GSK-3 modulates SHH-driven proliferation in postnatal cerebellar neurogenesis and medulloblastoma

Strict control of progenitor proliferation and differentiation are required for brain growth. Inadequate proliferation causes microcephaly whereas excessive proliferation may support tumorigenesis (Chizhikov and Millen, 2003; Grimmer and Weiss, 2006; Yang et al., 2009; Garel et al., 2011; Lang and Gershon, 2018). Cerebellar growth depends on the proliferation of cerebellar granule neuron progenitors (CGNPs), which continues into postnatal life (Ten Donkelaar and Lammens, 2009; Marzban et al., 2015). CGNPs proliferate rapidly in response to sonic hedgehog (SHH) signaling to give rise to the largest population of neurons in the brain (Wechsler-Reya and Scott, 1999). Disruptions in regulation of CGNP proliferation can give rise to severe neurological disorders, including cerebellar hypoplasia and medulloblastoma (Ten Donkelaar and Lammens, 2009; Roussel and Hatten, 2011). Although medulloblastomas are typically responsive to a combination of surgery, radiation and chemotherapy, this treatment causes long-term adverse effects and fails 10-20% of patients (Polkinghorn and Tarbell, 2007; Northcott et al., 2012). Understanding the mechanisms that regulate SHH-driven proliferation may lead to new insight into the pathogenesis of both cerebellar hypoplasia and tumorigenesis, and may lead to novel therapeutic approaches to medulloblastoma.

We examined the regulation of CGNPs and SHH-driven medulloblastoma by glycogen synthase kinase (GSK-3), which has been shown to downregulate proliferation in multiple cellular contexts. GSK-3 is a serine-threonine kinase that acts on the intracellular transducers of diverse signaling pathways, including WNT and SHH (Cole, 2012; Bengoa-Vergniory and Kypta, 2015). The mammalian genome includes two GSK-3 isozymes, GSK-3α and GSK-3β, which are encoded by distinct loci (Gsk3a and Gsk3b). In a study of mouse forebrain neural progenitors, co-deletion of Gsk3a and Gsk3b caused hyperproliferation and delayed neuronal differentiation (Kim et al., 2009; Morgan-Smith et al., 2014). A similar phenotype was seen with deletion of GSK3A and GSK3B in human breast epithelial cells (Zhao et al., 2014), where β-catenin (CTNNB; CTNNB1) mediated this mitogenic effect.

The growth suppressive role of GSK-3 in the embryonic forebrain and in gut and breast epithelia suggest a role for GSK-3 in downregulating proliferation. This possibility is supported by studies showing that GSK-3 destabilizes MYCN, a required effector of mitogenic SHH signaling (Knoepfler and Kenney, 2006). However, an alternative possibility is suggested by the pattern of WNT and SHH pathway activation in medulloblastoma.

Transcriptomic studies of medulloblastomas resected from patients have defined subsets of tumors with activation of either WNT or SHH. However, no subset shows upregulation of both WNT and SHH. Although both SHH (Wechsler-Reya and Scott, 1999; Kenney and Rowitch, 2002) and WNT signaling (Harada et al., 1999; Jho et al., 2002; Zechner et al., 2003; Akiyoshi et al., 2006) are known to support proliferation in diverse cell types, the absence of medulloblastomas with simultaneous activation of WNT and SHH suggests that in specific cells of origin, these two pathways may not have additive oncogenic effects. Consistent with this possibility, SHH activation and WNT activation recapitulate medulloblastomas from distinctly different pre-malignant cells that derive from different regions of the rhombic lip (Schüller et al., 2008; Gibson et al., 2010; Vladoiu et al., 2019). In other cellular contexts, moreover, the WNT and SHH pathways are antagonistic. For example, endogenous WNT activation inhibits SHH signaling in human embryonic stem cells and in the dorsal neural tube (Li et al., 2009; Ulloa and Martí, 2010), and exogenous activation of WNT through forced expression of a constitutively active allele of CTNNB has been shown to reduce both SHH-driven CGNP proliferation and medulloblastoma growth (Schuller and Rowitch, 2007; Pöschl et al., 2013, 2014). The inhibitory effect of CTNNB on SHH-driven proliferation raises the possibility that GSK-3, which targets CTNNB for degradation, may be growth permissive, rather than growth suppressive. Studies using LiCl, which inhibits GSK-3 activity, have provided evidence to support this possibility (Sato et al., 2004; Yeste-Velasco et al., 2007; Zinke et al., 2015).

To resolve definitively whether and how GSK-3 exerts physiological modulation of SHH-driven proliferation in CGNPs and medulloblastoma, we analyzed Gsk3a null mice with conditional deletion of Gsk3b in CGNPs. We found that deletion of both Gsk3 loci caused profound cerebellar hypoplasia due to impaired CGNP proliferation. Deletion of both Gsk3 loci also blocked medulloblastoma growth in mice expressing SmoM2, a constitutively active allele of Smo that typically induces SHH-driven medulloblastoma with 100% penetrance. These findings show that GSK-3 can exert a diametrically opposite effect on growth that depends on the cellular context, and define a previously unknown role for GSK-3 regulation of cerebellar development that is preserved in SHH-driven medulloblastoma.

