REST promotes ETS1‐dependent vascular growth in medulloblastoma

Abstract

Expression of the RE1‐silencing transcription factor (REST), a master regulator of neurogenesis, is elevated in medulloblastoma (MB) tumors. A cell‐intrinsic function for REST in MB tumorigenesis is known. However, a role for REST in the regulation of MB tumor microenvironment has not been investigated. Here, we implicate REST in remodeling of the MB vasculature and describe underlying mechanisms. Using REST^TG mice, we demonstrate that elevated REST expression in cerebellar granule cell progenitors, the cells of origin of sonic hedgehog (SHH) MBs, increased vascular growth. This was recapitulated in MB xenograft models and validated by transcriptomic analyses of human MB samples. REST upregulation was associated with enhanced secretion of proangiogenic factors. Surprisingly, a REST‐dependent increase in the expression of the proangiogenic transcription factor E26 oncogene homolog 1, and its target gene encoding the vascular endothelial growth factor receptor‐1, was observed in MB cells, which coincided with their localization at the tumor vasculature. These observations were confirmed by RNA‐Seq and microarray analyses of MB cells and SHH‐MB tumors. Thus, our data suggest that REST elevation promotes vascular growth by autocrine and paracrine mechanisms.

Abbreviations
• ANG
• angiogenin
• ANGPT1
• angiopoietin‐1
• ANGPT2
• angiopoietin‐2
• BLI
• bioluminescence imaging
• CCL2 / MCP1
• C‐C motif chemokine ligand 2 / monocyte chemoattractant protein‐1
• CGNPs
• cerebellar granule neuron progenitors
• CXCL16
• C‐X‐C motif chemokine ligand 16
• CXCL8 / IL‐8
• C‐X‐C motif chemokine ligand 8 / interleukin‐8
• ETS1
• E26 oncogene homolog 1
• H&E
• hematoxylin and eosin
• HBMEC
• human brain microvascular endothelial cell
• HUVEC
• human umbilical vein endothelial cell
• IHC
• immunohistochemistry
• MB
• medulloblastoma
• NS/P
• neural stem/progenitor
• PBS‐T
• phosphate buffered saline/Tween‐20
• PGF
• placental growth factor
• PLAU / uPA
• plasminogen activator, urokinase / urokinase‐type plasminogen activator
• PLG
• angiostatin/plasminogen
• REST / NRSF
• RE1‐silencing transcription factor / neuron‐restrictive silencer factor
• SHH
• sonic hedgehog
• THBS1 / TSP1
• thrombospondin‐1
• THBS2 / TSP2
• thrombospondin‐2
• TME
• tumor microenvironment
• VEGFR1
• vascular endothelial growth factor receptor‐1
• VM
• vascular mimicry / vasculogenic mimicry
• WNT
• Wingless

Introduction

Medulloblastoma (MB) is the most common malignant brain tumor in children and frequently occurs in the cerebellum [1–3]. MBs are classified into Wingless (WNT), sonic hedgehog (SHH), Group 3, and 4 molecular subgroups [4,5]. Although patients with WNT‐driven MBs have good prognosis, subsets of patients with SHH tumors and most with Group 3/4 tumors have poor outcomes [1,6]. The underlying reasons are not well understood. Cerebellar granule neuron progenitors (CGNPs) are thought to be the cells of origin of SHH‐MB tumors [7]. The SHH signaling pathway is frequently deregulated in SHH‐driven MBs, and its activation promotes CGNP hyperproliferation [8]. This added to the lack of their terminal neuronal differentiation, which contributes to MB development [9,10].

Aberrations in chromatin remodeling are believed to drive MB tumors [11,12]. Our previous work showed elevated expression of the RE1‐silencing transcription factor (REST), a transcriptional repressor of neuronal differentiation genes, in human MB tumors, and found it to be correlated with poor patient prognosis [13,10,14,6]. REST's contribution to MB genesis was demonstrated through the generation of a novel transgenic mouse model (REST^TG), where human REST transgene could be conditionally expressed in CGNPs [10]. Compared with age‐matched wild‐type (WT) mice, REST^TG animals exhibited an expanded external granule layer (EGL), where CGNPs reside [10]. Ex vivo‐cultured CGNPs from REST^TG mice also showed poorly neurogenesis, suggesting that REST increases cell proliferation and blocks differentiation [10]. In the background of constitutive activation of SHH signaling (Ptch^+/−), REST^TG mice developed poorly differentiated tumors with 100% penetrance, accelerated kinetics of 10–90 days, and leptomeningeal dissemination when compared to Ptch^+/− mice, which highlighted a cell‐intrinsic role for the protein in tumor progression [10].

The tumor microenvironment (TME) plays an important role in tumorigenesis. Angiogenesis and vasculogenesis, which are important for the growth, progression, and metastasis of tumors, are controlled by an imbalance between pro‐ and antiangiogenic molecules that are secreted by endothelial cells, tumor cells, or other cells present in the TME [15–20]. These vessels are frequently structurally and functionally abnormal [17]. Brain tumor vasculature growth can occur through mechanisms such as co‐option, angiogenesis, vasculogenesis, vascular mimicry (VM), and tumor endothelial differentiation [21,22]. Abnormal vasculature in MBs has also been noted. For example, clusters of abnormal, thick‐walled arterial‐type vessels along with numerous variably joined small venous and capillary structures are seen in WNT‐MBs [23]. In SHH‐MBs, increased expression of proangiogenic factors has been described [24]. Functional studies have attributed a role for SHH ligand‐dependent stimulation of tumor stromal secretion of placental growth factor (PGF) and neuropilin (NRP) in SHH‐MB development [18].

