Genome wide ChIP-seq analysis of EZH2 mediated H3K27me3

Received: May 23, 2016; Revised: October 14, 2016; Accepted: November 30, 2016

© The Author 2017. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected].

Carcinogenesis, 2017, Vol. 38, No. 2, 152–161

doi:10.1093/carcin/bgw126Advance Access publication January 5, 2017Original Manuscript

152

original manuscript

Genome-wide ChIP-seq analysis of EZH2-mediated H3K27me3 target gene profile highlights differences between low- and high-grade astrocytic tumorsVikas Sharma1, Prit Benny Malgulwar1, Suvendu Purkait1, Vikas Patil2, Pankaj Pathak1, Rahul Agrawal3, Ritu Kulshreshtha3, Supriya Mallick4, Pramod Kumar Julka4, Ashish Suri5, Bhawani Shankar Sharma5, Vaishali Suri1, Mehar Chand Sharma1 and Chitra Sarkar1,*1Department of Pathology, All India Institute of Medical Sciences, New Delhi 110029, India, 2Department of Microbiology and Cell Biology, Indian Institute of Science, Bengaluru 560012, Karnataka, India, 3Department of Biochemical Engineering and Biotechnology, Indian Institute of Technology Delhi, New Delhi 110016, India, 4Department of Radiation Oncology and 5Department of Neurosurgery, All India Institute of Medical Sciences, New Delhi 110029, India

*To whom correspondence should be addressed. Tel: +91-11-26593371; Fax: +91-11-26588663; Email: [email protected]

Abstract

Enhancer of zeste homolog-2(EZH2) is a key epigenetic regulator that functions as oncogene and also known for inducing altered trimethylation of histone at lysine-27 (H3K27me3) mark in various tumors. However, H3K27me3 targets and their precise relationship with gene expression are largely unknown in astrocytic tumors. In this study, we checked EZH2 messenger RNA and protein expression in 90 astrocytic tumors of different grades using quantitative PCR and immunohistochemistry, respectively. Further, genome-wide ChIP-seq analysis for H3K27me3 modification was also performed on 11 glioblastomas (GBMs) and 2 diffuse astrocytoma (DA) samples. Our results showed EZH2 to be highly overexpressed in astrocytic tumors with a significant positive correlation with grade. Interestingly, ChIP-seq mapping revealed distinct differences in genes and pathways targeted by these H3K27me3 modifications between GBM versus DA. Neuroactive ligand receptor pathway was found most enriched in GBM (P = 9.4 × 10−25), whereas DA were found to be enriched in metabolic pathways. Also, GBM showed a higher enrichment of H3K27me3 targets reported in embryonic stem cells and glioma stem cells as compared with DAs. Our results show majority of these H3K27me3 target genes were downregulated, not only due to H3K27me3 modification but also due to concomitant DNA methylation. Further, H3K27me3 modification-associated gene silencing was not restricted to promoter but also present in gene body and transcription start site regions. To the best of our knowledge, this is the first high-resolution genome-wide mapping of H3K27me3 modification in adult astrocytic primary tissue samples of human, highlighting the differences between grades. Interestingly, we identified SLC25A23 as important target of H3K27me3 modification, which was downregulated in GBM and its low expression was associated with poor prognosis in GBMs.

IntroductionGlioblastoma (GBM) is the most common malignant primary brain tumor in humans representing over 70% of all brain malignancies with a median survival time of ~14 months (1). Besides known genetic alterations like mutations in isocitrate

dehydrogenase 1 (IDH1), TP53 and telomerase reverse tran-scriptase (TERT), amplification of epidermal growth factor receptor (EGFR) and platelet-derived growth factor receptor (PDGFR) and the sole epigenetic alteration, namely promoter

Downloaded from https://academic.oup.com/carcin/article-abstract/38/2/152/2709441by gueston 14 February 2018

mailto:[email protected]?subject=

V.Sharma et al. | 153

Carcinogenesis, 2017, Vol. 38, No. 2, 152–161

doi:10.1093/carcin/bgw126Advance Access publication January 5, 2017Original Manuscript

hypermethylation of O-6-methylguanine–DNA methyltrans-ferase, the overall outcome in the clinical setup for the patients still remains elusive and poor (2). Elucidation of underlying mechanisms that potentiate the aggressive behavior of GBM is important for developing new prognostic markers and thera-peutic strategies. Polycomb group of genes are epigenetic regu-lators that can function as transcriptional regulators of specific sets of genes through chromatin modification (3). Enhancer of zeste homolog 2 (EZH2) is the catalytic subunit of the polycomb repressive complex 2 (PRC2), which mediates silencing of the target genes that are involved in fundamental cellular processes via trimethylation of histone H3 on lysine 27 (H3K27me3), via their catalytic SET domain activity (4). It establishes a repressive chromatin structure that ultimately contributes to the regula-tion of development and lineage commitment of stem cells as well as tumor progression. Increased expression of EZH2 has recently been described in various hematological and solid malignancies including breast, prostate (5), urinary bladder (6), leukemia (7), gastric, lung and hepatocellular carcinoma (8). In addition, overexpression of EZH2 has been found to be associ-ated with aggressive behavior and poor clinical outcome in pros-tate (8,9), breast and urinary bladder carcinomas (4). Knockdown of EZH2 in prostate cancer cells has been found to be associ-ated with growth arrest and reduced metastatic potential in vivo (9,10).

H3K27me3 was initially reported to regulate homeotic (HOX) gene expression and in early steps of X chromosome inactiva-tion (11). Literature suggests involvement of H3K27me3 in vari-ous vital processes including transcription, replication, DNA repair and modulating wide range of cellular processes involved in deciding cell fate decisions (12,13). Recent studies have also highlighted the association of H3K27me3 modifications with a large number of human malignancies (8,14). Low H3K27me3 levels have been associated with poor outcome in breast, ovar-ian and pancreatic cancers (15). On the other hand, high levels have been associated with poor outcome in hepatocellular car-cinomas (16) and esophageal squamous cell carcinomas (17). In one important study, Velichutina et al. (18) using ChIP-on chip technology(chromatin immunoprecipitation combined with DNA microarray) reported H3K27me3-mediated silencing of cell cycle-related tumor suppressor genes in diffuse large B-cell lym-phoma. In another study, Ke et al. (19) using H3K27me3 ChIP-on chip and microarray gene expression data in prostate cancer cell lines (EP156T and PC3) highlighted its role in distinct bio-logical functions (like multicellular organismal process, system development, cell–cell communication and signaling). Thus, it is apparent from the literature that H3K27me3 modifications control several cancer-related phenomena. Interestingly, new epigenetic regulators involved in maintenance of H3K27me3 mark-like demethylases JMJD3 and UTX (20) has been reported.

