Research ArticleCancer

Transcriptional repressor REST drives lineage stage–specific chromatin compaction at Ptch1 and increases AKT activation in a mouse model of medulloblastoma

See allHide authors and affiliations

Science Signaling  22 Jan 2019:
Vol. 12, Issue 565, eaan8680
DOI: 10.1126/scisignal.aan8680

Targeting SHH-type medulloblastoma

Medulloblastoma is a brain tumor that occurs mostly in children and has few therapeutic options. Attempts to block the Sonic Hedgehog (SHH) pathway that drives various subtypes of the tumor have been unsuccessful clinically. Dobson et al. found a mechanism by which SHH signaling may be increased in some patients. In a new mouse model, the authors found that increased expression of the transcriptional repressor REST in cells that give rise to the disease supports a more aggressive disease course by promoting epigenetic repression of the genes that encode PTCH1 and PTEN, thereby enhancing proliferative and migratory signaling by SHH and the kinase AKT. These findings reveal potential therapeutic targets for patients with high-REST, SHH-type medulloblastoma.


In medulloblastomas (MBs), the expression and activity of RE1-silencing transcription factor (REST) is increased in tumors driven by the sonic hedgehog (SHH) pathway, specifically the SHH-α (children 3 to 16 years) and SHH-β (infants) subgroups. Neuronal maturation is greater in SHH-β than SHH-α tumors, but both correlate with poor overall patient survival. We studied the contribution of REST to MB using a transgenic mouse model (RESTTG) wherein conditional NeuroD2-controlled REST transgene expression in lineage-committed Ptch1+/− cerebellar granule neuron progenitors (CGNPs) accelerated tumorigenesis and increased penetrance and infiltrative disease. This model revealed a neuronal maturation context–specific antagonistic interplay between the transcriptional repressor REST and the activator GLI1 at Ptch1. Expression of Arrb1, which encodes β-arrestin1 (a GLI1 inhibitor), was substantially reduced in proliferating and, to a lesser extent, lineage-committed RESTTG cells compared with wild-type proliferating CGNPs. Lineage-committed RESTTG cells also had decreased GLI1 activity and increased histone H3K9 methylation at the Ptch1 locus, which correlated with premature silencing of Ptch1. These cells also had decreased expression of Pten, which encodes a negative regulator of the kinase AKT. Expression of PTCH1 and GLI1 were less, and ARRB1 was somewhat greater, in patient SHH-β than SHH-α MBs, whereas that of PTEN was similarly lower in both subtypes than in others. Inhibition of histone modifiers or AKT reduced proliferation and induced apoptosis, respectively, in cultured REST-high MB cells. Our findings linking REST to differentiation-specific chromatin remodeling, PTCH1 silencing, and AKT activation in MB tissues reveal potential subgroup-specific therapeutic targets for MB patients.


Medulloblastoma (MB), the most common malignant brain tumor of childhood, is a molecularly diverse embryonal tumor of the posterior fossa. Genomic studies over the past few years have confirmed at least four distinct subgroups [Wingless (WNT), Sonic Hedgehog (SHH), Group 3, and Group 4] (15). Retrospective studies have confirmed that long-term outcomes are subgroup specific (1). Patients with WNT-subtype MB tumors have a greater than 90% long-term survival in contrast to patients with group 3 tumors who have a 40 to 60% 5-year overall survival. Patients with SHH-driven subtype MB have an intermediate prognosis (24). Current treatment protocols include maximal surgical resection, craniospinal radiation therapy for children older than 3 years of age, and aggressive multiagent chemotherapy (1). Unfortunately, metastasis continues to be a major clinical challenge in at least three of the four subgroups (6). In addition, survivors of MB have substantial long-term toxicities because of therapy (1). Therefore, in recent years, there has been an earnest push toward understanding subgroup-specific MB biology to target the molecular events that drive tumorigenesis and metastasis (1, 3, 6, 7).

SHH MBs are the best characterized subtype of the disease. The availability of multiple genetically engineered mouse models has provided valuable insights into parallels with normal cerebellar development and has allowed identification of cerebellar granule neuron progenitors (CGNPs) as the cell of origin of these tumors (811). These studies have also implicated the developmentally important SHH signaling pathway as a driver of the early postnatal burst of proliferation of CGNPs in the cerebellar external granule layer (EGL) (12). The switch to terminal neuronal differentiation coincides with a decline in SHH pathway activity. Study of this switch in SHH pathway activity and its contribution to the transition from a proliferative to a differentiation program in CGNPs will shed light on MB genesis. SHH is a secreted ligand and binds Patched1 (PTCH1), a transmembrane receptor and tumor suppressor protein, to relieve its repressive interaction with the oncogene, Smoothened (SMO). The subsequent movement of the latter to the primary cilia allows signaling through GLI proteins to regulate the expression of cell cycle regulatory genes including cyclins (Ccnd1 and Ccnd2), the proto-oncogene N-Myc, and, through a feedback loop, Ptch1, Gli1, and Gli2 (13, 14). Current evidence supports regulation of SHH signaling at multiple steps (13). Of particular interest to this report is control of Ptch1 expression during neuronal lineage commitment in CGNPs, because data from Ptch1 heterozygous mouse models suggest that its loss of heterozygosity (LOH) is critical to tumor progression (8, 15, 16). A second point of focus of this study is regulation of pathway activity at the level of GLI proteins. Of the three GLI proteins expressed in vertebrates, GLI1 is a transcriptional activator, GLI2 toggles between an activator and a repressor, and GLI3 functions as a transcriptional repressor. Phosphorylation and ubiquitination are known to control the activity of these proteins in an Shh ligand–dependent manner (1719). In addition, GLI1 activity is regulated by its inhibitory acetylation by the p300–β-arrestin1 (ARRB1) complex (20, 21). Knockdown and overexpression studies in CGNPs suggest that Arrb1 plays a role in the shift of CGNPs from a proliferative to a differentiation state, and therefore, it is important to understand its regulation as well (21).

The RE1-silencing transcription factor (REST) is a transcriptional repressor of neuronal differentiation genes (2229). It also maintains cell proliferation and blocks cell cycle exit by preventing stabilization of the cyclin-dependent kinase inhibitor, p27, a key event in cells undergoing terminal neuronal differentiation (29, 30). REST is expressed in neural stem cells (NSCs), but its expression declines during neurogenesis (24, 3134). It is a DNA binding protein and serves as a scaffold for chromatin remodeling enzymes that repress gene expression including histone deacetylases 1 and 2 (HDAC1/2), the histone methyl transferase G9a, and the histone demethylase LSD1 (35). REST protein abundance is increased in more than 80% of human MBs (14). In a previously reported study, we observed a tendency for poor prognosis and an increased risk for developing disseminated disease in a small cohort of patients with large increases in REST expression in their tumors compared to normal cerebella (29). This observation was particularly intriguing in a subset of patients with desmoplastic MB who had an unexpectedly poor outcome when compared to the entire SHH subgroup (29).

In this study, we investigated the involvement of REST in the regulation of SHH signaling during neuronal differentiation of CGNPs and determined whether a perturbation of this process promotes MB progression. Human SHH MBs are subdivided into SHH-α, SHH-β, SHH-γ, and SHH-δ tumors, which differ in the age of the patients, propensity for metastasis, and overall survival (36). Through analysis of a publicly available dataset, we found that REST expression and activity, as measured by the expression of its target genes associated with neuronal differentiation, were markedly correlated with poor prognosis in specific SHH subgroups. This included a subset of SHH-α and a majority of the SHH-β subgroup of tumors (36). We then used a novel transgenic mouse model (RESTTG) that enabled conditional overexpression of human REST (hREST) transgene (Tg) in lineage-committed CGNPs to assess both the survival of mice bearing tumors with increased expression of REST and constitutive activation of SHH signaling and to also uncover the molecular mechanisms underlying REST-mediated tumorigenesis. Our data suggest that REST increases SHH signaling in proliferating cells but accelerates the decline in pathway activity in differentiating cells in a manner involving both genetic- and epigenetic-mediated changes in SHH pathway regulation and AKT activity. We propose that these events, at least in part, underlie the aggressive disease course and leptomeningeal dissemination of REST-expressing SHH MB tumors in mice. It may also explain the poor outcomes seen in patients with increased REST expression in their tumors, mainly those with the SHH-α and SHH-β subgroup of tumors (36). Last, our findings in cell culture models provide support for potential pharmacological targeting of G9a, HDACs, and AKT in high-REST MBs.