Here, we describe a role for REST in the control of MB vasculature. Employing a combination of transgenic and xenograft mouse models, analyses of publicly available transcriptomic data on human MB tumors, and functional studies, we demonstrate that REST drives increased expression of proangiogenic molecules, vascular endothelial growth factor (VEGF), and PGF. In vivo, tumors in Ptch^+/−/REST^TG mice and animals bearing human MB xenografts exhibit a significant increase in the number and size of blood vessels compared with control mice. Interestingly, REST elevation is also associated with increased expression of vascular endothelial growth factor receptor‐1 (VEGFR1) and the proangiogenic transcription factor, E26 oncogene homolog 1 (ETS1), in CGNPs of REST^TG mice compared with cells from WT cerebella. Human MB tumors engineered to express REST transgene had increased expression of molecules identified as ‘VEGF pathway genes’ by RNA‐Seq analyses and colocalized with endothelial cells in vitro and in vivo, suggesting that REST elevation promotes angiogenesis‐related gene expression changes in MB cells. Our studies are the first to implicate REST, a canonical regulator of neurogenesis, in the control of MB vasculature.

Materials and methods

Cell culture

Human MB cells (DAOY, DAOY‐REST (DAOY‐R), UW426, UW426‐REST (UW426‐R), UW228, UW228‐REST (UW228‐R), and D283) and mouse v‐Myc transformed neural stem cell (C17.2) and its isogenic derivative expressing human (h) REST transgene (ST2) were cultured as described previously [10,6]. CGNPs were isolated from WT and REST^TG mice and cultured as previously outlined [10]. Human umbilical vein endothelial cells (HUVEC; Cat# CC‐2519; Lonza, Alpharetta, GA, USA) were cultured in endothelial growth medium‐2 (EGM‐2) with recommended growth factors (Cat# CC‐3162; Lonza). Human brain microvascular endothelial cells (HBMEC; Cat# HEC02) and endo growth medium (EGM; Cat# MED001) were purchased from Neuromics (Edina, MN, USA) and cultured in complete medium contained EGM‐2 and EGM (4 : 1 ratio). 293T cells were grown in Dulbecco's Modified Eagle's Medium in the absence of serum.

In vivo assays

Animal experiments and procedures were done following approval by the Institutional Animal Care and Use Committee. DAOY or DAOY‐R cells (50 ,000 cells in 3 μL) stably expressing firefly luciferase (ffluc) were implanted into cerebella of 4‐ to 6‐month‐old NOD/SCID gamma null (NSG; NOD.Cg‐Prkdcscid Il2rgtm1Wjl/SzJ) mice (The Jackson Laboratory, Bar Harbor, ME, USA), using a guide screw [25]. Tumor growth was monitored by bioluminescence imaging (BLI) using the Caliper Life Sciences IVIS Spectrum IVIS200 in vivo imaging system (Caliper LIfe Sciences, Hopkinton, MA, USA) [10]. Mice were euthanized when signs of morbidity were noted [10]. Brains were collected and sectioned for IHC analysis. REST^TG mice and Ptch^+/−/REST^TG were generated as described [10].

Immunohistochemistry

Brains from WT, REST^TG, Ptch^+/−, Ptch^+/−/REST^TG mice, and animals bearing DAOY/DAOY‐R xenografts and high‐REST and low‐REST (LR/HR) patient‐derived orthotopic xenografts (PDOX), obtained from X. Nan Li (under a material transfer agreement), were formalin‐fixed and embedded in paraffin. 4‐μm‐thick brain sections were cut and used for IHC analysis. Deparaffinized, rehydrated sections were subjected to heat‐mediated antigen retrieval in citrate buffer / Tris‐EDTA buffer, quenched, and incubated with primary antibodies to CD31 (Cat# DIA310; Dianova GmbH, Hamburg, Germany; Cat# ab28364; Abcam, Cambridge, MA, USA), VEGFR1 Cat# AMAB90703; Sigma‐Aldrich, St. Louis, MO, USA), and ETS1 (Cat# ab26096; Abcam) at 4 °C, overnight. After washing, sections were incubated with secondary antibody conjugated to horseradish peroxidase (HRP) (Cat# 115‐035‐003 and 115‐035‐006; Jackson ImmunoResearch, West Grove, PA, USA) and developed using 3,3′‐diaminobenzidine substrate (Cat# SK‐4100; Vector Laboratories, Burlingame, CA, USA). Hematoxylin and eosin (H&E) counterstaining was then done. Stained sections were visualized under a microscope (Nikon ECLIPSE E200; Melville, NY, USA) and images acquired using an Olympus SC100 camera (Waltham, MA, USA). Images were processed using CellSens Entry imaging software (Olympus Life Sciences, Waltham, MA, USA). CD31‐positive blood vessels were quantified, and statistical analyses were performed among different groups.

