Gene Signatures Associated with Mouse Postnatal Hindbrain Neural Stem Cells and Medulloblastoma Cancer Stem Cells Identify Novel Molecular Mediators and Predict Human Medulloblastoma Molecular Classification

Medulloblastoma is the most common malignant brain tumor of childhood. Surgery alone is not sufficient to eradicate this tumor, and both radiotherapy and chemotherapy are considered inefficient treatments, mostly due to the high vulnerability of young patients to the toxic effects of antimitotic treatments. The restricted incidence of medulloblastoma during childhood relates to the peculiar ontogeny of the cerebellum, which begins early during embryonic development, reaching its final maturation only after birth (1). This delayed maturation results in a prolonged formative postnatal period, characterized by increased susceptibility to tumor formation. Whereas the majority of patients are diagnosed with medulloblastoma at a very early age, suggesting that medulloblastomas initiate from mutations occurring during embryonic development in hindbrain germinative regions (2, 3), 20% of medulloblastoma patients develop medulloblastomas in late adolescence or early adulthood, indicating that medulloblastoma might also arise postnatally.

Different subsets of neural precursors persist postnatally within the hindbrain and, as such, are potentially involved in medulloblastomagenesis. In the mouse, the best characterized progenitor population comprises the so-called granule cell progenitors (GCP) of the external granular layer (EGL), which are characterized by a peak of proliferation at postnatal day (P) 7, followed by progressive decline and exhaustion within the third postnatal week. GCPs proliferate robustly in vivo and in vitro in response to Sonic Hedgehog (Shh), and their deregulated postnatal proliferation has been causatively implicated in the development of the desmoplastic medulloblastoma variant, characterized by alterations in the Shh pathway (2). However, GCPs do not satisfy the requirements for stemness, as they are characterized by limited self-renewal and lack of multipotency. On the contrary, the other population of transiently neurogenic progenitors identified within the cerebellum white matter during the same postnatal window of EGL GCPs does contain bona fide neural stem cells (NSC) that can be cultured in vitro in the presence of epidermal growth factor (EGF) and fibroblast growth factor-2 (FGF-2) and display typical features of NSCs, such as self-renewal and in vitro/in vivo multipotency (4–6).

Of note, a third subset of hindbrain neural progenitors has been retrieved in proximity of the IV ventricle (IVv), around which the cerebellum proper develops. IVv-derived progenitors are endowed with persistent neurogenic activity throughout adulthood (7, 8). However, the establishment of long-term self-renewing NSC lines from the IVv has never been reported. Interestingly, oncogenic transformation of progenitors located in the dorsal brainstem, as IVv NSCs are, has been associated with the development of a subtype of the “classic” medulloblastoma variant, whose main molecular feature is activation of the WNT pathway (3).

Cancer stem cells (CSC) are responsible for tumor growth and maintenance and, being resistant to standard therapies, are likely involved in tumor relapse. CSCs can be isolated from brain tumors and cultured in vitro long term in the presence of EGF and FGF-2. By this method, putative CSC lines have been established from desmoplastic medulloblastomas developing in Ptch^+/− mice (9), alone or in combination with p53 loss (10). Medulloblastoma CSCs, characterized by expression of the CSC marker CD15 (9, 11), display the essential requirement for a CSC, that is, the capability to generate new tumors, which closely mimic the human disease in terms of in vivo growth and phenotype.

To identify novel molecular determinants to be pursued as diagnostic markers and/or therapeutic targets in medulloblastoma CSCs, we set out to establish long-term expanding postnatal NSC lines from the IVv as well as from the cerebellum white matter and to compare them with medulloblastoma CSCs. IVv NSCs were capable of differentiating into the 3 main lineages of the central nervous system (CNS) bothin vitro and in vivo, thus fulfilling the requirements to qualify as bona fide NSCs. By applying the same culture conditions to mouse medulloblastomas derived from Ptch^+/− mutants with or without the concurrent loss of p53, we isolated several medulloblastoma CSC lines endowed with self-renewal, multipotency, and different tumorigenic potential. When subjected to microarray-based gene expression profiling, normal cerebellum white matter and IVv-derived SCs as well as medulloblastoma CSCs were characterized by distinctive gene signatures that hold predictive potential when tested against human data sets as previously shown for other stem cell types (12, 13). Most importantly, some of the genes upregulated in medulloblastoma CSCs were also expressed in human medulloblastoma and play a relevant role in the control of medulloblastomagenesis, thus validating medulloblastoma CSCs as a valuable preclinical model for medulloblastoma.

The ventricular zone lining the IVv contains mitotically active cells (Fig. 1A) whose proliferation can be reactivated in vivo by infusion of mitogens (14). To assess whether IVv NSCs might give rise to long-term expanding NSC lines, we cultured the IVv region from the mouse hindbrain at P7, P14, P21, and P30 (Fig. 1B) according to the NeuroSphere Assay. Cerebellum white matter and the subventricular zone lining the 2 lateral ventricles of the forebrain were used as controls.