To investigate GSK-3 function in CGNPs, we bred Gsk-3α^−/− and Math1-Cre/Gsk-3β^loxP/loxP mice. Within the cerebellum, the MATH1 (ATOH1) lineage includes the CGNPs and the neurons of the deep cerebellar nuclei (Wang et al., 2005). CGNPs in Math1-Cre mice undergo recombination at LoxP sites beginning from embryonic day (E)12.5, in a rostro-caudal progression from the anterior to posterior regions of the cerebellar cortex (Machold and Fishell, 2005; Wang et al., 2005). We did not detect phenotypic changes in either the Gsk-3α^−/− or Math1-Cre/Gsk-3β^loxP/loxP genotypes. However, crossing these genotypes to generate Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/loxP (Gsk-3^DKO) mice resulted in severe cerebellar growth failure (Fig. 1A, P21), demonstrating that GSK-3 is required for cerebellar development, and that either Gsk3a or Gsk3b can meet this requirement.

Gsk-3^DKO mice were born with expected Mendelian frequency. The pups were viable in the neonatal period but by postnatal day (P)12 exhibited severe tremors, were unable to be weaned at P20-22, and did not survive beyond this age. Gsk-3^DKO mice were born with normal cerebellar architecture, with CGNPs arrayed in an external granule layer (EGL) along the outside of the primordial cerebellum, as expected (Fig. 1A, P0). By P3, however, the EGL was markedly thinned and cerebellar foliation was markedly reduced, indicating that the CGNP population did not expand appropriately from P1 to P3 in Gsk-3^DKO mice (Fig. 1A, P3). The differentiated progeny of CGNPs, the cerebellar granule neurons (CGNs) of the internal granule layer (IGL) were also markedly reduced. In control cerebella, the IGL became progressively more densely populated from P1 to P7, but no accumulating population was readily discernible in this region in the Gsk-3^DKO mice (Fig. 1A, P2-21). The absence of an IGL demonstrated the failure of Gsk-3^DKO CGNPs to generate an appropriately large population of neurons.

The alteration of CGNP development in the Gsk-3^DKO mice disrupted the process of cerebellar foliation. However, in the P2 cerebellum where the phenotype of the Gsk-3^DKO becomes discernable, other cell types, including Bergmann glia and Purkinje neurons approximated their normal positioning along the internal margin of the EGL (Fig. S1). The radial processes of the Bergmann glia extended into the EGL of both control and Gsk-3^DKO cerebella. The relative preservation of these cell types is expected, because they are outside the Math1 lineage and do not undergo Gsk-3 deletion. However, it is possible that non-cell-autonomous effects alter these cells in ways that we did not detect.

We examined whether the severe hypoplastic phenotype seen in Gsk-3^DKO was caused by decreased proliferation or increased apoptosis. We compared apoptosis in Gsk-3^DKO and control cerebella either at P2, before the onset of hypoplasia, or at P3, using immunohistochemical detection of the apoptotic marker cleaved caspase-3 (cC3). We did not detect a significant increase in apoptosis at either time point (Fig. 1B,C). In contrast, we readily found cC3⁺ cells in cerebella of Math1-Cre/Atr^loxP/loxP (Atr^cKO) mice, which we have shown to have extensive CGNP apoptosis (Lang et al., 2016), providing a positive control for cC3 detection. To evaluate the possibility of caspase-independent cell death, we also compared terminal deoxynucleotidyl transferase dUTP nick end labeling staining at each time point and again did not detect increased cell death in Gsk-3^DKO cerebella (data not shown).

Although we did not find increased apoptosis in the Gsk-3^DKO EGL, we found a trend toward decreased expression of the proliferation marker PCNA at P2 and P3 (Fig. 1D,E). We also noted patches of increased differentiation in the Gsk-3^DKO EGL at P2, marked by expression of the differentiation marker CDKN1B (also known as p27). These changes suggested the possibility of premature CGNP cell cycle exit. Examination of RB (RB1) phosphorylation supported increased cell cycle exit, as we found markedly reduced populations of cells expressing phosphorylated RB (p-RB; Fig. 1F,G) in the Gsk-3^DKO EGL.

We directly compared cell cycle exit in Gsk-3^DKO and control CGNPs by in vivo pulse-labeling with the DNA base analogs 5-bromo-2′-deoxyuridine (BrdU) and 5-ethynyl-2′-deoxyuridine (EdU). We administered BrdU intraperitoneally (IP) at P1 or P2, then administered EdU IP 24 h later, and harvested brains 2 h after EdU administration. Brains were fixed, embedded and sectioned in the sagittal plane. We identified CGNPs by their position in the EGL and defined the cell cycle exit fraction as the proportion of CGNPs that were labeled by BrdU at P2 but not labeled by EdU at P3. Both the BrdU⁺ and EdU⁺ fractions of CGNPs were significantly decreased in the EGL of Gsk-3^DKO mice compared with controls, confirming reduced CGNP proliferation at P2 and P3 (Fig. 2A,B). Moreover, the cell cycle exit fraction was significantly increased in Gsk-3^DKO CGNPs (Fig. 2C), indicating that the CGNPs that were proliferating in these mice at P2 showed reduced self-renewal relative to CGNPs in the P2 controls.

