Atoh1 Inhibits Neuronal Differentiation and Collaborates with Gli1 to Generate Medulloblastoma-Initiating Cells

Granule neuron progenitors (GNPs) are specified during embryogenesis and undergo transient proliferation in the external germinal layer (EGL) of the developing postnatal cerebellum, after which they exit the cell cycle, migrate inwardly, and differentiate into mature granule neurons (1). Sonic Hedgehog (Shh), released from Purkinje neurons underlying the EGL, negatively regulates the Patched receptor (Ptch1) expressed on GNPs, thereby relieving its repression of Smoothened (Smo) and triggering a signal transduction pathway that targets Gli transcription factors to induce GNP proliferation. Deregulated pathway activation resulting from Ptch1 inactivation in GNPs, or from mutations affecting downstream transducers in the Shh signaling pathway, can induce medulloblastoma (MB) in mice and humans (2). Mutations constitutively activating the Shh signaling pathway can occur either in multipotent cells within the cerebellar ventricular zone or in lineage-committed GNPs in the EGL. MBs manifesting a Shh gene expression “signature” preferentially arise from, and retain the molecular and phenotypic features of, proliferating GNPs (3–7). Yet, despite constitutive activation of the Shh signaling pathway, MB-prone GNPs can still undergo cell cycle exit and further neuronal differentiation, suggesting that additional lineage-specific factors are required to maintain potential tumor-initiating progenitors in an undifferentiated, continuously proliferating state.

The basic helix-loop-helix transcription factor Atoh1 (Math1) is a candidate for playing this role. Deletion of Atoh1 severely compromises GNP histogenesis during cerebellar development (8) and prevents MB development in vivo (9). Atoh1 expression is restricted to proliferating GNPs within the EGL and is increased in that subset of MBs with Shh pathway mutations (3, 10–12), suggesting that it may contribute directly to MB development (13, 14). However, the role of Atoh1 in preventing differentiation of GNPs and enhancing MB formation is in stark contrast to its role as a tumor suppressor and effector of differentiation in other tissues, including the colon (15, 16), the sensory epithelium of the inner ear (17), and the fly retina (18). These opposing outcomes may be governed by the availability of other tissue-specific transcription factors and coregulators with which Atoh1 interacts.

We previously reported that the bone morphogenic proteins (BMPs), BMP2 and BMP4, antagonize MB formation in mice by triggering the proteasomal degradation of Atoh1; however, the ensuing cell cycle exit and neuronal differentiation of BMP2/4-treated MB cells can be overridden by enforced Atoh1 expression (19). Here, we show that although Gli1 expression enables GNPs to proliferate in the absence of Shh, enforced Atoh1 expression is unable to substitute for Shh signaling. We also now show that by inhibiting neuronal differentiation, constitutive Atoh1 expression greatly enhances the pro-proliferative effects of Shh/Gli1 signaling in transforming primary GNPs to MBs.

cDNAs for Gli1, Atoh1, Atoh1-ER [a 3′ fusion to an estrogen-responsive element engineered to be induced by 4-hydroxytamoxifen (4-HT)], and Cre recombinase were cloned into mouse stem cell virus–internal ribosome entry site (IRES)–lgreen fluorescent protein (GFP) or red fluorescent protein (RFP) vectors. Purified GNPs or MB cells infected during the preplating stage or within 24 hours of plating (4 × 10⁶ cells per well) were quantified for marker expression and injected into cerebral cortices of C57BL/6 or CD-1 nu/nu mice (all procedures are described in ref. 20). Nuclear magnetic resonance imaging (MRI) was performed on anesthetized animals (2–3% isoflurane in O₂) using a 7-Tesla Clinscan scanner (Bruker BioSpin MRI GmbH) equipped with a 12S gradient (BGA12S) and four-channel phased-array surface coil. Turbo Spin Echo protocols (T_R 2,500–3,800 ms; T_E 39–42 ms) produced T2-weighted images (sagittal, coronal, and axial) using a matrix of 320 × 320 and field of view of 25 × 25 mm. Images were read using Syngo MR B15 software (Siemens).

