To study regulation of Rest gene expression, cerebella of postnatal day 8 (P8) C57BL6 mice were harvested and cultivated in the presence of recombinant sonic hedgehog (r-SHH) for 10 days (proliferation conditions-prolif) or in the absence of r-SHH and in the presence of nerve growth factor (NGF) for 5 days (differentiation conditions-diff) (n=7, each). RNA-Seq analysis was performed followed by Principal Component Analyses (PCA) to demonstrate a clear separation of gene expression patterns in CGNPs cultured under prolif and diff conditions ( Figure 1A ). Genes that showed a significant difference in expression were subjected to pathway analysis to confirm that nervous system development and function were indeed the most altered under these conditions ( Figures 1B–D ). The expression of Rest, a major regulator of nervous system development, showed a significant decline (p=0.048) in gene expression during differentiation ( Figures 1E, F ). Interestingly, two isoforms (-201, -202), which differed in their transcription start site, could be mapped to the Rest locus under prolif conditions ( Figures 1F, G ). Of these, only Rest-201 showed a significant decline (p=0.04) in expression under differentiation conditions ( Figure 1G ). These data were confirmed by qRT-PCR analyses ( Figures 1H–J ) As seen in Figure S1A , CGNPs cultivated in the presence of r-Shh formed neurospheres, whereas their transfer to differentiation conditions caused adherence to the culture dish, development of neurite-like extensions. A decline in Rest levels was also confirmed by qRT-PCR analysis, which was associated with the expression of target genes such as Syn1, Scg10 and Tubb3 ( Figure 1K ). Western blotting also confirmed a decline in REST protein during differentiation ( Figure 1L ). Finally, immunohistochemistry (IHC) of cerebellar sections from P3, P8 and P12 mice (n=3), showed loss of REST staining in the external granule layer (EGL) by P12, which correlated with a decrease in KI-67 and a reduction in the thickness of this layer, and an increase in the neuronal differentiation marker, MAP2 ( Figures 1M-O , S1B–D ). These results indicate that Rest transcription, specifically that of Rest-201, is downregulated during neurogenesis.

Since REST is a developmentally regulated molecule, we looked for bivalent histone modifications in the upstream regulatory region of the human REST gene in the ENCODE database. In the mouse forebrain the Rest URR displayed the activating H3K4me3 and H3K27me3 marks ( Figure 2A ). In the human non-neural tumor cell line, K562, a significant peak of histone H3-lysine (K)-4 trimethyl (me3) was observed at the transcriptional start site (TSS), whereas in the neural tumor cell line, NT2-D1, a peak of histone H3K27me3 was observed ( Figure S2A ). To ask if these modifications could be detected in CGNPs, chromatin immunoprecipitation (ChIP) assays were performed, using anti-histone H3K4me1, me2 and me3, and anti-MLL4 antibodies. Rabbit IgG was used as a control and for normalization. A significant level of histone H3K4me2 (p=0.0006) and me3 (p=0.025), but not me1 were detected at the start of the coding sequence (CDS) ( Figure 2B ). The -1kb region showed a significant increase in histone H3K4me1 (p=0.02) and me2 (p=0.02), but not me3 relative to IgG ( Figure 2B ). This was correlated with a 150-fold increase in MLL4 binding relative to IgG at the -1kb site (p=0.001) ( Figure 2C ). Western blotting confirmed MLL4 expression in proliferating CGNPs, although a slower migrating form of the protein was also seen under differentiation conditions ( Figure 2D ). A significant level (p=0.01) of histone H3K27 acetylation (H3K27Ac) mark was observed at the -1kb region, consistent with active Rest gene expression ( Figure 2E ). To ask if a JQ1-sensitive BET-domain family member (BRD2, 3, 4) was involved in reading the K3K27Ac mark, we treated proliferating CGNPs with a range of concentrations of the drug (2.5, 5, 10 μM). qRT-PCR revealed a dramatic drop in Rest gene expression (p<0.0001) at all 3 doses studied. These data confirmed that a JQ1-sensitive BET-domain family confirmed that BRD2, 3 or 4 was required to read the histone H3K27Ac mark ( Figure 2F ). Interestingly, RNA-Seq analyses showed that while mRNA levels for Brd1-4 were not significantly altered between proliferation and differentiation conditions, BRD2 and BRD4 proteins were below the level of detection in proliferating CGNPs ( Figures S2B, 4D, S2C ). These data suggest that BRD2 and BRD4 proteins may be rapidly degraded, possibly to limit their activity. Similar experiments could not be performed for BRD3 due to lack of antibodies that can recognize the mouse protein.