We used the GSK-3 inhibitor CHIR-98014 (CHIR98) to determine whether a physiologically achievable reduction in GSK-3 activity, such as genetic co-deletion of Gsk3a and Gsk3b, could reduce SHH-driven CGNP proliferation. We cultured freshly explanted CGNPs for 24 h in media with or without SHH, or with SHH plus increasing doses of CHIR98. We then lysed cells and, using western blotting, quantified phosphorylated CTNNB (p-CTNNB) as a measure of GSK-3 inhibition, p-RB as a measure of proliferation, CDKN1A as a measure of cell cycle arrest, and cC3 as a measure of apoptosis (Fig. 3A). We found that CHIR98 produced dose-dependent reductions of both p-CTNNB and p-RB, reducing p-RB in SHH-treated CGNPs as effectively as SHH deprivation (Fig. 3B). GSK-3 inhibition through CHIR98 increased CDKN1A protein levels in CGNPs compared with controls (Fig. 3B). The decrease in proliferation was not accompanied by increased apoptosis, as CHIR98 did not induce a significant or dose-related increase in cC3 (Fig. 3B). In parallel cellular quantifications, we found that CHIR98 reduced the number cells showing p-RB expression, EdU incorporation and p-HH3 expression, and fewer cells were observed in S phase and M phase compared with SHH-treated controls (Fig. 3C,D). Treatment of CGNPs with the GSK-3 inhibitors LY2090314 (LY209), AZD1080 or LiCl did not decrease p-RB or p-HH3 levels or reduce EdU incorporation as effectively as CHIR98 at similar concentrations (Fig. S2). These results show that modulation of SHH-driven proliferation by GSK-3 is seen outside of the context of genetic deletion and can be achieved within the dynamic range of physiological GSK-3 activity.

To identify the transcriptomic changes set in motion by Gsk3a/b co-deletion, we subjected Gsk-3^DKO cerebella to expression microarray analysis. We compared whole cerebella of Gsk-3^DKO mice at P1, before the onset of severe hypoplasia, with two different, age-matched control groups, M-Cre/Gsk-3α^+/−/Gsk-3β^loxP/loxP, which retain a single Gsk3a allele but have no Gsk3b alleles in CGNPs, and No Cre/Gsk-3α^+/−/Gsk-3β^loxP/+, which retain a single Gsk3a allele and retain two Gsk3b alleles in CGNPs. We identified 68 genes that were differentially expressed in the Gsk-3^DKO cerebella compared with both controls, with false discovery rate (FDR) adjusted P-value (q-value) less than 0.05 and absolute fold change (|FC|) greater than or equal to 2. Fourteen genes were decreased in the Gsk-3^DKO and 54 genes were increased (Fig. 4A, Table S1). Among the 14 decreased genes, we noted Gsk3a, consistent with the genetic disruption, and three neuronal markers, Chrna3, Chrnb4, and Cadps2, consistent with the reduced neuronal population of Gsk-3^DKO cerebella.

We used gene set enrichment analysis (GSEA) to define biological pathways implicated in the Gsk-3^DKO phenotype by the differential pattern of gene expression. GSEA identified the WNT and p53 pathways as activated, with upregulation of WNT markers including Tcf7, Lef1, Wif1 and Wnt10a (enrichment score=0.5224; q-value=0.037), and upregulation of p53 markers Pmaip1, Gadd45a and Cdkn1a (enrichment score=0.7996; q-value=0.034) (Fig. 4B,C). Immunohistochemistry confirmed increased expression of CDKN1A and LEF1 proteins in Gsk-3^DKO CGNPs identified by morphology and position in the EGL (Fig. 4D,E). To confirm the identity of LEF1-expressing cells as CGNPs, we used MATH1 immunofluorescence to identify the proliferative CGNP population in LEF1 co-stained sections (Fig. 4E). These double label studies showed absence of LEF1 in control CGNPs, and expression of LEF1 in Gsk-3^DKO cerebella in a band that includes the MATH1⁺ CGNPs and adjacent, radially inward cells that we interpret to be the differentiating CGNP population. In both the Gsk-3^DKO and controls, TCF7 was localized to cells outside the CGNP population, identified by their morphology and expression of FABP7 as Bergmann glia (Fig. 4F). The cellular distributions of LEF1, CDKN1A and TCF7 show that disrupting GSK-3 in CGNPs caused both cell-autonomous and non-cell-autonomous WNT activation, and cell-autonomous, CGNP-specific induction of CDKN1A. To resolve the roles of WNT activation, p53 (TRP53) and CDKN1A in the pathogenesis of the GSK-3^DKO phenotype, we performed a series of co-deletion experiments.