At P14, P21, and P30, cell lines could be generated only from IVv and subventricular zone, confirming that sustained proliferation of cerebellum white matter cells ceases by the first postnatal week (5). However, at P7, long-term expanding cell lines could be established from all 3 regions; therefore, we analyzed NSCs at this time point. Cells were isolated from wild-type (wt), Ptch^+/−p53^+/+ and Ptch^+/−p53^+/− mice. Cerebellum white matter and IVv NSCs gave rise to a much lower frequency of primary neurospheres than cells from the subventricular zone, independently of genotype (Fig. 1B). Clonal NSC lines were established that self-renewed up to 40 subculturing passages in vitro, again independently of genotype and, also, of the region of origin (Fig. 1B). The 3 regions and the clonal NSC lines derived from them comprised a higher frequency of CD15/SSEA1-IR cells than CD133/Prom1-IR cells (Fig. 1C). All NSC populations gave rise to neurons, astrocytes, and oligodendrocytes, thus showing full multipotency (Fig. 1D and Supplementary Table S1A). However, electrophysiologic analysis of terminally differentiated neurons obtained from region-specific NSCs indicated that they were intrinsically different in their electroresponsiveness (Fig. 1E). In fact, cerebellum white matter and IVv NSC-derived neurons were characterized by inward fast Na⁺ and slow Ca²⁺ currents and outward inactivating K⁺ currents normally retrieved in functionally mature neurons. On the contrary, most neurons obtained from subventricular zone NSCs displayed K⁺ steady-state currents typical of immature neuronal cells (Fig. 1E and Supplementary Table S1B).

To formally prove the stem cell nature of IVv NSCs, GFP-labeled clonal IVv NSCs were transplanted into the cerebellum of neonatal mice (Fig. 1F). Similar to cerebellum white matter NSCs, IVv NSCs differentiated into astrocytes, neuron-like cells, and oligodendrocytes, which dispersed throughout the distinct layers of the cerebellum (Supplementary Table S1C). Thus, by showing extensive self-renewal ability and in vitro/in vivo multipotency, IVv-derived cells qualified as bona fide NSCs and, together with cerebellum white matter–derived NSCs, revealed intrinsic regional differences in terms of in vitro and in vivo differentiation potential as compared with subventricular zone NSCs.

Wt and Ptch^+/−p53^+/+ NSCs isolated from the 3 different regions were subjected to whole-genome transcriptome analysis at subculturing passages between 10 and 15. Notably, wt and Ptch^+/−p53^+/+ NSCs clustered together, with no genes resulting differentially expressed, suggesting that Ptch heterozygosity did not influence the global transcriptome of NSCs (not shown). Gene profiles of hindbrain NSCs were sharply distinguishable from that of subventricular zone NSCs, with 562 differentially expressed probe sets (Fig. 2A and Supplementary List S1). Forebrain-restricted transcription factors, such as Emx2 and FoxG1, were overexpressed in subventricular zone NSCs; by contrast, transcription factors involved in hindbrain development, such as En2 and Pax3, were upregulated in cerebellum white matter and IVv NSCs (Fig. 2B).In silico data were validated in cultured NSCs (Fig. 2B and C) and, most relevantly, in tissue sections comprising the regions of origin (Fig. 2D and E). Thus, in vitro cultured region-specific NSCs maintained their in vivo regionalization and their positional identity. Most remarkably, NSCs from the IVv, that is, an extracerebellar region, were molecularly indistinguishable from cerebellum white matter-derived NSCs.

Given that many medulloblastomas arise postnatally, we challenged postnatal hindbrain NSCs with a pro-oncogenic stimulus, such as chronic exposure to insulin, to test whether they might act as potential medulloblastoma cells of origin (15). We transplanted C57BL/6 mice with syngeneic pancreatic islets and GFP-labeled hindbrain NSCs under the kidney capsule. In line with previous studies in which tumor suppressor-deleted postnatal cerebellum white matter NSCs were tumorigenic (6), hindbrain NSCs developed GFP-IR tumors, which were characterized by the presence of rosette-like structures, typical of medulloblastoma/primitive neuroectodermal tumor (PNET) (Fig. 2F).

To investigate whether the gene signatures distinguishing hindbrain- versus forebrain-derived NSCs were enriched in distinct molecular subtypes of human medulloblastoma, we carried out Gene Set Enrichment Analysis (GSEA) on a publicly available data set (GSE21140), which contained expression data from 103 human medulloblastoma samples (16). We selected 155 and 142 genes that were upregulated at least 2-fold in hindbrain and forebrain NSCs, respectively. Interestingly, the expression of genes upregulated in cerebellum/IVv NSCs was more strongly associated with the desmoplastic medulloblastoma subgroup than the subventricular zone gene signature (Fig. 2G, Supplementary Table S2 and Supplementary List S1). Accordingly, the same gene signature was exclusively upregulated in the Shh molecular subgroup, which comprises many desmoplastic medulloblastomas. When the global 297-gene signature was tested against all the medullo-blastoma subgroups simultaneously, it again sharply separated the Shh subgroup from the others (Supplementary Fig. S1).

