Genomic Characterization of Medulloblastoma

Paul A. Northcott

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Department of Laboratory Medicine & Pathobiology University of Toronto

Paul A. Northcott

Doctor of Philosophy

Department of Laboratory Medicine & Pathobiology University of Toronto

2010 Abstract

Medulloblastoma is the most common malignant pediatric brain tumour. Although survival rates have improved in recent years, long-term survivors exhibit a significantly diminished quality of life complicated by neurological, endocrine, intellectual, and social sequelae as a result of conventional therapies. In order to improve the current outlook for patients with medulloblastoma, rational, targeted therapies that are more efficient and less toxic are required.

Despite insight gained from the study of hereditary tumour syndromes and candidate gene approaches, the molecular basis of medulloblastoma remains poorly defined, with more than half of all cases remaining unaccounted for at the genetic level. The intent of my PhD research program was to use high-resolution genomics in an attempt to gain an improved understanding of the medulloblastoma genome and potentially uncover novel genes and pathways driving its pathogenesis. By applying a combination of single nucleotide polymorphism (SNP) arrays, exon arrays, and microRNA arrays to a large cohort of primary medulloblastoma samples, we have identified novel oncogenes and tumour suppressors, implicated deregulation of the histone code as an important event in the pathogenesis of medulloblastoma, and refined the definition of medulloblastoma subgroups.

This thesis demonstrates the extent of heterogeneity that exists in the medulloblastoma genome, showing that relatively few genomic aberrations are common when studying medulloblastoma as a single disease. In spite of this heterogeneity, we have identified novel candidate genes and processes that may serve as potential targets for future therapies. Importantly, we have established an improved method of classifying medulloblastomas into distinct molecular variants, showing that certain genomic changes are enriched and occasionally restricted to a ii specific subgroup. Finally, in addition to genomic differences, we have confirmed that medulloblastoma subgroups differ in their demographics and clinical behavior, and propose that medulloblastoma subgroup affiliation should become an integral component of patient stratification in the future.

iii

Acknowledgments

First and foremost, I would like to thank my supervisor, Dr. Michael Taylor, for believing in me and giving me the opportunity to lead an outstanding PhD project in both a challenging and rewarding environment. I am also most grateful to have been co-supervised by Dr. Jim Rutka, who has provided me with important guidance and inspiration throughout the course of my thesis. I extend my gratitude to my additional supervisory committee members, Dr. Steve Scherer and Dr. Brent Derry, for their valuable insight and input towards my research. I am deeply indebted to all Taylor and Rutka Lab members, past and present, for their support, sharing of reagents and expertise, and willingness to contribute to my projects over the years. I would specifically like to acknowledge Paul Kongkham, Yukiko Nakahara, Xiaochong Wu, Stephen Mack, Adrian Dubuc, and John Peacock for their direct contributions to this thesis. I also thank Christian Smith for his invaluable “IT” support offered during the course of my PhD. From The Centre for Applied Genomics, I am grateful to Lars Feuk, Christian Marshall, John Wei, Chao Lu, Lan He, Kozue Otaka, and Xiaolin Wang for their amazing support with our genomics projects. From the Imaging Facility, I thank Paul Paroutis for assistance with microscopy and artwork. None of my PhD success would have been possible without strong collaborations with investigators and trainees from other institutions. I specifically thank Stefan Pfister, Hendrik Witt, Andrey Korshunov, Anna Kenney, Africa Fernandez, Charles Eberhart, Sidney Croul, and David Ellison for their collaboration. I would like to extend my gratitude to my family, including my mother Anne, and father Peter, for their endless support and encouragement to pursue success in life. Finally, I am most grateful to my wife Kara, and daughter Addison, for giving me daily inspiration and the motivation and strength to work hard and seek enjoyment in all aspects of life.