GSK-3 negatively regulates WNT/β-catenin signaling by phosphorylating CTNNB, which promotes its proteasomal degradation (Hagen and Vidal-Puig, 2002). To determine whether CTNNB is required for the developmental regulation of SHH-driven CGNPs, we bred mice with conditional alleles of Ctnnb (Ctnnb^loxP/loxP) with Math1-Cre mice to generate Math1-Cre/Ctnnb^loxP/loxP (Ctnnb^cKO) animals with Ctnnb deletion in the CGNP population. At P7, Ctnnb^cKO mice showed foci of hyperproliferation in the EGL, marked by increased thickness of the PCNA⁺ and p-RB⁺ domains of the EGL (Fig. 5A). This hyperproliferation did not lead to tumor formation, and by P15 all CGNPs in Ctnnb^cKO mice had exited the cell cycle and migrated to the IGL, as in Ctnnb-intact controls (Fig. 5A, P15). These findings are consistent with a physiological role for CTNNB in CGNPs that can be compensated by additional, redundant mechanisms that limit CGNP proliferation.

To determine the role of CTNNB in the Gsk-3^DKO phenotype, we crossed Ctnnb^cKO and Gsk-3^DKO mice to generate the genotype Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/loxP/Ctnnb^loxP/loxP (Gsk-3/Ctnnb^TKO). We found that Gsk-3/Ctnnb^TKO mice were markedly less symptomatic than Gsk-3^DKO mice. Whereas all Gsk-3^DKO mice failed to survive beyond P22, no Gsk-3/Ctnnb^TKO mice showed early mortality before the study endpoint of P40 (Fig. 5B). Gsk-3/Ctnnb^TKO mice were ataxic compared with Gsk-3-intact controls, but they were markedly less ataxic, with significantly less-frequent falls, compared with littermates with Gsk3a/b co-deletion and intact Ctnnb (Fig. 5C).

Consistent with improved motor function, cerebellar growth and organization were largely rescued by co-deletion of Ctnnb. In contrast to Gsk-3^DKO mice, CGNPs in Gsk-3/Ctnnb^TKO mice at P3 and P7 formed a thick EGL (Fig. 5D). The Gsk-3/Ctnnb^TKO EGL contained an outer layer of CGNPs expressing p-RB and PCNA, and an inner layer of CDKN1B⁺ CGNPs, as in Gsk-3-intact controls (Fig. 5E). Like Gsk-3-intact controls, Gsk-3/Ctnnb^TKO CGNPs did not express CDKN1A (Fig. 5E). Quantitative studies showed that the p-RB⁺ fraction of CGNPs in Gsk-3/Ctnnb^TKO mice was similar to that of age-matched controls (Fig. 5F).

Although proliferation was rescued by Ctnnb co-deletion, CGNP migration remained profoundly abnormal. Migration defects in Gsk-3/Ctnnb^TKO cerebella became more apparent over time, as CGNPs failed to move across the Purkinje cell layer to form an IGL and instead formed ectopic collections in the molecular layer (Fig. 5D, adult). To determine whether the CGNP migration failure was caused by a lack of radial glial processes, we compared FABP7 immunofluorescence in Gsk-3/Ctnnb^TKO and control mice. In both genotypes, radial glia were identified; however, in the Gsk-3/Ctnnb^TKO the processes appeared to be less numerous (Fig. 5G). These findings suggest that radial glial, which are present in the Gsk-3/Ctnnb^TKO mice, may present an anatomical substrate for migration, but also leave open the possibility that CGNP-glial interactions may not function normally. Purkinje neurons, labeled by calbindin immunofluorescence, were positioned appropriately along the inner margin of the EGL in the Gsk-3/Ctnnb^TKO mice at P7 (Fig. 5G). However, by P21 this positioning was disrupted, suggesting a non-cell-autonomous effect of the CGNP migration failure (Fig. 5D).

CDKN1A is known to be induced by p53. To determine whether the growth restriction of Gsk-3^DKO cerebella requires signaling through p53, we tested whether co-deletion of Trp53 would restore the proliferation of Gsk-3^DKO CGNPs. We co-deleted Trp53, Gsk3a and Gsk3b by crossing mice with a conditional allele of p53 (Trp53^loxP/loxP) with Gsk-3^DKO mice to generate the genotype Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/loxP/p53^loxP/loxP (Gsk-3/p53^TKO). Gsk-3/p53^TKO mice developed severe, symptomatic cerebellar hypoplasia similar to the Gsk-3^DKO phenotype and did not survive beyond P20-22. Like Gsk-3^DKO, Gsk-3/p53^TKO mice did not demonstrate increased apoptotic rates but showed reduced PCNA and p-RB and increased CDKN1B and CDKN1A expression throughout the EGL (Fig. 6A). These studies demonstrate that p53-mediated transcription is not required for the induction of CDKN1A, nor for the premature cell cycle exit and growth failure in Gsk-3^DKO CGNPs.