However, there are limited reports available for these genes in few tumors like urothelial (21), colon (22) and breast (23) and the detail mechanistic role and outcome of these enzymes (UTX/KDM6A, JMJD3/KDM6B) are not explored in detail. Thus, these molecules were not considered for the analysis in the present study in correlation with H3K27me3.

In gliomas, few genome-wide binding profiles available for H3K27me3 have been generated in cell lines (24–26). Although these studies provide information about global H3K27me3 map, they might not be optimal when investigating many important H3K27me3 modifications existing in human tumor samples (being restricted to cell line population only). Bender et al. (27) generated H3K27me3 map by utilizing four pediatric high-grade glioma tissues (two H3.3 mutants and two wild-types). However, as the authors focused only on H3K27me3 target genes in pedi-atric cases to discriminate variations in H3.3 K27 mutant tumors versus wild-type, this data cannot be applied to diverse group of adult glioma cases. Analyzing H3K27me3 target genes only in two cases may not represent the tumor heterogeneity observed in defining H3K27me3 modifications in GBM cases and other high-grade glioma populations. Thus, it is warranted to evaluate the global status of H3K27me3 target genes in high-grade glio-mas (GBMs) and correlate these profile changes with respect to normal brain as well as low-grade gliomas [diffuse astrocytoma (DAs)] in adults.

Further, no study till date has analyzed H3K27me3 profile in primary human tumor samples of low- and high-grade gli-omas, especially in relation to EZH2 expression. Hence, in the present study, we first evaluated EZH2 messenger RNA (mRNA) and protein expression in different grades of gliomas. Next, we analyzed and compared EZH2/polycomb complex-mediated H3K27me3 target genes in GBMs and DAs using ChIP followed by next-generation sequencing (ChIP-seq). Although this study specifically tries to identify distinct H3K27me3 modification-mediated epigenetic profile in low- and high-grade astrocytic tumors, it does not aim to distinguish between biologically vari-ous forms of GBM subtypes on the basis of molecular alterations (like IDH1, TP53, TERT mutation or EGFR/PDGFR amplification).Target genes so obtained in GBMs were further compared with H3K27me3 modification profile reported previously in non-neo-plastic brain, embryonic stem cells (ESCs) and glioma stem cells (GSCs) profile. H3K27me3 target genes were also analyzed for expression stats using The Cancer Genome Atlas (TCGA) data as well as with reported DNA methylation profiles. In addition to neuroactive pathways, SLC25A23, a solute carrier molecule was identified as an important H3K27me3 target, whose low expres-sion was associated with poor survival as compared with high expression, both in our cohort and in TCGA. To the best of our knowledge, this is the first comprehensive analysis focused to understand H3K27me3 modification status in adult GBM and DA tissue samples and also deliver a novel prognostic marker for GBMs regulated by H3K27me3.

Materials and methodsAstrocytic tumors (52 GBM, 24 AA and 14 DA) were selected for this study after ethical approval from Institute Ethical Committee. Age, sex and clinical history of all the patients were noted. Total RNA was iso-lated using mirVana™ miRNA isolation kit (Life Technologies, Carlsbad, CA; cat. no. AM1560), first-strand complementary DNA was synthe-sized by SuperScript® VILO™ complementary DNA synthesis kit (Life Technologies), and real-time PCR-based mRNA expression studies were performed using EZH2 and SLC25A23-specific primers (Supplementary Table 1, available at Carcinogenesis online) using SYBR green-based chemistry. The average of expression of four genes (TBP, RPL13A, RPL35A

Abbreviations

DA diffuse astrocytomaESC embryonic stem cellsEZH2 enhancer of zeste homolog-2fNSCs fetal neural stem cellsGB gene bodyGBM glioblastomaGSC glioma stem cellsH3K27me3 trimethylation of histone at lysine-27mRNA messenger RNAPMT promoterTSS transcription start site


154 | Carcinogenesis, 2017, Vol. 38, No. 2

and GARS) was used as reference for normalization of transcript levels as described by Valente et al. (28). An average of eight control normal brains was used to determine the normal range of expression for ref-erence and target genes. Level of EZH2 and SLC25A23 fold change was calculated using 2 t-∆∆C

method. Immunohistochemistry for EZH2 was done using monoclonal antibody for EZH2 (M/S Abcam, dilution 1:100) and labeled streptavidin–biotin kit (Universal) was used as a detec-tion system (M/S Dako, Glostrup, Denmark). ChIP immunoprecipitation was done on 13 tissue samples (11 GBM and 2 DA) using Low Cell ChIP kit (M/s Diagenode, Belgium) using ChIP-grade antibodies: H3K27me3 (M/s Abcam), Histone H3 (M/s Abcam) and immunoglobulin G antibody (M/s Abcam) per Low Cell ChIP kit guidelines with some modifications. DNA isolation from immunoprecipitates was done using IPure kit (M/s Diagenode, Belgium) per manufacturer’s instructions. For validation of ChIP, quantitative PCR reactions were performed in LC480 system (M/s Roche, Switzerland) for promoter of HOXA9 (positive control) and GAPDH (negative control) and expressed as percentage enrichment (Supplementary Table 1, available at Carcinogenesis online). Library gen-eration for ChIP-precipitated DNA was performed using True Seq DNA preparation kit (Ilumina, San Diego, CA) per manufacturer’s protocol. Sequencing was performed on HiSeq2000 next-generation sequenc-ing platform (Illumina), using 51 bp single-end protocols. A complete description of ChIP-seq bioinformatics analysis used is provided in Supplementary Text Materials and methods, available at Carcinogenesis online. Additionally, previously published data sets for H3K27me3 were sourced from the Gene Expression Omnibus (GEO) and/or published lit-erature. Raw sequencing reads were extracted for normal brain from Gene Expression Omnibus: GSE17312 (for frontal and temporal lobes) (29). Data for human ESCs, fetal neural stem cells (fNSCs) and GSCs were obtained from Guenther et al. (30), Yoo et al. (25) and Lin B et al. (26), respectively. For downstream analysis, both in-house and public data were processed in the identical way, wherever possible. Validation of ChIP-seq data was performed on four individual GBM tissues. Pathway analysis, gene expression, methylation analysis, heat map generation, cell culture, Western blot and survival and statistical analysis details of this study can be found in Supplementary Text Materials and methods, available at Carcinogenesis online.

