Defining the Molecular and Cellular Mechanisms of EZH2's Dynamic Role In

Home , CBX2 (gene), CBX7 (gene), CBX8, EZH1, H3F3B (gene), HOXA6

Defining the Molecular and Cellular Mechanisms of EZH2's Dynamic Role in

Tumourigenesis

Thomas Mortimer

University College London and The Francis Crick Institute

PhD Supervisor: Dr Paola Scaffidi

A thesis submitted for the degree of Doctor of Philosophy University College London

July 2019

Declaration

I, Thomas Mortimer confirm that the work presented in this thesis is my own. Where information has been derived from other sources, I confirm that this has been indicated in the thesis.

Abstract

Epigenetic regulators are recurrently hijacked by cancer cells to enhance and maintain their malignant properties. Remarkably, epigenetic regulators are often co-opted by cancer cells in their wild-type unmutated form without any discernible changes to their intrinsic properties. However, the molecular mechanisms that underlie the co-option of these key regulators of cellular identity remain largely unclear. To generate a mechanistic paradigm for this phenomenon, I investigate how the Polycomb component EZH2 transitions from its physiological role in regulating normal development and tissue maintenance, to its pathological role as a tumour promoter. Focussing on the role of EZH2 in the central nervous system, I show that oncogenic signalling redistributes EZH2 on chromatin, leading to transcriptional misregulation of key neural homeotic genes and a malignant rewiring of normal development programmes. This is achieved by de-repression of spinal cord-specifying HOX genes and repression of the forebrain and neural stem cell transcription factor EMX2, an expression switch observed in both glioblastoma cell lines and patient samples. By in vivo assays, I show that EMX2 is a potent tumour suppressor in glioblastoma, suggesting that EMX2 repression by EZH2 is important for glioblastoma maintenance. Beyond its chromatin distribution, I also show that transformation alters the binding partners of EZH2, suggesting that dynamic changes to the composition of Polycomb complexes may promote tumour formation. Thus, by performing a detailed multi-omics characterisation of how neoplastic transformation alters the distribution, activity and binding partners of EZH2, I provide a unique insight into the molecular and cellular mechanisms underlying wild-type EZH2's dynamic role in tumourigenesis.

Impact Statement

Therapeutic targeting of epigenetic regulators is emerging as an effective treatment strategy in many cancer types. In particular, EZH2, a transcriptional repressor crucial in development and tissue maintenance, acts as a tumour promoter in various malignancies and is also a clinically-relevant therapeutic target. However, the full range of cancer types for which EZH2-targeted therapy may be an effective option remains unclear and there exists no means to identify the specific patients who may benefit. Moreover, by undetermined mechanisms, EZH2 acts as both a tumour suppressor and promoter in cancer, potentially confounding the use of EZH2-targeted therapy. To address these important issues, I performed a detailed characterisation of molecular mechanisms which enable EZH2 to acts as a tumour promoter in cancer, leading to discoveries that I believe are of both academic and clinical interest.

In the first part of this study, I reveal that in glioblastoma, transformation- induced redistribution of EZH2 causes gene expression changes at key neural transcription factors. This finding provides a clinically-relevant insight into the molecular mechanisms which drive glioblastoma growth, a pertinent finding given glioblastomas poor prognosis and our limited knowledge of its underlying mechanisms. Furthermore, based on EZH2’s mechanism of action in glioblastoma, I propose a stage-specific model to describe EZH2’s dichotomous tumour suppressive and promoting roles, presenting a theoretical framework which may help others research and interpret the tumour promoting roles of more epigenetic regulators. In addition, the extensive ChIP-seq, RNA- seq and proteomics data sets generated in this study will provide a valuable resource to others in the epigenetics community, potentially stimulating and complementing future research.

In the second part of this study, I describe how transformation alters the composition of PRC2 through changing the binding of specific accessory proteins. The role of PRC2 accessory proteins remains poorly defined, and hence this provides an important insight to how these proteins function in both normal and cancer cells. Moreover, I present preliminary evidence that PRC2 accessory proteins in their wild-type form can play tumour suppressive and promoting roles in cancer, providing a starting point for future research to

characterise in greater detail the function of these proteins in cancer - research which may have important clinical implications.

In the final part of this study, I present a detailed comparative analysis of genetic and pharmacological loss-of-function approaches for determining the genes regulated by EZH2. This controlled comparison will provide a reference point for others considering on embarking on similar studies, enabling them to make a rational and informed decision as to the experimental loss-of-function approach to employ.

Acknowledgements

I would like to begin first and foremost by thanking my Ph.D. supervisor Paola Scaffidi. Paola, your phenomenal support, superlative critiques and boundless creativity have made my Ph.D. not only scientifically formative and engaging, but also an incredible journey. I am immensely appreciative for the countless hours you have spent diligently proof-reading work, patiently explaining concepts and dishing out sage advice.

I have been unbelievably fortunate to have shared my Ph.D. with a cast of fantastic lab mates/friends. Josep, you have been as much part of my Ph.D. as my pipettes, indispensable but only good with small amounts of liquid… In all seriousness, I am lucky to have had someone with your sense of humour and drive to do my Ph.D. alongside. Cristina, you are the bedrock of this lab. Your support and advice has enabled me to overcome so many of my Ph.D.’s twists and turns I think I’ve lost count... Not only that, but working with you has been amazing fun! Elanor, you are a consummate scientist, and if I can achieve just a fraction of your professionalism and skill I will surely retire a happy man. Thanks for everything you’ve helped me with and all the great times along the way! Giannis, Puay, Marina and Thomas, it’s been a pleasure spending days, nights, weekends and everything in-between in the lab with you all. You are top notch scientists and are going to achieve incredible things. Of course a huge thanks to all past lab members, especially to Matt Burney whose tutoring in all things R and command line was invaluable for this project’s success.

I want to thank everyone in the FACS, Advanced sequencing, Genomics, Histopathology and Bioinformatics STPs, without whom none of this would have been possible. In particular, Harshil Patel for imparting his computational wisdom to help initiate me into the world of bioinformatics.

Claire, you may never know what EZH2 is, but you have helped me through every stage of this journey. I am enormously indebted to you.

Table of contents

Declaration…...... 2 Abstract...... 3 Impact Statement ...... 4 Acknowledgements ...... 6 Table of contents ...... 7 Table of figures ...... 11 Abbreviations ...... 14 Chapter 1. Introduction ...... 17 1.1 Architecture and mechanisms of Polycomb repressive complexes ...... 18 1.1.1 PRC2 composition and function ...... 19 1.1.2 PRC1 composition and activity ...... 24 1.1.3 PRC2 targeting to chromatin ...... 25 1.1.4 Synergistic activity of PRC1 and PRC2 ...... 26 1.1.5 Complex interplay between PRC2 and DNA methylation ...... 28 1.2 PRC2 in development and tissue maintenance ...... 29 1.2.1 PRC2 in mammalian development ...... 29 1.2.2 PRC2 in tissue maintenance ...... 32 1.3 Epigenetic mechanisms and cancer ...... 34 1.3.1 Mutagenic disruption of epigenetic mechanisms ...... 34 1.3.2 Tumour promotion by wild-type epigenetic regulators ...... 37 1.4 PRC2 in cancer ...... 38 1.4.1 Tumour promotion by wild-type PRC2 ...... 38 1.4.2 Oncogenic PRC2 mutations ...... 40 1.4.3 Tumour suppression by PRC2 ...... 40 1.4.4 PRC2 in neural malignancies ...... 42 1.4.5 Therapeutic targeting of PRC2 in cancer ...... 47

Chapter 2. Materials & Methods ...... 49 2.1 Molecular biology techniques ...... 49 2.1.1 Genomic DNA extraction ...... 49 2.1.2 Polymerase chain reaction ...... 49 2.1.3 Restriction enzyme digests ...... 50 2.1.4 Gel electrophoresis ...... 50 2.1.5 Column purification of DNA ...... 51 2.1.6 DNA ligation ...... 51 2.1.7 Transforming competent bacteria ...... 51 2.1.8 Plasmid extraction ...... 52 2.1.9 Generation and source of plasmid constructs ...... 52 2.1.10 RNA extraction ...... 54 2.1.11 Reverse transcription and ChIP quantitative PCR ...... 54 2.1.12 DNA methylation analysis ...... 55 2.2 Cell culture and virus production ...... 56 2.2.1 Cell culturing conditions ...... 56 2.2.2 Production of lentivirus and transduction of cell lines ...... 56

2.2.3 Fluorescence activated cell sorting (FACS) ...... 57 2.2.4 Generation and assessment of polyclonal knock-out cell lines ... 57 2.2.5 Generation of clonal EZH2 knock-out cell lines ...... 57 2.2.6 Proliferation assays ...... 58 2.3 Protein immunodetection ...... 58 2.4 Tumour xenograft studies ...... 59 2.4.1 Tumour transplantation assays ...... 59 2.4.2 Tumour cell isolation and derivation of cell lines ...... 60 2.5 Chromatin immunoprecipitation sequencing experiments ...... 61 2.5.1 Chromatin immunoprecipitation (ChIP) ...... 61 2.5.2 ChIP library preparation and sequencing...... 62 2.5.3 ChIP-seq analysis ...... 62 2.6 RNA sequencing experiments ...... 65 2.6.1 RNA library preparation and sequencing ...... 65 2.6.2 RNA-seq analysis ...... 66 2.7 RNA in situ hybridization ...... 67 2.8 Proteomics experiments ...... 68 2.8.1 Stable isotope labelling with amino acids in cell culture (SILAC) 68 2.8.2 Immunoprecipitation ...... 68 2.8.3 Mass spectrometry analysis ...... 68 2.9 Analysis of publically available datasets ...... 69 2.9.1 Generation of ChIP-seq correlation heatmaps ...... 69 2.9.2 Analysis of DNA methylation in cancer cell lines ...... 70 2.10 Human tissue samples ...... 70 2.11 Statistical analyses ...... 70

Chapter 3. Wild-type EZH2 re-wires neural development programmes to drive glioblastoma ...... 71 3.1 Modelling neural neoplastic transformation ...... 74 3.2 Neoplastic transformation alters the chromatin binding profile of EZH2 ...... 78 3.2.1 Total EZH2 activity on chromatin is unaffected by transformation ...... 78 3.2.2 Profiling EZH2 and H3K27me3 genome-wide distribution by ChIP- seq ...... 79 3.2.3 Neoplastic transformation redistributes EZH2 on chromatin ...... 82 3.2.4 A fraction of transformation-induced changes to EZH2 binding are high-confidence ...... 86 3.2.5 EZH2 targeting to CpG islands is altered by transformation ...... 88 3.2.6 Differential binding of EZH2 is enriched at genes essential for neural development ...... 89 3.3 Transformation-induced redistribution of EZH2 affects the expression of a small, but important group of genes ...... 92 3.3.1 Transformation substantially alters the EZH2-regulated transcriptome ...... 92 3.3.2 DNA methylation may act to redundantly repress EZH2-bound genes ...... 95 3.3.3 A fraction of genes differentially bound by EZH2 undergo changes in expression ...... 101 3.4 Redistribution of EZH2 induces a transcriptional switch at neural homeotic genes ...... 105 3.4.1 EZH2 redistribution represses EMX2 and derepresses HOXB9 105

3.4.2 Transformation-induced repression of EMX2 likely involves the loss of enhancer activity ...... 110 3.4.3 A switch from EMX2 to HOXB9 expression occurs during transformation in glioblastoma ...... 112 3.4.4 EZH2 mediates EMX2 silencing in glioblastoma ...... 115 3.4.5 EZH2 mediates EMX2 repression in many cancer types ...... 117 3.4.6 EMX2 and HOX genes are misregulated in glioblastoma patients ...... 118 3.4.7 EZH2 mediates EMX2 repression in glioblastoma tumours ...... 121 3.4.8 Repression of EMX2 by EZH2 may be necessary for tumour maintenance in glioblastoma ...... 128

Chapter 4. PRC2 composition is altered by neoplastic transformation ...... 131 4.1 Neoplastic transformation alters the composition of PRC2 ...... 134 4.1.1 Immunoprecipitation of endogenous EZH2 captures all known PRC2 interactors ...... 134 4.1.2 Transformation alters the protein composition of PRC2 ...... 136 4.1.3 Gene expression changes may mediate a transformation-induced shift in PRC2 composition ...... 139 4.2 Sub-stoichiometric PRC2 components have distinct roles in tumours ...... 141 4.2.1 Transformation-induced changes to sub-stoichiometric component binding match their functions in tumours ...... 141 4.2.2 Sub-stoichiometric PRC2 components have context-dependent functions in cancer ...... 143 4.3 Changes to PRC2 composition do not mediate a tumour- promoting redistribution of EZH2 ...... 147

Chapter 5. Systematic comparison of experimental approaches for defining PRC2-regulated genes ...... 149 5.1 Disruption of PRC2 function by EZH2 knock-down, knock-out and pharmacological inhibition ...... 152 5.2 Disrupting PRC2 with different loss-of-function approaches has distinct effects on gene expression ...... 155 5.3 Pharmacological inhibition is the most effective LOF approach for identifying PRC2 target genes ...... 158

Chapter 6. Discussion ...... 164 6.1 EZH2 is hijacked by neoplastic transformation through redistribution on chromatin ...... 164 6.1.1 EZH2 overexpression upon transformation does not drive its tumour promoting role ...... 165 6.1.2 What drives EZH2 redistribution upon transformation? ...... 167 6.1.3 EZH2 redistribution induces only a small number of gene expression changes ...... 168 6.1.4 Mapping cancer-relevant changes in H3K27me3 deposition .... 169 6.1.5 Transformation-induced redistribution: an explanation for EZH2’s dual role in cancer? ...... 170 6.2 EZH2 redistribution switches the expression of important homeotic genes in glioblastoma ...... 172

6.2.1 EZH2 redistribution derepresses HOX genes upon transformation ...... 172 6.2.2 EMX2 repression by EZH2 may be important in glioblastoma maintenance ...... 173 6.2.3 Ras signalling initiates EZH2 redistribution at neural homeotic genes ...... 174 6.3 Transformation induces changes to PRC2 composition that may promote tumour growth ...... 176 6.4 Pharmacological inhibition is an effective tool for defining PRC2- regulated genes ...... 178 6.5 Concluding remarks ...... 179

Chapter 7. Appendix ...... 181 7.1 Appendix 1 - Oligonucleotides used in this study ...... 181 7.2 Appendix 2 - Cell culturing conditions ...... 182 7.3 Appendix 3 - Sources of publically available data ...... 183 7.4 Appendix 4 - ChIP-seq correlation heatmap datasets (Figure. 10) ...... 184 7.5 Appendix 5 - NCI-60 cell lines (Figure. 23B)...... 185

Reference List ...... 186

Table of figures

Figure 1. Composition and activity of the PRC2 ...... 21 Figure 2. Different modalities of PRC2 recruitment to chromatin...... 26 Figure 3. Aberrant expression of lineage specifying factors in the midbrain upon Ezh2 KO ...... 32 Figure 4. Mutation of epigenetic regulators in cancer ...... 36 Figure 5. The diverse functions of PRC2 in cancer ...... 42 Figure 6. Tumour promotion by wild-type EZH2 in neural malignancies ...... 45 Figure 7. Tumour suppression by EZH2 in neural malignancies ...... 46 Figure 8. Antithetic functions of wild-type EZH2 in neural development/tissue maintenance and cancer ...... 72 Figure 9. The model of neoplastic transformation used in this study ... 75 Figure 10. The epigenetic landscapes of astroglial and fibroblastic cells are related ...... 77 Figure 11. Total EZH2 activity is unaffected by neoplastic transformation ...... 78 Figure 12. EZH2 and H3K27me3 ChIP-seq quality control ...... 80 Figure 13. EZH2 and H3K27me3 ChIP-seq generates high quality datasets for the neoplastic transformation model ...... 81 Figure 14. EZH2 genome-wide distribution in de novo transformed cells ...... 83 Figure 15. Transformation induces an extensive redistribution of EZH2 binding ...... 85 Figure 16. Transformation-induced changes in EZH2 binding are primarily low-magnitude ...... 87 Figure 17. EZH2 affinity for GC rich sequences, such as CpG islands, is altered by transformation ...... 89 Figure 18. EZH2 redistribution is focussed at neural development genes ...... 91 Figure 19. Profiling the EZH2-regulated transcriptome using pharmacological EZH2 inhibition ...... 92 Figure 20. Transformation markedly alters the EZH2-regulated transcriptome ...... 94 Figure 21. A fraction of EZH2-bound genes respond to EZH2i ...... 96

Figure 22. EZH2 target genes insensitive to EZH2i are often marked by DNA methylation ...... 98 Figure 23. Redundant repression by DNA methylation may occur at a large portion of EZH2-bound genes ...... 99 Figure 24. Redundant repression by EZH2 and DNA methylation may enhance gene silencing ...... 100 Figure 25. EZH2 redistribution affects the expression of a small set of genes ...... 102 Figure 26. EZH2 does not repress CDKN2A upon transformation ...... 104 Figure 27. EMX2 and HOXB9 expression in the embryonic nervous system ...... 105 Figure 28. EMX2 is repressed and HOXB9 derepressed by EZH2 redistribution ...... 106 Figure 29. Transformation induces a general misregulation of repressive HOX chromatin ...... 107 Figure 30. EMX2 and HOXB9 are bona fide targets of EZH2 ...... 109 Figure 31. Transformation-induced silencing of EMX2 is concurrent with collapse of neighbouring enhancers ...... 111 Figure 32. Neural transformation leads to repression of EMX2 and de- repression of HOXB9 ...... 114 Figure 33. EZH2 maintains EMX2 repression in glioblastoma ...... 116 Figure 34. EZH2 maintains EMX2 repression in many cancer types ... 117 Figure 35. EMX2 is repressed and HOXB9 derepressed in glioblastoma tumours ...... 119 Figure 36. HOX genes are de-repressed in glioblastoma ...... 121 Figure 37. EZH2 and EMX2 expression anti-correlates in patient tumours ...... 123 Figure 38. EMX2 and EZH2 expression has an antithetic relationship in tumour cells ...... 125 Figure 39. EZH2 mediates EMX2 repression at all grades of glioma ... 127 Figure 40. EMX2 re-expression impairs glioblastoma cell growth ...... 129 Figure 41. EMX2 is a potent tumour suppressor in glioblastoma ...... 130 Figure 42. Effect of transformation on the composition of PRC2 ...... 132 Figure 43. Endogenous EZH2 co-immunoprecipitation captures the PRC2 interactome ...... 135 Figure 44. Workflow for quantifying transformation-induced changes in PRC2 composition ...... 136

Figure 45. Transformation induces a shift in PRC2 composition ...... 138 Figure 46. Changes in sub-stoichiometric component expression mirror the shift in PRC2 composition ...... 140 Figure 47. PRC2 components are depleted by shRNAs in transformed cells ...... 141 Figure 48. PRC2 sub-stoichiometric components have divergent effects on tumour growth ...... 142 Figure 49. shRNA knock-down depletes PRC2 components in PDX cell lines ...... 144 Figure 50. PRC2 sub-stoichiometric components have context- dependent roles in cancer ...... 146 Figure 51. A transformation-induced shift in PRC2 composition does not drive EZH2 redistribution ...... 148 Figure 52. Loss-of-function approaches for disrupting PRC2 function ...... 150 Figure 53. PRC2 function is disrupted by genetic and pharmacological LOF approaches ...... 154 Figure 54. EZH2 LOF induces large numbers of small-magnitude changes in gene expression ...... 155 Figure 55. EZH2 LOF by different approaches induces distinct expression changes ...... 157 Figure 56. EZH2i derepresses many more PRC2 target genes than genetic LOF ...... 159 Figure 57. EZH2 KD is insufficient to derepress high-magnitude PRC2 target genes ...... 161 Figure 58. Commonly identified PRC2 target genes are upregulated to a greater extent by EZH2i ...... 163 Figure 59. Model of EZH2 co-option by neoplastic transformation to drive glioblastoma growth ...... 165

Abbreviations

C Cas9 - CRISPR associated protein 9 CCLE - Cancer cell line encyclopaedia ChIP-seq - Chromatin immunoprecipitation sequencing CNS - Central nervous system Co-IP - Co-immunoprecipitation CRISPR - Clustered regularly interspaced short palindromic repeats

D DEG - Differentially expressed gene

E EED - Embryonic ectoderm development ESC - Embryonic stem cell EZH2 - Enhancer of zeste homologue 2 EZH2i - EZH2 pharmacological inhibition

F FAIRE-seq - Formaldehyde-assisted isolation of regulatory elements sequencing

G GBM - Glioblastoma GOF - Gain-of-function GSC - Glioblastoma stem cell GSEA - Gene set enrichment analysis

H H3K27me3 - Tri-methylated lysine 27 of histone H3 H3K27ac - Acetylated lysine 27 of histone H3 H3K4me3 - Tri-methylated lysine 4 of histone H3

HMTase - Histone methyl transferase

I IP - Immunoprecipitation

K KO - Knock-out KD - Knock-down

L LincRNA - Long non-coding RNA LOF - Loss-of-function

M mRNA - Messenger RNA

N NPC - Neural progenitor cell NSC - Neural stem cell

P PCR - Polymerase chain reaction PDX - Patient-derived xenograft PRC1 - Polycomb repressive complex 1 PRC2 - Polycomb repressive complex 2 PTM - Post translational modification

R RBBP4/7 - Retinoblastoma-binding protein 4 or 7 REMBRANDT - Repository of molecular brain neoplasia data RNA-FISH - RNA fluorescence in situ hybridisation RNAi - RNA interference RNA-seq - RNA sequencing RT - Room temperature RT-qPCR - Reverse transcription quantitative PCR

S SAM - S-adenosylmethionine sgRNA - Short-guide RNA shRNA - Short hairpin RNA SILAC - Stable isotope labelling with amino acids in cell culture SUZ12 - Suppressor of zeste 12 SVZ - Sub-ventricular zone

T TSS - Transcription start site

Chapter 1. Introduction

Epigenetic regulation is indispensable for instructing and maintaining correct cellular identity (Atlasi & Stunnenberg, 2017). Through differential modification and compaction of chromatin, epigenetic mechanisms allow the cell to regulate gene expression programmes in a spatially and temporally appropriate manner. These mechanisms are a ubiquitous feature of both normal development and tissue maintenance (Atlasi & Stunnenberg, 2017), and their essentiality is clearly demonstrated by the severe, and often lethal phenotypes, of abrogating epigenetic modifier function (Okano et al, 1999; Yu et al, 1995; O’Carroll et al, 2001). A core feature of the epigenetic regulatory machinery in metazoans are the Polycomb group (PcG) of proteins, which modulate many facets of the epigenetic landscape to shape a range of physiological and pathological processes (Schuettengruber et al, 2017).

1.1 Architecture and mechanisms of Polycomb repressive complexes

PcG genes were first identified nearly 70 years ago with the discovery of Polycomb in pioneering Drosophila mutagenesis experiments (Lewis, 1947). Mutation of Polycomb transformed anterior segments of the Drosophila embryo into more posterior ones, demonstrating an essential role for Polycomb in mediating correct body patterning. Ensuing genetic screens identified many others genes that exhibited similar mutant phenotypes, becoming collectively known as PcG genes. Subsequently, disruption of a second group of genes, the Trithorax group (TrxG), was found to transform posterior embryo segments into more anterior ones, a phenotype opposite to that of PcG genes (Kennison & Tamkun, 1988; Ingham, 1983). It was then established that PcG and TrxG proteins function in an antagonistic manner to maintain the correct expression patterns of body-plan-specifying HOX genes when the initiating transcription factors are removed, explaining the severe defects in body patterning their disruption induced (Lewis, 1978; Struhl & Akam, 1985; Ingham, 1983). Thus, the antagonistic activity of TrxG and PcG proteins provides a paradigmatic mechanism by which cells can retain a ‘memory’ of prior developmental signals (Ingham, 1985). These protein families have since been shown to be evolutionarily conserved from plants to mammals, emphasising the fundamental role they play in regulating the genome (Schuettengruber et al, 2017).

The PcG proteins primarily act as components of two complexes: Polycomb repressive complex 1 (PRC1) and Polycomb repressive complex 2 (PRC2) (Margueron & Reinberg, 2011). This subdivision is evolutionarily conserved, but notably the compositional diversity of PRC1 and 2 has markedly increased in higher organisms such as mammals, in line with their more complex gene regulatory requirements (Schuettengruber et al, 2017). For the purposes of this thesis, I will focus primarily on the mammalian Polycomb complexes, in particular PRC2.

1.1.1 PRC2 composition and function

1.1.1.1 PRC2 core complex

The core PRC2 complex in mammals is composed of the proteins enhancer of zeste homologue 1 or 2 (EZH1/2), suppressor of zeste 12 (SUZ12), embryonic ectoderm development (EED) and retinoblastoma-binding protein 4 or 7 (RBBP4/7) (Schuettengruber et al, 2017) (Fig. 1). EZH1 and EZH2 act in a mutually exclusive manner as the catalytic components of PRC2, transferring a methyl group from the cofactor S-adenosyl-L-methionine to lysine 27 on Histone H3 through the activity of their SET domains (Margueron & Reinberg, 2011). The histone methyl transferase (HMTase) activity of PRC2 was initially identified through a combination of biochemical purification, mutagenesis and chromatin immunoprecipitation experiments performed in Drosophila (Müller et al, 2002; Cao et al, 2002; Czermin et al, 2002; Kuzmichev et al, 2002). By assaying the catalytic activity of in vitro reconstituted PRC2, and E(Z) (Drosophila EZH1/2) alone, using nucleosomal arrays, the SET domain of EZH1/2 was shown to be a methyl transferase that specifically catalysed H3K27 methylation. Furthermore, the importance of HMTase activity for gene repression was confirmed by the observed upregulation of PRC2 targets, such as HOX genes, when the SET domain of E(Z) was mutated in vivo. Subsequent studies have shown that EZH1/2 catalyse sequential mono- (H3K27me1), di- (H3K27me2), and tri-methylation (H3K27me3) of lysine 27, with deposition of H3K27me3 at gene bodies and promoters acting as PRC2’s canonical mode of gene repression (Margueron & Reinberg, 2011; Pengelly et al, 2013). Removal of these methyl marks from H3K27 is catalysed through the activity of histone demethylases UTX and JMJD3 (Hong et al, 2007; Agger et al, 2007).

Although most research has focussed on the role of H3K27me3, recent evidence suggests that H3K27me1 and H3K27me2 are also functionally important chromatin marks. H3K27me1 is found at gene bodies and H3K27me2 covers large intergenic regions, where they promote transcription and prevent pervasive intergenic transcription, respectively (Lee et al, 2015; Ferrari et al, 2014). In line with their functions, the levels of the three modifications differ greatly; with H3K27me1, H3K27me2 and H3K27me3 found at ~4%, ~70% and ~7% of histone H3, respectively (Ferrari et al, 2014), indicating that di-methylated is the default state for histone H3.

It has been extensively reported that EZH1 and EZH2 are able to act in a redundant manner to repress genes (Ezhkova et al, 2011; Lui et al, 2016; Shen et al, 2008). However, PRC2-EZH2 has a substantially higher HMTase activity than PRC2-EZH1 (Margueron et al, 2008), and is activated by allosteric modulators and specific epigenetic features to which PRC2-EZH1 is not responsive (Lee et al, 2018a). In addition, PRC2- EZH1 has a higher affinity for nucleosomes and is able to induce gene repression independently of its catalytic activity (Margueron et al, 2008). Alongside these differing mechanisms of action, EZH1 and EZH2 exhibit divergent expression patterns across different tissue and cell types, and have distinct expression dynamics during cell state transitions (Laible et al, 1997; Ogawa et al, 1998; van Lohuizen et al, 1998; Ezhkova et al, 2011; von Schimmelmann et al, 2016). Specifically, EZH2 expression is restricted to proliferating cells, whilst EZH1 is ubiquitously expressed irrespective of proliferation rate in both proliferating and post-mitotic cells (Margueron et al, 2008). These expression differences are partly due to the linkage of EZH2, but not EZH1, expression to cellular proliferation rate through regulation of its transcription by the E2F pathway (Bracken et al, 2003). Thus, although PRC2-EZH1 and PRC2-EZH2 can act redundantly, they also have distinct mechanisms of action and functionalities.

SUZ12 and EED are also essential for PRC2’s HMTase activity, and loss or inhibition of these components severely impairs PRC2 function (Pasini et al, 2004; Qi et al, 2017). As with many PRC2 components, the role of SUZ12 in the PRC2 is remarkably multi- factorial. SUZ12 interacts with EZH2 to drive assembly of PRC2, promotes EZH2’s HMTase activity and enables allosteric regulation of HMTase activity by the active histone mark H3K4me3 (Ketel et al, 2005; Cao & Zhang, 2004; Schmitges et al, 2011). In addition, SUZ12 binds to non-coding RNAs, an interaction that regulates the association of PRC2 with genes (see: ‘PRC2 targeting to chromatin’) (Kanhere et al, 2010). EED binds to H3K27me3 via its WD40 domain, and in turn, allosterically activates the HMTase activity of EZH2 (Margueron et al, 2009). This allosteric activation allows the propagation of the H3K27me3 mark along chromatin, enabling maintenance of repressive domains and transmission of the mark from mother to daughter cells (Fig. 1). In addition to EED and SUZ12, proteins RBBP4 and 7 form a constitutive part of the PRC2 complex. However, dissection of the specific function of these proteins has proved challenging due to the many protein complexes they associate with (Zhang et al, 2013; Migliori et al, 2012). Nonetheless, the ability of these proteins to bind histones appears

to be important for recruiting PRC2 to nucleosomes and mediating allosteric interactions with histone marks (Nowak et al, 2011; Schmitges et al, 2011; Nekrasov et al, 2005).

Figure 1. Composition and activity of the PRC2 A schematic depicting the composition of the PRC2, including the core and accessory proteins. Red arrow indicates methylation of lysine 27 on histone H3 by EZH1/2, black unbroken arrow indicates an allosteric interaction between EED and H3K27me3, black broken arrows indicate accessory proteins interacting with the PRC2 core to form the PRC2.1 and PRC2.2 sub- complexes.

1.1.1.2 PRC2 accessory proteins and interacting RNAs

Associating with the PRC2 core are a number of accessory proteins that bind in a sub- stoichiometric manner and are required for complex targeting and allosteric regulation (Holoch & Margueron, 2017) (Fig. 1). A consensus has now emerged that binding of accessory proteins categorises PRC2 into two distinct sub-complexes: PRC2.1 and PRC2.2 (Hauri et al, 2016; Grijzenhout et al, 2016). PRC2.1 is defined by the mutually exclusive binding of proteins MTF2, PHF19 and PHF1, highly homologous members of the Polycomb-like (PCL) family of proteins. PCL proteins enable PRC2 to bind a number of features of the epigenetic landscape. Through their Tudor domains, PHF1, PHF19 and MTF2 are able to bind H3K36me3, a mark of active transcription (Ballaré et al, 2012; Musselman et al, 2012; Cai et al, 2013; Brien et al, 2012; Li et al, 2017). The interaction with H3K36me3 is hypothesised to allow spread of repressive H3K27me3 into transcriptionally active regions, a model supported by PHF19’s ability to recruit the

H3K36 demethylase NO66 (Brien et al, 2012). However, studies of PHF1’s interaction with H3K36me3 show it to inhibit PRC2 activity (Musselman et al, 2012), in line with the ability of increased H3K36me3 to deplete H3K27me3 (Popovic et al, 2014). Thus, the exact function of PCL proteins interaction with H3K36me3 still remains to be resolved. In addition, through their winged helix domain, PHF1 and MTF2 are able to bind unmethylated GC-rich DNA (Perino et al, 2018; Choi et al, 2017; Li et al, 2017). This interaction is particularly important as PRC2 shows a strong preference for binding at CpG islands (Mendenhall et al, 2010; Ku et al, 2008; Riising et al, 2014), regions of DNA with high CpG dinucleotide content often associated with gene promoters (Deaton & Bird, 2011). This interaction, and its significance in targeting PRC2, will be discussed in greater depth in ‘PRC2 targeting to chromatin’.

In spite of their sequence homology, PCL proteins appear to have distinctive cellular functions and modes of action. PHF1 is highly expressed in quiescent cells, where it binds to and stabilises p53 in a chromatin-independent manner, inducing cellular quiescence (Brien et al, 2015). In contrast, PHF19 and MTF2 are expressed in proliferating cells due to regulation of their transcription by the E2F pathway (Brien et al, 2015). In embryonic stem cells (ESCs), depletion of MTF2 leads to enhanced self- renewal characteristics (Walker et al, 2010), whilst depletion of PHF19 increases spontaneous differentiation (Hunkapiller et al, 2012). Moreover, in an in vitro model of neural differentiation, PRC2-MTF2 is enriched in ESCs, whilst a switch to PRC2-PHF19 is initiated upon differentiation to neural progenitor cells (Kloet et al, 2016). Taken together, this indicates striking differences in function between PCL proteins, the mechanistic underpinnings of which still remain to be fully elucidated.

Adding additional complexity, PRC2.1 can be further sub-divided by whether it interacts with EPOP or PALI1/2, two recently identified PRC2 interactors (Conway et al, 2018; Beringer et al, 2016; Liefke et al, 2016). EPOP is a highly unstructured protein which links PRC2 to elongin BC, part of the transcription elongation-promoting elongin complex (Liefke et al, 2016; Beringer et al, 2016). This interaction is proposed to restrict PRC2 activity and maintain a basal level of transcription from PRC2-repressed genes, potentially allowing them to escape PRC2 repression with greater ease when the appropriate signals are received. PALI1/2 are transcribed from within the locus of the nuclear co-repressor gene LCOR, and remain the least characterised PRC2 accessory proteins (Conway et al, 2018). Recent work has shown that PALI1/2 are essential for

mouse development and PRC2-PALI1/2 has an antagonistic relationship with PRC2- AEBP2, suggesting the existence of complex interactions between sub-classes of PRC2 (Conway et al, 2018). PRC2.2 is defined by concurrent binding of the proteins AEBP2 and JARID2 to the PRC2 core (Hauri et al, 2016; Grijzenhout et al, 2016). JARID2 is a member of the Jumonji C family of histone demethylases, but due to a lack of key residues JARID2 is catalytically inactive (Landeira & Fisher, 2011). Through its zinc finger domain, JARID2 exhibits weak binding affinity for GC-rich sequences of DNA, suggesting a mechanism by which PRC2 is targeted to CpG islands (Li et al, 2010). Also important for PRC2 targeting, JARID2 is able to bind non-coding RNAs and thus serve as a major conduit for PRC2 regulation by RNA (Kaneko et al, 2014a). Furthermore, JARID2 is tri-methylated by EZH2 at its K116 residue, a modification that is recognised by EED and is thought to stimulate EZH2 HMTase activity in a similar manner to H3K27me3 recognition by EED (Sanulli et al, 2015). This suggests a mechanism of PRC2 auto-activation that may allow de novo deposition of H3K27me3. In contrast to JARID2’s ability to promote PRC2 activity, AEBP2 appears to constrain it. This is evidenced by the increased H3K27me3 levels upon AEBP2 knock-out in ESCs and a homeotic transformation observed in AEBP2 KO mice that is reminiscent of those induced by loss of TrxG proteins (Grijzenhout et al, 2016). Similar to JARID2, AEBP2 is also a zinc finger-containing protein, and is suggested to employ this DNA affinity to target PRC2 on chromatin (Kim et al, 2009), however, convincing evidence demonstrating this is still lacking.

Compounding the complex architecture of PRC2, core and accessory proteins of the complex are able to interact with RNA (Davidovich & Cech, 2015). However, the exact function of RNA binding by PRC2 remains contentious. Initial studies identified interactions between PRC2 and non-coding RNAs RepA and HOTAIR as necessary for PRC2 targeting to the X chromosome and HOXD locus, respectively (Ogawa et al, 2008; Rinn et al, 2007). Although an attractive model for PRC2 targeting, subsequent studies have cast doubt on the veracity of these findings (Portoso et al, 2017; McHugh et al, 2015). Instead, PRC2 appears to be an extremely promiscuous binder of RNA (Zhao et al, 2010; Davidovich et al, 2013). This non-specific RNA interaction is thought to prevent PRC2 association at actively transcribing genes (Cifuentes-Rojas et al, 2014; Kaneko et al, 2014b), a hypothesis supported by the de novo binding of PRC2 to CpG islands after RNA depletion by ribonuclease treatment (Beltran et al, 2016).

1.1.2 PRC1 composition and activity

Similar to PRC2, PRC1 constitutes a core complex that shows strong evolutionarily conservation (Schuettengruber et al, 2017). The PRC1 core is formed by RING 1 proteins (RING1A and B), and one Polycomb group ring-finger domain protein (PCGF1- 6) (Gil & O’Loghlen, 2014). The E3 ubiquitin ligase activity of RING1A and RING1B catalyses mono-ubiquitination of lysine 119 of histone H2A (H2AK119ub), a chromatin mark which then mediates repression of the associated gene (Simon & Kingston, 2013). The ubiquitin-protein transferase activity of PRC1 was first identified through prospectively purifying the complex responsible for the then known H2AK119ub chromatin mark (Wang et al, 2004a). The role of this catalytic activity in gene repression was confirmed by the observation of Ubx upregulation (a known Drosophila PRC1 target gene) when levels of the Drosophila PRC1 catalytic component dRing were depleted using RNA interference (Wang et al, 2004a).

PRC1 has much greater compositional diversity than PRC2, and the complex can be broadly subdivided into canonical and non-canonical forms (cPRC1 and ncPRC1, respectively) (Gil & O’Loghlen, 2014). In addition to the core complex, cPRC1 contains one Chromobox protein (CBX2, CBX4, and CBX6–CBX8) and one Polyhomeotic homologous protein (PHC1-3). CBX proteins are able to bind H3K27me3 through their Chromodomain, leading to deposition of H2AK119ub at PRC2-repressed domains (Bernstein et al, 2006b). PHC proteins contain a SAM motif that allows them to self- associate and drive clustering of polycomb repressed domains (Isono et al, 2013). In contrast, ncPRC1 is defined by the binding of the protein RYBP to the PRC1 core. Alongside RYBP, a suite of accessory proteins associate with the ncPRC1 core, producing different versions of ncPRC1, a full description of which are outside the scope of this thesis (Blackledge et al, 2015). A key feature of ncPRC1 complexes is the absence of a CBX protein, meaning ncPRC1 is targeted to chromatin in a PRC2- independent manner. Although the mechanisms underlying ncPRC1 targeting remain to be fully defined, the discovery that ncPRC1 component KDM2B is able to bind CpG islands suggests a possible mechanism by which ncPRC1 associates with specific loci (Farcas et al, 2012).

1.1.3 PRC2 targeting to chromatin

In Drosophila, PRC2 activity is targeted to chromatin through transcription factor-directed binding of cis-regulatory sequences known as Polycomb responsive elements (PREs) (Steffen & Ringrose, 2014). However, attempts to identify a mammalian PRE have proved largely fruit(fly)less (Blackledge et al, 2015). Moreover, efforts to find transcription factors that target PRC2 in a sequence specific manner have uncovered only a handful of examples, all of which are relevant at specific genes and in restricted cellular contexts (Arnold et al, 2013; Dietrich et al, 2012; Herranz et al, 2008; Blackledge et al, 2015). Thus, loci-specific targeting does not appear to be a broadly relevant mechanism of directing PRC2 to chromatin.

One of the clearest determinants of whether Polycomb complexes will bind on DNA is the presence of a CpG island (Mendenhall et al, 2010; Ku et al, 2008), an association thought to be mediated by PCL proteins or JARID2 on PRC2, and KDM2B on PRC1 (Li et al, 2010, 2017; Farcas et al, 2012). Furthermore, two important studies investigating the temporal dynamics of PRC2 binding showed that gene repression precedes the binding of PRC2 (Hosogane et al, 2013; Riising et al, 2014). Specifically, in the work of Hosogane et al, deposition or loss of H3K27me3 at gene bodies was demonstrated to occur after Ras induced expression changes at these genes in mouse fibroblasts, indicating that PRC2 responds to expression changes as opposed to initiating them. In addition, global inhibition of transcription leads to recruitment of PRC2 to CpG islands genome-wide (Riising et al, 2014), suggesting that PRC2 binds to CpG islands by default, but is inhibited from doing so by transcription. This general affinity of PRC2 for CpG islands, its ability to recognise and respond to many chromatin features, and the paucity of evidence for sequence-specific targeting, led to the proposal of a ‘chromatin sampling’ model of Polycomb complex targeting (Klose et al, 2013). In this model, Polycomb complexes constantly sample the chromatin environment at CpG islands, incorporating information on transcriptional activity, chromatin modification and DNA methylation status to determine whether a site is bound and histone modification occurs. This model is supported by the ability of RNA to inhibit PRC2 binding (Davidovich et al, 2013; Beltran et al, 2016), and allosteric regulation of PRC2 by a multitude of histone modifications (Holoch & Margueron, 2017). Thus, PRC2 binding to chromatin is likely to be determined by a combination of general affinity for CpG islands, and the integration of the epigenetic features at each locus.

Figure 2. Different modalities of PRC2 recruitment to chromatin A schematic outlining the different mechanisms by which PRC2 HMTase activity is targeted to chromatin. Red arrows indicate deposition of a chromatin mark and black arrows indicate binding of a chromatin mark. Coloured spheres represent histones and ‘PCL’ indicates the proteins PHF19, MTF2 and PHF1. Txn factor – transcription factor. Figure based on a similar schematic from Aranda et al, 2015.

1.1.4 Synergistic activity of PRC1 and PRC2

PRC1 and PRC2, along with their repressive chromatin marks, show a highly concordant distribution on chromatin, indicating an intimate link between their repressive activity. Early studies exploring the interaction between PRC1 and PRC2 in Drosophila led to the proposal of a ‘hierarchical’ model of Polycomb complex activity, whereby PRC2 activity on chromatin leads to recruitment of PRC1 (Wang et al, 2004b). This model was based on initial experiments exploring PRC1 and PRC2 recruitment at the polycomb target gene Ubx (Wang et al, 2004b). At Ubx, PRC2 was shown to be recruited to PREs through

an interaction with PRE-binding transcription factors, such as Pho. This recruitment led to deposition of H3K27me3, which was bound by PRC1 via its H3K27me3-binding Pc component (CBX proteins in mammals), in turn leading to gene repression. Although an attractive model, recent studies have revealed that this hierarchical model is insufficient to fully describe the complex interaction between PRC1 and PRC2 (Blackledge et al, 2015). Targeting of PRC1 to loci devoid of PRC2 and H3K27me3 drives de novo formation of H3K27me3 domains, and depletion of PRC1 leads to a decrease in global H3K27me3 levels, together indicating that PRC1 can also recruit PRC2 to chromatin (Blackledge et al, 2014; Tavares et al, 2012; Cooper et al, 2014). PRC1-driven targeting has been proposed to act as follows: 1) ncPRC1 binds to CpG islands in a H3K27me3- independent manner through its KDM2B component and deposits H2AK119ub (Farcas et al, 2012); 2) PRC2.2 binds H2AK119ub via JARID2 leading to deposition of H3K27me3 (Kalb et al, 2014); 3) cPRC1 is recruited by CBX protein binding to H3K27me3, depositing further H2AK119ub and initiating a positive feedback loop to expand the Polycomb repressive domain (Holoch & Margueron, 2017).

The modes by which PRC1 and PRC2 mediate gene repression are remarkably diverse (Schuettengruber et al, 2017). Chromatin compaction by PRC1 is particularly important in repressing gene expression (Shao et al, 1999), and this occurs through interactions between specific ‘compaction regions’ on CBX proteins which act to bridge adjacent nucleosomes(Francis et al, 2004; Lau et al, 2017). In addition, Polycomb complexes are able to directly disrupt transcription through either blocking the release of RNA polymerase II (Dellino et al, 2004; Zhou et al, 2008) or inhibition of the catalytic activity of H3K27 acetyl transferase CBP (Tie et al, 2016). Alongside effects on local chromatin architecture, PHC proteins of the cPRC1 form homotypic interactions that drive higher- order clustering of Polycomb bound domains, a process crucial for enabling Polycomb mediated gene repression (Isono et al, 2013). Thus, PRC1 and PRC2 exhibit a complex molecular interplay which enables recruitment of both complexes to repressed loci which then drives gene repression by diverse modalities.

1.1.5 Complex interplay between PRC2 and DNA methylation

Of the many chromatin features PRC2 targeting has been linked to, a particularly strong anti-correlative relationship is observed between DNA methylation and H3K27me3, leading to the suggestion that DNA methylation may make a sequence refractory to PRC2 binding (Rose & Klose, 2014). Supporting this, PRC2 binds primarily at CpG islands which are largely unmethylated regions of the genome (Mendenhall et al, 2010; Ku et al, 2008), and depletion of DNA methylation is often followed by H3K27me3 deposition at newly hypomethylated loci (Lynch et al, 2012; Reddington et al, 2013). In addition, PHF1 and MTF2 selectively bind unmethylated CpG-containing DNA (Li et al, 2017), indicating a possible mechanism underlying DNA methylation sensitivity. Moreover, a direct interaction exists between DNA demethylase TET1 and PRC2 in ESCs, potentially allowing PRC2 to actively remove DNA methylation from its binding sites (Neri et al, 2013a). However, other studies suggest that DNA methylation and PRC2 activity are by no means incompatible. Both H3K27me3 and DNA methylation co-exist on the inactive X chromosome (Galupa & Heard, 2015), and targeting of PRC1 to hypermethylated pericentromeric chromatin leads to binding of PRC2 and formation of H3K27me3 domains (Cooper et al, 2014). Furthermore, a direct interaction has been demonstrated between the PRC2 and DNA methylation machinery (Viré et al, 2006), although subsequent work has suggested this interaction may be inhibited by competition with DNMT3L (Neri et al, 2013b). Importantly, the majority of studies exploring the relationship between DNA methylation and PRC2 have focussed on ESCs where a clear anti-correlative relationship does appear to exist. However, in instances when the relationship has been studied in the context of differentiated or cancer cells the relationship becomes less clear (Brinkman et al, 2012; Statham et al, 2012). Thus, the exact relationship between DNA methylation and PRC2 is complex and likely context- dependent, with clear inconsistencies in existing data which are yet to be fully resolved.

1.2 PRC2 in development and tissue maintenance

1.2.1 PRC2 in mammalian development

Since the initial observation in Drosophila that PcG proteins are essential for correct body patterning, subsequent research has established Polycomb complexes as a core conserved feature of developmental programmes in metazoans. In mammals, the importance of PRC2 function in development is exemplified by the embryonic lethality of disrupting core PRC2 components in mice (O’Carroll et al, 2001; Pasini et al, 2004; Faust et al, 1998), and the severe effects of accessory components loss (Vizán et al, 2015). Importantly, embryonic lethality occurs before gastrulation, demonstrating that PRC2 is required from the very earliest stages of mammalian development. In addition to mediating early development, tissue-specific roles for PRC2 have been identified in blood (Xie et al, 2014), brain (Pereira et al, 2010), muscle (Caretti et al, 2004), skin (Ezhkova et al, 2009) and heart development (Delgado-Olguín et al, 2012; He et al, 2012). Together, this indicates an indispensable role for PRC2 throughout mammalian development.

1.2.1.1 PRC2 in embryonic stem cells

As a result of the crucial role PRC2 plays in early embryogenesis, many subsequent studies have used ESCs as a model to unravel PRC2’s functions in the first stages of development. In initial experiments disrupting PRC2 core components in mouse ESCs, substantial de-repression of lineage-specific genes was observed (Boyer et al, 2006; Lee et al, 2006), leading to the conclusion that PRC2 maintained stem cell self-renewal. However, contradicting this assertion, ESCs depleted of core PRC2 components can be maintained in a stable undifferentiated state, but have a severely impaired ability to differentiate in a timely and correct manner (Pasini et al, 2007; Chamberlain et al, 2008). This suggests that PRC2 is dispensable for maintenance of ESC self-renewal, but necessary to mediate differentiation to specific lineages. Further supporting a role for PRC2 in enabling differentiation, PRC2 accessory proteins PHF19, MTF2 and JARID2 are also required for differentiation in ESCs (Ballaré et al, 2012; Zhang et al, 2011; Pasini et al, 2010). Nonetheless, a recent study investigating the effect of EZH2 knock-out (KO) in human ESCs found severe defects in both self-renewal and differentiation ability,

suggesting that the function of PRC2 may show a degree of species specificity (Collinson et al, 2016).