Table of Contents

!%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%"

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%"

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%"

& %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%&

&%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%& &%& %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% & &%' $ !# %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%' .)/). ))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))0 .)/)/ )))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))3 .)/)0 ))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))4 .)/)1 + ))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))5 &%( $ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%+ .)0). )))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))6 .)0)/ ))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))6 .)0)0 )))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))).. .)0)1 )))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))./ .)0)2 '%'& ! )))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))).0 &%) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% &* .)1). !%%# #"!% #% $ "# )))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))).3 .)1)/ *!" $# " $ "# (#!% #% '!!%* , -' "$ # % ! ", - #% !!%")))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))// .)1)0 $ "# #!"! # ! )))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))/3 .)1)1 $!""# $ "# )))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))0/ .)1)2 % ! #* "(! ", "-$ "# ))))))))))))))))))))))))00 .)1)3 % "(#$ "# )))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))06 &%* # %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% )' v

% #"

! ##

! ## ! ## !! #$ .(.(- #!""## !! (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((01 .(.(. -,, 1,, &!((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((01 .(.(/ &" !!((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((01 .(.(0 # !!"#& '"*+ (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((03 .(.(1 "# ""* +(((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((03 .(.(2 !#"# (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((03 .(.(3 !!(((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((04 .(.(4 "! (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((05 .(.(5 !!"## !!!((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((05 .(.(-, )" #"""$ * +(((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((1, .(.(-- "!!&!((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((1, .(.(-. &&!!$"&!!&! (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((1- .(.(-/ " $ #! #" (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((1- .(.(-0 ## # ! * +!"" !#" ((((((((((((((((((((((((1. .(.(-1 ## !#!"!" &(((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((1. .(.(-2 #"" ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((1/ !" $" .(/(- #!"(((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((1/ .(/(. # "" "%!!#!"(((((((((((((((((((((((((((((((11 .(/(/ #" # ""$"!$ " !"&!"&" #!" (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((34 !# &#

" &&

" && " && "! /(.(- ## !!"!"!((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((-,, /(.(. !""!"! "!((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( -,, vi

0(/(0 " ! ! &((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( .-- ,&, ! &&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*)* 0(0(. ! ! " )# " " " !$! !! (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( .-. 0(0(/ " !" !&! ! (((((((((((((((((((((((( .-5 0(0(0 " ") !#! !" !(((((((((((((((((((((((((((((((((((((( ..5 0(0(1 " ! !& ) #" !(((((((((((((((((((((((((((((((((((((((((((((((((((((((( ./1 0(0(2 !" ! " " &" ! ! &(((((((((((((((((((((((((((((((./1 ,&- ! &&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*+1

- ! &&&&&&&&&&&&&&&&&&&&&&&&&&&&*,-

-&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*,- -&* ! &&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*,- -&+ &&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*,. 1(/(. " !!"" (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( .02 1(/(/ .--2-- & ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( .02 1(/(0 & (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( .02 1(/(1 & (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( .03 1(/(2 &! %% & (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( .03 1(/(3 " !"& '!+ , ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( .03 1(/(4 & "!" (((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((( .03 -&, ! &&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*,0 1(0(. " ! " !&" ! (((((((((((((((((((((((((((((((((( .04 1(0(/ ).4*6/ # % "" " ! (((((((((((((((((((((((((((((.04 1(0(0 ).4*6/" "! !$!!#! " " ! ((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((((.05 1(0(1 ).4*6/ " "!& & "!" ((((((((((((((((((((((((((((((.15 1(0(2 ).4*6/" ! $ ! )& (((((((((((((((((((((((((((((((((((((( .2. -&- ! &&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*.*

.! $! ! &&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*..