Since REST is a developmentally regulated, potentially bivalent gene, we assessed levels of histone H3K27me3 deposited by the PRC2 complex. ChIP assays were performed using anti-histone H3K27me3, anti-EZH2 and anti-SUZ12 antibodies. Significant enrichment for histone H3K27me3 (p=0.0002) was noted at the -2kb region (10-fold), compared to IgG controls ( Figure 3A ). A small, but significant level of occupancy by the PRC2 complex protein, SUZ12, was observed at the CDS ( Figure 3B ). EZH2 was not bound to any of the sites examined in the Rest upstream regulatory region (URR) in proliferating CGNPs ( Figure 3B ). Western blot analyses confirmed the expression of EZH2 in proliferating and differentiating CGNPs, whereas SUZ12 was expressed at higher levels in differentiating cells compared to proliferating cells ( Figure 3C ). Interestingly, a distinct faster migrating SUZ12 band was seen in differentiating cells ( Figure 3C ).

Since chromatin binding of EZH2/SUZ12 was quite low and work from other groups has shown interaction between PRC2 and G9A/G9A-like protein (GLP) histone methyl transferases (27), we investigated if the latter played a role in silencing Rest gene expression in proliferating CGNPs. As shown in Figure 3D , the -1kb region showed 8-fold and 10-fold enrichment for the histone H3K9 mono (me1) (p=0.008) and dimethyl (me2) (p=0.002) marks, relative to IgG ( Figure 3D ). Histone H3K9 trimethylation (me3) was not detected at any of these sites ( Figure 3D ). Neither G9A, GLP, nor the histone H3K9 trimethyl transferases (SUV39H1 or SETDB1) showed binding at the Rest URR ( Figure 3E ). Western blotting revealed substantial expression of G9A and SUV39H1 and little to no expression of SETDB1 and GLP, respectively, in proliferating CGNPs ( Figure 3F ). In differentiating cells, SETDB1 was absent, G9A expression was dramatically reduced, and GLP and SUV39H1 were highly expressed ( Figure 3F ).

Histone lysine demethylases are important in the cross talk between activating and repressing complexes (28, 29). For example, the histone H3K27me3 demethylase (KDM6A), is known to be a part of the MLL4 complex, and demethylates H3K27me3 to facilitate H3K4me3 (28). However, KDM6A was not detected in proliferating CGNPs and at much higher levels under differentiating conditions ( Figure S2D ). Also, binding of KDM6A to the Rest URR was not seen in proliferating CGNPs ( Figure S2E ). KDM7A is a dual H3K27 and H3K9 demethylase (30, 31). KDM7A was expressed at low levels in proliferating CGNPs and higher in differentiating CGNPs ( Figure 3G ). To study its role in controlling REST expression in proliferating CGNPs, we knocked down its expression using 2 different shRNAs (#1 and #2). Surprisingly, KDM7A loss caused a (7.5-10-fold) increase in REST gene expression (p=0.004) in proliferating CGNPs ( Figure 3H ). Western blot revealed a 2.5-fold increase in REST protein levels under these conditions ( Figure S2F ). qRT-PCR confirmed that expression of Scg10 and Tubb3 expression were not significantly altered ( Figure 3I ). Collectively, the above data show that a balance between the activities of MLL4/BRD proteins and G9A-GLP/PRC2 methyltransferases/reader and the KDM7A demethylase regulate Rest expression in proliferation CGNPs.