1.2.1.2 Dynamic redistribution of PRC2 during development

In studies profiling the chromatin landscape of ESCs, a striking co-localisation was observed between H3K27me3 and H3K4me3 (a chromatin mark associated with transcriptional activation) at the promoters of key developmental genes (Bernstein et al, 2006a). These regions, termed ‘bivalent domains’, resolved to the repressive or activating mark upon differentiation, an event hypothesised to enable rapid initiation of lineage-specific transcriptional programmes (Bernstein et al, 2006a). In line with the ‘poised’ state of these genes, promoters with bivalent domains are also occupied by paused RNA polymerase II (Brookes & Pombo, 2009). However, depletion of MLL2, the protein depositing H3K4me3 at bivalent domains, has no effect upon a cells ability to induce transcriptional changes when differentiation is induced (Denissov et al, 2014; Hu et al, 2013). Moreover, paused RNA polymerase II is a ubiquitous feature of PRC2 repressed sites even after differentiation has been initiated (Ferrai et al, 2017; Brookes et al, 2012). Thus, the exact function of bivalent chromatin domains in ESCs still remains to be fully determined. Interestingly, in adult gut epithelium, genes marked with bivalent domains in villus cells are those which have been repressed by PRC2 during differentiation from crypt stem cells (Jadhav et al, 2016). This indicates that bivalent domains are features of both embryonic and adult cell populations, with diverse functions in regulating gene expression.

Alongside resolution of bivalent domains, redistribution of H3K27me3 is also observed at other chromatin sites during early development (Gifford et al, 2013; Xie et al, 2013; Hawkins et al, 2010). During differentiation from ESCs to the three germ layers (ectoderm, mesoderm and endoderm), H3K27me3 is observed to replace DNA methylation within many CpG poor regions (Gifford et al, 2013), once again emphasising the intimate relationship between PRC2 and DNA methylation. In addition, comparison of H3K27me3 distribution between human ESCs and various differentiated cell types shows a general expansion of H3K27me3 domains induced by differentiation (Hawkins et al, 2010). Thus, dynamic changes in H3K27me3 distribution are a recurrent feature of early developmental processes.

1.2.1.3 PRC2 in neural development

Beyond ESCs, one of the best studied roles of PRC2 in mammals is its regulation of central nervous system (CNS) development. Proper development and maintenance of the CNS is reliant upon the appropriate activity of epigenetic regulators (Lister et al, 2013), as seen in the epigenetic basis of many neurological diseases (Jakovcevski & Akbarian, 2012). During CNS development, epigenetic mechanisms interpret external cues to give patterned expression of region-specific transcription factors, a process essential in forming a correctly structured CNS (Barber & Rastegar, 2010). Concomitantly, epigenetic regulators initiate and maintain stem cell hierarchies in the CNS, inducing spatially and temporally appropriate cellular differentiation (Juliandi et al, 2010). PRC2 is involved in both aspects of CNS development, performing crucial functions in the developing forebrain, midbrain, hindbrain and spinal cord (Hirabayashi et al, 2009; Zemke et al, 2015; Pereira et al, 2010; Di Meglio et al, 2013).

In the forebrain, EZH2 maintains the self-renewal ability of neural progenitor cells (NPCs) (Pereira et al, 2010), and restricts NPC neurogenic ability to promote a transition from neurogenesis to astrogenesis, a crucial switch during cerebral cortex development (Hirabayashi et al, 2009). In a similar manner to ESC differentiation, NPC differentiation to neuronal and astrocytic lineages is accompanied by H3K27me3 changes at key lineage-specifying genes, such as transcription factors Eomes and Sox6 (Albert et al, 2017). Repression of lineage-specifying factors also underlies PRC2’s role in the developing mid-brain (Zemke et al, 2015). Here, EZH2 represses forebrain-specifying transcription factors such as Foxg1 and Pax6 to maintain midbrain identity (Fig. 3).

As in Drosophila body patterning, PRC2 is an essential regulator of HOX gene expression in neural development. In the pontine nucleus of the hindbrain, EZH2 defines spatially restricted HOX gene expression patterns to regulate neuronal migration and control connectivity between brain regions (Di Meglio et al, 2013). Moreover, in the spinal cord, EZH2 contributes to motor neuron subtype specification by repressing HOX genes in a region-specific manner and thus maintaining their distinct expression patterns (Mazzoni et al, 2013). Together, PRC2 is essential across the developing CNS to repress lineage-specifying factors and balance cellular self-renewal and differentiation.

Figure 3. Aberrant expression of lineage specifying factors in the midbrain upon Ezh2 KO Microscopy images reproduced from Fig. 4A of Zemke et al, 2015, showing upregulation of forebrain transcription factor Foxg1 upon conditional knock-out of Ezh2 in the murine midbrain. Asterisk indicates region in schematic of the mouse brain (top left) where the microscopy images were taken from. Scale bars: 500μm (upper panel), 40μm (lower panel). FB – Forebrain, MB – Midbrain. For additional details please see original publication.

1.2.2 PRC2 in tissue maintenance

PRC2 is also required in post-natal life to mediate tissue maintenance by regulating adult stem cells (Hwang et al, 2014; Rhodes et al, 2018; Ezhkova et al, 2011; Juan et al, 2011) and their post-mitotic progeny (Bodega et al, 2017; Ai et al, 2017). In adult stem cells, PRC2 acts to regulate self-renewal/differentiation and proliferation, a dual function that can be clearly observed in neural stem cells (NSCs) (Hwang et al, 2014). In the adult brain, NSCs constitute a small population of astrocytes, found within a region of the cerebrum known as the sub-ventricular zone (SVZ), which retain neurogenic capacity throughout adult life (Bond et al, 2015). EZH2 is expressed in NSCs, and its KO impairs NSC proliferation due to upregulation of the anti-proliferative gene Cdkn2a (Hwang et al, 2014). However, although proliferation of EZH2-/- NSCs can be rescued by depleting Cdkn2a, the neurogenic capacity of NSCs remains impaired, indicating that EZH2 plays separable roles in regulating proliferation and self-renewal/differentiation (Hwang et al, 2014). Further supporting the importance of PRC2 in regulating these two facets of adult stem cell biology, dual EZH1 and EZH2 KO in mouse skin leads to progressive degeneration of hair follicles due to de-repression of the Cdkn2a locus, an event that is

also accompanied by de-repression of non-skin lineage-specific genes (Ezhkova et al, 2011). Moreover, depletion of EZH2 in skeletal muscle stem cells leads to defects in stem cell proliferation and concomitant de-repression of genes associated with non- muscle cell lineages (Juan et al, 2011). Thus, PRC2 is essential for ensuring tissue homeostasis through controlling proliferation and self-renewal of adult stem cells.

1.3 Epigenetic mechanisms and cancer

An expansive body of evidence now illustrates that epigenetic mechanisms are frequently disrupted and hijacked in cancer (Shen & Laird, 2013; Wainwright & Scaffidi, 2017). Moreover, targeting epigenetic regulators is emerging as an attractive, and often efficacious, approach for treating previously intractable malignancies (Dawson, 2017). As a result, research is happening apace to gain clinically-relevant insights into how normal epigenetic mechanisms, so crucial in mediating development and tissue maintenance, can be disrupted or co-opted in a tumour promoting manner.

1.3.1 Mutagenic disruption of epigenetic mechanisms

The connection between epigenetics and cancer was first identified nearly 40 years ago with the description of recurrent DNA hypomethylation in cancer versus normal cells (Feinberg & Vogelstein, 1983). Since these initial correlative studies, extensive evidence has accrued outlining the direct functional importance of epigenetic mechanisms in driving all aspects of cancer biology (Flavahan et al, 2017). Particularly significant in establishing this link has been DNA sequencing studies which have exhaustively profiled the mutational landscape of cancer (Kandoth et al, 2013; Forbes et al, 2011), demonstrating that epigenetic modifiers are recurrently mutated genes (Shen & Laird, 2013) (Fig. 4). The frequency with which epigenetic modifiers are mutated is strikingly high both in individual malignancies and across cancer (Fig. 4). For example, the histone methyltransferase MLL is mutated in ~80% of infant leukaemia cases (Aplan, 2006), whilst components of the SWI/SNF chromatin remodelling complex are mutated in ~20% of all cancers (Kadoch et al, 2013). Importantly, in most instances, mutations of epigenetic regulators are loss-of-function (Shen & Laird, 2013) (Fig. 4). This indicates that epigenetic regulators act primarily as tumour suppressors within normal cells, in line with their normal function ensuring the faithful undertaking of development and tissue maintenance.

Mutagenic events disrupt epigenetic modifiers and chromatin components representing all aspects of the epigenetic landscape, including: DNA methylation (Abdel-Wahab et al, 2009), histone modification (Morin et al, 2010), nucleosome remodelling (Kadoch et al, 2013) and higher-order chromatin structures (Hnisz et al, 2016) (Fig. 4). Importantly, in many instances, these mutagenic events are requisite for tumour formation and growth

(Moran-Crusio et al, 2011; Souroullas et al, 2016). Probably the most compelling evidence for the role of epigenetics in cancer has come from ‘oncohistone’ mutations; exquisitely specific mutations that affect the amino acid residues on histone tails targeted by epigenetic regulators (Nacev et al, 2019). For example, the lysine 27 and 36 residues on histone H3 are frequently mutated to methionine in paediatric glioma and sarcoma, respectively (Khuong-Quang et al, 2012; Behjati et al, 2013). Both mutations act in a dominant negative manner to inhibit their respective chromatin modifiers, altering the epigenetic landscape and driving tumour formation (Lewis et al, 2013; Lu et al, 2016).

Mutations often also affect epigenetic modulators, proteins upstream of epigenetic modifiers which act to regulate their activity (Feinberg et al, 2016). An excellent example of such mutations are those affecting the metabolic enzyme isocitrate dehydrogenase 1/2 (IDH1/2) in solid and haematopoietic malignancies (Waitkus et al, 2018). Mutations to IDH1/2 are gain-of-function and lead the mutant protein to produce the oncometabolite 2-hydroxyglutarate, a molecule which causes global changes to the chromatin landscape through competitively inhibiting the α-ketoglutarate-dependent dioxygenases required for DNA and histone de-methylation. In glioma, 2-hydroxyglutarate produced by mutant IDH1 potently inhibits the TET family of 5'-methylcytosine hydroxylases, inducing hypermethylation of cohesin and CTCF binding sites. This in turn leads to disruption of higher order chromatin structure, and enables aberrant gene-enhancer interactions which promote tumour growth through upregulation of the glioma oncogene PDGFRA (Flavahan et al, 2016). Taken together, it is clear that mutation of epigenetic modifiers, chromatin components and epigenetic modulators are an essential event in many cancer types.

Figure 4. Mutation of epigenetic regulators in cancer Figure 4: A heatmap illustrating the frequency and type of mutations affecting epigenetic regulators across cancer. Cancers are grouped as epithelial, haematologic and other cancers. Cancer types: 1-Colorectal, 2-Gastric, 3-Breast, 4-Endometrial (Endometrioid), 5-Endometrial (Serous), 6-Ovarian (Clear Cell), Ovarian (Endometrioid), 8-Kidney (Clear Cell), 9-Bladder, 10- Lung, 11-Liver, 12-Pancreas, 13-Prostate, 14-Head and Neck, 15-Thyroid, 16-Esophageal, 17- NUT Midline Carcinoma, 18-Multiple Myeloma, 19-Diffuse Large B-cell Lymphoma, 20-Follicular Lymphoma, 21-Burkitt Lymphoma, 22-T-cell Lymphoma, 23-CLL, 24-ALL, 25-ALL early T precursor, 26-MDS, 27-MDS/MPN, 28-MPN, 29-AML, 30-Mesothelioma, 31-Melanoma, 32- Malignant Rhabdoid, 33-Pancreatic Neuroendocrine, 34-CNS Primitive Neuroectodermal, 35- GBM, 36-Glioma, 37-DIPG, 38-Medulloblastoma, 39-Meningioma, 40-Spinal Meningioma, 41- Neuroblastoma. Figure adapted from Shen and Laird, 2013.

1.3.2 Tumour promotion by wild-type epigenetic regulators

Remarkably, many epigenetic regulators can act as tumour promoting factors in the absence of any function-altering mutation (Wainwright & Scaffidi, 2017). This tumour- promoting role of wild-type epigenetic regulators has been recurrently observed for different classes of regulators across a range of cancer types (Shi et al, 2013; Suvà et al, 2009; Zuber et al, 2011), and the extent of the phenomenon is likely to be underestimated given the challenges of assigning importance to wild-type unmutated proteins. Strikingly, many epigenetic regulators co-opted to promote tumour formation are also frequently inactivated by loss-of-function mutations, indicating that epigenetic regulators can play apparently antithetic roles between normal and cancer cells (e.g. EZH2, SWI/SNF components).

An archetypal example of wild-type tumour promotion is exhibited by the chromatin reader protein BRD4. In studies investigating MLL-rearranged acute lymphoblastic leukaemia, BRD4 was found to be co-opted by cancer cells to drive expression of the oncogene c-MYC. This transcriptional promotion is crucial for tumour growth and has subsequently been shown to be important in other malignancies (Zuber et al, 2011; Venataraman et al, 2014; Shi et al, 2014; Dawson et al, 2011). Moreover, the clinical relevance of wild-type BRD4’s tumour promoting role is demonstrated by the anti-tumour activity that BRD4 inhibitors exhibit in patients with haematological cancers (Herait et al, 2014). However, the detailed dissection of BRD4’s tumour promoting function is an exception, as largely, our understanding of how wild-type epigenetic regulators promote tumour formation remains incomplete. This lack of knowledge precludes rational therapeutic targeting, and hence, there is an acute need for detailed mechanistic characterisation of how cancer cells hijack epigenetic regulators to promote tumour growth.

1.4 PRC2 in cancer

In recent years, an extensive body of evidence has accumulated showing that core and individual components of the PRC2 are key players in tumour establishment and maintenance (Comet et al, 2016). Remarkably, PRC2 has both tumour suppressive and promoting roles in cancer, indicating that the role of PRC2 is highly context-dependent. In this section I will survey the multitude of roles which PRC2 plays in cancer, focussing in particular on PRC2 in neural malignancies.

1.4.1 Tumour promotion by wild-type PRC2

The link between cancer and PRC2 was first established in prostate cancer, where expression levels of wild-type EZH2 were found to inversely correlate with patient survival, and EZH2 disruption in cancer cell lines impaired proliferation (Varambally et al, 2002). A similar overexpression of EZH2, and a concurrent dependency upon EZH2 activity, was subsequently identified in breast cancer (Kleer et al, 2003), melanoma (Zingg et al, 2015), ovarian cancer (Lu et al, 2010) and many other malignancies (Kim & Roberts, 2016). Supporting the functional importance of EZH2 overexpression, ectopic expression of EZH2 enhances cell proliferation in vitro (Bracken et al, 2003; Kleer et al, 2003) and augments tumour growth in vivo (Zingg et al, 2018; Li et al, 2009b; Gonzalez et al, 2014). In addition, EZH2 amplification is a recurrent event in many cancer types (Comet et al, 2016), and the EZH2-regulatory microRNA miR-101 is frequently silenced/deleted in prostate cancer (Varambally et al, 2008), suggesting potential mechanisms by which EZH2 overexpression might occur. Efforts to identify the key tumour suppressors repressed by EZH2 overexpression have identified many possible candidates, including: WNT agonist DKK1 (Hussain et al, 2009), angiogenesis factor VASH1 (Lu et al, 2010), cell cycle regulator p16/CDKN2A (Bracken et al, 2007; Wilson et al, 2010; Mohammad et al, 2017) and many other genes (Kim & Roberts, 2016). Particular interest has focussed on CDKN2A given its crucial role in the cell cycle (Gil & Peters, 2006) and repression by PRC2 in normal development (Ezhkova et al, 2009; Hwang et al, 2014). However, EZH2 expression is regulated by cellular proliferation rate to maintain H3K27me3 homeostasis (Bracken et al, 2003), and therefore, the observed overexpression in many instances likely represents a response to cancer cell hyperproliferation (Wassef & Margueron, 2017). Further, increased EZH2 levels in cancer are rarely accompanied by a matched growth in global H3K27me3 levels, supporting a

homeostatic role for EZH2 overexpression in hyper-proliferating cancer cells (Wassef & Margueron, 2017). Together, this demonstrates that functional-dependency on wild-type EZH2 is a recurrent feature of many cancer types, however, whether this is endowed by EZH2 overexpression is unclear.

In many instances, PRC2 gains tumour promoting ability due to mutational events which remove constraints on its normal activity. An excellent example of this is seen in cancers harbouring LOF mutations in the SWI/SNF chromatin remodelling complex (Wilson et al, 2010; Bitler et al, 2015; Fillmore et al, 2015; Kim et al, 2015). Here, the functional dependency on EZH2 is thought to arise through expansion of EZH2-mediated repression when the antagonistic activity of the SWI/SNF complex is removed, leading to repression of tumour suppressor genes such as PIK3IP1 and promotion of tumour growth (Bitler et al, 2015; Wilson et al, 2010). This tumour promoting role for EZH2 has been demonstrated to exist in malignant rhabdoid tumours (SMARCB1 LOF) (Wilson et al, 2010), ovarian cancer (ARID1A LOF) (Bitler et al, 2015), lung cancer (SMARCA4 LOF) (Fillmore et al, 2015) and more generally for LOF mutations of multiple SWI/SNF components in many cancer types (Kim et al, 2015). As ~20% of cancers have a mutation in a SWI/SNF component (Kadoch et al, 2013), this represents a synthetic lethal relationship of significant therapeutic interest. These findings are particularly striking given the conserved antagonistic relationship between PcG and TrxG (of which SWI/SNF is a member) in development (Schuettengruber et al, 2017), suggesting that loss of other epigenetic regulators which antagonise normal PRC2 function may endow similar dependencies in other cancer types.

Interestingly, the tumour promoting function of EZH2 can occur in a PRC2-independent manner. In castration-resistant prostate cancer, EZH2 acts as a co-activator to support transcriptional activation by the androgen receptor (Xu et al, 2012). A similar tumour promoting co-activator function is observed in breast cancer, where EZH2 upregulates NOTCH1 expression and signalling (Gonzalez et al, 2014), and also activates NFκB gene targets (Lee et al, 2011). In addition to PRC2-independent functions of EZH2 on chromatin, EZH2 also promotes tumour formation by methylating non-histone substrates (Xu et al, 2012; Kim et al, 2013). Thus, the mechanisms by which EZH2 can promote tumour growth are diverse and by no means restricted to its canonical chromatin function.

1.4.2 Oncogenic PRC2 mutations

Providing further convincing evidence for the tumour promoting role of PRC2 is the existence of oncogenic GOF mutations. Sequencing studies show that ~20% germinal centre diffuse large cell B-cell lymphomas and ~10% of follicular lymphomas harbour a heterozygous point mutation at residue Y641 on EZH2, a residue that forms part of the catalytic SET domain (Morin et al, 2010). This mutation enhances the ability of EZH2 to convert H3K27me2 to H3K27me3, but decreases methylation of H3K27me0 and H3K27me, the opposite substrate preference to wild-type EZH2 (Sneeringer et al, 2010). In concert with the remaining wild-type copy of EZH2, mutant EZH2 acts to hypermethylate H3K27 genome-wide (Yap et al, 2011). Similar mutations have also been found to exist in melanoma (Hodis et al, 2012). Importantly, EZH2 GOF mutations are sufficient to drive formation of both lymphoma and melanoma in mice, demonstrating their oncogenic properties (Souroullas et al, 2016). Mechanistically, unrestricted H3K27 tri-methylation by EZH2 is thought to drive tumour formation through blocking B-cell differentiation (Béguelin et al, 2013) and repressing tumour suppressors such as the mTOR regulator SESTRIN (Oricchio et al, 2017), potentially by altering the topology of higher order chromatin structures to manipulate gene expression programmes (Donaldson-Collier et al, 2019). In addition to EZH2, recent sequencing studies in autonomous thyroid adenomas, a benign tumour of the thyroid gland, identified a hotspot Q571R mutation on EZH1 in 27% of tumour samples (Calebiro et al, 2016). Functional investigations revealed that this mutation enhanced the levels of H3K27me3 and rate of proliferation in cells, indicating that it is a tumour promoter in this rare neoplasm. Together, the existence of EZH2 and EZH1 GOF mutations provides convincing evidence that enhanced H3K27me3 levels can have oncogenic consequences.

1.4.3 Tumour suppression by PRC2

Surprisingly, alongside the expansive tumour promoting functions of PRC2, sequencing studies have revealed that PRC2 acts as a tumour suppressor in a number of malignancies (Kim & Roberts, 2016). In myeloid disorders, malignancies arising from bone marrow stem or progenitor cells, core and accessory PRC2 components, most frequently EZH2, are inactivated by heterozygous and homozygous LOF mutations (Ueda et al, 2012; Ernst et al, 2010; Score et al, 2012; Nikoloski et al, 2010). Importantly, patients with homozygous EZH2 LOF mutations have poorer survival than those with

heterozygous mutations (Ernst et al, 2010), and inactivation of EZH2 in a mouse model of myelodysplastic syndrome enhances initiation and progression of the disease (Sashida et al, 2014), together indicating that PRC2 LOF mutations are functionally important in this malignancy. Further corroborating the tumour suppressive role of PRC2, EZH2 and SUZ12 LOF mutations have been identified to occur in ~40% of paediatric and ~25% of adult T-cell acute lymphoblastic leukaemia (T-ALL) (Ntziachristos et al, 2012). Demonstrating the functional importance of this mutation, deletion of EZH2 in p53-null hematopoietic cells leads to a differentiation block on T-cells and subsequent formation of leukaemia (Wang et al, 2018). Interestingly, this loss of PRC2 leads to DNA hypermethylation at key developmental genes, preventing them from being activated during subsequent differentiation (Wang et al, 2018), a mechanism reminiscent of the differentiation block instigated by EZH2 GOF mutations in lymphoma (Béguelin et al, 2013).

Although LOF mutations were first identified in haematological malignancies, it is becoming increasingly apparent that inactivation of PRC2 is also relevant in solid tumours. Loss of SUZ12 or EED occurs in ~80% of malignant peripheral sheath nerve tumours (MPSNTs), a rare cancer of nerve linings (Zhang et al, 2014; Lee et al, 2014; De Raedt et al, 2014). SUZ12 and EED mutations cooperate with loss of Ras GTPase- activating protein NF1 to amplify oncogenic Ras pathway signalling (De Raedt et al, 2014). Moreover, similar dual NF1-EED/SUZ12 LOF is also observed in melanoma and glioblastoma (GBM) (De Raedt et al, 2014), indicating a broader relevance for PRC2 loss in promoting tumour formation. Interestingly, in MPNSTs EZH2 mutations are not observed, a disparity attributed to the functional redundancy between EZH1 and EZH2 (Wassef et al, 2019). Alongside these cancer-specific examples of PRC2 LOF, a pan- cancer survey of PRC2 component mutation frequency revealed that LOF mutations occur in 23 different cancer types (Comet et al, 2016). Thus, mutagenic inactivation of PRC2 is a recurrent and functionally relevant event in promoting tumour initiation and maintenance, directly contrasting with the tumour promoting ability of PRC2 in other cancer contexts.

Figure 5. The diverse functions of PRC2 in cancer A schematic summarising the roles for PRC2 components in different cancer types. It is important to note that this list is not exhaustive, and only the most established examples of tumour suppression and tumour promotion by PRC2 components have been included. GOF – gain-of- function mutation, LOF – loss-of-function mutation, ATA – Autonomous Thyroid Adenomas, DLBCL - Diffuse Large B-cell Lymphoma, DIPG – Diffuse Intrinsic Pontine Glioma, FL – Follicular Lymphoma, MDS/MPN – Myelodysplastic syndrome/Myeloproliferative Neoplasm, MPNST – Malignant Peripheral Nerve Sheath Tumours, T-ALL - T-cell Acute Lymphoblastic Leukaemia.

1.4.4 PRC2 in neural malignancies

Neural malignancies serve as a microcosm of PRC2’s functional complexity in cancer. PRC2 has been assigned both tumour suppressive and tumour promoting roles in neural malignancies, with these opposing functions often observed even in the same tumour type - a disparity that still lacks any degree of mechanistic understanding. Here, I

describe the current state of our knowledge regarding the diverse functions which PRC2 plays in neural malignancies, focussing in particular on the role of PRC2 in GBM.

1.4.4.1 Tumour promotion by PRC2

To date, PRC2 has been implicated as a tumour promoter in the neural malignancy neuroblastoma (Tsubota et al, 2017; Li et al, 2018a; Chen et al, 2017) and GBM (Natsume et al, 2013; Suva et al, 2009; Kim et al, 2013a), with a particularly expansive body of evidence supporting a role for wild-type EZH2 in GBM. GBM is the most aggressive form (grade IV) of malignant glioma, and the commonest type of primary brain tumour in adults, affecting ~3.2 per 100,000 adults in the population (Davis, 2016). GBM tumours arise primarily within the cerebrum, and generally have a very poor prognosis, with a 5-year survival of only ~5% (Delgado-López & Corrales-García, 2016). Frequently, the tumours are driven by amplification of the PDGFR or EGFR growth factor receptor and inactivation of tumours suppressors p53 and pRB (Mao et al, 2012), although a host of other less frequent mutations also co-occur with these primary driver events (Brennan et al, 2013). Disruption of EZH2 activity by genetic and pharmacological approaches within in vitro and in vivo models of GBM, leads to abrogation of cell proliferation and tumour initiation/maintenance (Suva et al, 2009; de Vries et al, 2015; Jin et al, 2017), demonstrating a functional dependency on EZH2 activity in GBM. Interestingly, although short-term shRNA knock-down of wild-type EZH2 in GBM xenografts leads to tumour regression, long-term disruption eventually leads to tumours with a more aggressive phenotype than control tumours (de Vries et al, 2015). This indicates that EZH2 can exhibit functional plasticity in GBM, with consequences for targeting this protein therapeutically. In addition, as in other malignancies, a strong correlation has been identified between the expression of PRC2 components, in particular EZH2, and poor patient outcome in GBM (Zhang et al, 2015; Crea et al, 2010; Orzan et al, 2011; Li et al, 2013). However, whether overexpression of PRC2 components is required for acquisition of tumour promoting function is unclear.

As a result of EZH2’s central role in normal stem cells (Di Croce & Helin, 2013), many studies have focussed on the activity of EZH2 within glioblastoma stem cells (GSCs). In GSCs, EZH2 has been assigned functions in blockading differentiation (Lee et al, 2008), maintaining self-renewal (Suva et al, 2009) and endowing epigenetic plasticity (Natsume et al, 2013) (Fig. 6B). Specifically, in contrast to normal neural stem cells, in GSCs EZH2

uniquely represses BMPR1B, a receptor of the bone morphogenic protein pathway, leading to a blockade on astroglial differentiation which in turn promotes tumour growth (Lee et al, 2008). Furthermore, EZH2 mediates the inter-conversion of GSCs and non- GSCs within tumours, endowing cancer cells with an epigenetic plasticity necessary for adaptation to the tumour microenvironment (Natsume et al, 2013). Interestingly, EZH2 is also able to support GSCs by targeting non-histone substrates. EZH2 methylates the signalling protein STAT3, increasing STAT3 activity and in turn promoting oncogenic signalling (Kim et al, 2013a). Importantly, in all instances, tumour promotion by EZH2 occurs solely in its wild-type unmutated form, indicating that cancer cells co-opt the wild- type activity of EZH2 to potentiate transformative genetic events and/or adapt to their surrounding microenvironment.

As in GBM, wild-type EZH2 is an important tumour promoter in neuroblastoma, a paediatric neural malignancy thought to arise from cells of the neural crest (Matthay et al, 2016). Numerous studies have shown that genetic and pharmacological disruption of EZH2 impairs tumour growth in both in vitro and in vivo models of neuroblastoma (Tsubota et al, 2017; Li et al, 2018a; Chen et al, 2017) (Fig. 6A). Mechanistically, in MYCN-driven neuroblastoma, MYCN upregulates EZH2 expression which leads to suppression of neural differentiation through repression of pro-differentiation genes such as IGFBP3 (Chen et al, 2017). Thus, wild-type EZH2 is an important tumour promoter in neural malignancies, often acting through regulating the differentiation capacity of cancer cells. However, the mechanism(s) by which wild-type EZH2 attains its tumour promoting roles remains unclear.

Figure 6. Tumour promotion by wild-type EZH2 in neural malignancies A. Data reproduced from Fig. 3H of Chen et al, 2018, showing the effect of treating neuroblastoma xenografts with an EZH2 pharmacological inhibitor (GSK126). Xenografts generated from the SH- SY-5Y neuroblastoma cell line. Error bars are mean ±SEM. N = 8 tumours for Vehicle and EZH2i. B. Date reproduced from Fig. 4B of Suva et al, 2009, showing the effect of EZH2 siRNA knock- down upon the ability of glioblastoma stem cells to form tumours in the brains of mice. Dark purple region indicates the presence of a tumour. GSC – glioblastoma stem cell. For additional details please see original publications.

1.4.4.2 Tumour suppression by PRC2

Alongside the established tumour promoting role of PRC2 in neural malignancies, many examples have now emerged of PRC2 also acting as a tumour suppressor. The most striking example of this tumour suppressive ability has come from the recent discovery of K27M mutations on histone H3.1/3.3 which occur in ~80% of diffuse intrinsic pontine gliomas (DIPG), a rare and aggressive paediatric cancer of the brainstem, and ~20% of paediatric gliomas (Schwartzentruber et al, 2012; Wu et al, 2012). Subsequent studies demonstrated that, although the K27M mutation only affects a fraction of the cells total histone H3, the mutant histone drives a global decrease in H3K27me3 levels by inhibiting EZH2 activity in a dominant negative manner (Lewis et al, 2013). Confirming the functional importance of suppressing EZH2 activity, introduction of the K27M mutation into a mouse model of DIPG potentiates tumour growth (Mohammad et al, 2017) (Fig. 7A). Interestingly, a recent meta-analysis of sequencing data has identified that K27M

mutations also occur in melanoma and acute myeloid leukaemia (Nacev et al, 2019), suggesting that dominant negative inhibition of EZH2 is a tumour promoting strategy used in multiple cancers. In addition to H3K27 mutations, heterozygous loss of EED and SUZ12 co-occurs with loss of Ras GTPase-activating protein NF1 in human GBM samples, with ~14% of GBMs containing NF1 and SUZ12 mutations (De Raedt et al, 2014). Generation of mice with heterozygous mutations of Nf1, Suz12 and p53 led to formation of GBM in 54% of mice, whilst dual Nf1/p53 heterozygous KO mice had a penetrance of only 15%, indicating the importance of Suz12 loss in driving GBM formation (De Raedt et al, 2014) (Fig. 7B). Furthermore, EZH2 is also affected by LOF mutations in ~1% of GBM patients (Brennan et al, 2013), however, the functional significance of these mutations is yet to be determined. Thus, disruption of PRC2 in gliomas is a general mechanism for promoting tumour initiation and maintenance.

Figure 7. Tumour suppression by EZH2 in neural malignancies A. Data reproduced from Fig. 1B of Mohammad et al, 2017, showing that a H3K27M mutation enhances the progression of tumour growth in a mouse model of diffuse intrinsic pontine glioma. B. Data reproduced from De Raedt et al, 2014, showing that heterozygous deletion of Suz12 substantially shortens the latency of glioblastoma formation in Nf1+/- p53+/- mice. For further details on figures please see original publications.

PRC2 has also been assigned a tumour suppressive function in the neural malignancy medulloblastoma. Medulloblastoma is a paediatric malignancy that develops within the cerebellum and is constituted by four molecularly distinct groups (Sonic hedgehog, WNT, group 3, and group 4) (Gajjar & Robinson, 2014). Initial sequencing studies identified

that genetic amplification of EZH2 frequently increased levels of H3K27me3 in group 3 medulloblastomas, suggesting that enhanced gene repression could be an important event in driving tumour growth (Robinson et al, 2012). However, directly contradicting this result, deletion of EZH2 in a mouse model of group 3 medulloblastoma led to accelerated tumourigenesis through de-repression of the proto-oncogene Gfi1, indicating that despite its overexpression, EZH2 was acting as a tumour suppressor in medulloblastoma (Vo et al, 2017). This example is particularly pertinent in light of the many studies which have assigned functional importance to EZH2 based upon its overexpression.

1.4.5 Therapeutic targeting of PRC2 in cancer

As a result of PRC2’s established tumour promoting role in many cancer types, there has been great interest in developing pharmacological approaches to enable therapeutic targeting of PRC2. As such, several groups have developed small molecule inhibitors (EPZ-6438, GSK126 and EI1) which compete with S-adenosylmethionine (SAM), the substrate of EZH2, to inhibit EZH2’s HMTase activity (Knutson et al, 2012; McCabe et al, 2012; Qi et al, 2012). These compounds are able to specifically inhibit EZH2 with a ~5000-10,000-fold selectivity over other HMTases, and ~50-150-fold selectivity over EZH1, and induce an almost complete depletion of H3K27me3. Importantly, treatment of B-cell lymphoma cell lines and xenograft models harbouring EZH2 GOF mutations with these inhibitors halts cell proliferation and leads to tumour regression, respectively (Knutson et al, 2012; McCabe et al, 2012). Moreover, inhibitor treatment of xenograft models of malignant rhabdoid tumours harbouring SWI/SNF mutations also leads to tumour regression (Knutson et al, 2013a). In light of the success of these compounds in treating tumour models, many are now being used in pre-clinical and clinical trials to explore their efficacy in treating human disease (Kim & Roberts, 2016). The most advanced of these compounds, EPZ-6438, has now completed phase I clinical trials, in which, it induced complete and partial responses in both B-cell lymphoma and SWI/SNF mutant malignant rhabdoid sarcoma, demonstrating the clinical efficacy of inhibiting EZH2 (Italiano et al, 2018).

As an alternative strategy, compounds have also been developed which disrupt the interaction between EED and EZH2, and effectively inhibit H3K27me3 deposition as a result (Qi et al, 2017; Kim et al, 2013). Similar to the SAM-competitive inhibitors,

disrupting the EZH2-EED interaction inhibits proliferation of B-cell lymphoma cell lines harbouring EZH2 GOF mutations (Qi et al, 2017). Moreover, these inhibitors are effective at overcoming resistance mutations acquired in response to treating cells with SAM- competitive inhibitors (Qi et al, 2017). Given that the tumour promoting role of PRC2 is partly independent of its catalytic activity in the context of SWI/SNF mutations (Kim et al, 2015), disrupting formation of the PRC2 complex may offer a more effective approach to treating these tumours. Thus, the generation of high quality EZH2-specific inhibitors provides a powerful tool for therapeutic targeting of PRC2 and research into the basic biology of PRC2.

Chapter 2. Materials & Methods

2.1 Molecular biology techniques

2.1.1 Genomic DNA extraction

Cells (1x105-1x106) growing on culture dishes were harvested by trypsinisation and subsequently collected by centrifugation at 300xg for five minutes. The supernatant was discarded and the cell pellet was washed by resuspending in 10ml PBS followed by further centrifugation. To extract genomic DNA from the cell pellet, a DNeasy Blood & Tissue Kit (Qiagen - #69506) was used in accordance with the manufacturer’s instructions for cultured cells. Genomic DNA was eluted in 50µl nuclease free water (ThermoFisher - #10457884), quantified using a NanoDrop 2000 spectrophotometer and subsequently stored at -20ºC.

2.1.2 Polymerase chain reaction

Unless otherwise stated, for all amplifications of DNA by polymerase chain reaction (PCR), PfuUltra Hotstart DNA Polymerase (Agilent - #600322) was used and all oligonucleotides were sourced from ThermoFisher. All PCR reactions were performed on a C1000 Touch Thermal Cycler (Bio-Rad). Specific PCR conditions were tailored to the annealing temperature of the oligonucleotides and length of region to be amplified, but generally the below conditions were used:

Component Amount 10x PfuUltra Hotstart DNA pol 2µl 10µM Oligonucleotides (fwd and rev) 1µl DNA template 100ng (genomic DNA) Nuclease free water Up to 20µl

Temperature (ºC) Duration Cycles 95 2 minutes 1 95 30 seconds

Primer Tm – 5 30 seconds 35-40 72 1 minute/kb 72 5 minutes 1

2.1.3 Restriction enzyme digests

To perform restriction digest of plasmids, buffers and restriction enzymes were sourced from New England BioLabs in all instances. The below general protocol was employed for digestion of plasmids:

Component Amount Plasmid DNA 1µg NEB restriction enzymes 1µl each 10x NEB buffer 5µl Nuclease free water Up to 50µl

Digests were incubated at 37ºC for one hour to fully digest plasmid DNA. The resulting digestion was then separated and analysed by gel electrophoresis.

2.1.4 Gel electrophoresis

To make an agarose gel, 1% w/v agarose powder (Merck - #A9539) was combined with Tris-Borate-EDTA buffer (100mM Tris, 100mM Boric acid, 2mM EDTA) and subsequently microwaved to dissolve the agarose powder. The resulting molten agarose was then mixed with SYBR safe (ThermoFisher - # S33102) to allow DNA visualisation and poured into a mould to solidify. Samples to be run on the gel were combined with loading dye (ThermoFisher - #R0611) and then pipetted into the solidified agarose gel which had been placed in an electrophoresis tank containing Tris-Borate-EDTA buffer. The samples were then run through the gel at 120V.

2.1.5 Column purification of DNA

To purify small DNA fragments from PCR, ChIP and restriction digests, a QIAquick PCR Purification Kit (Qiagen - # 28104) was used in accordance with the manufacturer’s instructions. If extraction from an agarose gel was required, gel fragments were excised using a clean scalpel and then DNA was extracted from the gel slices using a QIAquick Gel Extraction Kit (Qiagen - #28115) in accordance with the manufacturer’s instructions.

2.1.6 DNA ligation

DNA ligation was performed using T4 DNA Ligase (NEB - #M0202) using the reaction mixture shown below:

Component Amount T4 DNA Ligase 0.5µl 10x T4 DNA Ligase buffer 1µl Digested plasmid DNA 50ng Insert DNA 3x molar amount of plasmid Nuclease free water Up to 10µl

Ligations were incubated at room temperature (RT) for 15 minutes before being transformed into competent bacteria.

2.1.7 Transforming competent bacteria

For transformation of all plasmids, One Shot Stbl3 Chemically Competent E. coli (ThermoFisher - #C7373-03) were used. Plasmids were transformed into bacteria in accordance with the manufacturer’s instruction. Transformed bacteria were grown overnight at 37ºC on agar plates containing either 50µg/ml Ampicillin (Merck - #10835242001) or 75µg/ml Kanamycin (Merck - #10106801001) to select for transformed bacteria.

2.1.8 Plasmid extraction

Bacteria harbouring the relevant plasmid were inoculated into Luria broth containing either 50µg/ml Ampicillin or 75µg/ml Kanamycin (25µg/ml Zeocin (ThermoFisher - #R25001) was also added for pTRIPZ constructs to prevent recombination) and then grown at 37ºC with vigorous shaking overnight. The next day, bacteria cultures were centrifuged at 3000xg for 15 minutes to pellet bacteria and the supernatant was discarded. To extract plasmids from the bacterial cell pellet, a QIAprep Spin Miniprep Kit (Qiagen - #27104) was used in accordance with the manufacturer’s instructions. Plasmids were eluted in 30µl nuclease free water. To remove contaminating proteins, purified plasmids were mixed with three volumes of Phenol:Chloroform:Isoamyl alcohol (ThermoFisher - # 15593031), vortexed for 30 seconds and then centrifuged for five minutes at 20,000xg. The upper aqueous phase was isolated and mixed with 1/10 volumes sodium acetate and subsequently three volumes of 100% ethanol. The precipitated plasmid was placed at –80°C for 20 minutes to enhance DNA precipitation, and then pelleted by centrifugation at 20,000xg for 30 minutes at 4ºC. To wash the plasmid pellet, the supernatant was discarded and 150µl ice-cold ethanol was added, followed by centrifugation at 20,000xg for two minutes at 4ºC. The supernatant was then removed and the plasmid pellet was dried at 37ºC for five minutes. The dried plasmid pellet was resuspended in nuclease free water and plasmid concentration was quantified using a NanoDrop 2000 spectrophotometer before being stored at -20ºC.

2.1.9 Generation and source of plasmid constructs

To allow inducible expression of EMX2, EMX2 cDNA (coding sequence of: NM_004098.3) was amplified by PCR from a Precision LentiORF EMX2 plasmid (Dharmacon - #2018) and sub-cloned into a derivative of pTRIPZ (Dharmacon) using AgeI-BstBI restriction enzyme sites. The derivative of pTRIPZ contained an SV40-poly A signal and Blasticidin resistance gene, and had previously been generated in the Scaffidi laboratory (see: Torres et al, 2016). To control for EMX2 overexpression, empty pTRIPZ, expressing Puromycin resistance, mir30 cassette, rtTA3 and TurboRFP, was used as negative control in all experiments.

For CRISPR-Cas9-mediated knock-out experiments the below sgRNAs were used:

Target sgRNA sequence (5’-3’) On-target Off-target score score EZH2 ACACGCTTCCGCCAACAAAC 42 78 eGFP GAGCTGGACGGCGACGTAAA 45 79 HRASv12 GGACCAGTACATGCGCACCG 39 90

The sgRNA sequences were selected using the MIT sgRNA design tool (crispr.mit.edu) as the top hit against the genes first exon using default settings. On- and off-target scores were calculated using the IDT CRISPR tool with default settings (https://www.idtdna.com/site/order/designtool/index/CRISPR_CUSTOM). To generate constructs for the stable expression of sgRNAs, oligonucleotides containing the sgRNA sequences were cloned into either pLENTI_GFP_sgRNA (EZH2 and eGFP) or pLX- sgRNA (HRASv12) by golden gate assembly as previously described (Henser-Brownhill et al, 2017). To produce a construct for generate EZH2 knock-out clonal cell lines, an oligonucleotide containing the EZH2 sgRNA was cloned into pSpCas9(BB)-2A-GFP as previously described (Ran et al, 2013).

All other constructs used in this study were sourced as stated below:

Plasmid Source pTRIPZ PHF19 shRNA Dharmacon, Clone ID: V2THS_21282 pTRIPZ PHF1 shRNA Dharmacon, Clone ID: V3THS_357126 pTRIPZ MTF2 shRNA Dharmacon, Clone ID: V2THS_85999 pTRIPZ AEBP2 shRNA Dharmacon, Clone ID: V2THS_45836 pMD2.G Addgene, Plasmid number: 12259 psPAX2 Addgene, Plasmid number: 12260 pAdVAntage Promega, Catalogue number: E1711 pLENTI_GFP_sgRNA Addgene, Plasmid number: 53121 pLX-sgRNA Addgene, Plasmid number: 50662 pSpCas9(BB)-2A-GFP Addgene, Plasmid number: 48138

2.1.10 RNA extraction

Cells (1x106-5x106) growing on culture dishes were harvested by trypsinisation and subsequently collected by centrifugation at 300xg for five minutes. The supernatant was discarded and the cell pellet was washed by resuspending in 10ml PBS followed by centrifugation. RNA was extracted from all cell lines using an RNeasy Plus Mini Kit (Qiagen - #74104) in accordance with the manufacturer’s instructions. RNA was eluted in 30µl nuclease free water, quantified using a NanoDrop 2000 spectrophotometer and subsequently stored at -80ºC.

2.1.11 Reverse transcription and ChIP quantitative PCR

To produce cDNA, 0.5μg of RNA was reverse transcribed in a 25µl volume using a high capacity cDNA reverse transcription kit (ThermoFisher - #4368814) as per the manufacturer’s instructions.

Both RT-qPCR and ChIP-qPCR were performed using SsoAdvanced Universal SYBR Green Supermix (Bio-Rad - #172-5270) on a CFX96 real-time PCR detection system (Bio-rad). Reactions were performed using the below conditions:

Component Amount (µl) 2x SYBR Green Supermix 5 10µM Oligonucleotides (fwd and rev) 0.5 cDNA/chromatin 0.4 Nuclease free water 4.1

Temperature (ºC) Duration Cycles 95 30 seconds 1 95 10 seconds 34 60 30 seconds 65-95 5 seconds +0.5⁰C/cycle

In all RT-qPCR experiments, the housekeeping gene Cyclophilin A (PPIA) was used as a reference gene and all reactions were performed in technical triplicates. In all ChIP- qPCR experiments, input chromatin was used as a reference for enrichment and all

reactions were performed in technical triplicates. The primers used for RT-qPCR and ChIP-qPCR in this study can be found in Appendix 1.

2.1.12 DNA methylation analysis

Genomic DNA from de novo transformed cells was bisulphite converted using an EZ DNA Methylation-Direct kit (Zymo research – #D5020) in accordance with the manufacturer’s protocol. Primers were designed using MethPrimer (Li & Dahiya, 2002) that would amplify a region encompassing the promoter CpG islands of the relevant genes on bisulphite treated DNA (see Appendix 1 for primer sequences). These regions were then amplified by PCR using ZymoTaq PreMix (Zymogen - #E2003) to the below conditions:

Component Amount 2x ZymoTaq PreMix 10µl 10µM Oligonucleotides (fwd and rev) 1µl Bisulphite-converted genomic DNA 200ng Nuclease free water Up to 20µl

Temperature (ºC) Duration Cycles 95 10 minutes 1 95 30 seconds

Primer Tm – 5 30 seconds 35 72 1 minute/kb 72 7 minutes 1

The amplified DNA was then PCR purified and cloned into a pCR 2.1 Topo vector using TOPO TA Cloning (ThermoFisher - # 451641) in accordance with the manufacturer’s instructions. Resulting plasmids were transformed into competent bacteria and grown on agar plates. For each amplified region, plasmids isolated from individual colonies were interrogated by sanger sequencing to ascertain the degree of DNA methylation at individual sequences, and the resulting data was analysed with the software QUMA (Kumaki et al, 2008) using default settings.

2.2 Cell culture and virus production

2.2.1 Cell culturing conditions

All cell lines used in this study were cultured at 37ºC in 5% CO2 on cell culture dishes (Corning) using the media conditions outlined in Appendix 2. For cell lines transduced with pTRIPZ or inducibly expressing Cas9, tetracycline-free fetal bovine serum (Gibco - #2023-11) was used to supplement the media to stop undesired expression from these constructs. All cell lines were sourced as outlined in Appendix 2 and subsequently tested by STR profiling and confirmed as mycoplasma free.

2.2.2 Production of lentivirus and transduction of cell lines

To generate lentivirus, lentiviral packaging plasmids pMD2.G (0.5μg/6cm dish) and psPAX2(1.5μg/6cm dish), pAdVAntage (0.4μg/6cm dish), and the construct of interest (2μg/6cm dish) were transfected into HEK293T cells. To do this, plasmids were mixed with FugeneHD (Promega - #E2311) (12μl/4µg plasmids) in opti-MEM media (ThermoFisher - # 31985070) (200µl/6cm plate), briefly vortexed and left to stand at RT for 15 minutes, and then added to an 80% confluent dish of HEK293T cells. After 24 hours, virus was collected, diluted 1:1 in the relevant media and 8μg/ml Polybrene (Merck - #TR-1003-G) was added. The virus was applied to the desired cell line growing at ~50% confluency. After an additional 24 hours, the media was changed and selection was initiated. See below for relevant selection conditions:

Cell line Transduced construct Method of selection PDX cell lines pTRIPZ-shRNA 1μg/ml Puromycin (ThermoFisher - #A1113802) Transformed fibroblasts pTRIPZ-shRNA O/n treatment 1μg/ml Doxycycline (Merck - #D9891) then FACS of RFP+ cells U-87 MG/DBTRG-05MG pTRIPZ-EMX2 5μg/ml Blasticidin (ThermoFisher - #A1113902) U-87 MG/DBTRG-05MG pTRIPZ-empty 1μg/ml Puromycin Transformed fibroblasts pLENTI_GFP_sgRNA FACS of GFP-positive cells Transformed fibroblasts pLX-sgRNA 5μg/ml Blasticidin

For all antibiotic selections, a plate of untransduced cells was used as a positive control for full selection.