REST expression and activity are elevated in the SHH-α and SHH-β subgroup of tumors

REST is an important regulator of neurogenesis and negatively controls the expression of a number of genes involved in terminal neuronal differentiation (32, 3739). In MBs, which are poorly differentiated tumors, we and others have previously demonstrated an aberrant maintenance of REST expression in a small cohort of patient samples (22, 28, 29). In addition, knockdown of REST in MB cell lines blocks their tumorigenic potential in mouse orthotopic models, whereas its ectopic expression in v-Myc–immortalized NSCs promotes tumor progression (25, 28). These data suggest that REST contributes to the development of MB. To test this hypothesis, we studied the repressive activity of REST on neurogenesis in a larger cohort of MB samples in a publicly available database (GSE85217), by clustering samples based on transcriptome information on REST in combination with that of known regulators of its protein stability (BTRC, USP7, and USP15) and a subset of its target neuronal differentiation genes (SCN2A, SYP, SYN1, NEFM, NEFL, MAP2, and RBFOX3/NEUN; Fig. 1A and fig. S1) (36). The WNT group of MBs (n = 70 patients, 29 males and 35 females; age range, 2 to 56 years old; median age, 10.8 years old; 6 patients with known metastasis) were divided into three clusters (fig. S1A). Tumors in cluster 1 exhibited significantly higher REST expression compared to samples in clusters 2 and 3 (fig. S1B). However, increased REST mRNA was not associated with a significant difference in the 5- or 20-year survival of patients (fig. S1C). Group 3 MBs (n = 144 patients, 99 males and 38 females; age range, 1 to 49 years old; median age, 5.1 years old; 43 patients with known metastasis) were divided into five clusters based on their terminal differentiation (fig. S1D). Of these, clusters 1 and 4 had significantly increased expression of REST mRNA compared to the other clusters, but a correlation with 5- or 25-year overall survival of patients was not noted (fig. S1, E and F). Group 4 MBs (n = 326 patients, 216 males and 92 females; age range, 1 to 48 years old; median age, 8.0 years old; 101 patients with known metastasis) were similarly divided into six clusters (fig. S1G). Of these, samples in clusters 2 through 6 had significantly down-regulated expression of REST mRNA compared to the ones in cluster 1 (fig. S1H). Once again, REST mRNA expression did not correlate with 5- or 25-year overall survival of patients (fig. S1I).

Fig. 1 Clinical characteristics associated with gene expression profiles.

(A) Hierarchical clustering analysis of SHH MB patient samples using gene expression. Hierarchical clustering assay identified six distinct clusters based on expression profiles of neuronal differentiation markers (; dataset GSE85217). The blue to red color scale indicates the expression level (Z score). The key, clinical information [subtype, age, gender, and metastasis (mets) status] regarding patient samples, is provided beneath. NA, not available. (B) Gene expression profiles measured by microarray in six clusters. Each dot corresponds to one individual patient. Data show individual variability and means ± SD. ns, not significant. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. (C) Overall survival (OS) of six clusters in patients with SHH MB (P value; log-rank Mantel-Cox test). (D) Hierarchical clustering analysis of SHH MB patient samples using gene expression. Hierarchical clustering assay identified five distinct clusters based on expression profiles of REST target genes. The blue to red color scale indicates the expression level based on Z score. The clinical information (subtype, age, gender, and metastasis status) regarding patient samples was shown in the bottom panel. (E) Overall survival of five clusters in patients with SHH MB (P value; log-rank Mantel-Cox test).

Last, we performed a similar analysis of SHH MB samples (n = 223 patients, 128 males and 82 females; age range, <1 to 56 years old; median age, 8.8 years old; 26 patients with known metastatic disease). Samples were divided into six clusters, with a significant increase in REST mRNA amounts seen in clusters 1, 2, and 5, with cluster 1 exhibiting the highest transcript abundance (Fig. 1, A and B). Clusters 1 and 2 SHH-α tumors and cluster 5 SHH-β tumors exhibited significantly decreased expression of BTRC, the gene encoding a component of the ubiquitin-dependent proteasomal machinery that targets REST for degradation, suggesting a potential for REST protein stabilization and increased activity in these samples (Fig. 1B). The lowest expression of BTRC mRNA was seen in samples from cluster 2 (Fig. 1B). Expression of ubiquitin-specific protease–7 (USP7) mRNA, encoding a known REST-specific deubiquitylase (DUB), was significantly reduced only in cluster 2 (fig. S1K). A second DUB that controls REST protein stability is encoded by the USP15 gene (40). USP15 expression was significantly increased in cluster 5 compared to the other clusters (fig. S1K). The expression of a panel of neuronal differentiation genes (ATOH1, NEUROD2, MAP2, NEFM, SYN1, and SYP) was also investigated in the above clusters. We observed a significant reduction in ATOH1 and an up-regulation of other granule neuron markers (NEUROD2, MAP2, NEFM, SYN1, and SYP) in clusters 5 and 6 (Fig. 1B and fig. S1K). Together, these data suggest that REST mRNA expression is misregulated in samples in cluster 1, whereas a more modest increase in REST mRNA expression in conjunction with an impaired protein degradation machinery was seen in clusters 2 and 5 (Fig. 1B).

Unexpectedly, most SHH-α tumors (56 of 65) were found in clusters 1, 2, and 4, whereas cluster 5—with a notable expression of terminal neuronal differentiation genes—included 21 of a total of 35 SHH-β tumors (Fig. 1A). Most SHH-γ tumors (42 of 47) were found in clusters 4, 5, and 6, whereas SHH-δ tumors (n = 76) exhibited more heterogeneity in their differentiation phenotype and were scattered between clusters 1 and 4 (Fig. 1A). Patients with SHH-α and SHH-β tumors have poor prognosis and are at a higher risk for developing metastasis compared to those with SHH-γ and SHH-δ tumors (36). Consistent with this, patients with samples found in cluster 2 (SHH-α) and the more differentiated cluster 5 (SHH-β) had the worst 5-year overall survival (Fig. 1C). At 25 years, only a 50% overall survival rate was seen in clusters 1 to 5 (Fig. 1C). These data suggest that increased REST expression is associated with worse initial overall survival in a subset of patients with SHH-α (cluster 2) and a plurality of SHH-β (cluster 5) tumors.

To further evaluate the above association between increased REST activity and poor prognosis in SHH-type MB samples, we performed a clustering analysis of REST and a substantially expanded list of known differentiation and nondifferentiation REST target genes (n = 76) derived from a literature search (Fig. 1D and fig. S1L) (4148). Here, most SHH-β tumors (28 of 35) were found in cluster 2 and recapitulated our findings above with respect to the worst 5-year survival (Fig. 1, C and E). This cluster of tumors also exhibited increased REST expression and a significant decline in the expression of BTRC (fig. S1M). Most SHH-α tumors were found within clusters 1 and 4 (Fig. 1D). Expression of USP7 was not significantly different between the various clusters; however, USP15 expression was decreased in clusters 4 and 5 compared to the other clusters (fig. S1M). Collectively, the above data suggest that increased REST expression can occur both in early progenitors and in more neuronal lineage–committed cells, which define a subset of SHH-α and most SHH-β MBs, respectively. Three-fourths of SHH-α tumors (in clusters 1 and 4; 49 of 65) had notable losses in 9q (73.5%), 10q (46.9%), and 17p (40.8%), and gains in 9p (40.8%) (fig. S1J) (36). As expected, patients with SHH-β, SHH-γ, and SHH-δ exhibited fewer copy number changes (arm-level aberrations) compared to SHH-α tumors (fig. S1J) (36).

Generation and characterization of a novel genetically engineered mouse model with increased REST expression in CGNPs

The above findings led us to hypothesize that increased REST expression in CGNPs contributes to tumor progression and that lineage-committed cells could also undergo transformation. To test this postulate and to understand the specific contribution of REST to tumorigenesis, we first generated a novel transgenic mouse model (NeuroD2-REST), where conditional expression of hREST Tg in CGNPs is driven by a 1-kb fragment of the NeuroD2 promoter (Fig. 2A) (11). Tg presence was confirmed in NeuroD2-REST mice by quantitative polymerase chain reaction (qPCR), using primers against an epitope tag at the 5′ end of the hREST sequence and the 5′ end of the open reading frame using the same strategy as described in Fig. 2A for RESTTG mice. The latter were derived by crossing NeuroD2-REST mice with Math1-CreERT2 mice. Genotyping confirmed amplification of a specific 105-bp band in two RESTTG littermates (L1 and L2) but not in wild-type (WT) mice (Fig. 2A).

Fig. 2 Generation and characterization of a novel genetically engineered mouse model with increased REST expression in CGNPs.

