Distinct Neural Stem Cell Populations Give Rise to Disparate Brain Tumors in Response to N-MYC

MB spheres are MYCN-dependent and express neuronal markers

We isolated brain tumors from nine mice from the transgenic GTML model, and derived tumor spheres in neurobasal (NB) media supplemented with EGF and FGF. Eight representative sphere lines, GTML2-9, arose independently of SHH (Swartling et al., 2010), and showed elevated expression of MYCN protein and Luciferase (LUC) (Figure 1A and Figure S1A). SHH-dependent GTML tumors arise at < 5% frequency. A single line (GTML1) was isolated from a SHH-dependent GTML tumor and showed low levels of MYCN mRNA (described previously as GTML-T7 (Swartling et al., 2010)). While MYCN-low GTML1 cells showed no response to doxycycline (dox), proliferation of MYCN-high GTML spheres was blocked after 3–5 days of dox treatment, reflecting MYCN-dependence (Figs 1B and 1C, Figures S1B–S1D). This proliferation block, associated with loss of Cyclin D2, did not recover after dox wash out, as compared to controls (Figures S1D and data not shown). These observations are consistent with our earlier in-vivo observations in GMTL mice that withdrawal of MYCN in tumors led to senescence (Swartling et al., 2010).

To confirm SHH-independence of MYCN-positive GTML2-9 tumor spheres, we cultured spheres with the SMO inhibitor cyclopamine (Taipale et al., 2000), which blocks signaling downstream from SHH. GTML1 SHH-dependent spheres responded to SMO inhibition. However, no significant differences in proliferation were observed after treating GTML2-9 spheres with cyclopamine (Figures 1D and 1E and data not shown), or with SHH-N or SMO agonists that activate SHH signaling (not shown). These data are consistent with tumors arising through a SHH-independent pathway. However, it remains possible that signaling downstream of SMO could contribute to tumor formation.

Compared to WT cerebellar spheres and SHH-dependent GTML1 spheres, GTML2-4 tumor spheres with high levels of MYCN protein showed elevated levels of mRNA for Ngn1 and Syp (Figure 1F–1G). NGN1, an inducer of neurogenesis and a marker of immature GABAergic cells originating from the roof of the fourth ventricle, is typically expressed in MATH1-negative, classic MB (Farah et al., 2000; Salsano et al., 2007). Spheres from GTML 2-4 showed low levels of glial-lineage transcription factors Olig2 and Sox9 (Figures 1H and 1I), both of which are also typically low in human MB, and high in cultured normal human cerebellar NSCs (Alcock and Sottile, 2009; de Bont et al., 2008; Ligon et al., 2004). These marker data collectively verify that GTML2-4 spheres resemble SHH-negative human MB.

Freshly isolated GTML tumor cells can be transplanted orthotopically into the brains of immunocompromised mice to regenerate MB (Swartling et al., 2010). While SHH dependent GTML1 tumor spheres did not grow orthotopically, spheres from SHH independent tumors (GTML2 and GTML3) generated lethal tumors (Figure S1E). Transplanted GTML2 tumor spheres grew significantly more rapidly than transplanted GTML3 tumor spheres consistent with higher levels of MYCN protein in GTML2 cells (Figures S1D and S1E).

To determine whether NSCs isolated from normal cerebellum could represent cells of origin for GTML MB, we crossed Glt1-tTA mice to TRE-Cre:Rosa26-lsl-LacZ reporter animals, enabling us to visualize LacZ as a marker of cell fate. Indeed, NSCs cultured as spheres from these triply transgenic mice at postnatal ages were β-galactosidase-positive suggesting that the GLT1 promoter is active and that GTML tumors could originate from normal NSCs (Figures 1J and 1K).

