Polycomb Ezh2 controls the fate of GABAergic neurons in the embryonic cerebellum

During neural development, the proliferation and differentiation of progenitor cells is regulated to ensure that neuronal cell types are generated in a tightly controlled temporal sequence. This is guaranteed by combinatorial expression of transcription factors imparting cell type- and developmental-specific patterns. Increasing evidence indicates that histone post-translational modifications (PTMs) mediate the effects of transcription factors in dictating proper neural cell fate and neurodevelopment (Laugesen and Helin, 2014).

Tri-methylation of lysine 27 of histone H3 (H3K27me3) is associated with transcriptional repression in embryonic stem cells (Boyer et al., 2006; O'Carroll et al., 2001; Young, 2011) and tissues (Ezhkova et al., 2009; He et al., 2012; Juan et al., 2011; Snitow et al., 2015). This modification is mediated by the polycomb repressive complex 2 (PRC2) via its catalytic subunit enhancer of zeste 2 (Ezh2) (Cao et al., 2002; Kirmizis et al., 2004; Kuzmichev et al., 2002). The role(s) of Ezh2/PRC2 during neurodevelopment have been explored in a few contexts (Hirabayashi et al., 2009; Pereira et al., 2010) but further studies are needed for a detailed understanding of its function.

The embryonic cerebellum offers an ideal system in which to study neural fate commitment and differentiation because of its well-defined cytoarchitecture composed of few cell types. In the developing cerebellum, two distinct germinal zones give rise to all the neuronal subtypes. The ventricular zone generates GABAergic neurons, including inhibitory Purkinje cells and cerebellar interneurons (Golgi, basket and stellate cells). The anterior rhombic lip produces glutamatergic neurons such as granule cells, unipolar brush cells and excitatory deep cerebellar nuclei (Marzban et al., 2014; Roussel and Hatten, 2011; Zervas et al., 2005). Lineage tracing and birthdating studies have shown that cerebellar neurons are generated in a controlled spatial and temporal manner. A cadre of signaling molecules and transcription factors are involved in specifying the cerebellar primordium, neural lineage commitment, neuronal differentiation and cerebellar morphogenesis. Fibroblast growth factor 8 (Fgf8) and transcription factors Otx2, Gbx2, En1 and En2 are involved in the initial specification of the cerebellar primordium and cerebellar morphogenesis (Bilovocky et al., 2003; Cheng et al., 2010; Chi et al., 2003; Joyner et al., 1991; Li and Joyner, 2001; Meyers et al., 1998; Millen et al., 1994; Wurst et al., 1994), whereas transcription factors Ptf1a and Atoh1 are required for the commitment of neural lineages of the ventricular zone and rhombic lip, respectively (Wang et al., 2005; Hoshino et al., 2005; Machold and Fishell, 2005; Pascual et al., 2007). Transcription factor Pax2 is involved in both cerebellar primordium specification and interneuron commitment, albeit at different developmental stages (Bouchard et al., 2000; Maricich and Herrup, 1999; Weisheit et al., 2006). Transcription factor orphan nuclear receptor Rorα, and c-Ski family member and transcription regulator Corl2 (also known as Skor2), are both expressed in the Purkinje cells and are necessary for their differentiation and maturation (Gold et al., 2003; Hamilton et al., 1996; Minaki et al., 2008; Nakagawa et al., 1997; Nakatani et al., 2014). Mutations of these signaling molecules and transcription factors often lead to congenital cerebellar conditions including cerebellar agenesis and hypoplasia that is associated with Dandy–Walker syndrome (Gold et al., 2003; Hamilton et al., 1996; Millen et al., 2014; Nakagawa et al., 1997; Parisi and Dobyns, 2003). Therefore, it is pivotal to understand how these signaling molecules and transcription factors are regulated during cerebellar development.

In this study, we explored the function of Ezh2/PRC2 in the murine embryonic cerebellum by genetically inactivating Ezh2 in the cerebellar primordium and found that the mutant mice exhibited cerebellar hypoplasia. Using a combination of genome-wide approaches, we unveiled a network of gene expression and chromatin modifications underlying the phenotypic changes occurring in the absence of Ezh2. These findings indicate that Ezh2/PRC2 serves as a ‘gatekeeper’ to govern proper cerebellar neurogenesis in both cerebellar germinal zones.