To understand the epigenetic mechanisms underlying the decline in Rest transcription in differentiating CGNPs, ChIP assays were performed to first assess histone H3K4me3 and MLL4 occupancy at the Rest URR. Histone H3K4me1 was significantly enriched at -3kb (p=0.006), -1kb (p=0.0008) and CDS (p=0.03) ( Figure 4A ). Histone H3K4me2 and me3 were not detected at significant levels relative to IgG at the Rest URR ( Figure 4A ). A 8-fold increase in MLL4 binding to -3kb region (p=0.0005), and a small but measurable binding to the CDS (p=0.003), -1kb (p=0.02) and -2kb (p=0.06) regions was seen ( Figure 4B ). As shown in Figure 2D , MLL4 was expressed in differentiating CGNPs, although its migration pattern was different, suggesting the existence of one or more post-translational modification(s). Levels of H3K27Ac was elevated relative to IgG at -3kb region ( Figure 4C ). This correlated well with an increase in expression of BRD4 protein expression, and its significantly increased occupancy of the URR at -2kb (p=0.02) and -3kb (p=0.05), when compared to IgG controls ( Figures 4D, E ). The striking reduction in Rest expression (p<0.001) following treatment with various concentrations of JQ1 (2.5-10μM), also supports the requirement for a JQ1-sensitive BET-domain protein, potentially BRD4, in maintaining Rest transcription in differentiating CGNPs ( Figure 4F ).

In contrast to histone H3K4me3, histone H3K27me3 was significantly enriched relative to IgG at the CDS (6-fold, p=0.04), -1kb (100-fold, p=0.02), -2kb (5-fold, p=0.0001) and -3kb (300-fold, p=0.13) regions of the Rest URR ( Figure 5A ). Expression of EZH2 and SUZ12 was readily detected in differentiating CGNPs by Western blotting ( Figure 3C ). However, only EZH2 occupancy was observed in differentiating cells, and a 2-fold (p=0.045) and 7-fold (p=0.0003) enrichment, respectively, was seen at CDS and -1kb region of the Rest URR ( Figure 5B ). In contrast to histone H3K27me3, enrichment of histone H3K9 methylation was strongest at the CDS and -2kb regions, whereas these marks were present at lower levels at -1kb. Specific elevation of histone H3K9me2 (40-fold, p=0.02) at the CDS, histone H39me1 at -1kb (3-fold, p=0.006) and histone H3K9me3 (30-fold, p=0.04) at -2kb of the Rest URR, were noted in differentiating CGNPs ( Figure 5C ). Relative to control IgG, the histone H3K9 mono- and dimethyl-transferase, G9A showed significant occupancy at the CDS and -1kb (p=0.02 and 0.01, respectively). Its heterodimeric partner, GLP and the histone H3K9 trimethyl transferase SUV39H1 showed a 3-fold (p=0.04) and an approximately 25-fold (p=0.0007) enrichment at the -1kb region ( Figure 5D ). Significant binding of SETDB1 was not found at the Rest URR ( Figure 5D ).

To determine the relevance of these repressive histone modifications in silencing REST transcription, differentiating CGNPs were treated for 8 hours with either 5μM Tazemetostat (Taze) – an Ezh2 inhibitor, or 5μM UNC-0638 – a G9a inhibitor, alone, or in combination (5μM each) to account for redundant use of the two repressive complexes. As shown in Figure S3A , treatment with Taze alone resulted in a significant decrease in H3K27me3 at -3kb (p<0.0001) and CDS (p<0.0001). At the same time, an increase in histone H3K9me3 deposition at -3kb (p=0.02) and at -1kb (p=0.002), whereas a significant reduction in this modification was noted at -2kb (p<0.0001) and CDS (p<0.0001) ( Figure S3A ). UNC-0638 treatment resulted in a significant decrease in histone H3K9me3 at -2kb (p<0.0001) and CDS (p=0.002) regions of the Rest URR ( Figure S3B ). This mark was significantly increased at -3kb (p=0.03) and -1kb (p<0001) ( Figure S3B ). Histone H3K27me3 was significantly reduced at -3kb (p<0.0001), -1kb (p=0.002) and CDS (p<0.0001), but elevated at -2kb (p=0.007), following UNC-0638 treatment ( Figure S3B ). These data indicated a potential for redundancy in these two repressive marks.