Immunofluorescence

Paraffin‐embedded brain sections were deparaffinized and rehydrated in a Gemini AS auto‐stainer (Thermo Fisher Scientific, Waltham, MA, USA), and washed in PBS, and antigen retrieval was performed using eBioscience™ IHC Antigen Retrieval Solution—High pH (Cat# 00‐4956‐58; Thermo Fisher Scientific, Pittsburgh, PA, USA) by heating to 98 °C for 15 min in LabVision™ PT module™ (Thermo Fisher Scientific, Pittsburgh, PA, USA). Slides were placed in distilled water, washed with 0.1% Tween‐20 in PBS (PBS‐T), blocked in 1% FBS in PBS for 1 h, and then coincubated with goat anti‐luciferase (1 : 150, Cat# NB100‐1677SS; Novus Biologicals, Centennial, CO, USA) and mouse anti‐CD31 (1 : 250, Cat# Dia‐310; DIANOVA GmbH). After washing in PBS‐T, slides were incubated with donkey anti‐goat (1 : 300, Cat# 705‐586‐147) and donkey anti‐rat (1 : 300, Cat# 712‐546‐153; Jackson ImmunoResearch) antibodies for 1 h. Washed slides were mounted in Fluorogel II mounting media and visualized under a fluorescence microscope.

In vitro angiogenesis assay

In vitro angiogenesis assay (tube assay) was performed by first placing matrigel (Cat# 354230; Thermo Fisher Scientific, Waltham; 100 μL/well) in 96‐well sterile culture plates [26]. HUVEC (5 × 10⁴ cells/25 μL) cells were mixed with endothelial medium and with conditioned medium derived from DAOY/ DAOY‐R/, UW228/ UW228‐R, UW426/UW426‐R (1 : 1 ratio), placed on the matrigel, and incubated in a CO₂ incubator at 37 °C. In other experiments, HBMECs were incubated with conditioned medium from DAOY‐R cells transduced with lentivirus expressing shRNA against ETS1 (shETS1‐1) or a nonspecific sequence (shControl). After 16 h, cells were incubated with Calcein‐AM (Cat# C3100MP; Thermo Fisher Scientific, Waltham) for 30 min and rinsed with the endothelial cell culture medium. The number of tubes formed in matrigel was determined by fluorescence microscopy and image analysis/quantification as described [26]. To distinguish between colocalized MB and endothelial cells, MB cells and HBMECs cells were loaded with cell tracker red and green (Cat#s C34552 and C2925; Invitrogen, Carlsbad, CA, USA), respectively. Cells were coincubated in matrigel for 16 h followed by fluorescence microscopy and quantification of tube formation and colocalization.

qRT‐PCR

RNA was extracted from MB cell lines using Quick‐RNA MiniPrep Kit (Cat# D4008; Zymo Research, Irvine, CA, USA). Equal amounts of RNA were reverse‐transcribed into cDNA using the iScript cDNA Synthesis Kit (Bio‐Rad, Hercules, CA, USA), and qRT‐PCR was performed in triplicate as described [10]. Relative mRNA expression, normalized to 18S ribosomal RNA, was determined by the comparative $2^{‐\mathrm{\Delta }\mathrm{\Delta }{\mathrm{C}}_t}$ method. Normalized mRNA expression was graphed as fold change compared with parental cell line.

Primer sequences are as follows:

hREST‐Forward: 5’‐GGCAGCTGCTGTGATTACCT‐3’

hREST‐Reverse: 5’‐AGTTGTTATCCCCAACCGGC‐3’

h18s‐Forward: 5’‐CGGCGACGACCCATTCGAAC‐3’

h18s‐Reverse: 5’GAATCGAACCCTGATTCCCCGTC‐3’

Western blot analysis

Cell lysates from human and mouse MB cells and CGNPs from WT and REST^TG mice brains were prepared in EBC lysis buffer [26]. Samples were subjected to SDS/PAGE and western blot analyses with the following primary antibodies: REST (Cat# 07579; Millipore, Billerica, MA, USA), VEGFR1 (Cat# 2479), ETS1 and alpha‐tubulin (Cat#s 14069 and 9099, respectively; Cell Signaling Technology, Danvers, MA, USA), and beta‐actin (Cat# ab40742; Abcam). After washing and incubation with the corresponding HRP‐conjugated secondary antibodies (Jackson ImmunoResearch), membranes were developed using SuperSignal (Cat# 34075 and Cat# 34087; Thermo‐Scientific, Waltham, MA) followed by autoradiography.

Lentivirus production and transduction

Lentiviral constructs expressing GFP and shRNAs against ETS1 were purchased from institutional shRNA and ORFeome Core. Lentiviral particles were prepared by cotransfection of 293T cells with plasmids pax2 and MD2 using OptiMEM®I and Lipofectamine® 2000 Reagent (Cat#s 31985‐062 and 11668‐019, Thermo Fisher Scientific, Waltham). DAOY‐R cells were transduced with shRNA control or shETS1‐expressing lentivirus for 72 h, and GFP‐positive cells were sorted by flow cytometry. ETS1 knockdown was confirmed by western blotting, and cells were used in in vitro angiogenesis assays. Nucleotide sequences for the shRNAs are as follows:

shETS1‐1: 5′‐CTTGATATGGTTTCACATC‐3′

shETS1‐2: 5′‐TAATTGATACCCGGCCCTG‐3′

shControl: 5′‐ATCTCGCTTGGGCGAGAGT‐3′

Human angiogenesis proteome profiler assay

Proteome profiler human angiogenesis array (Cat# ARY007; R&D Systems Minneapolis, MN, USA) was used to measure levels of pro‐ and antiangiogenic molecules using conditioned media from DAOY, UW228, UW426/UW426‐R, and D283 cells. Quantification was done by imagej analysis (https://imageJ.nih.gov).