To address the role of PRC2 in the developing cerebellum, we selectively excised the Ezh2 gene by crossing Ezh2^fl/fl mice (Su et al., 2003) with Pax7-Cre mice (Keller et al., 2004) to generate the conditional null genotype Pax7-Cre; Ezh2^fl/fl. For convenience, we named these conditional knockout (cKO) mice Ezh2^cKO. The Ezh2^cKO mice display skeletal muscle defects resulting from Pax7 expression and Ezh2 deletion in adult muscle stem cells (satellite cells) (Juan et al., 2011; Woodhouse et al., 2013). In addition to satellite cells (Jostes et al., 1990; Seale et al., 2000), Pax7 expression is also observed in the dorsal neural tube (Jostes et al., 1990). Cells with floxed alleles created by Pax7-Cre expression are inherited in dorsal neural tissues including the whole cerebellar primordium (Keller et al., 2004). We started investigating Ezh2 and Pax7 expression by immunostaining the cerebella of embryonic day (E)10.5 embryos (Fig. 1). At this stage, Ezh2 was detected in almost every nucleus of the whole embryonic brain. Consistent with previous reports (Keller et al., 2004), Pax7 expression was restricted to the dorsal region (Fig. 1). At both E10.5 (Fig. 1) and E11.5 (data not show), Ezh2 expression was comparable between Ezh2^cKO and littermate controls in the cerebellum. Transient retention of Ezh2 protein in the nuclei of cells expressing Cre-recombinase was reported when Ezh2 was genetically deleted in the cerebral cortex (Pereira et al., 2010). Reduced Ezh2 expression in the cerebellar primordium of Ezh2^cKO mice was noted starting from E12.5 (Fig. 1). In agreement with decreased Ezh2 accumulation, H3K27me3 was reduced at E12.5 and by E14.5 very few cells remained H3K27me3-positive in the Ezh2^cKO cerebellum (Fig. 2A,B). Consistent with the selective expression of Pax7 in the dorsal structures (Fig. 1), H3K27me3 was not apparently affected in the ventral part of the Ezh2^cKO embryonic brain (Fig. 2B, arrows).

Because reduction of Ezh2 and H3K27me3 in the cerebellar primordia began at around E12.5, we investigated possible developmental defects by histology analysis comparing the morphology between Ezh2^cKO mice and littermate controls from E12.5 onward. At E15.5, the cerebellar anlage of Ezh2^cKO embryos was discernibly smaller, missing the characteristic budding external granular layer (EGL) emerging from the rhombic lip (Fig. 2C). This phenotype became more evident during development and by E17.5, Ezh2^cKO embryos had a much smaller cerebellum lacking a distinct EGL (Fig. 2C). Gross cerebellar defects were evident in postnatal day (P)8 pups with the cerebellum of Ezh2^cKO mice lacking majority of the vermis (Fig. 2D, upper panel). A para-sagittal section through the hemisphere region of the cerebellum showed that the foliation pattern in Ezh2^cKO mice was much less intricate than that of littermate controls (Fig. 2D, lower panel). These defects persisted throughout adulthood. Nissl staining documented impoverished foliation (Fig. S1A) and magnetic resonance imaging (MRI) confirmed absence of the cerebellar vermis in adult Ezh2^cKO mice (Fig. S1B).

The phenotypic defects of the Ezh2^cKO embryos prompted us to investigate the functional consequences of Ezh2 deletion on the cerebellar transcriptome and epigenome. Therefore, RNA-seq and H3K27me3 ChIP-seq were conducted on E13.5 cerebella from Ezh2^cKO mice and littermate controls. To precisely identify and isolate Pax7-Cre-expressing regions, we generated a reporter mouse by breeding the Rosa26-YFP locus (Srinivas et al., 2001) into Ezh2^cKO mice to obtain Ezh2^cKO:YFP mice. YFP-positive cerebella were identified and isolated under a fluorescence dissecting microscope (Fig. S2A,B).

