Intertumoral and Intratumoral Heterogeneity as a Barrier for Effective Treatment of Medulloblastoma

Brain tumors remain the leading cause of cancer-related mortality in childhood.1,2 Medulloblastoma is the most common malignant pediatric brain tumor, representing up to 20% of newly diagnosed central nervous system malignancies.3,4 Current therapy consists of surgical resection, craniospinal irradiation with a posterior fossa boost, and adjuvant chemotherapy.5 Although overall survival in these patients has reached 70% to 80%, a high degree of treatment-induced morbidity remains.6-8 Furthermore, patients presenting with metastatic dissemination or those who later relapse still experience significant mortality.9 The past decade has seen an ever-growing interest in the molecular characterization of medulloblastoma, with the recent delineation of 4 core subgroups of medulloblastomas, each with distinct molecular, demographic and clinical features.10-19 The identification of these subgroups underscores the significant intertumoral heterogeneity that exists between patients and highlights the need to treat medulloblastoma subgroups as 4 distinct diseases. Following the advent of next-generation sequencing technology, 4 recent publications further highlight the critical importance of examining medulloblastoma in a subgroup-specific manner.20-23 These sequencing efforts presented the first comprehensive examination of the mutational landscape of medulloblastoma across a combined cohort of > 310 patients. Their findings offered insights into the cellular origin of the disease and the hijacked developmental pathways that contribute to the initiation, maintenance, and progression of medulloblastoma. Furthermore, in an effort to dissect the heterogeneity of medulloblastoma, an international collaboration was established, the Medulloblastoma Advanced Genomics International Consortium (MAGIC), comprising > 50 centers from around the globe. In its seminal work, the MAGIC consortium was able to analyze 1087 primary medulloblastoma samples at the copy number level to identify putative somatic driver events. This collective effort has led to the identification of novel actionable targets that can be pursued in future clinical trials and presented the first recurrent gene fusion in medulloblastoma.24 Besides the intertumoral heterogeneity appreciated in these large series of medulloblastomas, evidence suggests additional complexity within a given patient’s tumor. Intratumoral heterogeneity arises from the clonal architecture of the tumor and results in a bicompartmental disease: the primary tumor and the metastatic disease.25 This review highlights recent efforts to understand the intertumoral and intratumoral heterogeneity in medulloblastoma. Furthermore, this review highlights one of the most pressing challenges in treating medulloblastoma, which is leptomeningeal metastases arising from divergent clonal selection from the primary compartment. Thus, this review aims to discuss the remaining challenges and to offer clinical insights into future trial design and drug development. CURRENT PATIENT RISK STRATIFICATION BASED ON METASTATIC STATUS AND HISTOLOGY AND THE NEED FOR FURTHER CLINICALLY RELEVANT SUBCLASSIFICATION Patients presenting with medulloblastoma are currently stratified into different treatment arms based on clinicopathological variables, including extent of surgical resection, age at diagnosis, and metastatic status.26-28 Patients with significant residual disease, presenting at < 3 years of age, and having dissemination beyond the primary site are classified as having high-risk disease; the remainder are considered to be at standard risk. After treatment consisting of maximal safe resection, craniospinal irradiation, and adjuvant chemotherapy, cure rates have peaked at 85% for average-risk patients and around 70% for high-risk patients.1,5,29 Regrettably, significant clinical sequelae remain, including neurocognitive, endocrine, and psychosocial deficits, as well as an increased risk of future malignancies.6-8 Histological classifications have been used to inform clinical outcomes. According to the World Health Organization Classification of Tumors of the Central Nervous System, 5 major histological variants exist: classic, desmoplastic, large-cell, anaplastic, and medulloblastoma with extensive nodularity.30 Of these variants, classic histology is the most prevalent, whereas desmoplastic/medulloblastoma with extensive nodularity histology has the best survival outcomes.31,32 There are several shortcomings with respect to histological classification. Primarily, discrepancies between pathologists in the determination of anaplasia provide a significant challenge when guiding clinical decisions or generalizing the results of clinical trials to the community. Furthermore, although > 70% of medulloblastomas exhibit classic histology, response to current treatment strategies is highly variable, indicating remarkable biological heterogeneity within this group of patients. On the basis of these limited prognostic criteria, some patients may be overtreated and result in unnecessary morbidity. Several molecular genetic markers have been proposed to refine clinically based stratification, including MYC, ERBB2, CTNNB1, and