(A) Schema to describe generation of a conditional RESTTG mouse model. Primers targeting the 6× His/3× hemagglutinin (HA) tag and the 5′ end of the hREST complementary DNA (cDNA) were used for genotyping. Agarose gel of PCR product from WT, line 1 (L1), and line 2 (L2) is shown. (B and C) CGNPs harvested from p8 pups that received tamoxifen (TX) injections on p2, p3, and p4 were cultured for up to 15 days. Cells were collected and analyzed for (B) REST Tg mRNA expression by qRT-PCR analyses and (C) REST protein abundance by Western blotting. Data are means ± SD of three individual pups. Representative Westerns are shown. a.u., arbitrary units. (D) Neuronal differentiation in CGNPs was evaluated by qRT-PCR measurement of Syn1 mRNA expression. Data are means ± SD of three individual pups. (E) H&E staining of brain tissue from p8 WT (n = 3) or RESTTG (n = 3) pups injected with tamoxifen. (F and G) Sections were analyzed by IHC for (F) REST and (G) NeuN expression using specific antibodies to assess protein expression in CGNPs in the EGL and granule neurons of the IGL (n = 3). For (B) and (C), bars represent means ± SD of fold changes relative to WT controls. P values for qRT-PCR were calculated by paired two-tailed t test of ΔCp values: Significance is indicated as ns. *P < 0.05, **P < 0.01, ***P < 0.001, or ****P < 0.0001. Densitometry was obtained using Image Lab software. Scale bars, 50 μm (×10; E) and 20 μm (×40; F and G).

hREST Tg expression was induced in vivo by intraperitoneal injections of tamoxifen of WT and RESTTG littermates. CGNPs were harvested from brains of postnatal day 8 (p8) progeny, grown in culture for 10 days. Suspended CGNPs were collected and replated in fresh growth media supplemented with recombinant SHH protein for an additional 5 days. CGNPs were collected, and quantitative reverse transcription PCR (qRT-PCR) analyses were performed. A statistically significant induction of hREST expression was seen specifically in CGNPs from three lineages of RESTTG mice but not in CGNPs from a control WT mouse (Fig. 2B), confirmed additionally by Western blot analyses of lysates from harvested CGNPs (Fig. 2C). Overall, the increase in hREST Tg expression ranged from 2- to 13-fold and was comparable to that previously demonstrated in human MB samples (28). As expected, a decrease in the expression of the REST-target neuronal differentiation gene, Syn1, was seen in CGNPs from RESTTG mice when compared to cells from age-matched WT mice (Fig. 2D).

We also assessed p8 RESTTG mice for changes in postnatal cerebellar development after tamoxifen administration and hREST Tg induction as compared to age-matched control WT littermates. Hematoxylin and eosin (H&E) staining identified areas of substantial EGL expansion in 90% of RESTTG animals compared to WT controls (Fig. 2E and fig. S2A). Immunohistochemistry (IHC) indicated visibly increased REST abundance in areas of EGL expansion in RESTTG mice compared to WT animals (n = 3) (Fig. 2F and fig. S2B). Unlike cells in the internal granule layer (IGL) of control WT mice, a noticeable increase in REST abundance was observed in IGL cells in cerebella from RESTTG animals and correlated with decreased abundance of a neuronal differentiation marker (NeuN) in CGNPs from the same tissue (Fig. 2, F and G, and fig. S2, B and C). Thus, increased REST expression was associated with abnormal EGL expansion and blockade of neuronal differentiation.

Increased REST expression in the context of constitutive SHH signaling activity worsens survival and promotes tumors with leptomeningeal spread

To examine the contribution of REST to progression of SHH-driven MBs, RESTTG mice were crossed with Ptch+/− mice, and hREST Tg expression was induced in the resulting progeny (Ptch+/−/RESTTG) by tamoxifen injection (Fig. 3A). Kaplan-Meier curves revealed a marked decrease in the survival of the 13 Ptch+/−/RESTTG mice; all died within 10 to 90 days of Tg induction (Fig. 3B). In contrast, 21.7% of RESTTG and 16.1% of Ptch+/− mice (n = 23 and n = 31, respectively) died by 14 months of age, whereas all 45 WT mice survived (Fig. 3B). Gross examination of brains of Ptch+/−/RESTTG mice revealed tumor burden in 100% of the mice (Fig. 3C; in which a representative brain from a 40-day-old mouse is shown). Tumor development in about 16% of Ptch+/− mice occurred with a considerably delayed latency between 141 and 332 days of age (Fig. 3B). The above results indicate that increased REST expression in the context of activated SHH signaling not only increased tumor penetrance but also caused a sharp decrease in tumor latency. H&E staining of the cerebella of tumor-bearing Ptch+/−/RESTTG mice revealed small blue cell tumors with leptomeningeal dissemination (Fig. 3D and fig. S3A). In contrast, tumors in Ptch+/− mice were more localized (Fig. 3D and fig. S3A).

Fig. 3 Increased REST abundance alters the kinetics and penetrance of SHH-driven MB development.

(A) Schema to describe generation of Ptch+/−/RESTTG mice. (B) Survival of WT (n = 45), RESTTG (n = 23), Ptch+/− (n = 31), and Ptch+/−/RESTTG (n = 13) mice after tamoxifen administration to induce REST Tg expression in RESTTG and Ptch+/−/RESTTG mice was assessed using Kaplan-Meier analysis. (C) Representative gross images of brains from p40 WT, RESTTG, Ptch+/−, and Ptch+/−/RESTTG mice are shown (n = 3). The red oval indicates cerebellar tumor in p40 Ptch+/−/RESTTG mice. Scale bar, 2 mm. (D) H&E staining of brain tissue from Ptch+/− and Ptch+/−/RESTTG animals (n = 3) is shown. (E) Immunodeficient mice bearing cerebellar xenografts of human DAOY cells expressing endogenous REST (n = 9) or hREST (DAOY-REST; n = 11) were monitored for tumor growth by bioluminescent imaging (BLI). Images of representative mice and relative flux for the entire cohort are shown before euthanasia on day 47 due to tumor burden. P values were obtained using Student’s t test. H&E staining of brain tissue from (F) DAOY and DAOY-REST xenografts (n = 3) and (G) low-REST and high-REST PDOX (n = 3) are shown. Scale bars, 50 μm (×10; D, F, and G).

Validation of these REST-dependent changes in tumor behavior was obtained using isogenic “low”-REST (DAOY) and “high”-REST (DAOY-REST) tumors obtained by implanting these cell lines in the cerebella of immunodeficient mice. DAOY-REST cells gave rise to larger tumors as assessed by bioluminescence imaging and in a larger cohort of animals compared to animals with DAOY cell tumors (Fig. 3E). The entire cohort of mice implanted with DAOY-REST tumors also developed infiltrative tumors and extracranial tumors compared to this phenotype only in 20% of mice implanted with DAOY cells (Fig. 3F and fig. S3B). Because the validity of DAOY as a bona fide “SHH” MB cell line is frequently questioned, mice bearing patient-derived orthotopic xenografts (PDOXs) were also examined. These studies revealed that high-REST–expressing SHH-driven MB tumors also displayed a propensity for leptomeningeal dissemination similar to DAOY-REST cells when compared to animals with low-REST SHH and DAOY tumors (Fig. 3G and fig. S3C).

Augmented REST expression causes Ptch1 loss of heterozygosity in tumors

Loss of heterozygosity (LOH) in Ptch1 has been suggested as a driver of tumor development in Ptch+/− mice (8, 15, 16). Work from other groups has also shown that deletion of both alleles of the Ptch1 gene in CGNPs and NSCs accelerates MB development (7, 16, 41, 49). To determine the status of Ptch1 expression in tumors from Ptch+/− and Ptch+/−/RESTTG animals, IHC analyses were performed using antibodies to PTCH1. These studies revealed strong PTCH1 staining in two of three tumor sections from Ptch+/− animals (Fig. 4A and fig. S4A). Unexpectedly, one of these tumors exhibited an increase, whereas the others had decreased REST abundance (Fig. 4A and fig. S4A). In Ptch+/−/RESTTG tumors, there was a decrease in PTCH1 staining in two of the three brains examined, and all three had high abundance of REST (Fig. 4A and fig. S4A). Sections of DAOY tumors (one of three) exhibited higher PTCH1 abundance compared to DAOY-REST tumors (one of three) (Fig. 4B and fig. S4B). REST abundance once again was increased in one of three DAOY tumors and all three DAOY-REST tumors (Fig. 4B and fig. S4B). In brain sections of mice bearing SHH-driven PDOX (n = 3), two low-REST tumors had a higher intensity of PTCH1 staining compared to one of three high-REST tumors (Fig. 4C and fig. S4C). Furthermore, transcriptome analyses of the SHH group of patient tumors (described in Fig. 1A) identified samples in cluster 5 (SHH-β) as having statistically significantly decreased expression of PTCH1 and GLI1 compared to the other clusters (Fig. 4D). SHH signaling activity as measured by expression of GLI2, GLI3, and SMO also suggested a decline in pathway in cluster 5 (SHH-β) tumors (fig. S4D). In addition, KI-67 staining was substantially increased in Ptch+/−/RESTTG, DAOY-REST, and PDOX–high-REST tumors when compared to Ptch+/−, DAOY, and PDOX–low-REST tumor sections (n = 3), respectively (Fig. 4, A to C, and fig. S4, A to C), suggesting that augmentation of REST increases proliferation in tumor cells.