Transduction of N-myc^T58A into the GFAP-positive NSCs boosts proliferation

GFAP is a stem cell and astroglial marker (Doetsch et al., 1999) and shows tight co-expression with GLT1 both in cerebellar and cerebral cells in vivo (Schmitt et al., 1996) as well as in our NSC cultures (Figure S1F). We therefore investigated how GFAP-positive populations of NSCs would respond to N-myc, using mice transgenic for the avian tv-a retroviral receptor driven by the GFAP promoter (Gtv-a) (Holland et al., 2000). We dissected cells at embryonic day 16 (E16) and postnatal day 0 (P0) from three different brain regions of Gtv-a mice (Figure 2A): luminal parts of forebrain ventricular zone (< 1 mm), total cerebellum, and the top layer (< 1 mm) isolated from the ventricular zone region of the dorsal brain stem. This isolated region encompassed both VZ brain stem cells and the lower rhombic lip (LRL) structure that contains candidate cells of origin for MB (Gibson et al., 2010).

Dissected cells were cultured in neurobasal media for one week and subsequently transduced with harvested RCAS viruses containing either a GFP reporter, Flag-tagged wild-type N-myc^WT, or mutationally-stabilized Flag-tagged N-myc^T58A (Figure S2A–E). After differentiation for 72h without growth factors, most GFP-transduced NSCs from both E16 and P0 cerebellar and forebrain NSCs remained positive for the NSC marker GFAP, but were negative for the neuronal marker, TUJ1, confirming selective transduction of GFAP-positive NSCs (Figures 2B and S2F).

To further characterize the types of cells transduced, we sorted GFP-positive and negative NSC fractions from E16 cerebellum, and P0 cerebellum and forebrain, 72h after transduction with RCAS-GFP. The percentage of GFP-positive cells ranged from 2.5% up to 8% of the entire cell population (Figures 2B and S2F). Interestingly, GFP-sorted single NSCs from these conditions produced significantly more secondary spheres than GFP-negative NSCs after two weeks in culture (Figure S2G). These data are consistent with selective transduction of GFAP-positive cells within these primary cultures, showing increased self-renewal as compared with more differentiated populations of cells.

We next analyzed cells transduced with RCAS-N-myc constructs. N-myc^T58A-transduced cerebellar NSCs stained intensely for Flag on Western Blot as compared to N-myc^WT-transduced cells (Figure 2C), likely reflecting stabilization of N-MYC protein due to decreased proteasomal degradation (Salghetti et al., 1999). As compared to cerebellar NSCs transduced with GFP or N-myc^WT, NSCs transduced with N-myc^T58A and GTML2 tumor spheres were both resistant to blockade of new protein synthesis by cycloheximide (CHX, Figures S2H–S2K) consistent with stabilization of N-MYC^T58A or overexpression of MYCN in these cells.

While GFP- and N-myc^WT-transduced NSCs showed moderate proliferation when cultured on coated plates, transduction of N-myc^T58A into NSCs from P0 cerebellum promoted increased proliferation, associated with high levels of the fourth ventricular zone marker NGN1 (Figures 2C and 2D). Similarly, transduction of N-myc^T58A into P0 forebrain NSCs, as well as E16 NSCs from both cerebellum and forebrain, promoted proliferation, as compared to both vector GFP control and N-myc^WT-transduced NSCs (Figure 2E and not shown).

Because NSCs in the brain stem may also generate brain tumors (Gibson et al., 2010), we next tested VZ-derived NSCs from dorsal brain stem. N-myc^T58A did not clearly drive increased proliferation in E16 and P0 brain stem cultures in this system (not shown), prompting us to test VZ-derived NSCs from E14 dorsal brain stem/LRL (henceforth called E14 LRL). Transduction of E14 LRL NSCs cells with N-myc^T58A led to increased proliferation as compared to GFP- and N-myc^WT-transduced NSCs (Figure 2F).

Culturing cerebellar NSC for a week in neurobasal media typically removes any residual GNPs (Klein et al., 2005; Sutter et al., 2010). Transduction of N-myc^WT or N-myc^T58A retroviruses into purified P0 GNPs did not drive proliferation, presumably because very few PAX6 positive GNPs also expressed GFAP and tv-a, thus evading viral transduction (Figures S2L–S2M).