.&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*.. .&* ! % " " ! &&&&&&&&&&&&&&&&&&&&&&&*.. .&+ # ! !! ! '! (&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&*./ vii

!"# $%&'()%*'+%,-.((%&/$0%12%3)45(('6(.07'-.%8)9)(':-)27""""""""""""""""""""""""""""""""""""""""""""""""";!< !"= >?)%@575+)%'*%3)45(('6(.07'-.%A)2'-1B0""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""";C; ! ! &)*)+)2B)0! """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""";CC!

viii

List of Tables

Chapter 1

Table 1. Genetically engineered mouse models of medulloblastoma...... 10

Table 2. Most prominent cytogenetic aberrations in medulloblastoma ...... 21

Table 3. Candidate oncogenes recurrently amplified in medulloblastoma ...... 23

Table 4. Alterations of miRNA expression found in medulloblastoma ...... 36

Chapter 2

Table 1. Known oncogenes amplified in medulloblastoma...... 67

Table 2. Recurrent regions of amplification...... 68

Table 3. Homozygous deletion of known and candidate tumor suppressor genes...... 72

Table 4. Regions of recurrent, focal homozygous deletion targeting a single RefSeq...... 75

Table 5. Regions of recurrent homozygous deletion in medulloblastoma...... 79

Table 6. Novel genetic events converge on H3K9 in medulloblastoma...... 82

Chapter 3

Table 1. Ingenuity pathway analysis of subgroup-specific genes in medulloblastoma ..... 115

List of Figures

Chapter 1

Figure 1. The Shh signaling pathway...... 4

Figure 2. Evolution of genomic technologies and their application to the study of medulloblastoma...... 17

Figure 3. Prominent genomic aberrations in medulloblastoma...... 20

Figure 4. Transcriptional profiling of medulloblastoma...... 28

Figure 5. Overview of miRNA biogenesis...... 34

Chapter 2

Figure 1. Array-based profiling of medulloblastoma: summary of the literature...... 56

Figure 2. The medulloblastoma genome...... 57

Figure 3. Gains and losses in the medulloblastoma genome...... 58

Figure 4. CNAs on chromosome 17 in medulloblastoma...... 60

Figure 5. LOH in the medulloblastoma genome...... 62

Figure 6. Identification of statistically significant targets of gain and loss in medulloblastoma using GISTIC...... 65

Figure 7. Amplification of known oncogenes in medulloblastoma...... 69

Figure 8. Recurrent targeting of OTX2 in medulloblastoma...... 70

Figure 9. Homozygous deletion of CDKN2A/CDKN2B and LRP1B is limited to medulloblastoma cell lines...... 73

Figure 10. Recurrent focal deletion of TSC1 in medulloblastoma...... 76

Figure 11. Recurrent homozygous deletions target the histone lysine methyltransferase, EHMT1, in medulloblastoma...... 80

Figure 12. CNAs target genes controlling histone lysine methylation in medulloblastoma...... 84-85

Figure 13. Transcriptional deregulation of genes controlling histone lysine methylation in medulloblastoma...... 86

Figure 14. Re-expression of L3MBTL3 in the DAOY medulloblastoma cell line...... 89

Figure 15. H3K9 in the developing cerebellum...... 92

Figure 16. Model: role of H3K9 methylation in EGL development and medulloblastoma. 95

Chapter 3

Figure 1. Transcriptional profiling of medulloblastoma identifies four non-overlapping molecular subgroups with distinct demographics...... 102

Figure 2. Unsupervised HCL of medulloblastomas consistently identifies four distinct subgroups...... 104-106

Figure 3. Molecular classification of medulloblastoma by transcriptional profiling...... 109

Figure 4. Classification of medulloblastoma using PAM...... 111

Figure 5. Consensus NMF analysis of medulloblastoma expression data...... 113

Figure 6. Gene set enrichment analysis (GSEA) of medulloblastoma subgroups: overlap with chemical and genetic perturbation gene sets...... 116

Figure 7. Subgroup-specific genetic events in the medulloblastoma genome...... 119

Figure 8. Copy number summary plots for medulloblastoma subgroups...... 121

Figure 9. GISTIC analysis of medulloblastoma subgroups identifies statistically significant subgroup-specific CNAs...... 123