Fig. 4 REST represses the expression of the gene encoding PTCH1.

(A to C) Cerebellar sections of (A) tumor-bearing Ptch+/− and Ptch+/−/RESTTG mice (n = 3), (B) DAOY and DAOY-REST xenografts (n = 3), and (C) human SHH subgroup PDOX (n = 3) were analyzed by IHC for REST, PTCH1, and Ki-67 expression using specific antibodies. (D) PTCH1 and GLI1 mRNA expression profiles in SHH-type MB patient samples measured by microarray. Hierarchical clustering based on expression of neuronal differentiation markers divided the SHH-type MB patient samples into six distinct clusters. Each dot corresponds to one individual patient. (E) Ptch1 and Gli1 mRNA expression was measured in CGNPs from WT (white bars) and RESTTG (gray bars) mice after culturing in proliferation (prolif) or differentiation (diff) media. Data are means ± SD from three (WT) or two (RESTTG) pups. Graph shows fold change compared to WT proliferating controls. Scale bars, 20 μm (×40). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and ns by paired two-tailed t test of ΔCp values.

REST and Gli1 competitively control Ptch1 gene expression in a lineage commitment context–specific manner

Because cells in human SHH-β tumors appeared to be more differentiated, we examined the strength of SHH signaling, as measured by Gli1 and Ptch1 mRNA expression, during differentiation of WT and RESTTG CGNPs in culture. qRT-PCR analysis was performed using mRNA extracted from WT and RESTTG CGNPs and cultivated under conditions supporting proliferation [medium supplemented with SHH, epidermal growth factor (EGF), and fibroblast growth factor (FGF)] and conditions supporting differentiation [medium lacking SHH, EGF, and FGF, but containing nerve growth factor (NGF)]. As expected, a five- and twofold decline in Gli1 and Ptch1 mRNA expression, respectively, was observed under differentiation conditions relative to their expression in proliferating cells (Fig. 4E). A similar analysis performed with CGNPs from RESTTG mice showed a small but statistically significant increase in the abundance of Ptch1 and Gli1 mRNA relative to each in proliferating WT cells (Fig. 4E). However, under differentiation conditions, Gli1 and Ptch1 mRNA expression is markedly less in CGNPs from RESTTG mice than in WT cells (Fig. 4E). As expected, REST was constitutively expressed in both proliferating and differentiating RESTTG CGNPs and was associated with lack of expression of the REST target gene, Scg10 (fig. S4, E and F).

Because Ptch1 is a target of GLI1-mediated transcriptional activation and because a search of the upstream regulatory region of Ptch1 gene in mice identified a REST binding RE1 element proximal to GLI1 binding sites in the Ptch1 promoter, we explored whether Ptch1 is a direct target of REST (Fig. 5A). To this end, we examined the binding of GLI1 and REST to the mouse Ptch1 promoter in proliferating CGNPs from WT and RESTTG mice and further compared it to that under differentiation conditions (Fig. 5A). Chromatin immunoprecipitation (ChIP) assays performed using antibodies to GLI1 and REST and control immunoglobulin G (IgG) revealed substantial, and comparably similar, REST binding to the RE1 site upstream of the Ptch1 promoter under proliferation and differentiation conditions in WT CGNPs (Fig. 5A). Although significantly less binding, if any compared with IgG controls, was observed in RESTTG CGNPs, REST occupancy of the RE1 site upstream of the Ptch1 promoter was increased twofold in differentiating RESTTG cells compared to its proliferating counterpart (Fig. 5A). Curiously, whereas substantial GLI1 binding to the RE1 site was observed in proliferating WT CGNPs, its binding was markedly less (by eightfold) in proliferating CGNPs from RESTTG mice (Fig. 5A). GLI1 binding to the Ptch1 promoter was decreased in both WT and RESTTG CGNPs under differentiating relative to proliferative conditions (Fig. 5A). Thus, in REST-overexpressing GCNPs, the binding of GLI1 to the Ptch1 promoter was compromised when compared to WT cells, more so under differentiating conditions.

Fig. 5 Transcription factor binding and resulting histone modification changes leads to Ptch1 loss of heterozygosity.

(A) Schematic representation of mPtch1 promoter with RE1 site and adjacent Gli1 binding site are shown. REST and GLI1 binding to RE1 site on mPtch1 promoter measured by ChIP-qPCR in WT and RESTTG proliferating (prolif) and differentiating CGNPs. Data are represented as fold change over IgG (n = 3 for WT and n = 6 for RESTTG). (B) Enrichment of H3Ac over IgG at mPtch1 promoter in proliferating and differentiating (diff) CGNPs. Bars represent fold change of H3Ac over IgG in the samples (n = 3 for WT and n = 6 for RESTTG). (C) Enrichment of H3K4-me3 evaluated by ChIP-qPCR at the mPtch1 TSS site in WT and RESTTG proliferating and differentiating CGNPs (n = 3). (D) Enrichment of mono-, di-, and trimethylation at histone H3 Lys9 (H3K9-me1, H3K9-me2, and H3K9-me3) evaluated by ChIP-qPCR at the mPtch1 RE1 site in WT and RESTTG proliferating and differentiating CGNPs (n = 3). (E) Arrb1 mRNA expression was measured in WT and RESTTG CGNPs after culturing with proliferation or differentiation media. Data are means ± SD from three (WT) or two (RESTTG) pups. Graph shows fold change compared to WT proliferating controls. (F) ARRB1 mRNA expression profile in human SHH-type MB patient samples measured by microarray. Hierarchical clustering based on expression of neuronal differentiation markers divided the SHH-type MB patient samples into six distinct clusters. Each dot corresponds to one individual patient. (G) Enrichment of H3K4-me3 at hPTCH1 TSS and enrichment of other histone modifications at hPTCH1 RE1 site using ChIP-qPCR from a high-REST PDOX sample. (H) DAOY MB cell line treated with either the HDAC inhibitor MS275 (0.625 to 5 μM) or the G9a inhibitor UNC0638 (0.5 to 5 μM), or a combination of both, and MTT assay was performed at 48 hours after treatment to measure cell viability. Data are means ± SD of independent triplicates. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and ns by two-way analysis of variance (ANOVA) in GraphPad and either Dunnett’s method (A to D) or Sidak or Tukey’s test (H) for multiple comparisons or by paired two-tailed t test of ΔCp values (E).

GLI1 and REST are canonical transcriptional activator and repressor, respectively. The abundance of the activating histone H3 acetylation mark at the RE1 site was unchanged between proliferating and differentiating WT CGNPs (Fig. 5B). However, a twofold decline in its abundance was seen in proliferating CGNPs from RESTTG animals when compared to proliferating control WT CGNPs, with an additional threefold decrease in this mark noted in differentiating RESTTG CGNPs (Fig. 5B). Trimethylation of Lys4 in histone H3 (H3K4-me3) is a mark that is found at transcriptionally active promoters (50). At the transcriptional start site (TSS) in the Ptch1 promoter, a 100-fold decrease in its abundance was noted between proliferating and differentiating WT CGNPs (Fig. 5C). A 10-fold reduction in its levels was seen in proliferating RESTTG CGNPs relative to WT cells, which was maintained under differentiation conditions (Fig. 5C). Thus, increased abundance of REST in RESTTG CGNPs, under both proliferation and differentiation conditions, led to an overall decrease in both Ptch1 promoter activity and histone acetylation around the upstream RE1 site, raising the possibility that aberrant REST expression creates a more closed chromatin architecture.