Protein detection was performed (12, 20) with antibodies to Tuj1 (Covance), p27^Kip1 (BD), NeuN (Millipore), NeuroD1, p21^Cip1, actin (Santa Cruz), Atoh1 (Developmental Studies Hybridoma Bank), GFAP and synaptophysin (DAKO), Ki67, Gli1 (Rockland), p53 (Cell Signaling), MycN (Calbiochem), and Pax6 (Covance). The percentage of NeuN- and p27^Kip1-stained nuclei within at least 600 GFP-positive cells was calculated. Where indicated, human recombinant BMP4 (R&D; 100 ng/mL) and 4-HT (Invitrogen; 2 μmol/L) were added to cultures. Cultured GNPs were incubated with 10 μmol/L bromodeoxyuridine (BrdUrd) for 1.5 hours with incorporation quantified by staining with a fluorochrome-conjugated antibody to BrdUrd (APC-BrdUrd kit, BD Biosciences) and detected by fluorescence-activated cell sorting (FACS) analysis. Sorting of GNPs marked with GFP and RFP was done as previously described (19), except that infected cells were grown in culture for 2 days before sorting and injection into the cerebella of recipient mice. Confocal imaging of immunofluorescence staining of tumors was performed with a Marinas spinning disc confocal imaging system (Intelligent Imaging Innovations/3i), consisting of a CSUX confocal head (Yokogowa Electric Corporation); 404, 488, and 561 nm DPSS lasers (Coherent, Inc.); and a Carl Zeiss Axio Observer motorized inverted miscroscope (Carl Zeiss MicroImaging). Images were acquired with a Zeiss Plan-Neofluar 40× 1.3 numerical aperture differential interference contrast objective on a Evolve EMCDD camera (Photometrics), using SlideBook 5.0 software (3i). Images are maximum intensity projections of z images taken at 0.5 μm intervals.

Gene expression profiles generated using GeneChip Mouse Genome 430 2.0 arrays (Affymetrix; refs. 3, 11, 14) were normalized using an Affymetrix Mas5 algorithm, and differentially expressed genes were functionally classified using DAVID Bioinformatics Resource 2008 (21). For determining kinetic responses to 4-HT, we applied k-means clustering using Spotfire (v9.1.1.). Quantitative reverse transcription-PCR (QRT-PCR) primers and probes are described in Supplementary Table S1.

MBs arise spontaneously, but with low penetrance and relatively long latency, in mice lacking one allele of the Ptch1 gene encoding the Shh receptor (2); however, tumorigenesis is accelerated when the Cdkn2c gene (encoding the Cdk4/6 inhibitor p18^Ink4c) is also inactivated (20). To investigate a role for Atoh1 in MB development, we used retroviral vector–mediated gene transfer to enforce expression of either Atoh1 or Gli1 (a direct pro-proliferative target of the Shh signaling pathway; ref. 22) in primary GNPs purified from the cerebella of postnatal (P) days P5 to P7 Ptch1^+/−;Cdkn2c^−/− (C57BL/6 × 129) mice. Both vectors expressed GFP synthesized in cis from an IRES. Triturated GNPs purified on Percoll step gradients were transduced with the vectors and injected immediately thereafter into the cerebral cortices of immunocompromised CD-1 nu/nu recipient animals. FACS analysis of infected GNPs performed at the time of transplantation indicated that 20 ± 3% of the donor cells expressed GFP, enabling us to estimate that individual recipients received ∼4 × 10⁵ marked cells. Mice observed for subsequent tumor development were sacrificed after initial signs of morbidity, and necropsies were performed to document MB formation.