Figure 10. Molecular heterogeneity in SHH-driven medulloblastomas...... 125

Figure 11. Validation of medulloblastoma subgroups by immunohistochemistry...... 126

Figure 12. Prognostic significance of medulloblastoma subgroups...... 127

Figure 13. Survival probability of subgroups and histological subtypes of medulloblastoma...... 130

Chapter 4

Figure 1. Recurrent amplification of the miR-17/92 locus in primary human medulloblastomas...... 139

Figure 2. Overexpression of miR-17/92 in human and murine medulloblastomas...... 141

Figure 3. Unsupervised HCL of 90 medulloblastomas identifies four unique subgroups. 143

Figure 4. miR-17/92 is overexpressed in SHH-dependent medulloblastomas and tumours with elevated MYC family expression...... 144

Figure 5. miR-17/92 upregulation in specific medulloblastoma subgroups...... 146

Figure 6. MYC family expression in specific medulloblastoma subgroups...... 147

Figure 7. SHH and N-myc drive the expression of miR-17/92 in cerebellar neural precursor cells resulting in mitosis...... 149

Figure 8. Multiple mechanisms lead to deregulation of miR-17/92 in medulloblastoma... 153

Chapter 5

Figure 1. The future of medulloblastoma genomics...... 164

xii

Abbreviations

˚C degrees Celsius

5-aza 5-aza-2’-deoxycytidine

ANOVA analysis of variance

Array-PRIMES array-based profiling of reference-independent methylation status

AT/RT atypical teratoid rhabdoid tumours

ATRA all-trans-retinoic acid

BAC bacterial artificial chromosome bHLH basic helix-loop-helix bHLHZ basic helix-loop-helix-zipper

BMP bone morphogenic protein bp base pairs

CB cerebellum

CDK cyclin dependent kinase cDNA complementary DNA

CGH comparative genomic hybridization

CGNP cerebellar granule neuron precursor

CGP chemical and genetic perturbation database

ChIP chromatin immunoprecipitation

CKI cyclin-dependent kinase inhibitor cm centimetre

CNAs copy number aberrations

CNS central nervous system

CNV copy number variants

CTA cancer-testis antigen

xiii d days

DM double minute

DMH differential methylation hybridization

DNA deoxyribonucleic acid

EGL external granule layer

ES cells embryonic stem cells ext-EGL external layer of the EGL

FBS fetal bovine serum

FISH fluorescence in situ hybridization g acceleration due to gravity

G-banding Giemsa banding

GABA gamma-aminobutyric acid

Gb gigabase

GISTIC Genomic Identification of Significant Targets in Cancer

GLAD Gain and Loss Analysis of DNA

GSEA Gene Set Enrichment Analysis h hour

H&E hematoxylin and eosin

HA hemagglutinin

HCL hierarchical clustering

HGF hepatocyte growth factor

HMM hidden Markov model

HRP horseradish peroxidase

HSR homogeneously staining region i{17}q isochromosome 17q

IGL internal granule layer xiv int-EGL internal layer of the EGL

IPA Ingenuity Pathway Analysis

Kb kilobases

LCA large-cell/anaplastic

LCR segmental duplications

LOCKs large organized chromatin K9 modifications

LOH loss of heterozygosity

M molar

M+ metastatic

M0 non-metastatic

Mb megabases

MB medulloblastoma

MedIP methylation-dependent immunoprecipitation

MEFs mouse embryo fibroblasts mg milligram min minute miRNA microRNA ml millilitre

ML molecular layer

MO month old

MSigDB Molecular Signatures Database mu microns n sample size

NBCCS nevoid basal-cell carcinoma syndrome next-gen next-generation ng nanogram xv

NGS normal goat serum nm nanometre

NMF Non-negative matrix factorization

NSC neural stem cell

OS overall survival

P postnatal day

P probability

PAM Prediction analysis of microarrays

PBS Phosphate Buffered Saline

PCA principal component analysis

PCR polymerase chain reaction

PFA paraformaldehyde

PFS progression-free survival

PI propidium iodide

PNET primitive neuroectodermal tumour

PNS peripheral nervous system qPCR quantitative polymerase chain reaction qRT-PCR quantitative reverse-transcriptase polymerase chain reaction

