Gliomas are primary brain tumours that are thought to derive from neuroglial stem or progenitor cells. On the basis of their histological appearance, they have been traditionally classified as astrocytic, oligodendroglial or ependymal tumours and assigned WHO grades I–IV, which indicate different degrees of malignancy. Tremendous progress in genomic, transcriptomic and epigenetic profiling has resulted in new concepts of classifying and treating gliomas. Diffusely infiltrating gliomas in adults are now separated into three overarching tumour groups with distinct natural histories, responses to treatment and outcomes: isocitrate dehydrogenase (IDH)-mutant, 1p/19q co-deleted tumours with mostly oligodendroglial morphology that are associated with the best prognosis; IDH-mutant, 1p/19q non-co-deleted tumours with mostly astrocytic histology that are associated with intermediate outcome; and IDH wild-type, mostly higher WHO grade (III or IV) tumours that are associated with poor prognosis. Gliomas in children are molecularly distinct from those in adults, the majority being WHO grade I pilocytic astrocytomas characterized by circumscribed growth, favourable prognosis and frequent BRAF gene fusions or mutations. Ependymal tumours can be molecularly subdivided into distinct epigenetic subgroups according to location and prognosis. Although surgery, radiotherapy and alkylating agent chemotherapy are still the mainstay of treatment, individually tailored strategies based on tumour-intrinsic dominant signalling pathways and antigenic tumour profiles may ultimately improve outcome. For an illustrated summary of this Primer, visit: http://go.nature.com/TXY7Ri
Brain tumours are characterized by high morbidity and mortality owing to their localization and often locally invasive growth. Most neoplastic brain lesions are metastases arising from cancers outside the central nervous system (which are 5–10-times more common than primary brain tumours)1. Gliomas and meningiomas are the most common types of primary brain tumour2,3 (Fig. 1). Gliomas account for almost 30% of all primary brain tumours, and 80% of all malignant ones, and are responsible for the majority of deaths from primary brain tumours. Tumours in this heterogeneous group are thought to arise from neuroglial stem or progenitor cells and are classified histologically into astrocytomas, oligodendrogliomas, mixed oligoastrocytic gliomas or ependymomas based on morphological similarities to the neuroglial cell types found in the normal brain (Fig. 2). Further classification occurs on the basis of location (for example, in the pons or optic nerve), characteristic differentiation patterns (such as pilocytic or myxopapillary) and features of anaplasia (including high mitotic activity, microvascular proliferation or necrosis). The absence or presence of anaplastic features is used for assigning malignancy grades from WHO grade I to IV, with WHO grade I indicating the least malignant behaviour. In adults, common gliomas include infiltrative astrocytomas of various grades (diffuse astrocytoma (WHO grade II), anaplastic astrocytoma (WHO grade III), glioblastoma (WHO grade IV)), oligodendrogliomas and the controversial group of mixed oligoastrocytomas. Other gliomas such as pilocytic astrocytomas, pleomorphic xanthoastrocytomas and ependymomas occur less frequently; patients harbouring such tumours tend to have a better prognosis. In children, the most common types of glioma are pilocytic astrocytomas and diffuse midline gliomas including diffuse intrinsic pontine gliomas of various grades.
In recent years, there has been substantial progress in understanding the molecular pathogenesis of gliomas both in adults and in children. These advances have resulted in improved diagnostics and classification systems based on mutational profiles, which will complement the histology-based classification. Furthermore, the improved understanding of the molecular pathogenesis of these tumours has identified new molecular targets and delineated therapeutic strategies that offer the potential for improved outcome in the near future. However, at present, classic prognostic factors — including patient age, general and neurological performance status and the extent of tumour resection — still determine outcome. In this Primer, we discuss the epidemiology, molecular pathogenesis, diagnosis and management of gliomas in adults and children.
The annual incidence of primary brain tumours in the United States between 2007 and 2011 was 21.4 per 100,000 individuals, with an incidence of gliomas of 6.6 per 100,000 people of which about half were glioblastomas3. The incidence (per 100,000 individuals) of other gliomas was considerably lower: 0.34 for pilocytic astrocytomas, 0.55 for diffuse astrocytomas, 0.36 for oligodendroglial tumours and 0.42 for ependymomas (Fig. 1). There is regional variation, with the incidence rate for gliomas in Japan being less than half of that in Northern Europe or the United States3,
Many environmental factors have been studied, but ionizing radiation (exposure to a therapeutic dose) is the only definite factor that is recognized as a causative agent6,
A relationship between the presence of allergy or atopic disorders with increased immunoglobulin (IgE) levels and decreased glioma risk has been suggested16; the underlying mechanism has not been elucidated. Although an inverse association between infection and glioma risk has been shown in some studies, this remains unproven. In addition, the link between cytomegalovirus titre and glioma risk that has been suggested in some studies has not been confirmed in others6.
Gliomas are associated with rare familial syndromes. Neurofibromatosis type 1 is associated with an increased incidence of pilocytic astrocytomas (in particular, optic pathway gliomas), neurofibromatosis type 2 is associated with spinal ependymomas and tuberous sclerosis is associated with subependymal giant cell astrocytomas (SEGAs). Furthermore, Li–Fraumeni syndrome shows an increased incidence of gliomas and Turcot syndrome shows an increased incidence of gliomas and medulloblastomas. Although biologically important, known genetic predisposition syndromes account for only ≤1% of all gliomas17.
Whole-genome sequencing in families with a history of glioma has identified mutations in protection of telomeres protein 1 (POT1), which is a member of the telomere shelterin complex that is involved in DNA and tripeptidyl-peptidase 1 (TPP1) binding; these mutations have been specifically linked to familial oligodendrogliomas18. Genome-wide association studies noted that single-nucleotide polymorphisms (SNPs) are associated with glioma risk19,20, which include a retinoic acid modulator CCDC26 on 8q24 (Ref. 21), pleckstrin homology-like domain family B member 1 (PHLDB1) on 11q23.3 (Ref. 22), the TP53 (tumour protein p53) polyadenylation site rs78378222 on 17p13.1 (Ref. 23), rs11979158 and rs2252586 in the epidermal growth factor receptor (EGFR) gene on chromosome 7 (Ref. 24), and rs4977756 in the cyclin-dependent kinase inhibitor 2A (CDKN2A)–CDKN2B locus19,20. The genetic variant rs55705857 at 8q24.21 has a strong association with isocitrate dehydrogenase (IDH)-mutant tumours and oligodendrogliomas with an odds ratio of approximately 6 (Ref. 21), which is rarely observed in low-penetrance association studies. This variant resides within the CCDC26 locus, but the mechanism of action has not been elucidated25. Genome-wide association studies have also established telomerase reverse transcriptase (TERT) and telomerase RNA component (TERC) — which are both involved in regulating telomere length — as interesting candidate genes for increased glioma risk26,27.