Mice receiving Ptch1^+/−;Cdkn2c^−/− GNPs infected with a control vector encoding GFP alone developed GFP-marked tumors after a mean latency of 160 days, whereas enforced Gli1 expression significantly shortened the latency and increased the penetrance of MB development (Fig. 1A). Surprisingly, enforced expression of Atoh1 in Ptch1^+/−;Cdkn2c^−/− GNPs had an even more profound effect on tumorigenesis (Fig. 1A). In later studies, transduced donor cells explanted from a backcrossed Ptch1^+/−;Cdkn2c^−/− pure C57BL/6 strain proved equally tumorigenic when transplanted either into the cerebral cortices or cerebella of healthy syngeneic animals. Histopathologic analyses confirmed that all tumors were MBs (Fig. 1B, a and g; Supplementary Fig. S1A) that expressed Pax6, a marker of GNPs (Fig. 1B, f and l; Supplementary Fig. S2). We confirmed that Gli1 and Atoh1 were expressed in these tumors (Supplementary Fig. S1B). Although MBs induced by Gli1 overexpression retained and expressed the wild-type Ptch1 allele, those induced by Atoh1 exhibited no detectable Ptch1 expression (Supplementary Fig. S1C). This implies that, unlike Gli1, Atoh1 does not increase the threshold of Shh signaling and, like tumors arising spontaneously in Ptch1^+/− mice (10, 20), requires complete loss of Ptch1 expression to ensure MB formation.

MBs induced after Gli1 transduction continued to express detectable levels of the neuronal differentiation markers Tuj1, p27^Kip1, NeuroD1, and NeuN (Fig. 1B, b–e; Supplementary Fig. S1B), whereas those induced by Atoh1 expressed reduced levels of these proteins (Fig. 1B, h–k; Supplementary Figs. S1A,B and S2). To quantify differences in gene expression between MBs induced by overexpression of Gli1 and Atoh1, RNAs extracted from each of three MBs arising from Ptch1^+/−;Cdkn2c^−/− GNPs transduced with retroviral vectors expressing either Gli1 or Atoh1 were subjected to Affymetrix microarray analysis (three chips per sample). Of the total genes that exhibited significant (>2-fold) upregulation or downregulation in tumors overexpressing either Gli1 or Atoh1, about a quarter responded concordantly, whereas many more were differentially regulated (Supplementary Fig. S3A). To elucidate categorical differences in gene expression between the two MB subsets, genes undergoing >2-fold upregulation or downregulation in Atoh1-induced MBs versus Gli1-induced MBs were assigned to functional groups using Gene Ontology (DAVID database; ref. 21). Scrutiny of all differentially regulated genes revealed that the vast majority of them control processes of cell adhesion, morphogenesis, development, neurogenesis, and neuronal differentiation, but not cell proliferation or apoptosis (Supplementary Table S2). A subgroup of genes encoding proteins well recognized to be involved in neuronal differentiation, migration, and adhesion were perturbed by Atoh1 overexpression, whereas genes canonically regulating GNP cell cycle progression were expressed at equivalent levels in both subsets of MBs (Table 1). More specifically, Tuj1, p27^Kip1, NeuroD1, and Tag1, all markers of neuronal differentiation, were downregulated in MBs expressing Atoh1 compared with those overexpressing Gli1. In contrast, Pax6, MycN, and Zic1 that identify GNPs were unchanged (Supplementary Fig. S3B). Thus, although some genes were regulated similarly, the overall consequences of enforced Gli1 and Atoh1 expression in tumor-prone Ptch1^+/−;Cdkn2c^−/− GNPs were markedly different, with Atoh1 having more profound effects on neuronal differentiation.

To determine whether Atoh1 and Gli1 might induce MBs even in the absence of other predisposing mutations, we infected primary GNPs from the P5-P7 cerebella of normal C57BL/6 mice with the vector expressing Gli1/GFP alone or together with an Atoh1 vector coexpressing RFP in lieu of GFP. Wild-type GNPs infected with a control vector expressing GFP alone or Atoh1/RFP (∼2.2 × 10⁵ marked donor cells) failed to induce tumors in the cortices of syngeneic C57BL/6 recipients within 6 months of transplantation (Fig. 2A). In contrast, enforced expression of Gli1/GFP in GNPs resulted in “green” tumors that killed 80% of recipient mice within 200 days postinjection (t_1/2 = 159 days; Fig. 2A). Remarkably, mice receiving primary GNPs engineered to overexpress both Gli1 and Atoh1 invariably and rapidly succumbed to tumors that uniformly expressed both fluorescent markers (t_1/2 = 26.5 days; Fig. 2A and C, b; Table 2). Because mice can survive even with high tumor burdens, we used MRI to monitor tumor growth. No tumors were detectable 5 to 11 days after transplantation of 224,000 GNPs overexpressing Gli1 alone (Fig. 2B, a and c), whereas 46,000 GNPs coexpressing Gli1 and Atoh1 induced detectable tumor masses as early as 5 days after transplantation (Fig. 2B, b and d). Tumor growth in these cases was so aggressive that recipient animals had to be sacrificed within 2 to 3 weeks of transplantation (Table 2).