2.2.3 Fluorescence activated cell sorting (FACS)

Cells to be sorted were harvested from culture dishes by trypsinisation and subsequently collected by centrifugation at 300xg for five minutes. The supernatant was discarded and the cell pellet washed by resuspending in 10ml PBS and centrifugation was repeated. The washed cell pellet then was resuspended in ice cold sorting buffer (15mM HEPES, 1% Bovine Serum Albumin, 2mM EDTA, 100mM Penicillin/Streptomycin in PBS) at ~1x106 cells/ml and filtered through a 70µm cell filter (Merck - #CLS431751) into a FACS tube (ThermoFisher - #10585801). FACS was then performed using an Avalon cell sorter (Propel Labs), with cells being were sorted into 1ml ice cold 1:1 MEM media (Merck - #M2279):tetracycline-free FBS. After sorting, cells were centrifuged at 300xg for five minutes, the supernatant was discarded and cells were plated into the relevant media.

2.2.4 Generation and assessment of polyclonal knock-out cell lines

Transformed cells expressing either an EZH2, HRASv12 or eGFP targeting sgRNA and inducibly expressing Cas9, were treated with 1μg/ml doxycycline for 21 days to induce gene knock-out. To assess CRISPR-Cas9 editing efficiency, genomic DNA was extracted from the cells and a ~500bp region surrounding the sgRNA target site was amplified by PCR (see Appendix 1 for primer sequences). The resulting amplified region was PCR purified and the sequence was determined by sanger sequencing. The sequencing traces were then analysed to determine the efficiency of gene editing using the online tool TIDE (Brinkman et al, 2014).

2.2.5 Generation of clonal EZH2 knock-out cell lines

A confluent 10cm plate of transformed fibroblast cells was harvested by trypsinisation, and subsequently collected by centrifugation at 300xg for five minutes. The cells were then resuspended in 100 µl of serum-free MEM medium, into which 10µg of pSpCas9(BB)-2A-GFP containing an EZH2-targeting sgRNA and 10µg of Salmon Sperm DNA (ThermoFisher - # 15632011) were added. The cell-plasmid mixture was

then transferred to an electroporation cuvette (VWR - #732-1136) and electroporated using an ECM 830 Square Wave Electroporation System (Harvard Apparatus) running at 180mV with 5x1ms pulses and a 500ms interval. Electroporated cells were then transferred into cell culture media and left to recover overnight. The next day, GFP positive cells were isolated by FACS and then grown in a clonal manner to generate EZH2 knock-out cell lines. To test for EZH2 knock-out, genomic DNA was extracted from the clonal cell lines and a region surrounding the sgRNA cut site was amplified by PCR. The amplified DNA was isolated by PCR purification and subsequently cloned into a TOPO TA vector as described for ‘DNA methylation analysis’. Plasmids isolated from a minimum of six bacterial colonies were interrogated by sanger sequencing to ascertain whether an out-of-frame deletion was present at both EZH2 alleles. Clonal cell lines which were found to harbour appropriate mutations were then further tested by western- blotting to ensure full depletion of EZH2 protein levels.

2.2.6 Proliferation assays

The glioblastoma cell lines U-87 MG and DBTRG-05MG, expressing RFP or EMX2 from integrated pTRIPZ constructs in an inducible manner, were pre-treated ±1μg/ml Doxycycline for seven days and then plated at a density of 10,000 cells/well in triplicate on a 6-well plate (ThermoFisher - # CLS3516). After 16 hours, the plates were phase imaged using an IncuCyte S3 (Essen bio) to allow normalisation to time zero of cell plating. Cells were then grown ±1μg/ml Doxycycline for eight days. To quantify the final cell number, plates were stained with SYTOX (ThermoFisher - #S7020) and cell nuclei were counted with IncuCyte image analysis software. The endpoint cell counts were then normalised relative to time zero based on the object count calculated from the initial phase images.

2.3 Protein immunodetection

To extract protein from cultured cells, cells growing on culture dishes were harvested by trypsinisation and subsequently collected by centrifugation at 300xg for five minutes. The supernatant was discarded and the cell pellet was washed by resuspending in 10ml of ice cold PBS. The PBS-cell suspension was centrifuged at 300xg for five minutes and the PBS wash was repeated. The washed cell pellet was resuspended in high salt buffer

(50mM Tris pH 7.5, 300mM NaCl, 0.5% IgePal and 1mM EDTA) containing protease inhibitors (Cell Signalling Technologies - #5871) at ~1x107 cells/ml and incubated on ice for 20 minutes. This was followed by three cycles of sonication (30 secs on/off) using a chilled Bioruptor Pico sonicator (Diagenode) and lysates were then clarified by centrifugation at 20,000xg for 30 minutes at 4ºC. Protein levels were quantified by Bradford assay (Bio-Rad - #500-0201) in accordance with manufacturer’s instructions and the samples were then diluted to ~2µg protein/µl in 1x LDS sample buffer (ThermoFisher - #NP0008) containing reducing agent (ThermoFisher - #NP0004). The samples were boiled at 95ºC for seven minutes to denature the proteins. Samples (~20- 30µg protein) were then run on a 4-12% bis-tris gel (ThermoFisher - #NP0321PK2) at 200V for ~40 minutes and subsequently transferred using a iBlot2 system (Life technologies) running programme 0. Membranes were blocked with 0.1% Tween 20 (Merck - #P9416) in PBS (PBT) + 5% milk (Marvel) for 45 minutes at RT, and then blotted with the appropriate primary antibody at the relevant concentration in PBT + 5% milk overnight at 4ºC: αEZH2 (Cell Signalling - #5246S, 1:1000), αH3K27me3 (Upstate - #07-449, 1:5000), αHistone H3 (Abcam - #ab1791, 1:40,000), αSUZ12 (Cell Signalling – #D39F6, 1:1000). The membrane was then washed three times for 10 minutes in PBT, before being blotted with an anti-rabbit horse radish peroxidase secondary antibody (Vector - #PI-1000, 1:5000) in PBT + 5% milk for 45 minutes at RT. The membrane was washed again and then developed by treating with ECL western-blotting substrate (ThermoFisher - #32106) for five minutes. The membrane was imaged using an Amersham imager 600 (GE healthcare), and images were quantified using ImageJ software utilising the ‘gels’ tool.

2.4 Tumour xenograft studies

2.4.1 Tumour transplantation assays

All tumour xenograft studies were performed in male NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ mice sourced from internal breeding colonies. In all instances, mice were injected at 6-8 weeks of age. Prior to injection, hair was removed from the flanks of the mice using clippers, and injections were performed subcutaneously using Terumo 0.5ml 27g Insulin Syringes. After tumour appearance, the size was measured weekly using digital callipers and volume was calculated as L*W2/2, in which L = longest edge of tumour and W =

shortest edge of tumour. Animal studies were conducted in accordance with the Francis Crick Institute project license PPL 70/8167.

For glioblastoma transplantation assays, DBTRG-05MG cells, expressing either EMX2 cDNA or RFP as a control, were trypsinised from cell culture dishes and collected by centrifugation at 300xg for five minutes. The supernatant was discarded, and the cells were washed by resuspending in 10ml sterile PBS and then collected by further centrifugation. Washed cells were resupended in sterile PBS at a concentration of 7x106/ml, and 3.5x105 cells (50μl) were subsequently injected subcutaneously into both flanks of each mouse.

For transplantation assays monitoring the effect of PRC2 accessory component knock- down, cell lines transduced with pTRIPZ-shRNA constructs were treated ±1μg/ml Doxycycline in culture for four days to pre-induce mRNA knock-down. Cell lines were then prepared for injection as in glioblastoma transplantation assays, and 5x105 (PAXF 1998 and LXFL 1674 cell lines) or 1x105 (transformed fibroblasts, GXA 3067 and LXFA 677 cell lines) cells were injected in a 50μl volume into both flanks of each mouse. For the duration of the experiments, mice injected with cells pre-induced for mRNA knock- down were provided with drinking water containing 2μg/ml Doxycycline and supplemented with 1% sucrose (changed every 2-3 days) to maintain mRNA knock- down.

2.4.2 Tumour cell isolation and derivation of cell lines

Prior to tumour extraction, tumour-bearing mice were sacrificed through cervical dislocation followed by terminal bleeding. Immediately after sacrifice, hair was shaved from the region surrounding the tumour to expose a site for subsequent resection. The tumours were excised using a sterile scalpel and tweezers, then placed in a petri dish and bathed in 1ml sterile PBS. The tumour was cut into pieces 2-3mm in diameter using a clean scalpel, transferred to a 15ml falcon tube (Merck - #CLS430791) and then collected by brief (<15 seconds) centrifugation at 300xg. The supernatant was discarded and the tumour was resuspended in four volumes (relative to the tumour) of 0.5 x Liberase (Merck - #5401020001) in warm RPMI media (ThermoFisher - #11875093). This mixture was then transferred to a GentleMACS C tube (Miltenyi Biotec - #130-093- 237) and dissociated using human tumour programme one on a GentleMACS tumour

dissociator (Miltenyi Biotec). The resulting mixture was then incubated at 37ºC for 30 minutes with fast agitation to further dissociate the tumour. The mixture was processed with human tumour programme two on the tumour dissociator, after which, 10,000 U DNAse I (Merck – #04716728001) was added to the mixture and it was then incubated at 37ºC for 30 minutes with fast agitation. The dissociation mixture was then processed using human tumour programme three on the tumour dissociator. To remove undigested tumour, the dissociated mixture was filtered through a 70µm cell filter (Merck - #CLS431751) into 50ml falcon tubes. Filtered mixtures were then transferred to 15ml falcon tubes and centrifuged at 300xg for five minutes. The supernatant was discarded and cells were resuspended in 10ml of RPMI media. Centrifugation and RPMI media resuspension was repeated three times to wash the cell pellet.

To derive cell lines, dissociated tumour cells were plated onto cell culture dishes in the presence of 1mg/ml Neomycin (ThermoFisher - #N1142) to remove any contaminating murine cells.

2.5 Chromatin immunoprecipitation sequencing experiments

2.5.1 Chromatin immunoprecipitation (ChIP)

ChIP was performed for H3K27me3 and EZH2 in the same manner for all cell lines investigated in this study. In all instances, 2x107 cells were fixed with 30ml 1% formaldehyde (Merck - #252549) in cell culture media for 10 minutes at RT with slow rotation. The fixation was quenched by adding 125mM glycine followed by slow rotation for a further five minutes. Fixed cells were washed twice in ice-cold PBS and then resuspended in 1.8ml IP buffer, 1:1 SDS-containing buffer (100mM NaCl, 0.2% NaN3, 5mM EDTA pH 8.0, 50mM Tris-HCl pH 8.0, 0.5% SDS): triton-containing dilution buffer

(500mM Tris-HCl pH 8.6, 100mM NaCl, 0.2% NaN3, 5mM EDTA pH 8.0, 5% Triton X- 100) containing protease inhibitors (Cell Signalling Technologies - #5871). The suspension was incubated on ice for 20 minutes to lyse the cells. Chromatin was then sheared to 200-400bp using a cooled Bioruptor Pico Sonicator (Diagenode), with 10-15 cycles of 30 secs on/off. The lysate was then clarified by 30 minutes of centrifugation at 20,000xg and protein concentration was quantified by Bradford assay. To perform ChIP, EZH2 (Cell Signalling - #5246S, 1:40), H3K27me3 (Upstate - #07-449, 1:100) or Rabbit IgG control antibodies (Abcam - #ab46540, 1:100) were mixed with 1mg of chromatin

lysate, then made up to 1ml with IP buffer and slowly rotated at 4⁰C overnight. To generate technical replicates, all ChIPs were performed in duplicate using the same chromatin lysate. The following day, 30μl of protein-G magnetic beads (ThermoFisher - #10003D) (washed twice in IP buffer) were pipetted into the ChIP and slowly rotated at 4⁰C for four hours. Using a magnetic rack, the bead-antibody complexes were washed three times with ice-cold wash buffer one (0.1% SDS, 1% Triton X-100, 2mM EDTA pH 8.0, 150mM NaCl, 20mM Tris-HCl) and once with ice-cold wash buffer two (wash buffer one with 500mM NaCl). Chromatin was eluted by rapidly shaking the beads at 65⁰C overnight in 110μl of 0.1M NaHCO3 1% SDS solution. To provide a control from ChIP enrichment, input chromatin was also de-crosslinked in parallel. Chromatin was then isolated by performing PCR purification and eluted in 30μl nuclease free water. The resultant DNA was stored at -20ºC.

2.5.2 ChIP library preparation and sequencing

For ChIP samples that were to be interrogated by next generation sequencing, a quality control ChIP-qPCR was first performed using primers targeting a positive control region within the promoter of the WT1 gene and a negative control region in the promoter of housekeeping gene GAPDH (see Appendix 1 for primer sequences). Positive and negative controls were selected based on publically available ChIP-seq datasets. Prior to preparation of sequencing libraries, the quantity and integrity of the chromatin was assessed using a BioAnalyser 2100 (Agilent). Library preparation was then undertaken with 5-10ng of DNA using an Illumina TruSeq ChIP kit (Illumina - #IP-202-1012) in accordance with manufacturer’s protocol. ChIP-sequencing (ChIP-seq) samples were multiplexed six samples/lane and sequenced using a HiSeq 2500 sequencer (Illumina) performing 101bp paired-end runs.

2.5.3 ChIP-seq analysis

Sequencing run quality was assessed using the package FASTQC (Andrews S, 2010) and all runs were determined to be of a good quality. However, replicate one of the H3K27me3 ChIP from transformed fibroblast cells and replicate two of the EZH2 ChIP from transformed fibroblast cells did not have sufficient reads after the first run. As a result, these libraries were resequenced to obtain additional reads, which were then

merged with those from the original run to make a single fastq file. Adapter trimming was undertaken with cutadapt (version 1.9.1) (Martin, 2011) using parameters “--minimum- length=25 --qualitycutoff= 20 -a AGATCGGAAGAGC -A AGATCGGAAGAGC”. BWA (version 0.6.2) (Li & Durbin, 2009) using default parameters was employed to perform genome-wide mapping of the adapter-trimmed reads to the human hg19 genome. Duplicate marking was undertaken using the function ‘MarkDuplicates’ in picard (version 2.1.1) (Broad Institute) and duplicate reads were then excluded. Alignments were subsequently filtered to remove reads which mapped to DAC Blacklisted Regions from the ENCODE/DAC (ENCODE Project Consortium 2012) downloaded from the UCSC. Further filtering was performed to exclude read pairs that mapped to different chromosomes, ambiguously mapped, were discordant, or had a mismatch >4 in any read. After read filtering, replicate one of the H3K27me3 ChIP from transformed fibroblast cells did not have sufficient coverage. Hence, a third replicate for that sample was sequenced and the BAM files of replicate one and three were merged to reach sufficient coverage. Tiled data format (tdf) files for ChIP-seq visualisation were produced using the ‘count’ function in IGVTools (version 2.3.75) (Broad Institute) with default parameters.

To make the ChIP-seq data compatible with downstream applications, it was necessary to convert it to a single-ended format. Thus, BAM files were first converted to BED paired- end format using the bamtobed function in BEDTools (version 2.26 in all instances) (Quinlan & Hall, 2010) with default parameters. The BED files were then converted to a single-ended format by retaining only the chromosomal coordinates from the beginning of the first read and end of the second. Using the single-ended data set, peak calling was performed on both ChIP-seq replicates for all samples using SICER (version 1.1) (Zang et al, 2009) running the non-default parameters: “redundancy threshold = 1, window size = 200, fragment size = 110, effective genome fraction = 0.75, gap size = 400 or 600 (EZH2 and H3K27me3, respectively), FDR <0.0001”. A consensus set of peaks for EZH2 and H3K27me3 in each cellular state was generated by merging replicate peaks using ‘mergePeaks’ in the homer package (version 4.8.3 in all instances) (Heinz et al, 2010) and extracting the sum of overlapping peaks. Peaks separated by ≤250bp were merged into a single peak using the ‘merge’ function in BEDTools with default parameters. H3K27me3 and EZH2 consensus peak sets for each cellular state were intersected and only EZH2 peaks which overlapped a H3K27me3 peak for ≥25% of their width were retained for further analysis. To associate EZH2 peaks with genomic features, EZH2 peaks were mapped to gene deserts (gene depleted regions at least

1Mb long), promoters (±3kb of transcription start site), genebodies (from end of promoter until 1kb downstream of transcription termination site) or other intergenic regions (regions not belonging to the other categories) using region_analysis.py (Shen laboratory), running default parameters. As a reference to determine the genomic features which EZH2 peaks were enriched at, a BED file containing equally spaced 10kb windows across the entire human genome was also analysed.

Consensus EZH2 peak sets from each cellular state were intersected with ‘mergePeaks’ (homer) using default parameters, giving cell type unique and common peaks, from which Venn diagrams of EZH2 binding distribution were generated. To calculate tag enrichment at these sets of unique and common peaks, tag directories were compiled for the combined reads of both EZH2 ChIP-seq replicates from each cellular state with ‘makeTagDirectory’ (homer) using default parameters. Relative EZH2 enrichment between cell states was then calculated at all common and unique EZH2 peaks using ‘getDifferentialPeaks’ (homer) with the following non-default parameters: “-F = 0, -P = 1, -tagAdjust = 0, -tagAdjustBg = 0”. EZH2 peaks with a fold-change ≥1.5 and a p-value ≤1e-20 (defined as: ‘large-magnitude’ differential peaks) relative to the compared cellular state were then considered for further analysis. The selected peaks were annotated to the nearest transcription start sites (TSS) of an antisense, protein-coding, or lincRNA gene using the Refseq Hg19 TSS annotation and gene biotypes sourced from ensembl. If a peak was equidistant from, or overlapped with, more than one TSS then all TSSs were recorded. Only peaks ±5kb of a TSS were considered for further analysis.

Overlaps between the genes associated with large-magnitude differential EZH2 peaks (±5kb TSS) and existing gene signatures was calculated using ‘compute overlaps’ from the Broad Institute (software.broadinstitute.org/gsea/msigdb/annotate.jsp). To assign genes into specific functional categories, each gene was manually investigated using the genecards database (genecards.org) and subsequently assigned to a specific category based on existing literature describing the proteins function.

ChIP-seq metaprofiles were generated with ngsplot (version 2.63) (Shen et al, 2014) using the following parameters: “normalisation = bin, colour scaling = global, fragment length = 300”. To generate correlation heatmaps, the ChIP-seq peaks called on each replicate were compared using DiffBind (version 2.8.0) (Ross-Innes et al, 2012) within the R programming environment (version 3.5.1) using standard analysis setting. Briefly,

peak sets derived from each ChIP-seq mark were intersected and a consensus peak set was generated. The enrichment of sequencing reads within each consensus peak was compared between samples, leading to the generation of a correlation heatmap. To compare the nucleotide content of EZH2 peaks either common or unique to untransformed and transformed cells, the nucleotide content of each peak was calculated using the ‘nuc’ function in BEDTools. To explore the intersection of EZH2 peaks with CpG islands, CpG island annotations were sourced from UCSC (based on the epigenomic predictions of Bock et al, 2007). The ‘intersect’ function from BEDTools was then used to identify EZH2 peaks that had any overlap with a predicted CpG island.

2.6 RNA sequencing experiments

2.6.1 RNA library preparation and sequencing

For RNA-sequencing (RNA-seq) experiments involving EZH2 pharmacological inhibition, cell lines were first treated with either EPZ-6438 (Selleckchem - #S7128), dissolved in DMSO, or DMSO as a control for 12 days. To maintain repression throughout the treatment period, drug and media was replaced every three days. For RNA-seq experiments involving shRNA knock-down of EZH2, cell lines inducibly expressing an EZH2 targeting shRNA were treated ±1µg/ml Doxycycline for seven days to deplete mRNA levels. To ensure mRNA knock-down throughout the treatment period, Doxycycline was refreshed every two days. To assess the extent of EZH2 knock-down, a test RT-qPCR and western-blot were performed to confirm a decrease in mRNA and protein levels, respectively. For RNA-seq experiments involving EZH2 knock-out by CRISPR-Cas9 genome editing, cell lines were prepared as described in ‘Generation and assessment of polyclonal knock-out cell lines’. To ensure complete EZH2 knock-out, Doxycycline was refreshed every two days during the treatment period.

RNA was extracted from cells as described in ‘RNA extraction’. RNA integrity was assessed using a TapeStation 4200 (Agilent), and all samples were found to have an RNA integrity number ≥8, indicating the RNA was high quality with little degradation. Libraries for sequencing were prepared using the KAPA Stranded RNA-Seq Kit with RiboErase (Roche - #07962282001) in accordance with the manufacturer’s instructions. Library quality was confirmed using a BioAnalyser 2100 (Agilent). RNA sequencing was carried out on an HiSeq 4000 sequencer (Illumina) and generated ~20 million 75bp

single-end reads per sample. To reduce sequencing lane biases, the library of each sample was divided across two lanes, generating two fastq files for each sample. These fastq files were subsequently merged for the next analysis steps.

2.6.2 RNA-seq analysis

The reads from RNA-seq were adapter-trimmed using cutadapt as previously specified for ChIP-seq. The RSEM package (version 1.2.29) (Li & Dewey, 2011) in combination with the STAR alignment algorithm (version 2.5.1b) (Dobin et al, 2013) was used for mapping and subsequent gene-level counting of sequenced reads with respect to hg19 RefSeq genes downloaded from the UCSC Table Browser (Karolchik et al, 2004) on 14/04/2016. The non-default parameters used were “--star-output-genome-bam -- forward-prob 0”. Differential expression analysis was performed for all comparisons made in the study using the DESeq2 package (version 1.10.1) (Love et al, 2014) within the R programming environment (version 3.2.3). GSEA for gene signatures identified using RNA-seq was performed in an identical manner to that described in ‘ChIP-seq analysis’.

To characterise transformation-induced changes to the EZH2-regulated gene set, in cells from each stage of de novo transformation EZH2-regulated genes were first identified by differential expression analysis of RNA-seq data from cells treated with DMSO or the EZH2-specific inhibitor EPZ-6438 (EZH2i). A gene was defined as differentially expressed (DEG) if it had an FDR ≤0.01, Log2FC ≥1 and maximal TPM ≥1. The DEGs from untransformed and transformed cells were intersected and separated into cell type- specific and common EZH2-regulated genes. DEGs upregulated by EZH2i in a cell type- specific manner were then intersected with genes associated with a differential EZH2 peak in the same condition (see: ‘ChIP-seq analysis’). The genes found in both data sets were retained for further analysis. To identify which of these genes had expression changes associated with differential EZH2 binding, differential expression analysis was performed comparing the DMSO-treated condition of untransformed and transformed cells. DEGs were defined in the same manner as for the EZH2i analysis.

To compare the EZH2-regulated gene set identified by genetic and pharmacological LOF approaches, differential expression analysis was first performed to identify genes misregulated upon EZH2 disruption by each method. For all LOF approaches, genes

were defined as differentially expressed if they had an FDR ≤0.001. For EZH2i, differential expression analysis was performed comparing RNA-seq from DMSO and EZH2i treated transformed cells. For EZH2 KO, differential expression analysis was performed comparing RNA-seq from transformed cells with polyclonal EZH2 KO to that from transformed cells expressing Cas9 alone and also transformed cells expressing an eGFP-targeting sgRNA. In this instance, the final list of DEGs was generated by intersecting the DEGs identified by comparing EZH2 KO to the two separate controls and retaining only those genes which were identified as DEGs in both instances. For EZH2 KD, differential expression analysis was performed comparing RNA-seq from transformed cells with EZH2 KD induced and the same cell line in its uninduced form. The lists of DEGs from each approach were then intersected to identify the degree of similarity between approaches. The PRC2 target genes identified by each LOF approach were determined by intersecting upregulated DEGs with genes that had a peak of H3K27me3 ±5kb promoter (see: ‘ChIP-seq analysis’).

2.7 RNA in situ hybridization

Formalin fixed paraffin embedded (FFPE) tissue sections from primary resections of adult glioblastoma tumours were stained using an RNAscope® Multiplex Fluorescent Reagent Kit v2 (ACDBio - #323100) in accordance with the manufacturer’s instructions for FFPE brain tumour samples. The following probes were used for sample staining: Hs-EMX2 (ACDBio - #320269-C2), Hs-HOXB9 (ACDBio - #473521), Hs-EZH2 (ACDBio - #405491). As a positive control for EMX2 expression, a cell pellet containing a 1:1 mixture of HEC59 endometrial cancer cells endogenously expressing EMX2 and U-87 MG cells overexpressing EMX2 cDNA was fixed in 10% neutral buffered formalin, subsequently paraffin embedded and sections were then stained for EMX2 in parallel to glioblastoma samples. Similarly, as a positive control for HOXB9 expression, a cell pellet containing a 3:1 mixture of PC9 lung adenocarcinoma cells and transformed fibroblasts overexpressing HOXB9 was used. The cell pellet used as a positive control for HOXB9 staining was also employed as a negative control for EMX2 staining, whilst the EMX2 positive control cell pellet was used as a negative control for HOXB9 staining. Cell pellets were stained in accordance with the manufacturer’s instructions for FFPE cell pellets.

2.8 Proteomics experiments

2.8.1 Stable isotope labelling with amino acids in cell culture (SILAC)

For SILAC, the relevant cell lines were grown for nine days in DMEM for SILAC (ThermoFisher - #88364) supplemented with light or heavy L-Arginine (147.5 mg/L) and L-Lysine (91.25 mg/L), and 10% dialysed Fetal Bovine Serum (BioSera - # FB-1001D). Labelling efficiency was assessed after nine days by mass spectrometry and found to be >95% in all instances. Unlabelled L-Arginine and L-Lysine, (Sigma-Aldrich), and

13 15 13 15 heavy L-Arginine [ C6, N4] and L-Lysine [ C6, N2] in their hydrochloride forms (CK Isotopes) were used for proteome labelling.

2.8.2 Immunoprecipitation

Protein was extracted and quantified from relevant cell lines as described for ‘Protein immunodetection’. For initial test immunoprecipitations, 1mg of unlabelled whole cell lysate from transformed cells was used, whilst for SILAC experiments, 500µg of ‘heavy’ and ‘light’ labelled lysates were combined for all the relevant comparisons. Cell lysates were then mixed with either an αEZH2 (Cell Signalling - #5246S, 1:300) or a rabbit isotype control antibody (Abcam - #ab46540, 1μg) and made up to a total volume of 0.5ml with ice cold high salt buffer (50mM Tris pH 7.5, 300mM NaCl, 0.5% IgePal and 1mM EDTA) containing protease inhibitors (Cell Signalling Technologies - #5871). The immunoprecipitation was then rotated overnight at 4°C. Subsequently, 20μl of magnetic protein-G beads (ThermoFisher - #10003D) (washed twice in high salt buffer) were added to the immunoprecipitation and rotated for four hours at 4°C. The beads were then washed 4x with ice-cold high salt buffer and then resuspended in 30μl 2x LDS sample buffer (ThermoFisher - #NP0008) containing reducing agent (ThermoFisher - #NP0004). Co-immunoprecipitated proteins were then eluted by boiling the beads for seven minutes at 95⁰C.

2.8.3 Mass spectrometry analysis

Eluted proteins were separated by sodium dodecyl sulphate polyacrylamide gel electrophoresis using a 4-12% bis-tris gel (ThermoFisher - #NP0321PK2). Samples were run at 200V until the running front had migrated approximately 2cm into the gel, and the

gel was subsequently coomassie stained with InstantBlue (Expedeon - #ISB1L) in accordance with the manufacturer’s instructions. From the stained gel, eight horizontal gel slices were excised from each lane, and the proteins in each slice were then digested in-gel with trypsin (Promega - #V5111) using a Janus liquid handling system (Perkin Elmer). Tryptic peptides were analysed by liquid chromatography-mass spectrometry using an Orbitrap Velos or Orbitrap-Q exactive mass spectrometer coupled to an Ultimate 3000 uHPLC equipped with an EASY-Spray nanosource (ThermoFisher) and acquired in data-dependent mode. The data were searched against the human Uniprot database using the Andromeda search engine. Raw data were processed using MaxQuant (version 1.6.0.1) (Cox & Mann, 2008), using intensity based absolute quantification (iBAQ) or SILAC selected as the quantification algorithm. The proteinGroup.txt output table was imported into Perseus software (version 1.4.0.2)

(Tyanova et al, 2016) for further statistical processing. iBAQ intensities were Log10 transformed, and missing values were imputed using default Perseus settings by drawing from a simulated noise distribution with a down shift of 1.8 and a width of 0.3.

SILAC ratios were Log2 transformed, and missing values were imputed using default Perseus settings as described above.

2.9 Analysis of publically available datasets

The sources of publically available data used in this study are described in Appendix 3.

2.9.1 Generation of ChIP-seq correlation heatmaps

Processed BAM files containing ChIP-seq data for H3K27ac, H3K27me3 and H3K4me3 in different primary cell types, and the relevant input files, were downloaded from the ENCODE consortium data portal (encodeproject.org/files). The specific data sets used can be found in Appendix 4. The BAM files of ChIP-seq and input replicates were then merged using the ‘merge’ function in SAMtools (version 1.3.1) (Li et al, 2009a) with default parameters. For all merged ChIP-seq BAM files, peak calling was performed relative to input as described in ‘ChIP-seq analysis’ using the same parameters as previously employed for H3K27me3 ChIP-seq. To generate correlation heatmaps, peak sets of a ChIP-seq mark in different primary cell types were compared using DiffBind as described in ‘ChIP-seq analysis’.

2.9.2 Analysis of DNA methylation in cancer cell lines

Processed gene level methylation data for all NCI-60 cell lines was downloaded from National Cancer Institute via the cell miner database (discover.nci.nih.gov/cellminer/) and genes lacking methylation data in any cell line were excluded. Data for genes with an EZH2 peak ±5kb TSS, as identified by ChIP-seq, was selected. Genes with an EZH2 peak were further subdivided into genes responsive/unresponsive to EZH2i as identified by RNA-seq analysis. Responsive genes were defined as those upregulated with a

Log2FC ≥1, FDR ≤0.01 and a maximal TPM ≥1 upon EZH2i treatment.

2.10 Human tissue samples

GBM tissue sections were kindly provided by Prof Sebastian Brandner at University College London Hospital with necessary ethical consent provided by BRAIN UK (Ref: 18/008).

2.11 Statistical analyses

All statistical tests used are indicated in the appropriate figure legends. Unless otherwise stated, all error bars represent ±standard error of the mean for the number of replicates indicated by N in the relevant figure legend. In boxplots, the top, middle and bottom box delimiters represent the 75th, 50th and 25th percentiles of the data, respectively. Top and bottom whiskers show the 75th percentile + 1.5*interquartile range and 25th percentile – 1.5*interquartile range, respectively.

Chapter 3. Wild-type EZH2 re-wires neural development programmes to drive glioblastoma

Introduction

It is now established that an intimate link exists between mechanisms underlying normal developmental processes and those required for cancer initiation and maintenance (Ma et al, 2010). In particular, epigenetic mechanisms, essential for mediating normal development and maintaining tissue homeostasis, are often found to be misregulated in cancer (Dawson et al, 2012). Extensive sequencing studies have shown that epigenetic regulators are frequently mutated across a range of different cancer types (Shen & Laird, 2013). Strikingly, the majority of these mutations are loss-of-function (LOF) (Shen & Laird, 2013), suggesting that many epigenetic regulators are tumour suppressors in normal cells. Remarkably, epigenetic regulators affected by LOF mutations can often also act as tumour promoters in their wild-type unmutated state within established tumours (Wainwright & Scaffidi, 2017), indicating that they have dichotomous functions between normal and cancer cells. Despite these antithetic functions of wild-type epigenetic regulators having been independently described, very little is known about the molecular mechanisms which enable a switch between these roles at the transition from the physiological to pathological cell state.

An excellent example of an epigenetic regulator with a dichotomous function in normal and cancer cells, is EZH2. Through dynamically repressing specific gene sets, EZH2 mediates both normal developmental processes and tissue maintenance (Di Croce & Helin, 2013) - the importance of which is evidenced by the severe phenotypes of abrogating EZH2 function in the embryo or adult (O’Carroll et al, 2001; Ezhkova et al, 2009). Concurrently, and in direct contrast to its roles in normal cells, EZH2 in its wild- type unmutated form also acts as a tumour promoter in a range of different cancer types (Kim & Roberts, 2016). This dichotomous role of EZH2 is particularly evident within the central nervous system (CNS). In the embryonic and adult CNS, EZH2 is essential for controlling the expression of lineage specifying factors and regulating the balance between stem cell self-renewal and differentiation (Pereira et al, 2010; Hwang et al, 2014) (see: ‘Introduction’). In contrast, wild-type EZH2 is also a potent tumour promoter in neural malignancies such as glioblastoma (GBM, grade IV glioma), where it provides epigenetic plasticity and maintains the cancer stem cell population (Suva et al, 2009;

Natsume et al, 2013). How EZH2 is able to act as both a mediator of normal development/tissue maintenance and a tumour promoter, whilst remaining in its wild- type unmutated state, is unclear. To shed light on this hijacking of EZH2 activity, I set out to perform a detailed characterisation of the molecular mechanisms that enabled it to occur (Fig. 8). In dissecting these mechanisms, I hoped to provide a paradigm for how wild-type epigenetic regulators can be hijacked to promote and maintain neoplastic transformation.

Figure 8. Antithetic functions of wild-type EZH2 in neural development/tissue maintenance and cancer

Aims

In this chapter, I describe experiments in which I characterised the molecular mechanisms underlying wild-type EZH2’s functional switch from mediator of development/tissue maintenance to tumour promoter in the normal and malignant CNS. To achieve this, I performed a comprehensive multi-omics characterisation of EZH2 activity at different stages of transformation using an established model of neoplastic transformation. Subsequently, I validated the molecular mechanisms identified using this discovery system with cellular GBM models and publically available cancer data sets. Finally, I demonstrated the functional importance of these molecular mechanisms using in vitro and in vivo models of GBM.

Results

3.1 Modelling neural neoplastic transformation

To identify the mechanisms underlying EZH2’s switch from essential mediator of development/tissue maintenance in the normal CNS to tumour promoter in GBM, I needed to directly compare the activity of EZH2 in normal and cancer cells using a CNS- relevant system. The ideal methodology to do this, would have entailed isolating the normal cell-type from which GBM originates, transforming it in a controlled manner using genetic events found in GBM, and subsequently performing a detailed dissection of how EZH2 function differs between the untransformed and transformed states. However, confounding this approach, the identity of the cell-of-origin in GBM currently remains unclear (Alcantara Llaguno & Parada, 2016). In fact, studies attempting to define a single cell-of-origin have shown that cells across the neural differentiation hierarchy can be transformed to generate tumours (Alcantara Llaguno & Parada, 2016), suggesting there may be no single cell-of-origin. Furthermore, expression profiling of GBM tumours has revealed extensive inter-tumour heterogeneity, likely reflecting the diverse cell types from which these tumours originate (Verhaak et al, 2010). Moreover, although GBM murine models exist, I wanted to characterise EZH2 behaviour specifically in human adult cells. I decided on this approach to avoid identifying murine-specific mechanisms, and to enable me to use the many cell lines and datasets available for adult GBM. Obtaining normal neural cells from adult humans is extremely challenging, and commercially available primary neural cells are derived from foetal sources. These would have been unsuitable for studying adult GBM as the paediatric disease has divergent genetic, epigenetic and clinical features (Jones et al, 2016). Thus, uncertainty as to GBM’s cell-of-origin, a dearth of suitable human GBM models and a lack of appropriate reagents meant I would need to employ an alternative approach to dissect the molecular mechanisms underlying EZH2’s changing function.

To overcome these limitations, I chose to employ a cellular model of neoplastic transformation which allows direct comparison of isogenic normal and malignant cells, and has been previously shown to be relevant for discovering GBM-related mechanisms. (Fig. 9). In this system, human adult dermal fibroblasts are first immortalised through the expression of telomerase (henceforth: untransformed cells), overcoming issues of cellular senescence which often hamper studies using primary cells. Subsequently, the expression of simian virus 40 (SV40) large and small T-antigens leads to the inhibition

of tumour suppressors p53 and pRB, and transitions the untransformed cells to a pre- neoplastic state (henceforth: pre-neoplastic cells). Finally, expression of the oncogene HRASV12 constitutively activates the RAS/MAPK pathway, leading to full neoplastic transformation (henceforth: transformed cells) and enabling cells to form tumours in immunocompromised mice (Hahn et al, 1999). This experimental system has been used to identify epigenetic mechanisms found in GBM, most notably in the cancer stem cell population (Son et al, 2009; Scaffidi & Misteli, 2011; Suvà et al, 2014; Torres et al, 2016). Furthermore, recent studies classifying GBM based on gene expression profiles have a identified a major subtype defined by a mesenchymal expression profile, which includes a range of fibroblastic markers (Verhaak et al, 2010; Hölzel et al, 2010). Across all GBM subtypes, these mesenchymal features are associated with resistance to therapy in patients, demonstrating their clinical relevance and illustrating the functional importance of a fibroblastic phenotype in GBM (Bhat et al, 2013; Halliday et al, 2014). Importantly, hyper-activation of the RAS/MAPK pathway and genetic loss of TP53 and RB1 are some of the most frequent events in GBM (Mao et al, 2012), making this model of de novo transformation relevant for studying GBM. Thus, despite being an atypical system to study brain-related mechanisms, this represented a tractable, validated and relevant system in which to explore the switch in EZH2 function during neural neoplastic transformation.

Figure 9. The model of neoplastic transformation used in this study Schematic depiction of the fibroblast-based model of neoplastic transformation used in this study. Red strikethrough represents inhibition of tumour suppressors p53 and pRB by SV40 large and small T-antigens.

To further confirm this systems relevance, I compared the epigenetic landscapes of different primary cell types to ascertain whether similarities existed between neural and fibroblastic cells. Using publicly available ChIP-seq data, I generated a correlation

heatmap for three functionally distinct chromatin marks (H3K27me3, H3K27ac and H3K4me3) to visualise the similarity of chromatin mark distribution between different cell types (Fig. 10). For this analysis, I selected cell types from a diverse range of tissues and functional classes (e.g. epithelial, mesenchymal) to produce an accurate picture of how the neural cell epigenetic landscape related to other cell types. Reassuringly, in all instances, the analysis clustered haematopoietic, epithelial and mesenchymal cell types into distinct groups, validating the computational approach used to compare epigenetic landscapes (Fig. 10). Importantly, the analysis also revealed that astroglia, a neural cell type which resembles those constituting GBM, clustered with mesenchymal/fibroblastic cell types (Fig. 10). Moreover, the epigenetic landscape of astroglia was related to that of dermal fibroblasts (Fig. 10), the cell type used in the chosen model system. Together, this suggests that astroglial cells share a similar epigenetic landscape to mesenchymal cells, further justifying the use of the fibroblast-based system to study epigenetic events in neural neoplastic transformation. Importantly, although this model represented a suitable system in which to identify candidate molecular mechanisms, I planned to extensively validate any mechanism in normal and malignant cells from the CNS to confirm its relevance.

Figure 10. The epigenetic landscapes of astroglial and fibroblastic cells are related Correlation heatmap of ChIP-seq data from three different chromatin marks in eight primary cell types. All data sets were sourced from the encyclopaedia of DNA elements (ENCODE). PBMC – Peripheral blood mononuclear cells, CD4-T – CD4 T-cells, Mam – Mammary epithelia cells, Ker – Keratinocytes, Astro – Astroglial cells, Osteo – Osteoblasts, D-fib – Dermal fibroblasts, L-Fib – Lung fibroblasts.

3.2 Neoplastic transformation alters the chromatin binding profile of EZH2

3.2.1 Total EZH2 activity on chromatin is unaffected by transformation

To identify how EZH2 function is affected by neoplastic transformation, I first assessed EZH2 activity at each stage of de novo transformation. Previous studies have suggested that overexpression of EZH2, an event which often accompanies neoplastic transformation, may drive EZH2 to act as a tumour promoter through enhanced repression of existing target genes, such as the tumour suppressor CDKN2A/p16 (Gil & Peters, 2006; Wilson et al, 2010; Mohammad et al, 2017). To test whether a transformation-induced change in total EZH2 activity could underlie its role in cancer cells, I quantified EZH2 and H3K27me3 protein levels at each stage of de novo transformation. Levels of EZH2 increased upon transformation, whilst H3K27me3 remained unaltered (Fig. 11). This observation is in line with the known regulation of EZH2 expression by proliferation rate to maintain homeostatic levels of H3K27me3 (Bracken et al, 2003). Together, this suggests that a change of total EZH2 activity on chromatin is unlikely to underlie its gain of tumour-promoting ability. Of note, transformation-induced changes to EZH2 activity may be affecting global levels of H3K27me1 and 2. However, the biological function of these marks remains poorly defined, and they have no known regulatory role in either the normal CNS or neural malignancies, thus they were not examined in this study.

Figure 11. Total EZH2 activity is unaffected by neoplastic transformation

Figure 11: Western-blot quantification of EZH2 and H3K27me3 levels at each stage of de novo transformation. Histone H3 is used as a loading control. The graph (right) shows the normalised densitometric value of the western-blot bands (left). UT: untransformed, PN: pre-neoplastic, TR: transformed.

3.2.2 Profiling EZH2 and H3K27me3 genome-wide distribution by ChIP- seq

Redistribution of EZH2’s repressive activity is important for mediating cell state transitions in both the embryo and adult (Ezhkova et al, 2009; Di Croce & Helin, 2013). Given the functional importance of EZH2 redistribution in normal cells, I hypothesised that similar binding changes could be essential for EZH2 to become a tumour promoter. To characterise whether transformation induced changes in EZH2 distribution, I performed chromatin immunoprecipitation sequencing (ChIP-seq) to map EZH2 binding and its associated repressive mark, tri-methylated lysine 27 of histone H3 (H3K27me3), at all stages of de novo transformation. Quantitative PCR analysis of the chromatin isolated by ChIP showed a strong enrichment of EZH2 and H3K27me3 at the known Polycomb target gene WT1 relative to a negative control region in the promoter of SOX11, confirming the quality of ChIP experiments in all cellular states (Fig. 12A). Subsequently, next generation sequencing was performed on the isolated chromatin to profile EZH2 and H3K27me3 distribution genome-wide. All sequencing runs had a high per-base sequencing quality, low duplication level and yielded >10 million unique reads per sample, indicating that the sequencing was of excellent quality (Fig. 12B). Alignment and visualisation of the data revealed a strong enrichment of EZH2 and H3K27me3 at many conserved Polycomb binding sites, including WT1 (Fig. 13A). Furthermore, hierarchical clustering of EZH2 and H3K27me3 ChIP-seq data sets demonstrated that technical replicates were highly concordant (Fig. 13B). Together, I conclude that the ChIP-seq data sets generated are of sufficient quality to interrogate the effect of transformative events on EZH2’s distribution on chromatin.

Figure 12. EZH2 and H3K27me3 ChIP-seq quality control A. ChIP-qPCR quantification of EZH2 and H3K27me3 binding at the promoters of WT1 (positive control) and SOX11 (negative control) in ChIP samples from de novo transformed cells. Only data from replicate one of each cell line is shown. Values are expressed relative to input chromatin and represent mean ±SEM from three technical replicates. UT: untransformed, PN: pre-neoplastic, TR: transformed. B. Number of unique paired-end reads generated by next generation sequencing from each EZH2 and H3K27me3 ChIP sample. See ‘Materials & Methods’ for full description of read filtering pipeline. UT: untransformed, PN: pre-neoplastic, TR: transformed.

Figure 13. EZH2 and H3K27me3 ChIP-seq generates high quality datasets for the neoplastic transformation model A. ChIP-seq signal for EZH2 and H3K27me3 at known Polycomb target WT1. ChIP-seq signal normalised to sequencing depth is shown. UT: untransformed, PN: pre-neoplastic, TR: transformed. B. Correlation heatmap of EZH2 (left) and H3K27me3 (right) ChIP-seq duplicates from untransformed (UT), pre-neoplastic (PN) and transformed (TR) cells. Correlation values are calculated by comparison of EZH2/H3K27me3 read density across a set of common EZH2/H3K27me3 peaks.

3.2.3 Neoplastic transformation redistributes EZH2 on chromatin

To identify whether the distribution of EZH2 on chromatin was altered by transformation, I first set out to map the binding sites of EZH2 and H3K27me3 at each stage of de novo transformation. To do this, I first defined regions of EZH2 and H3K27me3 enrichment (false-discovery-rate (FDR) ≤0.0001) in each ChIP-seq replicate using the published peak caller SICER (Xu et al, 2014). By intersecting replicates and retaining the sum of overlapping peaks, I then generated consensus sets of EZH2 binding sites and H3K27me3 enriched regions for each cellular state (Fig. 14A). In line with EZH2’s canonical histone methyl transferase function, intersection of EZH2 and H3K27me3 consensus peaks revealed that >95% of EZH2 binding sites overlapped with a region of H3K27me3 enrichment in all cellular conditions (Fig. 14B-C).

Overall, each cellular condition had ~6000-8000 H3K27me3-associated EZH2 binding sites, a number similar to previous studies (Ku et al, 2008; Riising et al, 2014) (Fig. 15B). Interestingly, the transformed cells had ~30% fewer EZH2 binding sites than either the untransformed or pre-neoplastic cells (Fig. 15B). Previous work using this model has shown that transformation generates a cell population with heterogeneous self-renewal capacity (Scaffidi & Misteli, 2011). In part, this is explained by heterogeneous expression of linker histone H1.0, a protein which drives considerable changes to the epigenetic landscape and transcriptome when expressed (Torres et al, 2016). As EZH2 binding is sensitive to changes in transcription and the epigenetic landscape, this will also induce heterogeneity in EZH2 binding profiles, limiting EZH2 enrichment at any given site and hence, as observed, decreasing the total number of peaks detected. Inspection of the putative non-canonical sites (i.e. H3K27me3-independent binding) showed that most had associated regions of H3K27me3 enrichment which had not been called as peaks due to low ChIP-seq signal, suggesting that these were not true non-canonical sites (Fig. 14D). This lack of non-canonical EZH2 sites contrasts with previous studies which have shown that non-canonical functions of wild-type EZH2 on chromatin are essential for driving forms of prostate and breast cancer (Xu et al, 2012; Gonzalez et al, 2014; Kim et al, 2018).

Figure 14. EZH2 genome-wide distribution in de novo transformed cells A. A schematic outlining how consensus peak sets were derived for EZH2 and H3K27me3 in each cellular state of the neoplastic transformation model. UT: untransformed, PN: pre-neoplastic, TR: transformed. B. Percentage of EZH2 binding sites in each cellular state that have >25% overlap with a region of H3K27me3 enrichment. UT: untransformed, PN: pre-neoplastic, TR: transformed. C. ChIP-seq signal from a representative locus displaying concordant profiles of EZH2 and H3K27me3. ChIP-seq signal is shown normalised to sequencing depth. Blue bars indicate regions called as an EZH2 binding site. UT: untransformed, PN: pre-neoplastic, TR: transformed. D. ChIP-seq signal of EZH2 and H3K27me3 at two putative non-canonical EZH2

binding sites (H3K27me3-independent). ChIP-seq signal is shown normalised to sequencing depth. Blue bars indicate regions called as an EZH2 binding site. UT: untransformed.

To characterise the genomic features bound by EZH2, I annotated H3K27me3- associated EZH2 binding sites to the genomic region in which they were found e.g. gene bodies, gene deserts etc. This revealed that EZH2 binding was enriched at gene promoters, in line with EZH2’s canonical repressive function (Fig. 15A). For the next stage of analysis, I selected only EZH2 binding sites associated with H3K27me3, allowing me to focus on regions where EZH2 was likely to have repressive activity.

To identify EZH2 binding sites affected by transformation, I performed a comparison between the sites present at each stage of de novo transformation. Strikingly, this revealed that only a partial overlap existed between EZH2 binding sites (Fig. 15B). In particular, the untransformed and transformed states had <50% of EZH2 binding sites in common, indicating that transformation dramatically alters EZH2’s distribution (Fig. 15B). EZH2 redistribution occurred at all stages of transformation, but was especially pronounced at the transition from the untransformed to pre-neoplastic state, demonstrating that inhibition of p53 and pRB was a particularly strong driver of EZH2 redistribution (Fig. 15B). Together, characterisation of EZH2 distribution during de novo transformation shows that attainment of tumorigenic ability is accompanied by an extensive redistribution of EZH2 binding sites.