Analysis of the H3K27me3 ChIP-seq datasets revealed that the H3K27me3 mark was enriched at ∼13,000 genomic regions. More specifically, H3K27me3 peaks were present at ∼5700 intergenic, ∼2200 intragenic (including gene body and introns), as well as ∼4000 promoter regions (±2000 bp from the transcription start site, TSS) (Fig. 3A). In Ezh2^cKO cerebellar genomes, H3K27me3 was largely reduced compared with those of control embryos (Fig. 3A; Table S1). Gene ontology analysis of the genes associated with H3K27me3⁺ TSS returned terms related to ‘developmental protein’, ‘homeobox’, ‘DNA binding’ and ‘transcription regulation’ (Fig. 3B). Among H3K27me3⁺ TSS, there were those corresponding to genes for transcription factors involved in either conferring cell pluripotency (Nanog) or regulating differentiation of non-neuronal cell lineages, such as Myod1 (skeletal muscle), Hand1 and Hand2 (neural crest and the second heart field), Nkx2 and T-box genes (mesoderm), and over 150 homeobox-containing genes including HoxA, -B, -C and -D clusters (Table S1). In addition, promoters of genes encoding transcription factors and signaling regulators crucial for cerebellar development were H3K27me3⁺. These included Ptf1a, Atoh1, En1, Otx2, Gbx1/2, Gli1/3, Lhx1/2/5, Pax2/5/8, Hoxa1/2, Zic1/2 and 32 members of the Wnt and Fgf families (Fig. 3C; Table S1). Overall, these findings suggest a potential role of H3K27me3 in regulating a large number of non-neuronal as well as neuronal developmental transcription factors.

Analysis of RNA-seq datasets revealed that 481 transcripts were upregulated and 98 transcripts were downregulated in E13.5 cerebella of Ezh2^cKO embryos compared with littermate controls (Fig. 4A; Table S2). Of the 481 upregulated genes, ∼70% (339/481) were occupied by H3K27me3 (Fig. 4B; Table S3), indicating that these genes might be directly repressed by H3K27me3 in normal development. Most of the genes whose promoters were marked by H3K27me3 and that were related to non-neuronal cell lineage specification remained repressed in spite of decreased H3K27me3 in Ezh2^cKO embryos (Myod1, Mef2d, Gata2-6, Hand1). This indicates that at this developmental stage, cerebellar cells have already committed to the neural fate, and that reducing H3K27me3 is not sufficient to derepress non-neural lineage regulators. Gene ontology analysis of the 481 upregulated transcripts in Ezh2^cKO cerebella returned terms such as ‘developmental protein’, ‘homeobox’, ‘transcription regulation’ and ‘DNA binding’ (Fig. 4C; Table S4). Inspection of the gene ontology terms enriched in Ezh2^cKO embryos permitted the identification of 87 transcripts encoding transcription factors (Fig. S3). This class of transcripts corresponds to 18% (87/481) of the total genes upregulated in Ezh2^cKO embryos, suggesting a relevant role of Ezh2 in top-down control of developmental regulators.

Among these transcripts, transcription factors involved in ventricular zone neurogenesis, particularly in Purkinje cell and cerebellar interneuron specification and commitment, showed significant expression changes in the Ezh2^cKO cerebella (Fig. S3). Cerebellar interneuron marker genes (Pax2, Pax5, Pax8) were upregulated by ∼twofold, whereas genes known to be involved in Purkinje cell commitment (Rora, Olig2, Olig1) were downregulated by ∼twofold (Fig. S3; Table 1), suggesting an irregular fate transition between these two neuronal subtypes in Ezh2^cKO cerebella.