Results and discussion

EZH2 mRNA and protein expression

In the present study, we first quantified the levels of EZH2 in DA, AA and GBM at transcript level using quantitative real-time PCR. The expression of EZH2 mRNA was found significantly higher in all grades of astrocytic tumors as compared with nor-mal brain. Moreover, its expression showed a significant posi-tive correlation with tumor grade, being significantly increased in high-grade tumors(grade IV > grade III > grade II; Figure 1a and b). Similar to the mRNA expression profile, immunopo-sitivity for EZH2 protein was found not to be expressed in the normal brain, reactive glial tissue and the background non-neo-plastic glia. However, immunopositivity for EZH2 protein was infrequent (with low labeling index) in DA tumors and became more frequent and widespread in higher grades (AA and GBM; Figure 1c). Consistent with the present EZH2 status, increased EZH2 expression in higher grade of gliomas has been demon-strated by Smits et al. (31), Orzan et al. (32) and Ott et al. (33).

Extracting ChIP-seq data and quality control

Because EZH2 is an important epigenetic regulator, we hypoth-esized that this high level of EZH2 may play a fundamental role in gliomagenesis by inducing epigenetic reprogramming of H3K27me3 modification marks. Hence, we performed ChIP to identify the genome-wide binding profile associated with these H3K27me3 modifications using primary tissue samples of both low-grade (DA) and high-grade (GBM) astrocytic tumors. Further, an initial quality control was performed to assess the target enrichment in H3K27me3 precipitated DNA in tissue samples using quantitative real-time PCR. The results showed an average enrichment of ~15-fold for positive target (HOXA9) as compared with negative target (GAPDH) in precipitated DNA (Figure 1d).

Figure 1. Expression of EZH2 and H3K27me3 ChIP-seq data in gliomas: (a) EZH2 mRNA expression status in normal brain, DAs, AAs and GBMs; (b) statistical signifi-

cance for EZH2 mRNA expression; (c) photomicrograph showing lack of EZH2 expression in normal brain (NB), immunonegativity in DA, nuclear immunopositivity in

AA and diffuse nuclear positivity in GBMs (×200); (d) H3K27me3 ChIP quality check using ChIP quantitative PCR for HOXA9 (positive target) enrichment in comparison

with GAPDH (negative control). IgG and histone 3 used as negative and positive antibody controls, respectively; (e) bar graph representing H3K27me3 ChIP-seq peaks in

individual GBM and DA samples; and (f) genome-wide representation of H3K27me3 ChIP peaks observed in integrated genome visualizer (IGV) for GBM and DA samples.



ChIP-precipitated samples from the tissues were sequenced using HiSeq2000 platform, which generated an average of 21.51 million reads (range 12–28 million reads; Figure 1e and f). In addi-tion to the present data, we also analyzed previously published ChIP-seq data sets for H3K27me3 profile for frontal and tempo-ral lobes of non-neoplastic brain (29). Alignment of reads against version hg19 of the human genome resulted in an average of 16 million (73.48%) reads aligned to the reference sequence. Next, to analyze uniquely mapped reads, a model-based anal-ysis for ChIP-sequencing (MACS) algorithm was utilized. We obtained an average of 3386 targets (range 2292–4333) for GBMs and 3754 targets (range 3438–4070) for DAs. Furthermore, in order to validate quality of sequencing data, six selected target binding sites (found enriched in majority of the samples) for H3K27me3 were analyzed on four tissue samples using ChIP-qPCR (Supplementary Figure 1a and b, available at Carcinogenesis online). Our ChIP-quantitative PCR rate for the targets was good, indicating an efficient quality of the sequencing data.

Downstream pipeline for identification of preferential H3K27me3 gene targets

In order to keep data unambiguous, only top 1000 H3K27me3 peaks (on the basis of cutoff value ofP < 0.001) of each GBM and DA sample were selected for downstream analysis (suggested and modified method per Zhang et al. (34). Detailed description of informatics pipeline is described in Supplementary Figure 2, available at Carcinogenesis online. For GBM samples, all selected 1000 H3K27me3 peaks (n = 11) were combined and the removal of repetitive targets was done. This resulted in 7043 targets in coding region (gene coding) and 3957 targets of noncoding region (includes microRNA, ribosomal RNA, transfer RNA, Piwi-interacting RNA and small nucleolar RNA). As the main focus of this study was to analyze the effect of H3K27me3 modification on gene expression, only coding region in H3K27me3 targets were considered for analysis. Further, as suggested by Young et al. (35), these 7043 H3K27me3 targets in coding region were classified into three distinct H3K27me3 ‘genomic enrichment sites’, namely transcriptional start site (TSS), promoter (PMT) and gene body (GB). Repetitive targets in each enrichment region were next removed while keeping only one representative peak with most significant P value among replicates. This resulted in 4657 total unique targets of H3K27me3 modifications, of which GB region accounted for the largest with 59% (n = 2755), fol-lowed by PMT region with 23% (n = 1059) and TSS region with 18% (n = 843). For representing the most vital targets regulated by H3K27me3, a maximum of 1000 targets (with least P value) under each ‘genomic enrichment sites’ (i.e. GB, PMT and TSS regions) were finally selected. These criteria resulted in 1000 tar-gets each for GB and PMT regions and 843 targets for TSS region (Supplementary Table 2, available at Carcinogenesis online). For all downstream analysis, these selected targets are represented as ‘target genes’ under GBM profile (i.e. GB_GBM, PMT_GBM and TSS_GBM).