Figure 15. Transformation induces an extensive redistribution of EZH2 binding A. The fraction of H3K27me3-associated EZH2 binding sites found at different genomic features. ‘Whole genome’ indicates the distribution of features across the whole genome as a reference for enrichment of EZH2 peaks. Gene desert: gene depleted regions at least 1Mb long; Promoter: ±3kb of transcription start site (TSS); Genebody: from end of promoter until 1kb downstream of

transcription termination site; Other intergenic: regions not belonging to the other categories. B. Venn diagrams displaying the intersection between H3K27me3-associated EZH2 binding sites called at each stage of de novo transformation.

3.2.4 A fraction of transformation-induced changes to EZH2 binding are high-confidence

In spite of the extensive changes to EZH2 distribution upon transformation, it is unlikely that all events contribute to EZH2 becoming a tumour promoter in cancer cells. Thus, I decided to employ a multi-step filtering strategy that would identify functionally important changes in EZH2 binding. As a first filtering step, I identified sites which undergo a major change in EZH2 binding. To achieve this, I first merged the consensus EZH2 peak sets from the untransformed and transformed cellular states, generating a combined set of common and unique peaks. To find which peaks had large-magnitude changes in EZH2 binding, I calculated the relative enrichment of EZH2 at all peaks by comparing the number of ChIP-seq tag counts intersecting the peak in each cellular state. In line with intersection of EZH2 binding sites (Fig. 15B), a large portion of binding sites showed an enrichment of EZH2 in either condition, further supporting an extensive genome-wide redistribution of EZH2 (Fig. 16A). However, most of these changes were low-magnitude, and of 11,166 binding sites tested, only 703 (6%) exhibited high-confidence differential binding of EZH2 (defined as: fold-change ≥1.5 and p-value ≤1e-20) (Fig, 16A). No clear bias was observed for high-confidence gains or losses of EZH2 upon transformation, supporting a transformation-driven EZH2 redistribution as opposed to a change in total repressive activity. Similar quantification of EZH2 enrichment at binding sites in untransformed and pre-neoplastic or pre-neoplastic and transformed cells revealed comparable patterns (Fig. 16B). Interestingly, most high-confidence changes in EZH2 binding occurred at the transition from the untransformed to pre-neoplastic state (untransformed to pre-neoplastic – 528, pre-neoplastic to transformed - 120) (Fig. 16B), in line with my observations from comparing EZH2 binding sites (Fig. 15B). Moreover, this shows that attainment of full tumorigenic ability is associated with high-confidence changes at only ~1% of EZH2 binding sites. Together, this indicates that only a minor fraction of transformation-induced changes in EZH2 binding are of high-confidence, and hence potentially important for EZH2’s switch to tumour promoter.

Figure 16. Transformation-induced changes in EZH2 binding are primarily low-magnitude A. Volcano plot displaying the relative enrichment of EZH2 binding at all sites called in either untransformed or transformed cells. Numbers written within the volcano plot indicate differential binding sites that have a p-value ≤1e-20 and a fold-change in normalised tag count ≥1.5. Enrichment significance calculated based on a Poisson distribution using the ‘getDifferentialPeaks’ function in homer (see: ‘Materials & Methods’). P-values that were >1e-100 were altered to 1e-100 for display reasons. ChIP-seq tracks (right) show representative high- and low-magnitude EZH2 differential peaks. ChIP-seq signal is shown normalised to sequencing depth. Blue bars denote regions called as an EZH2 binding site. UT: untransformed, TR:

transformed. B. Volcano plots generated as in (A) comparing EZH2 enrichment at all sites detected in untransformed or pre-neoplastic cells (left) and pre-neoplastic or transformed cells (right).

3.2.5 EZH2 targeting to CpG islands is altered by transformation

Having defined a high-confidence set of differential EZH2 binding sites (fold-change ≥1.5 and p-value ≤1e-20, Fig. 15A – red dots), I then set out to characterise these regions of major EZH2 redistribution. DNA sequence and its associated chromatin features are often key determinants of protein binding at specific genomic loci. Thus, I calculated the nucleotide content for common, untransformed-specific and transformed-specific EZH2 binding sites. For this analysis, EZH2 binding sites present in both untransformed and transformed cells were used as common sites (Fig. 15B), whilst those with a fold-change ≥1.5 and p-value ≤1e-20 between states were used as cell type-specific sites (Fig. 16A – red dots). Strikingly, transformed-specific EZH2 sites had a significantly (p <0.0001) lower guanine/cytosine (GC) content than common or untransformed-specific sites (Fig. 17A). This suggests that an altered affinity for GC rich DNA sequences, or an associated chromatin feature, may be driving a portion of EZH2 redistribution. PRC2 binds to DNA preferentially at CpG islands (Ku et al, 2008) (see: ‘Introduction’), regions of high CpG dinucleotide content frequently present at gene promoters (Deaton & Bird, 2011). This binding modality is essential in allowing EZH2 to deposit H3K27me3 at gene promoters and maintain gene repression (Riising et al, 2014; Li et al, 2017). To understand if disruption of EZH2 localisation at CpG islands could explain the differing nucleotide content of EZH2 binding sites, I analysed how frequently cell-type specific/common sites intersected with CpG islands. Interestingly, ~40% of common and untransformed- specific sites intersected with a CpG island, but <20% of transformed-specific sites overlapped (Fig. 17B). The frequent intersection between EZH2 binding and CpG islands is in line with the known preferential binding of PRC2 at these sites; however, the greater tendency for EZH2 to be lost rather than gained at CpG islands suggests a transformation-induced alteration of EZH2 targeting to CpG islands. Inspection of regions where EZH2 was lost from CpG islands confirmed this observation (Fig. 17C). Together, I conclude that transformation alters EZH2 targeting to GC-rich regions such as CpG islands, potentially affecting the expression of associated genes.

Figure 17. EZH2 affinity for GC rich sequences, such as CpG islands, is altered by transformation A. Quantification of GC nucleotide content at EZH2 binding sites. Common: EZH2 binding sites detected in both untransformed and transformed cells, UT differential: high-confidence EZH2 binding sites enriched in untransformed cells (see Fig. 15A), TR differential: high-confidence EZH2 binding sites enriched in transformed cells (see: Fig. 15A). B. Quantification of EZH2 binding sites that intersect a CpG island. Definition of ‘Common’, ‘UT differential’ and ‘TR differential’ EZH2 binding sites as in (A). C. EZH2 ChIP-seq signal showing representative examples of CpG islands where an EZH2 binding site is lost on transformation. ChIP-seq signal is shown normalised to sequencing depth. UT: untransformed, TR: transformed.

3.2.6 Differential binding of EZH2 is enriched at genes essential for neural development

To identify if EZH2 redistribution was focussed at a functionally-related gene set, I annotated differential EZH2 binding sites to the nearest transcription start site (TSS) of a protein coding, anti-sense or lincRNA gene. I chose these specific gene classes as

their expression can be accurately quantified by total RNA-seq, a technique I planned to employ as a final step in my filtering of differential binding sites. I then selected gene loci with a differential binding site within 5kb of their TSS for gene set enrichment analysis (GSEA). To perform GSEA, I used the ‘Investigate Gene Sets’ function from the Molecular Signatures Database (Broad Institute) to compute overlaps between my experimental gene lists and existing genes signatures. GSEA identified a highly significant enrichment (FDR <1e-5) of differential site-associated genes in a number of neural-related gene signatures, including ‘central nervous system development’, ‘neurogenesis’ and ‘forebrain development’ (Fig. 18A). This linkage between EZH2 redistribution and neural development further suggested that this model system was a suitable selection for studying neural malignancies. To explore the specific classes of neural gene associated with differential EZH2 binding, I functionally classified the differential site-associated genes that were present in neural signatures. Of these genes, >50% were transcription factors, matching the EZH2-mediated repression of lineage specifying transcription factors in the developing CNS (see: ‘Introduction’) (Fig. 18B). This suggested the intriguing possibility that transformation-induced EZH2 redistribution could rewire core neural development programs. Inspection of ChIP-seq data confirmed marked changes in EZH2 levels at the promoters of many key neural transcription factors, such as IRX5 and SIM2 (Fig. 18C). Altogether, I conclude that EZH2 redistribution upon transformation is focussed at genes involved in mediating neural development, potentially rewiring neural transcriptional programmes.

Figure 18. EZH2 redistribution is focussed at neural development genes A. Neural-related GSEA gene signatures enriched amongst genes with a high-confidence, differential EZH2 binding site at their promoter (±5kb TSS). Gene signatures were obtained from GSEA Curated and Gene Ontology gene sets. The following gene signature names were altered from their original Molecular Signature Database descriptor for clearness: ‘CpG high promoters marked with H3K27me3 in the brain’ - ‘MEISSNER_BRAIN_HCP_WITH_H3K27ME3’, ‘CpG high promoters marked with H3K27me3 and H3K4me3 in the brain’ – ‘MEISSNER_BRAIN_HCP_WITH_H3K4ME3_AND_H3K27ME3’, ‘CpG high promoters marked with H3K27me3 in neural progenitor cells’ - ‘MIKKELSEN_NPC_HCP_WITH_H3K27ME3’. B. Functional classification of differential site-associated genes (total genes: 37) that are found in the gene signatures ‘Central Nervous System Development’ and ‘Neurogenesis’. C. ChIP-seq signal from EZH2 at neural transcription factor genes differentially bound by EZH2 upon transformation. ChIP-seq signal is shown normalised to sequencing depth. UT: untransformed, PN: pre-neoplastic, TR: transformed.

3.3 Transformation-induced redistribution of EZH2 affects the expression of a small, but important group of genes

3.3.1 Transformation substantially alters the EZH2-regulated transcriptome

Having identified that transformation redistributes EZH2 binding, I then wanted to specifically find alterations which affect gene expression. To do this, I began by characterising the EZH2-regulated transcriptome at each stage of de novo transformation through treating cells with the EZH2-specific pharmacological inhibitor EPZ-6438 (EZH2i) (Knutson et al, 2013b), or DMSO as a control. EZH2i treatment led to an almost complete depletion of H3K27me3 in all cell types, generating cells which I then profiled by total RNA-seq (Fig. 19A). All sequencing runs had a high per-base sequencing quality, low ribosomal RNA contamination, acceptable levels of read duplication and yielded >20 million reads. In addition, principle component analysis revealed high concordance between replicates, confirming the quality of these datasets (Fig. 19B).

Figure 19. Profiling the EZH2-regulated transcriptome using pharmacological EZH2 inhibition A. Western-blot analysis of H3K27me3 levels after treatment of cells with an EZH2 inhibitor (EZH2i) or DMSO as control for 12 days. Histone H3 is used as a loading control. UT: untransformed, PN: pre-neoplastic, TR: transformed. B. Principle component analysis of RNA- seq data from DMSO and EZH2i-treated cells. UT: untransformed, PN: pre-neoplastic, TR: transformed.

To identify genes regulated by EZH2 in each cellular state, I undertook differential expression analysis comparing the EZH2i and DMSO treated conditions. This analysis identified 900-1200 genes which were differentially expressed upon treatment with

EZH2i (Log2 fold-change ≥1 or ≤-1, FDR ≤0.01 and maximal transcripts per million (maxTPM) ≥1) (Fig. 20A). In line with the canonical repressive function of EZH2, ~90% of genes were upregulated upon treatment with EZH2i (Fig. 20A). GSEA of EZH2- regulated genes revealed they were significantly enriched (FDR <1e-47) amongst existing Polycomb gene signatures such ‘Benporath SUZ12 targets’ and ‘Genes with promoter H3K27me3 in ES cells’, validating the use of a pharmacological approach to characterise the EZH2-regulated transcriptome (Fig. 20B). Importantly, this also indicated that many PRC2-regulated genes are shared between different cell types, suggesting that identifying mechanisms underlying PRC2 function in one cell type could be relevant in others. EZH2-regulated genes were also significantly enriched (FDR <1e- 28) in neural-related signatures at all stages of de novo transformation, further confirming the importance of EZH2 in regulating neural development programmes within this system (Fig. 20B). To focus on potential direct targets of EZH2, I selected the genes upregulated by EZH2i for further analysis. Comparison of genes upregulated by EZH2i revealed that a substantial portion of differentially expressed genes were unique to each cellular state (Fig. 20C). In particular, only 35% of genes were commonly de-repressed by EZH2i in untransformed and transformed cells (Fig. 20C). Together, this demonstrates that transformation substantially alters the EZH2-regulated transcriptome, likely due to the accompanying redistribution of EZH2 at many genomic sites.

Figure 20. Transformation markedly alters the EZH2-regulated transcriptome

A. Quantification of genes differentially expressed (FDR ≤0.01, Log2FC ≥1/≤-1 and maxTPM ≥1) upon EZH2i treatment in each cellular state, as detected by RNA-seq. UT: untransformed, PN: pre-neoplastic, TR: transformed. B. PRC2 and neural GSEA gene signatures enriched amongst genes differentially expressed upon EZH2i treatment. Gene signatures are derived from GSEA Curated and Gene Ontology gene sets. The following gene signature name was amended from its original molecular signature database descriptor for clearness: ‘CpG high promoters marked with H3K4me3 and H3K27me3 in the brain’ – ‘MEISSNER_BRAIN_HCP_WITH_H3K4ME3_AND_H3K27ME3’. C. Venn diagrams showing the overlap between the genes EZH2i upregulates

(FDR ≤0.01, Log2FC ≥1 and maxTPM ≥1) in each cellular state.

3.3.2 DNA methylation may redundantly repress EZH2-bound genes

Surprisingly, when I overlapped genes with EZH2 bound at their promoter (within 5kb

TSS) with genes upregulated by EZH2i (Log2FC ≥1, FDR ≤0.01 and maxTPM ≥1), I found that only a fraction (14-24%) of EZH2-bound genes responded to EZH2i (Fig. 21A). Interestingly, this difference was independent of EZH2 or H3K27me3 levels at a genes promoter in all cell types (Fig. 21B). To confirm that the putative EZH2i insensitive genes were not being upregulated below the cut off of my gene expression filter, I calculated the expression change of these genes between DMSO and EZH2i treated cells. EZH2i led to a median increase in expression of only ~0.05 TPM at the genes classed as EZH2i insensitive, demonstrating that, despite being bound by EZH2, these genes were indeed unaffected to EZH2i (Fig. 21C).

Figure 21. A fraction of EZH2-bound genes respond to EZH2i

Figure 21: A. Number of EZH2-bound genes that are differentially expressed (FDR ≤0.01,

Log2FC ≥1 and maxTPM ≥1) upon EZH2i treatment (sensitive) or are unaffected (insensitive). UT: untransformed, PN: pre-neoplastic, TR: transformed. B. Average density profile of EZH2 and H3K27me3 at EZH2i sensitive and insensitive genes. Light red and grey areas represent the SEM of the average profile. C. Expression change upon treatment with EZH2i of all genes classified as EZH2i insensitive. Expression change was calculated as mean TPM EZH2i treated cells minus mean TPM DMSO treated cells. UT: untransformed, PN: pre-neoplastic, TR: transformed.

Instead, I hypothesised that a redundant epigenetic mechanism may be acting in conjunction with EZH2 to repress EZH2-bound genes insensitive to EZH2i. DNA methylation at promoter-associated CpG islands is a common mechanism of gene repression (Smith & Meissner, 2013), and previous work has shown that a direct interaction exists between PRC2 components and the DNA methylation machinery (Viré et al, 2006). To identify whether DNA methylation could be cooperating with EZH2 to redundantly repress genes, I performed bisulphite sequencing in transformed cells across the promoter-associated CpG islands of EZH2i sensitive and insensitive genes. Genes sensitive to EZH2i had lower levels of CpG methylation than their insensitive counterparts in all instances, suggesting that DNA methylation may cooperate with H3K27me3 to repress a sub-group of EZH2 target genes (Fig. 22A). Inspection of these genes showed them to have clear EZH2 and H3K27me3 peaks that intersected with the genes promoter (Fig. 22B).

Figure 22. EZH2 target genes insensitive to EZH2i are often marked by DNA methylation A. Bisulphite sequencing analysis quantifying DNA methylation of promoter-associated CpG islands at representative EZH2-bound genes sensitive or insensitive to EZH2i. Each row represents the sequence of an individual DNA molecule. Black and white circles represent methylated and unmethylated CpGs, respectively. For display reasons only the first 30 CpGs of CALB2 are shown, the remaining CpGs are also unmethylated. B. ChIP-seq signal for EZH2 and H3K27me3 at TNFRSF21 (EZH2i sensitive) and SPO11 (EZH2i insensitive). Black bars mark the CpG islands interrogated by bisulphite sequencing in (A). ChIP-seq signal normalised to sequencing depth is shown. TR: transformed.

To explore whether EZH2i insensitive genes were generally marked by DNA methylation, and to extend the analysis to glioma cell lines, I obtained whole-genome bisulphite sequencing data from glioma cell lines in the NCI60 data set and compared the promoter methylation status at all EZH2i sensitive and insensitive genes. All six glioma cell lines analysed showed higher levels of DNA methylation at EZH2i insensitive genes relative to EZH2i sensitive genes (Fig. 23A). Further, the same pattern was observed in an additional 54 cell lines from seven cancer types, suggesting this relationship was not cancer-type specific (Fig. 23B). Together, I conclude that high levels of DNA methylation often occur at EZH2 target genes insensitive to EZH2i, suggesting that these repressive mechanisms may act redundantly to restrict the genes sensitive to EZH2i, potentially limiting the transcriptional consequences of EZH2 redistribution.

Figure 23. Redundant repression by DNA methylation may occur at a large portion of EZH2- bound genes A. Average DNA methylation at the indicated sets of genes in NCI60 glioma cancer cell lines. Data sourced from the National Cancer Institute NCI60 cancer cell line dataset. B. As in (A) but for NCI60 non-glioma cancer cell lines. Cell lines from breast, lung, colon, leukaemia, melanoma, ovarian and prostate cancer are shown. Whiskers were removed for clarity. The cell lines indicated by x-axis numbers are shown in Appendix 5.

To explore whether the presence of both EZH2 and DNA methylation affected the degree to which a gene was repressed, I analysed the expression of EZH2i sensitive and insensitive genes. In line with these genes being EZH2 targets, <20% were expressed with a TPM >1 (Fig. 24); however, at all stages of de novo transformation, EZH2i sensitive genes (repressed by EZH2 alone) showed a low, but detectable, basal level of expression (median TPM ~0.25), whilst EZH2i insensitive genes (EZH2-DNA methylation repressed) were generally fully silenced (median TPM <0.1) (Fig. 24). This provides correlative evidence that redundant repression by EZH2 and DNA methylation may more effectively repress gene expression. Interestingly, basal expression of EZH2i-sensitive genes decreased as de novo transformation progressed (Fig. 24). Thus, I conclude that redundant repression of EZH2-bound genes by DNA methylation could lead to more robust repression of gene expression, although further functional studies will be required to confirm this relationship.

Figure 24. Redundant repression by EZH2 and DNA methylation may enhance gene silencing Quantification of expression by genes sensitive or insensitive to EZH2i as determined by RNA- seq. Expression of each gene was calculated as the mean transcripts per million (TPM) from three biological replicates. Four asterisks indicate p-value <0.0001 (One-way ANOVA followed by Tukey’s multiple comparisons test). UT: untransformed, PN: pre-neoplastic, TR: transformed.

100

3.3.3 A fraction of genes differentially bound by EZH2 undergo changes in expression

To identify differential EZH2 binding sites linked to altered gene expression, I integrated the data sets defining regions of differential EZH2 binding (Fig. 16A) and the EZH2- regulated transcriptome (Fig. 20C) to find sites which met three criteria: 1) The differential site was present at a genes promoter in untransformed or transformed cells (within 5kb

TSS); 2) The binding site was associated with a gene that was derepressed (Log2FC ≥1, FDR ≤0.01 and maxTPM ≥1) when the relevant cell type was treated with EZH2i i.e. EZH2 was required to maintain its repression; 3) The gene showed an appropriate change in expression upon transformation i.e. gene de-repression accompanied loss of EZH2 binding.

Application of the first filter showed that differential EZH2 binding sites were associated with the promoters of 151 and 71 genes in untransformed and transformed cells, respectively (Fig. 25A). The smaller number of promoter-associated transformed- specific sites was in contrast to the greater number of total transformed-specific sites (total sites: untransformed-specific – 313, transformed-specific - 390). Given that CpG islands are found primarily at gene promoters, this difference likely reflects the transformation-induced loss of EZH2 from CpG islands I had observed (Fig. 15B). When I applied the second filter, I found that only 47/222 genes required EZH2 to maintain their repression (Fig. 25A), in agreement with the insensitivity of ~75% of EZH2-bound genes to EZH2i (Fig. 21A). Finally, application of the third filter selected genes with an appropriate expression change upon transformation, giving a final set of 21 genes (untransformed – 14, transformed – 7) (Fig. 25B). Together, this demonstrates that only a fraction of transformation-induced EZH2 redistribution alters gene expression, suggesting that EZH2 gains tumour-promoting ability by changing the expression of only a small gene set.

101

Figure 25. EZH2 redistribution affects the expression of a small set of genes A. Schematic representation of the multi-step filtering strategy used to detect functionally important changes in EZH2 distribution induced by neoplastic transformation. Blue and red numbers represent the number of peaks/genes present after each filtering step in untransformed and transformed cells, respectively. UT: untransformed, TR: transformed. B. Heatmap showing the relative expression of genes identified in (A) at each cellular stage of de novo transformation. UT: untransformed, PN: pre-neoplastic, TR: transformed. Expression levels represent row normalised TPM values.

102

In spite of its small size, the signature of 21 genes showed a remarkable enrichment for functionally important gene categories (Fig. 25B). In line with earlier GSEA analysis (Fig. 18A), expression of neural transcription factors (EMX2, SIM2, HOXB9, NR6A1 and HOXA11) and regulators of normal neuronal function (CACNG8, SLC30A3, TENM4) was altered by EZH2 redistribution (Fig. 25B). This further supported the existence of an EZH2-mediated rewiring of neural developmental programmes. In addition, genes with established roles in GBM development (PREX1, BCL2) and CBX2, a component of the polycomb repressive complex 1 that binds to H3K27me3, also had their expression altered. Importantly, many EZH2-induced gene expression changes were specifically associated with the cell attaining tumourigenic ability (occurring at the pre-neoplastic to transformed cell transition), supporting their importance in tumour initiation and maintenance.

Interestingly, the final list of genes did not include the tumour suppressor CDKN2A/p16 (Fig. 25B). This contrasts with previous work which has suggested that repression of CDKN2A/p16 by EZH2 is a crucial event in many different cancer types (Gil & Peters, 2006; Wilson et al, 2010; Mohammad et al, 2017). Expression of CDKN2A remained constant across transformation, and its expression was not responsive to treatment with EZH2i (Fig. 26A). Furthermore, neither EZH2 nor H3K27me3 were present at the CDKN2A gene locus (Fig. 26B). Together, this indicates that wild-type EZH2’s attainment of tumour-promoting ability does not require repression of CDKN2A/p16. Of note, the SV40 large and small T-antigens expressed in this model of transformation act to inhibit the activity of Rb, a key tumour suppressor stabilised by p16 (LaPak & Burd, 2014). Thus, the selective pressure that drives EZH2 to repress CDKN2A/p16 is likely removed by Rb inhibition in this model of transformation, potentially explaining why no EZH2-mediated expression is observed.

103

Figure 26. EZH2 does not repress CDKN2A upon transformation A. Expression of CDKN2A/p16 in the indicated cellular states, treated with EZH2i or DMSO, as detected by RNA-seq. Expression values represent mean ±SEM from three biological replicates. B. ChIP-seq signal for EZH2 and H3K27me3 across the CDKN2A locus. ChIP-seq signal is shown normalised to sequencing depth. UT: untransformed, PN: pre-neoplastic, TR: transformed.

104

3.4 Redistribution of EZH2 induces a transcriptional switch at neural homeotic genes

3.4.1 EZH2 redistribution represses EMX2 and derepresses HOXB9

Amongst the final gene set, the homeotic genes empty spiracles homeobox 2 (EMX2) and homeobox B9 (HOXB9) stood out due to their distinct, non-overlapping roles in neural development. EMX2 is crucial for correct patterning of the forebrain (Cecchi 2002), and is also expressed in neural stem cells (NSCs) of the sub-ventricular zone (SVZ) from embryogenesis to adulthood (Cecchi, 2002) (Fig. 27). In NSCs, EMX2 is essential for restricting proliferation by limiting the frequency of symmetric cell divisions, in turn regulating the size of the NSC pool (Galli et al, 2002). Importantly, EZH2 is also expressed in SVZ NSCs where it too controls NSC proliferation and self-renewal (Hwang et al, 2014). In contrast, HOXB9 defines neuron identity in the developing spinal cord (Nolte & Krumlauf 2013) (Fig. 27). Initiating and maintaining correct spatial and temporal patterns of HOX gene expression is necessary to form the CNS (Nolte & Krumlauf, 2013). In the spinal cord, EZH2 maintains these pattern by depositing domains of repressive H3K27me3 across specific HOX gene loci to endow correct cellular identity (Mazzoni et al, 2013). Ectopic expression of HOX genes within the CNS leads to homeotic transitions and loss of cellular identity (Zhang et al, 1994; Shah et al, 2004), emphasising the importance of maintaining tight control on HOX gene expression. Together, the divergent functions of EMX2 and HOXB9 in normal CNS development indicate that an EZH2- mediated switch in their expression would lead to substantial changes in cellular identity, potentially transitioning cells to a state more favourable to GBM initiation and maintenance.

Figure 27. EMX2 and HOXB9 expression in the embryonic nervous system

105

Interrogation of RNA-seq expression data showed that EMX2 was specifically repressed at the transition from the pre-neoplastic to transformed state, and silencing was accompanied by the appearance of EZH2 and H3K27me3 peaks spanning the EMX2 promoter (Fig. 28A). Interestingly, EMX2-OS, the gene transcribed from the opposite strand to EMX2, was also repressed by EZH2 upon transformation, suggesting that these genes are co-ordinately regulated by EZH2 (Fig. 25B). In contrast, HOXB9 was strongly derepressed at the pre-neoplastic to transformed transition, with an associated loss of broad EZH2 and H3K27me3 peaks covering the whole HOXB9 locus (Fig. 28B). In addition to HOXB9, I also observed changes in EZH2 binding and H3K27me3 enrichment at sites across all four HOX clusters, with a general loss of EZH2 and H3K27me3 from posterior HOX genes (HOX6-13) (Fig. 29). Furthermore, my final signature of genes misregulated by EZH2 redistribution also included HOXA11 (Fig. 25B). Thus, despite choosing to focus on HOXB9, EZH2 binding changes at HOXB9 appeared to be a symptom of a broader misregulation of repressive chromatin at HOX clusters.

Figure 28. EMX2 is repressed and HOXB9 derepressed by EZH2 redistribution ChIP-seq signal for EZH2 and H3K27me3 (left) at EMX2 (A) and HOXB9 (B) loci. ChIP-seq signal is shown normalised to sequencing depth. mRNA expression of EMX2 and HOXB9 (right) as detected by RNA-seq. Expression values represent mean ±SEM from three biological replicates. UT: untransformed, PN: pre-neoplastic, TR: transformed. TPM: transcripts per million.

106

Figure 29. Transformation induces a general misregulation of repressive HOX chromatin ChIP-seq signal of EZH2 at HOX loci for all cellular stages of de novo transformation. ChIP-seq signal normalised to sequencing depth is shown. Only protein coding HOX genes are labelled. Note the loss of EZH2 from posterior HOX genes in transformed cells. UT: untransformed, PN: pre-neoplastic, TR: transformed.

107

To investigate if HOXB9 expression and EMX2 repression was retained in tumours, I generated cell lines from tumours initiated by injection of transformed cells into immunocompromised mice. I chose to derive tumour cell lines, as opposed to analysing sorted tumour cells, in order to allow a more controlled comparison with cultured de novo transformed cells and to remove non-proliferative/differentiated cells present in the tumour. Subsequently, I profiled gene expression in these cell lines by total RNA-seq to compare the expression profiles of cells able to form tumours and the initial transformed population. Importantly, EMX2 was silenced and HOXB9 expressed in all tumour-derived cell lines, further supporting the cancer relevance of an EMX2-HOXB9 expression switch (Fig. 30A). To independently validate that EMX2 and HOXB9 were direct targets of EZH2, I analysed their expression by reverse-transcription quantitative PCR (RT-qPCR) in untransformed and transformed cells treated with EZH2i. As expected, EZH2i treatment led to significant (p <0.01) upregulation of HOXB9 in untransformed cells and EMX2 in transformed cells, confirming them as direct EZH2 targets (Fig. 30B). As additional validation, I performed EZH2 knock out in transformed cells using CRISPR- Cas9 genome editing (Fig. 30C). EMX2 expression was significantly (p <0.0001) upregulated upon EZH2 knock-out, demonstrating direct repression of EMX2 by EZH2 (Fig. 30E). To identify if EZH2 also maintained EMX2 repression in tumours, I treated four tumour-derived cell lines with EZH2i (cell lines as Fig. 30A). In all instances, EZH2i drove a striking upregulation of EMX2 expression, indicating that EZH2 was also required to maintain EMX2 repression in tumours (Fig. 30F). Together, I conclude that transformation drives a redistribution of EZH2 that leads to the repression of EMX2 and a concomitant de-repression of HOXB9.

The initiation of EMX2 repression at the transition from pre-neoplastic to transformed cells indicates that oncogenic signalling from HRASv12 is ultimately responsible for redistributing EZH2 to the EMX2 locus (Fig. 28A). To test whether repression of EMX2 relied on continued signals from this initiating stimulus, I used transformed cells with a CRISPR knock-out of HRASv12 generated by Josep Monserrat Sanchez in the Scaffidi laboratory (Fig. 30D). Knock out of HRASv12 had no effect on EMX2 expression, suggesting that, once instigated, repression of EMX2 by EZH2 was independent of the initiating stimulus, confirming a stable epigenetic transition (Fig. 30E).

108

Figure 30. EMX2 and HOXB9 are bona fide targets of EZH2 A. Expression of EMX2 and HOXB9, as detected by RNA-seq, in cell lines derived from tumours generated by de novo transformed cells. Values represent mean ±SEM from two biological replicates. TPM: transcripts per million. B. RT-qPCR showing the expression levels of EMX2 and HOXB9 in the indicated cells treated with EZH2i or a DMSO control for 12 days. Values represent mean ±SEM from three technical replicates. Two asterisks indicate p-value <0.01 (one-tailed unpaired Student’s t-test). UT: untransformed, TR: transformed. C. Quantification by western-blot of EZH2 and H3K27me3 levels after CRISPR-Cas9 knock-out of EZH2 in transformed cells. EZH2 knock-out was performed using a population of cells expressing Cas9 and an EZH2-targeting sgRNA, as a result, not all cells had full EZH2 knock-out, leading to the residual level of EZH2 observed in the western blot (see: ‘Materials & Methods’). Histone H3 is used as a loading control. D. Quantification by western-blot of pan-RAS levels (no antibody is available against HRAS) after

109

CRISPR-Cas9 cutting HRASv12 locus. Cell line and western-blot generated by Josep Monserrat Sanchez in the Scaffidi laboratory. E. RT-qPCR analysis of EMX2 expression in transformed cells after CRISPR-Cas9 mediated knock-out of EZH2, HRASv12 or an eGFP control. Values represent mean ±SEM from three technical replicates. Four asterisks indicate p-value <0.0001 (one-way ANOVA followed by Dunnet’s multiple comparisons test). F. RT-qPCR showing the expression levels of EMX2 in four tumour-derived cell lines (as in: (A)) treated with EZH2i or a DMSO control for eight days. Values represent mean ±SEM from three technical replicates.

3.4.2 Transformation-induced repression of EMX2 likely involves the loss of enhancer activity

Expression of key developmental regulators is often tightly controlled through the activity of enhancer elements to ensure correct temporal and spatial expression patterns (Long et al, 2016). As such, during normal development, the expression of EMX2 and HOX genes is frequently regulated through the activity of their associated enhancers (Theil et al, 2002; Suda et al, 2010; Montavon & Duboule, 2013). Thus, I hypothesised that changes to the activity of these developmental enhancers could be linked to EZH2 redistribution at EMX2 and HOX genes. Previously, others in the group had characterised genome-wide changes to enhancers induced by transformation, also using de novo transformed fibroblasts (unpublished data). I therefore employed these datasets to investigate whether transformation altered enhancer activity at EMX2 and HOX genes. Remarkably, upon transformation, I observed loss of H3K27ac (chromatin mark of active enhancers), loss of H3K4me1 (marker of poised enhancers) and chromatin closure (assayed by FAIRE-seq) in five regions surrounding the EMX2 locus (Fig. 31A-B). This suggested the intriguing possibility that collapse of an EMX2-associated enhancer network may be involved in its repression upon transformation.

110

Figure 31. Transformation-induced silencing of EMX2 is concurrent with collapse of neighbouring enhancers A. ChIP-seq signal for H3K27ac, H3K4me1 and FAIRE-seq in the regions surrounding the EMX2 locus. ChIP-seq signal is shown normalised to sequencing depth. Numbers 1-5 indicate the five putative enhancers identified around the EMX2 locus. UT: untransformed, TR: transformed. B. Enlarged view of the five enhancers identified around the EMX2 locus. Numbers 1-5 indicate enhancers as denoted in (A). UT: untransformed, TR: transformed. C. Relationship between the boundaries of topologically associated domains (TADs) and the putative EMX2 enhancers. TAD boundaries (blue bars) were determined from IMR90 human lung fibroblasts (Dixon et al, 2012). ChIP-seq signal is shown normalised to sequencing depth. UT: untransformed, TR: transformed.

111

The ability of enhancers to regulate a gene is controlled by the relationship of gene and enhancer within the genomes three-dimensional structure. More specifically, enhancers can only regulate gene expression if they are present in the same topologically associated domain (TAD), self-interacting regions that define higher-order genome structure. To examine if these enhancers were in the same TAD as EMX2, I interrogated previously determined TAD boundaries from a fibroblast cell line. All five putative enhancers fell within the same TAD as the EMX2 locus, confirming their potential to control EMX2 expression (Fig. 31C). Together, I conclude that transformation induced changes in EMX2 expression were associated with the inactivation of a surrounding network of enhancers. Further investigations will be required to determine which, if any, of these enhancers are needed to drive expression of EMX2, and whether their inactivation upon transformation is an initiating event for EZH2 recruitment to the EMX2 promoter.

3.4.3 A switch from EMX2 to HOXB9 expression occurs during transformation in glioblastoma

To understand if the same expression switch occurs during transformation in the CNS, I began by comparing EMX2 and HOXB9 expression between normal neural cells and GBM cell lines. To characterise gene expression in the normal CNS, I employed available datasets from primary neural cells of the cerebrum - a region of the brain from which GBM commonly originates (Tamimi & Juweid, 2017). EMX2 was expressed in both fetal and mature astroglia, whilst HOXB9 was repressed in all cell types (Fig. 32A), in line with the expression patterns observed in untransformed cells (Fig. 28A-B). Recent evidence has suggested that GBM may originate in the astroglia-like NSCs of the SVZ (Lee et al, 2018b). To test expression patterns of EMX2, HOXB9 and EZH2 in this putative cell-of-origin, I interrogated gene expression datasets from in vitro-derived human neural progenitor cells and mouse SVZ NSCs. In both data sets, EMX2 and EZH2 were expressed, whilst HOXB9 was fully silenced (Fig. 32B-C), in line with the expression patterns observed in untransformed cells (Fig. 28A-B).

In contrast to normal neural cells, analysis of gene expression data from a panel of 62 glioma cell lines sourced from the cancer cell line encyclopaedia (CCLE) (Barretina et al, 2012) showed the opposite expression patterns. EMX2 was unexpressed in nearly all cell lines (61/62, RPKM ≤1), whilst HOXB9 was frequently expressed at high levels

112

(19/62, RPKM ≥5) (Fig. 32D). Many other HOX genes, from all four HOX clusters were also derepessed in glioma (Fig. 32E), in line with the general redistribution of EZH2 at the HOX clusters which I had observed by ChIP-seq (Fig. 26A). De-repression of HOX gene expression was especially frequent within the HOXB cluster, suggesting that this locus may be particularly susceptible to transformation-induced de-repression (Fig. 32E). Together, I conclude that neural neoplastic transformation induces silencing of EMX2 and de-repression of HOX genes.

113

Figure 32. Neural transformation leads to repression of EMX2 and de-repression of HOXB9

114

Figure 32: A. EMX2 and HOXB9 expression levels in primary human neural cells, as detected by RNA-seq. Data sourced from Brainseq2 (see: ‘Materials & Methods’). Expression values represent mean ±SEM from four, twelve, one and five biological replicates of fetal astroglia, mature astroglia, neurons and oligodendrocytes, respectively. FPKM: fragments per kilobase million. B. Expression levels of EZH2, EMX2 and HOXB9 in neural progenitor cells, as detected by RNA-seq. Error bars represent upper and lower bounds of 95% confidence limits. Data sourced from ENCODE. TPM: transcripts per million. C. Expression levels of EZH2, EMX2 and HOXB9 in SVZ active neural stem cells (aNSCs) and quiescent neural stem cells (qNSCs), as detected by microarray. Data sourced from Codega et al, 2014. Expression values represent mean ±SEM from three biological replicates. D. EMX2 and HOXB9 expression levels in 62 glioma cell lines. Data sourced from CCLE. Each spot represents a cell line. RPKM: reads per kilobase million. E. HOX gene expression levels in 62 glioma cell lines as detected by RNA-seq. Data sourced from the CCLE. Numbers above bars represent cell lines in which expression of the corresponding genes is detected (reads per kilobase million (RPKM) ≥1).

3.4.4 EZH2 mediates EMX2 silencing in glioblastoma

To confirm that EZH2 represses EMX2 in malignant cells, I treated a panel of six GBM cell lines with EZH2i. In all GBM cell lines, EZH2i led to a dose-responsive upregulation of EMX2 expression, with increases ranging from 2-250 fold (Fig. 33A). This confirmed that EZH2 was indeed necessary to maintain EMX2 repression in GBM. To demonstrate that EZH2 acted directly at EMX2’s promoter, I performed ChIP-qPCR on the M059K GBM cell line and interrogated the EZH2 enrichment at the promoters of EMX2 and HOXB9. As expected, EZH2 showed strong enrichment at the promoter of EMX2, but only a slight enrichment at the promoter of HOXB9 (Fig. 33B), mirroring the binding pattern of de novo transformed cells (Fig. 28B). The slight enrichment of EZH2 at HOXB9 was in line with the low levels of EZH2 binding observed in ChIP-seq of transformed cells, suggesting that either a subpopulation of cells retains EZH2 repression at HOXB9 after transformation or low-level, non-repressive binding of EZH2 persists in all cells (Fig. 28B).

Although all GBM cell lines had upregulated EMX2 upon treatment with EZH2i, it was striking that the magnitude of upregulation varied from 2-250 fold. As I had previously observed that DNA methylation often compliments EZH2 repression, I hypothesised that the difference in responsiveness to EZH2i could be explained by the same mechanism. Thus, I interrogated publically available DNA methylation data from the CCLE for the

115

GBM cell lines I had treated with EZH2i. As expected, the magnitude of EMX2 upregulation induced by EZH2i treatment showed a robust anti-correlation with the average level of DNA methylation at the EMX2 locus (Fig. 33C). This suggests that multiple epigenetic mechanisms are employed to silence EMX2, indicating the importance of maintaining its repression in GBM.

Figure 33. EZH2 maintains EMX2 repression in glioblastoma A. EMX2 expression levels quantified by RT-qPCR in five different GBM cell lines upon treatment with 1μM, 3μM or 10μM of EZH2i or a DMSO control for eight days. Values represent mean ±SEM from three technical replicates. B. EZH2 binding at the promoters of EMX2, HOXB9 and GAPDH (negative control) quantified by ChIP-qPCR in M059K GBM cells. Values represent mean ±SEM from two biological replicates. C. Relationship between the magnitude of EMX2 upregulation by EZH2i and the average DNA methylation level at the EMX2 promoter in GBM cell lines. Values for EMX2 upregulation are from treatment with 10μM EZH2i (see: (A)). DNA methylation values are averages from CpGs across the promoter of EMX2 in GBM cell lines. All data sourced from the CCLE.

116

3.4.5 EZH2 mediates EMX2 repression in many cancer types

Having confirmed that EZH2 represses EMX2 in GBM, I then wanted to explore whether the same interaction was relevant in other cancer types. To do this, I treated seven patient-derived xenograft (PDX) cell lines with similar genetic backgrounds to my initial model (p53-/- and expressing oncogenic RAS) and four established cancer cell lines with EZH2i, and then monitored expression of EMX2 by RT-qPCR. Strikingly, EZH2i led to a dose-responsive upregulation of EMX2 in nine cell lines from six different cancer types (Fig. 34A). A particularly strong and consistent response was observed in lung cancer cell lines where EMX2 was upregulated up to 190-fold, suggesting that EZH2-mediated repression of EMX2 may be particularly important in lung cancer (Fig. 34A). ChIP-qPCR analysis of EZH2 and H3K27me3 levels at the EMX2 promoter in a lung cancer PDX cell line confirmed that EMX2 was directly bound by EZH2 in this cancer-type (Fig. 34B). Together, this demonstrates that EZH2-mediated repression of EMX2 occurs across many cancer types.

Figure 34. EZH2 maintains EMX2 repression in many cancer types

A. Heatmap showing the Log2 fold change in EMX2 expression detected by RT-qPCR after treatment of PDX-derived (blue) and established cancer cell lines (black) with EZH2i for eight days. Values are from three technical replicates. B. Quantification of EZH2 and H3K27me3 binding at EMX2 and GAPDH (negative control) promoters by ChIP-qPCR in LXFL 1674 PDX cells. Values represent mean ±SEM from two biological replicates.

117

3.4.6 EMX2 and HOX genes are misregulated in glioblastoma patients

To investigate whether repression of EMX2 and de-repression of HOXB9 were relevant events in a clinical setting, I interrogated publically available gene expression data from patient GBM samples. The repository of molecular brain neoplasia data (REMBRANDT) contains microarray expression data from 214 GBM tumours and 21 normal brain samples, representing one of the largest data sets available for GBM. As such, I began by exploring REMBRANDT data to compare the expression of EMX2 and HOXB9 between normal brain and GBM. As predicted, EMX2 was strongly repressed in GBM relative to normal brain, and in many instances was almost completely silenced (53/214 patients, expression <20% normal brain mean), whilst HOXB9 showed a small, but significant (p <0.01), upregulation (Fig. 35A). As a cautionary note, ‘normal’ brain samples do not necessarily capture the gene expression profile of the cell type from which the tumour originates, and thus this comparison may not reveal true transformation-induced changes in expression. Nonetheless, EMX2 was consistently expressed and HOXB9 was completely silenced in normal neural cell types relevant to GBM (Fig. 32A-C). Hence, the observation that in many tumours EMX2 was robustly silenced, and HOXB9 was derepressed, supported the occurrence of an EMX2-HOXB9 expression switch in patient tumours.

The increase of HOXB9 expression in GBM tumours, although significant (p <0.01), was small in magnitude. To investigate the expression patterns of HOXB9 in more detail, I obtained samples from resections of GBM tumours to perform RNA fluorescence in situ hybridisation (RNA-FISH) against HOXB9 mRNA. As a positive control, I stained cell pellets containing a mixture of cells either overexpressing HOXB9 or with no HOXB9 expression. As expected, the positive control showed high HOXB9 signal, but only in a fraction of cells, whilst a negative control cell pellet had no signal (Fig. 35B). When I stained GBM tumour samples, it revealed a high, but extremely heterogeneous, expression of HOXB9 throughout the tumour (Fig. 35C). This suggests that the slight upregulation of HOXB9 seen by microarray analysis of bulk tumours was the averaging of expression from a small HOXB9hi population and large numbers of HOXB9lo cells. Together, I conclude that HOXB9 is upregulated to physiologically relevant levels in a sub-population of GBM cells.

118

Figure 35. EMX2 is repressed and HOXB9 derepressed in glioblastoma tumours

119

Figure 35: A. EMX2 and HOXB9 expression in normal brain and GBM patient samples, as detected by microarray. Data sourced from REMBRANDT. Two asterisks indicate p-value <0.01, four asterisks indicate p-value <0.0001 (two-tailed Mann Whitney U test). Error bars represent mean ±standard deviation (SD). All values are shown relative to the mean of normal brain samples. N: 21 - normal brain, 214 - GBM. B. Visualisation of HOXB9 mRNA (red) by RNA-FISH in negative (left) and positive (right) control cell pellets. Composition of control cell pellets is outlined in the methods section. Nuclei were counterstained with DAPI. Scale bar: 20 µm. C. Visualisation of HOXB9 mRNA (red) by RNA FISH in a human GBM tumour. White arrow (bottom left) indicates auto fluorescence from red blood cells. Nuclei were counterstained with DAPI. Scale bar: 100 µm.

In addition to HOXB9, I found that many other HOX genes were derepressed in GBM, suggesting disruption of repressive chromatin at HOX clusters is also relevant in tumours (Fig. 36B). Once again, this emphasised that the changes I had observed at HOXB9 in fibroblasts were just an indicator of general alterations to repressive chromatin at HOX clusters in GBM. Interestingly, the HOX genes showing the greatest upregulation in GBM were the same genes which had the highest levels of expression in glioma cell lines (see: HOXB1, HOXB2, HOXB7, HOXC6, HOXC10 and HOXD10 in Fig 32E and 36B), suggesting that glioma cell lines are an excellent proxy for identifying important facets of the tumour transcriptome.

120

Figure 36. HOX genes are de-repressed in glioblastoma Relative expression of HOX genes in normal brain and GBM patient samples, as detected by microarray. Data sourced from REMBRANDT. Values are shown relative to the median of normal brain samples. N: 21 - normal brain, 214 – GBM.

3.4.7 EZH2 mediates EMX2 repression in glioblastoma tumours

To identify whether EZH2 mediates EMX2 repression in GBM, I characterised the covariance of EZH2 and EMX2 expression in GBM tumours. This revealed a strong and significant (p <0.0001) anti-correlative relationship between EZH2 and EMX2, in line with EZH2 mediating EMX2 repression (Fig. 37A). To explore whether this same relationship existed within individual tumours, I interrogated expression data from laser-

121

microdissected regions of GBM tumours and adjacent normal tissue sourced from the ivy allen glioblastoma atlas (Puchalski et al, 2018). As expected, EZH2 expression was strongly upregulated in tumour regions vs matched adjacent normal tissue (7/8 tumours), whilst EMX2 expression was repressed (6/8 tumours), further supporting EZH2- mediated repression of EMX2 (Fig. 37B). Furthermore, the genetically and spatially matched normal brain samples provided additional well controlled evidence for EMX2 repression in GBM versus normal brain. Surprisingly, HOXB9 showed no expression in either GBM tumour or adjacent matched normal tissue (data not shown), possibly due to the high but heterogeneous expression of HOXB9 in GBM tumours not being captured by the restricted regions sampled by laser microdissection. However, interrogation of expression at other HOX genes revealed that many showed a strong and significant (p <0.0001) upregulation between normal tissue and tumour, in line with HOX cluster misregulation in GBM (Fig. 37C).

122

Figure 37. EZH2 and EMX2 expression anti-correlates in patient tumours

123

Figure 37: A. Covariance between EMX2 and EZH2 expression levels in GBM patient samples, as detected by microarray. Data sourced from REMBRANDT. Correlation coefficient (r) and p- value of the covariance are shown (Spearman rank correlation). Every dot is a patient. N: 214. B- C. Expression levels of EMX2 (left top), EZH2 (right top), HOXA10 (left bottom) and HOXB7 (right bottom) in tumour or adjacent normal regions laser microdissected from human GBM tumours, as detected by RNA-seq. Data sourced from the Ivy Glioblastoma Atlas. FPKM: fragments per kilobase million. The significance of the differential expression in normal and tumour regions is indicated (two-way ANOVA). Error bars represent mean ±SEM. N: 3 regions sampled for each GBM tumour, 2, 3, 2, 3, 2, 3, 3 and 1 regions for normal tissue of patients 1-8, respectively.