To test this possibility, CGNPs cultivated under differentiation conditions and co-exposed to Taze and UNC-0638, as described above. As seen in Figure 5E , UNC-0638 and the combination resulted in a 1.5-fold increase in Rest expression (p=0.003 and 0.01, respectively). Surprisingly, treatment with Taze alone promoted a significant reduction (p=0.005) in Rest gene expression compared to untreated controls ( Figure 5E ). The unexpectedly weak de-repression of Rest gene expression raised the probability of a redundant recruitment of the repressive histone H3K9me apparatus or that of EZH1 upon loss of EZH2 activity, or alternatively the requirement for an activating mechanism following loss of these repressive mechanisms, to upregulate Rest gene expression. Additionally, the roughly equal increase in Rest gene expression in the presence of UNC-0638 alone and the combination of UNC-0638 and Taze, suggested that the activity of the G9A/GLP/SUV39H1 complex may be more important for the silencing of Rest transcription in differentiating CGNPs. To further understand these results, ChIP experiments were performed to measure changes in histone H3K9me3 and histone H3K27me3 following treatment with both drugs (2.5μM, each). Histone H3K9me3 showed a significant reduction in enrichment at CDS (p<0.0001) and -1kb (p=0.0001) as would be expected following UNC-0638 treatment ( Figure 5F ). However, histone H3K27me3 was significantly enriched at -1kb (3-fold) and -2kb (7-fold) (p<0.0001, each) in the presence of both drugs compared to the no treatment control ( Figure 5G ). These findings suggest that an alternative mechanism, such as EZH1 recruitment may be the source of the persistent H3K27me3 at the Rest URR following EZH2 inhibition. A 5-fold increase in histone H3K4me3 observed at -1kb following treatment with both drugs, also suggests that the failure to promote a stronger recruitment of the MLL4 transcriptional activating machinery may contribute to the absence of a more robust upregulation of Rest expression ( Figure 5H ).

The demethylase KDM6A, which showed strong expression in differentiating CGNPs, exhibited significant binding at -2kb (5-fold, p<0.001) and -3kb (50-fold, p=0.04) regions of the Rest URR ( Figures 5I , S2D ). Its pharmacological inhibition by the drug GSKJ4 caused a significant downregulation of Rest expression (p=0.007) in differentiating CGNPs ( Figure 5J ). Surprisingly, a decrease in histone H3K27me3 (p<0.0001) at the -2kb site ( Figure 5K ), and an increase in histone H3K9me3 at -1kb (100-fold, p<0.0001) and -3kb (250-fold, p<0.0001) sites, highlight a novel cross-talk between the histone H3K9-me apparatus and the H3K27me demethylase, KDM6A ( Figure 5L ). Additionally, knockdown of KDM7A, in differentiating CGNPs caused a 5-fold downregulation of Rest mRNA levels 5-fold (p=0.0008, p=0.0002), and surprisingly, a significant reduction in the expression of REST target genes, Scg10 (p<0.001) and Tubb3 (p<0.001). These findings suggest that KDM7A, which is known to be required for neurogenesis, may control other targets required for the activation of expression of REST target genes ( Figures 5M, N ). Thus, our data with primary mouse CGNPs indicate that Rest mRNA homeostasis is orchestrated by the MLL4/BRD4, PRC2/G9A/GLP/SUV39H1 complexes and the demethylases-KDM6A and KDM7A. Most importantly, KDM7A had a differentiation stage-specific role in the regulation of Rest expression.

To ask if this regulatory mechanism is perturbed in human SHH medulloblastoma tumors, a publicly available database was probed for expression of the above complex members. In our previous work, we showed the sub grouping of human SHH MBs into 6 clusters based on their neuronal differentiation status (32, 33). The distribution of the samples belonging to the published subtypes (α, β, γ, δ) between these clusters is also indicated in Figure 6A . It is known that patients with SHH-α and -β tumors have the worst survival (1, 3). REST expression was elevated in clusters 1, 2 and 5 (1). Here we studied the expression of the chromatin modifiers (MLL4, BRD4, EZH2, SUZ12, G9A, GLP1, SUV39H1, KDM6A and KDM7A) implicated in the regulation of the mouse Rest gene. Expression data shown in Figure 6A is relative to the average expression (Z-score) of each gene across all other groups of medulloblastoma samples (SHH group Clusters from (32) In cluster 2, where REST expression is elevated, and the tumors are more immature, a number of patients tumors consistently expressed G9A and KDM7A at levels higher or lower than the Z-score ( Figure 6A ). In cluster 5, where REST is elevated and the tumors are known to be more committed to neuronal differentiation, expression of the repressive enzymes- EZH2, SUZ12, G9A GLP, SUV39H1 was lower than the Z-score ( Figure 6A ). Expression of the demethylases KDM6A and KDM7A was higher than the Z-score ( Figure 6A ). A positive correlation between REST and MLL4 (p=0.03, r=0.40), REST and SUV39H1 (p=0.02, r=0.43) and REST and KDM7A (p=0.001, r=0.56) was observed in cluster 2 ( Figure S4A ). In cluster 5, REST and SUV39H1 were negatively correlated (p=0.046, r=-0.32), while REST and KDM7A exhibited a positive correlation (p=0.002, r=0.48) ( Figure S4A ).