To assess whether the Shh pathway, normally active in GCPs, might be operating also in hindbrain NSCs, we took advantage of the genetic configuration of the Ptch targeting vector, which contains a sequence encoding a nuclear β-galactosidase (LacZ) and the neomycin resistance gene. If Ptch^+/−p53^+/+ cerebellum white matter and IVv NSCs express Ptch in vitro, they should be β-Gal positive and should survive neomycin selection.

The expression of Ptch-LacZ by β-Gal staining was retrieved in the subventricular zone, in the EGL, in the white matter, and in scattered cells around the IVv of P7 Ptch^+/− mice in vivo (Supplementary Fig. S2A). Primary cells from subventricular zone, cerebellum white matter and IVv self-renewed long-term in the presence of sublethal concentration of G418, thus indicating that cerebellum white matter and IVv NSCs did express Ptch (not shown). However, when wt and Ptch^+/− cerebellum white matter and IVv NSCs, cultured for more than 6 subculturing passages (SP), were exposed to 10 μmol/L cyclopamine, a Shh pathway inhibitor, no difference in terms of cell survival and proliferation were observed (Supplementary Fig. S2B). Accordingly, Shh downstream targets were not expressed in long-term cultured NSC lines, suggesting that they might grow independent of Shh signaling activation (Supplementary Fig. S2C and D). To formally prove this concept, we exposed to cyclopamine wt and Ptch^+/− subventricular zone, cerebellum white matter and IVv NSC lines starting at SP0 (i.e., primary culture) up to SP14. A remarkable reduction in survival was retrieved in all NSC lines at very early SPs, which was not observed in the same NSCs after additional SPs (Supplementary Fig. S2D). Accordingly, overexpression of the Shh pathway downstream targets Gli1 and Gli2 as well as downregulation of Gli3 were detected in short-term cultured NSCs, being absent in long-term cultured NSCs (Supplementary Fig. S2E).

Cells isolated from adult Ptch^+/−p53^+/+, Ptch^+/− p53^+/−, and Ptch^+/− p53^−/−medulloblastomas were cultured under the NeuroSphere Assay conditions (9). Part of each cell suspension was grown in the presence of serum and in the absence of mitogens, to generate serum-dependent cancer cell lines to be exploited as nonstem cell controls (17). After 30 to 60 days in vitro, stable clonal CSC lines could be obtained (n = 10 from Ptch^+/−p53^+/+ medulloblastomas, n = 4 from Ptch^+/− p53^+/−medulloblastomas, and n = 2 from Ptch^+/− p53^−/− medulloblastomas). Some CSC lines grew as small clusters similar to neurospheres, whereas other as adherent cells. Medulloblastoma cells were capable to expand in culture not only in the presence of both EGF and FGF2 (Fig. 3A) but also in the presence of either growth factor alone (data not shown). Importantly, all cell lines expanded for more than 100 passages, with a doubling time between 2 and 5 days. Medullo-blastoma CSC lines were heterogeneous in terms of growth rate and clonal efficiency, which was independent from the genotype (Fig. 3A and B). Similar to normal NSCs, all clonally derived medulloblastoma cell lines differentiated into Tuj1-IR neuron-like cells, GFAP-IR astrocyte-like cells, and NG2/O4-IR oligodendrocyte-like cells, with few cells aberrantly colabeled with neuronal and glial markers (ref. 18; Fig. 3C and Supplementary Table S3). When exposed to cyclopamine, only a few medulloblastoma CSC lines showed a very modest decrease in survival, which was in line with the low expression of Shh downstream targets in the same cells (Supplementary Fig. S3A–C). To test the presumptive independence of medulloblastoma CSCs from Shh signaling for in vitro growth, we generated stable medulloblastoma CSC lines in which Smo was silenced by lentiviral-based short hairpin RNA (shRNA; Supplementary Fig. S3D). Notably, Smo inhibition, and, as such, Shh pathway inactivation, did not significantly affect medulloblastoma CSC proliferation and survival (Supplementary Fig. S3D), thus suggesting that medulloblastoma CSCs might rely on different growth pathways for efficient growth.

Because the majority of Ptch^+/− tumors lose expression of the Ptch wt allele (9), we tested whether our medulloblastoma cells were also characterized by LOH for Ptch (Fig. 3D). By DNA genotyping, LOH for Ptch was detected in all Ptch^+/− p53^+/+ medulloblastoma CSCs due to the deletion of a large fragment of the genomic locus. Conversely, Ptch LOH was absent in their tumor of origin and in the tail tissue. Intriguingly, Ptch LOH was never observed in Ptch^+/−p53^−/− and Ptch^+/−p53^+/− medulloblastoma CSCs, which, conversely, showed LOH for p53(Fig. 3D). Both Ptch and p53 LOH resulted in the absence of the corresponding transcripts (data not shown).