TP53.33 Several chromosomal aberrations such as monosomy 6, isochromosome 17q, and loss of 17p raise further possibilities for clinical diagnostics.34-36 β-Catenin nuclear positivity and MYC amplifications are currently being used to assist in defining standard-risk and high-risk patients, respectively, in addition to clinical features. However, the inclusion of robust molecular markers and cytogenetic aberrations in current clinical staging remains sporadic, and their prognostic effects remain controversial. This prognostic variability has motivated the development of molecularly based strategies in stratifying patients according to their transcriptomes. Patients with similar gene expression profiles exhibit similar biological behavior, providing a rational scheme for the design of targeted therapies. This paradigm shift in the approach taken to classify patients has given rise to the identification of 4 distinct subgroups of medulloblastoma. DELINEATION OF 4 SUBGROUPS OF MEDULLOBLASTOMA Integrated genomic profiling of large numbers of primary tumor samples was first popularized through the subclassification of lung, breast, and glial malignancies, among others.37-41 More recently, this strategy was adopted by the medulloblastoma community. In 2002, Pomeroy et al42 successfully demonstrated that desmoplastic medulloblastoma was distinct from classic medulloblastoma and that the atypical rhabdoid tumor is a distinct biological entity. Furthermore, using a gene signature, this model was more accurate in outcome prediction than conventional metastatic status. Subsequently, multiple groups around the world have published on this topic, with studies spanning several medium (n = 46) to large (n = 194) cohorts of patient samples.10-15,17,23,43-47 These studies concluded that medulloblastoma does not represent 1 disease but rather comprises at least 4 distinct molecular subgroups (Table 1).TABLE 1: Clinical and Histopathological Features of Different Subgroups of MedulloblastomaaA recent global consensus was reached validating the existence of 4 core medulloblastoma subgroups. The first 2 subgroups are called WNT and SHH on the basis of their aberrant pathway activation. The other 2 subgroups are less characterized and hence are generically called groups 3 and 4. Since their discovery, these findings have transformed our collective view of medulloblastoma and are already beginning to revolutionize how this disease is studied both at the laboratory bench and at the hospital bedside.15,48,49 A meta-analysis of 550 medulloblastomas from 7 independent studies using gene expression approaches concluded that independently of current stratification features, subgrouping is overwhelmingly robust in terms of prognostication.43,44 The WNT subgroup, the smallest of the 4 subgroups, is characterized by the activation of WNT signaling and confers the best prognosis. WNT-driven tumors frequently have loss of chromosome 6, harbor β-catenin activating mutations, and exhibit classic histology.33,50,51 The SHH subgroup is characterized by aberrations in the Sonic Hedgehog signaling pathway, with mutations converging on signaling intermediates, including PTCH1 and SUFU.28,52,53 These patients exhibit intermediate prognosis with a cytogenetic profile including predominantly loss of chromosome 9q and, in much lower frequencies, 17p loss, 3q gain, and 10q loss.43 Interestingly, these patients exhibit a bimodal age distribution predominantly affecting infant and adult patients but infrequently children. Unsupervised clustering from 4 independent transcriptomic data sets demonstrated that adult SHH patients segregate distinctly with infant SHH patients.28 These traits are maintained at the molecular and cytogenetic levels. For example, homeobox family genes are upregulated in adult SHH tumors, whereas neuronal development genes such as ID2, ZIC5, and ZIC2 are overexpressed in infant cases.28,43 In terms of cytogenetics, pediatric patients have a preponderance of MYCN amplification and chromosome 10q deletion compared with adult cases. Further molecular characterization of SHH medulloblastoma is warranted and will likely reveal additional heterogeneity. Group 3 remains the subgroup with the worst prognosis, with an overall 5-year survival of < 50%. Frequently of the large-cell/anaplastic histology, 50% of group 3 tumors are metastatic at presentation, and many exhibit MYC amplification.13,54 Group 4 disease consists of classic medulloblastoma patients in that it targets mainly boys with classic histology, with very frequent isochromosome 17q in the tumor, and exhibits an intermediate survival of about 75% with multimodal treatment12,13,43 (additional cytogenetic aberrations summarized in Table 1). It is clear now that in addition to histology and clinical features, molecular subgroups will be a mainstay in risk stratification. Further research using more sensitive technologies may reveal additional heterogeneity and subtypes within the subgroups. Functional validation of these subgroups is ongoing, and already recent studies have shown distinct cells of origin among the different subgroups.46,55 This body of literature also underscores the heterogeneity and the need for subgroup-specific therapy. The fact that these subgroups