To clarify whether N-myc could drive self-renewal and substitute for growth factors, we cultured NSCs and tumor spheres in limiting dilution for one week with or without EGF and FGF (10 cells/well). To distinguish effects on NSCs from effects on neural progenitors, 100 cells/well were re-plated in a secondary sphere formation assay for a second week. N-myc^WT-, N-myc^T58A- and GFP-transduced control secondary spheres retained strict growth factor dependence and showed no significant differences in sphere number (Figures 2G–2I, Figures S2N–S2P). By contrast, GTML tumor spheres generated secondary spheres even when cultured without growth factors (Figure 2J). MYCN blocked differentiation in GTML tumor spheres, as spheres with high MYCN protein levels neither differentiated nor extended processes upon growth factor starvation or doxycycline treatments (not shown). In contrast and as expected, SHH-dependent GTML1 cells and control cerebellar NSCs transduced with GFP virus showed multipotent differentiation into both GFAP-positive astrocytes and TUJ1-positive neurons (Figures 2B, S2F and not shown).

An N-MYC neuronal program depends on both developmental age and regional origin

To characterize the potential for N-myc to induce distinct differentiation programs as a function of both NSC age and regional origin, we analyzed the glial markers SOX9 (Stolt et al., 2003) and GFAP, 72 hours after transduction of N-myc^T58A into cerebellar NSCs at E16 and P0 and forebrain NSCs at P0. N-myc^T58A-transduced NSCs all showed increased N-myc mRNA levels as by Affymetrix exon array analysis of distinct NSC populations. P0 cerebellar NSCs transduced with N-myc^T58A showed reduced mRNA expression of Gfap and protein levels of SOX9 as compared to E16 cerebellar NSCs; whereas P0 forebrain NSCs showed elevated levels of Gfap and SOX9 (Figure 3A).

As SOX9 has been reported to also act downstream of the SHH/SMO to induce and maintain NSCs (Scott et al., 2010), we next asked whether N-myc^T58A affected SHH signaling. GFP-transduced control NSCs from E16 and P0 cerebellum and forebrain all responded to SMO inhibition through cyclopamine treatment (Figures 3B and 3C), consistent with reports that forebrain NSCs, cerebellar radial glia, and cerebellar NSCs are sensitive to SHH signals in vivo (Ahn and Joyner, 2005; Huang et al., 2010). While N-myc^T58A-transduced E16 cerebellar cells and N-myc^T58A-transduced P0 forebrain cells retained sensitivity to cyclopamine, N-myc^T58A-transduced P0 cerebellar and E16 forebrain cells, like GTML tumor spheres, were resistant (Figures 3D and 3E).

In comparing cyclopamine sensitive and resistant N-myc^T58A-transduced cerebellar cultures, we again used Affymetrix exon arrays to analyze expression of Smo, a co-receptor for the SHH signal; Sfrp1, a marker of SHH-driven MB; Gli1, a marker of SHH activation, and Gli3, a marker of SHH repression. Expression of Smo was unchanged in comparing P0 and E16 cultures (Figures S3A and S3B). Cyclopamine-resistant P0 cultures showed repression of SHH signaling. Both Sfrp1 and Gli1 were decreased, while Gli3 was upregulated in P0 N-myc^T58A cultures, as compared to E16 cerebellar NSCs. Collectively, these studies demonstrate that N-myc^T58A promotes SHH independence (downstream of SMO) in postnatal cerebellar cells, in embryonic forebrain cells, and in GTML tumors.