To address this possibility, we measured the activity of G9a, a histone H3K9 methyl transferase associated with the REST-CoREST complex, which canonically catalyzes the repressive monomethylation (me1) and dimethylation (me2) of histone H3K9. In rare instances, G9a also promotes histone H3K9 trimethylation (me3) (51, 52). To assess whether G9a regulates Ptch1 gene expression, we used ChIP assays to measure histone H3K9-me1, H3K9-me2, and H3K9-me3 in the cognate promoter in proliferating and differentiating CGNPs from WT and RESTTG mice. H3K9-me1 and H3K9-me2 marks decreased six- and threefold, respectively, at the Ptch1 promoter upon induction of neurogenesis in WT CGNPs (Fig. 5D). Under these conditions, a sevenfold increase in histone H3K9-me3 was also seen in WT CGNPs (Fig. 5D). In contrast, a fivefold increase and a threefold decrease in histone H3K9-me1 and H3K9-me2 abundance, respectively, relative to control WT cells were seen at the Ptch1 promoter in CGNPs from RESTTG animals (Fig. 5D). Onset of differentiation was accompanied by maintenance of H3K9-me1 and H3K9-me3 abundance and a threefold increase in H3K9-me2 abundance in CGNPs from RESTTG animals (Fig. 5D). These data are suggestive of a more repressive chromatin structure at the Ptch1 promoter in RESTTG CGNPs compared to WT cells, whether proliferating or differentiating. The lack of a change in abundance of the histone H3K9-me1 and H3K9-me3 marks in proliferating and differentiating RESTTG CGNPs suggests that increased REST expression leads to an overall premature compaction of the Ptch1 promoter (Fig. 5D).

The above results were unexpected in a few aspects. First, histone H3K9-me3 was seen at the Ptch1 promoter during neuronal differentiation of both WT and RESTTG CGNPs. We therefore performed coimmunoprecipitation assays to examine whether REST formed a complex with Suv39H1, a histone H3K9 trimethyl transferase; however, we did not detect a significant association between REST and Suv39H1 in the DAOY MB cell line, although both proteins were detected by Western blotting (fig. S5). Second, despite the reduction in GLI1 binding, decreases in histone H3 acetylation and histone H3K4-me3, and an increase in histone H3K9-me1 and H3K9-me3 in proliferating RESTTG CGNPs relative to WT cells, a corresponding decrease in Ptch1 transcript abundance was not seen (Fig. 4E). To explain this conundrum, we explored whether a negative regulator of GLI1 activity was down-regulated in proliferating RESTTG CGNPs. Expression of Arrb1, encoding β-arrestin1, known to facilitate inhibitory acetylation of GLI1 protein by the histone acetyltransferase p300 was significantly (10-fold) decreased in proliferating RESTTG CGNPs compared to WT cells (Fig. 5E) (20, 21, 53). Its expression was increased in both cell types grown under differentiating conditions, but relatively and significantly less so in RESTTG CGNPs (Fig. 5E). In human MB tissue samples, ARRB1 expression was significantly less in SHH-α tumors than in SHH-β tumors (Fig. 5F). Existence of associated histone acetylation and activating and repressive histone methylation events at the RE1 site and at the promoter was also confirmed through ChIP assays in a PDOX tumor sample (Fig. 5G).

Last, to confirm that histone H3K9 methylation and histone acetylation play a role in the viability of human SHH MB cells, DAOY and UW228 MB cells were treated with a G9a inhibitor (UNC0638) and an HDAC inhibitor (MS275) either alone or in combination at various doses. Treating cells with UNC0638 alone had only a small effect in one cell line and a counterintuitive prosurvival effect in the other (Fig. 5H), whereas treating cells with MS275 or the combination had a statistically significant cytotoxic effect (Fig. 5H), including an enhanced and synergistic effect with the combination (table S1). Thus, these findings collectively indicate that Ptch1 is a REST target gene, that REST and GLI1 competitively control Ptch1 gene expression, and that pharmacologically targeting HDACs and G9a activity can synergistically decrease the proliferation of MB cells in culture.

REST overexpression induces phosphatase tensin homolog (PTEN) loss and AKT activation

Another hallmark of our Ptch+/−/RESTTG animals was leptomeningeal dissemination and infiltrative behavior of tumor cells. Several studies have demonstrated the relevance of hyperactivation of AKT signaling in MB metastasis and dissemination (54, 55). To examine the status of AKT signaling in tumors from Ptch+/− and Ptch+/−/RESTTG animals, brain sections from these animals were stained with antibodies to Ser473–phosphorylated AKT and analyzed by IHC (n = 3 each). Whereas tumors in both Ptch+/− and all Ptch+/−/RESTTG animals exhibited phosphorylation of AKT at Ser473, it was markedly stronger in the Ptch+/−/RESTTG brain sections, indicating a REST elevation–dependent hyperactivation of AKT (Fig. 6A and fig. S6A). Brain slices from mice bearing isogenic DAOY and DAOY-REST xenografts or PDOXs of low- and high-REST–expressing, SHH-driven MB tumors also suggested an increase in AKT activity that was dependent on REST abundance (Fig. 6, B and C, and fig. S6, B and C).

Fig. 6 Increased REST expression in the context of constitutive SHH signaling results in increased AKT activation.

(A to C) IHC was performed with phosphorylated (p)–AKTSer473 or PTEN-specific antibodies in (A) Ptch+/− and Ptch+/−/RESTTG tumors (n = 3), (B) DAOY and DAOY-REST xenografts (n = 3), and (C) human SHH subgroup PDOX (n = 3). Scale bars, 20 μm (×40). (D) PTEN mRNA expression profile was measured by microarray. Hierarchical clustering based on expression levels of neuronal differentiation markers divided the SHH MB patient samples into six distinct clusters. Each dot corresponds to one individual patient. (E) Pten mRNA expression was measured in WT and RESTTG CGNPs after culturing in proliferation or differentiation media. Graph shows fold change compared to WT proliferating controls. Data are means ± SD from three (WT) or two (RESTTG) pups. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and ns by paired two-tailed t test of ΔCp values. (F and G) Western blot analysis of p-AKTSer473 and total AKT abundance in WT and RESTTG CGNPs after culturing with proliferation or differentiation media unperturbed (F) or after 5 hours of treatment with MK2206 at 1 or 5 μM (G). Histone H3 served as loading control. Images are representative of n = 2 independent experiments.

Copy number changes leading to loss of PTEN are a clinically relevant characteristic of human SHH-β tumors, and various studies have shown that PTEN loss contributes to MB tumor progression in mice (36, 5658). Therefore, we investigated whether AKT hyperactivation in tumors from Ptch+/−/RESTTG animals was associated with a decline in PTEN abundance. We did not detect PTEN staining in tumors from Ptch+/−/RESTTG mice, compared with two of three tumors from Ptch+/− animals (Fig. 6A). The correlation between REST and PTEN was also evaluated in DAOY/DAOY-REST tumors and high- and low-REST PDOXs. PTEN abundance was detectable in two of three DAOY and one of three DAOY-REST tumors (Fig. 6B and fig. S6B). In PDOX tumors, PTEN abundance was detectable in all three low-REST tumors and two of three high-REST tumors (Fig. 6C and fig. S6C). PTEN expression was decreased in cluster 2 (SHH-α) and cluster 5 (SHH-β) tumors, significantly so relative to all others except cluster 3 (Fig. 6D).

Changes in Pten expression were also assessed during proliferation and differentiation of CGNPs from WT and RESTTG mice by qRT-PCR analyses. Its expression was increased sevenfold in differentiating cells relative to that in proliferating WT CGNPs (Fig. 6E). However, a slightly less robust fivefold up-regulation was observed in differentiating RESTTG CGNPs relative to that in proliferating cells from RESTTG mice (Fig. 6E). Similar to Pten, expression of Akt1 transcript was up-regulated in differentiated CGNPs from WT and RESTTG mice compared to proliferating counterparts; however, the increase was less robust (3.5-fold) in RESTTG cells than in the five- to sevenfold change in WT cells (fig. S6E), and expression of Akt2 and Akt3 mRNA was increased only in differentiating WT CGNPs (fig. S6E). In human tumors, the expression of AKT1 and AKT3 transcripts was not increased in cluster 1, 2, or 5 (fig. S6D), but that of the AKT2 transcript was significantly increased in cluster 2 compared to cluster 5 (fig. S6D).

Subsequently, Western blotting in WT and RESTTG CGNPs showed similar amounts of phosphorylated (activated) AKT in proliferating compared with differentiating WT CGNPs (Fig. 6F). However, the abundance of phosphorylated AKT was increased 2.5-fold in proliferating RESTTG CGNPs relative to WT cells and was similarly high in differentiating RESTTG CGNPs (Fig. 6F). However, total AKT protein abundance was also increased 2.5-fold in differentiating RESTTG CGNPs compared to that in proliferating cells (Fig. 6F). Thus, both Pten and Akt1 were up-regulated at the transcriptional level during neurogenesis in WT and RESTTG CGNPs, albeit to a lesser extent in RESTTG CGNPs, consistent with the increase in total AKT protein abundance and its phosphorylated form in differentiating RESTTG CGNPs relative to WT cells. AKT phosphorylation was blocked in proliferating and differentiating WT and RESTTG CGNPs by treatment with the AKT inhibitor MK2206 at a dose of 5 μM (Fig. 6G, right panel), but not as efficiently at a lower dose of 1 μM in differentiating RESTTG relative to WT CGNPs (Fig. 6G, left panel). Unexpectedly, at the same drug concentration, AKT phosphorylation was twofold lower in proliferating RESTTG CGNPs compared to WT CGNPs (Fig. 6G, left panel).