RNA‐Seq analyses

The quality of sequencing reads was evaluated using NGS QC Toolkit (v2.3.3) [27], and high‐quality reads were extracted. The human reference genome sequence and gene annotation data, hg38, were downloaded from Illumina iGenomes. (https://support.illumina.com/sequencing/sequencing_software/igenome.html). The quality of RNA‐sequencing libraries was estimated by [28] mapping the reads onto human transcript and ribosomal RNA sequences (Ensembl release 89) using Bowtie (v2.3.2). STAR (v2.5.2b) [29] was employed to align the reads onto the human and viral genomes. SAMtools (v1.9) [30] was employed to sort the alignments, and HTSeq Python package was employed to count reads per gene [31]. DESeq2 R Bioconductor package was used to normalize read counts and identify differentially expressed (DE) genes [31,32]. KEGG [33] pathway data were downloaded using KEGG API (https://www.kegg.jp/kegg/rest/keggapi.html), and gene ontology (GO) data were downloaded from NCBI FTP (ftp://ftp.ncbi.nlm.nih.gov/gene/DATA/gene2go.gz). The enrichment of DE genes to pathways and GOs was calculated by Fisher's exact test in R statistical package.

Gene expression profile in MB patient samples and MB cell lines

Microarray data sets containing gene expression values of MB tumors were obtained from Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo). The GSE85217 data set, which contains Affymetrix Human Gene 1.1 ST Array profiling of 763 primary MB samples, was used to evaluate gene expression. Microarray data were normalized using the robust multiarray average method. The expression data for each gene were Z‐score‐transformed. Hierarchical clustering based on the expression of neuronal differentiation markers divided the 223 SHH‐type MB patient samples into six distinct clusters (Clusters 1–6) as described previously [10]. We also analyzed the publically available data sets (GSE86574, GSE107405, GSE37382, GSE50765) in GEO and from the data set provided by Cho et al., in R2: Genomics Analysis and Visualization Platform (hgserver1.amc.nl/cgi‐bin/r2/main.cgi) for gene expression analysis [34–37].

Statistical analysis

The experimental data reported are mean ± SD of a minimum of three samples. P‐value of < 0.05 was considered to be statistically significant. P‐values for comparisons between every pairwise combination among clusters (1–6) based on gene expression status were obtained using the unpaired t‐test with Welch's correction using the graphpad prism version 7.0 (GraphPad, San Diego, CA, USA). 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). Student's t‐test and ANOVA were performed for significance between groups.

Results

REST^TG mice exhibit increased cerebellar vasculature

We had previously shown that conditional REST elevation in CGNPs caused an abnormal expansion of the cerebellar EGL in REST^TG mice compared with WT animals [10]. A more careful examination of H&E‐stained sections revealed an increased presence of vascular structures in the cerebella of REST^TG mice compared with WT mice (Figs 1A and S1A). These observations were confirmed by IHC, which revealed a twofold increase in CD31‐positive vessels in the cerebella of REST^TG mice relative to that in WT animals (Fig. 1A). A significant increase in lumen diameter and branching was also seen in the cerebella of REST^TG mice compared with WT cerebella (Figs 1A and S1A). These findings suggest a REST‐dependent increase in cerebellar vasculature.

REST elevation drives tumor vasculature

A role for REST in the progression of SHH‐driven MBs was first described by our previous work, where we showed that in Ptch^+/− mice with constitutive activation of SHH signaling, REST elevation (Ptch^+/−/REST^TG) promoted tumors with 100% penetrance, accelerated kinetics of 10–90 days, and leptomeningeal dissemination [10]. To determine whether REST elevation also contributed to modulation of tumor vasculature, IHC assessment of CD31 staining of tumor‐bearing cerebella of Ptch^+/− and Ptch^+/−/REST^TG mice was performed. While H&E staining showed larger and more infiltrative tumors in Ptch^+/−/REST^TG mice in contrast to Ptch^+/− mice, CD31 staining confirmed a twofold increase in blood vessels in Ptch^+/−/REST^TG tumors compared with Ptch^+/− tumors (Figs 1B and S1B). Once again, Ptch^+/−/REST^TG tumors exhibited a demonstrable increase in the number of vessels and vessel diameter relative to tumors in Ptch^+/− mice (Figs 1B and S1B).

We also validated our findings from genetically engineered mice in human MB cells. As a first step, we performed RNA‐Seq analyses of three commonly used MB cell lines, DAOY, UW228, and UW426, to compare their gene expression profile with that of published transcriptomic data (GSE86574) from normal cerebellum, human SHH, Group 3, and Group 4 MB tumors, as well as ONS76 MB cell line. Using data from GSE86574, we performed hierarchical clustering analysis of expression of genes involved in hedgehog pathway. DAOY cells clustered with SHH‐MBs, indicating significant similarity in their expression profiles with respect to hedgehog pathway markers with that of SHH‐MBs (Fig. 2A). In addition, we also observed clustering of DAOY cells with SHH‐MBs using subtype‐specific marker genes used for NanoString analyses [38,39] (Fig. S2A,B). These 22 subtype‐specific genes and 30 hedgehog marker genes were sufficient to divide a cohort of 763 MB patient tumors into the four MB subgroups (Fig. S2C,D) [4]. In addition, a recent study showed the expression profile of DAOY cells to be similar to that of SHH‐MB patient tumors by hierarchical clustering assay using a 22 genes NanoString panel [40]. These data provide direct transcriptomic proof that DAOY cells are representative of SHH‐MBs. We also analyzed the expression profiles of the above 22 subtype‐specific genes in another published data set (GSE107405) and in our RNA‐Seq data, to show that DAOY, UW228 and UW426 cells clustered separately from cell lines derived from Group 3 or Group 4 MB patients (Fig. S2E,F) [35]. With respect to most of 22 subtype‐specific marker genes, these MB cell lines showed‐ expression patterns that were similar to SHH‐MB subtype patient tumors (Fig. S2G–J). Collectively, these results confirm that DAOY, UW228, and UW426 cells retain features of SHH‐MBs.