N-myc^T58Ainitiates brain tumors from NSCs

To clarify whether N-MYC could induce brain tumors from NSCs, and whether the differentiative programs shown in Figure 3 would persist in vivo, we transduced embryonic and postnatal NSCs with N-myc^WT or N-myc^T58A and transplanted cells orthotopically into the brains of nude mice. In contrast to N-myc^WT NSCs, N-myc^T58A NSCs from E16 and P0 cerebellum or E16 and P0 forebrain all formed brain tumors (Figures 4A and 4B). N-myc^T58A cerebellar NSCs generated tumors at higher incidence and penetrance than the N-myc^T58A forebrain NSCs (Figures 4A and 4B). We also analyzed LRL-derived NSC. Consistent with our in-vitro data, N-myc^T58A E16 and P0 LRL NSCs failed to generate brain tumors, while N-myc^T58A –transduced E14 LRL NSCs generated brain tumors in 40% of mice when transplanted orthotopically (Figure 4C).

Orthotopic transplantation of N-myc^T58A P0 cerebellar NSCs into cerebellum resulted in massive tumors with morphology of human MB, including Homer Wright (neuroblastic) rosettes, and prominent cell wrapping characteristic of LC/A tumors (Figures 4D, 4E and Figure S4A)). The histology of N-myc^T58A P0 cerebellar tumors also resembled GTML tumors (Swartling et al., 2010), sometimes showing LC/A histopathology (Figures S4B–S4C). Similarly, orthotopic transplantation of N-myc^T58A P0 forebrain NSCs into cerebrum induced aggressive forebrain tumors (Figure 4F) that were more invasive than the cerebellar tumors (Figure 4G). The P0 forebrain tumors resembled diffuse gliomas in terms of distant spread of individual tumor cells along white matter tracts (Figure 4G and S4D). E14 LRL tumors were typically characterized by marked cellular and nuclear pleomorphism (Figure 4H–I).

While no differences in proliferation (KI67) and vascularity (CD34) were found among the P0 forebrain and P0 cerebellar tumors, P0 cerebellar tumors were significantly more apoptotic than P0 forebrain tumors (Figures S4E–S4J). P0 cerebellar tumors were generally negative or showed only scattered cells positive for the astrocytic markers GFAP, OLIG2, and SOX9 (Figures 4J–4L) and moderate immunoreactivity for the neuronal marker SYP (Figure 4M). In contrast, P0 forebrain tumor cells were strongly positive for GFAP, OLIG2 and SOX9 (Figures 4N–4P) and negative for SYP (Figure 4Q). LRL tumors showed an intermediate phenotype with few scattered cells positive for the astrocytic markers GFAP, moderate positivity for OLIG2 and SOX9 (Figures 4R–4T) and scattered islands of SYP positive cells (Figure 4U).

The histology of N-myc^T58A tumors induced from E16 cerebellar, E16 forebrain NSCs generally resembled N-myc^T58A tumors induced from P0 cerebellar and P0 forebrain NSCs, respectively (Figures S4K and S4L). The majority of tumors derived from E16 cerebellar NSCs were SOX9 positive (Figure S4M). Most embryonic forebrain tumors showed distinct areas of glial differentiation (Figure S4N). Lack of nuclear β-catenin expression in LRL tumors (Figure S4O) suggests a WNT-independent tumor subtype, as compared to a recently reported model for MB also derived from LRL cells originating from brainstem (Gibson et al., 2010). These and tumors from E16 or P0 cerebellum were similarly insensitive to the GSK3β inhibitor TWS119 at 1μM (not shown) demonstrating WNT-independence, although it remains possible that N-myc^T58A affects the ability of GSK3β to phosphorylate and destabilize β-catenin.

Regional origin exerts a dominant effect on brain tumor type

Transplantation of normal cerebellar NSCs into forebrain VZ generates mostly GFAP-positive astroglial cells (Klein et al., 2005). To determine if the regional brain niche was responsible for the generated tumor phenotypes we injected N-myc^T58A –transduced P0 cerebellar NSCs into the forebrain and N-myc^T58A –transduced P0 forebrain NSCs into the hindbrain. We studied the same markers as in Figure 4 and quantified positive cells (Table S1). We included three controls each of site-matched tumors for E16 and P0 cerebellum, tumors from E16 forebrain, P0 forebrain and E14 brain stem; and three GTML tumors.