Out of 98 downregulated transcripts in Ezh2^cKO cerebella, 50 (54%) were identified as components of the gene ontology term ‘membrane’ (P-value: 9.7×10⁻⁸) (Table S5). Analysis of these transcripts revealed that they are involved in regulating Purkinje cell adhesion, migration and dendrite projection (Table S6). Furthermore, other known Purkinje cell markers (Skor2, Foxp2, Pcp4) were also downregulated (Table 1). These findings suggested a potential loss of Purkinje cells in Ezh2^cKO cerebella. To evaluate this possible phenotype, we investigated the number and distribution of cerebellar interneurons and Purkinje cells in both Ezh2^cKO cerebella and in littermate controls. E13.5 and E14.5 cerebella sagittal sections were immunostained with antibodies against Pax2 and Rorα to identify interneurons and Purkinje cells, respectively (Fig. 5A-D). Pax2⁺ cells were increased in Ezh2^cKO cerebella, adopting a deeper position similar to that normally observed at later stages of development (Englund et al., 2006) (Fig. 5A,C). By contrast, while being correctly positioned in the deep part of cerebellar primordium of both genotypes, Rorα⁺ Purkinje cells were decreased in the Ezh2^cKO cerebella (Fig. 5B,D). In addition, to quantify the Pax2⁺ and Rorα⁺ cells of the whole cerebellar primordium, we FACS-sorted YFP⁺ cells from Ezh2^cKO:YFP cerebella and littermate controls at E13.5 (Fig. S2B,C). YFP⁺ cells were plated and fixed immediately, then immunostained with antibodies against Pax2 and Rorα (Fig. S4A,B). After normalizing for total DAPI⁺ nuclei, cells from Ezh2^cKO:YFP cerebella were found to contain a higher percentage of Pax2⁺ cells and lower percentage of Rorα⁺ cells (Fig. S4A,B, graphs). Taken together, these findings showed that loss of Ezh2 lead to the loss of Purkinje cells and gain of cerebellar interneurons at early embryonic stages.

Ezh2 regulates cell proliferation through repression of the Ink4B-Arf-Ink4A tumor suppressor locus (also known as the Cdkn2b-Cdkn2a locus) in cultured cells and tissues (Bracken et al., 2007; Ezhkova et al., 2009; Juan et al., 2011). Because of the EGL reduction and cerebellar hypoplasia (Fig. 2C,D), we queried expression and H3K27me3 occupancy of the Ink4B-Arf-Ink4A locus. Expression of the individual components of the Ink4B-Arf-Ink4A locus, Cdkn2a, Cdkn2b and Cdkn2c, was increased and their H3K27me3 occupancy was reduced in Ezh2^cKO cerebella (Fig. 6A). As determined by quantitative PCR (qPCR), Cdkn2a expression remained elevated in E13.5-E15.5 Ezh2^cKO cerebella (Fig. 6B). Cdkn2a levels were increased in regions corresponding to the EGL and deep cerebellar nuclei of E17.5 Ezh2^cKO mice (Fig. S5). Consistent with transcriptional derepression, H3K27me3 was decreased at the Cdkn2a promoter regions of Ezh2^cKO cerebella at E13.5 and E14.5 (Fig. 6B). We confirmed the EGL phenotype (Fig. 2C,D) by immunostaining for the granule precursor cell (GPC) marker Pax6 (Fig. 6C) (Engelkamp et al., 1999). At E14.5, Ezh2^cKO cerebella had comparable number of Pax6⁺ GPCs as control cerebella. However, at E15.5, the EGL in Ezh2^cKO was dramatic reduced in thickness. The number of Pax6⁺ GPCs was largely reduced in Ezh2^cKO cerebella (Fig. 6C), indicative of either cell death or lack of cell proliferation. As Cdkn2a acts as a potent cell-cycle inhibitor of the G1-S transition (Nobori et al., 1994), we evaluated the proliferative state of EGL by EdU labeling (Zeng et al., 2010). We administrated EdU in pregnant mice at E13.5 (Fig. 7A) and E14.5 (Fig. 7B), respectively, and collected embryos after 24 h. Co-staining the embryonic brain sections with EdU and anti-Ki67 (mitotic marker) antibody revealed the proliferation state of EGL. We found that ∼55% EdU⁺ cells had already exited the cell cycle in E14.5 Ezh2^cKO EGL, whereas ∼85% EdU⁺ cells still maintained proliferative state in controls (Fig. 7A). At E15.5, ∼80% EdU⁺ cells exited the cell cycle in Ezh2^cKO EGL, while ∼94% EdU+ cells still maintained proliferative state in controls (Fig. 7B). Apoptosis, evaluated by TUNEL assay, was not increased in Ezh2^cKO cells (Fig. S6). Thus, reduced proliferation, rather than cell death, appears to be responsible for decreased representation of granule precursor cells in Ezh2^cKO cerebella.