To study the effect of constitutive REST expression on the biology of DAOY, UW228, and UW426 cells, and specifically with respect to genes involved in vascular development, we generated LR/HR isogenic pairs of the three MB cell lines and performed RNA‐sequencing analysis. Interestingly, we observed that all three isogenic pairs of MB cell lines retain their overall gene expression landscape even after overexpressing REST (Fig. 2B,C), suggesting that altered REST expression influences a restricted number of biologically relevant genes rather than creating global expression changes. To further functionally characterize these genes whose expression is modulated by REST elevation, we conducted pathway enrichment analysis and defined KEGG pathways that are enriched for REST‐driven gene expression changes (Fig. 2D). These enriched genes defined pathways with roles in cancer development (Fig. 2E), hedgehog signaling (Fig. 2F), cell cycle regulation (Fig. 2G), VEGF signaling (Fig. 2H), hippo signaling (Fig. S3A), and MAPK signaling (Fig. S3B).

Based on our data described in Figs 1 and 2, we further investigated the possibility that REST elevation contributes to modulation of vasculature. For this, DAOY and DAOY‐REST (DAOY‐R) cell lines engineered to express firefly luciferase (ffluc) were injected into the cerebellum of NOD/SCID mice (n = 5, each) and tumor growth was monitored by BLI. REST expression in these cells was confirmed by RT‐PCR and Western blotting (Fig. 3A,B). Although both DAOY and DAOY‐R cells formed tumors, the latter grew more rapidly and formed larger tumor masses at the time of euthanasia (72 days) (Figs 3C and S4). Brains were harvested from both cohort of animals, sectioned, and studied by H&E staining, and IHC using anti‐CD31 antibody, to identify tumors and vasculature, respectively (Figs 3D and S5A). Quantitation of CD31‐positive structures showed a twofold increase in the number and diameter of vessels in DAOY‐R tumors compared with DAOY tumors (Fig. 3D, right panel). Similar differences in vasculature were also noted in sections of mice brains bearing HR‐ and LR‐PDOX tumors (Figs 3E and S5B).

CD31 expression was also quantified in a publicly available transcriptome database (GSE85217) of human SHH‐MB samples [4]. As seen in Fig. 3F, CD31 gene expression was higher in SHH‐α, β, γ tumor samples compared with SHH‐δ tumors. CD31 and REST expression showed a strong overall positive correlation (r = 0.33, P < 0.0001) in SHH subgroup of MB samples and in SHH‐β (r = 0.36, P = 0.04), γ (r = 0.40, P = 0.005) subtypes (Figs 3G and Fig S10A). SHH‐MBs were also divided into six clusters, based on the expression of REST target neuronal differentiation genes, where a significant increase in REST mRNA levels was seen in clusters 1, 2 (SHH‐α), and 5 (SHH‐β) tumors [10]. The pattern of CD31 expression paralleled that of REST, with higher levels seen in clusters 1, 2, and 4, relative to clusters 3, 4, and 6 (Fig. 3H). These findings were confirmed using GSE37382 and GSE50765 data sets [37]. Finally, comparisons of tumor samples with normal cerebella made using GSE data sets revealed upregulation of CD31 in MB samples and a positive correlation with REST expression in SHH‐MBs (Fig. S6A–C). Thus, the above observations indicate that REST supports tumor vasculature in mice and its expression in subsets of SHH tumors is strongly correlated with the expression of CD31 (Fig. S10A–D).

REST elevation promotes secretion of proangiogenic factors in MBs

To investigate whether REST expression modulated secreted pro‐ or antiangiogenic factors, we incubated a Proteome Profiler Human Angiogenesis Array membrane with antibodies that can detect the presence of up to 55‐ human angiogenesis‐related secreted proteins‐ in conditioned media from DAOY, UW228, UW426, and D283 cells. Densitometry was used to quantitate levels of these secreted proteins (Figs 4A and S7A; Table S1). DAOY and UW228 cells expressed proangiogenic molecules, angiogenin (ANG), angiopoietin‐1 (ANGPT1), C‐X‐C motif chemokine ligand 16 (CXCL16), IL‐8 (CXCL8), MCP1 (CCL2), PLGF (PGF), uPA (PLAU), and VEGF. The levels of ANG, ANGPT1, CXCL16, CCL2, and PGF were higher in DAOY than in UW228 cells and paralleled REST levels in these cells (Fig. 4A) [10]. The above proangiogenic molecules, except ANGPT1, CCL2, and PLAU, were also expressed in UW426 cells (Fig. 4A). ANGPT1, CXCL8, PGF, and PLAU were also detected in D283 cells, a Group 3/4 MB cell line (Fig. S7A). Antiangiogenic molecules angiopoietin‐2 (ANGPT2), angiostatin‐2 (PLG), and thrombospondin‐2 (TSP2/THBS2) were not detected in DAOY, UW228, and UW426 cells (Fig. 4A). REST dependency for the above changes was established by comparing the secretome of UW426 and UW426‐REST cells, which confirmed an increase in the levels of the proangiogenic molecules and ligands for VEGFR1 VEGF and PGF under conditions of REST elevation (Figs 4A and S7B).