For all tumors, we also examined protein expression of KCNA1, a Kv1.1 voltage-gated potassium channel-encoding gene associated with Group 4 MB (Taylor et al., 2011) as well as the LRL and WNT markers OLIG3 and β-catenin, respectively (Gibson et al., 2010). The profiles of GFAP, Nestin and SYP in tumors correlated generally with region from which NSCs were isolated, rather than region of the brain into which we injected transduced cells. Like the LRL tumors (Figure S4O), all tumors examined were negative for nuclear β-catenin (Table S1). Intriguingly, SOX9 and OLIG2 showed a modest and differential expression pattern in “misplaced” tumors, with higher levels of protein expression in misplaced hindbrain tumors and lower expression levels in misplaced forebrain tumors (Table S1). Our data still suggests that region of origin exerts a dominant effect on tumor type.

SOX9 marks SHH-dependent MB and malignant glioma

Transduction of N-myc^T58A into murine NSCs led to regional and age-dependent changes in levels of SOX9 protein (Figure 3A). To further clarify the expression of SOX9 in MB, we analyzed array data from 103 human MBs segregated into WNT, SHH, Group 3 or Group 4 tumors (Northcott et al., 2011; Taylor et al., 2011). Clustering and expression patterns were consistent with both SOX9 and MYCN as markers for WNT- and SHH-driven human tumors (Figure 5A). NGN1 expression was increased in all subtypes except for the SHH group, marked by the human MATH1 ortholog, ATOH1, and supported by previous data discriminating these subclasses (Cho et al., 2011; Kool et al., 2008; Northcott et al., 2011; Thompson et al., 2006). Consistent with the findings of SOX9 as a marker of SHH-driven tumors, levels of SOX9 levels were highest in infant and adult MB patients (Figure S5), which are typically SHH-dependent (Northcott et al., 2011). Furthermore, SOX9 was only observed in a few scattered cells in a representative human classic MB (Perry et al., 2009) previously reported to lack MYCN amplification (Figure 5B), while a human MYCN-amplified MB showed high levels of SOX9 (Figure 5C).

The pathology and marker-expression in P0 forebrain tumors is characteristic of aggressive human malignant glioma with components of primitive neuroectodermal tumors (MG-PNET). We recently described 53 cases of human MG-PNET, detailing both poor survival (essentially identical to human high-grade glioma) and common amplification of MYC and MYCN (43%) in PNET-like foci (Perry et al., 2009). Since MG-PNET may arise in the setting of relapsed glioblastoma (GBM), we next analyzed data from The Cancer Genome Atlas (TCGA) (2008). While OLIG2 and SOX9 were overexpressed in half or more than half of the GBM samples, expression of neuronal markers like NGN1 and SYP were downregulated in most samples (Table S2). Among 424 human GBMs, 20% overexpressed MYCN, consistent with a role for MYCN in the pathogenesis of some GBMs (Table S2). Irrespective of MYCN amplification, human MG-PNET samples showed high level and distinct nuclear expression of SOX9 (Figures 5D and 5E), consistent with results from glioma cell lines (Swartling et al., 2009) and TCGA data (Table S2).

Alignment of murine tumors and classification using human MB subgroup identifiers

Having demonstrated that SOX9 can mark SHH-driven human MB, we next applied these observations to murine N-myc^T58A MB and GTML tumors, using Affymetrix exon arrays. Unsupervised clustering revealed that N-myc^T58A tumors from E16 cerebellum, P0 cerebellum and P0 forebrain NSCs were closely aligned to the NSCs from which they were derived (Figure 6). N-myc^T58A tumors and their corresponding NSCs were more similar to GTML tumors than to normal cerebellum (Figure 6). Within the N-myc^T58A tumor group, the region of origin (cerebellum vs forebrain) was a larger separator than the age of isolation of the initiating cells. Originating NSCs, N-myc^T58A tumors, and GTML tumors were all more similar to each other than to normal cerebellum. The signature of NSCs vs tumor was greater than the difference between NSC types, precluding the use of gene expression signatures to identify the most similar potential originating cell.