Similar to most other tumours, gliomas are essentially genetic diseases of single cells. The specific patterns of genetic alterations shape the clinical features of brain tumours and are increasingly used for classification and diagnostic purposes (Fig. 4).
Gliomas with circumscribed growth
This group of tumours includes three main entities — pilocytic astrocytoma, pleomorphic xanthoastrocytoma and SEGA — that preferentially manifest in children and young adults and show a slow and usually well-demarcated growth. The increased risk for pilocytic astrocytoma in patients with neurofibromatosis type 1 indicates a role of mutations in the neurofibromin 1 (NF1) gene — the product of which is a bona fide tumour suppressor that controls the mitogen-activated protein kinase (MAPK) pathway28. Recurrent genetic alterations in MAPK pathway genes other than NF1 mutations were also identified in sporadic pilocytic astrocytomas, most prominently truncating duplication and subsequent oncogenic fusion of the genes encoding the serine/threonine protein kinases BRAF or RAF1, activating BRAFV600E point mutations and point mutations in the genes encoding the GTPase KRAS, which might affect codons other than the classic ‘hotspot’ codons 12, 13 and 61 (Refs 29,30). In previously unexplained tumours, next-generation sequencing identified not only additional hits in the same pathway but also revealed no recurrent genetic hits in any other pathway, indicating that pilocytic astrocytoma is a single-pathway disease31,32. In support of the genetic data, activated BRAF is sufficient to cause gliomas in mice, which recapitulate the human disease in terms of morphology and benign growth behaviour33. Collectively, these findings open up attractive new avenues of targeted treatment. Specific MAPK pathway inhibitors (that target MEK1 (also known as MAP2K1) or MEK2 (also known as MAP2K2)) are currently under investigation for the treatment of relapsed pilocytic astrocytoma34.
In addition, targeting oncogene-induced senescence and driving tumour cells into apoptosis will be central translational research questions to address in the future. Indeed, oncogene-induced senescence provides a potential explanation for spontaneous phases of tumour stasis and halted tumour growth that may be observed during clinical follow-up, and it has been identified as a frequent phenomenon in pilocytic astrocytomas35,36.
Pleomorphic xanthoastrocytoma is a diagnostically challenging entity, as the high degree of cellular pleomorphism can be misinterpreted as evidence for malignancy. However, although most pleomorphic xanthoastrocytomas behave like other WHO grade II tumours, occasional cases may indeed undergo malignant progression. The histological criteria for grading are not well defined. Approximately 60% of pleomorphic xanthoastrocytomas are driven by activating BRAFV600E mutations37, which are frequently accompanied by homozygous deletion of CDKN2A and loss of CDKN2B38,39. In mouse models, the combination of BrafV600E and Cdkn2a loss in neural progenitors synergistically induces gliomas40,41.
SEGA is a hallmark lesion in patients with tuberous sclerosis, a rare genetic condition with autosomal dominant inheritance. Thus, inactivation of either tuberous sclerosis 1 (TSC1) or TSC2 — which encode hamartin and tuberin, respectively — owing to mutations and loss of heterozygosity is commonly seen in these tumours and drives tumour growth. This in turn causes aberrant activation of mammalian target of rapamycin (mTOR) signalling42. In line with this behaviour, loss of Tsc1 in mouse neural stem and progenitor cells resulted in SEGA-like lesions in the lateral ventricles43. Moreover, treatment with the mTOR inhibitor everolimus showed encouraging results in patients with tuberous sclerosis who have SEGA44,45.
Diffuse WHO grade II and III gliomas
Diffuse WHO grade II and III gliomas are the most common gliomas in young adults and are characterized by diffuse infiltration of the brain parenchyma and an inherent tendency to recurrence and malignant progression. Although histological classification traditionally separated astrocytic, oligodendroglial and mixed oligoastrocytic tumours, recent molecular profiling studies have revealed evidence for only two molecularly distinct subtypes. These are characterized either by mutations in TP53 that are often accompanied by a mutation of the α-thalassemia/mental retardation syndrome X-linked (ATRX) gene, indicating an ‘astrocytic’ genotype, or by co-deletion of chromosomal arms 1p and 19q associated with TERT promoter mutation, which are hallmark alterations of an ‘oligodendroglial’ genotype46,47. This molecular classification obviates the need for a mixed oligoastrocytoma subtype46. Diffuse gliomas of both genotypes carry mutations in IDH1 or, less commonly, in IDH2 (Refs 48,49). IDH mutation probably represents the tumour-initiating alteration in the vast majority of diffuse gliomas except for rare tumours that arise in patients with a germline TP53 mutation who secondarily acquire IDH mutations, in particular the otherwise rare IDH1R132C mutation50. IDH enzymes normally catalyse the oxidative decarboxylation of isocitrate to α-ketoglutarate, thereby generating NADPH from NADP+. By contrast, mutant IDH catalyses the NADPH-consuming reduction of α-ketoglutarate to R(-)-2-hydroxyglutarate. Thus, IDH-mutant gliomas show increased R(-)-2-hydroxyglutarate levels and decreased NADPH production, the latter of which probably predisposes cells to oxidative stress51. Increased levels of R(-)-2-hydroxyglutarate in IDH-mutant cells competitively inhibit the activity of α-ketoglutarate-dependent dioxygenases, including histone demethylases and the ten-eleven translocation (TET) family of 5-methylcytosine hydroxylases52. This inhibition in turn leads to increased histone methylation and hypermethylation of multiple CpG islands in the DNA, which is a characteristic epigenetic marker of IDH-mutant gliomas that has been referred to as the glioma CpG island methylator phenotype (g-CIMP)53,54 (Fig. 5). However, IDH mutation alone is not sufficient to cause glioma development55. Instead, IDH-mutant neuroglial stem or progenitor cells probably need to acquire additional genetic hits to transform into astrocytomas (TP53 and ATRX mutations) or oligodendrogliomas (1p/19q co-deletion and TERT promoter mutations). 1p/19q co-deleted oligodendrogliomas frequently carry mutations in the Drosophila homologue of capicua (CIC) gene on 19q13.2, whereas a mutation of the far upstream element-binding protein 1 (FUBP1) gene on 1p31.1 is restricted to a smaller subset of cases56,57. Malignant progression to secondary WHO grade IV glioblastoma, which is mostly restricted to tumours with an astrocytic genotype, involves homozygous deletion of the CDKN2A–CDKN2B locus and other genomic alterations, such as losses of the long arm of chromosome 10 (Refs 58,59). To date, only a few preclinical models are available for IDH-mutant gliomas with or without 1p/19q co-deletion60,
Overall, WHO grade II and III gliomas are rare in children and commonly lack IDH mutations as well as 1p/19q co-deletion in this age group. Duplications in the myeloblastosis (MYB) gene have been suggested as a prototypic feature both for diffuse astrocytoma and for angiocentric glioma in children64. At this stage, it seems unclear whether these are separate entities in children.