In the case of DA samples, a similar approach of peak sorting was performed, which resulted in 1383 targets in coding and 617 targets in noncoding regions. Removal of duplicates and clas-sification of each coding targets into three ‘genomic enrichment sites’ resulted in 82% targets in GB region (n = 1085), 13% in PMT (n = 185) and 4% in TSS (n = 55) region. Similar to approach fol-lowed in GBM, a maximum of top 1000 targets under GB, PMT and TSS profiles were selected, which finally resulted in 1000 tar-gets for GB, 185 for PMT and 55 for TSS regions (Supplementary Table 2, available at Carcinogenesis online). For all downstream analysis, these selected targets are denoted as ‘target genes’

under DA profile (i.e. GB_DA, PMT_DA and TSS_DA). In both GBM and DA samples, maximum ‘target genes’ for H3K27me3 were observed in GB region followed by PMT and TSS (GB > PMT > TSS). Although, top 1000 target genes in GB region are included for analysis, the complete GB targets for GBM and DA are also provided (Supplementary Table 3, available at Carcinogenesis online).

Distribution of binding sites of the H3K27me3 target genes in GBM and DA

In order to find out whether the H3K27me3 target genes were modified at single or at multiple ‘genomic enrichment sites’, the targets of GB, PMT and TSS regions were compared in both GBMs and Das (Figure 2a). In GBM cases, 95 H3K27me3 target genes were found to be shared by 3 genomic sites that included genes such as PAX5, WNT and HOX family members. Further, >200 H3K27me3 target genes were found to be shared between 2 genomic sites in GBM (PMT and TSS = 256; PMT and GB = 225; GB and TSS = 216). On the contrary, none of the H3K27me3 tar-get genes of DAs were common to all the three genomic enrich-ment sites and nearly negligible(≤2) H3K27me3 target genes were shared between two genomic sites (PMT and TSS = 1; PMT and GB = 2; GB and TSS = 2). Therefore, in GBMs, a large set of H3K27me3 target genes were marked at multiple genomic sites unlike in low-grade DAs. Moreover, analyzing DNA meth-ylation status of H3K27me3 targets in GBM using previous stud-ies (36) showed that targets in TSS regions (either alone or in combination) were associated with high levels of DNA meth-ylation [GB_TSS (17.3%) > PMT_TSS (16.7%) > TSS(12.2%) > GB_PMT_TSS (10.5%) > GB (8.56%) > GB_PMT (8.4%) > PMT (6.6%) (Supplementary Table 4), available at Carcinogenesis online]. However, PMT region showed the low levels of DNA methyla-tion (6.6%). Interestingly, the target genes harboring H3K27me3 modification at all the three regions (i.e. GB, PMT and TSS) were found to show a moderate level (10.5%) of DNA methylation lev-els. Thus, targets with H3K27me3 modification at TSS regions seems to be regulated by both H3K27me3 and DNA methylation, whereas targets in PMT regions are majorly regulated by the effect induced by H3K27me3 modification alone.

Comparison of H3K27me3 profile of GBM and DA versus normal brain

We next studied the differences observed in H3K27me3 meth-ylation profile in GBMs with normal brain (29). For GBMs, 91.3% (n = 2597/2843) target genes were uniquely expressed in com-parison with normal brain (Supplementary Figure 3a, available at Carcinogenesis online). On the basis of ‘genomic enrichment sites’, 34% (n = 881/2597) target genes were found to be present in GB, 35% (n = 909/2597) in PMT and 31% (n = 807/2597) in TSS region (Supplementary Figure 3a, available at Carcinogenesis online). Thus, the proportions of overlapping targets among GBMs with respect to normal brain were <8.7% (n = 246/2843). Analysis of H3K27me3 target genes that were found spe-cifically in GBMs included homeobox genes, LHX-4/-6, CDX2, POU4F1, neuronal differentiation target genes like NEUROD-1/ -2, NEUROG-1/-2/-3, brain-specific cadherin CDH22 and regula-tors of cancer-related genes like TGFB1, WNT9A, YAP1, VEGFA, HES7, BMP3 and SNAI3. In the case of DAs, >95% (n = 1172/1240) of targets were noticeably different in comparison with nor-mal brain (Supplementary Figure 3b, available at Carcinogenesis online). Sorting the genes basis of ‘genomic enrichment sites’ revealed that, 93.7% of GB region targets (n = 937/1000), 97.8% of PMT region targets (n = 181/185) and 98.1% of TSS region target (n = 54/55), genes were different in comparison with normal brain



(Supplementary Figure 3b, available at Carcinogenesis online). Further, 5% of DA target genes were shared with normal brain (n = 68/1240). Notably, H3K27me3 target genes specifically found in DA were found to be involved in metabolic pathways like UDP glucuronosyltransferase 1 family members (UGT1A10/-5/-6/-7/-8/-9), aldehyde dehydrogenase 3 family members (ALDH3A-1/-2), calcium signaling pathway members (CACNA1A/-B/-C and RYR1/-2), glutamate receptors like GRIA4, GRIK3/-4 and cancer-related molecules like AXIN2, COL4A, WNT9A and LAMA1/-2). Thus, it is apparent from the present data that in comparison with nor-mal brain, a large shift in H3K27me3 modification is observed in gliomas, and these target genes may play an important role in gliomagenesis and progression.

Comparison of H3K27me3 target gene profile between GBM and DA

We next investigated whether a distinct set of H3K27me3 regu-lated genes can differentiate GBM versus DA. On comparison, a large proportion of H3K27me3 target genes in GBMs were found to be different from DA, i.e. 93.2% (n = 2651/2843; Figure 2b). On

site basis evaluation, 85.4% (n = 854/1000) of GB region target genes were found to be different in GBM in comparison with DA. Also, differences were noted in PMT (96.1% [n = 961/1000]) and TSS genomic site (99.1% [n = 836/843]). Notably, only 6.8% (n = 192/2843) of GBM target genes were found to be shared with DA cases. This suggests significant differences in the H3K27me3 modification profile in low-grade (DA) when compared with high–grade (GBM) astrocytic tumors.

Comparison of H3K27me3 profile of GBM and DA with ESCs, fNSCs and GSCs H3K27me3 profile

High-grade tumors are reported to be associated with more malignant characteristics and higher percentage of stem cell-like characters as compared with low grades. Because EZH2-mediated H3K27me3 modification has been reported to be associated with such properties, thus, in order to draw a rela-tionship between H3K27me3 map of GBMs and DAs with ESCs, the present data were compared with a set of published human embryonic data of Guenther et al. (30), which has a comprehen-sive analysis of H3K27me3 profile for six ESCs (human ESCs).