To investigate the relationship between EZH2 and EMX2 expression in individual tumour cells, I performed dual colour RNA-FISH against EMX2 and EZH2 mRNA in GBM samples. As a positive control, I used a cell pellet containing a mix of cells either overexpressing EMX2 or expressing EMX2 at endogenous levels. As expected, the positive control showed mixed regions of high and moderate EMX2 expression across the cell pellet, whilst a negative control cell pellet had no signal (Fig. 38A). Staining of GBM samples revealed homogeneous, high EZH2 and low EMX2 expression across the tumour (Fig. 38B), in line with observations from gene expression data (Fig. 35A). However, in the few regions where EMX2 and EZH2 were both expressed, there was evidence of an antithetic relationship between the expression of both genes at the single cell level (Fig. 38C). Confirming this anti-correlative relationship will require a detailed quantitative analysis of EZH2 and EMX2 expression in tumour cells imaged from across a panel of patient samples, a potential future direction for this study. Together, this data indicates that EZH2 is likely responsible for repressing EMX2 expression in GBM tumours.

124

Figure 38. EMX2 and EZH2 expression has an antithetic relationship in tumour cells A. Visualisation of EMX2 mRNA by RNA-FISH in positive (right) and negative (left) control cell pellets. Composition of cell pellets is described in ‘Materials & Methods’. Nuclei were counterstained with DAPI. Scale bar: 20 µm. B-C. Visualisation of EZH2 (green) and EMX2 mRNA (red) in a human GBM tumour by RNA FISH. Nuclei were counterstained with DAPI. Scale bar: 20 µm.

125

Having identified that a strong anti-correlative relationship existed between EZH2 and EMX2 in GBM, I explored whether the same was true at all glioma grades (GBM is grade IV glioma - the most aggressive form with the poorest prognosis). To do this, I used data from the Chinese glioma genome database which contains gene expression profiles from grade II-IV gliomas. At all three grades, I observed a significant (p <0.01 in all instances) inverse relationship between EZH2 and EMX2 expression, confirming that EZH2- mediated repression of EMX2 likely occurs in all forms of glioma (Fig. 39A). Furthermore, the expression of EMX2 significantly (p <0.0001) decreased at higher glioma grades, whilst EZH2 expression increased, suggesting that expression of EMX2 was a clinically relevant factor in glioma (Fig. 39B). Interrogation of HOX gene expression revealed that many genes showed increased expression with tumour grade, indicating that a general deregulation of HOX gene expression was a clinically relevant feature in glioma (Fig. 39C). Together, this suggests that repression of EMX2 by EZH2 occurs at all stages of glioma progression, and expression of EMX2 and HOX genes is a clinically relevant feature in glioma.

126

Figure 39. EZH2 mediates EMX2 repression at all grades of glioma

127

Figure 39: A. Covariance between EMX2 and EZH2 expression levels in glioma patient samples as detected by RNA-seq. Data sourced from the Chinese Glioma Genome Atlas. P-value and correlation coefficient (r) of the covariance are shown (Spearman rank correlation). Every dot is a patient. N: 109 - grade II, 72 - grade III, 144 - grade IV. Samples with EMX2 or EZH2 expression of 0 are not shown, but are included in significance calculations. B-C. Expression of EMX2 (top left), EZH2 (top right) and HOX genes (bottom) in glioma patient samples, grouped by tumour grade, as detected by RNA-seq. Expression data was not available for HOXA3, HOXA9, HOXB1, HOXC6, HOXC12 and HOXD12. Data sourced from the Chinese Glioma Genome Atlas. Four asterisks indicate p-value <0.0001 (one-way ANOVA test). Error bars represent mean ±SD. N: as in (A). FPKM: fragments per kilobase million.

3.4.8 Repression of EMX2 by EZH2 may be necessary for tumour maintenance in glioblastoma

Having confirmed that EZH2-mediated silencing of EMX2 occurs in GBM, I hypothesised that maintenance of EMX2 repression was essential for wild-type EZH2’s tumour- promoting role. One prediction of this model is that re-expression of EMX2 in GBM should lead to impairment of tumour initiation and maintenance. To test this, I re- expressed EMX2, or RFP as a control, in two different GBM cell lines. In both cell lines, re-expression of EMX2 halved the number of cells over the course of the experiment, demonstrating that repression of EMX2 was indeed important for maintaining cell growth in GBM (Fig. 40). Interestingly, the relative decrease in cell number took eight days to manifest, suggesting that re-expression of EMX2 was not directly affecting the cell cycle and instead was altering the cells long-term proliferative potential, in line with EMX2’s ability to inhibit symmetric cell division in NSCs (Galli et al., 2002).

128

Figure 40. EMX2 re-expression impairs glioblastoma cell growth Proliferation assay quantifying the effect of EMX2 over expression in U-87 MG and DBTRG-05MG GBM cells, RFP is used as a control. Values represent mean ±SEM from three biological replicates. The cell number fold change after eight days of proliferation with expression of EMX2 or RFP is shown relative to an uninduced control. Two asterisks indicate p-value <0.01 (one-tailed unpaired Student’s t-test).

As a more stringent test of the functional consequences of EMX2 re-expression, I examined the effect of EMX2 ectopic expression in a GBM xenograft model. To ensure the physiological relevance of this assay, I re-expressed EMX2 in the GBM cell line DBTRG-05MG at levels comparable to those seen in the pre-neoplastic cells of the neoplastic transformation model (Fig. 41A). I then transplanted 350k cells expressing ectopic EMX2, or RFP as a control, into immunocompromised mice. Cells expressing EMX2 were unable to form tumours over the time course of the experiment, whilst cells expressing RFP initiated tumours in all instances (Fig. 41B). Corroborating this, independently-derived RFP and EMX2 expressing cell lines showed a similar result. Together, this shows that EMX2 is an extremely potent tumour suppressor in GBM and its repression by EZH2 is likely to be an important event for GBM initiation and maintenance.

129

Figure 41. EMX2 is a potent tumour suppressor in glioblastoma A. RT-qPCR quantification of relative EMX2 levels in DBTRG-05MG GBM cells, untransduced (endogenous) or transduced with pTRIPZ-EMX2 (exogenous). Pre-neoplastic fibroblasts acted as a standard for endogenous EMX2 levels in non-transformed cells. Note that expression of the doxycycline-inducible pTRIPZ construct was not induced and the detected mRNA represents leaky expression. Values represent mean ±SEM from three technical replicates. B. Transplantation assay comparing the growth kinetics of DBTRG-05MG induced tumours expressing EMX2 at levels comparable to those expressed in normal cells, or expressing RFP as a control. Values represent mean ±SEM from six biological replicates. Three asterisks indicate p- value <0.001 (two-tailed single sample Student’s t-test).

130

Chapter 4. PRC2 composition is altered by neoplastic transformation

Introduction

The ability to form functionally diverse sub-complexes is an important feature of many epigenetic regulators. Through controlling the balance between these different sub- complexes, a cell is able to initiate changes in the epigenetic landscape that drive normal development (Gil & O’Loghlen, 2014; Meeks & Shilatifard, 2017; Mathur & Roberts, 2018). In mammals, compositional diversity is particularly marked in the Polycomb repressive complexes, PRC1 and PRC2, where an array of proteins bind in a sub- stoichiometric manner to the core complexes and endow specific functionalities (Blackledge et al, 2015). In particular, these proteins can bind certain histone modifications, DNA sequences and RNA structures, in turn controlling the distribution of the complexes catalytic activity (Blackledge et al, 2015). For example, during differentiation of embryonic stem cells, the PRC1 component CBX7 is repressed and other CBX homologues are upregulated, leading PRC1 to switch from repressing lineage-specific to pluripotency genes (O’Loghlen et al, 2012). Although previous work has focussed primarily on the more compositionally-diverse PRC1 complex, emerging evidence suggests that the PRC2 also undergoes alterations to its configuration during normal development (Kloet et al, 2016).

Remarkably, components of PRC1/2 sub-complexes crucial for mediating normal development, are also important factors in a range of malignancies (Gil & O’Loghlen, 2014; Vizán et al, 2015) (see: ‘Introduction’). Specifically, loss/gain-of-function mutations, and transformation-induced expression changes, can affect these components in cancer. This indicates that hijacking or impairing the function of specific PRC1/2 sub-complexes can be an important event in tumorigenesis. Given the importance of PRC1/2 composition changes in mediating cell-state transitions in normal development (O’Loghlen et al, 2012; Kloet et al, 2016), and the established role of sub-complex- specific components in cancer (O’Loghlen et al, 2012; Vizán et al, 2015), I hypothesised that transformation-induced changes to the make-up of wild-type PRC1/2 may accompany, and possibly promote, transformation (Fig. 42). The effect of transformation on wild-type PRC2 composition is particularly pertinent in light of the tumour-promoting redistribution of EZH2 that transformation also induces (see previous chapter). Although

131

oncogenic signalling driven by HRASv12 is the upstream factor initiating EZH2 redistribution in this instance, the downstream changes this causes to alter EZH2 binding sites remains unclear. Hence, I decided to characterise how the composition of wild-type PRC2 is affected by transformation. In doing so, I hoped to provide insights into how transformation is able to redistribute EZH2 activity, and potentially draw parallels to changes in complex composition occurring during normal development.

Figure 42. Effect of transformation on the composition of PRC2

132

Aims

In this chapter, I describe experiments in which I characterised transformation-induced changes to the protein composition of wild-type PRC2. To do this, I first employed quantitative proteomic techniques to identify changes in PRC2 composition occurring upon transformation in de novo transformed cells. Subsequently, using gene expression and proteomic data I explored how changes to complex composition occur. Using in vivo tumour models, I then validated the functional importance of changes to PRC2 composition, and identified context-dependent roles for sub-stoichiometric components of the PRC2. Finally, I explored whether a link exists between the tumour-promoting EZH2 redistribution identified in the previous chapter and changes in PRC2 composition.

133

Results

4.1 Neoplastic transformation alters the composition of PRC2

4.1.1 Immunoprecipitation of endogenous EZH2 captures all known PRC2 interactors

To identify transformation-induced changes to the composition of PRC2, I decided to perform immunoprecipitation (IP) of endogenous PRC2 from untransformed and transformed cells followed by proteomic analysis. I chose to undertake proteomic characterisation as it allowed unbiased identification of changes to PRC2 composition, overcame the limited availability of high quality antibodies for PRC2 components and offered the possibility of discovering novel interactors. To perform this analysis, I once again employed untransformed and transformed cells from the de novo transformation model (Fig. 9). This system recapitulates transformation-induced changes to wild-type EZH2 function found in cancer cells (see previous chapter), making it a relevant choice for studying the composition of PRC2, and results from the proteomic analysis could be integrated with existing ChIP-seq and RNA-seq datasets generated for this system. To validate whether transformation-induced changes to PRC2 composition were also retained in tumours, I included in my workflow comparisons between untransformed cells and two cell lines derived from tumours generated through injection of transformed cells into immunocompromised mice.

As a first step in characterising the composition of PRC2, I began by developing a co-IP protocol able to capture endogenous EZH2 and its binding partners. I selected EZH2 as my bait protein as it is a core component of PRC2 essential for gene repression, and I had previously validated an EZH2 antibody for use in co-IP (see previous chapter). In addition, specifically characterising transformation-induced changes in the binding partners of EZH2 could help to explain its tumour-promoting redistribution. For the IPs, I used whole-cell lysates to enable identification of the composition of nucleoplasmic and chromatin-bound PRC2. In addition, to retain the maximum number of interactors, I lysed cells in high salt buffer, a low-stringency buffer that induces minimal disruption to protein interactions (see: Materials & Methods’ for further details). As an initial test, I performed IP of endogenous EZH2 using lysates from transformed cells. Western blotting of the resulting eluate revealed that the core PRC2 component SUZ12 was effectively co- immunoprecipitated, indicating that my protocol was capturing known PRC2 interactors

134

(Fig. 43A). To confirm that the full suite of PRC2 components was being captured, I characterised the protein composition of the eluates from EZH2 and isotype-control IgG IPs by mass spectrometry. Reassuringly, all known PRC2 interactors, including the recently identified proteins EPOP and PALI (Beringer et al, 2016; Conway et al, 2018), were detected exclusively in the EZH2 IP (Fig. 43B). As expected, the core PRC2 components SUZ12, EED and RBBP7/4 showed that highest signal intensity, whilst sub- stoichiometric complex members such as PHF19 and MTF2 were detected with a lower signal, in line with their relative frequency of binding to the PRC2 (Fig. 43B). Together, I conclude that my co-IP protocol is suitable to characterise changes in PRC2 composition occurring upon transformation.

Figure 43. Endogenous EZH2 co-immunoprecipitation captures the PRC2 interactome A. Western-blot quantification of SUZ12 levels in eluates from co-IPs performed with an EZH2 (α-EZH2) or isotype control antibody (Ctl IgG) using transformed cell lysate. 1:150 and 1:300 represent the concentration of EZH2 antibody used in the EZH2 co-IP. Input (left) was exposed for longer period than the co-IP samples (right). B. Quantification by mass spectrometry of proteins present in the co-IP eluates from (A). For display reasons, proteins detected in only one sample were assigned an arbitrary low iBAQ value based on a normal distribution for the other sample. All PRC2 components were detected only in the EZH2 co-IP sample. iBAQ: intensity based absolute quantification.

135

4.1.2 Transformation alters the protein composition of PRC2

Tightly regulated changes to the ratios of Polycomb sub-complexes are recurrent events during normal development, and similar mechanisms may be relevant in tumourigenesis. However, a precise comparison of protein abundance cannot be performed for samples which have been analysed independently on a mass-spectrometer (as in Fig. 43B) due to confounding inter-run variability - preventing the detection of subtle transformation- induced changes to PRC2 composition. In contrast, stable isotope labelling by amino acids in cell culture (SILAC) is a technique which enables precise comparison of protein abundance between samples by mass spectrometry. In SILAC, cells are cultured in media containing only 13C (‘heavy’) or 12C (‘light’) labelled Arginine and Lysine, leading the cells proteome to become isotopically labelled. This labelling allows identical proteins from ‘heavy’ and ‘light’ cells to be distinguished by their mass difference in a mass- spectrometer. Hence, co-IP coupled with mass-spectrometry for two populations of cells can be performed and analysed as a single sample, overcoming inter-run variability and allowing accurate quantification of relative protein levels. Thus, I employed SILAC to compare the composition of PRC2 in untransformed and transformed/tumour-derived cells (Fig. 44).

Figure 44. Workflow for quantifying transformation-induced changes in PRC2 composition Schematic outlining the workflow for quantifying the effect of transformation on PRC2 composition. UT: untransformed, TR: transformed, m/z: ion mass.

To label the proteome of untransformed and transformed/tumour-derived cells, I cultured cells with ‘heavy’ or ‘light’ amino acid-containing media for seven days. Subsequently, I mixed equal quantities of labelled cell lysates and performed IP of endogenous EZH2 to compare the composition of PRC2 in untransformed and transformed/tumour-derived

136

cells. To control for alterations in PRC2 composition induced by SILAC labelling, I also immunoprecipitated EZH2 from a mix of ‘heavy’ and ‘light’ labelled lysates of untransformed cells. All IP eluates were interrogated by mass-spectrometry to characterise relative quantities of PRC2 components isolated from each cell type. As expected, in the control IP, EZH2 co-immunoprecipitated with equal quantities of all PRC2 interactors in the ‘heavy’ and ‘light’ labelled cells, indicating that SILAC did not affect the composition of the PRC2 (Fig. 45A). In IPs comparing untransformed and transformed/tumour-derived cells, the core PRC2 components SUZ12, EED and RBBP4/7 co-immunoprecipitated with EZH2 to an equal extent in both cell types, indicating that, as expected, the core PRC2 complex was not altered by transformation (Fig. 45B-C). However, specific sub-stoichiometric components did show reproducible changes in their association with EZH2 (Fig. 45B-C). In particular, relative to untransformed cells, both transformed and tumour-derived cells exhibited a ~2-fold decrease in PHF19 and PHF1 binding to EZH2, whilst MTF2 and AEBP2 showed a concurrent ~2-fold increase (Fig. 45B-C). PHF1, PHF19 and MTF2 are closely related homologues which constitute part of the PRC2 in a mutually exclusive manner (Hauri et al, 2016), in turn enabling it to bind the chromatin mark H3K36me3 and CG-rich DNA (Musselman et al, 2012; Li et al, 2017) (see: ‘Introduction’). Interestingly, despite displaying sequence and functional homology, evidence suggests that the three proteins play divergent roles during normal development (Li et al, 2011; Brien et al, 2015; Kloet et al, 2016), indicating that changes to their relative association with PRC2 could substantially alters its function. AEBP2 is a zinc finger-containing protein which stimulates the catalytic activity of PRC2 via its nucleosome binding ability (Lee et al, 2018a) (see: ‘Introduction’). Interestingly, AEBP2 binds SUZ12 in a manner mutually exclusive to PHF1 (Youmans et al, 2018), demonstrating that AEBP2 directly competes with PHF1, PHF19 and MTF2 to be part of the PRC2. Together, I concluded that transformation induces a shift in the ratios of functionally distinct PRC2 sub-complexes present within a cell.

137

Figure 45. Transformation induces a shift in PRC2 composition A-B. Quantification by mass-spectrometry of proteins isolated by co-IP of endogenous EZH2 in mixes of SILAC labelled cell lysates from untransformed (A) or untransformed and transformed/tumour-derived (B) cells. Ratio values for all proteins are normalised to the ratio of EZH2 to control for differences in EZH2 protein levels between cell types. Untransformedheavy: untransformed cells with a ‘heavy’ amino acid labelled proteome; Untransformedlight: untransformed cells with a ‘light’ amino acid labelled proteome. In (B) notice the positions of MTF2, AEBP2, PHF1 and PHF19. C. Summary of changes to PRC2 composition induced by transformation. Values represent mean + SEM from one co-IP of untransformed vs transformed cells and two co-IPs of untransformed vs tumour-derived cells.

138

4.1.3 Gene expression changes may mediate a transformation-induced shift in PRC2 composition

Having identified a transformation-induced shift in PRC2 composition, I reasoned this change could be attributable to either alteration of the transcription or translation of PHF1, PHF19, MTF2 and AEBP2. Interrogation of mRNA expression data from untransformed and transformed cells identified expression changes of PRC2 components concordant with alterations to PRC2 composition (Fig. 46). Specifically, expression of MTF2 and AEBP2 was upregulated upon transformation, whilst PHF19 and PHF1 were repressed (Fig. 46). The magnitude of differential PHF1, MTF2 and AEBP2 expression mirrored the degree of change in their association with PRC2, indicating that transformation-induced gene expression changes could be leading to a shift in PRC2 composition. However, the expression of PHF19 decreased by only 17% upon transformation, a change insufficient to explain its ~2-fold decrease in PRC2 binding (Fig. 45). Nonetheless, this disparity could potentially be explained by increased competition for PRC2 binding by the greater levels of MTF2 and AEBP2. Accurate quantification of protein levels for these PRC2 components was not possible due to the poor quality of existing antibodies, hindering western-blotting, and the low protein levels of sub-stoichiometric components preventing the use of proteomic approaches. As a result, this precluded confirmation that differential gene expression led to matched alterations in protein levels. Together, I conclude that changes in the expression of sub- stoichiometric components may contribute to the PRC2’s compositional shift upon transformation.

139

Figure 46. Changes in sub-stoichiometric component expression mirror the shift in PRC2 composition Expression of sub-stoichiometric PRC2 components in the indicated cellular states, as detected by RNA-seq. Expression values represent mean + SEM from three biological replicates.

140

4.2 Sub-stoichiometric PRC2 components have distinct roles in tumours

4.2.1 Transformation-induced changes to sub-stoichiometric component binding match their functions in tumours

Alterations to epigenetic regulator sub-complexes are important events in promoting malignancies. This led me to hypothesise that the transformation-induced shift in PRC2 composition could be similarly important. Hence, I set out to deplete specific PRC2 components in transformed cells, and observe whether the effect on tumour growth matched transformation-induced changes to PRC2 composition. To do this, I transduced constructs into transformed cells which allowed inducible expression of shRNAs targeting PHF1, PHF19, MTF2 and AEBP2. Induction of hairpin expression led to a >50% depletion of mRNA levels for all PRC2 components, a level I deemed sufficient to investigate the function of PRC2 sub-stoichiometric components (Fig. 47).

Figure 47. PRC2 components are depleted by shRNAs in transformed cells RT-qPCR showing the expression levels of AEBP2, MTF2, PHF1 and PHF19 in transformed cells ±shRNA knock-down for eight days. Values represent mean +SEM from three technical replicates.

To assess the effect of depleting sub-stoichiometric components, I transplanted transformed cells with or without knock-down into immunocompromised mice and monitored the growth of resulting tumours. Strikingly, the effect of depleting a specific component mirrored whether its binding to PRC2 had increased or decreased upon transformation (Fig. 48A-B). Specifically, knock-down of MTF2 and AEBP2, which were enriched among EZH2-intactors in transformed cells, impaired tumour growth relative to

141

an uninduced control, whilst knock-down of PHF19, which was instead depleted, enhanced it. PHF1 knock-down, although not reaching significance, exhibited a strong trend towards increasing tumour growth, in line with its depletion among EZH2- interactors in transformed cells (Fig. 48A-B). Together, this indicates that a transformation-induced shift from PHF1/PHF19-containing to AEBP2/MTF2-containing PRC2 complexes may promote the growth of tumours induced by de novo transformed cells.

Figure 48. PRC2 sub-stoichiometric components have divergent effects on tumour growth

142

A. Transplantation assays comparing the growth kinetics of transformed cell-induced tumours ±shRNA knock-down of the indicated PRC2 component. Knock-down of PRC2 components was pre-induced for eight days prior to transplantation into immunocompromised mice. Time 0 represents the day at which tumours were first visible. Outlier tumours were excluded from the analysis and are not displayed (Grubbs test performed on final tumour volumes, tumours excluded if alpha <0.05). Values represent mean ±SEM for six (AEBP2 both conditions, MTF2 both conditions, PHF19 induced and PHF1 uninduced) or five (PHF19 uninduced, PHF1 induced) biological replicates. One asterisk indicates p-value <0.05, two asterisks indicate p-value <0.01 (one-tailed unpaired Student’s t-test). B. Final tumour volumes from two independent transplantation experiments performed as in (A). Outlier tumours excluded as in (A). Black line represents the median value for each group of tumours. Two asterisks indicate p-value <0.01 (one-tailed unpaired Student’s t-test).

4.2.2 Sub-stoichiometric PRC2 components have context-dependent functions in cancer

Having identified a tumour-promoting shift in PRC2 composition within de novo transformed cells, I wanted to explore whether the differential role of sub-stoichiometric proteins could also be observed in patient-derived samples. To do this, I decided to focus on MTF2 and PHF19 as they exhibited the strongest transformation-induced changes in EZH2 binding, and their depletion induced significant (p <0.01), but opposing, effects on tumour growth. To investigate the function of these proteins across cancer types, I employed cell lines derived from gastric and pancreatic PDXs with similar genetic backgrounds to the de novo transformed cells (p53-/-, expressing oncogenic RAS), and expressing sub-stoichiometric components (Fig. 49A). I specifically selected gastric and pancreatic PDX cell lines as these cancers originate from different cell and tissue types, but are driven by similar genetic events to each other and the de novo transformed cells (Kamisawa et al, 2016; Van Cutsem et al, 2016), allowing me to dissect whether transformative events or cell-of-origin determines the function of sub-stoichiometric components in tumours. Into these cell lines, I transduced constructs allowing inducible expression of PHF19 or MTF2-targeting shRNAs (shRNAs as used for transformed cells); induction of these hairpins led to a >60% knock-down of MTF2 and PHF19 in both cell lines (Fig. 49B), a magnitude similar to that seen in transformed cells (Fig. 47).

143

Figure 49. shRNA knock-down depletes PRC2 components in PDX cell lines A. RT-qPCR quantifying the expression of sub-stoichiometric components AEBP2, MTF2, PHF1 and PHF19 in two PDX cell lines GXA 3067 (gastric cancer) and PAXF 1998 (pancreatic cancer) relative to transformed cells. Values represent mean ±SEM from two technical replicates. B. RT- qPCR quantifying the expression levels of MTF2 and PHF19 in PDX cell lines GXA 3067 and PAXF 1998 ±shRNA knock-down for eight days. Values represent mean +SEM from three technical replicates.

To interrogate the effect of depleting PHF19 and MTF2, I performed transplantation assays with the PDX cell lines in a similar manner to transformed cells (Fig. 50). Interestingly, depletion of MTF2 and PHF19 exhibited diverse effects on the growth of PDX-derived tumours. In pancreatic xenograft tumours, depletion of MTF2 or PHF19 did not affect tumour growth, indicating that the tumour was either agnostic to the function of these proteins or the extent of shRNA knock-down was insufficient (Fig. 50A). In contrast, depletion of PHF19 in gastric xenograft tumours severely impaired growth, whilst MTF2 depletion once again had no effect (Fig. 50A). The divergent responses to

144

PHF19 depletion in tumours generated from transformed, gastric and pancreatic cells, in spite of their similar PHF19 expression levels (Fig. 49A) and extent of knock-down (Fig. 47/49B), suggest that PHF19 could play a context-dependent role in cancer. To further explore the context dependence of PHF19 activity, I performed additional transplantation assays using two PDX cell lines derived from non-small-cell lung cancer. Strikingly, lung xenograft tumours derived from each PDX cell line showed differential responses to depletion of PHF19, with one (LXFA 677) unaffected and the other (LXFL 1674) exhibiting impaired growth (Fig. 50B). This indicated that the response to PHF19 depletion varies within, as well as between, cancer types. These divergent effects on tumour growth could be attributable to cell/tissue-specific off-target effects of shRNA knock-down, a possibility that would need to be excluded through the use of additional shRNAs. Alternatively, the specific set of genes repressed by the PHF19-containing complex may differ between cell and tissue types, leading to distinct functional effects of derepressing the PHF19 repressed gene set. Altogether, as a result of confounding experimental factors, it was not possible to conclude whether the tumour function of MTF2 and PHF19 identified in de novo transformed cells was conserved in patient samples and across different cancer types.

145

Figure 50. PRC2 sub-stoichiometric components have context-dependent roles in cancer A-B. Transplantation assays comparing the growth kinetics of PDX cell line-induced tumours ±shRNA knock-down of MTF2 or PHF19. Knock-down of PRC2 components was pre-induced for eight days prior to transplantation into immunocompromised mice. Time 0 represents the day at which tumours were first visible. Outlier tumours were excluded from the analysis and are not displayed (Grubbs test performed on final tumour volumes, tumours excluded if alpha <0.05). Values represent mean ±SEM for six (PAXF 1998 PHF19/MTF2 uninduced), four (GXA 3067 all conditions, PAXF 1998 PHF19/MTF2 induced, LXFL 1674 PHF19 induced and LXFA 677 PHF19 induced), three (LXFL 1674 PHF19 uninduced) or two (LXFA 677 PHF19 uninduced) biological replicates. PDX model cancer type: PAXF 1998 – pancreatic, GXA 3067 – gastric, LXFL 1674 – non-small-cell lung, LXFA 677 – non-small-cell lung. One asterisk indicates p-value <0.05, three asterisks indicate p-value <0.001 (one-tailed unpaired Student’s t-test performed on final tumour volumes).

146

4.3 Changes to PRC2 composition do not mediate a tumour- promoting redistribution of EZH2

Transformation induces redistribution of EZH2 binding sites, which in turn causes transcriptional changes essential for tumour formation (see previous chapter). Thus, given that PHF1, PHF19, MTF2 and AEBP2 are responsible for targeting PRC2 to different chromatin locations, I hypothesised that the transformation-induced shift in PRC2 composition is driving EZH2 redistribution. This model predicts that depletion of the sub-stoichiometric components responsible for targeting EZH2 to new loci upon transformation (i.e. AEBP2 and MTF2) would specifically lead to misregulation of the relevant target genes. As a result, I interrogated how the expression of genes repressed by EZH2 specifically in transformed cells (Fig. 25B), were affected by depletion of different sub-stoichiometric components (Fig. 51). Specifically, I selected the genes EMX2, LBH and PELI2 to investigate, as EMX2 repression promoted tumour growth in GBM (see previous chapter), and LBH and PELI2 exhibited particularly strong changes in EZH2 binding and concomitant repression. For all genes investigated, depletion of sub-stoichiometric components led to upregulation of gene expression, in line with these genes being PRC2 targets in transformed cells (Fig. 51). However, in all instances upregulation of the same gene was induced by depletion of sub-stoichiometric components with opposing functions in tumours (Fig. 51). In particular, EMX2 and LBH were upregulated by depletion of AEBP2, MTF2, PHF1 and PHF19, whilst MTF2 and PHF1 depletion also induced a slight upregulation of PELI2 (Fig. 51). This indicated that repression of genes silenced by EZH2 only in transformed cells is not mediated solely by PRC2 sub-complexes with tumour-promoting functions. As all genes repressed by EZH2 specifically in transformed cells were not tested, I cannot exclude the possibility that other genes may be repressed in a PRC2 sub-complex-specific manner. Together, I concluded that the transformation-induced switch in PRC2 composition was unlikely to maintain the tumour-promoting redistribution of EZH2 I had previously identified, indicating that the switch in PRC2 composition promotes tumour formation by an independent mechanism.

147

Figure 51. A transformation-induced shift in PRC2 composition does not drive EZH2 redistribution RT-qPCR showing the expression levels EMX2, LBH and PELI2 in transformed cells ±shRNA knock-down of AEBP2, MTF2. PHF1 and PHF19 for 17 days. Values represent mean + SEM from three technical replicates.

148

Chapter 5. Systematic comparison of experimental approaches for defining PRC2-regulated genes

Introduction

The genes controlled by a chromatin regulator defines its role in a cell, making experimental determination of these gene sets essential for understanding regulator function. The target genes of chromatin regulators are often identified through profiling their chromatin binding sites (e.g. ChIP-seq) and the transcriptional changes induced by suppressing regulator function through genetic or pharmacological loss-of-function (LOF) approaches (e.g. gene knock-out followed by RNA-seq). When disrupting a chromatin regulator, the choice of LOF approach is often driven by the constraints of a specific experiment; however, in many instances approaches are used interchangeably, with selection instead being determined by reagent availability and investigator preference. Yet this interchangeable usage discounts the distinct transcriptional and phenotypic effects of each LOF approach (Housden et al, 2017). In particular, partial suppression (pharmacological inhibition, mRNA knock-down) or full ablation (genetic knock-out) of protein function can have differing transcriptional effects. Moreover, pharmacological inhibition disrupts only the catalytic activity of regulators, whilst genetic LOF approaches also abrogate non-catalytic functions. To date, few controlled comparisons have been performed comparing the effectiveness of different LOF approaches in identifying chromatin regulator target genes, precluding rational selection of the most appropriate method.

Through its gene repressive activity, PRC2 acts as an essential regulator of both normal and pathological cellular states (Di Croce & Helin, 2013; Kim & Roberts, 2016). Hence, there has been particular interest in accurately determining the specific genes regulated by PRC2 to understand the mechanisms underlying its diverse functions. Previously, I generated transcriptional data sets profiling the effect of EZH2 pharmacological inhibition on cells at each stage of de novo transformation, allowing me to determine changes to the PRC2-regulated gene set upon transformation (see: ‘Chapter 3’). However, pharmacological inhibition disrupts only the catalytic activity of EZH2, leaving the protein intact - a potentially important factor given the known non-catalytic functions of EZH2 (Kim et al, 2015, 2018). Thus, I was interested to understand whether disrupting EZH2

149

protein levels as well as catalytic activity would change the number and type of PRC2- regulated genes identified. To do this, I set out to perform a controlled comparison between the genes deregulated by pharmacological EZH2 inhibition and genetic EZH2 LOF in the same cell type (Fig. 52). In doing so, I hoped to define differences in the PRC2-regulated gene sets identified by each approach, and determine the origin of any variability.

Figure 52. Loss-of-function approaches for disrupting PRC2 function Schematic representation of the genetic and pharmacological loss-of-function approaches used to disrupt the function of PRC2 to identify its target genes. Red strikethrough denotes abrogation of the indicated process. A red cross denotes inhibition of gene expression. RISC: RNA-induced silencing complex, LOF: loss-of-function.

150

Aims

In this chapter, I describe experiments in which I systematically compared different LOF approaches for determining PRC2-regulated genes. To do this, I first disrupted PRC2 function in de novo transformed cells by targeting EZH2 using genetic (knock-down and knock-out) and pharmacological LOF approaches. Subsequently, I determined the transcriptional effects of EZH2 LOF by RNA-seq, and then compared the putative PRC2 regulated genes identified with each approach. By then integrating the RNA-seq data sets with H3K27me3 and EZH2 ChIP-seq, I compared the number and properties of direct PRC2 target genes identified by each approach. Together, I show that pharmacological inhibition is the most effective LOF approach for defining PRC2- regulated genes.

151

Results

5.1 Disruption of PRC2 function by EZH2 knock-down, knock-

out and pharmacological inhibition

Previously, I compared by RNA-seq the gene expression changes induced through treating cells at each stage of de novo transformation with the EZH2 pharmacological inhibitor EPZ-6438 (EZH2i). By integrating this differential expression analysis with EZH2 and H3K27me3 ChIP-seq, I was able to identify changes to the PRC2-regulated gene set upon transformation (see: ‘Chapter 3’). To compare how this EZH2i-derived gene set differed from those identified by genetic LOF approaches, I performed mRNA knock- down (KD) and gene knock-out (KO) of EZH2 in transformed cells.

To disrupt EZH2 by genetic KO, I initially generated three clonal EZH2 KO cell lines through CRISPR-Cas9 genome editing of transformed cells. However, analysis by RT- qPCR of EMX2 expression, a PRC2-repressed gene in transformed cells (see: Chapter 3), revealed striking differences in the extent of EMX2 upregulation in each EZH2 KO cell line (Fig. 53A). This indicated that each clone was having a distinct transcriptional response to EZH2 KO, an effect that would likely confound accurate classification of genes misregulated by EZH2 KO. Instead, I decided to perform polyclonal KO of EZH2 to surmount this issue. To do this, I transduced an EZH2-targeting sgRNA into transformed cells which inducibly expressed Cas9 nuclease (see: ‘Materials & Methods’). Induction of Cas9 expression led to a gene editing efficiency of 75% (Fig. 53B), which in turn induced EZH2 KO and a substantial depletion of both EZH2 and H3K27me3 protein levels (Fig. 53D). As expected given the polyclonal nature of this approach, both unedited DNA and residual EZH2/H3K27me3 was observed (Fig. 53B&D). However, the 75% gene editing efficiency and 85% decrease in H3K27me3 levels, indicates that the majority of cells have homozygous EZH2 LOF, meaning transcriptional changes induced by complete EZH2 depletion will still be observed (Fig. 53B&D). As a control for gene expression changes induced by EZH2 KO, I also transduced Cas9-inducible transformed cells with a control sgRNA that targets an eGFP gene in the same plasmid, thus controlling for transcriptional effects arising from Cas9 expression and DNA cutting. To provide additional stringency, I also used transformed cells expressing only Cas9 as further control cell line to account for the effects of eGFP overexpression in the first control.

152

To disrupt EZH2 by mRNA KD, I transduced transformed cells with a construct allowing the inducible expression of an EZH2-targeting shRNA. I chose to perform inducible as opposed to constitutive shRNA KD as this enabled me to identify the effects of EZH2 KD using the same shRNA-transduced cell line as a control, overcoming confounding gene expression differences introduced by transducing the shRNA expressing construct. Induction of hairpin expression led to an 85% decrease in levels of EZH2 mRNA (Fig. 53C), and a substantial depletion of EZH2 and H3K27me3 protein levels (Fig. 53D). Together, I concluded that genetic and pharmacological approaches disrupted PRC2 function in transformed cells to a comparable extent.

To compare the PRC2-regulated gene sets identified by EZH2i and genetic EZH2 LOF approaches, I generated gene expression data sets by RNA-seq for the EZH2 KO/KD cell lines and their relevant control cell lines, complementing those I had already generated for EZH2i. All sequencing runs had a high per-base sequencing quality, low ribosomal RNA contamination, acceptable levels of read duplication and yielded >20 million reads. Thus, these high quality gene expression data sets provided me with a robust start point to interrogate the relative efficacy of each LOF approach.

153

Figure 53. PRC2 function is disrupted by genetic and pharmacological LOF approaches A. Relative expression of EMX2 in three different clonal EZH2 KO cell lines as determined by RT- qPCR. Expression values represent mean ±SEM from three technical replicates. WT: wild-type. B. Screenshot from TIDE analysis (see: ‘Materials & Methods’) of gene editing by CRISPR-Cas9 at the EZH2 locus. For EZH2 KO, expression of an EZH2-targeting sgRNA was induced in cells expressing Cas9 for 21 days to induce LOF mutations. The graph indicates the proportion of CRIPSR-Cas9 edited DNA molecules in the cell population relative to an uncut control. C. Relative expression of EZH2 in transformed cells ±expression of an EZH2-targeting shRNA for seven days, as detected by RNA-seq. Expression values represent mean ±SEM from three biological replicates. NT: not treated. D. Western-blot quantification of EZH2 and H3K27me3 levels in transformed cells ±EZH2 disruption by pharmacological inhibition (left), genetic KO (middle) and mRNA KD (right). For pharmacological inhibition, cells were treated with an EZH2 inhibitor (EZH2i) or DMSO as control for 12 days. EZH2 KO and KD were performed as in (B) and (C), respectively. Wild-type indicates cells only expressing Cas9. Intensity values for the H3K27me3 band are displayed relative to the control condition. Histone H3 is used as a loading control. NT: not treated.

154

5.2 Disrupting PRC2 with different loss-of-function approaches has distinct effects on gene expression

With the expression datasets generated from each LOF approach, I then performed differential expression analysis to identify genes which were misregulated by EZH2 disruption. To capture all differential expression, and to overcome the different extents of gene up/down-regulation each approach may induce, I classified genes as differentially expressed if they exhibited any expression change with a FDR ≤0.01. In transformed cells, LOF approaches induced differential expression of ~3500-5000 genes, with EZH2 KD inducing the most extensive changes and EZH2 KO the least (Fig. 54A). As expected, due to the incomplete inhibition of protein function within the cell population, expression changes were generally low-magnitude (Fig. 54B). Downregulated genes showed a median expression decrease <1.4-fold in all instances, indicating that derepressing PRC2 targets has limited secondary effects on gene expression irrespective of the LOF approach used (Fig. 54B). The magnitude of gene upregulation was generally greater than down-regulation, as expected given the canonical repressive function of EZH2, but varied significantly (p <0.0001) between the LOF approaches. Specifically, EZH2i induced a median upregulation ~2x greater than EZH2 KO or KD, suggesting that it is a more potent means for derepressing PRC2 targets (Fig. 54B).

Figure 54. EZH2 LOF induces large numbers of small-magnitude changes in gene expression A. Quantification of genes differentially expressed (FDR ≤0.01) upon EZH2i, EZH2 KO and EZH2 KD in transformed cells, as detected by RNA-seq. B. Expression change of up and downregulated genes indicated in (A). Four asterisks indicate p-value <0.0001 (One-way ANOVA followed by Tukey’s multiple comparisons test).

155

Comparison of the DEGs induced by each LOF approach revealed a surprisingly small overlap between gene sets (Fig. 55A). In particular, only ~5% of upregulated and ~0.5% of downregulated genes were shared between all three approaches, suggesting that most were either PRC2-regulated genes misregulated by a specific LOF approach(es) or PRC2-independent off-target effects. To attempt to differentiate between these possibilities, I utilised ChIP-seq data from transformed cells (see: ‘Chapter 3’) to compare the enrichment of H3K27me3 at upregulated gene promoters. Comparison of H3K27me3 enrichment at genes upregulated by one or multiple LOF approaches, showed that genes uniquely upregulated by KD or KO had negligible H3K27me3 enrichment (Fig. 55B). This indicates that differential expression of these genes is either a secondary effect of PRC2 disruption or, more likely, an off-target effect of the LOF approach. In contrast, genes uniquely identified by EZH2i showed a strong enrichment of H3K27me3, with average H3K27me3 levels equivalent to high-confidence PRC2 target genes identified by all three LOF approaches, suggesting that EZH2i uniquely identifies many direct PRC2 targets (Fig. 55B). Together, I conclude that disrupting PRC2 function by mRNA KD, gene KO and pharmacological inhibition has strikingly distinct effects on gene expression, indicating that these approaches cannot be considered equivalent methods for probing the function of chromatin regulators.

156

Figure 55. EZH2 LOF by different approaches induces distinct expression changes A. Venn diagrams showing the overlap between genes upregulated (left) or downregulated (right) (FDR ≤0.01) by EZH2i, EZH2 KO and EZH2 KD in transformed cells. B. Average density profile of H3K27me3 at genes upregulated by one or multiple EZH2 LOF approaches. Light coloured areas represent the SEM of the average profile.

157

5.3 Pharmacological inhibition is the most effective LOF approach for identifying PRC2 target genes

Having identified a striking difference in average H3K27me3 enrichment at genes upregulated by each LOF approach, I decided to specifically characterise the PRC2 target genes each method identifies. For each LOF approach, I classified genes as PRC2 targets if they were both upregulated and had a promoter peak of H3K27me3 (within 5kb of the TSS). In agreement with differences observed in average H3K27me3 enrichment across all upregulated genes (Fig. 55B), ~40% of all genes upregulated by EZH2i were PRC2 targets, contrasting with ~20% and ~10% for EZH2 KO and KD, respectively (Fig. 56A). Furthermore, whilst EZH2i upregulated ~20% of all genes with H3K27me3 enrichment their promoter, EZH2 KO and KD only upregulated ~10% (Fig. 56B). Thus, there are striking differences between the sets of genes identified as PRC2-regulated when each LOF approach is used. It is important to note that this analysis may underestimate the fraction of upregulated genes which are direct PRC2 targets, as more distal PRC2 binding could also be mediating gene repression.

Comparison of PRC2 target genes identified by each LOF approach revealed that ~50% of genes identified using EZH2i were not detected by any other approach (Fig. 56C). In addition, the majority of PRC2 target genes detected using EZH2 KO or KD were also identified by EZH2i (KO - 77%, KD - 65%), and of the few PRC2 target genes which were uniquely identified by genetic LOF (KO -100, KD - 129), only 12 genes were commonly identified by both KO and KD (Fig. 56C). This indicates that depleting EZH2 protein through genetic LOF approaches upregulates relatively few additional PRC2 target genes compared to inhibiting catalytic activity alone. Together, I conclude that pharmacological inhibition is more effective at derepressing PRC2 target genes than genetic LOF approaches, making it the most efficacious choice for determining PRC2- regulated gene sets.

158

Figure 56. EZH2i derepresses many more PRC2 target genes than genetic LOF A. Percentage of genes upregulated by EZH2i, EZH2 KO and EZH2 KD in transformed cells which are PRC2 targets (H3K27me3 peak ±5kb TSS). Numbers above bars indicate PRC2 target genes derepressed by each approach. B. Percentage of genes with H3K27me3 enrichment at their promoter which are upregulated by the different EZH2 LOF approaches. C. Venn diagram showing the overlap of PRC2 target genes identified by each approach. D. Expression of EZH1, as detected by RNA-seq, in transformed cells with EZH2 LOF by either pharmacological inhibition, mRNA KD or gene KO. For EZH2i, ‘Control’ indicates DMSO treated cells. For EZH2 KO, ‘Control’ indicates cells expressing only Cas9. For EZH2 KD, ‘Control’ indicates cells with the shRNA uninduced. Values represent mean ±SEM from three biological replicates. TPM: transcripts per million.

159

To rationalise the smaller number of PRC2 target genes identified by EZH2 KD and KO, I hypothesised three possible explanations: i) depletion of EZH2 protein or mRNA triggers downstream compensatory mechanisms that maintain the repression of many PRC2 target genes upon EZH2 KO and KD; ii) EZH2 wild-type cells remaining after induction of CRISPR-Cas9 EZH2 KO mask expression changes occurring in EZH2 KO cells; iii) H3K27me3 depletion by EZH2 KD is insufficient to drive de-repression of high- magnitude PRC2 target genes. Supporting my first hypothesis, EZH2 homologue EZH1 has been shown in certain biological contexts to act redundantly with EZH2 (Ezhkova et al, 2011; Lui et al, 2016; Shen et al, 2008). However, EZH1 expression was unaffected by EZH2 LOF in transformed cells, indicating that it is unlikely EZH1 is compensating for EZH2 loss (Fig. 56D). Nonetheless, a change in EZH1 protein levels or other H3K27me3- independent mechanisms could be compensating for EZH2 loss, possibilities that will require further investigation. In line with the second hypothesis, western-blot and sequencing analysis confirm that EZH2 wild-type cells remain after EZH2 KO is induced, suggesting this is also a viable explanation for the fewer genes identified by EZH2 KO than EZH2i (Fig. 53B&D).

To explore the possibility that EZH2 KD was insufficient to derepress high-magnitude PRC2 target genes, I compared the enrichment of EZH2 and H3K27me3 at the promoters of PRC2 target genes uniquely identified by EZH2i or commonly by both EZH2i and EZH2 KD (genes indicated in Fig. 56C). Supporting my hypothesis, EZH2i- unique PRC2 target genes had a higher average enrichment of EZH2 and H3K27me3 than common genes (Fig. 57A). Furthermore, inspection of EZH2-unique PRC2 target genes revealed large domains of strong EZH2 and H3K27me3 enrichment at their promoters (Fig. 57C). As expected, no such difference was seen when comparing EZH2i unique and EZH2i/EZH2 KO common PRC2 target genes, in line with the complete depletion of H3K27me3 that EZH2 KO induces (Fig. 57B). Together, this suggests that residual H3K27me3 remaining after EZH2 KD decreases the number of direct PRC2 targets that are derepressed by this approach, restricting its utility for defining the PRC2- regulated gene set.

160

Figure 57. EZH2 KD is insufficient to derepress high-magnitude PRC2 target genes A-B. Average density profile of H3K27me3 and EZH2 at PRC2 target genes upregulated by EZH2i alone or multiple EZH2 LOF approaches. Light coloured areas represent the SEM of the average profile. C. ChIP-seq signal for EZH2 and H3K27me3 at genes derepressed by EZH2i but not EZH2 KD. ChIP-seq signal normalised to sequencing depth is shown.

161

To investigate how the LOF approach used affected the magnitude of PRC2 target de- repression, I analysed the extent of gene upregulation at PRC2 targets commonly identified by LOF approaches. In line with the failure of EZH2 KD to derepress high- magnitude PRC2 targets (Fig. 57A), ~80% of PRC2 target genes commonly identified by EZH2i and EZH2 KD were upregulated to a greater extent by EZH2i (Fig. 58). A similar, but less striking, pattern was also observed for EZH2 KO relative to EZH2i (Fig. 58). In this instance, the lesser upregulation by EZH2 KO is likely due to the remaining EZH2 wild-type cells in the polyclonal EZH2 KO population which would lessen the observed transcriptional effects of EZH2 loss. Thus, as a result of its ability to drive almost total depletion of H3K27me3 and its homogeneous repressive action across the entire cell population, EZH2i represents the most effective means for identifying PRC2- regulated genes.

162

Figure 58. Commonly identified PRC2 target genes are upregulated to a greater extent by EZH2i Scatter plots comparing the degree of upregulation of PRC2 target genes commonly derepressed by EZH2 KD and EZH2i (left), EZH2 KO and EZH2i (right), EZH2 KO and EZH2 KD (bottom). Each black dot represents a gene. Grey dotted lines are included to indicate the point of equivalent gene upregulation between LOF approaches.