Glioblastoma (WHO grade IV)
WHO grade IV glioblastomas are the most frequent and most malignant gliomas that predominantly manifest in patients >50 years of age3 (Fig. 4). However, glioblastomas can also occur in children, adolescents and young adults, with increasing evidence that tumours in these age groups are genetically distinct from those in elderly patients. Recent large-scale profiling studies suggest that glioblastomas comprise at least six molecular subgroups that are characterized by distinct mutations and DNA methylation profiles65. These include two subgroups of paediatric glioblastomas characterized by missense mutations in the H3F3A gene (encoding histone H3.3) that affects either H3F3A K27, particularly in pontine and thalamic tumours with poor outcome, or H3F3A G34, mainly in hemispheric lesions and associated with global DNA hypomethylation65. A third glioblastoma subgroup is defined by an IDH mutation associated with g-CIMP and a proneural mRNA expression profile. This subgroup most commonly manifests in young adults, is linked to improved outcome and includes secondary glioblastomas derived by progression from pre-existing IDH-mutant WHO grade II or III gliomas. These H3F3A-mutant and IDH-mutant subgroups also frequently carry TP53 and ATRX mutations65. A fourth glioblastoma subgroup is molecularly defined by the platelet-derived growth factor receptor-α (PDGFRA) gene amplification and is most prevalent in adolescents and young adults. This subgroup has been referred to as ‘receptor tyrosine kinase I’ (RTK I) class65 and shares the proneural expression profile with IDH-mutant tumours. Most glioblastomas in patients >50 years of age correspond to the remaining two epigenetically distinct subgroups designated as ‘classic’ (or ‘RTK II’) and ‘mesenchymal’ classes65. These classes are characterized by distinct mRNA expression profiles (classic versus mesenchymal). In addition, the RTK II class demonstrates EGFR amplification more frequently than other glioblastoma subtypes65.
In general, about half of EGFR-amplified glioblastomas carry a genomic rearrangement characterized by deletion of exons 2–7, which encode key parts of the extracellular domain of the receptor. The constitutively active EGFR variant III (EGFRvIII) carries an extracellular epitope encoded by the fusion site of exons 1 and 8 that might serve as a tumour-specific therapeutic target, for example, for peptide-based immunotherapy. Other chromosomal and genetic alterations that are frequently detected in adult glioblastomas include copy number gains on chromosome 7, monosomy of chromosome 10, phosphatase and tensin homologue (PTEN) tumour-suppressor gene mutations, homozygous deletion of the CDKN2A–CDKN2B locus, and TERT promoter mutations66. Moreover, amplification of certain proto-oncogenes — in particular, CDK4 and CDK6, the murine double minute 2 (MDM2) and MDM4 genes or the hepatocyte growth factor receptor (MET) gene — occurs in subsets of the tumours67,68. Collectively, genetic alterations in glioblastomas converge on the aberrant activation of several tumour-promoting pathways, including the MAPK and protein kinase B (also known as AKT) signalling cascades as well as TERT activation or ATRX mutation. In addition, important tumour-suppressive pathways are impaired, in particular p53 and retinoblastoma protein 1 (RB1)-dependent regulation of apoptosis and cell cycle progression68. Moreover, rapid tumour growth leading to tissue hypoxia not only causes necrosis but also upregulates hypoxia-inducible factor 1 (HIF1) and HIF2, which increase the production of vascular endothelial growth factor (VEGF), the key factor driving aberrant angiogenesis in glioblastoma69. Another important feature of glioblastoma pathogenesis is the influence of the tumour on the patient's immune system, including suppression of an effective antitumour immune response, for example, by upregulation of transforming growth factor-β (TGFβ)70.
Diffuse intrinsic pontine gliomas
Diffuse intrinsic pontine gliomas are mostly high-grade astrocytic gliomas, including glioblastomas, that preferentially develop in children and are associated with poor outcome. The vast majority carry the K27M mutation in H3F3A or, less commonly, a mutation in the related histone H3.1 (HIST1H3B) gene, whereas activin A receptor 1 (ACVR1) mutations are present in approximately 20% of patients71,72. In addition, mutations in TP53 or the magnesium/manganese-dependent protein phosphatase 1D (PPM1D) gene, as well as ATRX and the death domain-associated protein (DAXX) gene, are frequent32,65,71. Recent preclinical data suggest a potential of epigenetic therapy in H3F3A K27-mutant tumours using specific inhibition of jumonji domain-containing protein 3 (JMJD3; also known as KDM6B), which functions as a demethylase of trimethylated lysine 27 on histone H3 (Ref. 73).