Using identifiers for subtypes of human MB (Northcott et al., 2011), E16 MB (generated from N-myc^T58A-transduced E16 cerebellar NSCs) showed a distinct SHH-pathway profile (Figure S6A) while both P0 MB (generated from N-myc^T58A-transduced P0 cerebellar NSCs) and GTML tumors presented significantly higher expression of the Group 4 identifier KCNA1, as compared to E16 MB. By contrast, the other subgroup identifiers, DKK1 (WNT), SFRP1 (SHH), and NPR3 (Group 3), did not significantly delineate these groups (Figure S6B).

E16 N-myc^T58A cerebellar tumors model SHH-dependent MB, while P0 tumors model SHH-independent disease

Expression levels of Sox9, N-myc and Math1 further distinguished the various N-myc^T58A tumors using both Affymetrix arrays as well as real-time qPCR, as compared to expression levels in adult cerebellum (Figures 7A–7C and Figure S7A and S7B). E16 MB showed high levels of Sox9 mRNA, N-myc, and of the EGL marker Math1 suggesting a SHH- rather than a WNT-dependent subtype (Figures 7A–7C). In contrast, P0 MB displayed lower expression of Sox9, N-myc and Math1. Similar to P0 MB, E14 LRL tumors (generated from N-myc^T58A-transduced prenatal LRL NSCs) showed low N-myc and Math1 levels consistent with a SHH-independent origin (Figures 7A–7C).

Interestingly, forebrain N-myc^T58A P0 tumors (P0 glioma) showed high levels of Sox9 and N-myc and absent Math1 (Figures 7A–7C), consistent with transformation of a forebrain cell type that shows SHH-dependence independently of Math1 (Akazawa et al., 1995). Expression analysis substantiated the glioma phenotype observed in the forebrain tumors, in which levels of Gfap were significantly elevated as compared to hindbrain tumors (Figure S7B). Moreover, Nestin (Nes) was among the top 30 genes overexpressed in P0 glioma as compared to P0 MB (Table S3). P0 MB showed a low immunoreactivity of Nestin while P0 glioma showed elevated Nestin expression (Figure S7C–D and Table S1). As Nestin is a strong marker for glioma and upregulated in 99% of human GBM samples in the TCGA database (Table S2), increased Nestin expression suggests a glioma-like differentiation pattern in forebrain tumors compared to the MB-like pattern seen in cerebellar tumors.

In agreement with SOX9 as a marker for SHH-dependence, cultured P0 MB spheres with low levels of SOX9 were resistant to cyclopamine (Figure 7D). Conversely, both E16 MB and P0 glioma with high SOX9 levels were cyclopamine-sensitive (Figure 7D). Similar dependence on SHH was also observed in N-myc^T58A NSCs prior to orthotopic implantation and tumor formation (Figures 3B–E). These results suggest that N-myc^T58A initiates SHH-dependent transformation in both E16 cerebellar NSCs and P0 forebrain NSCs, while transformation of E16 forebrain NSCs and P0 cerebellar NSCs (and possibly in brain stem NSCs) occurs through a SHH-independent pathway.

SOX9 promotes self-renewal and generates GLI2-expressing tumors at shorter latency

Our data demonstrates that P0 MBs arise through a SOX9- and SHH- independent program. To establish whether SOX9 plays a functional role in transforming P0 cerebellar NSCs, we generated SOX9 RCAS viruses (Figure S8A) and selectively transduced Gtv-a positive P0 N-myc^T58A cerebellar NSCs or P0 MB spheres (isolated from individual P0 MBs). Forced expression of SOX9 suppressed proliferation in P0 N-myc^T58A cerebellar cells and in P0 MB spheres (Figure S8B), which have low levels of SOX9 (Figures 3A, 4I and 7A). In contrast, similar forced expression of SOX9 did not affect E16 MB and P0 glioma, which have high SOX9 levels at baseline (Figures S8C–D).