SHH subgroup human MB cell lines (DAOY, UW426, and UW228) also exhibited a similar association between REST abundance and AKT activation as measured by Western blotting (Fig. 7A). Knockdown and overexpression approaches were then taken to establish a dependency of AKT activation on REST. REST knockdown in UW228 and DAOY cells with each of two REST-specific short hairpin RNAs (shRNAs) decreased the abundance of phosphorylated AKT at Ser473 (Fig. 7B), whereas total AKT abundance was also slightly, but at least with the second shRNA not proportionally, decreased (Fig. 7B and fig. S7).

Fig. 7 REST-dependent AKT phosphorylation in MB cell lines.

(A) Western blotting for basal protein abundance of REST, p-AKTSer473, total AKT, and histone H3 (control) in DAOY, UW426, and UW228 cells. Representative blots are shown; long/short indicates exposure times. (B and C) Western blotting for total and p-AKTSer473 protein abundance after either shRNA-mediated REST knockdown in UW228 and DAOY cells (B) or REST overexpression in DAOY cells (C). Blots are representative of three experiments. (D) MTT assay–derived proliferation of UW228 and DAOY cells treated with various doses of MK2206 for 24, 48, or 72 hours. Data are means ± SD of three independent assays. (ns), *P < 0.05, **P < 0.01, ***P < 0.001, or ****P < 0.0001. (E) Western blotting for abundance of p-AKTSer473, total AKT, cleaved caspase-3, cleaved PARP, and histone H3 (loading control) to assess induction of apoptosis after treatment of UW228 cells with MK2206 (5 μM) for 12 or 24 hours. Blots are representative of three experiments.

Conversely, REST overexpression in DAOY cells led to an increase in the abundance of both total and Ser473–phosphorylated AKT (Fig. 7C). Pharmacological inhibition of AKT activity in UW228 and DAOY cells by the AKT inhibitor MK2206 caused a dose- and time-dependent decrease in the viability of both cells (Fig. 7D). However, DAOY cells, which have relatively greater REST expression, were less sensitive to MK2206 than were UW228 cells, as determined by IC50 (half-maximal inhibitory concentration) measurements (Fig. 7D and table S2). As expected, AKT inhibition with MK2206 promoted caspase-3 activation and PARP [poly(adenosine 5′-diphosphate–ribose) polymerase] cleavage, indicating activation of apoptosis (Fig. 7E). Thus, aberrant REST expression was associated with loss of PTEN expression and increased AKT activation in both mouse and human MBs.

In conclusion, our findings collectively suggest that increased REST expression in SHH MBs supports a more aggressive disease course (Fig. 8A) by promoting epigenetic-mediated repression of PTCH1 and a decrease in PTEN, thereby enhancing SHH and AKT signaling, respectively (Fig. 8, B and C). Hyperactivation of AKT and loss of PTEN have been described in human MBs and are known to drive tumor progression in mouse models (56).

Fig. 8 Increased REST expression drives progression of SHH-driven MBs.

(A) Schematic representation of tumor characteristics obtained from CGNPs with perturbed SHH signaling in the presence or absence of increased REST expression. (B) Graphical representation of REST-dependent chromatin remodeling of the Ptch1 gene in WT or RESTTG CGNPs during proliferation and differentiation. (C) Model depicting REST regulation of AKT signaling in SHH-driven MBs.


Whether misregulation of neurogenesis contributes to MB tumorigenesis had not been evaluated until the discovery that ectopic expression of REST in NSCs recapitulated the poorly differentiated phenotype of human tumors in mice (23, 28). Increased REST expression in tumors was associated with a decrease in survival of a small subset of patients with desmoplastic MBs, which are SHH-driven tumors (2, 4, 29). Our analysis of human SHH tumor transcriptome data yielded an unexpected result in that increased REST expression and activity correlated with two clusters of SHH tumor samples with poor prognostic significance for patients (29). One was more immature in its lineage commitment and aligned with an SHH-α molecular profile, whereas the other displayed significant expression of neuronal differentiation markers and aligned with an SHH-β molecular profile. These results suggest that MBs can arise from immature and more lineage-committed CGNPs, which is concordant with not only our findings here but also a previous underappreciated observation that Smo1 oncogene expression in lineage-committed cells promoted transformation (11, 58). There is growing acceptance that cells at various stages of lineage commitment can undergo transformation (16, 5961). Of importance, down-regulated expression of BTRC, a proteasome component that degrades REST and GLI1 proteins, in human SHH-α and SHH-β MB clusters, and its association with increased activity of these proteins, underscores the need to consider transcriptomic changes in conjunction with proteomic alterations (6265).

Because tumor formation was not observed in RESTTG animals, the reason for their decreased survival needs investigation. However, mice with constitutive activation of SHH signaling and increase REST abundance formed tumors with accelerated kinetics and penetrance, highlighting its importance in driving tumor progression. The finding that increased REST abundance in CGNPs causes haploinsufficiency and not LOH is consistent with the current consensus that mono-allelic loss of Ptch1 leads to preneoplastic lesions only (15, 16, 66, 67). However, in the context of Ptch1 haploinsufficiency, REST caused Ptch1 LOH and promoted rapid MB formation, which aligns with studies that have shown deletion of both alleles of Ptch1 in CGNPs and NSCs to accelerate MB kinetics in mice (16, 68). Last, increased REST abundance was also seen in Ptch+/− tumors, suggesting that deregulation of REST may be a second “hit.” However, onset of tumors in older mice suggests a greater relevance to the adult SHH-δ subgroup. The issue of potential tumor heterogeneity in our model system also remains to be verified.

Another important finding of our work is that a novel and complex interaction between REST and GLI1 controls SHH signaling. In our study, normal neurogenesis of CGNPs was accompanied by a decline in SHH pathway activity, as measured by Ptch1 and Gli1 expression. The first point of convergence between REST and SHH pathway was noted at the Ptch1 gene and involved an antagonistic interaction between REST and GLI1 proteins for occupancy and chromatin remodeling at the Ptch1 promoter. The second point of convergence occurred further downstream at GLI1 and involved a REST-dependent silencing of the Arrb1 gene, which encodes a negative regulator of GLI1 protein activity (53). In proliferating cells, GLI1 appeared to prevail over REST at the Ptch1 locus, possibly because of a lack of an inhibitory effect of β-arrestin on GLI1 protein. In differentiating cells, up-regulation of the gene encoding β-arrestin (Arrb1) and decreased GLI1 activity were associated with a decline in SHH signaling. We propose that this regulation is perturbed in tumors with increased REST expression in the context of Ptch1 haploinsufficiency. Specifically, we suggest that in tumors arising from proliferating Ptch+/−/RESTTG CGNPs, premature chromatin compaction at Ptch1 and consequent functional Ptch1 LOH in conjunction with increased GLI1 activity promoted by loss of Arrb1 expression facilitated SHH pathway activity and drove a human SHH-α–like tumor. In contrast, in tumors arising consequent to increased REST expression in more neuronal lineage–committed Ptch+/−/RESTTG CGNPs, a further compaction of the Ptch1 locus associated with diminished GLI1 activity associated with up-regulation of Arrb1 results in decreased SHH signaling and is reminiscent of human SHH-β tumors. This apparent tumor heterogeneity in our system needs further investigation. This body of work also adds alterations in histone H3K9 methylation as an important epigenetic event in MB etiology; nevertheless, several questions remain. REST-dependent augmentation of G9a activity and a noncanonical increase in H3K9-me3 at the Ptch1 promoter are inexplicable, although enhanced chromatin retention may provide a possible explanation (69, 70). The synergistic decline of human MB cell line proliferation in vitro upon treatment with MS275 and UNC0638 may be due to a direct effect on REST-HDAC activity or indirectly through modulation of Arrb1 activity, or both, and remains to be explored.

The REST-AKT phosphorylation link also ties chromatin remodeling to leptomeningeal MB dissemination. It is supported by data from a seminal screen that implicated REST, Pten, and Akt as separate candidate drivers of metastasis in SHH-driven MB mouse models (71, 72). Increased AKT activity and genetic deletion of Pten in the context of Ptch haploinsufficiency are known drivers of murine MB metastasis (19, 33, 55, 7175). We propose that REST-dependent loss of Pten expression may also account for increased phosphorylation of AKT in MBs. In human SHH-type MBs, genetic loss of PTEN is seen in SHH-β tumors (36). However, our data argue that down-regulation of PTEN expression can also occur in a subset of SHH-α tumors.