Gene expression microarray data derived from SHH‐MBs also confirmed these findings [4]. SHH‐MB samples were divided into two groups (117 HR and 106 LR tumors) based on the average Z‐score of their REST expression. The expression of 410 angiogenesis‐related genes listed in Table S2 was studied between these two cohorts using a volcano plot, and genes with significantly differential expression were identified (P < 0.05) (Fig. 4B). Of these, 136 genes showed higher expression and 46 genes showed lower expression in samples with higher REST levels (Table S3). Of the 11 proteins examined in Fig. 4A, genes encoding ANG, ANGPT2, PGF, THBS2, CD31, VEGFR1, and ETS1 had significantly higher expression in HR MBs (Fig. 4B).

Finally, in vitro angiogenesis/tube formation assay was carried out using conditioned media from isogenic pairs of DAOY/DAOY‐R, UW228/UW228‐R, and UW426/UW426‐R cells. HUVECs labeled with cell tracker green dye were placed on matrigel and incubated with the conditioned media from the above isogenic pairs of cells to monitor tube formation. Fluorescence microscopy showed that conditioned medium from DAOY and DAOY‐R cells supported a twofold and threefold increase in tube formation, respectively, relative to control HUVECs cultivated in unconditioned growth medium (Fig. 4C). Similar REST‐dependent increases in tube formation were noted with UW228/UW228‐R and UW426/UW426‐R cells (Fig. 4C). Together, these data support a role for REST elevation in promoting angiogenesis in vitro.

REST elevation drives ETS1‐dependent increase in VEGFR1 expression

We next asked whether the levels of VEGFR1, a cognate receptor for PGF and VEGF, are modulated in a REST‐dependent manner. IHC showed that tumors in Ptch^+/−/REST^TG mice and animals with HR‐PDOX had higher VEGFR1 expression compared with Ptch^+/− and LR‐PDOX, respectively (Figs 5A,B and S8A,B). Western blotting of cell lysates from DAOY/DAOY‐R and UW426/UW426‐R cell pairs showed a clear enhancement of VEGFR1 levels in the higher REST context compared with the parental cells (Fig. 5C). VEGFR1 levels were also similarly increased in ST2 cells with constitutive hREST expression, relative to parental C17.2 cells (Fig. 5D, left panel). Likewise, relative to WT CGNPs, cells from REST^TG mice exhibited higher VEGFR1 protein levels (Fig. 5D, right panel). Tubulin served as a loading control for these assays (Fig. 5C,D).

Then, a role for REST in tube formation was confirmed by co‐incubating cell tracker green‐labeled HBMCs with cell tracker red‐labeled DAOY/DAOY‐R cells. Surprisingly, REST elevation was associated with a significant increase (threefold) in the co‐localization of MB cells and endothelial cells (Fig. 5E). Indeed, co‐immunofluorescence staining of ffluc‐expressing DAOY and DAOY‐R tumor sections from NSG mice, with anti‐CD31 and anti‐luciferase antibodies, also revealed a fourfold increase in REST‐dependent colocalization (yellow) of tumor cells (green) with endothelial cells (red) (Fig. 5F). These findings raise the possibility that REST elevation in MB cells could lead to an endothelial cell‐like phenotype or VM, a phenomenon in which cancer cells form blood vessels independent of, or in association with endothelial cells in tumors [41,42].

Like CD31, VEGFR1 gene expression was significantly higher in SHH‐α, SHH–β, SHH–γ MBs compared with SHH‐δ tumors, and VEGFR1 and REST expression showed a strong overall positive correlation (r = 0.29, P < 0.0001) in SHH subgroup of MB samples and in SHH‐γ (r = 0.30, P = 0.04) and SHH‐δ subtypes (r = 0.23, P = 0.047) (Figs 5G,H and S10E). A trend toward significance was noted in SHH‐α (r = 0.22, P = 0.07) and SHH‐β (r = 0.31, P = 0.07) (Fig. S10E). Higher VEGFR1 expression was observed in clusters 1, 2, 5, and 6 compared with clusters 3 and 4 in the differentiation‐based grouping of tumor samples (Fig. 5I). Among WNT, Group 3, and Group 4 MB tumor samples, a positive correlation between REST and VEGFR1 was detected when all Group 4 tumors were collectively considered (r = 0.16, P = 0.005), but a statistically significant correlation could not be detected in the individual subtypes of Group 4 tumors (Figs 5G,H and S10E–H). These results indicate that VM may be unique to SHH‐MBs, although the limited availability of subgroup/subtype information in other patient tumor data sets precluded further investigation of this observation (Fig. S8C,D).

If MBs with REST elevation do indeed mimic endothelial cells, we reasoned that transcription factors directing endothelial specification such as ETS1 may also be expressed in tumor cells [43]. Again, IHC confirmed that Ptch^+/−/REST^TG mice and HR‐PDOX brain sections had higher ETS1 expression relative to Ptch^+/−, and LR‐PDOX samples, respectively (Figs 6A,B and S9A,B). Western blotting performed with lysates from WT/REST^TG CGNPs, as well as DAOY/DAOY‐R and UW426/UW426‐R cells, showed an increase in ETS1 levels, which paralleled REST levels, showing that ETS1 expression was REST‐dependent (Fig. 6C,D). The above findings taken in conjunction with VEGFR1 being a known ETS1 target gene, led us to assess whether REST‐dependent increase in VEGFR1 expression in DAOY‐R cells is mediated by ETS1 [44,45]. Knockdown experiments showed that reduction of ETS1 in DAOY‐R cells using two different shRNA‐ETS1 constructs did indeed result in a decrease in VEGFR1 levels in these cells (Fig. 6E). REST levels remained unaffected, relative to actin controls. In vitro angiogenesis assays showed that a reduction in ETS1 expression in DAOY‐R caused a significant decline in tube formation, as well as a decrease in tumor endothelial cell colocalization at the tubes, confirming the involvement of ETS1 in this process under conditions of REST elevation (Fig. 6F,G).