Figure 2. Venn diagram showing comparison of H3K27me3 target genes: (a) comparison of target genes of three genomic enrichment sites—GB, PMT and TSS—in GBMs

and Das and (b) comparison of target genes GBMs versus DAs.



Approximately 50% (n = 1405/2843) of GBM H3K27me3 target genes falling in each genomic site were found to be shared with H3K27me3 profile of ESCs (Supplementary Figure 4a, available at Carcinogenesis online). This is in agreement with the obser-vation by Lin et al. (26), where a comparison showed ~47.5% of target genes to be shared between them. On the contrary, only 21% (n = 258/1240) of DA gene targets were shared with ESC data (Supplementary Figure 4b, available at Carcinogenesis online). Thus, GBMs share a higher percentage of H3K27me3 modified target gene profile similar to ESCs in comparison with DA, and majority of these targets are linked with maintenance of stem cell-like state in GBMs.

Recent reports have suggested that H3K27me3 modifications are critical in conversion of gliogenic transition during brain development from neural stem cells (37), and this could serve as a cell of origin for brain tumor stem cells (38,39). We com-pared H3K27me3 ChIP-seq data for fNSCs (25) with the present data and found ~35% (n = 1459/2242) of GBM genes and 17.4% (n = 1019/1235) of DA genes to be shared with fNSCs–H3K27me3 data (Supplementary Figure 4c and d, available at Carcinogenesis online). Thus, our results demonstrate that a higher percentage of fNSCs–H3K27me3 target genes are exhibited in GBMs as com-pared with DAs, indicating the presence of fetal stem cell-like state to be present in GBMs.

Further, GBM is also known to have a small population of self-renewing, tumorigenic cancer stem cells called as GSCs that contribute to tumor initiation and therapeutic resistance (40). We hypothesized that GSCs when compared with ESCs uti-lize comparatively more repressive regions via epigenetic con-trol mechanism. We compared the present H3K27me3 profile data of GBM and DA with H3K27me3 profile published on eight GSCs cell line by Lin et al. (26). In comparison with GBM data, an average of 53.8% (n = 1527/2843) of H3K27me3 target genes was found to be shared with GSCs (Supplementary Figure 4e,

available at Carcinogenesis online). For DA cases, an average of 34.2% (n = 423/1240) of target genes were found to be shared with GSCs H3K27me3 data (Supplementary Figure 4f, avail-able at Carcinogenesis online). These observations suggest that H3K27me3-modified profile of stem cells are more enriched in GBMs when compared with DAs, indicating GBMs to have more stem cell-like state than DAs. Hence, these genes may have a vital role in influencing higher proliferation and may define aggressive behavior and therapeutic resistance associated with high-grade gliomas as compared with low grades.

Correlation of H3K27me3 target gene profile in GBMs with gene expression and methylation data from TCGA

H3K27me3 modifications are traditionally known to be a repres-sive mark and are generally associated with silenced promoters. Thus, the relationship between gene expression and H3K27me3 modifications was also evaluated in this study using expression data of GBMs from TCGA keeping ≤2-fold as cutoff for upreg-ulated and downregulated expression. The analysis showed around 23.2% (n = 522/2247) H3K27me3 target genes were down-regulated and only 9.2% (n = 207/2247) were upregulated, rest 67.5% (n = 1518/2247) remained unchanged in terms of expres-sion (Figure 3a and b). KEGG analysis of the target genes that were upregulated were noted in metabolic regulation (CYP27B1, PRIM2 and SHMT2), pathways in cancers (EGFR, FZD2, CDKN1A, MMP2 and FOS), Antigen-processing and -presentation (B2M, CIITA) and glioma-specific (CDK4, CDK6) genes.

DNA methylation represents another most widely studied epigenetic mark in gliomas that occurs at cytosine residues in the context of CpG dinucleotides. Although DNA and histone meth-ylation are different sets of epigenetic modifications, they may be dependent on each other to determine the final gene expres-sion profile. Further, cross-talk between these modifications may

Figure 3. Analysis of GB, PMT and TSS specific H3K27me3 targets in GBM TCGA database. (a) Bar graph showing gene expression analysis performed using TCGA data

on the basis of the three genomic enrichment sites; (b) gene expression heat map for H3K27me3 target genes in GB, PMT and TSS region analyzed using TCGA data; and

(c) bar graph showing percentage of genes with both DNA hypermethylation and H3K27me3 modifications and genes with only H3K27me3 modifications.



play an important role in the establishment of chromatin diver-sity within the genome. However, because DNA methylation is also known to play a vital role in repression of gene expression, we took another step to analyze whether the negative regula-tion of genes was independently regulated by H3K27me3 alone or was a combined effect of H3K27me3 as well as DNA promoter hypermethylation. To test this, we analyzed the downregulated target genes for their DNA methylation status using avail-able data from Gene Expression Omnibus database (36). It was interesting to note that of the 408 downregulated target genes with methylation data available, majority 74.02% (n = 302/408) showed both DNA hypermethylation and H3K27me3 silencing marks, whereas only 25.98% (n = 106/408) were observed to have H3K27me3 silencing mark alone (Figure 3c). Thus, this highlights that majority of the H3K27me3 targets were downregulated by a combinatorial effect of H3K27me3 modifications as well as DNA methylation.

Neuroactive ligand-receptor pathway is enriched in GBMs

In order to identify H3K27me3-regulated pathways specific to GBM and DA groups, the H3K27me3 target gene reported in all the three genomic enrichment profile (GB, PMT and TSS) were combined for GBM and DA and multiple copy of repetitive tar-gets as well as those shared with normal brain were removed. This resulted in 2241 H3K27me3 target genes for GBM and 1235 for DA. By comparing these targets, we observed 86.65%

(1942/2241) H3K27me3 target genes to be exclusive to GBM and 75.79% (936/1235) gene targets to DA. KEGG pathway enrich-ment analysis was done using KEGG for these H3K27me3 gene targets specific to GBM and DA, respectively. In GBM, the most enriched pathway was neuroactive ligand-receptor pathway (P = 9.4 × 10−25) that was enriched with 73 genes (Figure 4a). The other top pathways were pathways in cancer (P = 3.4 × 10−16), calcium signaling pathway (P = 1.13 × 10−10), WNT signaling (P = 1.66 × 10−7) and MAPK signaling pathway (P = 1.67 × 10−7). In DA, we observed 37 pathways to be enriched with significant P values (Figure 4a). The top pathways included were ascorbate and aldarate metabolism (P = 3.07 × 10−5), pentose and glucu-ronate interconversions (P = 6.94 × 10− 5), starch and sucrose metabolism (P = 2.89 × 10− 4) and so on.