When injected into immunosuppressed animals, both intracranially (striatum and/or cerebellum) and subcutaneously, either in adult or early postnatal animals, Ptch^+/− p53^+/+ medulloblastoma CSCs gave rise to slowly growing grafts that never evolved into full-blown lesions (data not shown). Conversely, Ptch^+/−p53^+/− and Ptch^+/−p53^−/− medulloblastoma CSC lines reproducibly established tumors, with a take efficiency of 70% and 100% (intracranially and subcutaneously, respectively). Of note, Ptch LOH did not influence the tumorigenic potential of medulloblastoma CSC lines, as all CSC lines that were tumorigenic retained the expression of the wt Ptch allele. Rather, loss of p53, occurring by genomic deletion or by somatic LOH (Fig. 3D), was associated with medulloblastomagenesis. Intracranial tumors were detected by MRI after 10 to 12 weeks after transplantation and fully developed within 30 weeks (data not shown). Conversely, subcutaneous tumors were observed as early as at 2 weeks after transplantation and progressed very rapidly. Only NeuroSphere Assay–cultured medulloblastoma CSC lines generated tumors, whereas matched serum-dependent, nonstem cancer cell lines did not.

CSC-derived intracranial tumors closely resembled spontaneous Ptch^+/− tumors, although their morphology was slightly more malignant and pleomorphic, with marked atypia and the presence of infiltrating cellular nests; they also expressed markers typical of human medulloblastoma, such as Tuj1, NeuN, synaptophysin, and GFAP (Fig. 3E). In a few circumstances, CSC-derived intracranial tumors escaped the site of implantation, through either leptomeningeal or extracranial dissemination. Tumors infiltrating the meninges expressed lower levels of neuronal and glial markers, whereas extracranial tumors were highly undifferentiated, with signs of pleomorphism and increased malignancy. CSC-derived subcutaneous tumors were the most highly undifferentiated, showing a myxoid and heterogeneous morphology, and containing high numbers of cells expressing the putative CSC marker Sox2. Tuj1-IR cells were also detected, while NeuN- and synaptophysin-IR cells were absent. Thus, the site of tumor formation affects the histologic phenotype of experimental CSC-derived medulloblastomas, with a progressive malignant evolution from lowly malignant intracranial tumors to highly malignant extra-CNS neoplasias.

To identify dysregulated genes that may play a role in the pathogenesis of medulloblastoma, we first compared the global gene expression profiles of medulloblastoma CSCs with that of normal NSCs. Unsupervised clustering analysis showed that medulloblastoma CSCs shared many more molecular determinants with hindbrain NSCs than with subventricular zone NSCs. Notably, many more genes were differentially expressed between hindbrain NSCs and Ptch^+/−p53^−/− medulloblastoma CSCs than between hindbrain NSCs and Ptch^+/−p53^+/+ medulloblastoma CSCs (Fig. 4A).

These gene expression data were then exploited to generate gene signatures that reflected the behavior of the distinct populations of medulloblastoma CSCs (Fig. 4B). The first signature was upregulated exclusively in tumorigenic Ptch^+/− p53^−/− medulloblastoma CSCs versus hindbrain NSCs (MB1_EXCLUSIVE, 980 genes, Supplementary List S2), the second exclusively in nontumorigenic Ptch^+/− p53^+/+ medulloblastoma CSCs versus hindbrain NSCs (MB2_EXCLUSIVE, 105 genes, Supplementary List S2), and the third comprised the probe sets whose expression was shared by both tumorigenic and nontumorigenic CSCs versus hindbrain NSCs (MB1_MB2 COMMON, 81 genes, Supplementary List S2). The MB1_EXCLUSIVE and MB1_MB2 COMMON gene signatures were specifically enriched in the desmoplastic subgroup of human medulloblastomas. However, although MB1_MB2 COMMON gene signature was associated with the Shh molecular subgroup, MB1_EXCLUSIVE gene signature was enriched in the WNT molecular subgroup (Fig. 4B and Supplementary Fig. S4). Accordingly, the WNT pathway mediator β-catenin was expressed in Ptch^+/− p53^+/− spontaneous tumors, in Ptch^+/−p53^−/− CSC-derived intracranial and subcutaneous tumors in vivo, and in Ptch^+/− p53^−/− medulloblastoma CSCs in vitro (Supplementary Fig. S5A and B). Transcripts for several WNT pathway mediators included in the MB1_EXCLUSIVE gene signature were also significantly upregulated in Ptch^+/− p53^−/− medulloblastoma CSC lines as compared with Ptch^+/− p53^+/+ medulloblastoma CSCs (Supplementary Fig. S5C). Thus, human medulloblastomas can be distinguished based on their degree of resemblance to distinct mouse medulloblastoma CSC molecular phenotypes.

Early B-cell factors (Ebfs), such as Ebf2 and Ebf3, are expressed in the cerebellum during embryonic development and regulate progenitor cell proliferation and differentiation. Interestingly, they were highly expressed in medulloblastoma CSCs and in their corresponding tumor tissues (Fig. 4C). Ebf3 expression, in particular, was observed in cycling progenitors located in the mouse EGL as early as P3, peaked at P7, and decreased around P15. Accordingly, Ebf protein expression was retrieved in EGL progenitors and also in cells in the prospective white matter and around the IVv (Fig. 4E). In line with gene expression data, Ebf protein expression was high in both spontaneous and CSC-derived intracranial tumors, being very low in intrameningeal and subcutaneous CSC-derived tumors (Fig. 4E).