Ependymomas, although rare, pose a considerable challenge mainly because histological grading into WHO grade II or III tumours is poorly linked to patient outcome and because they are rather resistant to chemotherapy, limiting treatment options to maximal safe surgical resection and radiotherapy. Patients with neurofibromatosis type 2 are at increased risk for ependymomas, in particular spinal intramedullary tumours. In sporadic cases, somatic NF2 mutations are also prevalent in spinal intramedullary ependymomas74. No other significant genetic hits have been identified in sporadic tumours in any of the other anatomical compartments before the era of next-generation sequencing. Only recently, a highly recurrent fusion gene involving the nuclear factor-κB (NF-κB) downstream intermediate transcription factor p65 (encoded by RELA) and a previously uncharacterized fusion partner, C11orf95, was identified in most supratentorial ependymomas both in children and in adults75. In posterior fossa (PF) ependymomas, even when screening the entire exome for genetic hits, no single recurrently mutated gene was identified76. However, two distinct molecular subgroups of posterior fossa ependymomas — designated PF-A and PF-B — were identified based on transcriptional profiles and DNA methylation patterns77,78. These subgroups differ with regard to demographics, location within the posterior fossa, clinical course, cytogenetic events and aberrantly activated molecular pathways. Although PF-B tumours associated with good prognosis are possibly copy number driven and show a high degree of genomic instability, PF-A tumours associated with worse prognosis are characterized by a very stable genome. It is suspected that the PF-A tumours are primarily driven by epigenetic mechanisms76,
Further exploiting promoter methylation fingerprints as a very robust ‘memory’ of the cell of origin, a total of nine distinct molecular subgroups of ependymoma were identified across all three anatomical compartments (the spine, posterior fossa and supratentorial region) with three subgroups in each compartment. In the spine, in addition to myxopapillary ependymomas, there is a subgroup of classic ependymomas as well as a group enriched for subependymomas, which was also found in the posterior fossa and supratentorial region. PF-A and PF-B groups were confirmed in the posterior fossa. In the supratentorial space, RELA-positive tumours and a second robust subgroup characterized by YES-associated protein 1 (YAP1) gene fusions was identified79. In conclusion, apart from RELA–C11orf95 or YAP1 fusion-positive supratentorial ependymomas and NF2-mutant spinal intramedullary ependymomas, oncogenic drivers are yet to be identified for the remaining subgroups, and it is very likely that at least for some subgroups these drivers will not be found in the genetic code but may primarily involve epigenetic mechanisms. There is primary evidence that a vast majority of high-risk ependymomas comprises only two subgroups: namely, the RELA and PF-A subgroups79.
Diagnosis, screening and prevention
The clinical suspicion of a brain tumour is commonly based on the development of neurological deficits over a period of weeks to months or the new onset of seizures in a previously healthy individual. These symptoms demand a neurological work-up that includes a neuroimaging assessment. The imaging method of choice to detect a brain tumour is MRI without and with contrast enhancement (Fig. 6). CT is less frequently used today. The clinical value of other imaging techniques, such as magnetic resonance spectroscopy or PET, remains to be established. PET seems to be of value in tumour grading and can aid surgical planning80.
The definitive diagnosis of a glioma can only be made by histology81 (Box 1). Histological diagnoses based on tumour resections are more reliable than diagnoses based on biopsies because of limited tissue and the risk of sampling error and possible undergrading. However, we want to emphasize that, although many features inherent to morphology-based histological classification provide a useful stratification for clinicians to define risk groups, a classification system that is based on morphology alone leaves much unknown and incomplete, especially in the context of recent advances in determining molecular markers of gliomas82. The determination of several molecular markers is increasingly requested by clinical neuro-oncologists and implemented in treatment guidelines83. Of note, the salient molecular markers can also be determined in biopsy specimens84.
Molecular diagnostic markers
IDH mutation. A key conceptual understanding in the molecular pathogenesis of glioma was made with the identification and characterization of IDH mutations (Fig. 4). IDH mutations are much more common in WHO grade II and III gliomas (60–80%) than in glioblastomas (5–10%)48,85. Those glioblastomas that are IDH mutant tend to manifest in younger patients (≤50 years of age) compared with IDH wild-type tumours, and they usually lack the molecular profile of monosomy 10, trisomy 7 and EGFR amplification that is frequent in IDH wild-type glioblastomas. The consensus to date is that the previous characterization of glioblastoma as primary (that is, without clinical diagnosis of a lower-grade precursor lesion) versus secondary (that is, a glioblastoma progressing from a previously diagnosed WHO grade II or III lesion) correlates closely with the absence and presence of an IDH mutation, respectively86. However, not all patients with IDH-mutant glioblastomas have a history of pre-existing lower-grade gliomas, whereas progression from pre-existing IDH wild-type lower-grade gliomas to secondary glioblastomas may also occur, for example, in IDH wild-type anaplastic astrocytoma or pleomorphic xanthoastrocytoma. Additional work has led to the concept that, although IDH-mutant and wild-type gliomas cannot be discriminated under the microscope, they do in fact represent very different tumour entities in terms of molecular pathogenesis and potential therapeutic targets. For diagnostic purposes, the most common IDH mutation in gliomas, IDH1R132H, is readily detectable by immunohistochemistry using a mutation-specific antibody87. Direct (pyro)sequencing is most commonly used to detect other IDH1 or IDH2 mutations88 and is recommended for WHO grade II or III lesions and glioblastomas in young adults that are negative by IDH1R132H immunohistochemistry.
1p/19q co-deletion. IDH-mutant diffuse gliomas are subdivided into two main groups, which are defined either by 1p/19q co-deletion and TERT promoter mutation or by TP53 mutation and the loss of nuclear expression of ATRX89. Co-deletion of 1p/19q deserves special mention not only because of its association with oligodendroglioma histology and diagnostic value but also because of its clinical association with patient outcome in the context of genotoxic therapy: that is, its role as a predictive biomarker (see below). Diagnostically, 1p/19q co-deletion may be demonstrated by different techniques, with fluorescent in situ hybridization (FISH) and microsatellite analysis for loss of heterozygosity being most commonly used for routine diagnostic purposes90.