To determine whether CDKN1A mediates the growth suppressive effect of WNT hyperactivation in Gsk3a/b-deleted cerebella, we bred Cdkn1a^−/− and Gsk-3^DKO mouse lines to generate Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/loxP/Cdkn1a^−/− (Gsk-3/Cdkn1a^TKO) mice. In contrast to Gsk-3/Ctnnb^TKO animals, Gsk-3/Cdkn1a^TKO mice showed severe cerebellar hypoplasia (Fig. 6B), with ataxia and early mortality similar to Gsk-3^DKO mice.

The absence of rescue in Gsk-3/Cdkn1a^TKO cerebella demonstrates that CDKN1A is not necessary for the CGNP proliferation defect in Gsk3a/b-deleted mice. Although it remains possible that CDKNA1 induction may be sufficient to stop CGNP proliferation, our data show that additional mechanisms predominantly mediate growth suppression downstream of CTNNB in Gsk-3^DKO mice. Taken together, these rescue studies show that SHH-driven CGNP proliferation requires GSK-3 to suppress CTNNB-mediated WNT signaling, whereas CGNP migration requires GSK-3 for a yet uncharacterized mechanism. Consistent with this interpretation, cortical neuron migration is controlled by GSK-3-mediated, WNT-independent processes (Morgan-Smith et al., 2014).

To determine whether Gsk3a/b co-deletion restricted SHH-driven proliferation in medulloblastoma as well as in normal cerebellar development, we examined the effect on SmoM2-induced tumorigenesis of deleting Gsk3a, Gsk3b, or both loci. Mice with the genotype Math1-Cre/SmoM2^loxP/+ (M-Smo) develop medulloblastoma with 100% frequency by P12 and die by P50 (Schüller et al., 2008; Crowther et al., 2016). We crossed Math1-Cre, Gsk-3^DKO and SmoM2 mouse lines mice to generate the genotype Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/loxP/SmoM2^loxP/+ (M-Smo/Gsk-3^DKO). Like Gsk-3^DKO mice, M-Smo/Gsk-3^DKO animals could not survive beyond weaning at P20-22. The identically short survival of Gsk-3^DKO mice and M-Smo/Gsk-3^DKO mice shows that Gsk3a/b deletion consistently limits survival to 3 weeks and precludes using survival time to study the effect of Gsk3a/b deletion on medulloblastoma growth in M-Smo mice. However, we were able to analyze tumor growth in the pre-weaning period. For this purpose, we compared cerebellar pathology between M-Smo, M-Smo/Gsk-3^DKO and littermate controls with the genotypes Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/+/SmoM2^loxP/+ (M-Smo/Gsk-3α^KOβ^HET) or Math1-Cre/Gsk-3α^−/+/Gsk-3β^loxP/loxP/SmoM2^loxP/+ (M-Smo/Gsk-3α^HETβ^KO).

Tumor growth in M-Smo/Gsk-3^DKO mice was markedly reduced, compared with M-Smo, M-Smo/Gsk-3α^KOβ^HET and M-Smo/Gsk-3α^HETβ^KO genotypes (Fig. 7A,B). PCNA staining at P12 and P15 demonstrated small foci of proliferating cells in the otherwise hypoplastic M-Smo/Gsk-3^DKO cerebella. Co-staining with CDKN1B also demonstrated pockets of differentiating cells that were absent in the GSK-3 intact tumors (Fig. 7C). Immunohistochemistry showed that GSK-3 expression persisted in these small proliferating foci, and was absent in the larger population of non-proliferating cells (Fig. 7D). The overall reduced growth in Gsk3a/b-deleted medulloblastomas, and the close consistently retained GSK-3 in p-RB⁺ tumor cells, together show that GSK-3 is required for SHH-driven proliferation in medulloblastoma.

We investigated whether apoptosis or altered cell cycle regulation played a role in suppressing medulloblastoma growth in M-Smo/Gsk-3^DKO mice, by staining M-Smo/Gsk-3^DKO tumors for cC3 and p-RB. We did not detect widespread cC3 expression that would indicate apoptosis (Fig. S3). However, we found significantly increased p-RB expression in M-Smo/Gsk-3^DKO tumors, compared with M-Smo/Gsk-3α^KOβ^HET or M-Smo/Gsk-3α^HETβ^KO controls (Fig. 7D,E). These data indicate that Gsk3a/b deletion blocked tumor growth by restricting cell cycle progression, rather than by increasing cell death. We also noted induction of CDKN1A in M-Smo/Gsk-3^DKO tumors, indicating that the GSK-3/CTNNB/CDKN1A axis that we noted in CGNPs remains intact in SHH-driven medulloblastoma.

The growth restriction imposed in tumors by Gsk3a/b deletion suggested that pharmacological GSK-3 inhibition might exert a similarly growth suppressive effect. To avoid the complexity of in vivo pharmacokinetics, we used an in vitro culture assay to examine the function of GSK-3 in promoting medulloblastoma growth. Briefly, we harvested primary M-Smo tumors from P15 mice, dissociated the tumor cells and plated them in serum-free media. We then cultured the dissociated tumors with or without GSK-3 inhibitor CHIR98. After 24 h, we collected cells for western blot or flow cytometry.