163

Chapter 6. Discussion

6.1 EZH2 is hijacked by neoplastic transformation through redistribution on chromatin

As our understanding of cancer epigenetics develops, it is increasingly apparent that mutagenic alterations are not pre-requisite for epigenetic regulators to attain a central function in tumour biology (Wainwright & Scaffidi, 2017). Epigenetic regulators in their wild-type unmutated form are recurrently co-opted by cancer cells to facilitate unlimited proliferation, differentiation blockades, and adaptation to internal and external stresses (Sharma et al, 2010; Zingg et al, 2018; Shi et al, 2013; Wainwright & Scaffidi, 2017). However, the tumour promoting role of wild-type epigenetic regulators juxtaposes with their normal physiological function in mediating correct development and tissue maintenance. This apparent contradiction is epitomised by the PRC2 catalytic component EZH2, which in its wild-type form acts both as a crucial regulator of physiological processes and a potent tumour promoter in a variety of malignancies (Schuettengruber et al, 2017; Kim & Roberts, 2016). In this study, I investigate the role of EZH2 in the normal and malignant CNS as a paradigm for how epigenetic regulators can transition between these apparently contradictory roles upon neoplastic transformation. By performing an integrated multi-omics analysis, I dissect the changing function of EZH2 in a model of de novo transformation, and find that transformation induces a striking redistribution of EZH2 on chromatin. This redistribution leads to tumour promoting expression changes at key neural homeotic genes, enabling a switch from EZH2’s physiological to pathological function (Fig. 59). Importantly, although I focus on the role of EZH2 in the CNS, these findings suggest a generalisable mechanism by which other wild-type epigenetic regulators could attain tumour promoting functions.

The tumour promoting redistribution of EZH2 I observe has precedent in PRC2’s mode of action in normal physiology. In both development and tissue maintenance, gene expression changes critical for correct cellular differentiation are often driven by redistribution of PRC2 and H3K27me3 (Sher et al, 2012; Ezhkova et al, 2009; Jadhav et al, 2016). Furthermore, in the absence of core PRC2 components ESCs lose the ability to differentiate in a correct manner, emphasising the physiological importance of PRC2 in driving cell state transitions (Pasini et al, 2007; Chamberlain et al, 2008). Thus, a

164

parallelism exists between the mechanism through which PRC2 mediates differentiation in normal cells and establishes the neoplastic state in cancer cells.

Figure 59. Model of EZH2 co-option by neoplastic transformation to drive glioblastoma growth

6.1.1 EZH2 overexpression upon transformation does not drive its tumour promoting role

Wild-type EZH2’s tumour promoting ability has often been attributed to its increased expression levels in cancer cells relative to their normal untransformed counterparts (Kim & Roberts, 2016). Supporting a direct link between overexpression and tumour promotion, a strong anti-correlation often exists between EZH2 expression levels and patient prognosis (Varambally et al, 2002; Kleer et al, 2003). Moreover, overexpression or hyperactivation of EZH2 can promote cell proliferation and tumour growth (Gonzalez et al, 2014; Zingg et al, 2018; Souroullas et al, 2016). However, EZH2 expression and cellular proliferation rate are intimately linked via the E2F pathway, which enables

165

homeostatic maintenance of H3K27me3 levels (Bracken et al, 2003). This indicates that upregulation of EZH2 in cancer cells may in fact be a hard-wired response to cellular hyperproliferation with no intrinsic tumour promoting role (Wassef et al, 2016; Bracken et al, 2003). In line with this, I observed that although EZH2 expression increases upon de novo transformation, there is no accompanying change in the levels of H3K27me3. Instead, H3K27me3 and EZH2 are redistributed across the genome by transformation leading to functionally important expression changes at key homeotic genes (Fig. 59).

This redistribution-based mechanism provides an explanation for how a functional dependency on EZH2 can arise without changes in total H3K27me3 levels. Given that many studies purporting to show that EZH2 overexpression is a tumour promoting event have not reported accompanying increases of H3K27me3 (Kleer et al, 2003; Varambally et al, 2002; Zingg et al, 2018), it is possible that transformation-induced redistribution could also underlie EZH2’s tumour promoting role in those instances. For example, in prostate cancer EZH2 overexpression is concurrent with de novo PRC2-mediated repression of ADRB2, an essential regulator of β-adrenergic signalling (Yu et al, 2007a). Although repression of ADRB2 has been attributed to EZH2 overexpression, no global increase in H3K27me3 levels has been shown to accompany it. Thus, transformation- induced redistribution could be an alternative mechanism for ADRB2 repression to occur. As a result, when assigning a tumour promoting role to EZH2 overexpression, it is important to also assay the levels of H3K27me3 to understand whether the EZH2 increase observed is a homeostatic response or has the potential to enhance levels of gene repression.

The effects of EZH2 overexpression contrasts with EZH2 GOF mutations, which induce a substantial increase in H3K27me3 levels that drives tumour formation, demonstrating an oncogenic role for enhanced H3K27me3 (Souroullas et al, 2016). Furthermore, whilst GOF mutations can induce melanoma and lymphoma with high penetrance in mice, EZH2 overexpression has either modest or neutral effects on tumour growth, emphasising the lack of equivalence between these two modes of EZH2 disruption. (Koppens et al, 2017; Souroullas et al, 2016).

166

6.1.2 What drives EZH2 redistribution upon transformation?

The exact drivers of PRC2 targeting in physiological conditions remain to be fully determined. Thus, identifying the specific molecular mechanisms that drive EZH2 to change its binding profile upon transformation is challenging. Nonetheless, likely to be of central importance is the general responsiveness of PRC2 to changes in a genes transcriptional state, as demonstrated by the recruitment of PRC2 to CpG islands genome-wide upon global inhibition of transcription (Riising et al, 2014). Given that transformation induces substantial transcriptional changes, PRC2 will likely respond by dissociating from promoter CpG islands of derepressed genes and binding those of silenced genes. In doing so, EZH2 would deposit H3K27me3 at newly repressed genes, acting to stabilise and maintain the neoplastic gene expression programme. Thus, the specific redistribution that EZH2 undertakes upon transformation will in part be determined by the downstream transcriptional effects of the genetic events that drive cancer.

One of the most striking patterns in EZH2 redistribution I observed is a disproportional loss of EZH2 from CpG islands upon transformation. This suggests that transformation is altering the ability of PRC2 to bind CpG islands, either through altering island accessibility or modifying PRC2 affinity for these regions. Previously, others have observed that CpG islands marked by H3K27me3 in normal untransformed cells are recurrently targeted for DNA methylation upon transformation (Schlesinger et al, 2007; Widschwendter et al, 2007; Ohm et al, 2007). This is thought to represent a switch from a reversible mechanism of gene repression to one that gives permanent silencing, potentially locking a cell in a pro-tumourigenic stem cell state (Ohm et al, 2007). It is possible that the eviction of EZH2 from CpG islands which I observed could result from polycomb-mediated repression being supplanted by hypermethylation of CpG islands. Supporting this, CpG island methylation is a recurrent feature of glioblastoma (Malta et al, 2018). However, it remains to be determined whether a similar phenomenon occurs in the de novo transformation model I used in this study. Moreover, this mechanism would contradict the co-occurrence of DNA methylation and H3K27me3 I found at many EZH2 target genes in the same transformation model.

Another candidate mechanism for driving EZH2 redistribution is the change in PRC2 sub-unit composition which I found accompanies transformation. The accessory proteins

167

which change their association with PRC2 upon transformation (PHF19, PHF1, MTF2 and AEBP2) all have demonstrated roles in regulating PRC2 targeting to chromatin (Holoch & Margueron, 2017). Moreover, these proteins have distinct functions in physiological conditions, as indicated by the divergent phenotypes that arise from genetically disrupting them, (Holoch & Margueron, 2017; Vizán et al, 2015). However, to my knowledge, no evidence exists that these accessory proteins have deviating chromatin binding profiles or a differential affinity for CpG islands that could explain the loss of EZH2 from these regions. In addition, I found that accessory proteins with enhanced and depleted binding to PRC2 upon transformation are both required for EZH2-mediated repression of genes upon transformation, suggesting that the PRC2 composition switch does not drive binding to these new sites. Clarifying the extent to which a switch in PRC2 accessory proteins could be driving EZH2 redistribution will require ChIP-seq of each accessory protein in the same cell line to allow a controlled comparison of their genome-wide distribution.

6.1.3 EZH2 redistribution induces only a small number of gene expression changes

In profiling how transformation affects EZH2 distribution, I observed that although transformation alters EZH2 binding at thousands of sites across the genome, only a fraction of these changes demonstrably affect gene expression. Why does such extensive EZH2 redistribution have only a small effect on gene expression? One important factor is the extent to which transformation changes EZH2 binding at a given locus. By analysing the sites differentially bound by EZH2, I revealed that most have only small changes in EZH2 enrichment upon transformation. These low-magnitude changes in EZH2 binding are unlikely to be sufficient to mediate changes in gene expression, suggesting a possible reason for the disparity between changes in EZH2 binding and alterations to gene expression.

Functional redundancy is recurrently used by biological systems to make them robust against cell intrinsic or extrinsic perturbations. Hence, the existence of redundant repressive mechanisms could explain why EZH2 redistribution largely fails to induce gene expression changes. In line with this possibility, I observed that EZH2 target genes are often also DNA methylated. This suggests that at many genes, H3K27me3 and DNA methylation may cooperate to maintain repression. Supporting this, in instances where

168

genes are targeted by both repressive mechanisms, I observe that expression is generally fully silenced (Fig. 21A). This contrasts with genes repressed by H3K27me3 alone that retain a low basal level of transcription, indicating that the co-occurrence of these marks enhances gene expression.

This putative cooperative repression contrasts with the purported mutually antagonistic relationship between DNA methylation and EZH2’s HMTase activity (Rose & Klose, 2014). Previous studies have shown that depleting DNA methylation leads to deposition of H3K27me3 at hypomethylated loci (Lynch et al, 2012; Reddington et al, 2013), and PRC2 has a binding preference for CpG islands, largely unmethylated regions of DNA (Mendenhall et al, 2010; Ku et al, 2008). However, DNA methylation is not universally incompatible with the deposition of H3K27me3, as demonstrated by bisulphite sequencing of H3K27me3 ChIP DNA from cancer cells and studies of the inactive X chromosome (Statham et al, 2012; Heard & Martienssen, 2014). My findings support a co-operative relationship between EZH2 and DNA methylation at a portion of EZH2 target genes in cancer cells, and indicate a possible mechanism by which the effects of EZH2 redistribution on gene expression are mitigated. However, to formally demonstrate that the genes refractory to EZH2 binding changes are reliant upon DNA methylation to maintain their repression would require dual disruption of DNA methylase and EZH2 HMTase activity - a possible experiment for a continuation of this study.

Alongside redundant repressive mechanisms, the absence of appropriate transcription factors, or other transcriptional activators, may prevent gene expression changes from occurring. Supporting this, despite EZH2 binding being lost upon transformation from large domains encompassing many genes in all HOX clusters, only HOXB9 and HOXA11 are upregulated (Fig. 25B). This suggests that although loss of EZH2 makes chromatin competent for transcription, it is often not sufficient to drive transcription unless the gene-specific activating proteins are present.

6.1.4 Mapping cancer-relevant changes in H3K27me3 deposition

To unveil the mechanisms underlying the tumour promoting functions of PRC2, many other studies have also performed genome-wide characterisation of PRC2 target genes in normal and cancer cells (Ngollo et al, 2017; Rheinbay et al, 2013; Yu et al, 2007a). This approach was initially employed to investigate the tumour promoting role of EZH2

169

in prostate cancer (Yu et al, 2007a, 2007b). Here, the transcriptional effects of modulating EZH2 levels in normal prostate cells were compared to the promoters occupied by SUZ12 in prostate cancer cell lines, allowing the identification of ADRB2, an essential mediator of β-adrenergic signalling, as a functionally important PRC2 target in prostate cancer (Yu et al, 2007a). Since, similar comparisons between normal and cancer cells have been extensively performed, including in: breast cancer (Li et al, 2018b; Zuo et al, 2011), pancreatic cancer (Jia et al, 2013), colorectal cancer (Enroth et al, 2011) and GBM (Lin et al, 2015; Rheinbay et al, 2013). In particular, two independent studies in GBM, have compared the distribution of H3K27me3 in GBM stem cells and normal astrocytes. In both studies, changes in H3K27me3 distribution were observed at developmental transcription factor genes (Rheinbay et al, 2013; Lin et al, 2015). In particular, loss of H3K27me3 from ASCL1, a basic helix-loop-helix transcription factor, was important in driving upregulation ASCL1 which in turn promoted WNT signalling (Rheinbay et al, 2013). Importantly, neither study detected changes in H3K27me3 distribution at EMX2 or the HOX clusters, redistribution events detected by my approach and subsequently validated in a range of patient data.

Despite providing interesting insights into wild-type PRC2’s roles in cancer, many of these studies, including those in GBM, compared the distribution of H3K27me3 using non-isogenic cell lines or tumour and normal tissue samples sourced from different patients. These genetic differences introduce confounding noise into the experiments, lowering sensitivity and therefore decreasing the ability to detect functionally important changes in H3K27me3 distribution. Thus, this has prevented an accurate assessment of the changes that transformation is inducing to PRC2 binding and activity. In contrast to previous studies, I compared isogenic normal and malignant cells, allowing me to undertake a controlled characterisation of the changes transformation induces in EZH2 distribution and activity. This approach enabled me to identify a functionally important redistribution of EZH2 binding at neural homeotic genes, providing a unique insight into the mechanisms underlying the hijacking of wild-type EZH2 in GBM.

6.1.5 Transformation-induced redistribution: an explanation for EZH2’s dual role in cancer?

A remarkable feature of EZH2’s role in cancer is its ability to act both as a tumour suppressor, as evidenced by LOF mutations affecting EZH2 and other PRC2

170

components, and a tumour promoter (Kim & Roberts, 2016).This dichotomous function has been previously attributed to the tissue-specific roles of EZH2, and its differing responses to the genetic driver events initiating the disease (Comet et al, 2016). However, my results suggest an alternative cancer stage-specific model for the apparently antithetic roles that EZH2 plays in cancer. I propose that in normal untransformed cells, EZH2 acts as a tumour suppressor by acting in conjunction with other chromatin regulators to present an epigenetic barrier to neoplastic transformation. By repressing genes associated with alternative cellular lineages, such as HOX genes in astroglial cells, EZH2 ensures maintenance of the correct cellular phenotype (Fig. 59). This tumour suppressive role of EZH2 would be instructed by specific developmental or tissue maintenance signals that maintain the correct distribution of EZH2’s repressive activity. However, upon mutagenic disruption of EZH2 or other PRC2 components, this barrier to alternative cellular states would be relieved, enabling oncogenic genetic events to more easily drive full neoplastic transformation. In instances where oncogenic drivers are able to overcome the restrictive epigenetic barrier presented by wild-type EZH2, I propose that cancer cells co-opt EZH2 to promote and maintain the neoplastic phenotype. This co-option would be driven by an oncogenic-signalling-induced redistribution of EZH2 activity on chromatin, leading EZH2 to repress a new set of genes, such as EMX2 in GBM, corrupting normal developmental programmes and thus maintaining a neoplastic gene expression programme (Fig. 59).

Supporting this model, recent work investigating the role of EZH2 in acute myeloid leukaemia (AML) demonstrates that during disease induction EZH2 acts as a tumour suppressor, whilst upon progression to the maintenance phase EZH2 becomes a tumour promoter (Basheer et al, 2019). Furthermore, this shifting function in AML is attributed to the different gene expression programmes regulated by EZH2 at each stage of disease progression, in line with my findings. Together, this cancer stage-specific model of EZH2 function provides a possible explanation for the dichotomous roles EZH2 performs in cancer.

171

6.2 EZH2 redistribution switches the expression of important homeotic genes in glioblastoma

6.2.1 EZH2 redistribution derepresses HOX genes upon transformation

In this study, I observed that transformation leads to a general loss of H3K27me3 and EZH2 from all four HOX clusters, in particular from posterior HOX genes (HOX6-13). In line with this loss of repressive chromatin, I found substantial and recurrent upregulation of many HOX genes in expression data sets from GBM patient samples and cell lines. Furthermore, I showed that expression of many HOX genes positively correlates with tumour grade in glioma patient samples, supporting clinical relevance for this misregulation. In accordance with my findings, broad changes in the expression and epigenetic features of HOX clusters have also been observed in other studies investigating GBM. Quantification of HOX gene mRNA and protein levels in normal astrocytes and GBM cell lines revealed that nine different HOX genes are consistently upregulated in GBM cell lines (Abdel-Fattah et al, 2006). In addition, H3K27ac ChIP-seq and RNA-seq profiling in GBM stem cells and normal neural stem cells identified that in GBM stem cells H3K27ac is gained across large regions of HOX clusters, a change that is accompanied by matched upregulation of many HOX genes (Zhou et al, 2018). Emerging evidence also demonstrates that ectopic expression of many HOX genes, including HOXC10, HOXA13, HOXA6 and HOXB13, can promote glioma invasiveness and oncogenic signalling, in line with a general tumour promoting role for HOX genes in GBM (Duan et al, 2015; Guan et al, 2019; Guo et al, 2016). Thus, upregulation of HOX gene expression upon transformation may be an important tumour promoting event in GBM.

Intriguingly, an interesting parallel can be drawn between the broad reconfiguration of repressive chromatin at HOX clusters in GBM, and events occurring in normal neural development. During the determination of rostro caudal neural identity in the spinal cord, developmental signalling from the WNT and Fibroblast growth factor pathways drives clearance of H3K27me3 from large domains encompassing multiple genes on the HOX clusters, an event required for correct positional identity of motor neurons (Mazzoni et al, 2013). It is interesting to speculate that misregulation of normal developmental signalling pathways in GBM may cooperate with oncogenic signalling to drive the changes in repressive chromatin at HOX genes which I observe. In line with this, both

172

WNT and Fibroblast growth factor signalling is misregulated in GBM, and is required to maintain tumour growth (Zuccarini et al, 2018; Loilome et al, 2009; Haley & Kim, 2014). Thus, exploring the relationship between HOX gene expression and aberrant developmental signalling pathways could offer an interesting future research direction for this project.

6.2.2 EMX2 repression by EZH2 may be important in glioblastoma maintenance

In parallel to HOX gene upregulation, I found that transformation is also accompanied by EZH2-mediated repression of the neural homeotic gene EMX2. During neural development, EMX2 is expressed in the developing forebrain where it acts to drive correct formation of cerebral cortex regions such as the hippocampus and olfactory bulbs (Cecchi, 2002). As development progresses, EMX2 expression becomes restricted to neural stem cells of the sub-ventricular zone, where it is maintained into adult life (Cecchi, 2002). In neural stem cells, EMX2 regulates the size of the stem cell pool through negatively regulating proliferation by promoting asymmetric stem cell division (Galli et al, 2002). Importantly, alongside EMX2, EZH2 is also expressed in neural stem cells, where it opposes EMX2 by maintaining both neural stem cell proliferation and neurogenic competence (Hwang et al, 2014). These antithetic roles suggest an EMX2- EZH2 regulatory axis which maintains the neural stem cell population at an optimum size. However, upon neoplastic transformation, I found that EZH2 is co-opted to drive repression of EMX2, disrupting this regulatory axis and potentially enabling the unlimited cellular self-renewal required for GBM growth (Fig. 59). Supporting the functional relevance of EMX2 repression, I found that ectopic expression of EMX2 in GBM cell lines induces inhibition of proliferation and impairs in vivo tumour forming ability.

Further support comes from previous studies which have explored the role of EMX2 in GBM and other cancer types. In Lung, Endometrial, Gastric and Colon cancer EMX2 is repressed relative to normal tissue and has tumour suppressive ability (Yue et al, 2015; Qiu et al, 2013; Okamoto et al, 2010; Aykut et al, 2017; Li et al, 2012). Although EMX2 repression has not previously been observed in GBM, ectopic expression of EMX2 in GBM cell lines was shown to impair their tumour forming ability in the mouse cerebral cortex (Falcone et al, 2016). Moreover, ectopic expression of EMX2 in GBM cell lines leads to cell cycle arrest (Monnier et al, 2018; Falcone et al, 2016). Interestingly, de novo

173

DNA methylation appears to repress EMX2 expression in gastric cancer, demonstrating that cancer can employ alternative epigenetic mechanisms to repress this potent tumour suppressor (Okamoto et al, 2010), a finding I also observed in GBM cell lines (Fig. 33C). Thus, EZH2-mediated repression of EMX2 upon transformation is in agreement with EMX2’s emerging tumour suppressive role in cancer, and appears to be an important event in maintaining GBM growth.

6.2.3 Ras signalling initiates EZH2 redistribution at neural homeotic genes

I observed that EZH2-mediated repression of EMX2 and loss of EZH2 from HOX clusters occurs upon the expression of oncogenic HRASv12 in the de novo transformation model used in this study. This indicates that hyperactivated Ras pathway signalling, driven by HRASv12, is the initiating factor that drives EZH2 redistribution to mediate a switch in EMX2 and HOX expression. However, whether the ability of HRASv12 to drive this redistribution is specific to the downstream effects of the hyperactivated Ras pathway or a more general feature of neoplastic transformation is unclear. In a previous study by Hosogane et al, expression of oncogenic RAS in mouse 3T3 cells was shown to induce redistribution of EZH2 and H3K27me3 (Hosogane et al, 2013). Importantly, EZH2/H3K27me3 redistribution or expression changes does not happen at EMX2 and HOX genes in this system, arguing that redistribution of EZH2 at these genes is not a conserved feature of hyperactivating the Ras pathway. However, it should be noted that this study was performed in murine cells over a short time course, potentially confounding an accurate comparison with my findings. Moreover, the cancer types (Lung, Gastric, Endometrial and Colorectal) in which EMX2 has been identified to be silenced have some of the highest frequencies of Ras driver mutations in all malignancies (Hobbs et al, 2016), providing circumstantial evidence for Ras-specific induction of EZH2 redistribution at these genes. Thus, it is unclear whether the homeotic gene expression switch which I observe is pre-requisite on the hyperactivation of the Ras signalling pathway or whether silencing can also be achieved by other transformative genetic events. A possible future experiment to investigate the role of Ras signalling in more details would be to stratify GBM patients by their degree of Ras pathway activation and explore whether this correlated with EMX2 and HOX gene expression.

174

In the specific case of EMX2, I find that EZH2-mediated repression is accompanied by the collapse of a surrounding network of putative enhancers, suggesting an additional layer of complexity to the regulation of EMX2 expression. It is possible that signalling pathways downstream of oncogenic Ras drive the collapse of this enhancer network, initiating silencing of EMX2 expression which could lead to recruitment of PRC2 to maintain gene repression. Experimental disruption of these enhancers would clarify whether they are indeed key regulatory elements in controlling EMX2 expression in GBM. Enhancer-mediated gene regulation enables nuanced spatial and temporal control of transcription factor expression (Long et al, 2016). Thus, it is interesting to speculate that the putative enhancers I have identified surrounding the EMX2 locus could be important in regulating the expression of EMX2 expression patterns during forebrain development. The regulation of EMX2 expression in this process remains poorly defined, and further interrogation of these putative regulatory elements may reveal interesting mechanisms that control neural development.

175

6.3 Transformation induces changes to PRC2 composition that may promote tumour growth

Accessory proteins are key mediators of PRC2 targeting and allosteric regulation, and an intricate picture is now emerging of a diverse family of PRC2 sub-complexes which act in concert to regulate gene expression (Holoch & Margueron, 2017). Moreover, studies of PRC1 and PRC2 reveal that complex composition can be dynamic, and compositional changes often occur during cell differentiation (Kloet et al, 2016; O’Loghlen et al, 2012; Oliviero et al, 2016). However, it has remained unclear whether changes in PRC2 composition, often observed in normal development, are also relevant events in promoting tumour initiation and maintenance. By profiling changes to the composition of PRC2 in a transformation model, I revealed that transformation leads to an increase in PRC2-MTF2 and PRC2-AEBP2, and a concomitant decrease in PRC2- PHF1 and PRC2-PHF19. This provides, to my knowledge, the first evidence that transformation is able to induce a shift in the composition of wild-type PRC2. This switch in PRC2 composition is particularly striking in light of similar changes which have been observed in development. During differentiation of pluripotent embryonic stem cells to lineage-committed neural progenitor cells, PRC2-MTF2 is depleted and PRC2-PHF19 is enriched (Kloet et al, 2016). Moreover, in an embryonic carcinoma model of differentiation, PRC2-PHF1 has increased levels upon differentiation (Oliviero et al, 2016). Thus, intriguingly, the PRC2 composition shift I observed during transformation appears to be the inverse of changes occurring upon cellular differentiation.

The shift in PRC2 composition appears, in part, to be driven by transcriptional changes of each accessory component. Previously, the expression of PHF19 and MTF2 has been shown to be regulated by proliferation via the E2F pathway, whilst PHF1 is expressed primarily in quiescent cells (Brien et al, 2015). Given the increased proliferation rate transformation induces, this proliferation linkage aligns with the expression changes I observed for MTF2 and PHF1, but is contrary to the decrease in PHF19 expression, suggesting that additional competing pathways regulate PHF19 expression. Furthermore, differential expression does not appear sufficient to explain the full magnitude of binding changes observed, suggesting that additional mechanisms are required to mediate this switch. Thus, further work will be required to unravel the regulatory mechanisms that control this switch in PRC2 composition.

176

Supporting a functional relevance for this composition shift, I found that depleting accessory proteins which have enhanced binding to PRC2 upon transformation (MTF2 and AEBP2) impairs tumour xenograft growth, whilst depleting those with decreased binding (PHF1 and PHF19) enhances it. This shows that, in the context of this transformation model, MTF2/AEBP2 act as tumour promoters, and PHF1/PHF19 are tumour suppressors. However, arguing against generalisation of these tumour roles, I found that in tumour xenografts derived from cell lines of other cancer types, the tumour roles of PHF19 and MTF2 are not conserved and vary between, and even within, cancer types. This apparent context dependency aligns with existing literature which suggests dichotomous roles for accessory proteins in cancer. For example, PHF1 has been assigned both tumour promoting (Boulay et al, 2012; Liu et al, 2018) and suppressive roles (Yang et al, 2013; Brien et al, 2015) in different cancer contexts. Alternatively, the differing results I saw between cancer types may also be explainable by context-specific off-target effects of shRNA knockdown, a possibility that would need to be excluded by repeating the experiments with additional shRNAs. Together, I showed that transformation induces changes in PRC2 composition, likely through differential gene expression, that appear to mirror the specific function of an accessory protein in the tumour. It would be interesting to expand this study to other models of transformation to explore whether changes in PRC2 composition are generally predictive of accessory protein tumour function.

177

6.4 Pharmacological inhibition is an effective tool for defining PRC2-regulated genes

Accurate identification of PRC2-regulated genes enables determination of the mechanisms through which PRC2 maintains pathological and physiological cell states. To do this, genome-wide profiles (usually obtained by ChIP-seq) of H3K27me3 or PRC2 components are often integrated with transcriptomic analysis of gene expression changes induced by PRC2 component LOF. However, in this workflow, different genetic and pharmacological LOF approaches are often used interchangeably to disrupt PRC2 function. This interchangeable usage contrasts with the distinct mechanisms by which pharmacological inhibition, mRNA knock-down and gene knock-out disrupt protein function, and the growing evidence that each approach can have distinct phenotypic and transcriptional effects (Housden et al, 2017).

To perform a controlled comparison of these LOF approaches, I integrated transcriptomic analyses of EZH2 disruption by each approach with H3K27me3 ChIP-seq to define the different sets of PRC2-regulated genes they identify. I found that pharmacological inhibition derepresses the largest portion of PRC2 direct targets, induces a greater upregulation of target gene expression and is able to derepress genes marked by large domains of H3K27me3 which are refractory to genetic LOF approaches. Thus, pharmacological inhibition appears to enable a more comprehensive characterisation of PRC2-regulated genes than genetic methods. The greater efficacy of pharmacological inhibition versus genetic LOF approaches appears to have varied origins. Compared to mRNA knock-down, pharmacological inhibition induces a greater depletion of H3K27me3 levels, enabling de-repression of genes that remain silenced by residual H3K27me3 upon mRNA knock-down. Why EZH2 knock-out is not as effective as pharmacological inhibition is less clear. In part, remaining wild-type cells in the polyclonal knock-out population may mask expression changes. In addition, the loss of EZH2 protein may trigger compensatory mechanisms that could counteract the effects of EZH2 disruption - a mechanism that may also impair the efficacy of mRNA knock-down. As such, the greater homogeneity of PRC2 disruption, higher potency and lack of need for genetic manipulation, recommend pharmacological inhibition as an effective approach for defining the gene set controlled by PRC2. More generally, highly specific small molecule inhibitors are now available for an ever

178

increasing number of epigenetic regulators (Dawson, 2017), meaning a similar approach may also be relevant for investigating epigenetic regulators more broadly.

In considering these results, two assumptions made in the analysis must be noted. First, to enable RNA-seq normalisation, I assume that an average gene will not change its expression upon EZH2 LOF, a standard assumption used in differential expression analyses (Evans et al, 2018). As such, if EZH2 LOF does cause a global change in transcription, a possibility given EZH2’s genome-wide repressive activity, this normalisation technique would obscure a large portion of gene expression changes, preventing a correct characterisation of the transcriptional effects from EZH2 LOF. Notably, similar assumptions prevented identification of the role c-Myc plays as a general promoter of transcription (Lovén et al, 2012). Secondly, I assume that the time after initiating EZH2 LOF at which RNA is harvested provides an accurate picture of gene expression changes that a LOF technique induces. However, differing temporal dynamics of the expression changes induced by each technique may mean the time points selected do not accurately capture the full transcriptional consequences of EZH2 LOF. These assumptions represent important caveats for this analysis, and further experiments will be required to investigate whether they may explain the differences I observe between genetic and pharmacological LOF techniques.

6.5 Concluding remarks

The tumour promoting activity of epigenetic regulators is emerging as an important feature of cancer and an attractive therapeutic target in multiple malignancies. As such, it is increasingly pertinent that a detailed mechanistic understanding of these regulators is built to enable development of new therapies and accurate stratification of patients. Interestingly, many epigenetic regulators act as tumour promoters in their wild-type unmutated form, a phenomenon that is poorly understood and may represent an untapped source of therapeutic targets. In this thesis, I have provided a detailed dissection of the mechanisms which allow Polycomb component EZH2 to act as a tumour promoter in its wild-type unmutated state. In doing so, I offer a unique insight into the mechanisms by which epigenetic regulators in their wild-type form can attain tumour promoting ability. Specifically, I have shown that in glioblastoma, transformation induces a functionally important redistribution of EZH2 on chromatin that rewires normal neural transcriptional programmes and corrupts neural cell identity. This mechanism offers an exciting start point for future research to further unravel, and possibly disrupt, the intricate

179

mechanisms that corrupt cell identity to drive glioblastoma. Importantly, I believe a similar transformation-induced redistribution could also explain how other epigenetic regulators attain tumour promoting ability in their wild-type state, suggesting a broader relevance for my findings that go beyond the roles of EZH2 in cancer. These results lead me to propose a novel ‘stage-specific’ model that rationalises the dichotomous tumour suppressive and tumour promoting roles of EZH2 in cancer, a model that I believe can help us to rationalise EZH2’s antithetic roles in cancer. Together, this study offers an important insight into the mechanisms which underlie the activity of an important tumour promoter, and proposes general principals which are likely to be relevant to understanding the role of epigenetic regulators in cancer more generally.

180

Chapter 7. Appendix

7.1 Appendix 1 - Oligonucleotides used in this study

Oligonucleotide Sequence (5’-3’) EMX2 quantification RT-qPCR/ChIP-qPCR Forward CACACCCCCTATTCGCCTC EMX2 quantification RT-qPCR/ ChIP-qPCR Reverse CCAGATATCGGTAGCGGTGG Cyclophillin quantification RT-qPCR Forward GTCAACCCCACCGTGTTCTT Cyclophillin quantification RT-qPCR Reverse CTGCTGTCTTTGGGACCTTGT HOXB9 quantification RT-qPCR Forward GGAAACTTCGGCGGGCA HOXB9 quantification RT-qPCR Reverse TTTTTCCGGGAAGAGCGAGC HOXB9 ChIP-qPCR Forward CTGTCCCAGACCACTTGTCC HOXB9 ChIP-qPCR Reverse GCCAGGAGAGTCCGAATGAG EZH2 CRISPR KO test Forward ACAGGAAACGATTGCCATCC EZH2 CRISPR KO test Reverse GCACAAATGAGCACCTTTCTG GAPDH ChIP-qPCR Forward TCCAATTCCCCATCTCAGTC GAPDH ChIP-qPCR Reverse GCAGCAGGACACTAGGGAGT LYPD4 Methylation test primer Forward GATATAAGGAAAGGGGTGTAATTTGAGG LYPD4 Methylation test primer Reverse CCTTCCCTTCTACCAACTTCTTAAC TNFRSF21 Methylation test primer Forward TTGAGGAGAATTAGGGAGGAGGATT TNFRSF21 Methylation test primer Reverse AAAACACCCTCTCCCTCCCC SPO11 Methylation test primer Forward TTAAATTATTTTAAGGGATTTTAGGTTTAA SPO11 Methylation test primer Reverse TCCTCAAAACAACCAACAAAAACTC PPP4R4 Methylation test primer Forward GGTTATATGGAGGATTTGTAGGAGTT PPP4R4 Methylation test primer Reverse CCCTTAAAAATAAAAATCCAAACTAAAA CALB2 Methylation test primer Forward TTGTTATTGTTAGTAATAATAAGTGGTGGA CALB2 Methylation test primer Reverse CAAAATACTTCCATATTTCCAAAAACTAAA POU6F2 Methylation test primer Forward AGAAGTTTAGATTTTAGGGGAAGTTG POU6F2 Methylation test primer Reverse AACACTAAACAAAAAACCAAATAAC EZH2/H3K27me3 WT1.1 ChIP-qPCR Forward GGGATGAGAAACCAACCTGA EZH2/H3K27me3 WT1.1 ChIP-qPCR Reverse GGGTTAGAACAAGGACAGGG PELI2 quantification RT-qPCR Forward GGAGCCAGTGAAATACGGGG PELI2 quantification RT-qPCR Reverse GGATATCACATGGACGGTGCT LBH quantification RT-qPCR Forward CGCCCGTGTCATCCTCACT LBH quantification RT-qPCR Reverse CCTCAGTCATCTTGGCCGAT MTF2 quantification RT-qPCR Forward CAGTAAGTGCTCGGACTCGC MTF2 quantification RT-qPCR Reverse GTCCATCCTGCCCCTGTAGA AEBP2 quantification RT-qPCR Forward GATGTCGTCGGATGGGGAAC AEBP2 quantification RT-qPCR Reverse GAGTTGAACGCCCACTGGAA PHF19 quantification RT-qPCR Forward CTACCTCGGGAAGATCAAGAG PHF19 quantification RT-qPCR Reverse CTAGGCAGATGTTGCACTTGG PHF1 quantification RT-qPCR Forward TAGGTCTCCTACGTCTGCCC PHF1 quantification RT-qPCR Reverse GAAGCTGGGTCCCAAAGTGA SOX11 ChIP-qPCR Forward AACCTTCCTCCTCCAAAGGA SOX11 ChIP-qPCR Reverse GGAATATGGGTGGCAGGTTA

181

7.2 Appendix 2 - Cell culturing conditions

Cell Line Medium FBS NEAA L-Q (mM) P/S (mM) HEPES (mM) Source Untransformed immortalised fibroblasts MEM 15% - 100 100 - Scaffidi and Misteli (2012) Pre-neoplastic immortalised fibroblasts MEM 15% - 100 100 - Scaffidi and Misteli (2012) Transformed immortalised fibroblasts MEM 15% - 100 100 - Scaffidi and Misteli (2012) Tumour-derived cell lines MEM 15% - 100 100 - This study DBTRG-05MG RPMI 1640 10% - 100 100 25 Crick Institute Common Repository M059K DMEM 10% - 100 100 - Crick Institute Common Repository U-87 MG DMEM 10% 1% 100 100 - Crick Institute Common Repository T98G MEM 10% - 100 100 - Crick Institute Common Repository AGS Ham’s F12 10% - - 100 - Crick Institute Common Repository Patient-derived xenografts RPMI 1640 10% - 100 100 25 Charles River HEK293T DMEM 10% - 100 100 - Crick Institute Common Repository H1299 RPMI 1640 10% - 100 100 25 Crick Institute Common Repository LN-229 DMEM 10% - 100 100 - Crick Institute Common Repository HCT116 DMEM 10% - 100 100 - Crick Institute Common Repository A549 DMEM 10% - 100 100 - Crick Institute Common Repository Phoenix DMEM 10% - 100 100 - Crick Institute Common Repository

MEM – Minimal Essential Media, RPMI 1640 - Roswell Park Memorial Institute 1640 Media, DMEM – Dulbecco’s Modified Eagle Media, FBS – Fetal Bovine Serum, NEAA – Non-essential Amino Acids, P/S – Penicillin/Streptomycin.

182

7.3 Appendix 3 - Sources of publically available data

Data type Source Link NCI-60 gene level methylation values National Cancer Institute discover.nci.nih.gov/cellminer/ Normalised microarray mRNA expression Repository of Molecular Brain betastasis.com/glioma/rembrandt values from GBM and normal brain samples Neoplasia Data RNA-seq mRNA expression values for low and Chinese Glioma Genome Atlas cgga.org.cn:9091/gliomasdb high-grade glioma patients RNA-seq mRNA expression values for human BrainSeq2 - Zhang et al, 2016 brainrnaseq.org primary neural cells RNA-seq mRNA expression values from glioma Cancer Cell Line Encyclopaedia portals.broadinstitute.org/ccle/data cell lines Reduced representation bisulphite sequencing Cancer Cell Line Encyclopaedia portals.broadinstitute.org/ccle/data DNA methylation values from glioma cell lines RNA-seq expression values from laser Ivy Glioblastoma Atlas glioblastoma.alleninstitute.org/rnaseq/ microdissected GBM tumours search/index.html H3K27me3, H3K27ac and H3K4me3 ChIP-seq ENCODE encodeproject.org data from primary cells TAD boundaries from IMR90 lung fibroblasts Dixon et al, 2012 nature.com/articles/nature11082 Normalised microarray mRNA expression Codega et al, 2014 ncbi.nlm.nih.gov/pmc/articles/PMC4360885 values for SVZ NSCs RNA-seq expression values from neural ENCODE encodeproject.org progenitor cells CpG island annotation Bock et al, 2007 hgdownload.soe.ucsc.edu/goldenPath /hg19/database/cpgIslandExt.txt.gz

183

7.4 Appendix 4 - ChIP-seq correlation heatmap datasets (Figure. 10)

Cell type Histone mark ENCODE ChIP-seq file Ref ENCODE Input file Ref NHA H3K27me3 ENCFF000CHN, ENCFF000CHR ENCFF350LQZ NHA H3K27ac ENCFF100JEN, ENCFF384LZD ENCFF191SLR NHA H3K4me3 ENCFF651YTD, ENCFF027IJX MEC H3K27me3 ENCFF323ZZH, ENCFF813AGD ENCFF062WDT MEC H3K27ac ENCFF674SDM, ENCFF807VEH ENCFF903ATW MEC H3K4me3 ENCFF037RFR, ENCFF095KBB Osteoblast H3K27me3 ENCFF196FGA, ENCFF010EZQ Osteoblast H3K27ac ENCFF535AIK, ENCFF767YLX ENCFF000CUZ ENCFF814GWN,ENCFF608CBJ, ENCFF000CVB Osteoblast H3K4me3 ENCFF588XYH PBMC H3K27me3 ENCFF142FPZ PBMC H3K27ac ENCFF542XPY ENCFF289ZBB PBMC H3K4me3 ENCFF122XTO CD4 T H3K27me3 ENCFF442TLB, ENCFF443MBE CD4 T H3K27ac ENCFF057SES, ENCFF199ZCW ENCFF843SOW CD4 T H3K4me3 ENCFF870MPK NHLF H3K27me3 ENCFF926ZDR,ENCFF330AVF ENCFF208SHP NHLF H3K27ac ENCFF066AQB, ENCFF377BNX ENCFF250VFW NHLF H3K4me3 ENCFF148IFM, ENCFF689JQV Keratinocyte H3K27me3 ENCFF285QUM, ENCFF873WML ENCFF221LXY Keratinocyte H3K27ac ENCFF986JEU, ENCFF907JBE ENCFF288LLD Keratinocyte H3K4me3 ENCFF747OGU, ENCFF174BJB ENCFF694HYE, ENCFF255WHJ DF H3K27me3 ENCFF138XPE ENCFF197IFZ ENCFF005RXL DF H3K27ac ENCFF688WGV,ENCFF721HIJ ENCFF615VVU DF H3K4me3 ENCFF664FYA,ENCFF174XNM

NHA – Normal human astrocytes, MEC – Mammary epithelial cells, PBMC – Peripheral blood mononuclear cells, CD4 T – CD4 T-cells, NHLF – Normal human lung fibroblasts, DF – Dermal fibroblasts

184

7.5 Appendix 5 - NCI-60 cell lines (Figure. 23B)

Cell line number NCI60 cell line name Cell line number NCI60 cell line name 1 MALME.3M 28 SK.MEL.5 2 RPMI.8226 29 NCI.H226 3 M14 30 786.0 4 UO.31 31 HL.60.TB 5 NCI.H23 32 SW.620 6 HOP.62 33 DU.145 7 OVCAR.3 34 HOP.92 8 UACC.257 35 HCC.2998 9 OVCAR.4 36 MCF7 10 SK.MEL.2 37 MDA.MB.231 11 BT.549 38 MOLT.4 12 NCI.H522 39 OVCAR.5 13 T.47D 40 NCI.H322M 14 K.562 41 MDA.MB.435 15 SN12C 42 CCRF.CEM 16 NCI.H460 43 OVCAR.8 17 A549.ATCC 44 HT29 18 SK.OV.3 45 A498 19 HS.578T 46 MDA.N 20 SK.MEL.28 47 COLO.205 21 PC.3 48 LOX.IMVI 22 IGROV1 49 NCI.ADR.RES 23 EKVX 50 HCT.15 24 UACC.62 51 TK.10 25 CAKI.1 52 SR 26 ACHN 53 HCT.116 27 RXF.393 54 KM12

185

Reference List

Abdel-Fattah R, Xiao A, Bomgardner D, Pease C-S, Lopes M-B & Hussaini I (2006) Differential expression of HOX genes in neoplastic and non-neoplastic human astrocytes. J. Pathol. 209: 15–24 Abdel-Wahab O, Mullally A, Hedvat C, Garcia-Manero G, Patel J, Wadleigh M, Malinge S, Yao J, Kilpivaara O, Bhat R, Huberman K, Thomas S, Dolgalev I, Heguy A, Paietta E, Le Beau MM, Beran M, Tallman MS, Ebert BL, Kantarjian HM, et al (2009) Genetic characterization of TET1, TET2, and TET3 alterations in myeloid malignancies. Blood 114: 144–147 Agger K, Cloos PAC, Christensen J, Pasini D, Rose S, Rappsilber J, Issaeva I, Canaani E, Salcini AE & Helin K (2007) UTX and JMJD3 are histone H3K27 demethylases involved in HOX gene regulation and development. Nature 449: 731–734 Ai S, Peng Y, Li C, Gu F, Yu X, Yue Y, Ma Q, Chen J, Lin Z, Zhou P, Xie H, Prendiville TW, Zheng W, Liu Y, Orkin SH, Wang D-Z, Yu J, Pu WT & He A (2017) EED orchestration of heart maturation through interaction with HDACs is H3K27me3- independent. Elife 6: Albert M, Kalebic N, Florio M, Lakshmanaperumal N, Haffner C, Brandl H, Henry I & Huttner WB (2017) Epigenome profiling and editing of neocortical progenitor cells during development. EMBO J. 36: 2642–2658 Alcantara Llaguno SR & Parada LF (2016) Cell of origin of glioma: biological and clinical implications. Br. J. Cancer 115: 1445–1450 Aplan PD (2006) Chromosomal translocations involving the MLL gene: Molecular mechanisms. DNA Repair (Amst). 5: 1265–1272 Aranda S, Mas G & Di Croce L (2015) Regulation of gene transcription by Polycomb proteins. Sci. Adv. 1: e1500737 Arnold P, Scholer A, Pachkov M, Balwierz PJ, Jorgensen H, Stadler MB, van Nimwegen E & Schubeler D (2013) Modeling of epigenome dynamics identifies transcription factors that mediate Polycomb targeting. Genome Res. 23: 60–73 Atlasi Y & Stunnenberg HG (2017) The interplay of epigenetic marks during stem cell differentiation and development. Nat. Rev. Genet. 18: 643–658 Aykut B, Ochs M, Radhakrishnan P, Brill A, Höcker H, Schwarz S, Weissinger D, Kehm R, Kulu Y, Ulrich A & Schneider M (2017) EMX2 gene expression predicts liver metastasis and survival in colorectal cancer. BMC Cancer 17: 555

186

Ballaré C, Lange M, Lapinaite A, Martin GM, Morey L, Pascual G, Liefke R, Simon B, Shi Y, Gozani O, Carlomagno T, Benitah SA & Di Croce L (2012) Phf19 links methylated Lys36 of histone H3 to regulation of Polycomb activity. Nat. Struct. Mol. Biol. 19: 1257–1265 Barber BA & Rastegar M (2010) Epigenetic control of Hox genes during neurogenesis, development, and disease. Ann. Anat. - Anat. Anzeiger 192: 261–274 Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, Wilson CJ, Lehár J, Kryukov G V, Sonkin D, Reddy A, Liu M, Murray L, Berger MF, Monahan JE, Morais P, Meltzer J, Korejwa A, Jané-Valbuena J, Mapa FA, et al (2012) The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature 483: 603–7 Basheer F, Giotopoulos G, Meduri E, Yun H, Mazan M, Sasca D, Gallipoli P, Marando L, Gozdecka M, Asby R, Sheppard O, Dudek M, Bullinger L, Döhner H, Dillon R, Freeman S, Ottmann O, Burnett A, Russell N, Papaemmanuil E, et al (2019) Contrasting requirements during disease evolution identify EZH2 as a therapeutic target in AML. J. Exp. Med. 216: 966–981 Béguelin W, Popovic R, Teater M, Jiang Y, Bunting KL, Rosen M, Shen H, Yang SN, Wang L, Ezponda T, Martinez-Garcia E, Zhang H, Zheng Y, Verma SK, McCabe MT, Ott HM, Van Aller GS, Kruger RG, Liu Y, McHugh CF, et al (2013) EZH2 Is Required for Germinal Center Formation and Somatic EZH2 Mutations Promote Lymphoid Transformation. Cancer Cell 23: 677–692 Behjati S, Tarpey PS, Presneau N, Scheipl S, Pillay N, Van Loo P, Wedge DC, Cooke SL, Gundem G, Davies H, Nik-Zainal S, Martin S, McLaren S, Goody V, Robinson B, Butler A, Teague JW, Halai D, Khatri B, Myklebost O, et al (2013) Distinct H3F3A and H3F3B driver mutations define chondroblastoma and giant cell tumor of bone. Nat. Genet. 45: 1479–1482 Beltran M, Yates CM, Skalska L, Dawson M, Reis FP, Viiri K, Fisher CL, Sibley CR, Foster BM, Bartke T, Ule J & Jenner RG (2016) The interaction of PRC2 with RNA or chromatin is mutually antagonistic. Genome Res. 26: 896–907 Beringer M, Pisano P, Di Carlo V, Blanco E, Chammas P, Vizán P, Gutiérrez A, Aranda S, Payer B, Wierer M & Di Croce L (2016) EPOP Functionally Links Elongin and Polycomb in Pluripotent Stem Cells. Mol. Cell 64: 645–658 Bernstein BE, Mikkelsen TS, Xie X, Kamal M, Huebert DJ, Cuff J, Fry B, Meissner A, Wernig M, Plath K, Jaenisch R, Wagschal A, Feil R, Schreiber SL & Lander ES (2006a) A Bivalent Chromatin Structure Marks Key Developmental Genes in