O-6-methylguanine-DNA methyltransferase promoter methylation. O-6-methylguanine-DNA methyltransferase (MGMT) is a DNA repair protein that counteracts DNA damage induced by alkylating agents. Methylation of the MGMT-associated 5′-CpG island results in transcriptional gene silencing and sensitivity to alkylating agent chemotherapy, and thus has become one of the most studied biomarkers in neuro-oncology since it was first described more than a decade ago91,92. Interest in this gene has increased considerably since the publication of an analysis of a Phase III glioblastoma trial showing that patients with MGMT promoter-methylated glioblastoma had longer survival when treated with temozolomide chemotherapy than patients with tumours without MGMT promoter methylation93,94. MGMT promoter methylation as a predictive marker to select patients for treatment was subsequently studied and confirmed in elderly patients with glioblastoma. Indeed, analyses of tissue samples from two randomized controlled clinical trials (RCTs) in elderly patients with malignant gliomas showed that patients with MGMT promoter-methylated tumours had better outcomes with chemotherapy than with radiotherapy95,96. By contrast, patients with MGMT promoter-unmethylated tumours had reduced survival when treated with chemotherapy. These results strongly suggest that MGMT promoter methylation should be used to decide on treatment strategies for the individual elderly patients with regard to the inclusion of chemotherapy. With respect to diagnostic testing, methylation-specific PCR and pyrosequencing of sodium bisulfite-modified DNA are the most commonly used methods92. In addition, the MGMT methylation status can be determined by array-based large-scale DNA methylation profiling approaches, such as the Infinium (Illumina) methylation bead array technique97. Advantages and disadvantages of each method and the importance of quality assurance have been addressed elsewhere92.
TERT promoter mutation. Mutations of the TERT promoter have emerged as one of the most common molecular markers in gliomas. These mutations result in increased gene expression and are found in >90% of IDH-mutant and 1p/19q co-deleted oligodendroglial gliomas, and also at very high frequency — exceeding 70% — in IDH wild-type glioblastoma98.
The markers discussed above show complex interactions and are not independent variables. As stated, essentially all 1p/19q co-deleted tumours are also IDH mutant, although the converse is not true. This requires an analysis of 1p/19q co-deletion, separate from IDH mutation, to be evaluated carefully. In addition, in the setting of IDH mutation-associated g-CIMP, MGMT promoter methylation is often present. Much of the data examining the predictive value of MGMT have been generated in IDH wild-type or g-CIMP-negative tumours, and it is likely that this represents the subgroup of glioma for which MGMT testing is most helpful99. These considerations led to the development of a simple algorithm of standardizing diagnostic and management approaches100 (Fig. 7): is there an IDH mutation? If so, is 1p/19q co-deletion present? Furthermore, in the case of IDH wild-type status, is there evidence for a malignant genotype, for example, as reflected by the +7q/–10q marker? It is very likely that such an approach will gradually replace or at least complement the current morphology-based WHO classification101,102. Another study103 demonstrates that adult gliomas categorized as WHO grades II–IV can be classified into one of five groups using the combination of just three markers: TERT promoter mutation status, IDH mutation status and 1p/19q co-deletion status. The five groups have different ages at diagnosis, different associations with germline-risk SNPs, different profiles of somatic mutations and distinct outcomes103.
EGFRvIII. Approximately 20–25% of IDH wild-type glioblastomas — corresponding to approximately half of tumours with EGFR amplification — have a characteristic deletion mutation of the EGFR gene, referred to as EGFRvIII, which results in constitutive and ligand-independent receptor activity104. Given that this deletion results in an internal, in-frame-altered tumour-specific allele, EGFRvIII is being explored as an antigen for immune-based therapies105. To date, vaccination trials directed against the peptide sequence at the rearrangement breakpoint have shown some activity for EGFRvIII-positive glioblastoma106. With further promising data, EGFRvIII could become a potential selection marker for the initiation of vaccine therapy.
BRAF. For pilocytic astrocytoma, truncating fusions of BRAF with different partners, predominantly with the previously uncharacterized gene KIAA1549, are almost diagnostic107. As breakpoints and fusion partners vary, it remains a clinical challenge to establish standardized molecular tests for routine diagnostic applications that cover all possible fusions. FISH seems to lack sensitivity. For the three main fusion types, various reverse transcription PCR-based tests have been established that also seem to work for formalin-fixed paraffin-embedded material108. Furthermore, truncating fusions of BRAF are almost universally found in cerebellar pilocytic astrocytoma but are less frequent in supratentorial tumours. In this compartment, BRAF point mutations, mostly BRAFV600E, as well as fibroblast growth factor receptor 1 mutations are more frequent than in the cerebellum. For pilomyxoid astrocytomas, which are grade II tumours according to the current WHO classification, an identical range of aberrations has been observed as in pilocytic astrocytomas, indicating that pilomyxoid astrocytomas are a somewhat less-differentiated variant of pilocytic astrocytomas rather than a separate entity109. BRAFV600E mutation is detected by immunohistochemistry with a mutation-specific antibody110 or by direct (pyro)sequencing. Demonstration of BRAF fusions is most often accomplished by FISH or reverse transcription PCR107,108.
Tumours of ependymal differentiation comprise a variety of clinicopathological entities and, although the full range of these entities is outside the scope of this review, there are major advances in the molecular understanding of these tumours79. Nevertheless, differential diagnosis primarily remains based on morphology in clinical practice today.
Screening and prevention
Screening for gliomas in the setting of a predisposing syndrome has been suggested: for example, for optic nerve gliomas in the setting of neurofibromatosis type 1 (Ref. 111). Screening using imaging, such as MRI, in the absence of a known predisposing syndrome is more controversial. Screening for less-malignant gliomas could be considered given that these are slow growing and have a large window of opportunity for detection and intervention. When tumours are small, a greater proportional resection with less functional risk is more readily achieved112. However, evidence thus far113 does not support systematic screening for gliomas in the general population, based on the low incidence and the cost involved. Furthermore, the high rate of ‘false-positive’ MRI scans — that is, MRI changes that do not represent glioma or another malignant brain tumour — does not support systematic screening. In this regard, a concern is the rate of complications as a result of potentially unnecessary neurosurgical procedures. Conversely, a normal MRI provides reassurance only for short time periods as gliomas can arise rapidly — within weeks to months — to considerable sizes.
Recent genome-wide association studies have identified potential regions and SNPs that might be associated with increased glioma risk. Such approaches, using screening with a blood test, might overcome some of the limitations and challenges of MRI-based testing; although much more research is required to identify the increased risk associated with specific alleles and haplotypes. Finally, evidence is lacking that early detection and treatment of glioma leads to a survival benefit.