have unique clinical features and underlying biology strongly suggests that these patients need to be treated differently. Smoothened inhibitors, which target the Sonic Hedgehog pathway, are currently in phase II clinical trials for recurrent medulloblastoma and will likely be considered in the future for upfront targeted therapy of SHH subgroup patients.56,57 WNT subgroup patients, for example, may benefit from therapy de-escalation to reduce treatment-induced morbidity, whereas group 3 patients need to be prioritized for novel therapeutic interventions.44 RECENT GENOMIC ADVANCES REVEAL NOVEL SUBGROUP-SPECIFIC MOLECULAR ALTERATIONS The availability of next-generation sequencing and integrated genomics has recently provided significant insight into medulloblastoma pathogenesis and further given credence to subgroup-specific classification of the disease.20-22,58 These studies have also validated previously identified mutations while unveiling previously unappreciated novel driver genes.23,59-61 An examination of somatic copy number aberrations across the largest assembly of medulloblastoma samples in the world was performed and revealed several novel actionable targets in a subgroup-specific manner (Table 2).24TABLE 2: Molecular Features of Different Subgroups of MedulloblastomaaWNT Subgroup Medulloblastoma WNT subgroup medulloblastoma remains quite genomically bland with an absence of any focal recurrent somatic copy number aberrations. The most common somatic mutation is CTNNB1, which confirms the importance of β-catenin and downstream WNT signaling.43,62 Other recurrent somatic mutations found within this subgroup include TP53; the DEAD-box RNA helicase DDX3X, involved in cellular growth and division; and chromatin-modifier SMARCA4. The developmental origin of WNT tumors has recently been suggested to be the progenitor cells of the lower rhombic lip. A mouse model harboring activated Ctnnb1 in the Blbp-expressing radial glial cells in a Trp53 heterozygote background with activated PI3K signaling (Blbp-cre; Ctnnb1+/lox(Ex3); Trp53+/flx; Pik3caE545K) developed highly penetrant WNT medulloblastoma.46 SHH Subgroup Medulloblastoma Activation of Sonic Hedgehog signaling has long been known to play a pathogenic role in medulloblastoma. Somatic mutations targeting the SHH receptor PTCH1 are found only in this subgroup. Other somatically mutated genes include TP53 and MLL2 in 14% and 12% of patients, respectively.44 SHH medulloblastoma also exhibits frequent somatic copy number aberrations, especially targeting genes involved in the PI3K signaling cascade.24 This finding is especially notable in that inhibitors of the PI3K pathway are widely available and may be used in combination with SHH inhibitors to treat this disease variant.56,63,64 Several mouse models of this subgroup exist, targeting both the cerebellar granule neuron precursors and the neural stem cells located in the subventricular zone.65-67 Specifically, inactivation of patched in either the Atoh1 (marks cerebellar granule neuron precursor cells) or GFAP (marks neural stem cells) compartment leads to medulloblastoma development.55,68,69 Group 3 Medulloblastoma Because this molecular subgroup of medulloblastoma is associated with the worst prognosis, a great deal of attention has been focused on understanding the pathogenesis of group 3 medulloblastomas. This subgroup frequently harbors genomic instability and high-level amplification targeting the proto-oncogene MYC.43 Somatic mutations in this subgroup appear to converge on deregulation of the epigenome, mainly SMARCA4 and MLL2, validating previous work done using traditional Sanger-based sequencing method.23 More interesting is the identification of a novel PVT1-MYC fusion, the first recurrent gene fusion identified in medulloblastoma.24 This discovery offers new insights into the complex function of MYC in driving tumorigenesis. Furthermore, the finding that > 20% of group 3 tumors have aberrant transforming growth factor-β signaling provides a new opportunity for therapeutic intervention in these aggressive tumors. Two recent publications presented the first mouse models of group 3 medulloblastoma using orthotopic transplantation.70-72 In both cerebellar granule neuron precursor and neural stem cell populations, the activation of myc with concomitant p53 inactivation was able to generate the first group 3 preclinical mouse models. Group 4 Medulloblastoma Group 4 disease is characterized by amplification of MYCN (in addition to the SHH subgroup) and isochromosome 17q. Prior gene expression studies have reported that this particular subgroup exhibits a neuronal expression signature.12,73 Recurrent somatic mutations, similar to group 3, seem to converge on epigenetic aberrations, particularly histone modifiers.74-77 This feature can be identified by frequent KDM6A, MLL3, and ZMYM3 mutations. These genes may play a role in maintaining cells in an undifferentiated state, paralleling their role in normal stem cell function. Further research is needed to elucidate this pathogenic mechanism. One of the recurrently altered genes identified through the MAGIC cohort constitutes a tandem duplication of the SNCAIP gene, which is mutated in a subset of Parkinson disease patients.24,78 The biological relevance