EBF3 expression was detected in silico in human medullo-blastomas and in a small fraction of neuroblastomas, whereas it was not retrieved in glial tumors, suggesting a possible association between the expression of the gene and a neuronal tumor phenotype (Supplementary Fig. S5A and B). Accordingly, EBF3 transcript and EBF proteins were expressed in human medulloblastomas and not in glioblastoma multiforme (GBM, Fig. 4F and G and Supplementary Table S5). The pattern of expression of the EBF proteins varied among histologic variants of medulloblastoma. In particular, classic medullo-blastomas displayed intense EBF staining in the majority of tumor cells, whereas desmoplastic medulloblastomas showed EBF expression being restricted to desmoplastic portion of the tumor and excluded from the nodular part. Of note, EBF proteins were also strongly expressed in the pleomorphic giant cells found in large-cell anaplastic medulloblastomas.

To address the putative role of Ebf3 in medulloblastomagenesis, we transduced both NSCs and medulloblastoma CSCs with an Ebf3-FLAG lentiviral construct, obtaining efficient expression of Ebf3 transcript and protein (Fig. 5A and B). Mock- and Ebf3-tranduced NSC and CSC lines did not show any difference in proliferation, long-term self-renewal, migration, and apoptosis. However, under proliferative culture conditions (i.e., in the presence of EGF and FGF2), Ebf3-transduced NSC and CSC lines acquired the appearance of cells undergoing differentiation. In fact, whereas normal mock-transduced IVv, cerebellum, and subventricular zone NSCs were almost negative for lineage differentiation markers, Ebf3 overexpressing NSCs showed a consistent increase in the frequency of FLAG-positive cells expressing the neuronal marker Tuj1 and displaying a mature cellular morphology (Fig. 5C and Supplementary Fig. S7). Likewise, Ebf3 overexpressing FLAG-positive CSCs gave rise to Tuj1-IR cells displaying a more mature morphology than Tuj1-IR cells retrieved in mock CSC cultures (Fig. 5C). Upon differentiation, Ebf3 overexpressing NSC/CSC cultures mostly generated FLAG-positive neuronal-like cells featuring extended processes and well-developed neurites (Fig. 5D). The frequency of astrocytes and oligodendrocytes was similar between mock- and Ebf3-transduced NSC/CSCs under both proliferative and differentiative conditions (not shown).

Mock- and Ebf3-transduced CSC lines were then transplanted intracranially and subcutaneously into adult immune-deficient mice. In spite of very high efficiency of Ebf3-FLAG overexpression (Fig. 6A and B), significant differences in the tumorigenic ability of Ebf3 overexpressing CSCs with respect to mock CSCs were observed only upon subcutaneous transplantation. Indeed, the frequency of cells endogenously expressing Ebf in mock CSC-derived intracranial tumors was very high and comparable with that in Ebf3 expressing tumors. As such, Ebf3 overexpression in CSCs under intracranial transplantation settings was not functionally relevant. On the contrary, given that endogenous Ebf expression in mock CSC-derived subcutaneous tumors was low (Fig. 4E, Fig. 6B), enforced expression of Ebf3 in medulloblastoma CSCs accelerated the formation of highly malignant subcutaneous tumors (Fig. 6C and D). Whereas mock CSC-derived tumors were characterized by sarcomatoid/myxoid morphology and by the presence of Tuj1- and GFAP-IR cells, Ebf3-transduced CSC-derived tumors were more highly undifferentiated, showed a pleomorphic and malignant appearance, contained perivascular cellular rosettes, and comprised only Tuj1-IR cells. In line with these findings, both the mitotic index as measured by Ki67 staining and the frequency of Sox2-IR cells were significantly higher in Ebf3-transduced CSC-derived tumors than in mock CSC-derived tumors (Fig. 6E). Overall, subcutaneous mock CSC-derived tumors were reminiscent of classic medulloblastomas, and Ebf3 overexpressing tumors resembled highly malignant PNET-like anaplastic medulloblastomas.

To inhibit Ebf3 endogenous expression in medulloblastoma CSCs by RNA interference (RNAi), given that Ebf family members function as active heterodimers, we took advantage of a GFP-coding retroviral construct that allows the concomitant expression of siRNAs against Ebf1-, Ebf2-, and Ebf3 (19). Significant downregulation of Ebf2 and Ebf3 transcript and protein was observed in Ebf-silenced medulloblastoma CSCs (Fig. 6F). Mock- and Ebf-silenced medulloblastoma CSCs were then injected subcutaneously. Ebf-silenced tumors were significantly smaller than those generated by mock-transduced cells and, again, this defect was reflected in the histologic appearance of tumors (Fig. 6G). In fact, whereas mock CSC-derived tumors showed the typical sarcomatoid/myxoid morphology, tumors in which Ebfs were silenced were characterized by a homogeneous, well-differentiated morphology and reduced proliferation, with cells displaying large and eosinophil cytoplasm, suggestive of a less malignant phenotype.