Consistent with the above results, KDM7A gene expression was significantly down- and upregulated in clusters 1 and 5, respectively ( Figure 6B ). Finally, we divided SHH MBs into four groups based on the expression levels of REST and KDM7A, and found that low-REST/low-KDM7A expressions were associated with a significantly better overall survival (OS) (p=0.03) in SHH MB patients. In contrast, patients with high REST and low KDM7A or high REST and high KDM7A expression in their tumors, had the worst OS at 5 years (p=0.03), in line with seemingly opposite role of KDM7A on REST expression based on the differentiation status of the cells ( Figures 6C , S4B ). Thus, our results indicate a cross talk between activating and repressive modifiers during neuronal differentiation of CGNPs ( Figure 7 ). These data have implications for SHH MB tumors.

Cerebellar tissue from P8 pups of C57BL6 mice were dissected and triturated using 18.5 gauge needle with 10 ml syringe to form single cell suspension. CGNPs were expanded for 10 days accompanied by trituration every 4 days, in Neurobasal medium supplemented with B27, Glutamax, Antibiotic/Antimycotic, Heparin and 20ug/ml of Epidermal growth factor (EGF), Fibroblast growth factor (FGF), and then for 4 days in the presence of 0.1μg/ml recombinant mouse sonic hedgehog (rShh) (proliferation conditions) (34). Differentiation of CGNPS was achieved by culturing them in the absence of EGF and FGF and with the addition of 20μg/ml of neural growth factor (NGF) after their initial cultivation in the presence of Shh ligand as above. Differentiated cells were collected with TrypLE for further molecular analyses.

The pellets from proliferating and differentiated neural progenitors were used to isolate RNA using RNA mini kit in accordance with manufacturer’s instructions (Zymo Research, Irvine, CA). RNA was quantified using a Nano Drop 2000 (Thermo Fisher Scientific, MA). Equal concentrations of RNA were used to synthesize cDNA using iscript cDNA kit (Bio-Rad, Hercules, CA) and then measure gene expression.

Mouse brain tissues were fixed in 10% buffered formalin phosphate and embedded in paraffin. Eight-micrometer-thick brain sections were used for IHC analysis. Sections were deparaffinized with xylene, followed by rehydration with ethanol and water. Antigen retrieval was performed in citrate antigen retrieval buffer (pH 6) for 30 minutes at 95°C in the PT module (ThermoFisher). Sections were washed with 0.1% Triton X in PBS (PBST) and then treated with 3% H2O2 solution for 10 min to block the endogenous peroxidase. Nonspecific binding of antibodies was blocked with 3% BSA in PBS for 1 hour. The sections were then incubated with primary antibodies as indicated in Table 1 , diluted in blocking buffer, at 4°C overnight. The primary antibody was detected using a secondary antibody conjugated to horseradish peroxidase (HRP; the Jackson Laboratory) by incubating sections for 1 hour at room temperature. All incubations were performed under humidified conditions. Last, slides were washed with TBST and developed using the DAB kit (Vector laboratories) and counterstained with hematoxylin. After dehydration and mounting, slides were visualized under a Nikon ECLIPSE E200 microscope, and images were captured under 4x, 10x, 40x and 100x magnification with an Olympus SC100 camera. Analyses were performed using Olympus cellSens Entry software.