The serious prognosis as well as the high personal and socioeconomic impact of glioma demand that patients with gliomas are managed in specialized centres according to institutional standard operating procedures and national and international guidelines (Table 1). It is desirable and established in many centres of excellence that a multidisciplinary board preoperatively outlines the treatment strategy for each patient. The neurosurgeon takes the lead and decides on the appropriate neurosurgical procedure that commonly aims at a maximal and safe resection in one or two (post biopsy) steps for each patient with a glioma irrespective of suspected tumour grade and without a priori minimizing care in the elderly patient population. Biopsies are preferred in patients with extensive or multifocal disease. Some patients with relatively benign tumours, as indicated by neuroimaging and growth dynamics, prefer a wait-and-scan strategy without surgical intervention or histological verification of the diagnosis. However, predictive tests to identify at presentation patients who are candidates for this strategy are not available. After resection, an early postoperative contrast-enhanced MRI is the standard of care. This should not be replaced by intraoperative scans because such scans might be taken before the last surgical procedure and will, for instance, potentially underestimate ischaemic complications114.
Histological diagnosis and a set of molecular markers — IDH, g-CIMP, MGMT and 1p/19q status in adults — guide further clinical decisions. The postsurgical tumour board should tailor the need for radiotherapy and/or chemotherapy, and explore treatment within an RCT. The main prognostic factors in neuro-oncology — performance status, mental status and age — are also taken into consideration. The need for multiprofessional management potentially involving nursing, speech, physical and occupational therapy, and psycho-oncology should be explored early in the care continuum, in addition to tumour-specific management. Patients are followed up clinically — which involves neurological examination, but increasingly also includes cognitive and health-related quality-of-life assessments — and using neuroimaging, typically MRI. A similar algorithm is followed once more if and when the disease progresses83.
Challenges for imaging include not only the diagnostic accuracy for unknown lesions but also the sensitivity and, even more importantly, the specificity of follow-up images in the context of therapy-induced changes, termed pseudoresponse and pseudoprogression115. Both are challenges for daily clinical practice but also for study assessments. For example, anti-angiogenic or other agents might reduce the contrast permeability of the blood–brain barrier and thereby mimic response. In addition, therapies involving irradiation or immunotherapies can cause changes in contrast enhancement, or T2 or fluid-attenuated inversion recovery (FLAIR) sequences, mimicking tumour progression. The most important developments in MRI include the development of common structured protocols within the response assessment criteria, which are undergoing continuous iterative improvements115,116 and provide algorithms to manage these ‘pseudochanges’ (Refs 117,118). Furthermore, multiple attempts are made to integrate perfusion and diffusion, as well as T1 contrast subtraction maps or susceptibility weighted images, to enhance the predictive accuracy of MRI119,
Beyond tumour-specific therapeutic approaches, steroids are commonly used to control tumour-associated brain oedema, typically in the perioperative period and at progression if there is major tumour burden. As steroids have major adverse effects — including depression, osteoporosis and immunosuppression — treatment should follow the simple rule: as much as necessary, as little as possible123.
Finally, given that roughly half of all patients with brain tumours experience seizures during the course of disease, anticonvulsant drugs are often administered. When choosing such drugs, their adverse effect profiles and particularly drug–drug interactions need to be taken into account. Whether certain anticonvulsant drugs have an impact on outcome — notably, valproic acid — remains controversial124.
Gliomas with circumscribed growth
These childhood or young adulthood tumours are potentially cured with complete resection. At recurrence, another resection might also be advisable if feasible. Radiotherapy, commonly to a dose of 50.4 Gy in 1.8 Gy fractions, is limited to incompletely resected progressive tumours. The recent observation that all pilocytic astrocytomas show an alteration in the MAPK pathway31 might open up new avenues for therapy, for example, by specific MEK inhibition, although pathway interference needs close observation for unwanted tumour-propagating effects125. SEGAs are often associated with tuberous sclerosis and do not necessarily need treatment but should be excised if symptomatic or progressive. The mTOR inhibitor everolimus is an effective medical treatment44,126. The finding of frequent BRAFV600E mutations in pleomorphic xanthoastrocytomas may offer a new target for therapy based on BRAF inhibition127,128.
Diffuse WHO grade II and III gliomas
Although the WHO grade is a prognostic factor and has been traditionally used to decide whether patients may potentially be managed by observation alone (WHO grade II) or require treatment (WHO grade III), recent evidence challenges the distinction between WHO grade II and III gliomas100,101. Histological grading may be difficult and is subject to interobserver variation129. If only small biopsy specimens are available, histology may underestimate the WHO grade because of regional tumour heterogeneity.
Gross total resection — defined as removal of all T2 or FLAIR and contrast-enhanced MRI abnormalities — is achieved in up to 40% of patients with WHO grade II gliomas130. In the other patients, lesions are either too extensive or involve eloquent areas that cannot be removed without causing neurological deficits. Gross total resection is associated with improved seizure control, particularly in patients with a long history of epilepsy and insular tumours131,132. Quality measures to ensure both optimal functional outcome and maximal extent of resection in addition to standard navigation or image guidance include brain mapping techniques and awake surgery133. Indeed, morbidity of glioma resections is reduced when brain mapping techniques are used because this enables precision in defining the margin of the tumour and functionally important brain tissue133. The positive impact of extent of resection on neurological function, quality of life, progression-free survival and overall survival is beginning to be defined; there is increasing evidence that more extensive resection reduces the risk of malignant transformation and improves survival132,134,135. A comparison of strategies of early surgical resection with watchful waiting for patients with WHO grade II gliomas at regionally defined treatment centres with presumably little patient migration suggested an important advantage for patients receiving early surgical resection. In addition, the rate of malignant transformation was also reduced136. The timing of resection after a diagnosis has been established is also controversial, especially in patients who are young, or those who present with an isolated medically controlled seizure or with small tumours. The risks of malignant progression of the tumour to a higher WHO grade — associated with the acquisition of additional molecular genetic alterations and the challenges of treating a more extensive lesion later on — need to be balanced against the surgical risk.