To identify novel molecular targets involved in medulloblastomagenesis to be exploited for selective and effective therapies, we took advantage of the biologic relationship existing between medulloblastoma and the cell progenitor lineages from which it is thought to arise. In fact, different clinical/molecular medulloblastoma subtypes are believed to originate from mutations occurring in distinct stem/precursor cells located in cerebellar germinative regions, such as the rhombic lip derivative and the ventricular zone matrix (20), as well as in extracerebellar domains, as the dorsal brainstem (3), some of them persisting postnatally (2).

During embryonic development, 2 primary cerebellar neurogenic epithelia, that is, the rhombic lip and the IVv ventricular zone, segregate as early as embryonic day 9 (E9) and represent the birthplace of all GABAergic and glutamatergic lineages, respectively (1, 21). Early postnatally, the rhombic lip derivative located at the cerebellar surface, the EGL, undergoes a massive expansion due to GCP proliferation that peaks at P7 and gradually ceases, being completely exhausted within the third postnatal week in rodents and the first year in humans (22). Very recently, a second postnatal cerebellar neurogenic compartment has been identified within the cerebellar white matter (4, 5). Cerebellum white matter NSCs were shown to self-renew shortly in culture and to be multipotent in vitro and in vivo (4). We also generated P7 cerebellum white matter neurospheres, which were subcultured for many passages, giving rise to long-term expanding NSCs (6). Most interestingly, by applying the same culture conditions to cells obtained from the P7 brain parenchyma surrounding the IVv, from which short-term proliferating precursors were previously isolated (7, 8), we established long-term self-renewing multipotent bona fide IVv NSCs.

As previously shown for embryonic NSCs (23), the distinct postnatal NSC populations did maintain their original positional identity. In fact, both cerebellum and IVv NSCs overexpressed transcription factors, such as Pax3 and En2, which are well-known regulators of hindbrain development. Likewise, subventricular zone NSCs displayed enhanced expression of genes involved in forebrain development, for example, Emx2, Foxg1, and Lhx2. Most interestingly, IVv NSCs, which were isolated from an extracerebellar region, were molecularly indistinguishable from cerebellum white matter NSCs, thus suggesting that progenitors residing in different anatomic sites might have a common origin. Relevantly, when challenged oncogenically, hindbrain-derived postnatal NSCs generate medulloblastoma/PNET-like lesions, suggesting that multiple populations of hindbrain NSCs may be at the origin of medulloblastoma (3). Accordingly, IVv NSCs could be isolated not only from early postnatal mice but also from adults, in line with the observation that medulloblastomas deriving from brainstem progenitors show a distributed age of onset, ranging from late infancy to adulthood, with a peak in older children (24).

In agreement with previous observations in medullo-blastoma cell cultures and serum-cultured colon cancer cells (25, 26), the expression of downstream mediators of Shh pathway in NSCs and medulloblastoma CSCs was modest, and, accordingly, Shh pathway activation was marginally relevant for proliferation and self-renewal of NSC/CSCs in vitro. Indeed, we showed that addiction to Shh pathway occurs in NSCs only at early subculturing passages, when Shh pathway is still active. Accordingly, Shh inhibition in long-term cultured medulloblastoma CSCs by Smo RNAi does not influence CSC proliferation. These results are in contrast with studies reporting full activation of Shh pathway in normal NSCs (27) as well as in CSCs isolated from the same Ptch mouse model (28) or other brain tumors (29). Indeed, both NSC/CSC-derived neurospheres employed in those studies were maintained in vitro for very few subculturing passages. Short-term NSC/CSC cultures are mostly composed by committed progenitors (30). As such, lineage-restricted precursors, rather than NSC/CSCs might have been addicted to Shh signaling and, therefore, responded to Shh inhibition by reduced survival and proliferation. On the contrary, long-term propagated cultures, which are highly enriched in the NSC/CSC component and devoid of committed precursors (30), might not be Shh dependent and may self-maintain through different signaling pathways. Alternatively, chronic exposure of NSC/CSCs to mitogenic stimulation, in particular to FGF2 (31), might repress Shh pathway activation and promote the activity of different compensatory effectors.

Notably, although Shh pathway activation was not retrieved in postnatal NSCs, the cerebellum/IVv NSC gene signature is significantly associated with the Shh human molecular subgroup. This observation might imply cerebellum white matter and IVv NSCs as additional cells of origin for medulloblastomas, as they share the expression of the Shh pathway mediators with EGL GCPs, which are well known to be causatively involved in medulloblastoma tumorigenesis (32, 33).