Incubation of dissociated M-Smo tumor cells with CHIR98 decreased p-RB, decreased phosphorylated CTNNB (p-CTNNB), and increased CDKN1A (Fig. 7F,G). We also noted a decrease in the inhibitory phosphorylation of GSK-3β, in an apparent homeostatic response to GSK-3 inhibition. The decrease in p-RB in M-Smo tumor cells treated CHIR98 was dose dependent, and correlated with decrease in CCND2 expression (Fig. 7H,I, Fig. S4). These data show that SHH-driven medulloblastoma cells are sensitive to changes in GSK-3 activity that are within the range of modulation that can be achieved through small molecule inhibitors.

Our data show that modulating GSK-3 activity can modulate SHH-driven proliferation in cerebellar development and in medulloblastoma. Conditional deletion of both isoforms of GSK-3 inhibited cell cycle progression and promoted cell cycle exit in CGNPs and SHH-driven medulloblastomas in mice. Pharmacological inhibition of GSK-3 produced similar growth suppression. Our molecular studies identified WNT activation and CDKN1A upregulation as consequences of Gsk3a/b co-deletion. The persistence of CDKN1A expression and cerebellar hypoplasia in Gsk-3/p53^TKO mice shows that p53 activity is not required for either Cdkn1a upregulation or for the Gsk-3^DKO phenotype. Rescue studies with co-deletion of Ctnnb show that growth suppression and abnormal migration are aspects of the Gsk-3^DKO phenotype that can be dissociated, as Ctnnb co-deletion in Gsk-3/Ctnnb^TKO mice allowed Gsk3a/b-deleted CGNPs to proliferate but not to migrate normally.

Prior studies have alternatively suggested growth-promoting effects and growth-inhibiting effects of GSK-3, in both CGNPs (Knoepfler and Kenney, 2006; Penas et al., 2015) and other cell types (Kim et al., 2009). In CGNPs, the negative regulation of MYCN by GSK-3 suggested a growth-suppressive effect (Knoepfler and Kenney, 2006), whereas the destabilization of WEE1 by GSK-3 suggested a role in promoting cell cycle progression (Penas et al., 2015). Our Gsk-3^DKO studies make clear that GSK-3 is required for proliferation, and our Gsk-3/Ctnnb^TKO studies show that the regulation of the WNT pathway by GSK-3 is sufficient to mediate this effect.

The growth suppression elicited by WNT activation in CGNPs contrasts with the growth-promoting effects of WNT activation in diverse progenitors in both the CNS (Spittaels et al., 2002; Kim et al., 2009; Morgan-Smith et al., 2014) and the gut (Harada et al., 1999; Giles et al., 2003). This growth-suppressive response to WNT signaling in CGNPs may be due to their specific reliance of mitogenic SHH signaling. Diverse interactions between SHH and WNT signaling pathways have been proposed. Inhibitory interactions have been identified in the developing neural tube and in colon cancer cells, in which WNT signaling induces GLI3, which inhibits SHH-dependent transcription (Alvarez-Medina et al., 2008; Song et al., 2015). In contrast, activating interactions have also been described epidermal cells, where WNT activation induces GLI1, which activates SHH-dependent transcription (Wang et al., 2018). Our finding that WNT signaling interferes with SHH-driven proliferation is consistent with an antagonistic interaction between these pathways.

Our findings build on previous studies in which constitutive WNT signaling, induced by either a mutant allele of Ctnnb, or conditional deletion of Apc, impaired cerebellar and medulloblastoma growth (Lorenz et al., 2011; Pöschl et al., 2013). By activating WNT signaling through a loss-of-function mutation, rather than engineering a gain of function, the deletions of Apc (Lorenz et al., 2011) and Gsk-3 demonstrate a physiological role for WNT in the normal cerebellum. The hyperproliferation that we found in CGNPs with isolated deletion of Ctnnb further support the idea that GSK-3/WNT signaling is part of the normal regulation of cerebellar development. Importantly, GSK-3 is a kinase that is susceptible to specific small molecule inhibitors. Our data suggest a potential deleterious impact of exposure to GSK-3 inhibitors during the period of cerebellar development, as well as the potential for GSK-3 inhibitors to advance the treatment of SHH-driven tumors.

It is important to note that our data support the potential of GSK-3 inhibition specifically in the SHH-driven subset of medulloblastoma. If GSK-3 activity shows the same bivalence in medulloblastoma that is seen in development, other medulloblastoma subtypes may be promoted, rather than suppressed by GSK-3 inhibition. The potential effects of GSK-3 inhibition in Group 3 and Group 4 medulloblastomas are unknown and GSK-3 inhibition in WNT subgroup tumors might be expected to increase proliferation. Within the SHH subgroup, however, WNT activation through GSK-3 inhibition may provide a new, targeted approach to therapy.