RCAS Replication-Competent ASLV long terminal repeat with a Splice acceptor

RFLP restriction fragment length polymorphism

RLGS restriction landmark genomic scanning

RNA ribonucleic acid

RPM revolutions per minute

SAGE serial analysis of gene expression

SAS subarachnoid space

SD standard deviation xvi

SDS sodium dodecyl sulfate

SHH Sonic Hedgehog

SJMB-96 St. Jude Medulloblastoma-96 clinical trial

SKY spectral karyotyping

SNP single nucleotide polymorphism sPNET supratentorial primitive neuroectodermal tumour

SSH suppression subtractive hybridization

SubMap Subclass Mapping

SVZ subventricular zone

TCAG The Centre for Applied Genomics

TCGA The Cancer Genome Atlas

TMA tissue microarray

TSG tumour suppressor gene ug microgram um micrometre

UPD uniparental disomy

UTR untranslated region vol volume

VZ ventricular zone w weight

WHO World Health Organization

WNT Wingless

xvii 1

Chapter 1 Introduction 1 *

Medulloblastoma (World Health Organization (WHO) Grade IV) is an embryonal tumour of the cerebellum and the most common malignant childhood brain tumour1,2. Although five-year overall survival rates have improved dramatically of late, survivors typically face a variety of long-term neurological, neuroendocrine, and social sequelae as a result of conventional treatment regimens.3-6. It is therefore imperative we gain a better understanding of the genes and pathways driving medulloblastoma oncogenesis so future targeted therapies that are more effective and less toxic can be made available.

1.1 Clinical Aspects of Medulloblastoma

Medulloblastoma accounts for approximately 20% of all intracranial tumours in children and 40% of childhood posterior fossa tumours5,6. This malignancy occurs predominantly in young children, with a peak incidence ranging from 3-9 years of age. Approximately 10-15% of medulloblastomas are diagnosed in infants. Although medulloblastoma is known to occur in adult patients, these cases account for <1% of all adult central nervous system (CNS) tumours. Infants with medulloblastoma often have poor clinical outcomes, whereas adults typically have improved overall survival rates. The male:female ratio of medulloblastoma is approximately 1.5:1, and males over the age of 3 years have been reported to have a worse prognosis than females of the same age group7-9.

Current treatment stratification protocols for medulloblastoma place patients into either average- risk or high-risk categories based on three diagnostic criteria: patient age, extent of surgical resection, and metastatic status5. Patients under the age of 3, patients with >1.5cm2 residual

* Work in this Thesis Chapter contributed to the following publications: Northcott PA, Rutka JT, Taylor MD: Genomics of medulloblastoma: from Giemsa-banding to next-generation sequencing in 20 years. Neurosurg Focus 28:E6 (1-20), 2010. Fernandez LA, Northcott PA, Taylor MD, et al: Normal and oncogenic roles for microRNAs in the developing brain. Cell Cycle 8:4049-54, 2009. Rutka JT, Kongkham P, Northcott P, et al: The evolution and application of techniques in molecular biology to human brain tumors: a 25 year perspective. J Neurooncol 92:261-73, 2009. 2

disease, or patients with disseminated disease are all considered high-risk, while all others are deemed average-risk. With the application of treatment that includes surgical resection, craniospinal irradiation, and chemotherapy, cure rates of 80-85% for average-risk patients and up to 70% for high-risk patients have been obtained9,10.