The rapidity with which tumors coexpressing Atoh1/RFP and Gli1/GFP arose suggested that many, if not all, infected cells were capable of initiating MBs. We therefore conducted limiting dilution experiments to determine the minimal number of marked GNPs necessary to induce MBs in naïve recipient animals. Because sorting of donor cells before their transplantation reduces the efficiency of tumor induction, marked cells were injected immediately after vector transduction. We retained representative aliquots of these cell populations and retrospectively measured the number of primary GNPs that had been infected with retroviruses encoding Gli1/GFP (GFP+) or Gli1/GFP plus Atoh1/RFP (GFP+/RFP+) by FACS analysis 48 hours after infection. This allowed us to back-calculate the numbers of marked cells injected into the brains of recipient animals. Although >160,000 Gli1/GFP GNPs were required to induce tumors within 14 weeks, mice receiving as few as ∼200 GFP+/RFP+ cells succumbed to tumors as early as 1 month after transplantation (Table 2). Thus, Atoh1 expression enhanced the ability of Gli1 to transform GNPs by >800-fold and efficiently converted GNPs into tumor-initiating cells.

MBs arising from GNPs infected with retroviruses expressing Gli1/GFP plus Atoh1/RFP overexpressed Gli1 and Atoh1 (Supplementary Fig. S1B) and both GFP and RFP (Fig. 2C, a and b). They expressed Pax6, confirming that the tumor cells arose from GNPs (Fig. 2C, g; Supplementary Fig. S2), and relatively low levels of the neuronal differentiation markers Tuj1, NeuroD1, NeuN, and p27^Kip1(Fig. 2C, c–f; Supplementary Figs. S1B, S2). Spectral karyotyping of chromosomes of Gli1/Atoh1–accelerated tumors revealed no gross chromosomal anomalies (Supplementary Fig. S4A). Although p53 mutations significantly accelerate MBs induced by deregulated Shh signaling (20, 23), MBs induced by enforced coexpression of Gli1 and Atoh1 retained functional p53, which was rapidly induced, together with its transcriptional target p21^Cip1, after irradiation of primary tumor cells (Supplementary Fig. S4B). Together, these findings underscore the synergy between Atoh1 and the Shh signaling pathway in inducing MBs and further reveal that coexpression of Gli1 and Atoh1 is sufficient to guarantee the conversion of primary GNPs into MB-initiating cells.

After birth, Atoh1 expression in the cerebellum depends on Shh signaling and is detected only in proliferating GNPs within the EGL (8, 11). Wild-type P7 GNPs infected with retroviruses encoding Gli1/GFP or Atoh1/GFP were cultured with or without Shh. Although enforced expression of Gli1 readily stimulated GNP proliferation in the absence of Shh, GNPs infected with the control GFP vector or with a vector expressing Atoh1/GFP failed to proliferate without Shh stimulation and exited the cell cycle (Fig. 3A). However, enforced Atoh1 expression enabled a greater fraction of GNPs to reenter the cell cycle after transient Shh deprivation (Fig. 3B) and enhanced their overall mitogenic response when they were stimulated continuously by Shh during a 7-day culture period (Supplementary Fig. S5A). GNPs cultured in the continued presence of Shh can be induced to undergo further neuronal differentiation by BMP4 treatment (19). However, unlike GNPs transduced by a control GFP–expressing vector, those transduced by Atoh1/GFP resisted the effects of BMP4 in inducing the neuronal differentiation markers, NeuN and p27^Kip1 (Supplementary Fig. S5B). When P7 GNPs explanted from mice homozygous for a “floxed” Atoh1 allele (9, 24) were infected with a retrovirus encoding Cre recombinase and GFP, Atoh1 expression was lost (Fig. 3C, top), and the infected cells exited the cell cycle more rapidly (Fig. 3C, bottom) and underwent premature differentiation (Fig. 3D shows NeuN, a representative marker) even in the continued presence of Shh. Thus, although Atoh1, unlike Gli1 alone, is unable to drive proliferation itself, Atoh1 enhances the ability of Shh-driven GNPs to divide by inhibiting cell cycle exit and neuronal differentiation, and thereby maintains their sensitivity to its mitogenic effects.