Most relevantly, the genes characterizing Ptch^+/−p53^+/+ medulloblastoma CSCs were associated with the Shh molecular subgroup of human medulloblastomas, whereas those overexpressed in Ptch^+/−p53^−/− medulloblastoma CSCs were enriched in the WNT subgroup, thus emphasizing the exploitability of these medulloblastoma CSCs as in vitro preclinical model of Shh- and Wnt-dependent medulloblastomas, respectively. Moreover, the observation that the site of transplantation of medulloblastoma CSCs under experimental in vivo settings results in tumors endowed with varying degree of malignancy, ranging from well-differentiated desmoplastic intracranial tumors to highly undifferentiated anaplastic subcutaneous tumors, further underline the relevance of medulloblastoma CSCs as a versatile model, highly representative of the different variants of human medulloblastomas.

In line with a recent report (9), we established long-term self-renewing CSC lines from medulloblastomas, which spontaneously develop in Ptch heterozygous mice (34) and in compound Ptch/p53 mutants (10). However, only medullo-blastoma CSCs, characterized by genetic or somatic p53 loss, were able to generate fully developed tumors closely mimicking the different histologic variants of the human disease. Interestingly, the gene signature distinguishing Ptch^+/−p53^−/− medulloblastoma CSCs is enriched in the WNT subgroup, in agreement with the recent finding that, in medulloblastoma patients, p53 mutations are mostly found in favorable-risk WNT-subtype medulloblastomas (35).

During development, Ebf transcription factors are selectively expressed in early postmitotic neurons (36). However, we observed that Ebf3 expression was retrieved not only in postmitotic GCPs of the inner EGL but also in bromodeoxyuridine (BrdUrd)-incorporating GCPs undergoing neuronal fate choice in the outer EGL. Accordingly, enforced expression of Ebf3 in NSCs and medulloblastoma CSCs did not induce cell-cycle arrest or apoptosis, but, rather, premature and enhanced neuronal commitment, which was already evident under proliferative conditions and was maintained upon differentiation.

Most interestingly, enforced expression of Ebf3 in medulloblastoma CSCs greatly increased their tumorigenic ability, indicating that the Ebf3-mediated induction of a neuronal phenotype in medulloblastoma cells might be required to promote medulloblastoma initiation and progression, in agreement with recent reports proposing that the acquisition of granule cell neuronal commitment is essential for medulloblastomagenesis (32, 33). Of note, the molecular subgroup of human medulloblastomas showing a highly malignant behavior is associated with a neuronal signature (24).

Ebf3 expression is known to be undetectable in many human cancers, including leukemias, pancreatic cancers, head and neck squamous cell carcinomas and, most interestingly, GBMs. In many of these cancers, EBF3 inactivation is due to epigenetic silencing by promoter methylation, genomic deletion, or point mutations. As expected, Ebf3 expression in these tumor cells induces cell-cycle arrest and apoptosis, indicating that Ebf3 could act as tumor suppressor (37). By contrast, our findings suggest that Ebf3 expression in malignancies belonging to the neuronal lineage, such as medulloblastomas and, possibly, neuroblastomas, might play a pro-oncogenic role, in contrast with glial tumors and other cancers.

Transcriptional data derived from molecular comparisons between normal NSCs and medulloblastoma CSCs thus holds great potential as preclinical molecular tools for the identification of novel molecular determinants and meaningful gene signatures in medulloblastoma.

NSC cultures were established from the subventricular zone, the cerebellum, and the region surrounding the IVv of P7 mice, whereas CSCs were obtained from tumors developed in 3- to 6-months-old animals. NSC lines were established from tissues pooled from 4 to 6 mice, whereas CSC lines were from single tumors. NSC and CSC lines were cultured and propagated in standard medium as described by Galli and colleagues (38).

NSCs and CSCs were differentiated by withdrawal of mitogens from the culture medium and addition of 2% FBS for 7 days (38) and processed for immunocytochemistry. The lists of the antibodies used are available in the Supplementary Methods section.

See Supplementary Methods.

Digoxygenin-labeled riboprobes were transcribed from plasmids coding for Emx2, En2, and Ebf3 cDNAs. Sixteen micrometer cryostat sections of P3, P7, and P15 brains were processed as described in Supplementary Methods.

Tissue samples and NSC/CSC lines were enzymatically and/or mechanically dissociated to obtain a single cell suspension. Single cells were suspended in PBS/BSA 5 mg/mL 2 mmol/L EDTA pH 8 and incubated on ice for 20 minutes. Specific antibodies were diluted in the same solution and then incubated together with the cells for 30 minutes on ice. The acquisition was carried out on BD FACS CANTO II (Becton and Dickinson) and the analysis by FCS express 3.0 software. Antibodies used were fluorescein isothiocyanate (FITC)-conjugated CD15 (Becton and Dickinson) 1:20; PE-conjugated Prominin-1 (eBioscience) 1:50, APC-conjugated CD45 (Becton and Dickinson), 1:100, PE-conjugated Ter119 (Becton and Dickinson), 1:100.

Each sample was homogenized in 10× volume of radioimmunoprecipitation assay lysis buffer. Samples were then diluted in Laemmli's SDS sample buffer. Proteins were separated by electrophoresis on 8% SDS-PAGE and transferred onto trans-blot nitrocellulose membranes (Amersham). Ponceau staining (Sigma) was carried out to confirm that the samples were loaded equally. The primary antibodies/antisera used were mouse anti-Ebf3 (1:1,000; Abnova), mouse anti-GAPDH (glyceraldehyde-3-phosphate dehydrogenase, 1:10,000; Sigma), and goat anti-Smo (1:300; Santa Cruz).