Although forced expression of SOX9 suppressed proliferation in N-myc^T58A cerebellar cells and in P0 MB spheres, SOX9 actually drove increased self-renewal when culturing these at a limited dilution. Single cell clones from normal P0 cerebellar NSCs and P0 cerebellar NSCs transduced with N-myc^T58A were transduced with SOX9, and clones with high levels of SOX9 were selected (Figure 8A). SOX9 increased self-renewal in both normal and N-myc^T58A –transduced P0 cerebellar NSC clones, as compared to normal and N-myc^T58A –transduced P0 cerebellar NSC clones without forced expression of SOX9 (Figure 8B). Single cell spheres and spheres generated from an isolated P0 MB (P0C-T3, also cultured and propagated from a single cell clone) were dependent on growth factors EGF and FGF as compared to single cell clones subcloned out from the GTML2 sphere line (Figure 1A) that generated secondary spheres essentially independently of growth factors. Forced expression of SOX9 was also associated with higher levels of Gli2 (Figure 8C–E) suggesting that SOX9 can drive SHH-signaling in these cells.

Since forced expression of SOX9 drove self-renewal while blocking proliferation of cells in-vitro, we next asked how these potentially opposing activities would impact tumor formation in vivo. Forced expression of SOX9 and N-myc^T58A in P0 cerebellar NSCs generated orthotopic tumors at shorter latency and with increased tumor penetrance, as compared with NSCs expressing N-myc^T58A alone (Figure 8F). The SOX9 expressing tumor cells demonstrated marked cellular and nuclear pleomorphism, numerous mitoses, and numerous apoptotic bodies, consistent with a primitive neuroectodermal phenotype (Figure S8E). Focally tumor cells contained abundant cytoplasm and GFAP-immunostaining demonstrated a population of GFAP-positive tumor cells. (Figures S8E and S8F). These observations suggest that SOX9 can cooperate with N-MYC to promote malignant progression acting upstream of GLI2.

Forced expression of N-myc^T58A repressed SOX9 in P0 cerebellar NSCs. Is SOX9 similarly repressed in GTML tumors? In response to dox-mediated withdrawal of MYCN, some cells within GTML tumors undergo senescence (Swartling et al., 2010). Consistent with this result, dox-treatment of GTML3 tumor spheres led to death of most cells, while a few cells remained viable without proliferating (Figures S8G-S8H and data not shown). In response to withdrawal of MYCN, these viable cells showed elevated levels of SOX9 at levels similar to SHH-dependent GTML1 control cells (Figure 8G). This suggests SOX9 is repressed by forced expression of MYCN also in SHH-independent GTML tumors. Collectively, data in Figure 8 demonstrate that interactions between N-MYC and SOX9 regulate proliferation, differentiation, and self-renewal in P0 N-myc^T58AMB, and that similar interactions are relevant to MYCN-driven GTML MB.