In this study, we found that Ezh2 regulates embryonic cerebellum development by affecting the emergence of distinct neuronal cell types. Ezh2 ablation resulted in increased Pax2⁺ interneurons, reduced representation of Rorα⁺ Purkinje cells and granule precursor cells, and ultimately in a hypoplastic cerebellum. Bmi1, a member of the polycomb repressive complex 1 (PRC1) has been reported to play an important role in the expansion of granule precursor cells (Leung et al., 2004). However, PRC1 does not seem to control Purkinje cell formation and differentiation (Leung et al., 2004), suggesting that PRC1 and PRC2 exert distinct functions during cerebellar neurogenesis.

We employed a genome-wide approach to survey gene expression changes resulting from the knockout of Ezh2. Analysis of genes involved in early embryonic cerebellum development (E8.5 to ∼E13.5) revealed developmental dependency on Ezh2/PRC2 (Table 1). Even though occupied by H3K27me3, expression of genes controlling cerebellar specification and patterning developmental events occurring at earlier stages (E8.5-9.5), was not affected in Ezh2^cKO embryos (Table 1). This is likely a result of the fact that H3K27me3 accumulation was affected starting from E12.5 in Ezh2^cKO cerebella (Fig. 1A), at which stage the primordium has already been committed to cerebellar fate (Zervas et al., 2005). By contrast, genes involved in interneuron and Purkinje cell commitment and differentiation were upregulated and downregulated, respectively (Table 1, Fig. 5). These events occurred at E13.5, a critical developmental stage when ventricular zone precursor cells undergo a Purkinje cell-to-interneuron fate transition (Martinez et al., 2013). Taken together, these observations indicate that Ezh2/PRC2 regulates cerebellum development in a cell stage-dependent manner.

Purkinje cells and cerebellar interneurons derive from Ptf1a⁺ neural precursors in a temporally controlled manner through mechanisms that remain to be fully understood. Among Ptf1a⁺ ventricular zone precursors, Olig1/2⁺ and Gsx1⁺ precursor cells are commited to develop into Rorα⁺ Purkinje cells and Pax2⁺ interneurons, respectively. Olig1/2 and Gsx1 mutually represses each other during this temporal transition (Seto et al., 2014). Our study suggests a potential mechanism regulated by Ezh2 in controlling generation of Purkinje cells and cerebellar interneurons. Reduced H3K27me3 following Ezh2 knockout resulted in increased expression of Pax2 and generation of Pax2⁺ interneurons in Ezh2^cKO cerebella. Increased Pax2 expression might accelerate the turnover process from Gsx1⁺ precursor cells to Pax2⁺ interneurons, which in turn would counteract the generation of Olig1/2⁺ precursors and Rorα⁺ Purkinje cells.

An additional phenotype observed in Ezh2^cKO cerebella was the decreased proliferation of granule precursor cells of the EGL. Granule precursor cells contribute to the proper cerebellar growth and shape up the foliation pattern throughout embryonic and early postnatal stages (Martinez et al., 2013; Zervas et al., 2005). Purkinje cells are essential for granule cell proliferation through secretion of the mitogen Shh (Corrales et al., 2004; Lewis et al., 2004) and loss of Purkinje cells leads to cerebellar hypoplasia, as seen in Rorα and Skor2 mutant mice (Gold et al., 2003; Hamilton et al., 1996; Nakagawa et al., 1997; Nakatani et al., 2014). A similar phenotype occurred in Ezh2^cKO mice (Fig. 2D; Fig. S1). Combined with proliferation defects of granule cells associated with overexpression of cdkn2a, -b and -c at early embryonic stages (Fig. 6), loss of Purkinje cells in Ezh2^cKO mice would ultimately lead to a hypoplastic cerebellum. Elevated Ezh2 has been reported to be involved in the pathogenesis of ataxia-telengiectasia (A-T) caused by ATM protein deficiency. A-T postmitotic neurons have increased PRC2 and undergo cell death, indicating that controlled Ezh2 expression is crucial for neuronal function and survival (Li et al., 2013). Thus, in addition to its role in orchestrating topographic migration and connectivity of precerebellar neurons (Di Meglio et al., 2013) and neuron survival (Li et al., 2013), this study indicates that Ezh2 regulates gene expression patterns underlying the appropriate temporal neurogenesis required for proliferation and differentiation of neural precursor cells.