of this amplification needs to be functionally characterized further. One of the remaining hurdles in this subgroup is the lack of an animal model. Recent evidence suggests that an MYCN-driven mouse model independent of SHH signaling resembles a non-SHH/WNT model, although its specific subgroup affiliation remains to be definitively defined.71,79 These findings, identified with modern genomic technologies, uncover an astoundingly small number of somatic mutations in medulloblastoma, on the order of 10 to 12, compared with solid tumors in adulthood.44,80 This suggests that group 3 and 4 medulloblastoma may be a copy number--driven disease with the pathogenic process converging on a few key molecular processes, or it may be regulated through an epigenetic mechanism.74,76,81,82 Over the coming years, with the establishment of preclinical models recapitulating subgroup-specific disease, these findings will herald a new wave of targeted therapeutics. MEDULLOBLASTOMA METASTASES ARE BIOLOGICALLY DISTINCT FROM THEIR MATCHED PRIMARY TUMOR Recent evidence in adult glioblastomas highlights the intratumoral heterogeneity whereby closely intermingled cells can exhibit disparate driver genes.83-86 These findings support the notion that within an individual patient, there lie subclonal compartments with unique biological behaviors.25 This observation is particularly troubling because therapies aimed at eradicating the bulk of the tumor may in fact spare resistant clones that subsequently seed relapses. Although previous publications and current clinical management of human medulloblastoma assume that the primary tumor and its matched metastases respond to therapy in a similar manner, this notion may be jarringly false. In fact, a recent publication by Wu et al90 offers support for a bicompartmental model whereby the primary disease and the matched metastatic disease are inherently different. Failure to study the leptomeningeal disease as an important and separate entity may result in the ineffectiveness of targeted therapies. Functional genomics studies using Sleeping Beauty insertional mutagenesis technology have been successfully applied to the study of solid tumor initiation and progression.87-89 Recently, this system has been used to model a highly metastatic subset of medulloblastoma.90,91 Comparing the primary and matched tumors revealed that although different metastases are genetically very similar to each other, they deviate markedly from the primary tumor. Certain genetic events in the metastatic lesions were not present in the matched primary tumor, whereas other genetic events were completely restricted to the primary tumor. This finding aligns with the hypothesis that metastases arise from a restricted subclone of the primary tumor that has been selected in the metastatic tumor niche. This complex pattern of genetic variance may explain the existence of therapy-resistant clones, underscoring the difficulty in treating patients with metastatic disease. A silver lining is that many of the genes found to be enriched in the metastatic compartment target the PI3K pathway, implicating PI3K signaling as a potential driver for medulloblastoma dissemination and an avenue for therapeutic intervention. These observations add another degree of complexity as a result of intratumoral heterogeneity, which raises additional concerns regarding the management of the disease. Treatment decisions will need to be made while accounting for this added complexity, and molecular targeting will need to focus on multiple compartments, perhaps relying on combinatorial therapy. For example, biopsies of the metastatic disease, which are currently not done, may be required to characterize the metastatic lesion and to direct treatment to this particular compartment. Finally, in subgroups with a high incidence of metastatic dissemination such as group 3, safeguards must be put in place during the upfront treatment phase to limit the survival of resistant clones. FUTURE CONSIDERATIONS IN TRIAL DESIGN AND TARGETED THERAPIES With the first wave of medulloblastoma genomics studies coming to a conclusion, a wealth of new insight into its pathogenesis has been gained. The delineation of the 4 molecular subgroups is already beginning to affect the way this disease is studied. To facilitate inclusion of subgroup stratification in clinical trials, efforts have already been made on using inexpensive subgroup assignment with immunohistochemistry for for for group for group or expression from or using as as The challenges will be to each and to functionally their biological in subgroup-specific preclinical models. the of mutations are a few driver it will be up to the research to which events are to the initiation and of medulloblastoma. The of the medulloblastoma genomic data will be to our understanding of the data from the medulloblastoma epigenome, and, in the will the comprehensive data needed to targeted therapies. These future studies will reveal increased and further within It is that efforts such as MAGIC to and using collective and to a new age of molecular to patient For the a of Health II was a and is by the in is by from the of Health and the Brain the the and is by the from the of Health The have or interest in any of the or in this