187

Embryonic Stem Cells. Cell 125: 315–326 Bernstein E, Duncan EM, Masui O, Gil J, Heard E & Allis CD (2006b) Mouse Polycomb Proteins Bind Differentially to Methylated Histone H3 and RNA and Are Enriched in Facultative Heterochromatin. Mol. Cell. Biol. 26: 2560–2569 Bhat KPL, Balasubramaniyan V, Vaillant B, Ezhilarasan R, Hummelink K, Hollingsworth F, Wani K, Heathcock L, James JD, Goodman LD, Conroy S, Long L, Lelic N, Wang S, Gumin J, Raj D, Kodama Y, Raghunathan A, Olar A, Joshi K, et al (2013) Mesenchymal Differentiation Mediated by NF-κB Promotes Radiation Resistance in Glioblastoma. Cancer Cell 24: 331–346 Bitler BG, Aird KM, Garipov A, Li H, Amatangelo M, Kossenkov A V, Schultz DC, Liu Q, Shih I-M, Conejo-Garcia JR, Speicher DW & Zhang R (2015) Synthetic lethality by targeting EZH2 methyltransferase activity in ARID1A-mutated cancers. Nat. Med. 21: 231–238 Blackledge NP, Farcas AM, Kondo T, King HW, McGouran JF, Hanssen LLP, Ito S, Cooper S, Kondo K, Koseki Y, Ishikura T, Long HK, Sheahan TW, Brockdorff N, Kessler BM, Koseki H & Klose RJ (2014) Variant PRC1 Complex-Dependent H2A Ubiquitylation Drives PRC2 Recruitment and Polycomb Domain Formation. Cell 157: 1445–1459 Blackledge NP, Rose NR & Klose RJ (2015) Targeting Polycomb systems to regulate gene expression: modifications to a complex story. Nat. Rev. Mol. Cell Biol. 16: 643–649 Bock C, Walter J, Paulsen M & Lengauer T (2007) CpG Island Mapping by Epigenome Prediction. PLoS Comput. Biol. 3: e110 Bodega B, Marasca F, Ranzani V, Cherubini A, Della Valle F, Neguembor MV, Wassef M, Zippo A, Lanzuolo C, Pagani M & Orlando V (2017) A cytosolic Ezh1 isoform modulates a PRC2–Ezh1 epigenetic adaptive response in postmitotic cells. Nat. Struct. Mol. Biol. 24: 444–452 Bond AM, Ming G-L & Song H (2015) Adult Mammalian Neural Stem Cells and Neurogenesis: Five Decades Later. Cell Stem Cell 17: 385–95 Boulay G, Dubuissez M, Van Rechem C, Forget A, Helin K, Ayrault O & Leprince D (2012) Hypermethylated in Cancer 1 (HIC1) Recruits Polycomb Repressive Complex 2 (PRC2) to a Subset of Its Target Genes through Interaction with Human Polycomb-like (hPCL) Proteins. J. Biol. Chem. 287: 10509–10524 Boyer LA, Plath K, Zeitlinger J, Brambrink T, Medeiros LA, Lee TI, Levine SS, Wernig M, Tajonar A, Ray MK, Bell GW, Otte AP, Vidal M, Gifford DK, Young RA &

188

Jaenisch R (2006) Polycomb complexes repress developmental regulators in murine embryonic stem cells. Nature 441: 349–353 Bracken AP, Kleine-Kohlbrecher D, Dietrich N, Pasini D, Gargiulo G, Beekman C, Theilgaard-Mönch K, Minucci S, Porse BT, Marine JC, Hansen KH & Helin K (2007) The Polycomb group proteins bind throughout the INK4A-ARF locus and are disassociated in senescent cells. Genes Dev. Bracken AP, Pasini D, Capra M, Prosperini E, Colli E & Helin K (2003) EZH2 is downstream of the pRB-E2F pathway, essential for proliferation and amplified in cancer. EMBO J. 22: 5323–5335 Brennan CW, Verhaak RGW, McKenna A, Campos B, Noushmehr H, Salama SR, Zheng S, Chakravarty D, Sanborn JZ, Berman SH, Beroukhim R, Bernard B, Wu C-J, Genovese G, Shmulevich I, Barnholtz-Sloan J, Zou L, Vegesna R, Shukla SA, Ciriello G, et al (2013) The somatic genomic landscape of glioblastoma. Cell 155: 462–77 Brien GL, Gambero G, O’Connell DJ, Jerman E, Turner SA, Egan CM, Dunne EJ, Jurgens MC, Wynne K, Piao L, Lohan AJ, Ferguson N, Shi X, Sinha KM, Loftus BJ, Cagney G & Bracken AP (2012) Polycomb PHF19 binds H3K36me3 and recruits PRC2 and demethylase NO66 to embryonic stem cell genes during differentiation. Nat. Struct. Mol. Biol. 19: 1273–1281 Brien GL, Healy E, Jerman E, Conway E, Fadda E, O’Donovan D, Krivtsov A V, Rice AM, Kearney CJ, Flaus A, McDade SS, Martin SJ, McLysaght A, O’Connell DJ, Armstrong SA & Bracken AP (2015) A chromatin-independent role of Polycomb- like 1 to stabilize p53 and promote cellular quiescence. Genes Dev. 29: 2231–43 Brinkman AB, Gu H, Bartels SJJ, Zhang Y, Matarese F, Simmer F, Marks H, Bock C, Gnirke A, Meissner A & Stunnenberg HG (2012) Sequential ChIP-bisulfite sequencing enables direct genome-scale investigation of chromatin and DNA methylation cross-talk. Genome Res. 22: 1128–38 Brinkman EK, Chen T, Amendola M & van Steensel B (2014) Easy quantitative assessment of genome editing by sequence trace decomposition. Nucleic Acids Res. 42: e168–e168 Brookes E & Pombo A (2009) Modifications of RNA polymerase II are pivotal in regulating gene expression states. EMBO Rep. 10: 1213–1219 Brookes E, de Santiago I, Hebenstreit D, Morris KJ, Carroll T, Xie SQ, Stock JK, Heidemann M, Eick D, Nozaki N, Kimura H, Ragoussis J, Teichmann SA & Pombo A (2012) Polycomb associates genome-wide with a specific RNA

189

polymerase II variant, and regulates metabolic genes in ESCs. Cell Stem Cell 10: 157–70 Cai L, Rothbart SB, Lu R, Xu B, Chen W-Y, Tripathy A, Rockowitz S, Zheng D, Patel DJ, Allis CD, Strahl BD, Song J & Wang GG (2013) An H3K36 Methylation- Engaging Tudor Motif of Polycomb-like Proteins Mediates PRC2 Complex Targeting. Mol. Cell 49: 571–582 Calebiro D, Grassi ES, Eszlinger M, Ronchi CL, Godbole A, Bathon K, Guizzardi F, de Filippis T, Krohn K, Jaeschke H, Schwarzmayr T, Bircan R, Gozu HI, Sancak S, Niedziela M, Strom TM, Fassnacht M, Persani L & Paschke R (2016) Recurrent EZH1 mutations are a second hit in autonomous thyroid adenomas. J. Clin. Invest. 126: 3383–8 Cao R, Wang L, Wang H, Xia L, Erdjument-Bromage H, Tempst P, Jones RS & Zhang Y (2002) Role of Histone H3 Lysine 27 Methylation in Polycomb-Group Silencing. Science (80-. ). 298: 1039–1043 Cao R & Zhang Y (2004) SUZ12 Is Required for Both the Histone Methyltransferase Activity and the Silencing Function of the EED-EZH2 Complex. Mol. Cell 15: 57– 67 Caretti G, Di Padova M, Micales B, Lyons GE & Sartorelli V (2004) The Polycomb Ezh2 methyltransferase regulates muscle gene expression and skeletal muscle differentiation. Genes Dev. 18: 2627–38 Cecchi C (2002) Emx2: a gene responsible for cortical development, regionalization and area specification. Gene 291: 1–9 Chamberlain SJ, Yee D & Magnuson T (2008) Polycomb Repressive Complex 2 Is Dispensable for Maintenance of Embryonic Stem Cell Pluripotency. Stem Cells 26: 1496–1505 Chen L, Alexe G, Dharia N V., Ross L, Iniguez AB, Conway AS, Wang EJ, Veschi V, Lam N, Qi J, Gustafson WC, Nasholm N, Vazquez F, Weir BA, Cowley GS, Ali LD, Pantel S, Jiang G, Harrington WF, Lee Y, et al (2017) CRISPR-Cas9 screen reveals a MYCN-amplified neuroblastoma dependency on EZH2. J. Clin. Invest. 128: 446–462 Choi J, Bachmann AL, Tauscher K, Benda C, Fierz B & Müller J (2017) DNA binding by PHF1 prolongs PRC2 residence time on chromatin and thereby promotes H3K27 methylation. Nat. Struct. Mol. Biol. 24: 1039–1047 Cifuentes-Rojas C, Hernandez AJ, Sarma K & Lee JT (2014) Regulatory Interactions between RNA and Polycomb Repressive Complex 2. Mol. Cell 55: 171–185

190

Collinson A, Collier AJ, Morgan NP, Sienerth AR, Chandra T, Andrews S & Rugg-Gunn PJ (2016) Deletion of the Polycomb-Group Protein EZH2 Leads to Compromised Self-Renewal and Differentiation Defects in Human Embryonic Stem Cells. Cell Rep. 17: 2700–2714 Comet I, Riising EM, Leblanc B & Helin K (2016) Maintaining cell identity: PRC2- mediated regulation of transcription and cancer. Nat. Rev. Cancer 16: 803–810 Conway E, Jerman E, Healy E, Ito S, Holoch D, Oliviero G, Deevy O, Glancy E, Fitzpatrick DJ, Mucha M, Watson A, Rice AM, Chammas P, Huang C, Pratt-Kelly I, Koseki Y, Nakayama M, Ishikura T, Streubel G, Wynne K, et al (2018) A Family of Vertebrate-Specific Polycombs Encoded by the LCOR/LCORL Genes Balance PRC2 Subtype Activities. Mol. Cell 70: 408–421.e8 Cooper S, Dienstbier M, Hassan R, Schermelleh L, Sharif J, Blackledge NP, De Marco V, Elderkin S, Koseki H, Klose R, Heger A & Brockdorff N (2014) Targeting Polycomb to Pericentric Heterochromatin in Embryonic Stem Cells Reveals a Role for H2AK119u1 in PRC2 Recruitment. Cell Rep. 7: 1456–1470 Cox J & Mann M (2008) MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification. Nat. Biotechnol. 26: 1367–1372 Crea F, Hurt EM & Farrar WL (2010) Clinical significance of Polycomb gene expression in brain tumors. Mol. Cancer 9: 265 Di Croce L & Helin K (2013) Transcriptional regulation by Polycomb group proteins. Nat. Struct. Mol. Biol. 20: 1147–1155 Van Cutsem E, Sagaert X, Topal B, Haustermans K & Prenen H (2016) Gastric cancer. Lancet (London, England) 388: 2654–2664 Czermin B, Melfi R, McCabe D, Seitz V, Imhof A & Pirrotta V (2002) Drosophila enhancer of Zeste/ESC complexes have a histone H3 methyltransferase activity that marks chromosomal Polycomb sites. Cell 111: 185–96 Davidovich C & Cech TR (2015) The recruitment of chromatin modifiers by long noncoding RNAs: lessons from PRC2. RNA 21: 2007–22 Davidovich C, Zheng L, Goodrich KJ & Cech TR (2013) Promiscuous RNA binding by Polycomb repressive complex 2. Nat. Struct. Mol. Biol. 20: 1250–1257 Davis ME (2016) Glioblastoma: Overview of Disease and Treatment. Clin. J. Oncol. Nurs. 20: S2 Dawson MA (2017) The cancer epigenome: Concepts, challenges, and therapeutic opportunities. Science (80-. ). 355: 1147–1152

191

Dawson MA, Kouzarides T & Huntly BJP (2012) Targeting Epigenetic Readers in Cancer. N. Engl. J. Med. 367: 647–657 Dawson MA, Prinjha RK, Dittmann A, Giotopoulos G, Bantscheff M, Chan W-I, Robson SC, Chung C, Hopf C, Savitski MM, Huthmacher C, Gudgin E, Lugo D, Beinke S, Chapman TD, Roberts EJ, Soden PE, Auger KR, Mirguet O, Doehner K, et al (2011) Inhibition of BET recruitment to chromatin as an effective treatment for MLL-fusion leukaemia. Nature 478: 529–33 Deaton AM & Bird A (2011) CpG islands and the regulation of transcription. Genes Dev. 25: 1010–22 Delgado-López PD & Corrales-García EM (2016) Survival in glioblastoma: a review on the impact of treatment modalities. Clin. Transl. Oncol. 18: 1062–1071 Delgado-Olguín P, Huang Y, Li X, Christodoulou D, Seidman CE, Seidman JG, Tarakhovsky A & Bruneau BG (2012) Epigenetic repression of cardiac progenitor gene expression by Ezh2 is required for postnatal cardiac homeostasis. Nat. Genet. 44: 343–347 Dellino GI, Schwartz YB, Farkas G, McCabe D, Elgin SCR & Pirrotta V (2004) Polycomb silencing blocks transcription initiation. Mol. Cell 13: 887–93 Denissov S, Hofemeister H, Marks H, Kranz A, Ciotta G, Singh S, Anastassiadis K, Stunnenberg HG & Stewart AF (2014) Mll2 is required for H3K4 trimethylation on bivalent promoters in embryonic stem cells, whereas Mll1 is redundant. Development 141: 526–537 Dietrich N, Lerdrup M, Landt E, Agrawal-Singh S, Bak M, Tommerup N, Rappsilber J, Södersten E & Hansen K (2012) REST–Mediated Recruitment of Polycomb Repressor Complexes in Mammalian Cells. PLoS Genet. 8: e1002494 Dixon JR, Selvaraj S, Yue F, Kim A, Li Y, Shen Y, Hu M, Liu JS & Ren B (2012) Topological domains in mammalian genomes identified by analysis of chromatin interactions. Nature 485: 376–80 Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M & Gingeras TR (2013) STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29: 15–21 Donaldson-Collier MC, Sungalee S, Zufferey M, Tavernari D, Katanayeva N, Battistello E, Mina M, Douglass KM, Rey T, Raynaud F, Manley S, Ciriello G & Oricchio E (2019) EZH2 oncogenic mutations drive epigenetic, transcriptional, and structural changes within chromatin domains. Nat. Genet. 51: 517–528 Duan R, Han L, Wang Q, Wei J, Chen L, Zhang J, Kang C & Wang L (2015) HOXA13

192

is a potential GBM diagnostic marker and promotes glioma invasion by activating the Wnt and TGF-β pathways. Oncotarget 6: 27778–93 Enroth S, Rada-Iglesisas A, Andersson R, Wallerman O, Wanders A, Påhlman L, Komorowski J & Wadelius C (2011) Cancer associated epigenetic transitions identified by genome-wide histone methylation binding profiles in human colorectal cancer samples and paired normal mucosa. BMC Cancer 11: 450 Ernst T, Chase AJ, Score J, Hidalgo-Curtis CE, Bryant C, Jones A V, Waghorn K, Zoi K, Ross FM, Reiter A, Hochhaus A, Drexler HG, Duncombe A, Cervantes F, Oscier D, Boultwood J, Grand FH & Cross NCP (2010) Inactivating mutations of the histone methyltransferase gene EZH2 in myeloid disorders. Nat. Genet. 42: 722–726 Evans C, Hardin J & Stoebel DM (2018) Selecting between-sample RNA-Seq normalization methods from the perspective of their assumptions. Brief. Bioinform. 19: 776–792 Ezhkova E, Lien W-H, Stokes N, Pasolli HA, Silva JM & Fuchs E (2011) EZH1 and EZH2 cogovern histone H3K27 trimethylation and are essential for hair follicle homeostasis and wound repair. Genes Dev. 25: 485–498 Ezhkova E, Pasolli HA, Parker JS, Stokes N, Su I, Hannon G, Tarakhovsky A & Fuchs E (2009) Ezh2 Orchestrates Gene Expression for the Stepwise Differentiation of Tissue-Specific Stem Cells. Cell 136: 1122–1135 Falcone C, Daga A, Leanza G & Mallamaci A (2016) Emx2 as a novel tool to suppress glioblastoma. Oncotarget 7: 41005–41016 Farcas AM, Blackledge NP, Sudbery I, Long HK, McGouran JF, Rose NR, Lee S, Sims D, Cerase A, Sheahan TW, Koseki H, Brockdorff N, Ponting CP, Kessler BM & Klose RJ (2012) KDM2B links the Polycomb Repressive Complex 1 (PRC1) to recognition of CpG islands. Elife 1: e00205 Faust C, Lawson KA, Schork NJ, Thiel B & Magnuson T (1998) The Polycomb-group gene eed is required for normal morphogenetic movements during gastrulation in the mouse embryo. Development 125: 4495–506 Feinberg AP, Koldobskiy MA & Göndör A (2016) Epigenetic modulators, modifiers and mediators in cancer aetiology and progression. Nat. Rev. Genet. 17: 284–299 Feinberg AP & Vogelstein B (1983) Hypomethylation distinguishes genes of some human cancers from their normal counterparts. Nature 301: 89–92 Ferrai C, Torlai Triglia E, Risner‐Janiczek JR, Rito T, Rackham OJ, de Santiago I, Kukalev A, Nicodemi M, Akalin A, Li M, Ungless MA & Pombo A (2017) RNA

193

polymerase II primes Polycomb‐repressed developmental genes throughout terminal neuronal differentiation. Mol. Syst. Biol. 13: 946 Ferrari KJ, Scelfo A, Jammula S, Cuomo A, Barozzi I, Stützer A, Fischle W, Bonaldi T & Pasini D (2014) Polycomb-Dependent H3K27me1 and H3K27me2 Regulate Active Transcription and Enhancer Fidelity. Mol. Cell 53: 49–62 Fillmore CM, Xu C, Desai PT, Berry JM, Rowbotham SP, Lin Y-J, Zhang H, Marquez VE, Hammerman PS, Wong K-K & Kim CF (2015) EZH2 inhibition sensitizes BRG1 and EGFR mutant lung tumours to TopoII inhibitors. Nature 520: 239–242 Flavahan WA, Drier Y, Liau BB, Gillespie SM, Venteicher AS, Stemmer-Rachamimov AO, Suvà ML & Bernstein BE (2016) Insulator dysfunction and oncogene activation in IDH mutant gliomas. Nature 529: 110–114 Flavahan WA, Gaskell E & Bernstein BE (2017) Epigenetic plasticity and the hallmarks of cancer. Science (80-. ). 357: eaal2380 Forbes SA, Bindal N, Bamford S, Cole C, Kok CY, Beare D, Jia M, Shepherd R, Leung K, Menzies A, Teague JW, Campbell PJ, Stratton MR & Futreal PA (2011) COSMIC: mining complete cancer genomes in the Catalogue of Somatic Mutations in Cancer. Nucleic Acids Res. 39: D945–D950 Francis NJ, Kingston RE & Woodcock CL (2004) Chromatin Compaction by a Polycomb Group Protein Complex. Science (80-. ). 306: 1574–1577 Gajjar AJ & Robinson GW (2014) Medulloblastoma—translating discoveries from the bench to the bedside. Nat. Rev. Clin. Oncol. 11: 714–722 Galli R, Fiocco R, De Filippis L, Muzio L, Gritti A, Mercurio S, Broccoli V, Pellegrini M, Mallamaci A & Vescovi AL (2002) Emx2 regulates the proliferation of stem cells of the adult mammalian central nervous system. Development 129: 1633–44 Galupa R & Heard E (2015) X-chromosome inactivation: new insights into cis and trans regulation. Curr. Opin. Genet. Dev. 31: 57–66 Gifford CA, Ziller MJ, Gu H, Trapnell C, Donaghey J, Tsankov A, Shalek AK, Kelley DR, Shishkin AA, Issner R, Zhang X, Coyne M, Fostel JL, Holmes L, Meldrim J, Guttman M, Epstein C, Park H, Kohlbacher O, Rinn J, et al (2013) Transcriptional and Epigenetic Dynamics during Specification of Human Embryonic Stem Cells. Cell 153: 1149–1163 Gil J & O’Loghlen A (2014) PRC1 complex diversity: where is it taking us? Trends Cell Biol. 24: 632–641 Gil J & Peters G (2006) Regulation of the INK4b–ARF–INK4a tumour suppressor locus: all for one or one for all. Nat. Rev. Mol. Cell Biol. 7: 667–677

194

Gonzalez ME, Moore HM, Li X, Toy KA, Huang W, Sabel MS, Kidwell KM & Kleer CG (2014) EZH2 expands breast stem cells through activation of NOTCH1 signaling. Proc. Natl. Acad. Sci. 111: 3098–3103 Grijzenhout A, Godwin J, Koseki H, Gdula MR, Szumska D, McGouran JF, Bhattacharya S, Kessler BM, Brockdorff N & Cooper S (2016) Functional analysis of AEBP2, a PRC2 Polycomb protein, reveals a Trithorax phenotype in embryonic development and in ESCs. Development 143: 2716–2723 Guan Y, He Y, Lv S, Hou X, Li L & Song J (2019) Overexpression of HOXC10 promotes glioblastoma cell progression to a poor prognosis via the PI3K/AKT signalling pathway. J. Drug Target. 27: 60–66 Guo Y, Shao Y, Chen J, Xu S, Zhang X, Wang M & Liu H (2016) Effect of overexpression of HOX genes on its invasive tendency in cerebral glioma. Oncol. Lett. 11: 75–80 Hahn WC, Counter CM, Lundberg AS, Beijersbergen RL, Brooks MW & Weinberg RA (1999) Creation of human tumour cells with defined genetic elements. Nature 400: 464–468 Haley EM & Kim Y (2014) The role of basic fibroblast growth factor in glioblastoma multiforme and glioblastoma stem cells and in their in vitro culture. Cancer Lett. 346: 1–5 Halliday J, Helmy K, Pattwell SS, Pitter KL, LaPlant Q, Ozawa T & Holland EC (2014) In vivo radiation response of proneural glioma characterized by protective p53 transcriptional program and proneural-mesenchymal shift. Proc. Natl. Acad. Sci. U. S. A. 111: 5248–53 Hauri S, Comoglio F, Seimiya M, Gerstung M, Glatter T, Hansen K, Aebersold R, Paro R, Gstaiger M & Beisel C (2016) A High-Density Map for Navigating the Human Polycomb Complexome. Cell Rep. 17: 583–595 Hawkins RD, Hon GC, Lee LK, Ngo Q, Lister R, Pelizzola M, Edsall LE, Kuan S, Luu Y, Klugman S, Antosiewicz-Bourget J, Ye Z, Espinoza C, Agarwahl S, Shen L, Ruotti V, Wang W, Stewart R, Thomson JA, Ecker JR, et al (2010) Distinct Epigenomic Landscapes of Pluripotent and Lineage-Committed Human Cells. Cell Stem Cell 6: 479–491 He A, Ma Q, Cao J, von Gise A, Zhou P, Xie H, Zhang B, Hsing M, Christodoulou DC, Cahan P, Daley GQ, Kong SW, Orkin SH, Seidman CE, Seidman JG & Pu WT (2012) Polycomb Repressive Complex 2 Regulates Normal Development of the Mouse Heart. Circ. Res. 110: 406–415

195

Heard E & Martienssen RA (2014) Transgenerational epigenetic inheritance: myths and mechanisms. Cell 157: 95–109 Heinz S, Benner C, Spann N, Bertolino E, Lin YC, Laslo P, Cheng JX, Murre C, Singh H & Glass CK (2010) Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38: 576–89 Henser-Brownhill T, Monserrat J & Scaffidi P (2017) Generation of an arrayed CRISPR-Cas9 library targeting epigenetic regulators: from high-content screens to in vivo assays. Epigenetics 12: 1065–1075 Herait PE, Berthon C, Thieblemont C, Raffoux E, Magarotto V, Stathis A, Thomas X, Leleu X, Gomez-Roca C, Odore E, Roumier C, Bourdel F, Quesnel B, Zucca E, Michallet M, Recher C, Cvitkovic E, Rezai K, Preudhomme C, Facon T, et al (2014) Abstract CT231: BET-bromodomain inhibitor OTX015 shows clinically meaningful activity at nontoxic doses: interim results of an ongoing phase I trial in hematologic malignancies. Cancer Res. 74: CT231-CT231 Herranz N, Pasini D, Diaz VM, Franci C, Gutierrez A, Dave N, Escriva M, Hernandez- Munoz I, Di Croce L, Helin K, Garcia de Herreros A & Peiro S (2008) Polycomb Complex 2 Is Required for E-cadherin Repression by the Snail1 Transcription Factor. Mol. Cell. Biol. 28: 4772–4781 Hirabayashi Y, Suzki N, Tsuboi M, Endo TA, Toyoda T, Shinga J, Koseki H, Vidal M & Gotoh Y (2009) Polycomb Limits the Neurogenic Competence of Neural Precursor Cells to Promote Astrogenic Fate Transition. Neuron 63: 600–613 Hnisz D, Weintraub AS, Day DS, Valton A-L, Bak RO, Li CH, Goldmann J, Lajoie BR, Fan ZP, Sigova AA, Reddy J, Borges-Rivera D, Lee TI, Jaenisch R, Porteus MH, Dekker J & Young RA (2016) Activation of proto-oncogenes by disruption of chromosome neighborhoods. Science (80-. ). 351: 1454–1458 Hobbs GA, Der CJ & Rossman KL (2016) RAS isoforms and mutations in cancer at a glance. J. Cell Sci. 129: 1287–92 Hodis E, Watson IR, Kryukov GV, Arold ST, Imielinski M, Theurillat J-P, Nickerson E, Auclair D, Li L, Place C, DiCara D, Ramos AH, Lawrence MS, Cibulskis K, Sivachenko A, Voet D, Saksena G, Stransky N, Onofrio RC, Winckler W, et al (2012) A Landscape of Driver Mutations in Melanoma. Cell 150: 251–263 Holoch D & Margueron R (2017) Mechanisms Regulating PRC2 Recruitment and Enzymatic Activity. Trends Biochem. Sci. 42: 531–542 Hölzel M, Huang S, Koster J, Ora I, Lakeman A, Caron H, Nijkamp W, Xie J, Callens T,

196

Asgharzadeh S, Seeger RC, Messiaen L, Versteeg R & Bernards R (2010) NF1 is a tumor suppressor in neuroblastoma that determines retinoic acid response and disease outcome. Cell 142: 218–29 Hong S, Cho Y-W, Yu L-R, Yu H, Veenstra TD & Ge K (2007) Identification of JmjC domain-containing UTX and JMJD3 as histone H3 lysine 27 demethylases. Proc. Natl. Acad. Sci. 104: 18439–18444 Hosogane M, Funayama R, Nishida Y, Nagashima T & Nakayama K (2013) Ras- Induced Changes in H3K27me3 Occur after Those in Transcriptional Activity. PLoS Genet. 9: e1003698 Housden BE, Muhar M, Gemberling M, Gersbach CA, Stainier DYR, Seydoux G, Mohr SE, Zuber J & Perrimon N (2017) Loss-of-function genetic tools for animal models: cross-species and cross-platform differences. Nat. Rev. Genet. 18: 24–40 Hu D, Garruss AS, Gao X, Morgan MA, Cook M, Smith ER & Shilatifard A (2013) The Mll2 branch of the COMPASS family regulates bivalent promoters in mouse embryonic stem cells. Nat. Struct. Mol. Biol. 20: 1093–1097 Hunkapiller J, Shen Y, Diaz A, Cagney G, McCleary D, Ramalho-Santos M, Krogan N, Ren B, Song JS & Reiter JF (2012) Polycomb-Like 3 Promotes Polycomb Repressive Complex 2 Binding to CpG Islands and Embryonic Stem Cell Self- Renewal. PLoS Genet. 8: e1002576 Hussain M, Rao M, Humphries AE, Hong JA, Liu F, Yang M, Caragacianu D & Schrump DS (2009) Tobacco Smoke Induces Polycomb-Mediated Repression of Dickkopf-1 in Lung Cancer Cells. Cancer Res. 69: 3570–3578 Hwang WW, Salinas RD, Siu JJ, Kelley KW, Delgado RN, Paredes MF, Alvarez-Buylla A, Oldham MC & Lim DA (2014) Distinct and separable roles for EZH2 in neurogenic astroglia. Elife 3: e02439 Ingham PW (1983) Differential expression of bithorax complex genes in the absence of the extra sex combs and trithorax genes. Nature 306: 591–3 Ingham PW (1985) Genetic control of the spatial pattern of selector gene expression in Drosophila. Cold Spring Harb. Symp. Quant. Biol. 50: 201–8 Isono K, Endo TA, Ku M, Yamada D, Suzuki R, Sharif J, Ishikura T, Toyoda T, Bernstein BE & Koseki H (2013) SAM Domain Polymerization Links Subnuclear Clustering of PRC1 to Gene Silencing. Dev. Cell 26: 565–577 Italiano A, Soria J-C, Toulmonde M, Michot J-M, Lucchesi C, Varga A, Coindre J-M, Blakemore SJ, Clawson A, Suttle B, McDonald AA, Woodruff M, Ribich S, Hedrick E, Keilhack H, Thomson B, Owa T, Copeland RA, Ho PTC & Ribrag V (2018)

197

Tazemetostat, an EZH2 inhibitor, in relapsed or refractory B-cell non-Hodgkin lymphoma and advanced solid tumours: a first-in-human, open-label, phase 1 study. Lancet. Oncol. 19: 649–659 Jadhav U, Nalapareddy K, Saxena M, O’Neill NK, Pinello L, Yuan G-C, Orkin SH & Shivdasani RA (2016) Acquired Tissue-Specific Promoter Bivalency Is a Basis for PRC2 Necessity in Adult Cells. Cell 165: 1389–1400 Jakovcevski M & Akbarian S (2012) Epigenetic mechanisms in neurological disease. Nat. Med. 18: 1194–1204 Jia J, Parikh H, Xiao W, Hoskins JW, Pflicke H, Liu X, Collins I, Zhou W, Wang Z, Powell J, Thorgeirsson SS, Rudloff U, Petersen GM & Amundadottir LT (2013) An integrated transcriptome and epigenome analysis identifies a novel candidate gene for pancreatic cancer. BMC Med. Genomics 6: 33 Jin X, Kim LJY, Wu Q, Wallace LC, Prager BC, Sanvoranart T, Gimple RC, Wang X, Mack SC, Miller TE, Huang P, Valentim CL, Zhou Q, Barnholtz-Sloan JS, Bao S, Sloan AE & Rich JN (2017) Targeting glioma stem cells through combined BMI1 and EZH2 inhibition. Nat. Med. 23: 1352–1361 Jones C, Karajannis MA, Jones DTW, Kieran MW, Monje M, Baker SJ, Becher OJ, Cho Y-J, Gupta N, Hawkins C, Hargrave D, Haas-Kogan DA, Jabado N, Li X-N, Mueller S, Nicolaides T, Packer RJ, Persson AI, Phillips JJ, Simonds EF, et al (2016) Pediatric high-grade glioma: biologically and clinically in need of new thinking. Neuro. Oncol. 19: now101 Juan AH, Derfoul A, Feng X, Ryall JG, Dell’Orso S, Pasut A, Zare H, Simone JM, Rudnicki MA & Sartorelli V (2011) Polycomb EZH2 controls self-renewal and safeguards the transcriptional identity of skeletal muscle stem cells. Genes Dev. 25: 789–794 Juliandi B, Abematsu M & Nakashima K (2010) Epigenetic regulation in neural stem cell differentiation. Dev. Growth Differ. 52: 493–504 Kadoch C, Hargreaves DC, Hodges C, Elias L, Ho L, Ranish J & Crabtree GR (2013) Proteomic and bioinformatic analysis of mammalian SWI/SNF complexes identifies extensive roles in human malignancy. Nat. Genet. 45: 592–601 Kalb R, Latwiel S, Baymaz HI, Jansen PWTC, Müller CW, Vermeulen M & Müller J (2014) Histone H2A monoubiquitination promotes histone H3 methylation in Polycomb repression. Nat. Struct. Mol. Biol. 21: 569–571 Kamisawa T, Wood LD, Itoi T & Takaori K (2016) Pancreatic cancer. Lancet 388: 73– 85

198

Kandoth C, McLellan MD, Vandin F, Ye K, Niu B, Lu C, Xie M, Zhang Q, McMichael JF, Wyczalkowski MA, Leiserson MDM, Miller CA, Welch JS, Walter MJ, Wendl MC, Ley TJ, Wilson RK, Raphael BJ & Ding L (2013) Mutational landscape and significance across 12 major cancer types. Nature 502: 333–339 Kaneko S, Bonasio R, Saldaña-Meyer R, Yoshida T, Son J, Nishino K, Umezawa A & Reinberg D (2014a) Interactions between JARID2 and noncoding RNAs regulate PRC2 recruitment to chromatin. Mol. Cell 53: 290–300 Kaneko S, Son J, Bonasio R, Shen SS & Reinberg D (2014b) Nascent RNA interaction keeps PRC2 activity poised and in check. Genes Dev. 28: 1983–8 Kanhere A, Viiri K, Araújo CC, Rasaiyaah J, Bouwman RD, Whyte WA, Pereira CF, Brookes E, Walker K, Bell GW, Pombo A, Fisher AG, Young RA & Jenner RG (2010) Short RNAs Are Transcribed from Repressed Polycomb Target Genes and Interact with Polycomb Repressive Complex-2. Mol. Cell 38: 675–688 Karolchik D, Hinrichs AS, Furey TS, Roskin KM, Sugnet CW, Haussler D & Kent WJ (2004) The UCSC Table Browser data retrieval tool. Nucleic Acids Res. 32: 493D–496 Kennison JA & Tamkun JW (1988) Dosage-dependent modifiers of polycomb and antennapedia mutations in Drosophila. Proc. Natl. Acad. Sci. U. S. A. 85: 8136–40 Ketel CS, Andersen EF, Vargas ML, Suh J, Strome S & Simon JA (2005) Subunit Contributions to Histone Methyltransferase Activities of Fly and Worm Polycomb Group Complexes. Mol. Cell. Biol. 25: 6857–6868 Khuong-Quang D-A, Buczkowicz P, Rakopoulos P, Liu X-Y, Fontebasso AM, Bouffet E, Bartels U, Albrecht S, Schwartzentruber J, Letourneau L, Bourgey M, Bourque G, Montpetit A, Bourret G, Lepage P, Fleming A, Lichter P, Kool M, von Deimling A, Sturm D, et al (2012) K27M mutation in histone H3.3 defines clinically and biologically distinct subgroups of pediatric diffuse intrinsic pontine gliomas. Acta Neuropathol. 124: 439–447 Kim E, Kim M, Woo D-H, Shin Y, Shin J, Chang N, Oh YT, Kim H, Rheey J, Nakano I, Lee C, Joo KM, Rich JN, Nam D-H & Lee J (2013a) Phosphorylation of EZH2 Activates STAT3 Signaling via STAT3 Methylation and Promotes Tumorigenicity of Glioblastoma Stem-like Cells. Cancer Cell 23: 839–852 Kim H, Kang K & Kim J (2009) AEBP2 as a potential targeting protein for Polycomb Repression Complex PRC2. Nucleic Acids Res. 37: 2940–2950 Kim J, Lee Y, Lu X, Song B, Fong K-W, Cao Q, Licht JD, Zhao JC & Yu J (2018) Polycomb- and Methylation-Independent Roles of EZH2 as a Transcription

199

Activator. Cell Rep. 25: 2808–2820.e4 Kim KH, Kim W, Howard TP, Vazquez F, Tsherniak A, Wu JN, Wang W, Haswell JR, Walensky LD, Hahn WC, Orkin SH & Roberts CWM (2015) SWI/SNF-mutant cancers depend on catalytic and non-catalytic activity of EZH2. Nat. Med. 21: 1491–1496 Kim KH & Roberts CWM (2016) Targeting EZH2 in cancer. Nat. Med. 22: 128–34 Kim W, Bird GH, Neff T, Guo G, Kerenyi MA, Walensky LD & Orkin SH (2013b) Targeted Disruption of the EZH2/EED Complex Inhibits EZH2-dependent Cancer. Nat. Chem. Biol. 9: 643 Kleer CG, Cao Q, Varambally S, Shen R, Ota I, Tomlins SA, Ghosh D, Sewalt RGAB, Otte AP, Hayes DF, Sabel MS, Livant D, Weiss SJ, Rubin MA & Chinnaiyan AM (2003) EZH2 is a marker of aggressive breast cancer and promotes neoplastic transformation of breast epithelial cells. Proc. Natl. Acad. Sci. 100: 11606–11611 Kloet SL, Makowski MM, Baymaz HI, van Voorthuijsen L, Karemaker ID, Santanach A, Jansen PWTC, Di Croce L & Vermeulen M (2016) The dynamic interactome and genomic targets of Polycomb complexes during stem-cell differentiation. Nat. Struct. Mol. Biol. 23: 682–690 Klose RJ, Cooper S, Farcas AM, Blackledge NP & Brockdorff N (2013) Chromatin Sampling—An Emerging Perspective on Targeting Polycomb Repressor Proteins. PLoS Genet. 9: e1003717 Knutson SK, Warholic NM, Wigle TJ, Klaus CR, Allain CJ, Raimondi A, Porter Scott M, Chesworth R, Moyer MP, Copeland RA, Richon VM, Pollock RM, Kuntz KW & Keilhack H (2013a) Durable tumor regression in genetically altered malignant rhabdoid tumors by inhibition of methyltransferase EZH2. Proc. Natl. Acad. Sci. 110: 7922–7927 Knutson SK, Warholic NM, Wigle TJ, Klaus CR, Allain CJ, Raimondi A, Porter Scott M, Chesworth R, Moyer MP, Copeland RA, Richon VM, Pollock RM, Kuntz KW & Keilhack H (2013b) Durable tumor regression in genetically altered malignant rhabdoid tumors by inhibition of methyltransferase EZH2. Proc. Natl. Acad. Sci. U. S. A. 110: 7922–7 Knutson SK, Wigle TJ, Warholic NM, Sneeringer CJ, Allain CJ, Klaus CR, Sacks JD, Raimondi A, Majer CR, Song J, Scott MP, Jin L, Smith JJ, Olhava EJ, Chesworth R, Moyer MP, Richon VM, Copeland RA, Keilhack H, Pollock RM, et al (2012) A selective inhibitor of EZH2 blocks H3K27 methylation and kills mutant lymphoma cells. Nat. Chem. Biol. 8: 890–896

200

Koppens MAJ, Tanger E, Nacerddine K, Westerman B, Song J-Y & van Lohuizen M (2017) A new transgenic mouse model for conditional overexpression of the Polycomb Group protein EZH2. Transgenic Res. 26: 187–196 Ku M, Koche RP, Rheinbay E, Mendenhall EM, Endoh M, Mikkelsen TS, Presser A, Nusbaum C, Xie X, Chi AS, Adli M, Kasif S, Ptaszek LM, Cowan CA, Lander ES, Koseki H & Bernstein BE (2008) Genomewide Analysis of PRC1 and PRC2 Occupancy Identifies Two Classes of Bivalent Domains. PLoS Genet. 4: e1000242 Kumaki Y, Oda M & Okano M (2008) QUMA: quantification tool for methylation analysis. Nucleic Acids Res. 36: W170–W175 Kuzmichev A, Nishioka K, Erdjument-Bromage H, Tempst P & Reinberg D (2002) Histone methyltransferase activity associated with a human multiprotein complex containing the Enhancer of Zeste protein. Genes Dev. 16: 2893–2905 Laible G, Wolf A, Dorn R, Reuter G, Nislow C, Lebersorger A, Popkin D, Pillus L & Jenuwein T (1997) Mammalian homologues of the Polycomb-group gene Enhancer of zeste mediate gene silencing in Drosophila heterochromatin and at S. cerevisiae telomeres. EMBO J. 16: 3219–32 Landeira D & Fisher AG (2011) Inactive yet indispensable: the tale of Jarid2. Trends Cell Biol. 21: 74–80 LaPak KM & Burd CE (2014) The Molecular Balancing Act of p16 INK4a in Cancer and Aging. Mol. Cancer Res. 12: 167–183 Lau MS, Schwartz MG, Kundu S, Savol AJ, Wang PI, Marr SK, Grau DJ, Schorderet P, Sadreyev RI, Tabin CJ & Kingston RE (2017) Mutation of a nucleosome compaction region disrupts Polycomb-mediated axial patterning. Science (80-. ). 355: 1081–1084 Lee C-H, Holder M, Grau D, Saldaña-Meyer R, Yu J-R, Ganai RA, Zhang J, Wang M, LeRoy G, Dobenecker M-W, Reinberg D & Armache K-J (2018a) Distinct Stimulatory Mechanisms Regulate the Catalytic Activity of Polycomb Repressive Complex 2. Mol. Cell 70: 435–448.e5 Lee H-G, Kahn TG, Simcox A, Schwartz YB & Pirrotta V (2015) Genome-wide activities of Polycomb complexes control pervasive transcription. Genome Res. 25: 1170– 1181 Lee J, Son MJ, Woolard K, Donin NM, Li A, Cheng CH, Kotliarova S, Kotliarov Y, Walling J, Ahn S, Kim M, Totonchy M, Cusack T, Ene C, Ma H, Su Q, Zenklusen JC, Zhang W, Maric D & Fine HA (2008) Epigenetic-Mediated Dysfunction of the

201

Bone Morphogenetic Protein Pathway Inhibits Differentiation of Glioblastoma- Initiating Cells. Cancer Cell 13: 69–80 Lee JH, Lee JE, Kahng JY, Kim SH, Park JS, Yoon SJ, Um J-Y, Kim WK, Lee J-K, Park J, Kim EH, Lee J-H, Lee J-H, Chung W-S, Ju YS, Park S-H, Chang JH, Kang S-G & Lee JH (2018b) Human glioblastoma arises from subventricular zone cells with low-level driver mutations. Nature 560: 243–247 Lee ST, Li Z, Wu Z, Aau M, Guan P, Karuturi RKM, Liou YC & Yu Q (2011) Context- Specific Regulation of NF-κB Target Gene Expression by EZH2 in Breast Cancers. Mol. Cell 43: 798–810 Lee TI, Jenner RG, Boyer LA, Guenther MG, Levine SS, Kumar RM, Chevalier B, Johnstone SE, Cole MF, Isono K, Koseki H, Fuchikami T, Abe K, Murray HL, Zucker JP, Yuan B, Bell GW, Herbolsheimer E, Hannett NM, Sun K, et al (2006) Control of Developmental Regulators by Polycomb in Human Embryonic Stem Cells. Cell 125: 301–313 Lee W, Teckie S, Wiesner T, Ran L, Prieto Granada CN, Lin M, Zhu S, Cao Z, Liang Y, Sboner A, Tap WD, Fletcher JA, Huberman KH, Qin L-X, Viale A, Singer S, Zheng D, Berger MF, Chen Y, Antonescu CR, et al (2014) PRC2 is recurrently inactivated through EED or SUZ12 loss in malignant peripheral nerve sheath tumors. Nat. Genet. 46: 1227–1232 Lewis EB (1978) A gene complex controlling segmentation in Drosophila. Nature 276: 565–70 Lewis PH (1947) New mutants report. Dros. Inf. Serv: 21, 69 Lewis PW, Muller MM, Koletsky MS, Cordero F, Lin S, Banaszynski LA, Garcia BA, Muir TW, Becher OJ & Allis CD (2013) Inhibition of PRC2 Activity by a Gain-of- Function H3 Mutation Found in Pediatric Glioblastoma. Science (80-. ). 340: 857– 861 Li B & Dewey CN (2011) RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics 12: 323 Li D, Song H, Mei H, Fang E, Wang X, Yang F, Li H, Chen Y, Huang K, Zheng L & Tong Q (2018a) Armadillo repeat containing 12 promotes neuroblastoma progression through interaction with retinoblastoma binding protein 4. Nat. Commun. 9: 2829 Li G, Margueron R, Ku M, Chambon P, Bernstein BE & Reinberg D (2010) Jarid2 and PRC2, partners in regulating gene expression. Genes Dev. 24: 368–80 Li G, Wang D, Ma W, An K, Liu Z, Wang X, Yang C, Du F, Han X, Chang S, Yu H,

202

Zhang Z, Zhao Z, Zhang Y, Wang J & Sun Y (2018b) Transcriptomic and epigenetic analysis of breast cancer stem cells. Epigenomics 10: 765–783 Li G, Warden C, Zou Z, Neman J, Krueger JS, Jain A, Jandial R & Chen M (2013) Altered Expression of Polycomb Group Genes in Glioblastoma Multiforme. PLoS One 8: e80970 Li H & Durbin R (2009) Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25: 1754–1760 Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, Marth G, Abecasis G, Durbin R & 1000 Genome Project Data Processing Subgroup 1000 Genome Project Data Processing (2009a) The Sequence Alignment/Map format and SAMtools. Bioinformatics 25: 2078–9 Li H, Liefke R, Jiang J, Kurland JV, Tian W, Deng P, Zhang W, He Q, Patel DJ, Bulyk ML, Shi Y & Wang Z (2017) Polycomb-like proteins link the PRC2 complex to CpG islands. Nature 549: 287–291 Li J, Mo M, Chen Z, Chen Z, Sheng Q, Mu H, Zhang F, Zhang Y, Zhi X-Y, Li H, He B & Zhou H-M (2012) Adenoviral Delivery of the EMX2 Gene Suppresses Growth in Human Gastric Cancer. PLoS One 7: e45970 Li L-C & Dahiya R (2002) MethPrimer: designing primers for methylation PCRs. Bioinformatics 18: 1427–1431 Li X, Gonzalez ME, Toy K, Filzen T, Merajver SD & Kleer CG (2009b) Targeted Overexpression of EZH2 in the Mammary Gland Disrupts Ductal Morphogenesis and Causes Epithelial Hyperplasia. Am. J. Pathol. 175: 1246–1254 Li X, Isono K -i., Yamada D, Endo TA, Endoh M, Shinga J, Mizutani-Koseki Y, Otte AP, Casanova M, Kitamura H, Kamijo T, Sharif J, Ohara O, Toyada T, Bernstein BE, Brockdorff N & Koseki H (2011) Mammalian Polycomb-Like Pcl2/Mtf2 Is a Novel Regulatory Component of PRC2 That Can Differentially Modulate Polycomb Activity both at the Hox Gene Cluster and at Cdkn2a Genes. Mol. Cell. Biol. 31: 351–364 Liefke R, Karwacki-Neisius V & Shi Y (2016) EPOP Interacts with Elongin BC and USP7 to Modulate the Chromatin Landscape. Mol. Cell 64: 659–672 Lin B, Lee H, Yoon J-G, Madan A, Wayner E, Tonning S, Hothi P, Schroeder B, Ulasov I, Foltz G, Hood L & Cobbs C (2015) Global analysis of H3K4me3 and H3K27me3 profiles in glioblastoma stem cells and identification of SLC17A7 as a bivalent tumor suppressor gene. Oncotarget 6: 5369–81 Lister R, Mukamel EA, Nery JR, Urich M, Puddifoot CA, Johnson ND, Lucero J, Huang

203

Y, Dwork AJ, Schultz MD, Yu M, Tonti-Filippini J, Heyn H, Hu S, Wu JC, Rao A, Esteller M, He C, Haghighi FG, Sejnowski TJ, et al (2013) Global Epigenomic Reconfiguration During Mammalian Brain Development. Science (80-. ). 341: 1237905–1237905 Liu R, Gao J, Yang Y, Qiu R, Zheng Y, Huang W, Zeng Y, Hou Y, Wang S, Leng S, Feng D, Yu W, Sun G, Shi H, Teng X & Wang Y (2018) PHD finger protein 1 (PHF1) is a novel reader for histone H4R3 symmetric dimethylation and coordinates with PRMT5–WDR77/CRL4B complex to promote tumorigenesis. Nucleic Acids Res. 46: 6608–6626 van Lohuizen M, Tijms M, Voncken JW, Schumacher A, Magnuson T & Wientjens E (1998) Interaction of mouse polycomb-group (Pc-G) proteins Enx1 and Enx2 with Eed: indication for separate Pc-G complexes. Mol. Cell. Biol. 18: 3572–9 Loilome W, Joshi AD, ap Rhys CMJ, Piccirillo S, Angelo VL, Gallia GL, Riggins GJ & Riggins GJ (2009) Glioblastoma cell growth is suppressed by disruption of fibroblast growth factor pathway signaling. J. Neurooncol. 94: 359–366 Long HK, Prescott SL & Wysocka J (2016) Ever-Changing Landscapes: Transcriptional Enhancers in Development and Evolution. Cell 167: 1170–1187 Love MI, Huber W & Anders S (2014) Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15: 550 Lovén J, Orlando DA, Sigova AA, Lin CY, Rahl PB, Burge CB, Levens DL, Lee TI & Young RA (2012) Revisiting global gene expression analysis. Cell 151: 476–82 Lu C, Han HD, Mangala LS, Ali-Fehmi R, Newton CS, Ozbun L, Armaiz-Pena GN, Hu W, Stone RL, Munkarah A, Ravoori MK, Shahzad MMK, Lee J-W, Mora E, Langley RR, Carroll AR, Matsuo K, Spannuth WA, Schmandt R, Jennings NB, et al (2010) Regulation of Tumor Angiogenesis by EZH2. Cancer Cell 18: 185–197 Lu C, Jain SU, Hoelper D, Bechet D, Molden RC, Ran L, Murphy D, Venneti S, Hameed M, Pawel BR, Wunder JS, Dickson BC, Lundgren SM, Jani KS, De Jay N, Papillon-Cavanagh S, Andrulis IL, Sawyer SL, Grynspan D, Turcotte RE, et al (2016) Histone H3K36 mutations promote sarcomagenesis through altered histone methylation landscape. Science (80-. ). 352: 844–849 Lui JC, Garrison P, Nguyen Q, Ad M, Keembiyehetty C, Chen W, Jee YH, Landman E, Nilsson O, Barnes KM & Baron J (2016) EZH1 and EZH2 promote skeletal growth by repressing inhibitors of chondrocyte proliferation and hypertrophy. Nat. Commun. 7: 13685 Lynch MD, Smith AJH, De Gobbi M, Flenley M, Hughes JR, Vernimmen D, Ayyub H,