Gsk-3α^−/− and Gsk-3β^loxP/loxP mice were generously shared by Dr William Snider (University of North Carolina, UNC Neuroscience Center, Chapel Hill, NC, USA). Gsk-3α^−/− harboring an exon 2 deletion and Gsk-3β^loxP/loxP mice with exon 2 flanked by loxP sites have been previously described (MacAulay et al., 2007; Patel et al., 2008). Gsk-3α^−/− and Gsk-3β^loxP/loxP mice were bred with the Math1-Cre mouse line (Matei et al., 2005), generously shared by Dr David Rowitch (University of California San Fransisco, Pediatrics, School of Medicine, CA, USA) and Dr Robert Wechsler-Reya (Sanford Burnham Prebys Medical Discovery Institute, Tumor Initiation and Maintenance Program, San Diego, CA, USA), to produce Gsk-3^DKO mice (Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/loxP).

Conditional Ctnnb (Ctnnb^loxP/loxP) knockout mice were generously shared by Dr Eva Anton (University of North Carolina, UNC Neuroscience Center, Chapel Hill, NC, USA). These mice harbor loxP sites flanking exons 2-6 of Ctnnb as previously described (Brault et al., 2001). Ctnnb^loxP/loxP were bred with Math1-Cre mice to produce Math1-Cre/Ctnnb^loxP/loxP (Ctnnb^cKO) mice. Alternatively, Ctnnb^loxP/loxP mice were bred with Math1-Cre/Gsk-3α^+/−/Gsk-3β^loxP/loxP to generate Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/loxP/Ctnnb^loxP/loxP (Gsk-3/Ctnnb^TKO). Tp53^loxP/loxP mice were purchased from the Jackson Laboratory (Marino et al., 2000) and crossed with Math1-Cre/Gsk-3α^+/−/Gsk-3β^loxP/loxP mice to generate Math1-Cre/Gsk-3α^+/−/Gsk-3β^loxP/loxP/tp53^loxP/loxP (Gsk-3/p53^TKO) mice. Cdkn1a^−/− mice were purchased from Jackson Laboratories (Deng et al., 1995). These mice were bred with Math1-Cre/Gsk-3α^+/−/Gsk-3β^loxP/loxP to generate Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/loxP/Cdkn1a^−/− (Gsk-3/Cdkn1a^TKO) progeny. Littermates heterozygous for Gsk-3 or Cdkn1a were used as controls.

Math1-Cre/Gsk-3α^+/−/Gsk-3β^loxP/loxP were crossed with SmoM2^loxP/loxP mice purchased from Jackson Laboratories (Mao et al., 2006) to produce Math1-Cre/Gsk-3α^+/−/Gsk-3β^loxP/loxP/SmoM2^loxP/wt (M-Smo/Gsk-3α^HET/Gsk-3β^KO), Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/wt/SmoM2^loxP/wt (M-Smo/Gsk-3α^KO/Gsk-3β^HET) and Math1-Cre/Gsk-3α^−/−/Gsk-3β^loxP/loxP/SmoM2^loxP/wt (M-Smo/Gsk-3^DKO) mice. SmoM2^loxP/loxP mice were also crossed with Math1-Cre mice to produce Math1-Cre/SmoM2 (M-Smo) progeny.

All mice were genotyped by PCR using primers listed in Table S2. All animal experiments were carried out in accordance with established NIH practices and approved under University of North Carolina Institutional Animal Care and Use Committee Protocols #13-121.0, 15-306.0, 16-099.0 and 18-199. All mice for these experiments were of the species Mus musculus, maintained on the C57/BLK6 background.

CGNPs were isolated by size selection and explanted as previously described (Kenney et al., 2003; Lang et al., 2016). Briefly, cerebella were dissected from P5 pups, dissociated with papain using the Papain Dissociation System (LK003150, Worthington Biochemical Corporation) and selected on a density gradient of ovomucoid inhibitor, then allowed to adhere to coated culture wells in DMEM/F12 (11320, Life Technologies) with 25 mmol/l KCl, supplemented with heat inactivated FBS (HI-FBS) and N2. After 4 h, media was replaced with identical serum-free media. Cells were maintained in 0.5 mg/ml SHH (464SH, R&D Systems) or vehicle (0.1% bovine serum albumin in 1× PBS).

Where indicated, the specified dose of GSK-3 inhibitor (CHIR-980914, S2745; LY2090314, S7063; Selleckchem) was added to cultures after the first 4 h and cells were harvested 24 h after drug treatment. In vitro proliferation was assessed by EdU incorporation after a 1 h exposure to 20 µmol/L EdU. EdU was visualized using the Click-iT EdU Alexa Fluor Imaging Kits (C10337, Thermo Fisher Scientific) according to the manufacturer's protocol. Cell counts were performed using Olympus CellSens software. Tumor cultures from medulloblastomas freshly harvested from Math1-Cre/SmoM2 mice were prepared according to the CGNP culture protocol above, with SHH supplementation omitted.

Mouse brains were processed and immunostained as previously described (Crowther et al., 2016; Lang et al., 2016) using the antibodies listed in Table S2. EdU was visualized using the Click-iT TM EdU Alexa Fluor Imaging Kits (Thermo Fisher Scientific, C10337) according to the manufacturer's protocol. Immunostained sections were counterstained with 200 ng/ml 4′,6-diami-dino-2-phenylindole (DAPI; D1306, Thermo Fisher Scientific) in 1× PBS for 20 min. Stained slides were digitally acquired using an Aperio ScanScope XT (Aperio).