With the goal of extending the identification of Atoh1-regulated genes involved in neuronal differentiation, we performed a kinetic analysis with a 4-HT–inducible Atoh1-ER fusion protein (25). We first introduced Atoh1-ER into various cultured cell lines and confirmed that 4-HT treatment mobilized the transcription factor to the nucleus (Supplementary Fig. S6A), leading to its stabilization in complexes with other E-box binding proteins and to efficient induction of an Atoh1-responsive luciferase reporter gene (Supplementary Fig. S6B and C). We next introduced the validated Atoh1-ER vector into primary P7 GNPs, cultured them for 48 hours in medium containing Shh alone to establish infection, and then added BMP4 to the medium for an additional 24 hours. 4-HT was added either together with BMP4 (to induce Atoh1-ER for 24 h), or 16 or 20 hours afterward (yielding 8 and 4 h 4-HT induction periods, respectively); all induced cultures were harvested simultaneously, and total RNA was extracted for subsequent gene expression profiling. As an additional control, primary GNPs transduced with an Atoh1 mutant (R158G) that neither binds to DNA nor activates transcription (Supplementary Fig. S6C) were similarly treated and analyzed.

This protocol was based on previous observations that BMP4 induces proteasomal degradation of the endogenous Atoh1 protein to trigger neuronal differentiation, whereas enforced Atoh1 expression in these cells overrides these phenotypic effects (Supplementary Fig. S5B; ref. 19). Therefore, we reasoned that, in the absence of the endogenous Atoh1 protein, we might catalogue Atoh1-regulated genes that oppose the action of BMP4 on GNPs, acting to block their further neuronal differentiation and to maintain them in a Shh-responsive state. In keeping with these expectations, the expression of endogenous Atoh1 protein in GNPs cultured in the presence of Shh (Fig. 4A, a–c) was extinguished by BMP4 (Fig. 4A, d–f). Uninduced GNPs transduced with Atoh1-ER behaved similarly (Fig. 4A, g–l), but 4-HT treatment maintained robust expression of the fusion protein even in the face of BMP4 treatment (Fig. 4A, m–r).

Microarray gene expression profiling identified a total of 66 genes that were upregulated and 123 genes that were downregulated (fold changes >2) in response to Atoh1-ER induction. Representative “heat maps” of each of the top 50 genes in these categories (Fig. 4B) indicated that the upregulation or downregulation of these genes was not transient but progressed over the course of the 24-hour induction period. Introduction of the Atoh1 mutant defective in DNA binding had minimal effects on the expression of these same genes (designated “4M” in Fig. 4B), implying that viral infection and 4-HT treatment per se did not contribute significantly to their regulation. QRT-PCR was used to validate the expression of a short list of genes differentially regulated by Atoh1-ER (examples are illustrated in Supplementary Fig. S7). Notably, Gli1 gene expression was unaffected by induction of Atoh1-ER, consistent with the notion that it acts in a parallel pathway (19).

Of the 189 genes that showed >2-fold changes in expression, 138 could be functionally classified based on their annotation in Gene Ontology using the DAVID bioinformatics resource database (21). Only two biological processes were significantly affected (P < 0.01) by Atoh1-ER induction. By far the major class included 43 genes that were assigned to the cell differentiation category (P = 2.12E−06; Supplementary Table S3). Moreover, of only 15 genes listed within the cell proliferation category (P = 3.40E−03), 7 were coclassified as affecting cell differentiation as well (denoted by asterisks in Supplementary Table S3). Many of the genes have been previously implicated in BMP2/4 signaling (see Discussion). Therefore, not only is Atoh1 protein turnover and concomitant neuronal differentiation accelerated in response to BMP2/4 (19), but conversely, Atoh1 acutely counters these effects by rapidly inhibiting the expression of a multitude of BMP target genes.