The diagnosis of medulloblastoma is currently based on histological criteria. According to the 2007 WHO Classification of Tumours of the CNS, medulloblastoma can be classified into five recognizable subtypes: classic, desmoplastic/nodular, medulloblastoma with extensive nodularity (MBEN), anaplastic, and large-cell medulloblastoma 11,12. Classic medulloblastoma is by far the most common, followed by desmoplastic cases that make up 10-20%, and the large- cell/anaplastic (LCA) tumours (usually grouped together), which account for 5-10% of tumours. Two very rare subtypes of medulloblastoma include melanotic medulloblastoma and medullomyoblastoma. Although there is considerable heterogeneity and variability with regards to histological subtype and disease course, a multitude of reports have linked desmoplastic/nodular and MBEN cases with a better prognosis and LCA cases with a significantly worse outcome9,13-16.

1.2 Medulloblastoma: Normal Cerebellar Development Gone Awry

As an embryonal cerebellar tumour, there is substantial molecular, cellular, and biological evidence linking medulloblastoma pathogenesis to dysregulation of normal developmental processes12,17. To fully comprehend the cellular origin(s) and molecular mechanisms responsible for medulloblastoma pathogenesis, an understanding of the molecular and cellular processes required for normal cerebellar development is necessary.

Development of the cerebellum begins during embryonic life and continues to develop postnatally in both mice and humans. Two distinct germinal zones, the primary and secondary germinal zones, contain stem and progenitor cell populations responsible for populating the various cell types of the mature cerebellum. The primary germinal zone (i.e. ventricular zone; VZ) is located in the roof of the fourth ventricle, giving rise to GABAergic neurons, Purkinje cells, and Golgi neurons. Progenitor cells of the secondary germinal zone originate in the rhombic lip and give rise to cerebellar granule neuron precursors (CGNPs) that migrate rostrally across the cerebellum to form the external granule layer (EGL). The EGL persists until postnatal 3

day 15 (P15) in mice and into the second year of life in humans. As the cerebellum continues to develop, CGNPs comprising the EGL undergo a period of rapid and massive clonal expansion (peaking at P5-P7 in the mouse), before migrating inward, across the Purkinje cell layer, to eventually form the post-mitotic neurons of the internal granule layer (IGL).

Cerebellar development relies on the coordinated activity of multiple signal transduction pathways including the Sonic Hedgehog (Shh), Wingless (Wnt), and Notch cascades. These developmental pathways play a critical role in the expansion of neural precursor populations such as those found in the EGL and VZ, and inappropriate activation of these pathways contributes to medulloblastoma genesis. The role of these pathways in cerebellar development and evidence for their involvement in medulloblastoma is discussed below.

1.2.1 SHH

Over the past two decades there has been a considerable amount of literature strongly implicating CGNPs of the EGL as a probable cell of origin for at least a subset of medulloblastomas12,17-19. During normal cerebellar development, proliferation of CGNPs relies on mitogenic stimulation by neighboring Purkinje cells that secrete Shh ligand. Secreted Shh binds to its receptor, PTCH on target CGNPs causing derepression of the serpentine receptor SMO and leading to activation of downstream GLI transcription factors which trancriptionally upregulate pro-mitotic genes such as GLI1 and MYCN (Figure 1). Deregulation of the Shh pathway was first suggested in the pathogenesis of medulloblastoma when mutations in PTCH1 were reported in the germline of individuals with Gorlin syndrome (also known as nevoid basal-cell carcinoma syndrome, NBCCS)20,21. Kindreds with Gorlin syndrome develop characteristic bone cysts and multiple basal cell carcinomas, and are predisposed to a variety of tumour types including medulloblastoma. In addition to familial medulloblastomas, PTCH1 mutations are found in ~9% of sporadic cases12. Additional components of the Shh signaling apparatus are also targeted for mutation in medulloblastomas. Germline mutations in SUFU, another negative regulator of the Shh pathway, predispose to medulloblastoma formation and rare mutations have been reported in sporadic cases22. Activating mutations in SMO have also been identified in medulloblastoma, providing yet another mechanism by which the Shh pathway can be dysregulated23,24. Definitive evidence supporting a role for the Shh pathway in medulloblastoma owes to studies manipulating this pathway in mice. In 1997, Matthew Scott’s group deleted Ptch1 in the mouse, with 14% of 4