ETS1 expression was also studied using bulk transcriptomic data from human SHH‐MB samples described in Figs 2–4. First, ETS1 expression was significantly higher in SHH‐α, SHH–β, SHH–γ tumor samples compared with SHH‐δ tumors (Fig. 6H). ETS1 and REST expression showed a strong overall positive correlation (r = 0.37, P < 0.0001) in SHH subgroup of MB samples and in SHH‐β (r = 0.40, P = 0.02), SHH–γ (r = 0.43, P = 0.003), and SHH–δ (r = 0.32, P = 0.005) subtypes (Figs 6I and S10I). When SHH‐MBs were divided into the six neurogenesis‐based clusters, a significant increase in ETS1 mRNA levels was seen in clusters 1, 2, 5, and 6, relative to clusters 3 and 4 (Fig. 6J). Among WNT, Group 3, and Group 4 MB tumor samples, a positive correlation between REST and ETS1 was detected in Group 4 tumors (r = 0.31, P < 0.0001), with a significant correlation seen in Group 4 α (r = 0.38, P < 0.0001), β (r = 0.27, P = 0.005) and γ (r = 0.35, P < 0.0001) tumor subtypes (Fig. S10J‐L). Collectively, the above data suggest that REST‐dependent modulation of tumor vasculature is ETS1‐dependent, with a positive association between REST and ETS1 expression seen in subsets of SHH and Group 4 tumors (Fig. S10I‐L). We also detected positive correlations between REST and ETS1 expression in GSE50765 and GSE37382 [37] sample sets, respectively, but not in the data set provided by Cho et al. [36], likely due to the limitations and/or differences of probe sets and sample numbers among patient data sets (Fig. S9C,D). Overall, our data suggest that REST contributes to MB vasculature through cell‐extrinsic and cell‐intrinsic mechanisms (Fig. 7).

Discussion

REST is a canonical regulator of neurogenesis and plays a key role during normal brain development. It is this aspect of REST function that has been most widely studied and reported in the literature [46,14]. REST binds to the RE1 sequence found in the regulatory regions of many neuronal genes to silence their expression. REST controls neural development by regulating neural lineage specification. It promotes neural stem/progenitor (NS/P) self‐renewal while restricting their maturation into neurons [47]. REST shows differential expression during neural development, with its levels being highest in embryonic stem cells (ES) and gradually declining thereafter as cells transition through NS/P cells into mature neurons [47]. However, its expression is maintained in cells destined for glial specification, suggesting that REST levels may dictate other neural lineage choices [48,46]. Although most studies have focused on its function in neural cells, genome‐wide chromatin occupancy studies have identified a reasonable number of potential REST target genes, which are not involved in neural development [49]. Abnormal REST activity is implicated in the genesis of many neural cancers including MB, glioblastoma, diffuse intrinsic pontine glioma (DIPG) and neuroblastoma [10,50,51,26,6]. Its aberrant expression in these cancers has been associated with poor patient survival [51,26,6]. However, these studies were also mostly focused on the‐cell‐intrinsic functions of REST and attributed roles for the protein in the control of cell proliferation and/or blockade of neural lineage specification during tumor development [10].

The TME, which encompasses the vascular network, stromal cells, immune cells, extracellular matrix, and fibroblasts, plays a key role in tumor growth and progression [52]. Cell–cell communication between tumor cells and TME also influence tumor response to therapies [53]. For these reasons, the crosstalk between cancer cells and the TME has been a subject of intense research in many cancers. However, similar studies in pediatric brain cancers have been quite limited and mostly restricted to the study of vascular networks and their role in tumor progression and metastasis [17,54,18]. As stated above, although computational studies have suggested that the REST network may include genes implicated in modulation of the TME, very few follow‐up functional studies have been conducted. REST has been shown to control pericyte biology in‐ Ewing's sarcoma, a primitive neuro‐ectodermal tumor that occurs mostly in adolescents [55,56]. Our group was ‐ the first to demonstrate a role for the REST‐gremlin axis in controlling the vasculature of DIPG tumors [26]. In the current study, we provide the first demonstration ‐ that REST elevation controls MB vasculature. Indeed, RNAseq analyses‐ showed changes in hippo and MAPK signaling and confirmed our previous findings that REST elevation drives cell proliferation and represses PTCH expression in the more immature SHH‐α MBs, and surprisingly in the more differentiated SHH‐β tumors [57,10]. We also show that REST controls endothelial cell biology and MB vasculature in part by paracrine mechanisms. VEGF, VEGF165, PDGFA, VEGF121, Ang‐1 (ANGPT1), Ang‐2 (ANGPT2), VEGFC, TGFA, VEGF189, and VEGFB were some of the proangiogenic molecules previously described in MB tumors [24,58]. Here, we found PGF, ANG, ANGPT1, CXCL8, and CXCL16 to be secreted by MB cell lines. PGF produced by the cerebellar stroma in SHH tumors signals through neuropilin‐1 and promotes MB cell survival [18]. Although levels of antiangiogenic molecules were significantly lower in MB cell lines, this was not recapitulated in human MB samples. Our work is also the first to suggest a role for ETS1 in REST‐dependent angiogenesis in SHH‐MBs. Most importantly, our findings suggest that REST‐driven modulation of tumor vasculature may contribute to the increased incidence of metastasis and poor survival in patients with SHH‐α and SHH‐β MBs. ETS1 is a transcription factor and is a known regulator of angiogenic growth factors such as VEGFR1 [44,59,45]. Given the strong correlative parallels in REST and ETS1/CD31/VEGFR1 expression between SHH‐MBs and Group 4 MBs, similar mechanisms could be operational in these two subgroups of MBs. However, this needs further evaluation.