Neuroactive ligand-receptor signaling (KEGG: hsa04080) was the most prominent pathway found altered in GBMs. We analyzed the TCGA database for gene expression status of the 73 H3K27me3 target genes of this pathway, of which 27 genes showed significantly 2-fold changes (P < 0.05) in expression (Supplementary Table 5, available at Carcinogenesis online). Interestingly, all of these 27 genes were found to be downregu-lated (Figure 4b and c). We further speculated the involvement of DNA methylation in negative regulation of these genes. The DNA methylation status showed 74% (20/27) genes to be having significant DNA hypermethylation at these H3K27me3-modified genomic sites, thus suggesting a combinatorial silencing is mediated by both these epigenetic modifications. Interestingly,

Figure 4. Analysis for GBM-specific and DA-specific H3K27me3 targets/pathways. (a) KEGG pathways analysis showing top pathways altered in GBM and DA. The X-axis

and Y-axis indicate P value expressed in −log 10 scale and KEGG pathway, respectively. (b) Pictorial representation of neuroactive ligand-receptor interaction pathway

using KEGG. Boxes highlighted in black color are downregulated H3K27me3 target genes of neuroactive ligand-receptor pathway in GBMs. (c) TCGA gene expression

analysis for H3K27me3 target genes of neuroactive ligand-receptor pathway in GBMs.



among these low expression of CHRNA3 and ADRA2A were found to be associated with the worst outcome in GBMs (Supplementary Figure 5, available at Carcinogenesis online).

SLC25A23: H3K27me3 target gene with a prognostic role

Previous studies based on whole genome mapping of chro-matin modifications in ESCs have shown the presence of dis-tinct histone marks (i.e. H3K4me3 and H3K27me3) at certain genomic sites that are associated with active and repressed PMTs, respectively. Because PMT-mediated regulation of gene expression is a well-known phenomenon and the role of repres-sive histone marks at PMT region is a parallel reported event, we considered to analyze H3K27me3-downregulated targets from PMT region for further analysis. One such interesting gene was found belonging to the solute carrier family namely SLC25A23 that represents part of drug transport families. SLC molecules have recently gained therapeutic importance as an oncogene as well as tumor suppressor genes (26, 41–43). SLC25A23 is a cal-cium-dependent mitochondrial solute carrier, which is known to transport metabolites, nucleotides and cofactors through the mitochondrial inner membrane.

Literature suggests that deregulation of solute carrier pro-teins may lead to accumulation of certain molecules within the cell (or organelles) and thus may become lethal if the accumula-tion reaches toxic levels (26). Defects in many of the solute carrier genes are implicated in human diseases. Loss of SLC members in prostate and pancreatic cancers (41) and in gastric cancers (42) has been reported due to PMT hypermethylation. Recently, the role of SLC family members has also been reported in gliomas. Lin et al. (26) reported the existence of bivalent histone modifications

on SLC17A7, which leads to the loss of its expression in GBMs. Using ChIP-seq and cell culture studies in T98 and U87 cell lines, an overexpression of SLC17A7 gene led to decreased cell pro-liferation, migration and invasion. This suggested active role of SLCs in maintaining tumor suppressor activity, which has been less explored thus far. Hence, in the present study, we analyzed the role of H3K27me3 target gene SLC25A23 molecule in our in-house GBM cases as well as in TCGA cohort.

Gene expression analysis of SLC25A23 was performed in a cohort of 40 GBM samples and 6 controls, which showed signifi-cantly downregulation of SLC25A23 in GBMs as compared with the controls (P < 0.001; Figure 5a). Of these 40 GBM cases, sur-vival data for 28 cases (70%) were available. Kaplan Meir graph was plotted to correlate SLC25A23 mRNA expression with the patient outcome. The expression data were analyzed by keep-ing the median as cutoff (values above median was considered to be high expression and expression below median value was considered as low expression). The analysis showed that low SLC25A23 expression was associated with poorer outcome (P < 0.001; Figure 5b). The median time of progression free sur-vival for patients with low SLC25A23 was 31.8 weeks in con-trast to those with high SLC25A23 which was 86.2 weeks. For validation of our in-house results, the expression of SLC25A23 was seen in large TCGA database and was also correlated with survival. Similar to our results, we observed downregulation of SLC25A23 in all cases of GBM (P < 0.0001; Figure 5c) and a simi-lar trend to poor prognosis was also noted in TCGA database (P = 0.031; Figure 5d). This indicated that SLC25A23 may act as a novel prognostic marker for GBMs.

In order to confirm whether H3K27me3 plays a role in the regulation of SLC25A23, EZH2 knockdown (catalytic unit of

Figure 5. Analysis of H3K27me3 target genes, SLC25A23 in in-house and TCGA cohort. (a) Expression analysis of SLC25A23 in a cohort of 40 GBM samples and 6 controls

showed significantly downregulated of SLC25A23 in GBMs as compared with controls (P < 0.001). (b) Kaplan Meir analysis for in-house data showed low SLC25A23

expression to be associated with poorer outcome (P < 0.001). (c) Expression of SLC25A23 in large TCGA database for GBM. (d) Kaplan Meir analysis for TCGA database

also showed poor prognosis in TCGA database (P = 0.031). Knockdown of EZH2 in U87 resulted in increase of SLC25A23. (e) mRNA expression (using quantitative real-

time PCR) and (f) at protein level (using western blot).



polycomb repressive complex 2 complex which functions to induce H3K27me3) was perfomed in U87MG cell line and its effect on expression of SLC25A23 was analyzed. Interestingly, knockdown of EZH2 resulted in an increased SLC25A23 expres-sion (by atleast 4-fold), confirming a direct role of H3K27me3 modification in epigenetically silencing the SLC25A23 gene expression (Figure 5e and f). Previous studies have reported an active role of SLC25A23 in calcium ion uptake regulation with significant reactive oxygen species and cell death implications (44). Further, in cancers, downregulation of calcium ion homeo-stasis proteins have been reported to develop chemoresistance (45). Therefore, targeting these calcium ion transporters in order to enhance the proapoptotic potential of malignant cells may be a useful strategy in the treatment of gliomas. Thus, the pre-sent study highlights SLC25A23 as a potential new prognostic biomarker for GBMs and further studies are needed to establish its role.