Gene expression profiling of human MBs reveals that they can be subdivided into four distinct subgroups, two of which manifest either SHH or WNT pathway upregulation (26–28). In humans, less than a quarter of MBs manifest grossly aberrant SHH pathway activation, but most mouse models of MB, including those studied here, molecularly recapitulate the human SHH subgroup. Mouse MBs with a Shh signature seem to arise from GNPs in the postnatal cerebellum (3–7). In contrast, several lines of evidence suggest that WNT-driven MBs are generated from more primitive embryonic neuronal progenitors (27). The cells that initiate the other classes of human MBs have not been identified.

Shh drives the proliferation of GNPs during postnatal cerebellar development by inducing pro-proliferative genes that include Gli1, Gli2, the D-type cyclins Ccnd1 and Ccnd2, and Mycn (12, 29–31). Similarly, Atoh1 expression correlates with Shh signaling and is specifically increased in mouse and human MBs that exhibit Shh pathway mutations (3, 10–16). However, although Atoh1 expression is restricted to proliferating GNPs within the EGL and is essential for proper cerebellar histogenesis (8), Atoh1 is not a direct target of either Shh signaling or N-Myc in GNPs but functions in a parallel pathway (15, 19, 29). Notably, enforced expression of either N-Myc (31) or Atoh1 alone (this study) in primary GNPs is unable to induce MB, although both genes collaborate with the Shh signaling pathway to accelerate tumor onset. A major difference between them is that enforced N-Myc expression in primary GNPs grown in the presence of Shh is unable to counter the effects of BMP4 (15, 32), whereas Atoh1 can. Unlike GNPs overexpressing Gli1, those forced to express Atoh1, although able to proliferate in the presence of Shh, were unable to proliferate in its absence. Instead, when Shh was transiently removed from the culture medium, Atoh1 overexpression maintained quiescent GNPs in a prolonged Shh-responsive state, thereby synergizing with re-added Shh to enhance subsequent mitogenesis. Conversely, acute deletion of Atoh1 in GNPs limited their response to Shh and accelerated neuronal differentiation.

When GNPs explanted from MB tumor-prone Ptch1^+/−;Cdkn2c^−/− mice were transduced with vectors encoding either Gli1 or Atoh1 and transplanted into the brains of naïve animals, MB formation was more significantly accelerated by enforced Atoh1 expression. Consistent with the concept that Gli1 is a key arbiter of Shh signaling and that Atoh1 functions in a parallel pathway, MBs arising from Gli1-transduced donor cells did not inactivate the wild-type Ptch1 allele, whereas its expression was invariably lost in those MBs whose formation was enhanced by Atoh1. Expression of representative neuronal markers (Tuj1, NeuroD1, p27^Kip1, NeuN) was significantly reduced in response to Atoh1. Moreover, a detailed analysis of gene expression differences between MBs induced by overexpression of Gli1 and Atoh1 in this setting indicated that the majority of differentially regulated genes control the processes of cell adhesion, morphogenesis, development, neurogenesis, and neuronal differentiation, but not cell proliferation or apoptosis. Notably, the levels of expression of prototypic cell cycle genes, including Ccnd1, Ccnd2, and Mycn, each of which can accelerate MB formation in the Ptch1^+/−;Cdkn2c^−/− background (12), were not significantly different in MBs that were accelerated by Atoh1 versus Gli1. Together, these findings argue that Atoh1 overexpression inhibits the further differentiation of GNPs, thereby extending and enhancing their response to Shh/Gli1 pathway activation.