To quantify proliferating cell nuclear antigen (PCNA)-, CDKN1B/p27-, phosphorylated retinoblastoma (p-RB)-, CDKN1A-, cleaved caspase-3-, BrdU- or EdU-positive cells, the EGL region was manually annotated on each section, which was then subjected to automated cell counting using Tissue Studio (Definiens) for fluorescent slides. Quantification of cultured cells was performed by randomly imaging three sections of each well and performing cell counts with binary images using the particle analyzer in ImageJ/Fiji. P-values were determined by two-tailed Student's t-tests where indicated. Adjusted P-values for multiple comparisons were determined by ANOVA with Tukey HSD post-hoc testing where indicated.

BrdU and EdU experiments, mice were subjected to IP injection of BrdU (B23151, Thermo Fisher Scientific; 100 mg/kg in 25 µl of HBSS). After 24 h, mice were IP injected with EdU (A10044, Life Technologies; 40 mg/kg in 25 µl of HBSS) and dissected 2 h later. Brains were fixed in 4% paraformaldehyde in 1× PBS for 48 h at 4°C before being processed for histology.

RNA was purified from whole P1 cerebella according to the manufacturer's protocol (RNeasy Mini Kit, 74104, Qiagen). RNA quality and quantity were assessed by spectrophotometry, capillary gel electrophoresis and NanoDrop. Samples were prepared using the Ambion WT Expression Kit for High-Throughput Robotics (4440537) and the Affymetrix HT WT Terminal Labeling and Controls Kit (901647). We quantified transcripts using the Affymetrix Mouse Gene 2.1 ST 24-Arrays (902140, Affymetrix) and scanned with the Affymetrix GeneChip Scanner 3000 7G Plus. Data were processed using the Partek Genomics Suite 6.6 standard workflow for expression analysis. Differential gene expression analysis was performed by two-way ANOVA using Partek or with Significance Analysis of Microarrays (SAM) and Gene Set Analysis (GSA) with RStudio Version 1.1.456.

Gene sets collected from microarray analysis were subjected to GSEA using freely available software from the Broad Institute (Mootha et al., 2003; Subramaniana et al., 2005). The Molecular Signatures Database (MSigDB) were used to provide annotated gene sets for comparative analysis and P-values were estimated by 1000 permutations.

Cultured cells and whole cerebella were lysed by homogenization in RIPA buffer containing protease inhibitor cocktail, sodium fluoride (NaF) and sodium orthovanadate (Na₃VO₄). Protein concentrations were quantified using the bicinchoninic acid (BCA) method (23229, Thermo Fisher Scientific) and equal concentrations of protein were resolved on SDS-polyacrylamide gels followed by transfer onto polyvinylidene difluoride membranes. Immunological analysis was performed on the SNAP i.d. 2.0 Protein Detection System (Millipore) as per the manufacturer's protocol with the antibodies listed in Table S2. Western blots were developed using the enhanced chemiluminescent SuperSignal West Femto Maximum Sensitivity Substrate (34095, Thermo Fisher Scientific) and digitized using the C-DiGit blot scanner (LI-COR Biosciences). Quantification was performed using Image Studio Lite software (LI-COR Biosciences).

Tumors from M-Smo/Gsk-3^DKO mice and control, tumor-bearing littermates were harvested and dissociated as described for cell culture studies. Dissociated cells were re-suspended in HBSS supplemented with 6 g/l glucose and fixed with the Fix & Perm Cell Fixation & Cell Permeabilization Kit (Thermo Fisher Scientific). Cells were then stained for FxCycle Violet and p-RB using the antibodies listed in Table S2. Samples were run on the Becton Dickinson LSR Fortressa and data were analyzed with FlowJo V10.

We thank the UNC Center for Gastrointestinal Biology and Disease Histology Core for processing tissue sections and staining for H&E, (P30 DK 03987), Gabriela De la Cruz and Stephanie Cohen in the UNC Translational Pathology Laboratory (TPL) for help staining, digitizing, and quantifying sections. TPL is supported, in part, by grants from the NCI (5P30CA016086-42), NIH (U54-CA156733), NIEHS (5 P30 ES010126-17), UCRF, and NCBT (2015-IDG-1007). We also thank the UNC Flow Cytometry Core Facility for assistance with FACS. We are thankful to Jeremy Simon (UNC Neuroscience Bioinformatics Core) for his assistance with microarray analysis, who is supported by The Eunice Kennedy Shriver National Institute of Child Health and Human Development (U54HD079124) and NINDS (P30NS045892). We thank David Rowitch (UCSF, San Francisco, CA) and Robert Wechsler-Reya (Sanford-Burnham Medical Research Institute, La Jolla, CA) for the Math1-Cre mice, Dr William D. Snider (UNC Neuroscience Center) for the Gsk-3α^−/−/Gsk-3β^loxP/loxP mice, and Dr Eva Anton (UNC Neuroscience Center) for the Ctnnb^cKO mice.