ConclusionThis is possibly the first high-resolution genome-wide map of H3K27me3 modification in primary adult tissue samples using human gliomas. EZH2, which is overexpressed in all grades of gliomas, catalyzes the H3K27me3 and therefore the high lev-els of EZH2 may be playing a fundamental role in gliomagen-esis and progression by inducing epigenetic reprogramming of these H3K27me3 marks. In majority, H3K27me3 modifica-tion contributes to silencing/repression of the target genes. The present study highlights that majority (70%) of H3K27me3 target genes are downregulated not only due to H3K27me3 modification but also due to concomitant with DNA hyper-methylation. We identify three enrichment profiles in the genome-wide maps of H3K27me3 in gliomas, which also cor-relate with transcriptional activity. Thus, H3K27me3 modifica-tion in gliomas associated with gene silencing/repression is not restricted to PMT region but also present in GB and TSS regions. This suggests the importance of considering the pre-cise enrichment region of H3K27me3 for regulation of gene expression. The present study also reports significant differ-ences noted in the H3K27me3 profile between GBMs and Das: (i) Significantly more H3K27me3 targets are found in the cod-ing region of GBMs (n = 7043) versus Das (n = 1383). (ii) differ-ences are observed between GBM versus DA in the percentage of targets in each genomic enrichment region (GB = 59% GBM versus 82% DA; PMT = 23% GBM versus 13% DA; TSS = 18% GBM versus 4% DA). (iii) H3K27me3 targets are often shared between two to three genomic enrichment sites in GBMs. In contrast, there was negligible sharing in DAs. (iv) Both ESC- and GSC-specific H3K27me3 target genes are more enriched in GBMs (50% and 54%, respectively) as compared with Das (21% and 34%, respectively). Interestingly, the shared genes between GBM and ESCs include stem cell maintenance regula-tors and differentiation markers. Abundance of such shared targets in GBMs may suggest them for contributing to more stemness characteristics/s phenotypes as well as therapeutic resistance in GBMs. (v) More than 90% of the target genes are noted to be different between GBM and DA. From this it can be suggested that different pathways are targeted in low-grade versus high-grade gliomas. Indeed, pathway differences are found on KEGG analysis of H3K27me3 target genes between GBM and DA. GBMs are found enriched in neuroactive ligand-receptor pathways, whereas DA shows enrichment of meta-bolic pathways. For the first time, we show the downregulation of neuroactive ligand-receptor pathway in GBMs, which is

possibly attributable to H3K27me3 modification. The role of this pathway in gliomagenesis needs further study. Further, SLC25A23 appears to be a potential new prognostic biomarker for GBMs. This is a calcium-dependent mitochondrial sol-ute carrier gene that appears to be downregulated by EZH2-mediated H3K27me3 silencing. Further studies are needed to establish its role as a new biomarker. Thus, this study not only provides an improved understanding of the epigenetic altera-tions associated with gliomagenesis and progression but also offers a novel view of complexity of regulation of gene expres-sion in gliomas.

Supplementary materialSupplementary data are available at Carcinogenesis online.

FundingIndian Council of Medical Research (ICMR), India (no. 5/13/118/2011-NCD-III) and J C Bose fellowship of Department of Science and Technology, India, to Dr C.S. and Senior Research Fellowship [09/006(0397)/2009-EMR-I] award from Council of Scientific and Industrial Research (CSIR), India, to V.S.

AcknowledgementsWe thank Dr Shantanu Chowdhury and his student Mr Dhurjhoti Saha (Institute of Genomics and Integrative Biology, New Delhi) for helping in ChIP standardization. We also like to thank the technical staff of the Neuropathology laboratory, All India Institute of Medical Sciences(AIIMS), New Delhi, India, and all the consultants of the Neurosurgery department.Conflict of Interest Statement: None declared.

References 1. Van Meir, E.G. et al. (2010) Exciting new advances in neuro-oncology:

the avenue to a cure for malignant glioma. CA. Cancer J. Clin., 60, 166–193.

2. Huse, J.T. et al. (2014) The evolving role of molecular markers in the diagnosis and management of diffuse glioma. Clin. Cancer Res., 20, 5601–5611.

3. Di Croce, L. et al. (2013) Transcriptional regulation by Polycomb group proteins. Nat. Struct. Mol. Biol., 20, 1147–1155.

4. Deb, G. et al. (2014) EZH2: not EZHY (easy) to deal. Mol. Cancer Res., 12, 639–653.

5. Yang, Y.A. et al. (2013) EZH2, an epigenetic driver of prostate cancer. Protein Cell, 4, 331–341.

6. Martínez-Fernández, M. et al. (2015) EZH2 in bladder cancer, a promis-ing therapeutic target. Int. J. Mol. Sci., 16, 27107–27132.

7. Lund, K. et al. (2014) EZH2 in normal and malignant hematopoiesis. Leukemia, 28, 44–49.

8. Chase, A. et al. (2011) Aberrations of EZH2 in cancer. Clin. Cancer Res., 17, 2613–2618.

9. Yu, J. et al. (2007) A polycomb repression signature in metastatic pros-tate cancer predicts cancer outcome. Cancer Res., 67, 10657–10663.

10. Takeshita, F. et al. (2005) Efficient delivery of small interfering RNA to bone-metastatic tumors by using atelocollagen in vivo. Proc. Natl. Acad. Sci. USA, 102, 12177–12182.

11. Jenuwein, T. et al. (2001) Translating the histone code. Science, 293, 1074–1080.

12. Zhao, X.D. et al. (2007) Whole-genome mapping of histone H3 Lys4 and 27 trimethylations reveals distinct genomic compartments in human embryonic stem cells. Cell Stem Cell, 1, 286–298.