Shh-induced proliferation is opposed by a parallel BMP2/4-mediated signaling pathway that accelerates exit of GNPs from the cell division cycle and induces their concomitant differentiation (19, 32). BMPs trigger the proteasomal degradation of Atoh1, but enforced expression of Atoh1 overrides these BMP-mediated effects (19). To further investigate the mechanism by which Atoh1 antagonizes the differentiation-inducing effects of BMP2/4, we used a tamoxifen-inducible Atoh1-ER gene to override the effects of BMP4 on GNP gene expression and again applied microarray gene expression profiling to identify Atoh1-responsive genes. The protocol used for this experiment minimized expression of the endogenous cellular Atoh1 gene and allowed a kinetic analysis of genes that responded rapidly to conditional Atoh1 induction. An unsupervised functional classification of Atoh1-ER–responsive genes that exhibited at least 2-fold variations in expression identified 189 unique genes, 43 of which regulated cell differentiation and only 15 of which were annotated as genes controlling cell proliferation. Indeed, almost half of the latter were also categorized as governing differentiation processes. Included among these was a group of downregulated genes previously implicated in regulating BMP2/4 signaling, including Dlx1 and Dlx2 (33–35), Hhip (36), Runx2 (37–39), and Tbx1 (40), as well as genes such as Ntn1 and Slit that govern axon guidance (41), and Cxcr4, which encodes a G protein-coupled chemokine receptor that regulates GNP migration within the EGL (42). Otx2, the expression of which is essential for proper cerebellar development and has been found to be overexpressed in human MB cell lines (43, 44) was overexpressed in half of the MBs that arose spontaneously in tumor-prone mice exhibiting Shh pathway activation; however, its expression was not significantly changed in Gli1/Atoh1–overexpressing tumors. Moreover, although Flora and collaborators recently reported that Atoh1 can directly induce Gli2 (9), we failed to see an increase in Gli2 transcription after 4HT-mediated Atoh1-ER induction in primary GNPs. Pretreatment of cells with BMP before Atoh1 induction and/or variations in the timing of Gli2 induction in the different experimental settings might well account for the apparent discrepancies. We conclude that Atoh1 protein turnover and concomitant neuronal differentiation are not only accelerated in response to BMP2/4 signaling (19), but conversely, that Atoh1 counters these effects by inhibiting the expression of many differentiation-specific genes, BMP targets among them.

Most strikingly, the enforced coexpression of Atoh1 and Gli1 in primary GNPs explanted from the early postnatal cerebella of healthy C57BL/6 mice guaranteed their conversion into tumor-initiating cells, less than 200 of which induced MBs within 2 weeks of transplantation into the brains of naïve recipient animals. Remarkably, Gli1 and Atoh1 together seem sufficient to transform primary GNPs into MB-initiating cells. The fact that Atoh1 alone was incapable of inducing MBs in this assay but maximized the oncogenic potential of Gli1 underscores how anti-differentiative and pro-proliferative programs can synergize in initiating cancer.

No potential conflicts of interest were disclosed.

We thank Mary-Elizabeth Hatten, Michael Dyer, and Suzanne Baker for constructive criticisms of the manuscript; Jerold Rehg, Dorothy Bush, and David Elisson for immunohistochemistry and pathologic review; Richard A. Ashmun and Ann-Marie Hamilton-Easton for flow cytometric analysis; Melissa Johnson and John Killmar for help with orthotopic transplants; Jennifer Peters for confocal imaging of tumors; and Ziwei Zhang and Chris Calabrese for MRI. Robert Johnson, Jennifer Craig, Deborah Yons, Shelly Wilkerson, and Sarah Gayso provided excellent technical assistance. We thank Roger Y. Tsien (Department of Pharmacology, HHMI, University of California, San Diego, La Jolla, CA) for RFP cDNA, Jane E. Johnson (Center for Basic Neuroscience, UT Southwestern Medical Center, Dallas, TX) for the monoclonal antibody to Atoh1, and Huda Y. Zoghbi (Department of Molecular and Human Genetics, Baylor College of Medicine, Houston, TX) for supplying “floxed” Atoh1 mice.

Grant support: NIH grant CA-096832 (M.F. Roussel) and Cancer Core Grant CA-21765 (M.F. Roussel/C.J. Sherr), Gephardt Endowed Fellowship (O. Ayrault), American Brain Tumor Association Fellowship (H. Zhao), and American Lebanese Syrian Associated Charities of St. Jude Children's Research Hospital. C.J. Sherr is an Investigator of the Howard Hughes Medical Institute.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.