Mice were housed in a pathogen-free facility and all experiments were performed according to the National Institutes of Health (NIH) Animal Care and Use regulations. Pax7^+/Cre:Ezh2^fl/fl mice (Ezh2 conditional knockout, Ezh2^cKO) were generated by breeding Pax7-Cre mice (Pax7^+/Cre) (Keller et al., 2004) with Ezh2 flox allele mice (Ezh2^fl/fl) (Su et al., 2003). We crossed Rosa26-YFP (Rosa26-YFP^Y/Y) mice (Srinivas et al., 2001) to Ezh2^fl/fl for three generations to homozygosity (Ezh2^fl/fl:Rosa26-YFP^Y/Y) . Ezh2^fl/fl:Rosa26-YFP^Y/Y females were mated with heterozygous males (Pax7^+/Cre:Ezh2^fl/+) to obtain Pax7^+/Cre:Ezh2^fl/fl:Rosa26^YFP/+ (Ezh2^cKO:YFP) mice. Timed matings were carried out with heterozygous Pax7^+/Cre:Ezh2^fl/+ males crossed with homozygous females (either Ezh2^fl/fl or Ezh2^fl/fl:Rosa26^YFP/YFP). The day a vaginal plug was observed in females was counted as embryonic day 0.5. Genotypes were identified by allelic PCR on embryonic tissues or tail genomic DNA. Genotyping protocol and primers were described by Juan et al. (2011).

E10.5 to E17.5 embryonic brains were dissected under either white-field or fluorescence dissecting microscopes (Olympus SZX9, Leica Fluro). The genotypes were identified by allelic PCR on genomic DNA of embryonic tissues (tails and limbs). The embryonic brains were further fixed in 4% PFA and processed for paraffin- (E10.5-12.5) and/or cryo- (E12.5-17.5) embedding for histology and immunohistochemistry. The P8 brains were dissected out from pups and fixed in 4% PFA and processed for paraffin embedding. The E13.5 cerebellar primordia for RNA-seq and ChIP-seq experiments were precisely dissected under a fluorescence microscope based on the area with YFP expression. The meninges were removed and discarded. Images of histology and immunohistochemistry were taken under the Leica DMR or DM6000 microscope, respectively.

Sagittal sections were prepared on either a Microtome (Microm HM325) or Cryostat (Leica CM1860C) at 10-12 µm each. Sections on the approximately same axial level were subjected to H&E staining or immunofluorescence staining with antibodies described in Table S7. Adjacent sections were subjected to terminal deoxynucleotidyl transferase dUTP nick end-labeling (TUNEL) assay using a detection system (Roche-Applied Science) according to the manufacturer's instructions.

Pregnant mice (E13.5, E14.5) were weighed and administrated with EdU (30 mg/kg) by intraperitoneal injection. Embryos were harvested 24 h after injections. Embryonic brains were dissected and fixed (4% PFA), then cryo-embedded in OCT. The embryonic brains were sectioned at 10 µm each. Sections were subjected to anti-Ki67 antibody staining, then EdU staining by using Click-iT Edu imaging kit (Life Technologies Inc.). Nuclei were counted and imaged under the Leica DMR and Carl-Zeiss LSM710.

E13.5 YFP⁺ cerebellar primordia were dissociated by a modified protocol based on Wagner et al. (2011) and sorted by gating on the GFP channel of a BD FACSAria IIIu machine. The sorted YFP⁺ cells were plated in a 24-well coated BD culture plate at a density of ∼10,000 cells/well. The plated cells were slightly fixed by 4% PFA for 5 min and immunostained with antibodies against Pax2 and Rorα, and counter-stained with DAPI. Pax2⁺ and Rorα⁺ cells in ∼30-40 microscopic fields (∼1000 total cells) were counted and the ratios of Pax2⁺/DAPI⁺ and Rorα⁺/DAPI⁺ were determined for each sample.