Animals and Bioluminescent Imaging

Generation of GTML mice of the FVB/NJ strain has been previously described (Swartling et al., 2010). Tumor progression of doubly Glt1-tTA/MYCN-TRE transgenic mice was continuously followed by luciferase signaling using the IVIS Lumina (Caliper Life Sciences, Mountain View, CA) using Living Image 2.5 software (Caliper Life Sciences, Mountain View, CA). Mice were injected with 75 mg/kg of sodium luciferin (LUCNA, Gold Biotechnology, St. Louis, MO) in saline prior to imaging. The mice selected for tumor removal and neurobasal cell culture all had high scores (>5.0×10¹⁰ photons/cm² sec) from bioluminescence imaging suggesting a functional bi-directional transgene expressing luciferase as previously reported (Swartling et al., 2010). For cell fate experiments Glt1-tTA mice were crossed with TRE-Cre:R26R-LSL-LacZ mice obtained from Dr. Robert Blelloch, UCSF. The G-tva mice has been described before (Holland and Varmus, 1998) and were generously provided by Dr. Eric Holland and backcrossed at least 5 generations into FVB/NJ strain before experimental use. Athymic Nude-Foxn1nu mice were obtained by Simonsen Laboratories (Gilroy, CA) or Harlan Laboratories (Venray, The Netherlands) and used in experimental procedures as described below. Mice were maintained in the Animal Facilities at University of California, San Francisco and at Uppsala University, Uppsala. All animal procedures were performed in accordance with national guidelines and regulations and approved by the Institutional Animal Care and Use Committee of the University of California in San Francisco and Uppsala Ethical Committee on Animal Experiments in Uppsala.

Neural stem/progenitor cell and tumor cell culturing

Normal tissue or part of tumor was removed and mechanically dissolved in cold Hanks buffer without Calcium and Magnesium (Cell Culture Facility, UCSF, Mission Bay, San Francisco). After one more wash in cold Hanks buffer sample was incubated 20 minutes in activated papain (Worthington, Lakewood, NJ). Tumor cells were gently dissociated and cultured in neurobasal media (Gibco) without Vitamin A, supplemented with antibiotics (Cell Culture Facility, UCSF), B27 without Vitamin A (Invitrogen), L-glutamine and 20ng/ml EGF (Sigma) and 20ng/ml FGF-2 ((Peprotech, Rocky Hill, NJ, USA)). Cells were cultured and propagated on low adhesion plates (Corning) for 1 week, further passaged and used in experiments. SVZ neurospheres were established from FVB/N WT mice and similarly cultured as described above. Forebrains of E16.5 (described as E16) and P0.5 (described as P0) were coronally sectioned (100 μm thick sections) and sections 2.5–3.5 mm posterior of the base of olfactory bulb (OB) and the dorsal regions closest to the ventricles were isolated and dissociated as above. Total cerebellum was carefully removed excluding any parts of meninges, brain stem or midbrain, dissociated and similarly cultured as above. GNPs were enriched and cultured as previously described in 10ng/ml SHH-N (SHH N-terminal peptide, R&D Systems) containing serum-free media (Swartling et al., 2010).

Array Experiments

Expression analysis (as presented in Figure 5 and Figure S5) was performed on 103 primary human medulloblastoma using Affymetrix exon arrays as previously described (Northcott et al., 2009; Swartling et al., 2010). All tumor specimens were obtained in accordance with the Research Ethics Board at the Hospital for Sick Children (Toronto, Canada) and de-identified prior to analyses (Northcott et al. 2009). We studied the expression levels of SOX9, MYCN, MATH1, OLIG2 and NGN1 in four distinct molecular variants described as WNT, SHH, group C, and group D tumors as previously identified by multiple unsupervised analyses of these medulloblastoma samples (Northcott et al., 2010). RNA from tumors was isolated using Trizol (Invitrogen, Carlsbad, CA) and purified using the RNeasy Mini Kit (QIAGEN, Valencia, CA). For the GTML tumors, 1 μg of RNA was used as a starting template for the RiboMinus rRNA subtraction protocol (Invitrogen) followed by the ST labeling protocol (Affymetrix, Santa Clara, CA). For the transplanted tumors, 100 ng of RNA was used as a starting template for the Ambion WT Expression protocol (Applied Biosystems, Carlsbad, CA) followed by the WT Terminal Labeling protocol (Affymetrix). Labeled samples were hybridized to Affymetrix Mouse Exon 1.0 arrays. All arrays were normalized together with expression values calculated using RMA in the XPS package in R. Boxplots were generated in R using the standard graphics package. 2-way comparisons between groups were performed using Significance of Microarrays (SAM) (Tusher et al., 2001).