In addition to the above events, other molecules, such as N-Myc, may also be deregulated in our mouse model (28, 67). Although the status of N-Myc was not specifically examined here, it is a downstream target of GLI1, and we previously showed that REST knockdown decreases N-Myc abundance in MB cells (22, 76). From a therapeutic perspective, AKT signaling is known to promote MB cell survival (77). Therefore, sensitivity of high-REST SHH-driven MBs to drugs used in the clinic as standard of care needs to be carefully examined. Further preclinical investigation of the efficacy of G9a and HDAC inhibitors, either alone or in combination with AKT, SMO, and GLI inhibitors, against high-REST MBs is necessary.

In conclusion, growing evidence suggests that chromatin remodelers such as REST contribute to SHH-driven MBs (27, 29, 78, 79). Both immature and lineage-committed CGNPs appear to contribute to SHH-driven MB formation. Their molecular dissection will provide a rationale for testing unique combinations of pharmacological agents against human SHH tumor subgroups, especially SHH-α and SHH-β tumors, where prognosis continues to be grim.



hREST Tg was cloned into a modified pcDNA3.1-V5/His plasmid wherein the cytomegalovirus promoter was replaced by a 1-kb region of the NeuroD2 (ND2) promoter from plasmid pcS2SmoA1 (54). A 6× His/3× HA epitope tag was added to the N terminus of hREST to generate pcDNA3.1/ND2/REST, followed by insertion of a LoxP-1×Stop-LoxP site from the Lox-Stop-Lox TOPO plasmid between the ND2 promoter and hREST Tg (Addgene plasmid #11584) (29).


Plasmid pcDNA3.1/ND2/LoxP-REST was restriction-digested, and the ND2-LoxP-Stop-LoxP-hREST DNA fragment was used for pronuclear injection of embryos from C57/Bl6 mice (National Cancer Institute, Bethesda, MD) at the Institutional Genetically Engineered Mouse Facility. Three NeuroD2-REST transgenic founders were backcrossed for several generations and then crossed to mice conditionally expressing Cre recombinase under the Math1 promoter (Math1CreERT2; the Jackson Laboratory) to create RESTTG animals. hREST expression was induced by intraperitoneal injections of (100 μl of 2 mg/ml) tamoxifen (Sigma-Aldrich) on p2, p3, and p4. Immunocompromised nonobese diabetic–severe combined immunodeficient IL2rgammanull (NSG) mice (the Jackson Laboratory) were purchased for xenograft studies. The intracranial inoculation of cells into NSG mice was performed using a stereotactic device as described previously (13). All mice were housed and treated in accordance with the guidelines of The University of Texas MD Anderson Cancer Center’s Animal Care and Use Committee. NSG mice were subjected to bioluminescence imaging (BLI) weekly and were euthanized on day 47 or at the onset of symptoms. For BLI, mice were given intraperitoneal injections of d-luciferin (150 mg/kg; Promega, Madison, WI), anesthetized with 2.5% isoflurane, and imaged using the Xenogen Spectrum (IVIS-200) system.

Tumor samples

Histopathological review of MB samples and analyses were performed following Institutional Review Board approval to C. Hawkins from the Hospital for Sick Children, Toronto. Sections of paraffin-embedded tissue were studied using H&E and IHC staining performed by V. Rajaram (The University of Texas Southwestern Medical Center, Dallas, TX) and C. Hawkins. All antibodies are listed in table S3.

Statistical analysis

Statistical significance of overall survival between various mouse groups was calculated by log-rank (Mantel-Cox) test. For all qRT-PCR data, paired two-sided t tests were performed using GraphPad Prism version 7.0 for Windows (GraphPad Software Inc., San Diego, CA, USA). Data are shown as means ± SD of at least three independent samples. P < 0.05 was considered to be statistically significant. Significance is indicated as *P < 0.05, **P < 0.01, ***P < 0.001, or ****P < 0.0001; where necessary for clarity, lack of significance is indicated (ns). Hierarchical clustering was performed using ArrayTrack software available at using Ward’s method. Samples were divided into three to six clusters based on the dendrogram from hierarchical clustering to keep at least 15 patients in each cluster for further analyses. P values for comparisons between every pairwise combination among clusters based on gene expression status were obtained using the unpaired t test with Welch’s correction using GraphPad Prism version 7.0.

Gene expression profile in patient samples

Microarray datasets containing the gene expression values of patients with MB were obtained from Gene Expression Omnibus ( We used the GSE85217 dataset, which contained Affymetrix Human Gene 1.1 ST Array profiling of 763 primary MB samples to evaluate gene expression. Microarray data were normalized using the robust multiarray average method. The expression data for each gene were Z score–transformed.

IHC and microscopy

Mouse brain tissues were fixed in 10% buffered formalin phosphate and embedded in paraffin. Four-micrometer-thick brain sections were used for IHC analysis. Sections were deparaffinized with 100% histoclear solution or xylene, followed by rehydration with ethanol and water. Sections were treated with 3% H2O2 solution for 10 min to block the endogenous peroxidase. Antigen retrieval was performed using a pressure cooker (Aroma). Slides were placed in citrate buffer (pH 6.0) and incubated for 30 to 60 min under steam conditions. The samples were then cooled and washed in tris-buffered saline–Tween (TBST). Blocking was performed by incubating the sections in blocking buffer (1% bovine serum albumin + 5% normal goat serum in 1× TBST) for 1 hour. The sections were then incubated with primary antibodies as indicated 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 3,3′-diaminobenzidine (Vertex) as a substrate and counterstained with hematoxylin. After dehydration and mounting, slides were dried and visualized under a microscope. All antibodies used are listed in table S3. Stained slides were viewed using a Nikon ECLIPSE E200 microscope, and images were captured under ×4, ×10, and ×40 magnification with an Olympus SC100 camera. Analyses were performed using Olympus cellSens Entry software. Whole mouse brains were viewed under an Olympus SZ61 microscope. Image capture and analyses were performed as described above.

Cell culture

Human MB cell lines DAOY, UW228, and UW426 were cultured as described previously (29). Cerebellar tissues from p8 WT and RESTTG pups were dissected and triturated using an 18.5-gauge needle with a 10-ml syringe to form single cell suspension. The cells were grown for 10 days in Neurobasal medium supplemented with B27 (vitamin A), Glutamax, antibiotic/antimycotic, heparin, and EGF and FGF (20 μg/ml), triturating every 4 days to disrupt cell attachment to the plate. Neurospheres were grown for an additional 4 days in the presence of recombinant SHH at a concentration of 0.1 μg/ml. The proliferating cells were collected and used for further analysis. To differentiate the cells, the proliferating neurospheres were put back in culture containing NGF (20 μg/ml) without EGF and FGF. The cells were cultured in a six-well plate for 5 days, and the attached differentiated cells were collected after dissociation with TrypLE and used for further experiments. Unattached cells were discarded.

Plasmid and cell transfection

The pcDNA3.1/ND2/REST plasmid was stably transfected into DAOY and UW426 cells using Lipofectamine 2000 reagent in accordance with the manufacturer’s instructions (Invitrogen).

MTT assay

DAOY, UW228, DAOY–high-REST versus DAOY–low-REST, and UW426–high-REST versus UW426–low-REST cells were cultured in 96-well microplates and treated with various concentrations of the AKT inhibitor MK2206 (U.S. Biological Life Science), the HDAC inhibitor MS275 (catalog no. 13284, Cayman Chemical), and the G9a inhibitor UNC0638 (catalog no. 10734, Cayman Chemical), either alone or in combination for the time period indicated in the figures. After labeling with 3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide (MTT/thiazolyl blue tetrazolium bromide; Sigma-Aldrich), cell growth was analyzed using SpectraMax Plus 384 microplate reader (Molecular Devices LLC). Synergy between MS275 and UNC0638 was calculated using the fractional product method of Webb using the formula g1g2 = g12 following the procedure described previously (80).