In addition to angiogenesis, brain tumors utilize other mechanisms to acquire new blood vessels, including co‐option, vasculogenesis, and intussusception [17]. Plasticity of cancer cells enables them to mimic endothelial cells, thus leading to the formation of vessels [60]. This has been described in glioblastomas where stem‐like cells were found to differentiate into endothelial cells, and harbored the same genomic alterations as cancer cells [41]. In a study by Wang et al. [60] ~ 22% of MB tumors (n = 41) were shown to exhibit VM and was also associated with poorer clinical outcomes. Thus, there is support for VM in MBs, although mechanisms have not been defined. Our data suggest that REST elevation in tumor cells may promote VM by driving VEGFR1 expression and possibly activating the protein kinase C alpha pathway [42] (Fig. 7). However, the REST‐VM connection needs to be further investigated.

Antiangiogenic therapies have been under consideration for pediatric and adult brain tumors [61]. For recurrent glioblastoma multiforme (GBM), the median overall survival was 8.63 months for patients treated with bevacizumab, an anti‐VEGF antibody, and 8.91 months when bevacizumab was combined with irinotecan, a chemotherapeutic agent [62] Thus, this study showed bevacizumab alone is beneficial for GBM. Interestingly, recent clinical trials have also demonstrated the efficacy of bevacizumab for the treatment of recurrent MB when combined with chemotherapeutic agents temozolomide and irinotecan or with stereotactic radiosurgery [63,64]. Despite this promise, clinical use of antiangiogenic agents has not evolved [65,19]. The development of resistance to anti‐VEGF therapies could be an underlying reason [66]. VM may be yet another cause since angiogenesis inhibitors appear to block the formation of vessels by endothelial cells, but not those originating from tumor cells [67]. Therefore, targeting drivers of VM such as REST or ETS1 may alleviate resistance to conventional angiogenesis inhibitors and need to be further evaluated in preclinical studies. Our previous preclinical studies have demonstrated the feasibility of targeting REST activity through inhibition of associated chromatin remodeling enzymes—G9a/GLP, LSD1, and HDAC1/2 [13,68,14]. However, their effect on tumor vasculature was not studied. Downregulation of ETS proteins is correlated with regression of hyaloid vessel endothelial cells [69]. In newborn mice, administration of YK‐4‐279, an inhibitor of ETS and ETS‐related gene activity, decreased the number of hyaloid vessels [69]. YK‐4‐279 was also been shown to reduce tube formation by HUVECs in vitro, in a VEGFR1‐dependent manner [69]. Targeting ETS1 for proteolysis, by inhibiting the activity of its deubiquitylase USP9X, may be another interesting strategy, which should be explored [70]. Overall, antiangiogenesis approaches remain under‐investigated for brain tumor therapy.

Conclusion

The current study is the first to attribute a role for REST in the regulation of MB vasculature. We have provided evidence that its elevation promotes increased secretion of pro‐angiogenic factors, which allows vascular growth. In addition, MB cells with elevated REST expression display molecular and functional features of endothelial cells, suggesting that REST may alter cell fate decisions in MBs by modulating the expression of transcription factors that control angiogenesis, although mechanistic details remain to be delineated (Fig. 7). Targeting REST and ETS1 for the therapeutic modulation of tumor angiogenesis is a topic for future studies.

Acknowledgements

We would like to thank members of the Gopalakrishnan lab and Dr. Eugenie Kleinerman for helpful discussions. This work was supported by grants from the NIH (5R01‐NS‐079715‐01 and 5R03NS077021‐01), the American Cancer Society (RSG‐09‐273‐01‐DDC), the Cancer Prevention Research Institute of Texas (CPRIT‐RP150301), Addi’s Faith Foundation and the Rally Foundation for Childhood Cancers to V.G. RNA‐Seq library preparation and sequencing were conducted by the MD Anderson Cancer Center Science Park Next Generation Sequencing Core, supported by CPRIT Core Facility grant (RP170002).

Conflict of interest

The authors declare no conflict of interest.

Data Accessibility

The public data sets for gene expression analysis are available in GEO (https://www.ncbi.nlm.nih.gov/geo/) or R2: Genomics Analysis and Visualization Platform (https://hgserver1.amc.nl/cgi‐bin/r2/main.cgi).

Data Availability

The RNA‐Seq data of MB cell lines described in this study has been deposited in NCBI Gene Expression Omnibus (GEO) with the accession number GSE164887.

Author contributions

SS and SM were involved in conceptualization and performance of in vitro and in vivo experiments, data analysis, and generation of the manuscript. AH, JS, TD, and AS were involved in in vivo xenograft experiments, immunofluorescence, and BLI. YY was involved in in vitro studies, cell transfection, and microscopy. KS provided guidance on angiogenesis assays. AH, FBW, LX, and XX performed analyses of the RNA‐Seq data. XNL provided reagents. VK provided guidance on imaging analyses and manuscript review. VG was involved in study conceptualization, experimental design, data analysis, funding support, and writing of the manuscript and overall supervision of the project.

Consent for publication

Consent was obtained from all authors for publication.