13. Kouzarides, T. (2007) Chromatin modifications and their function. Cell, 128, 693–705.

14. Crea, F. et al. (2012) Polycomb genes and cancer: time for clinical appli-cation? Crit. Rev. Oncol. Hematol., 83, 184–193.



15. Wei, Y. et al. (2008) Loss of trimethylation at lysine 27 of histone H3 is a predictor of poor outcome in breast, ovarian, and pancreatic cancers. Mol. Carcinog., 47, 701–706.

16. Cai, M.Y. et al. (2011) High expression of H3K27me3 in human hepato-cellular carcinomas correlates closely with vascular invasion and pre-dicts worse prognosis in patients. Mol. Med., 17, 12–20.

17. He, L.R. et al. (2009) Prognostic impact of H3K27me3 expression on locoregional progression after chemoradiotherapy in esophageal squa-mous cell carcinoma. BMC Cancer, 9, 461.

18. Velichutina, I. et al. (2010) EZH2-mediated epigenetic silencing in ger-minal center B cells contributes to proliferation and lymphomagen-esis. Blood, 116, 5247–5255.

19. Ke, X.S. et al. (2009) Genome-wide profiling of histone h3 lysine 4 and lysine 27 trimethylation reveals an epigenetic signature in prostate carcinogenesis. PLOS One, 4, e4687.

20. Hübner, M.R. et al. (2010) Role of H3K27 demethylases Jmjd3 and UTX in transcriptional regulation. Cold Spring Harb. Symp. Quant. Biol., 75, 43–49.

21. Ahn J. et al. (2016) Target sequencing and CRISPR/Cas editing reveal simultaneous loss of UTX and UTY in urothelial bladder cancer. Onco-target, doi: 10.18632/oncotarget.11207 [Epub ahead of print].

22. Zha, L. et al. (2016) Epigenetic regulation of E-cadherin expression by the histone demethylase UTX in colon cancer cells. Med. Oncol., 33, 21.

23. Kim, J.H. et al. (2014) UTX and MLL4 coordinately regulate transcrip-tional programs for cell proliferation and invasiveness in breast cancer cells. Cancer Res., 74, 1705–1717.

24. Natsume, A. et al. (2013) Chromatin regulator PRC2 is a key regulator of epigenetic plasticity in glioblastoma. Cancer Res., 73, 4559–4570.

25. Yoo, S. et al. (2014) Differences among brain tumor stem cell types and fetal neural stem cells in focal regions of histone modifications and DNA methylation, broad regions of modifications, and bivalent pro-moters. BMC Genomics, 15, 724.

26. Lin, B. et al. (2015) Global analysis of H3K4me3 and H3K27me3 profiles in glioblastoma stem cells and identification of SLC17A7 as a bivalent tumor suppressor gene. Oncotarget, 6, 5369–5381.

27. Bender, S. et al. (2013) Reduced H3K27me3 and DNA hypomethylation are major drivers of gene expression in K27M mutant pediatric high-grade gliomas. Cancer Cell, 24, 660–672.

28. Valente, V. et al. (2009) Selection of suitable housekeeping genes for expression analysis in glioblastoma using quantitative RT-PCR. BMC Mol. Biol., 10, 17.

29. Bernstein, B.E. et al. (2010) The NIH roadmap epigenomics mapping consortium. Nat. Biotechnol., 28, 1045–1048.

30. Guenther, M.G. et al. (2010) Chromatin structure and gene expression programs of human embryonic and induced pluripotent stem cells. Cell Stem Cell, 7, 249–257.

31. Smits, M. et al. (2010) miR-101 is down-regulated in glioblastoma resulting in EZH2-induced proliferation, migration, and angiogenesis. Oncotarget, 1, 710–720.

32. Orzan, F. et al. (2011) Enhancer of Zeste 2 (EZH2) is up-regulated in malignant gliomas and in glioma stem-like cells. Neuropathol. Appl. Neurobiol., 37, 381–394.

33. Ott, M. et al. (2012) Promotion of glioblastoma cell motility by enhancer of zeste homolog 2 (EZH2) is mediated by AXL receptor kinase. PLoS One, 7, e47663.

34. Zhang, J.X. et al. (2013) Unique genome-wide map of TCF4 and STAT3 targets using ChIP-seq reveals their association with new molecular subtypes of glioblastoma. Neuro. Oncol., 15, 279–289.

35. Young, M.D. et al. (2011) ChIP-seq analysis reveals distinct H3K27me3 profiles that correlate with transcriptional activity. Nucleic Acids Res., 39, 7415–7427.

36. Etcheverry, A. et al. (2010) DNA methylation in glioblastoma: impact on gene expression and clinical outcome. BMC Genomics, 11, 701.

37. Hirabayashi, Y. et al. (2009) Polycomb limits the neurogenic compe-tence of neural precursor cells to promote astrogenic fate transition. Neuron, 63, 600–613.

38. Sanai, N. et al. (2005) Neural stem cells and the origin of gliomas. N. Engl. J. Med., 353, 811–822.

39. Stiles, C.D. et al. (2008) Glioma stem cells: a midterm exam. Neuron, 58, 832–846.

40. Lathia, J.D. et al. (2015) Cancer stem cells in glioblastoma. Genes Dev., 29, 1203–1217.

41. Park, J.Y. et al. (2007) Candidate tumor suppressor gene SLC5A8 is fre-quently down-regulated by promoter hypermethylation in prostate tumor. Cancer Detect. Prev., 31, 359–365.

42. Liu, X. et al. (2009) Promoter hypermethylation mediates downregula-tion of thiamine receptor SLC19A3 in gastric cancer. Tumour Biol., 30, 242–248.

43. Takeuchi, Y. et al. (2015) Exonic mutations in the SLC12A3 gene cause exon skipping and premature termination in Gitelman syndrome. J. Am. Soc. Nephrol., 26, 271–279.

44. Hoffman, N.E. et al. (2014) SLC25A23 augments mitochondrial Ca2+ uptake, interacts with MCU, and induces oxidative stress-mediated cell death. Mol. Biol. Cell, 25, 936–947.

45. Dubois, C. et al. (2013) Targeting apoptosis by the remodelling of cal-cium-transporting proteins in cancerogenesis. FEBS J., 280, 5500–5510.


Documents

Genome wide ChIP-seq analysis of EZH2 mediated H3K27me3