204

Sharpe JA, Sloane-Stanley JA, Sutherland L, Meek S, Burdon T, Gibbons RJ, Garrick D & Higgs DR (2012) An interspecies analysis reveals a key role for unmethylated CpG dinucleotides in vertebrate Polycomb complex recruitment. EMBO J. 31: 317–329 Ma Y, Zhang P, Wang F, Yang J, Yang Z & Qin H (2010) The relationship between early embryo development and tumourigenesis. J. Cell. Mol. Med. 14: 2697–701 Malta TM, de Souza CF, Sabedot TS, Silva TC, Mosella MS, Kalkanis SN, Snyder J, Castro AVB & Noushmehr H (2018) Glioma CpG island methylator phenotype (G- CIMP): biological and clinical implications. Neuro. Oncol. 20: 608–620 Mao H, Lebrun DG, Yang J, Zhu VF & Li M (2012) Deregulated signaling pathways in glioblastoma multiforme: molecular mechanisms and therapeutic targets. Cancer Invest. 30: 48–56 Margueron R, Justin N, Ohno K, Sharpe ML, Son J, Drury III WJ, Voigt P, Martin SR, Taylor WR, De Marco V, Pirrotta V, Reinberg D & Gamblin SJ (2009) Role of the polycomb protein EED in the propagation of repressive histone marks. Nature 461: 762–767 Margueron R, Li G, Sarma K, Blais A, Zavadil J, Woodcock CL, Dynlacht BD & Reinberg D (2008) Ezh1 and Ezh2 Maintain Repressive Chromatin through Different Mechanisms. Mol. Cell 32: 503–518 Margueron R & Reinberg D (2011) The Polycomb complex PRC2 and its mark in life. Nature 469: 343–349 Martin M (2011) Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet.journal 17: 10 Mathur R & Roberts CWM (2018) SWI/SNF (BAF) Complexes: Guardians of the Epigenome. Annu. Rev. Cancer Biol. 2: 413–427 Matthay KK, Maris JM, Schleiermacher G, Nakagawara A, Mackall CL, Diller L & Weiss WA (2016) Neuroblastoma. Nat. Rev. Dis. Prim. 2: 16078 Mazzoni EO, Mahony S, Peljto M, Patel T, Thornton SR, McCuine S, Reeder C, Boyer LA, Young RA, Gifford DK & Wichterle H (2013) Saltatory remodeling of Hox chromatin in response to rostrocaudal patterning signals. Nat. Neurosci. 16: 1191–1198 McCabe MT, Ott HM, Ganji G, Korenchuk S, Thompson C, Van Aller GS, Liu Y, Graves AP, III ADP, Diaz E, LaFrance L V., Mellinger M, Duquenne C, Tian X, Kruger RG, McHugh CF, Brandt M, Miller WH, Dhanak D, Verma SK, et al (2012) EZH2 inhibition as a therapeutic strategy for lymphoma with EZH2-activating

205

mutations. Nature 492: 108–112 McHugh CA, Chen C-K, Chow A, Surka CF, Tran C, McDonel P, Pandya-Jones A, Blanco M, Burghard C, Moradian A, Sweredoski MJ, Shishkin AA, Su J, Lander ES, Hess S, Plath K & Guttman M (2015) The Xist lncRNA interacts directly with SHARP to silence transcription through HDAC3. Nature 521: 232–236 Meeks JJ & Shilatifard A (2017) Multiple Roles for the MLL/COMPASS Family in the Epigenetic Regulation of Gene Expression and in Cancer. Annu. Rev. Cancer Biol. 1: 425–446 Di Meglio T, Kratochwil CF, Vilain N, Loche A, Vitobello A, Yonehara K, Hrycaj SM, Roska B, Peters AHFM, Eichmann A, Wellik D, Ducret S & Rijli FM (2013) Ezh2 Orchestrates Topographic Migration and Connectivity of Mouse Precerebellar Neurons. Science (80-. ). 339: 204–207 Mendenhall EM, Koche RP, Truong T, Zhou VW, Issac B, Chi AS, Ku M & Bernstein BE (2010) GC-Rich Sequence Elements Recruit PRC2 in Mammalian ES Cells. PLoS Genet. 6: e1001244 Migliori V, Mapelli M & Guccione E (2012) On WD40 proteins: Propelling our knowledge of transcriptional control? Epigenetics 7: 815–822 Mohammad F, Weissmann S, Leblanc B, Pandey DP, Højfeldt JW, Comet I, Zheng C, Johansen JV, Rapin N, Porse BT, Tvardovskiy A, Jensen ON, Olaciregui NG, Lavarino C, Suñol M, de Torres C, Mora J, Carcaboso AM & Helin K (2017) EZH2 is a potential therapeutic target for H3K27M-mutant pediatric gliomas. Nat. Med. 23: 483–492 Monnier A, Boniface R, Bouvet R, Etcheverry A, Aubry M, Avril T, Quillien V, Chevet E & Mosser J (2018) The expression of EMX2 lead to cell cycle arrest in glioblastoma cell line. BMC Cancer 18: 1213 Montavon T & Duboule D (2013) Chromatin organization and global regulation of Hox gene clusters. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 368: 20120367 Moran-Crusio K, Reavie L, Shih A, Abdel-Wahab O, Ndiaye-Lobry D, Lobry C, Figueroa ME, Vasanthakumar A, Patel J, Zhao X, Perna F, Pandey S, Madzo J, Song C, Dai Q, He C, Ibrahim S, Beran M, Zavadil J, Nimer SD, et al (2011) Tet2 Loss Leads to Increased Hematopoietic Stem Cell Self-Renewal and Myeloid Transformation. Cancer Cell 20: 11–24 Morin RD, Johnson NA, Severson TM, Mungall AJ, An J, Goya R, Paul JE, Boyle M, Woolcock BW, Kuchenbauer F, Yap D, Humphries RK, Griffith OL, Shah S, Zhu H, Kimbara M, Shashkin P, Charlot JF, Tcherpakov M, Corbett R, et al (2010)

206

Somatic mutations altering EZH2 (Tyr641) in follicular and diffuse large B-cell lymphomas of germinal-center origin. Nat. Genet. 42: 181–185 Müller J, Hart CM, Francis NJ, Vargas ML, Sengupta A, Wild B, Miller EL, O’Connor MB, Kingston RE & Simon JA (2002) Histone methyltransferase activity of a Drosophila Polycomb group repressor complex. Cell 111: 197–208 Musselman CA, Avvakumov N, Watanabe R, Abraham CG, Lalonde M-E, Hong Z, Allen C, Roy S, Nuñez JK, Nickoloff J, Kulesza CA, Yasui A, Côté J & Kutateladze TG (2012) Molecular basis for H3K36me3 recognition by the Tudor domain of PHF1. Nat. Struct. Mol. Biol. 19: 1266–72 Nacev BA, Feng L, Bagert JD, Lemiesz AE, Gao J, Soshnev AA, Kundra R, Schultz N, Muir TW & Allis CD (2019) The expanding landscape of ‘oncohistone’ mutations in human cancers. Nature 567: 473–478 Natsume A, Ito M, Katsushima K, Ohka F, Hatanaka A, Shinjo K, Sato S, Takahashi S, Ishikawa Y, Takeuchi I, Shimogawa H, Uesugi M, Okano H, Kim SU, Wakabayashi T, Issa J-PJ, Sekido Y & Kondo Y (2013) Chromatin Regulator PRC2 Is a Key Regulator of Epigenetic Plasticity in Glioblastoma. Cancer Res. 73: 4559–4570 Nekrasov M, Wild B & Müller J (2005) Nucleosome binding and histone methyltransferase activity of Drosophila PRC2. EMBO Rep. 6: 348–353 Neri F, Incarnato D, Krepelova A, Rapelli S, Pagnani A, Zecchina R, Parlato C & Oliviero S (2013a) Genome-wide analysis identifies a functional association of Tet1 and Polycomb repressive complex 2 in mouse embryonic stem cells. Genome Biol. 14: R91 Neri F, Krepelova A, Incarnato D, Maldotti M, Parlato C, Galvagni F, Matarese F, Stunnenberg HG & Oliviero S (2013b) Dnmt3L Antagonizes DNA Methylation at Bivalent Promoters and Favors DNA Methylation at Gene Bodies in ESCs. Cell 155: 121–134 Ngollo M, Lebert A, Daures M, Judes G, Rifai K, Dubois L, Kemeny J-L, Penault-Llorca F, Bignon Y-J, Guy L & Bernard-Gallon D (2017) Global analysis of H3K27me3 as an epigenetic marker in prostate cancer progression. BMC Cancer 17: 261 Nikoloski G, Langemeijer SMC, Kuiper RP, Knops R, Massop M, Tönnissen ERLTM, van der Heijden A, Scheele TN, Vandenberghe P, de Witte T, van der Reijden BA & Jansen JH (2010) Somatic mutations of the histone methyltransferase gene EZH2 in myelodysplastic syndromes. Nat. Genet. 42: 665–667 Nolte C & Krumlauf R (2013) Expression of Hox Genes in the Nervous System of

207

Vertebrates. Nowak AJ, Alfieri C, Stirnimann CU, Rybin V, Baudin F, Ly-Hartig N, Lindner D & Müller CW (2011) Chromatin-modifying Complex Component Nurf55/p55 Associates with Histones H3 and H4 and Polycomb Repressive Complex 2 Subunit Su(z)12 through Partially Overlapping Binding Sites. J. Biol. Chem. 286: 23388–23396 Ntziachristos P, Tsirigos A, Vlierberghe P Van, Nedjic J, Trimarchi T, Flaherty MS, Ferres-Marco D, da Ros V, Tang Z, Siegle J, Asp P, Hadler M, Rigo I, Keersmaecker K De, Patel J, Huynh T, Utro F, Poglio S, Samon JB, Paietta E, et al (2012) Genetic inactivation of the polycomb repressive complex 2 in T cell acute lymphoblastic leukemia. Nat. Med. 18: 298–302 O’Carroll D, Erhardt S, Pagani M, Barton SC, Surani MA & Jenuwein T (2001) The Polycomb-Group Gene Ezh2 Is Required for Early Mouse Development. Mol. Cell. Biol. 21: 4330–4336 O’Loghlen A, Muñoz-Cabello AM, Gaspar-Maia A, Wu H-A, Banito A, Kunowska N, Racek T, Pemberton HN, Beolchi P, Lavial F, Masui O, Vermeulen M, Carroll T, Graumann J, Heard E, Dillon N, Azuara V, Snijders AP, Peters G, Bernstein E, et al (2012) MicroRNA Regulation of Cbx7 Mediates a Switch of Polycomb Orthologs during ESC Differentiation. Cell Stem Cell 10: 33–46 Ogawa M, Hiraoka Y, Taniguchi K & Aiso S (1998) Cloning and expression of a human/mouse Polycomb group gene, ENX-2/Enx-2. Biochim. Biophys. Acta - Gene Struct. Expr. 1395: 151–158 Ogawa Y, Sun BK, Lee JT, Song J-J & Lee JT (2008) Intersection of the RNA interference and X-inactivation pathways. Science 320: 1336–41 Ohm JE, McGarvey KM, Yu X, Cheng L, Schuebel KE, Cope L, Mohammad HP, Chen W, Daniel VC, Yu W, Berman DM, Jenuwein T, Pruitt K, Sharkis SJ, Watkins DN, Herman JG & Baylin SB (2007) A stem cell–like chromatin pattern may predispose tumor suppressor genes to DNA hypermethylation and heritable silencing. Nat. Genet. 39: 237–242 Okamoto J, Hirata T, Chen Z, Zhou H-M, Mikami I, Li H, Beltran A, Johansson M, Coussens LM, Clement G, Shi Y, Zhang F, Koizumi K, Shimizu K, Jablons D, He B & He B (2010) EMX2 is epigenetically silenced and suppresses growth in human lung cancer. Oncogene 29: 5969–5975 Okano M, Bell DW, Haber DA & Li E (1999) DNA methyltransferases Dnmt3a and Dnmt3b are essential for de novo methylation and mammalian development. Cell

208

99: 247–57 Oliviero G, Brien GL, Waston A, Streubel G, Jerman E, Andrews D, Doyle B, Munawar N, Wynne K, Crean J, Bracken AP & Cagney G (2016) Dynamic Protein Interactions of the Polycomb Repressive Complex 2 during Differentiation of Pluripotent Cells. Mol. Cell. Proteomics 15: 3450 Oricchio E, Katanayeva N, Donaldson MC, Sungalee S, Pasion JP, Béguelin W, Battistello E, Sanghvi VR, Jiang M, Jiang Y, Teater M, Parmigiani A, Budanov A V., Chan FC, Shah SP, Kridel R, Melnick AM, Ciriello G & Wendel H-G (2017) Genetic and epigenetic inactivation of SESTRIN1 controls mTORC1 and response to EZH2 inhibition in follicular lymphoma. Sci. Transl. Med. 9: eaak9969 Orzan F, Pellegatta S, Poliani PL, Pisati F, Caldera V, Menghi F, Kapetis D, Marras C, Schiffer D & Finocchiaro G (2011) Enhancer of Zeste 2 (EZH2) is up-regulated in malignant gliomas and in glioma stem-like cells. Neuropathol. Appl. Neurobiol. 37: 381–394 Pasini D, Bracken AP, Hansen JB, Capillo M & Helin K (2007) The Polycomb Group Protein Suz12 Is Required for Embryonic Stem Cell Differentiation. Mol. Cell. Biol. 27: 3769–3779 Pasini D, Bracken AP, Jensen MR, Lazzerini Denchi E & Helin K (2004) Suz12 is essential for mouse development and for EZH2 histone methyltransferase activity. EMBO J. 23: 4061–71 Pasini D, Cloos PAC, Walfridsson J, Olsson L, Bukowski J-P, Johansen J V., Bak M, Tommerup N, Rappsilber J & Helin K (2010) JARID2 regulates binding of the Polycomb repressive complex 2 to target genes in ES cells. Nature 464: 306–310 Pengelly AR, Copur Ö, Jäckle H, Herzig A & Müller J (2013) A histone mutant reproduces the phenotype caused by loss of histone-modifying factor Polycomb. Science 339: 698–9 Pereira JD, Sansom SN, Smith J, Dobenecker M-W, Tarakhovsky A & Livesey FJ (2010) Ezh2, the histone methyltransferase of PRC2, regulates the balance between self-renewal and differentiation in the cerebral cortex. Proc. Natl. Acad. Sci. U. S. A. 107: 15957–62 Perino M, van Mierlo G, Karemaker ID, van Genesen S, Vermeulen M, Marks H, van Heeringen SJ & Veenstra GJC (2018) MTF2 recruits Polycomb Repressive Complex 2 by helical-shape-selective DNA binding. Nat. Genet. 50: 1002–1010 Popovic R, Martinez-Garcia E, Giannopoulou EG, Zhang Q, Zhang Q, Ezponda T, Shah MY, Zheng Y, Will CM, Small EC, Hua Y, Bulic M, Jiang Y, Carrara M,

209

Calogero RA, Kath WL, Kelleher NL, Wang J-P, Elemento O & Licht JD (2014) Histone Methyltransferase MMSET/NSD2 Alters EZH2 Binding and Reprograms the Myeloma Epigenome through Global and Focal Changes in H3K36 and H3K27 Methylation. PLoS Genet. 10: e1004566 Portoso M, Ragazzini R, Brenčič Ž, Moiani A, Michaud A, Vassilev I, Wassef M, Servant N, Sargueil B & Margueron R (2017) PRC2 is dispensable for HOTAIR ‐ mediated transcriptional repression. EMBO J. 36: 981–994 Puchalski RB, Shah N, Miller J, Dalley R, Nomura SR, Yoon J-G, Smith KA, Lankerovich M, Bertagnolli D, Bickley K, Boe AF, Brouner K, Butler S, Caldejon S, Chapin M, Datta S, Dee N, Desta T, Dolbeare T, Dotson N, et al (2018) An anatomic transcriptional atlas of human glioblastoma. Science (80-. ). 360: 660– 663 Qi W, Chan H, Teng L, Li L, Chuai S, Zhang R, Zeng J, Li M, Fan H, Lin Y, Gu J, Ardayfio O, Zhang J-H, Yan X, Fang J, Mi Y, Zhang M, Zhou T, Feng G, Chen Z, et al (2012) Selective inhibition of Ezh2 by a small molecule inhibitor blocks tumor cells proliferation. Proc. Natl. Acad. Sci. 109: 21360–21365 Qi W, Zhao K, Gu J, Huang Y, Wang Y, Zhang H, Zhang M, Zhang J, Yu Z, Li L, Teng L, Chuai S, Zhang C, Zhao M, Chan H, Chen Z, Fang D, Fei Q, Feng L, Feng L, et al (2017) An allosteric PRC2 inhibitor targeting the H3K27me3 binding pocket of EED. Nat. Chem. Biol. 13: 381–388 Qiu H, Yan Q, Luo X, Zhang H, Bao W & Wan X (2013) EMX2 is Downregulated in Endometrial Cancer and Correlated With Tumor Progression. Int. J. Gynecol. Pathol. 32: 193–198 Quinlan AR & Hall IM (2010) BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26: 841–842 De Raedt T, Beert E, Pasmant E, Luscan A, Brems H, Ortonne N, Helin K, Hornick JL, Mautner V, Kehrer-Sawatzki H, Clapp W, Bradner J, Vidaud M, Upadhyaya M, Legius E & Cichowski K (2014) PRC2 loss amplifies Ras-driven transcription and confers sensitivity to BRD4-based therapies. Nature 514: 247–251 Ran FA, Hsu PD, Wright J, Agarwala V, Scott DA & Zhang F (2013) Genome engineering using the CRISPR-Cas9 system. Nat. Protoc. 8: 2281–2308 Reddington JP, Perricone SM, Nestor CE, Reichmann J, Youngson NA, Suzuki M, Reinhardt D, Dunican DS, Prendergast JG, Mjoseng H, Ramsahoye BH, Whitelaw E, Greally JM, Adams IR, Bickmore WA & Meehan RR (2013) Redistribution of H3K27me3 upon DNA hypomethylation results in de-repression of Polycomb

210

target genes. Genome Biol. 14: R25 Rheinbay E, Suvà ML, Gillespie SM, Wakimoto H, Patel AP, Shahid M, Oksuz O, Rabkin SD, Martuza RL, Rivera MN, Louis DN, Kasif S, Chi AS & Bernstein BE (2013) An Aberrant Transcription Factor Network Essential for Wnt Signaling and Stem Cell Maintenance in Glioblastoma. Cell Rep. 3: 1567–1579 Rhodes CT, Zunino G, Huang S-WA, Cardona SM, Cardona AE, Berger MS, Lemmon VP & Lin C-HA (2018) Region specific knock-out reveals distinct roles of chromatin modifiers in adult neurogenic niches. Cell Cycle 17: 377–389 Riising EM, Comet I, Leblanc B, Wu X, Johansen JV & Helin K (2014) Gene Silencing Triggers Polycomb Repressive Complex 2 Recruitment to CpG Islands Genome Wide. Mol. Cell 55: 347–360 Rinn JL, Kertesz M, Wang JK, Squazzo SL, Xu X, Brugmann SA, Goodnough LH, Helms JA, Farnham PJ, Segal E & Chang HY (2007) Functional Demarcation of Active and Silent Chromatin Domains in Human HOX Loci by Noncoding RNAs. Cell 129: 1311–1323 Robinson G, Parker M, Kranenburg TA, Lu C, Chen X, Ding L, Phoenix TN, Hedlund E, Wei L, Zhu X, Chalhoub N, Baker SJ, Huether R, Kriwacki R, Curley N, Thiruvenkatam R, Wang J, Wu G, Rusch M, Hong X, et al (2012) Novel mutations target distinct subgroups of medulloblastoma. Nature 488: 43–8 Rose NR & Klose RJ (2014) Understanding the relationship between DNA methylation and histone lysine methylation. Biochim. Biophys. Acta - Gene Regul. Mech. 1839: 1362–1372 Ross-Innes CS, Stark R, Teschendorff AE, Holmes KA, Ali HR, Dunning MJ, Brown GD, Gojis O, Ellis IO, Green AR, Ali S, Chin S-F, Palmieri C, Caldas C & Carroll JS (2012) Differential oestrogen receptor binding is associated with clinical outcome in breast cancer. Nature 481: 389–393 Sanulli S, Justin N, Teissandier A, Ancelin K, Portoso M, Caron M, Michaud A, Lombard B, da Rocha ST, Offer J, Loew D, Servant N, Wassef M, Burlina F, Gamblin SJ, Heard E & Margueron R (2015) Jarid2 Methylation via the PRC2 Complex Regulates H3K27me3 Deposition during Cell Differentiation. Mol. Cell 57: 769–783 Sashida G, Harada H, Matsui H, Oshima M, Yui M, Harada Y, Tanaka S, Mochizuki- Kashio M, Wang C, Saraya A, Muto T, Hayashi Y, Suzuki K, Nakajima H, Inaba T, Koseki H, Huang G, Kitamura T & Iwama A (2014) Ezh2 loss promotes development of myelodysplastic syndrome but attenuates its predisposition to

211

leukaemic transformation. Nat. Commun. 5: 4177 Scaffidi P & Misteli T (2011) In vitro generation of human cells with cancer stem cell properties. Nat. Cell Biol. 13: 1051–61 von Schimmelmann M, Feinberg PA, Sullivan JM, Ku SM, Badimon A, Duff MK, Wang Z, Lachmann A, Dewell S, Ma’ayan A, Han M-H, Tarakhovsky A & Schaefer A (2016) Polycomb repressive complex 2 (PRC2) silences genes responsible for neurodegeneration. Nat. Neurosci. 19: 1321–1330 Schlesinger Y, Straussman R, Keshet I, Farkash S, Hecht M, Zimmerman J, Eden E, Yakhini Z, Ben-Shushan E, Reubinoff BE, Bergman Y, Simon I & Cedar H (2007) Polycomb-mediated methylation on Lys27 of histone H3 pre-marks genes for de novo methylation in cancer. Nat. Genet. 39: 232–236 Schmitges FW, Prusty AB, Faty M, Stützer A, Lingaraju GM, Aiwazian J, Sack R, Hess D, Li L, Zhou S, Bunker RD, Wirth U, Bouwmeester T, Bauer A, Ly-Hartig N, Zhao K, Chan H, Gu J, Gut H, Fischle W, et al (2011) Histone Methylation by PRC2 Is Inhibited by Active Chromatin Marks. Mol. Cell 42: 330–341 Schuettengruber B, Bourbon H-M, Di Croce L & Cavalli G (2017) Genome Regulation by Polycomb and Trithorax: 70 Years and Counting. Cell 171: 34–57 Schwartzentruber J, Korshunov A, Liu X-Y, Jones DTW, Pfaff E, Jacob K, Sturm D, Fontebasso AM, Quang D-AK, Tönjes M, Hovestadt V, Albrecht S, Kool M, Nantel A, Konermann C, Lindroth A, Jäger N, Rausch T, Ryzhova M, Korbel JO, et al (2012) Driver mutations in histone H3.3 and chromatin remodelling genes in paediatric glioblastoma. Nature 482: 226–231 Score J, Hidalgo-Curtis C, Jones A V., Winkelmann N, Skinner A, Ward D, Zoi K, Ernst T, Stegelmann F, Dohner K, Chase A & Cross NCP (2012) Inactivation of polycomb repressive complex 2 components in myeloproliferative and myelodysplastic/myeloproliferative neoplasms. Blood 119: 1208–1213 Shah V, Drill E & Lance-Jones C (2004) Ectopic expression of Hoxd10 in thoracic spinal segments induces motoneurons with a lumbosacral molecular profile and axon projections to the limb. Dev. Dyn. 231: 43–56 Shao Z, Raible F, Mollaaghababa R, Guyon JR, Wu C, Bender W & Kingston RE (1999) Stabilization of Chromatin Structure by PRC1, a Polycomb Complex. Cell 98: 37–46 Sharma S V., Lee DY, Li B, Quinlan MP, Takahashi F, Maheswaran S, McDermott U, Azizian N, Zou L, Fischbach MA, Wong K-K, Brandstetter K, Wittner B, Ramaswamy S, Classon M & Settleman J (2010) A Chromatin-Mediated

212

Reversible Drug-Tolerant State in Cancer Cell Subpopulations. Cell 141: 69–80 Shen H & Laird PW (2013) Interplay between the cancer genome and epigenome. Cell 153: 38–55 Shen L, Shao N, Liu X & Nestler E (2014) ngs.plot: Quick mining and visualization of next-generation sequencing data by integrating genomic databases. BMC Genomics 15: 284 Shen X, Liu Y, Hsu Y-J, Fujiwara Y, Kim J, Mao X, Yuan G-C & Orkin SH (2008) EZH1 Mediates Methylation on Histone H3 Lysine 27 and Complements EZH2 in Maintaining Stem Cell Identity and Executing Pluripotency. Mol. Cell 32: 491–502 Sher F, Boddeke E, Olah M & Copray S (2012) Dynamic Changes in Ezh2 Gene Occupancy Underlie Its Involvement in Neural Stem Cell Self-Renewal and Differentiation towards Oligodendrocytes. PLoS One 7: e40399 Shi J, Wang Y, Zeng L, Wu Y, Deng J, Zhang Q, Lin Y, Li J, Kang T, Tao M, Rusinova E, Zhang G, Wang C, Zhu H, Yao J, Zeng Y-X, Evers BM, Zhou M-M & Zhou BP (2014) Disrupting the Interaction of BRD4 with Diacetylated Twist Suppresses Tumorigenesis in Basal-like Breast Cancer. Cancer Cell 25: 210–225 Shi J, Whyte WA, Zepeda-Mendoza CJ, Milazzo JP, Shen C, Roe J-S, Minder JL, Mercan F, Wang E, Eckersley-Maslin MA, Campbell AE, Kawaoka S, Shareef S, Zhu Z, Kendall J, Muhar M, Haslinger C, Yu M, Roeder RG, Wigler MH, et al (2013) Role of SWI/SNF in acute leukemia maintenance and enhancer-mediated Myc regulation. Genes Dev. 27: 2648–62 Simon JA & Kingston RE (2013) Occupying Chromatin: Polycomb Mechanisms for Getting to Genomic Targets, Stopping Transcriptional Traffic, and Staying Put. Mol. Cell 49: 808–824 Smith ZD & Meissner A (2013) DNA methylation: roles in mammalian development. Nat. Rev. Genet. 14: 204–220 Sneeringer CJ, Scott MP, Kuntz KW, Knutson SK, Pollock RM, Richon VM & Copeland RA (2010) Coordinated activities of wild-type plus mutant EZH2 drive tumor- associated hypertrimethylation of lysine 27 on histone H3 (H3K27) in human B- cell lymphomas. Proc. Natl. Acad. Sci. 107: 20980–20985 Son MJ, Woolard K, Nam D-H, Lee J & Fine HA (2009) SSEA-1 Is an Enrichment Marker for Tumor-Initiating Cells in Human Glioblastoma. Cell Stem Cell 4: 440– 452 Souroullas GP, Jeck WR, Parker JS, Simon JM, Liu J-Y, Paulk J, Xiong J, Clark KS, Fedoriw Y, Qi J, Burd CE, Bradner JE & Sharpless NE (2016) An oncogenic Ezh2

213

mutation induces tumors through global redistribution of histone 3 lysine 27 trimethylation. Nat. Med. 22: 632–640 Statham AL, Robinson MD, Song JZ, Coolen MW, Stirzaker C & Clark SJ (2012) Bisulfite sequencing of chromatin immunoprecipitated DNA (BisChIP-seq) directly informs methylation status of histone-modified DNA. Genome Res. 22: 1120–7 Steffen PA & Ringrose L (2014) What are memories made of? How Polycomb and Trithorax proteins mediate epigenetic memory. Nat. Rev. Mol. Cell Biol. 15: 340– 356 Struhl G & Akam M (1985) Altered distributions of Ultrabithorax transcripts in extra sex combs mutant embryos of Drosophila. EMBO J. 4: 3259–64 Suda Y, Kokura K, Kimura J, Kajikawa E, Inoue F & Aizawa S (2010) The same enhancer regulates the earliest Emx2 expression in caudal forebrain primordium, subsequent expression in dorsal telencephalon and later expression in the cortical ventricular zone. Development 137: 2939–2949 Suva M-L, Riggi N, Janiszewska M, Radovanovic I, Provero P, Stehle J-C, Baumer K, Le Bitoux M-A, Marino D, Cironi L, Marquez VE, Clement V & Stamenkovic I (2009) EZH2 Is Essential for Glioblastoma Cancer Stem Cell Maintenance. Cancer Res. 69: 9211–9218 Suvà M-L, Riggi N, Janiszewska M, Radovanovic I, Provero P, Stehle J-C, Baumer K, Le Bitoux M-A, Marino D, Cironi L, Marquez VE, Clément V & Stamenkovic I (2009) EZH2 is essential for glioblastoma cancer stem cell maintenance. Cancer Res. 69: 9211–8 Suvà ML, Rheinbay E, Gillespie SM, Patel AP, Wakimoto H, Rabkin SD, Riggi N, Chi AS, Cahill DP, Nahed B V., Curry WT, Martuza RL, Rivera MN, Rossetti N, Kasif S, Beik S, Kadri S, Tirosh I, Wortman I, Shalek A, et al (2014) Reconstructing and reprogramming the tumor propagating potential of glioblastoma stem-like cells. Cell 157: 580 Tamimi AF & Juweid M (2017) Epidemiology and Outcome of Glioblastoma Tavares L, Dimitrova E, Oxley D, Webster J, Poot R, Demmers J, Bezstarosti K, Taylor S, Ura H, Koide H, Wutz A, Vidal M, Elderkin S & Brockdorff N (2012) RYBP- PRC1 Complexes Mediate H2A Ubiquitylation at Polycomb Target Sites Independently of PRC2 and H3K27me3. Cell 148: 664–678 Theil T, Aydin S, Koch S, Grotewold L & Rüther U (2002) Wnt and Bmp signalling cooperatively regulate graded Emx2 expression in the dorsal telencephalon. Development 129: 3045–54

214

Tie F, Banerjee R, Fu C, Stratton CA, Fang M & Harte PJ (2016) Polycomb inhibits histone acetylation by CBP by binding directly to its catalytic domain. Proc. Natl. Acad. Sci. 113: E744–E753 Torres CM, Biran A, Burney MJ, Patel H, Henser-Brownhill T, Cohen A-HS, Li Y, Ben- Hamo R, Nye E, Spencer-Dene B, Chakravarty P, Efroni S, Matthews N, Misteli T, Meshorer E & Scaffidi P (2016) The linker histone H1.0 generates epigenetic and functional intratumor heterogeneity. Science (80-. ). 353: aaf1644-aaf1644 Tsubota S, Kishida S, Shimamura T, Ohira M, Yamashita S, Cao D, Kiyonari S, Ushijima T & Kadomatsu K (2017) PRC2-Mediated Transcriptomic Alterations at the Embryonic Stage Govern Tumorigenesis and Clinical Outcome in MYCN- Driven Neuroblastoma. Cancer Res. 77: 5259–5271 Tyanova S, Temu T, Sinitcyn P, Carlson A, Hein MY, Geiger T, Mann M & Cox J (2016) The Perseus computational platform for comprehensive analysis of (prote)omics data. Nat. Methods 13: 731–740 Ueda T, Sanada M, Matsui H, Yamasaki N, Honda Z, Shih L-Y, Mori H, Inaba T, Ogawa S & Honda H (2012) EED mutants impair polycomb repressive complex 2 in myelodysplastic syndrome and related neoplasms. Leukemia 26: 2557–2560 Varambally S, Cao Q, Mani R-S, Shankar S, Wang X, Ateeq B, Laxman B, Cao X, Jing X, Ramnarayanan K, Brenner JC, Yu J, Kim JH, Han B, Tan P, Kumar-Sinha C, Lonigro RJ, Palanisamy N, Maher CA & Chinnaiyan AM (2008) Genomic Loss of microRNA-101 Leads to Overexpression of Histone Methyltransferase EZH2 in Cancer. Science (80-. ). 322: 1695–1699 Varambally S, Dhanasekaran SM, Zhou M, Barrette TR, Kumar-Sinha C, Sanda MG, Ghosh D, Pienta KJ, Sewalt RGAB, Otte AP, Rubin MA & Chinnaiyan AM (2002) The polycomb group protein EZH2 is involved in progression of prostate cancer. Nature 419: 624–629 Venataraman S, Alimova I, Harris P, Birks DK, Balakrishnan I, Remke M, Taylor MD, Handler M, Foreman NK, Vibhakar R, Handler M, Foreman NK & Vibhakar R (2014) Inhibition of BRD4 attenuates tumor cell self-renewal and suppresses stem cell signaling in MYC driven medulloblastoma. Oncotarget 5: 2355–71 Verhaak RGW, Hoadley KA, Purdom E, Wang V, Qi Y, Wilkerson MD, Miller CR, Ding L, Golub T, Mesirov JP, Alexe G, Lawrence M, O’Kelly M, Tamayo P, Weir BA, Gabriel S, Winckler W, Gupta S, Jakkula L, Feiler HS, et al (2010) Integrated Genomic Analysis Identifies Clinically Relevant Subtypes of Glioblastoma Characterized by Abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell

215

17: 98–110 Viré E, Brenner C, Deplus R, Blanchon L, Fraga M, Didelot C, Morey L, Van Eynde A, Bernard D, Vanderwinden J-M, Bollen M, Esteller M, Di Croce L, de Launoit Y & Fuks F (2006) The Polycomb group protein EZH2 directly controls DNA methylation. Nature 439: 871–874 Vizán P, Beringer M, Ballaré C & Di Croce L (2015) Role of PRC2-associated factors in stem cells and disease. FEBS J. 282: 1723–1735 Vo BT, Li C, Morgan MA, Shilatifard A, Sherr CJ, Roussel Correspondence MF, Theurillat I, Finkelstein D, Wright S, Hyle J, Smith SMC, Fan Y, Wang Y-D, Wu G, Orr BA, Northcott PA & Roussel MF (2017) Inactivation of Ezh2 Upregulates Gfi1 and Drives Aggressive Myc-Driven Group 3 Medulloblastoma. Cell Rep. 18: 2907–2917 de Vries NA, Hulsman D, Akhtar W, de Jong J, Miles DC, Blom M, van Tellingen O, Jonkers J & van Lohuizen M (2015) Prolonged Ezh2 Depletion in Glioblastoma Causes a Robust Switch in Cell Fate Resulting in Tumor Progression. Cell Rep. 10: 383–397 Wainwright EN & Scaffidi P (2017) Epigenetics and Cancer Stem Cells: Unleashing, Hijacking, and Restricting Cellular Plasticity. Trends in Cancer 3: 372–386 Waitkus MS, Diplas BH & Yan H (2018) Biological Role and Therapeutic Potential of IDH Mutations in Cancer. Cancer Cell 34: 186–195 Walker E, Chang WY, Hunkapiller J, Cagney G, Garcha K, Torchia J, Krogan NJ, Reiter JF & Stanford WL (2010) Polycomb-like 2 Associates with PRC2 and Regulates Transcriptional Networks during Mouse Embryonic Stem Cell Self- Renewal and Differentiation. Cell Stem Cell 6: 153–166 Wang C, Oshima M, Sato D, Matsui H, Kubota S, Aoyama K, Nakajima-Takagi Y, Koide S, Matsubayashi J, Mochizuki-Kashio M, Nakano-Yokomizo T, Bai J, Nagao T, Kanai A, Iwama A & Sashida G (2018) Ezh2 loss propagates hypermethylation at T cell differentiation–regulating genes to promote leukemic transformation. J. Clin. Invest. 128: 3872–3886 Wang H, Wang L, Erdjument-Bromage H, Vidal M, Tempst P, Jones RS & Zhang Y (2004a) Role of histone H2A ubiquitination in Polycomb silencing. Nature 431: 873–878 Wang L, Brown JL, Cao R, Zhang Y, Kassis JA & Jones RS (2004b) Hierarchical Recruitment of Polycomb Group Silencing Complexes. Mol. Cell 14: 637–646 Wassef M, Luscan A, Aflaki S, Zielinski D, Jansen PWTC, Baymaz HI, Battistella A,

216

Kersouani C, Servant N, Wallace MR, Romero P, Kosmider O, Just P-A, Hivelin M, Jacques S, Vincent-Salomon A, Vermeulen M, Vidaud M, Pasmant E & Margueron R (2019) EZH1/2 function mostly within canonical PRC2 and exhibit proliferation-dependent redundancy that shapes mutational signatures in cancer. Proc. Natl. Acad. Sci.: 201814634 Wassef M & Margueron R (2017) The Multiple Facets of PRC2 Alterations in Cancers. J. Mol. Biol. 429: 1978–1993 Wassef M, Michaud A & Margueron R (2016) Association between EZH2 expression, silencing of tumor suppressors and disease outcome in solid tumors. Cell Cycle 15: 2256–2262 Widschwendter M, Fiegl H, Egle D, Mueller-Holzner E, Spizzo G, Marth C, Weisenberger DJ, Campan M, Young J, Jacobs I & Laird PW (2007) Epigenetic stem cell signature in cancer. Nat. Genet. 39: 157–158 Wilson BG, Wang X, Shen X, McKenna ES, Lemieux ME, Cho Y-J, Koellhoffer EC, Pomeroy SL, Orkin SH & Roberts CWM (2010) Epigenetic Antagonism between Polycomb and SWI/SNF Complexes during Oncogenic Transformation. Cancer Cell 18: 316–328 Wu G, Broniscer A, McEachron TA, Lu C, Paugh BS, Becksfort J, Qu C, Ding L, Huether R, Parker M, Zhang J, Gajjar A, Dyer MA, Mullighan CG, Gilbertson RJ, Mardis ER, Wilson RK, Downing JR, Ellison DW, Zhang J, et al (2012) Somatic histone H3 alterations in pediatric diffuse intrinsic pontine gliomas and non- brainstem glioblastomas. Nat. Genet. 44: 251–253 Xie H, Xu J, Hsu JH, Nguyen M, Fujiwara Y, Peng C & Orkin SH (2014) Polycomb Repressive Complex 2 Regulates Normal Hematopoietic Stem Cell Function in a Developmental-Stage-Specific Manner. Cell Stem Cell 14: 68–80 Xie W, Schultz MD, Lister R, Hou Z, Rajagopal N, Ray P, Whitaker JW, Tian S, Hawkins RD, Leung D, Yang H, Wang T, Lee AY, Swanson SA, Zhang J, Zhu Y, Kim A, Nery JR, Urich MA, Kuan S, et al (2013) Epigenomic Analysis of Multilineage Differentiation of Human Embryonic Stem Cells. Cell 153: 1134–1148 Xu K, Wu ZJ, Groner AC, He HH, Cai C, Lis RT, Wu X, Stack EC, Loda M, Liu T, Xu H, Cato L, Thornton JE, Gregory RI, Morrissey C, Vessella RL, Montironi R, Magi- Galluzzi C, Kantoff PW, Balk SP, et al (2012) EZH2 Oncogenic Activity in Castration-Resistant Prostate Cancer Cells Is Polycomb-Independent. Science (80-. ). 338: 1465–1469 Xu S, Grullon S, Ge K & Peng W (2014) Spatial clustering for identification of ChIP-

217

enriched regions (SICER) to map regions of histone methylation patterns in embryonic stem cells. Methods Mol. Biol. 1150: 97–111 Yang Y, Wang C, Zhang P, Gao K, Wang D, Yu H, Zhang T, Jiang S, Hexige S, Hong Z, Yasui A, Liu JO, Huang H & Yu L (2013) Polycomb Group Protein PHF1 Regulates p53-dependent Cell Growth Arrest and Apoptosis. J. Biol. Chem. 288: 529–539 Yap DB, Chu J, Berg T, Schapira M, Cheng S-WG, Moradian A, Morin RD, Mungall AJ, Meissner B, Boyle M, Marquez VE, Marra MA, Gascoyne RD, Humphries RK, Arrowsmith CH, Morin GB & Aparicio SAJR (2011) Somatic mutations at EZH2 Y641 act dominantly through a mechanism of selectively altered PRC2 catalytic activity, to increase H3K27 trimethylation. Blood 117: 2451–2459 Youmans DT, Schmidt JC & Cech TR (2018) Live-cell imaging reveals the dynamics of PRC2 and recruitment to chromatin by SUZ12-associated subunits. Genes Dev. 32: 794–805 Yu BD, Hess JL, Horning SE, Brown GAJ & Korsmeyer SJ (1995) Altered Hox expression and segmental identity in Mll-mutant mice. Nature 378: 505–508 Yu J, Cao Q, Mehra R, Laxman B, Yu J, Tomlins SA, Creighton CJ, Dhanasekaran SM, Shen R, Chen G, Morris DS, Marquez VE, Shah RB, Ghosh D, Varambally S & Chinnaiyan AM (2007a) Integrative Genomics Analysis Reveals Silencing of β- Adrenergic Signaling by Polycomb in Prostate Cancer. Cancer Cell 12: 419–431 Yu J, Yu J, Rhodes DR, Tomlins SA, Cao X, Chen G, Mehra R, Wang X, Ghosh D, Shah RB, Varambally S, Pienta KJ & Chinnaiyan AM (2007b) A Polycomb Repression Signature in Metastatic Prostate Cancer Predicts Cancer Outcome. Cancer Res. 67: 10657–10663 Yue D, Li H, Che J, Zhang Y, Tolani B, Mo M, Zhang H, Zheng Q, Yang Y, Cheng R, Jin JQ, Luh TW, Yang C, Tseng H-HK, Giroux-Leprieur E, Woodard GA, Hao X, Wang C, Jablons DM & He B (2015) EMX2 Is a Predictive Marker for Adjuvant Chemotherapy in Lung Squamous Cell Carcinomas. PLoS One 10: e0132134 Zang C, Schones DE, Zeng C, Cui K, Zhao K & Peng W (2009) A clustering approach for identification of enriched domains from histone modification ChIP-Seq data. Bioinformatics 25: 1952–1958 Zemke M, Draganova K, Klug A, Schöler A, Zurkirchen L, Gay MH-P, Cheng P, Koseki H, Valenta T, Schübeler D, Basler K & Sommer L (2015) Loss of Ezh2 promotes a midbrain-to-forebrain identity switch by direct gene derepression and Wnt- dependent regulation. BMC Biol. 13: 103

218

Zhang J, Chen L, Han L, Shi Z, Zhang J, Pu P & Kang C (2015) EZH2 is a negative prognostic factor and exhibits pro-oncogenic activity in glioblastoma. Cancer Lett. 356: 929–936 Zhang M, Kim HJ, Marshall H, Gendron-Maguire M, Lucas DA, Baron A, Gudas LJ, Gridley T, Krumlauf R & Grippo JF (1994) Ectopic Hoxa-1 induces rhombomere transformation in mouse hindbrain. Development 120: 2431–42 Zhang M, Wang Y, Jones S, Sausen M, McMahon K, Sharma R, Wang Q, Belzberg AJ, Chaichana K, Gallia GL, Gokaslan ZL, Riggins GJ, Wolinksy J-P, Wood LD, Montgomery EA, Hruban RH, Kinzler KW, Papadopoulos N, Vogelstein B & Bettegowda C (2014) Somatic mutations of SUZ12 in malignant peripheral nerve sheath tumors. Nat. Genet. 46: 1170–1172 Zhang W, Tyl M, Ward R, Sobott F, Maman J, Murthy AS, Watson AA, Fedorov O, Bowman A, Owen-Hughes T, El Mkami H, Murzina N V, Norman DG & Laue ED (2013) Structural plasticity of histones H3–H4 facilitates their allosteric exchange between RbAp48 and ASF1. Nat. Struct. Mol. Biol. 20: 29–35 Zhang Z, Jones A, Sun C-W, Li C, Chang C-W, Joo H-Y, Dai Q, Mysliwiec MR, Wu L- C, Guo Y, Yang W, Liu K, Pawlik KM, Erdjument-Bromage H, Tempst P, Lee Y, Min J, Townes TM & Wang H (2011) PRC2 Complexes with JARID2, MTF2, and esPRC2p48 in ES Cells to Modulate ES Cell Pluripotency and Somatic Cell Reprograming. Stem Cells 29: 229–240 Zhao J, Ohsumi TK, Kung JT, Ogawa Y, Grau DJ, Sarma K, Song JJ, Kingston RE, Borowsky M & Lee JT (2010) Genome-wide Identification of Polycomb-Associated RNAs by RIP-seq. Mol. Cell 40: 939–953 Zhou D, Alver BM, Li S, Hlady RA, Thompson JJ, Schroeder MA, Lee J-H, Qiu J, Schwartz PH, Sarkaria JN & Robertson KD (2018) Distinctive epigenomes characterize glioma stem cells and their response to differentiation cues. Genome Biol. 19: 43 Zhou W, Zhu P, Wang J, Pascual G, Ohgi KA, Lozach J, Glass CK & Rosenfeld MG (2008) Histone H2A monoubiquitination represses transcription by inhibiting RNA polymerase II transcriptional elongation. Mol. Cell 29: 69–80 Zingg D, Debbache J, Peña-Hernández R, Antunes AT, Schaefer SM, Cheng PF, Zimmerli D, Haeusel J, Calçada RR, Tuncer E, Zhang Y, Bossart R, Wong K-K, Basler K, Dummer R, Santoro R, Levesque MP & Sommer L (2018) EZH2- Mediated Primary Cilium Deconstruction Drives Metastatic Melanoma Formation. Cancer Cell 34: 69–84.e14

219

Zingg D, Debbache J, Schaefer SM, Tuncer E, Frommel SC, Cheng P, Arenas- Ramirez N, Haeusel J, Zhang Y, Bonalli M, McCabe MT, Creasy CL, Levesque MP, Boyman O, Santoro R, Shakhova O, Dummer R & Sommer L (2015) The epigenetic modifier EZH2 controls melanoma growth and metastasis through silencing of distinct tumour suppressors. Nat. Commun. 6: 6051 Zuber J, Shi J, Wang E, Rappaport AR, Herrmann H, Sison EA, Magoon D, Qi J, Blatt K, Wunderlich M, Taylor MJ, Johns C, Chicas A, Mulloy JC, Kogan SC, Brown P, Valent P, Bradner JE, Lowe SW & Vakoc CR (2011) RNAi screen identifies Brd4 as a therapeutic target in acute myeloid leukaemia. Nature 478: 524–528 Zuccarini M, Giuliani P, Ziberi S, Carluccio M, Iorio P Di, Caciagli F & Ciccarelli R (2018) The Role of Wnt Signal in Glioblastoma Development and Progression: A Possible New Pharmacological Target for the Therapy of This Tumor. Genes (Basel). 9: 105 Zuo T, Liu T-M, Lan X, Weng Y-I, Shen R, Gu F, Huang Y-W, Liyanarachchi S, Deatherage DE, Hsu P-Y, Taslim C, Ramaswamy B, Shapiro CL, Lin H-JL, Cheng ASL, Jin VX & Huang TH-M (2011) Epigenetic silencing mediated through activated PI3K/AKT signaling in breast cancer. Cancer Res. 71: 1752–62

220