Embryonic cerebellar primordia were precisely dissected and immediately immersed in 1 ml of RNAlater (Ambion/Life Technologies Inc.). Total RNA was extracted by using RNeasy Mini kit (Life Technologies Inc.) that yielded about 4-8 µg/sample. RNA aliquots of 500 ng were employed for reverse transcription (High-Capacity cDNA Reverse Transcription kit, Life Technologies Inc.) and reverse transcription quantitative PCR (RT-qPCR; SYBR Green PCR Master Mix, Life Technologies Inc.). RNA-seq libraries were generated with 2-4 µg RNA/sample using the TruSeq Kit (Illumina Inc.). The libraries were subsequently amplified by PCR and run in HiSeq™2000 (Illumina Inc.). RNA-seq was repeated with cerebella of four littermate controls and four Ezh2^cKO embryos (n=4, 4). Primers for RT-qPCR on cdkn2a transcipts were described in Juan et al. (2011).

Freshly dissected cerebellar primordia were cross-linked by 1% formaldehyde at room temperature for 15 min, washed twice with 0.1 M glycine-HCl, and immediately used for sonication or stored at −80°C. The cross-linked tissue was immersed in ChIP RIPA buffer (1× PBS, 1% NP40, 0.5% sodium deoxycholate, 0.1% SDS, Roche Protease Inhibitor Cocktail) in a 1.5 ml microcentrifuge tube on ice and sonicated by a Vibro-Cell Sonicator at 75% for 10×45 sec with 30 sec intervals for each cycle. Each cerebellar primordium yielded around 400-500 µg chromatin. ChIP-qPCR was performed with 40-50 µg chromatin and the remaining chromatin was processed for ChIP-seq with anti-H3K27me3 antibody. H3K27me3 ChIP-seq was repeated with cerebella of two littermate controls and Ezh2^cKO mice. The protocol for H3K27me3 ChIP-qPCR on the Cdkn2a promoter is described in Juan et al. (2011) and for ChIP-seq in Mousavi et al. (2012).

For RNA-seq, raw sequences of 50 bp single-end reads were mapped to mouse genome (mm9 assembly) using TopHat (Trapnell et al., 2009) and gene transcript levels were determined via Cuffdiff in the form of FPKM (RPKM) values by correcting for multi reads and using geometric normalization (Trapnell et al., 2013). ChIP-seq data were mapped to the mouse genome (mm9 assembly) using the Bowtie algorithm (Langmead et al., 2009). Sequencing data generated from genomic DNA (input DNA) were used against the ChIP sample in calling enriched regions and to control for false discovery rate (FDR). Enriched regions for H3K27me3 were detected using the SICER algorithm at the FDR level of 5% with a window and gap size set to 200 bp and 600 bp, respectively (Zang et al., 2009). Downstream analyses to generate intensity profile around the transcriptional start site (TSS), and HeatMap for differentialy expressed genes and transcription factors were completed using custom written codes in MATLAB and R, respectively. Gene ontology analyses were done using the online bioinformatics resource DAVID (National Institute of Allergy and Infectious Diseases, NIH) (Huang et al., 2009a,b).

Adult mouse brains were imaged by the In Vivo NMR center at NIH MRI core facility on a 14 Tesla (Bruker Biospin Inc.) scanner. Images of frontal and dorsal optical sections were taken. Adult mouse brains were processed for Nissl staining according to Petralia and Wenthold (1999).

We thank Mario Capecchi and Alexander Tarakhovsky for sharing the Pax7-Cre and Ezh2^fl/fl animals, respectively. Jim Simone and Jeffrey Lay helped with cell sorting, and Gustavo Gutierrez-Cruz assisted with sequencing. Members of the NIAMS Laboratory Animal Care and Use Section and NIAMS Light Imaging Section are kindly acknowledged. We thank Ya-xian Wang, Ron Petralia (NIDCD), Joe Laakso and John Hammer (NHLBI), Assia Derfoul and Stefania Dell'Orso (NIAMS) for technical help and advice. The Pax7 hybridoma (Kawakami et al., 1997) were obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by the Department of Biological Sciences, The University of Iowa, Iowa City, IA 52242, USA.