RNA was harvested from pooled WT CGNPs and quality checked using RNA Nano assay using Bioanalyzer 2100 (Agilent Technologies - RNA RIN should be >8). Quantification was done using Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA). Library preparation was done with 1000 ng of RNA using TruSeq Stranded Total RNA Gold kit (Illumina RS-122-2303) following TruSeq Stranded Total RNA Reference guidelines. RNA-Seq library quality/quantity was checked using Tape Station 2200 D1000 assay (Agilent Technologies, Santa Clara, CA and QuantStudio 6 Flex Real time PCR System (Applied Biosystem, Waltham, MA), and sequenced (paired end-75 cycles) using Hiseq 3000 (Illumina, San Diego, CA). The raw sequencing reads were aligned to the human GRCm39 genome (Star Aligner version 2.7.3a) and quantified using featureCounts (subread v1.6.3) based on Ensembl gene model GRCm39 vM27. Genes with count per million (CPM) greater than 0 in all sample were selected and normalized using the voom method of the R package limma v3.50.0 (35). Batch effects were removed using the R package sva v3.42.0 (36). Differential gene expression analysis was performed using the R package limma v3.50.0. Principal component analysis was performed for the samples in R v4.1.1.

Functional class annotation analysis was performed on up/down-regulated genes by using the Ingenuity Pathway Analysis software. We analyzed biological processes, molecular functions, and cellular components that were relatively enriched by the gene lists of interest.

Quantitative RT-PCR was performed in triplicate using 2x SYBR-Green master mix (Bioline, Boston, MA) and run on a LightCycler 96 Real-Time PCR System (Roche Diagnostics GmbH, Mannheim, Germany). Relative mRNA expression was calculated using mouse 18S RNA as a control. The sequence of the primers used are listed in Table 2 .

Protein lysates from CGNPs were prepared in EBC lysis buffer (50 mM Tris, pH 8.0, 120 mM NaCl, and 0.5% NP-40) in the presence of protease inhibitors (Thermo Fisher Scientific, Waltham, MA) and phosphatase inhibitor cocktail-2 (Sigma, St. Louis, MO). Cleared lysates were used to measure protein concentration in supernatants using Bio-Rad protein assay dye reagent (Bio-Rad Laboratories, Hercules, CA). Samples were run on a SDS-PAGE and Western blot analyses were performed using primary antibodies against the proteins indicated in the table ( Table 3 ). Membranes were washed and incubated with the corresponding HRP-conjugated secondary antibodies (Jackson Immuno Research, West Grove, PA), and developed using SuperSignal (Cat#34075; Cat#34087; Thermo-Scientific, Waltham, MA) followed by autoradiography.

GNPs from proliferating and differentiating cells were fixed with formaldehyde, cross-linked, and processed for chromatin immunoprecipitation (ChIP) analyses as described previously (15). Cross-linked cells were resuspended in sonication buffer (50 mmol/L Tris-HCl (pH 8.0), 10 mmol/L EDTA (pH 8.0), 1% SDS, and protease inhibitors) and sonicated, and 10% of this material was saved as input DNA. The remainder of the samples was diluted 5-fold with ChIP dilution buffer [16.7 mmol/L Tris-HCl (pH 8.0), 167 mmol/L NaCl, 1.2 mmol/L EDTA (pH 8.0), 1.1% Triton X-100, and protease inhibitors], precleared, and incubated with various antibodies ( Table 4 ) overnight at 4°C. Following incubation with protein-A beads, washing, and elution, the cross-linking was reversed, and DNA was purified with a PCR Purification Kit (Zymo). Bound DNA was quantified by SYBRGreen qPCR using the listed primers using a Roche Lightcycler 96. Data were analyzed using the comparative 2−ΔΔCt method. Primers are listed in Table 5 .

3-5 million proliferating and differentiating progenitors were used for shRNA assay. Both proliferating and differentiating progenitors were treated with either nonsense control shRNA (shRNA-NC) or two different shRNA constructs targeting Kdm7a. 500μl of the virus was added to 1.5ml of cells in media and incubated overnight at 34°C. The cells were then moved to 37°C incubator and the media was changed after 24 hours and allowed to grow. After 72 hours of shRNA introduction, cells were harvested and RNA extraction was performed using Quick RNA mini kit (Zymo research). cDNA was synthesized using Iscript reverse transcriptase kit (Biorad, USA). RT-PCR was performed to analyze knockdown of Kdm7a, REST and normalized using 18S as a control. Data were analyzed using the comparative 2−ΔΔCt method.