Similarly, the initiation of radiotherapy or alkylating agent chemotherapy or both after surgery to prevent progression is debated. In patients aged >40 years, the presence of preoperative neurological deﬁcits, larger tumours (>6 cm diameter), tumours crossing the midline137,138 or a WHO grade III as opposed to a WHO grade II tumour are adverse prognostic factors associated with higher risk of progression and death, thus determining whether further treatment is necessary83,139.The historically most-established treatment modality for patients with WHO grade II and III gliomas is radiotherapy to the involved part of the brain, which is commonly applied at 50.4 Gy in 1.8 Gy fractions for WHO grade II gliomas or 54–60 Gy in 1.8–2.0 Gy fractions for anaplastic gliomas (WHO grade III). However, the European Organization for Research and Treatment of Cancer (EORTC) 22845 Phase III study in patients with newly diagnosed WHO grade II gliomas demonstrated that deferring radiotherapy until tumour progression did not compromise overall survival compared with early radiotherapy140. Alternatives to regular external photon beam radiotherapy are hypofractionated (a similar biological dose in fewer but larger fractions) or hyperfractionated (a similar biological dose in more but smaller fractions) regimens, or interstitial treatments in which radioactive seeds are implanted (mainly for young patients, patients that might need sedation for each fraction of regular radiotherapy or patients with spherical, small tumours in deep locations). These alternative concepts are typically used by institutional preference in the absence of randomized data83.
Alkylating agent chemotherapy is an essential part of first-line therapy for patients with high-risk WHO grade II tumours and all patients with WHO grade III gliomas with 1p/19q co-deletion141,
For patients with 1p/19q non-co-deleted WHO grade III gliomas, the current CATNON trial explores the role of temozolomide concomitant with radiotherapy or as a maintenance treatment to delay recurrence, or both146. Guidance for individual treatment decisions may be derived from IDH mutation and MGMT promoter methylation status: for IDH-mutant, 1p/19q non-co-deleted tumours, radiotherapy or chemotherapy are most commonly used, whereas for patients with a glioblastoma-like, IDH wild-type tumour, many centres opt for temozolomide plus radiotherapy followed by temozolomide alone versus radiotherapy alone based on MGMT promoter methylation status99. Treatment options at recurrence follow similar considerations and depend on pattern of failure and the type of initial treatment83 (Fig. 7).
The identification of IDH mutations as an early event in tumorigenesis in the majority of WHO grade II and III gliomas49,85,107,145 has prompted attempts to therapeutically inhibit IDH in a mutation-specific manner63 or by using the DNA methyltransferase inhibitor decitabine147. Furthermore, mutations resulting in a defined altered amino acid sequence of the encoded protein have raised interest in their suitability as immunotherapeutic targets148,149. Conceptually, these altered protein sequences may represent neo-epitopes that are suitable as tumour antigens. As for all cancer vaccines, particularly for gliomas, which are characterized by a profound immunosuppressive microenvironment shaped by glioma-derived immunosuppressive factors, translation into clinical trials ought to consider combinations with pharmacological approaches that target key immunosuppressive pathways in gliomas150,151.
Glioblastoma (WHO grade IV)
The standard of care for adult patients (aged up to 70 years and in good general and neurological condition) with newly diagnosed glioblastoma is surgery to the greatest extent of resection feasible, followed by concomitant and adjuvant temozolomide plus radiotherapy followed by temozolomide alone83,152. The goal of complete macroscopic resection may be reached easier with the help of novel techniques, such as fluorescence-guided visualization of tumour tissue with the aid of 5-aminolevulinic acid153,154. Either radiotherapy alone, or temozolomide alone or temozolomide plus radiotherapy followed by maintenance temozolomide, are the standard of care for elderly or frail patients without or with MGMT promoter methylation, respectively95,96. Neither dose-intensified temozolomide155 nor anti-angiogenic approaches using the integrin inhibitor cilengitide156 or the VEGFA-specific antibody bevacizumab157,158 have been shown to improve overall survival. The molecular genetic classification of glioblastoma summarized above65 has so far had no influence on the current standards of care.
Surgery and radiotherapy are the key therapeutic principles for patients with ependymoma. Surgery should be as radical as safely possible, and extent of resection is a prognostic factor in any localization. As these tumours are relatively radiosensitive, the timing, dose and extent of radiotherapy are a matter of debate. Timing is critical because especially young patients (50% are children) are likely to live long enough to be at risk for potential late neurotoxic effects. Extent and dosing of postsurgical radiotherapy are summarized in Table 2 (Refs 159,
Quality of life
Diagnosis of glioma may result in profound effects on quality of life owing to both the tumour itself and as a consequence of its treatment (Box 2). Depending on location and size, quality of life can be affected in terms of functional deficits, including motor dysfunction, impaired communication ability or decline in neurocognitive function; ultimately, the general health and physical and social functioning of the patient are also affected167. Surgery may be associated with acute functional deficits, whereas radiotherapy poses a major risk of long-term neurocognitive impairment, notably in patients with longer survival168. Chemotherapy may cause haematological and other acute toxicities. Common to all treatments is chronic fatigue169. Furthermore, commonly used routine medications such as corticosteroids and anti-seizure agents have their own set of well-recognized adverse effects, including drowsiness, fatigue, muscle weakness, weight gain and nausea. Long-term adverse effects of corticosteroid use include depression, osteoporosis, diabetes mellitus and hypertension. In addition, quality of life is likely to be affected in non-physical domains, including psychological, spiritual, behavioural and existential dimensions. Finally, families and care-givers can be profoundly affected by the illness, which can negatively impact their own quality of life.
There are several validated instruments used in clinical trials to assess quality of life in patients with brain tumours. Objective neurocognitive testing using defined sets of batteries170,171 needs to be distinguished from patient-reported outcomes; these measures frequently show different results. The EORTC QLQ-30 questionnaire has been supplemented by a brain cancer-specific module (BN-20) with 20 additional questions. Similarly, the Functional Assessment of Cancer Therapy-General (FACT-G) questionnaire and its brain module cover concerns that are relevant to patients with brain tumours172,173. Nevertheless, self-reporting can be difficult, and a patient-by-proxy evaluation may be of value or may be used instead. Indeed, in a situation of symptomatic tumour progression, formal quality-of-life evaluation may not be feasible. Some limitations of standardized quality-of-life evaluation became evident in two similar Phase III trials that evaluated the value of bevacizumab in addition to standard chemoradiotherapy in patients with newly diagnosed glioblastoma157,158. Although the same quality-of-life instruments were used in part, the studies came to opposite conclusions. The main limitations in all quality-of-life evaluations are the increasing frequency of missing data over time and the discontinuation of quality-of-life assessments at the time of tumour progression. Thus, although quality-of-life studies add meaningful information on clinical benefit above and beyond a progression-free or survival benefit, they add substantial complexity to clinical trials, and their interpretation can be difficult174. Increasingly, the care of patients with gliomas has a focus on quality of life with early supportive and palliative care interventions175, and family and carers need to be considered in this setting.