Mouse Ebf3 cDNA was cloned into the monocistronic transfer lentiviral vector (LV) pCCL.sin.cPPT.PGK.GFP.WPRE11. GFP was excided and substituted with the FLAG-Ebf3 cassette. Sister cultures were infected with the same vector coding for GFP, as mock control. NSCs and CSCs were transduced with 1 × 10⁷ TU/mL of each LV for 16 hours. Ebf3 silencing was achieved by using retroviral constructs coding for short-hairpin RNA directed against the 3 Ebf genes (19). CSCs were transduced through 3 rounds of infection. Smoothened silencing was done by exploiting commercially available LVs coding for gene-specific shRNA clones (Mission RNAi; Sigma-Aldrich). Infection of CSCs with shRNAs was carried out according to the manufacturer's instructions.

Total RNA from all samples was extracted using the RNeasy Mini kit (Qiagen). One microgram of total RNA was reverse transcribed by using first strand synthesis kit Superscript III RNaseH Reverse Transcriptase (Invitrogen) and OligodT primers. All the semiquantitative reverse transcriptase PCRs (RT-PCR) were carried out using home-designed specific primers and cDNAs were normalized on β-actin housekeeping gene. Quantitative real-time PCR was carried out by IQ SybrGreen technology (BioRad), following manufacturer's instructions. β-Actin was used as housekeeping gene. Mouse- and human-specific Ebf2 and Ebf3 primers were employed for qPCR (Qiagen, Sa Bioscience).

The subculturing passages at which medulloblastoma CSCs and NSCs were collected for molecular analysis were comprised between the 8th and 20th. Quality control of hybridization was done by Image Quality, MAplots, Boxplot and Density Plot, Array normalization was executed by the RMA and GCRMA algorithms. Divisive clustering algorithms were used to obtain dendrograms, in which the biologic samples were clustered on the basis of the differentially expressed genes. The hierarchical clustering algorithms employed were (i) distances (Euclidian, correlation), (ii) linkage (complete, single, mcquitty, ward, and centroid). The differentially expressed genes (DEG) were obtained based on (i) t test moderated empirical Bayes, (ii) P value [false discovery rate (FDR) adjusted 0.05], (iii) cut-off (1 log₂ Fold Change, FC). Microarray data are available at NCBI GEO (GEO accession number GSE37316).

GFP-labeled clonal cerebellum white matter and IVv NSC lines were injected into neonatal cerebella as described in Leto and colleagues (21).

For subcutaneous injection, 1 × 10⁶ to 2 × 10⁶ CSCs were resuspendend in 100 µL PBS and injected into the right flank of 45 to 60 days old nu/nu female mice. Mice were sacrificed at different time points comprised between 4 to 12 weeks postinjection, according to the cell line originally injected. For intracranial transplantation, 2 × 10⁵ CSCs were resuspended in Dulbecco's modified Eagle's medium and DNase (Sigma) and delivered into the right striatum or the cerebellum by stereotactic injection through a 5 µL Hamilton microsyringe. The following coordinates were used: AV = 0; ML =+2.5 mm; DV = −3.5 mm from bregma for intrastriatal injections and AV = −3.5; ML = 0; DV = 2 from interaural line for intracerebellar injections. Animals were sacrificed 2 to 8 months after transplantation. Transplantation of hindbrain NSCs under the renal capsule was done as described in Melzi and colleagues (15).

No potential conflicts of interest were disclosed.

Conception and design: D. Corno, R. Galli

Development of methodology: D. Corno, B. Cipelletti, V. Barili, R. Melzi, L. Sergi Sergi, L. Piemonti, R. Galli, G.G. Consalez, L. Croci, P.L. Poliani

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Corno, M. Cominelli, K. Leto, L.S. Politi, L. Piemonti, A. Bulfone, P. Rossi, F. Rossi, P.L. Poliani, R. Galli, G.G. Consalez, L. Croci

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Corno, M. Pala, K. Leto, F. Brandalise, L.S. Politi, P. Rossi, F. Rossi, P.L. Poliani, R. Galli

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Corno, L. Croci, A. Di Gregorio, R. Galli

Writing, review, and/or revision of the manuscript: P. Rossi, F. Rossi, R. Galli, G.G. Consalez

Study supervision: R. Galli

Design and analysis of Ebf3 gene expression in cerebellar granules: L. Croci, G.G. Consalez

Design of Ebf3 silencing: G.G. Consalez, D. Corno, R. Galli

The authors thank Laura Magri and Matteo Zanella for help and assistance with molecular analysis.

This work was supported by Fondazione Pierfranco and Luisa Mariani for Child Neurology and by Fondazione Guido Berlucchi for Cancer Research to R. Galli.

Received August 15, 2011; revised April 12, 2012; accepted April 12, 2012; published OnlineFirst May 3, 2012.