examining the role of ablative therapies, such as laser ablation, in combination with immune-checkpoint inhibitors (Supplementary Table 3). With regard to laser ablation, catheters are stereotactically implanted into glioblastomas and the tumours are heated in a conformal manner to a temperature that causes tumour cell death while minimizing injury to nonmalignant tissue. In an ongoing phase I/II trial (NCT02311582; Supplementary Table 3), the safety and efficacy of MRI-guided laser ablation combined with pembrolizumab are being evaluated in patients with recurrent glioblastoma.
The approach to chemotherapy might need to be re-examined in the context of immunotherapy. The immunosuppression associated with TMZ is emphasized by the fact that clinicians are required to anticipate an increased susceptibility to infections176. Several studies have evaluated the effects of chemotherapy on the antitumour immune response in glioblastomas. For example, preclinical studies in a mouse glioblastoma model have revealed that the use of high doses of chemotherapy (either TMZ or carmustine) blunts the antitumour immune responses invoked by anti-PD-1 antibodies132. Furthermore, the immunosuppressive effects of chemotherapy seem to be long-lasting because mice that survived after treatment with a combination of systemic chemotherapy and immunotherapy or systemic chemotherapy alone failed to generate an antitumour immune response upon tumour rechallenge132. Interestingly, the animals treated with systemic chemotherapy could not be rescued with anti-PD-1 therapy at the time of tumour rechallenge132. Hence, the current strategy for immunotherapy studies in patients with glioblastoma, in which the experimental agents are used after or even together with TMZ chemotherapy, might be suboptimal. Notably, when the mice were treated with local chemotherapy, synergistic activity with immunotherapy was observed, resulting in a durable immune response132. Moreover, transfer of CD8+ T cells from these mice partially rescued the response of mice previously treated with systemic chemotherapy to tumour rechallenge132. The mechanism underlying these observations was hypothesized to involve the increased release of tumour antigens after local chemotherapy without systemic immunosuppression, resulting in augmented antigen presentation.
Thus, sweeping generalizations about the immune effects of systemic chemotherapy must be avoided. The important clinical nuance must be recognized, considering that patients are treated with intense doses of systemic chemotherapy and then treated with a maintenance regimen of chemotherapy on 5 days out of a 28-day cycle. Indeed, the effects of monthly TMZ on the immune system might not be the same as those observed with the initial continuous 6-week course of treatment. Heimberger, Sampson, and colleagues46,65 have studied the effects of monthly TMZ on the antitumour immune response in patients receiving the EGFRvIII vaccine rindopepimut and found that TMZ preferentially ablated Treg cells over effector cells, thereby suggesting that the anticancer immune response is not hindered during maintenance chemotherapy.
In addition, when interpreting these interesting and important immunological findings in mice, we must be cognizant of the limitations of such preclinical models; while the GL261 model is commonly used, the findings might not necessarily translate to human glioblastoma. Hence, higher-fidelity models of the disease are needed, including genetically engineered mouse models and humanized mice, and confirmatory data from human tumour specimens should be sought in order to inform the design of clinical studies.
Immunotherapy is clearly a revolution in cancer care. Dramatic responses have been observed across various tumour types with immunotherapies, particularly immune-checkpoint inhibitors and CAR T cells. Clearly, however, not all tumours are susceptible to current immunotherapies, and even among those patients who do have a response, the effects are not always durable. Where glioblastoma falls in the spectrum of responsiveness will soon become clear because several studies with large patient cohorts are set to mature. Early data suggest that most glioblastomas are cold tumours. Hence, ongoing studies are using combination approaches with the aim of making these cold tumours hot and thus augmenting current immunotherapy strategies. The data will guide the way in which immunotherapy is implemented as part of the standard of care for patients with glioblastoma.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
ClinicalTrials.gov database: https://clinicaltrials.gov/
The authors thank A. Wu of the Johns Hopkins University School of Medicine for her help in formatting the manuscript.Competing interests
M.L. has received research funding from Accuray, Agenus, Altor, Arbor, BMS, Celldex, and Immunocellular, and has been a consultant for Agenus, Baxter, BMS, Boston Biomedical, Oncorus, Regeneron, SQZ Biotechnologies, Stryker, and Tocagen. M.W. has received research grants from Acceleron, Actelion, Bayer, Merck (EMD), MSD, Novocure, OGD2, PIQUR, and Roche, and has received honoraria for lectures, advisory board participation, or consulting from AbbVie, BMS, Celldex, Merck (EMD), MSD, Novocure, Pfizer, Roche, Teva, and Tocagen. Y.X. and C.B. declare no competing interests.