Research into the origin, classification and treatment of gliomas is evolving in many different directions. Beyond genetics and radiation exposure, no risk factors have emerged, but advanced genetic profiling may disclose not only novel risk factors but also protective traits, for example, in the field of immune responses including allergy.
Indeed, advances in the understanding of gliomas at a molecular level along with technological progress have led to the identification of key genetic alterations in glioblastoma. Panel, whole-exome or whole-genome sequencing is increasingly used to identify targeted agents, single or in combination, according to a dedicated algorithm to restrict treatment to patients with a glioma harbouring a specific alteration or signature; these approaches also have a key role in diagnostics. Concepts currently applying these molecular entry analyses are looking at inhibitors of phosphatidylinositol 3-kinase (PI3K), fibroblast growth factor or CD95 (also known as TNFRSF6). Given the extensive intratumoural heterogeneity and secondary changes over time176, as well as practical hurdles with the development of predictive biomarkers and combination treatments facing unwanted effects, the era of precision medicine (that is, targeted treatments) for patients with gliomas is only in its infancy. A selection of novel therapeutic approaches that are currently being explored is provided in Table 3.
In the future, several key issues will need to be resolved to successfully translate the new molecular findings into improved clinical management. For example, concerning brain tumour diagnostics, the morphology-based WHO classification (Box 1) needs to be revised by integrating molecular markers and histological features for better definition and eventually more precise classification of distinct brain tumour entities100,177. Standardization and quality control of molecular testing procedures will have to be implemented. The recent development of new approaches based on 450k DNA methylation bead array hybridization and next-generation sequencing of brain tumour-tailored gene panels, will have to be evaluated for widespread applicability and cost effectiveness in the routine diagnostic setting. Further challenges in interpretation include spatial and temporal heterogeneity when assigning molecular signatures to complex somatic genetic diseases such as gliomas176,178,179.
Immunotherapy is coming of age in neuro-oncology. There is now great hope for efficacy of various strategies based on immune checkpoint inhibition or active cellular immunotherapy, with two Phase III trials (ACT IV180, which is assessing a peptide that mimicks EGFRvIII106; and DCVax using tumour cell lysate co-incubated dendritic cells181) and various Phase II trials awaiting maturation of data or completion of recruitment. ICT-107 is another promising vaccine using peptides from several tumour-associated proteins (melanoma-associated antigen 1 (MAGE1; also known as MAGEA1); absent in melanoma 2 (AIM2); HER2; tyrosinase-related protein 2 (TRP2; also known as DCT); glycoprotein 100; and interleukin-13 receptor-α2) with pulsed peripheral blood mononuclear cell-derived dendritic cells182.
Furthermore, some non-mutated, tumour-associated peptides have been described to be shared to a high degree between different glioblastomas183 as their presentation may mainly depend on deregulated signalling pathways. The term actively personalized vaccines (APVACs) was coined by the Regulatory Research Group of the Association of Cancer Immunotherapy184. The basic principle that is followed for the composition of APVAC drug products is that the most relevant (that is, peptides with the highest tumour association) and, at the same time, the most immunogenic peptides should be selected for each patient to maximize the number of effective antitumour immune responses. This principle is followed in the current GAPVAC trial185.
The glioblastoma microenvironment is a hostile environment for antitumour immune responses. T cell function is paralysed by glioma-derived factors, such as TGFβ70,186 and catabolites of the essential amino acid tryptophan187, or by T cell suppressive pathways mediated by cytotoxic T lymphocyte antigen 4 (CTLA4) or programmed cell death protein 1 (PD1; also known as PDCD1)–PD1 ligand 1 (PDL1; also known as CD274 and B7H1). On the basis of promising results from other tumour entities, using antibodies against CTLA4 (such as ipilimumab188 and tremelimumab189) or those that target PD1 or PDL1 signalling (such as nivolumab or pembrolizumab), among others190, there may be an option to combine active immunotherapy with agents that block the immunosuppressive microenvironment in patients with glioblastoma to enable a peripheral antitumour immune response induced by vaccination to become effective.
Although the potential role of cytomegalovirus in glioblastoma continues to divide the scientific community, recent experimental and clinical data — which are compelling but still preliminary — raise the possibility of cytomegalovirus-targeted immunotherapy191. Although promising in terms of outcome for selected groups of patients, questions remain regarding the feasibility of immunotherapy in subgroups of patients, notably elderly patients with major tumour burden.
Oncolytic viruses also continue to be explored as therapeutic ‘weapons’ against gliomas, including poliovirus and measles virus. Some of these treatments have resulted in key inflammatory responses that may contribute to clinical efficacy or may even be the dominant mode of action of infective therapies192.
With respect to other emerging treatments, data from the interim analysis of the Novocure Phase III glioblastoma trial suggest that tumour-treating electric fields at 200 kHz applied through skin electrodes on the bald head could be an option for patients with newly diagnosed glioblastoma in addition to the standard of care193.
In general, it will be necessary to design new types of RCTs that enable clinical evaluation of biomarker-driven individualized approaches based on rather small patient cohorts194. Neuroimaging approaches for diagnostic work-up and follow-up need to be standardized not only to improve cross-trial comparisons but also to ensure access to best practice oncology to more parts of the world. The value of novel outcome measures — notably, preservation of cognitive function and patient-reported outcomes — needs to be weighed against the classic outcome parameters in oncology of response to treatment, progression-free survival and overall survival.
The development and clinical validation of new compounds that specifically target glioma-associated aberrant genes and pathways is mandatory. Eventually, combinatorial drug approaches targeting complementary oncogenic signal ling pathways — such as those involving the VEGF receptor, EGFR, MET or PI3K — will have to be developed and evaluated that promise synergistic effects on tumour growth based on predictions drawn from patient-specific mutational profiles. Moreover, the glioma microenvironment is a new emerging therapeutic target.
All of these issues can only be successfully addressed in close collaboration between neuropathologists, basic scientists including molecular biologists and immunologists, bioinformaticians and clinicians with a multidisciplinary approach, and tailored treatment strategies that are pursued in RCTs will require international collaboration.