ChIP and qPCR from proliferating and differentiating CGNPs

CGNPs from proliferating and differentiating cells were fixed with 1% formaldehyde, cross-linked, and processed for ChIP analyses according to the manual from Millipore. Briefly, cross-linked cells were washed with 1× phosphate-buffered saline (PBS), lysed using lysis buffer [50 mM tris-HCl (pH 8.0), 10 mM EDTA (pH 8.0), 1% SDS, and protease inhibitors], and sonicated, and 1% of this material was saved as input DNA. The remainder of the samples was diluted fivefold with ChIP dilution buffer [16.7 mM tris-HCl (pH 8.0), 167 mM NaCl, 1.2 mM EDTA (pH 8.0), 1.1% Triton X-100, and protease inhibitors], precleared, and incubated with various antibodies for 12 hours at 4°C. The complex was then incubated with protein-A beads (Millipore), washed, and eluted. After reversal of the cross-link, DNA was purified with a PCR purification kit (Zymo Research). Bound DNA was quantified by SYBRGreen qPCR using Roche LightCycler 96. Data were analyzed using the comparative 2−ΔΔCp method. The antibodies used for ChIP and the primers used for the mouse Ptch1 are listed in tables S3 and S4, respectively.

ChIP and qPCR from PDOX samples

ChIP assays were performed by the Center for Cancer Epigenetics Core facility with the following modifications to a previously described high-throughput ChIP protocol (81). Briefly, 100 mg of PDOX tissue was dissociated in Hanks’ balanced salt solution using a MACS dissociator to obtain single cell suspension. The cell suspension was cross-linked in 1% formaldehyde for 10 min at room temperature, followed by incubation with glycine for 5 min to stop cross-linking. Cells were collected and washed with ice-cold PBS, and crude nuclei were isolated using cell lysis buffer [5 mM Pipes (pH 8.0), 85 mM KCl, and 0.5% NP-40, supplemented with protease inhibitor] for 10 min, followed by centrifugation for 5 min. The nuclear pellet was lysed for 30 min on ice using lysis buffer [12 mM tris-HCl (pH 7.5), 6 mM EDTA (pH 8.0), and 0.5% SDS] supplemented with protease inhibitor. Chromatin lysates were fragmented with the Bioruptor (Diagenode) to obtain DNA fragments ranging from 200 to 600 bp. After centrifugation, the supernatant was collected and incubated with respective antibodies conjugated with Dynabeads Protein G (Invitrogen) overnight at 4°C. The immunocomplexes were collected using Dynamag, washed as described in the protocol, treated with ribonuclease and proteinase K, and reverse cross-linked overnight, followed by DNA extraction. The DNA region of interest was detected by SYBR green real-time qPCR using primers to the human PTCH1 promoter (table S4).


Whole cell lysates (WCLs) were prepared from DAOY cells. Briefly, DAOY cells were collected and washed in PBS (Corning Cellgro). Cells were cross-linked in 1% formaldehyde, rocked at room temperature for 10 min, and washed 2× in PBS. Cells were collected by centrifugation and incubated in lysis buffer [50 mM tris-HCl (pH 8.0), 50 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, and protease/phosphatase inhibitors] for 5 min on ice with periodic mixing. The lysates were clarified by centrifugation at 13,000g for 10 min at 4°C, and supernatants were aliquoted. WCLs were precleared at 4°C for 30 min with gentle rocking. Beads were collected by centrifugation at 1000 rpm for 1 min at 4°C. Supernatant was transferred to a new tube and diluted 1:4 with lysis buffer. Input (2.5%) was collected and stored at −80°C. Primary antibody or control IgG was added to the tube and incubated for 12 hours at 4°C with gentle rocking. The complex was then incubated with protein-A beads (Millipore) for 3 hours at 4°C, washed 4× [20 mM tris-HCl (pH 7.4), 50 to 300 mM NaCl, 0.1% NP-40, 1 mM dithiothreitol, 5 mM EDTA, 25% glycerol, and protease/phosphatase inhibitors], and eluted with lysis buffer plus SDS loading dye. Proteins were analyzed by Western blotting as previously described. Protein bands were developed using the SuperSignal West Dura Extended Duration Substrate and detected using the ChemiDoc Touch Imaging System (Bio-Rad).

Quantitative reverse transcription PCR

CGNPs from WT and RESTTG mice were harvested, cultured, and collected as described above. RNA was extracted using the Quick-RNA MiniPrep Kit (Zymo Research). Equal amounts of RNA up to 1 μg were reverse-transcribed into cDNA using the iScript cDNA Synthesis Kit (Bio-Rad). qRT-PCR was performed in triplicate with a 2× SensiMix SYBR & Fluorescein Kit (Bioline) using the LightCycler 96 Real-Time PCR System (Roche Diagnostics GmbH). Relative mRNA expression normalized to 18S ribosomal RNA was determined by the comparative 2−ΔΔCp method. Relative mRNA expression was graphed as fold change compared to WT controls. Primer sequences are listed in table S4.

Lentiviral infection

For shRNA: human embryonic kidney–293T cells were cotransfected with control or REST shRNA, together with packaging plasmid (PAX2) and envelope plasmid (MD2). Lentiviral particles were harvested 48 hours after transfection. DAOY and UW228 cells were transduced with the collected viral supernatant in the presence of polybrene (8 μg/ml) and incubated for 48 hours. Infected cells were cultured in medium containing puromycin (2 μg/ml) for selection up to 1 week.

Western blot analysis

DAOY, UW228, and UW426 cells were collected, and cell extracts were prepared for Western blot analysis. Briefly, cell extracts were prepared by incubation in lysis buffer [50 mM tris-HCl (pH 8.0), 50 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, and protease/phosphatase inhibitors] for 30 min on ice. The lysates were clarified by centrifugation at 13,000g for 10 min at 4°C, and the supernatants were collected and boiled in SDS loading buffer (Cell Signaling Technology). Proteins were separated by electrophoresis on 10% SDS-polyacrylamide gels (Bio-Rad), transferred to Hybond-P polyvinylidene difluoride membranes (GE Healthcare), and analyzed by Western blotting with the indicated primary antibodies (table S3) and HRP-conjugated goat anti-mouse or anti-rabbit secondary antibodies (Thermo Fisher Scientific). Protein bands were developed using the SuperSignal West Dura Extended Duration Substrate (Thermo Fisher Scientific) and detected using the Kodak Medical X-Ray Processor 104 (Eastman Kodak Company) and the ChemiDoc Touch Imaging System (Bio-Rad). Images were analyzed using Image Lab software version 5.2.1 (Bio-Rad).


Fig. S1. Clinical characteristics.

Fig. S2. Cerebellar architecture and protein abundance in age-matched RESTTG and WT mice.

Fig. S3. Increased REST expression contributes to infiltrative SHH-driven MB development.

Fig. S4. Differential regulation of PTCH1 expression by REST in MB tumors and CGNPs.

Fig. S5. REST and SUV39H1 do not coimmunoprecipitate in DAOY cells.

Fig. S6. Increased REST expression increases AKT phosphorylation at Ser473.

Fig. S7. REST-dependent AKT phosphorylation in cell lines.

Table S1. Synergy between MS275 and UNC0638.

Table S2. IC50 values for UW228 and DAOY cells treated with MK2206.

Table S3. Antibodies for IHC, qChIP (quantitative ChIP), and Western blotting assays.

Table S4. Primers for qChIP and qRT-PCR assays.


Acknowledgments: We thank M.C. Hung for helpful comments, X. Shi for providing reagents, and R. DuBois for support. Funding: This work was supported by grants from the NIH (5R01-NS-079715-01 and 5R03NS077021-01 to V.G. and R01 CA185402 to X.-N.L.), the American Cancer Society (RSG-09-273-01-DDC to V.G.), the Cancer Prevention Research Institute of Texas (CPRIT-RP150301 to V.G.), Addi’s Faith Foundation and the Rally Foundation for Childhood Cancers (to V.G.), and The University of Texas MD Anderson Cancer Center-CCE Scholar Program (to T.H.W.D.). Author contributions: T.H.W.D.: Conceptualization and performance of experiments, analysis of data and generation, and editing of the manuscript. R.-H.T.: Conceptualization and performance of experiments, analysis of data, and editing of the manuscript. S.M., J.S., S.S., P.T., B.K., Y.Y., A.S., A.R.H., K.C., and J.B.-A.: Experimental design, execution, and analyses. M.K., L.Q., S.K., S.G., R.R.L., J.F., T.J.M., and X.-N.L.: Provided important reagents and helped with editing of the manuscript. C.H. and V.R.: Assisted with review of pathology, analysis of data, and editing of the manuscript. V.G.: Conceptualization, experimental design, data analysis, and writing and editing of the manuscript. Competing interests: S.K. is a member of the Data Safety Monitoring Board of the Nationwide Children’s Hospital. All other authors declare that they have no competing interests. Data and materials availability: The microarray data underlying our analyses are previously published and publicly available from Gene Expression Omnibus (; dataset GSE85217). All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.

Stay Connected to Science Signaling

Navigate This Article