Figure 1. The Shh signaling pathway. Schematic diagram illustrating major components of the Shh signaling pathway in mammals. During cerebellar development, Purkinje cells secrete SHH ligand which binds to the PTCH1 receptor located on target CGNPs. SHH-PTCH1 interaction leads to derepression of the transmembrane protein SMO, inhibiting turnover of GLI family transcription factors mediated by cytosolic proteins, FU and KIF7/KIF27. During active Shh signaling, GLI transcription factors translocate to the nucleus to activate target genes, which include components of the Shh pathway such as PTCH1, GLI1, and HHIP, as well as the MYCN proto-oncogene. GLI1 and GLI2 primarily function as transcriptional activators, whereas GLI3 is a transcriptional repressor. SUFU negatively regulates GLI activity through a poorly understood mechanism. 5

heterozygous animals (Ptch1+/-) developing posterior fossa tumours by 10 months of age that resembled human medulloblastoma25. Subsequent studies have shown that the penetrance of medulloblastomas on the Ptch1+/- background can approach 100% by crossing these mice with Tp53-/- mice (i.e. Ptch1+/-;Tp53-/-), which fail to develop medulloblastoma on their own26. Importantly, elegant studies by Rob Wechsler-Reya’s group have shown that Ptch1+/- mice develop pre-neoplastic foci on the surface of the EGL that express the granule cell lineage marker Math1 and elevated levels of Shh target genes Gli1, Mycn, and CyclinD127. These preneoplastic cells are believed to represent CGNPs that have failed to complete their normal course of expansion, migration, and differentiation and thus develop into medulloblastoma. Since these early studies, the Ptch1+/- mouse model has become an indispensible tool for labs studying the molecular events that cooperate with aberrant Shh signaling during medulloblastoma pathogenesis and for studies aimed at gaining a better understanding of medulloblastoma biology. More recently, Wechsler-Reya and colleagues have shed further light in this area through targeted deletion of PTCH1 in both Math1-expressing neuronal progenitors (i.e. CGNPs) and neural stem cells (NSCs; i.e. GFAP-expressing cells)28. Conditional loss of PTCH1 in either CGNPs or NSCs resulted in universal medulloblastoma onset by 10-11 and 4 weeks, respectively. It is important to note that NSCs give rise to CGNPs, as well as most other types of neurons and glia in the cerebellar cortex. Despite this, disruption of PTCH1 in NSCs resulted in medulloblastomas but not other CNS tumours such as astrocytomas, suggesting that Shh-driven tumours are associated with neuronal lineage commitment.

Additional Shh-driven mouse models of medulloblastoma have further established the importance of this pathway in the human disease. For example, Sufu+/-;Tp53-/- mice develop medulloblastomas that appear to arise in the EGL and exhibit constitutive activation of the Shh pathway29. The median survival period of Sufu+/-; Tp53-/- animals (~4 months) is longer than that of Ptch1+/-;Tp53-/- animals (~2 months), suggesting that Ptch1 is a more potent tumour suppressor gene (TSG) than Sufu in the context of medulloblastoma. Transgenic expression of an activated form of Smoothened (ND2:SmoA1) in CGNPs resulted in hyperproliferation of these cells and ~48% of mice developing medulloblastoma with a median age of onset of ~25 weeks30. Similar to tumours formed on Ptch1 and Sufu mutant genotypes, cerebellar tumours arising in ND2:SmoA1 mice exhibit an activated Shh expression ‘signature’, as well as aberrant Notch pathway expression. An ‘improved’ homozygous variant of the ND2:SmoA1 model was 6

published by the same group in 2008, with ~94% of transgenics developing medulloblastoma by 2 months of age31. Importantly, these tumours had a propensity for leptomeningeal spread to the brain and spine, a hallmark of the human disease that had not been previously recapitulated in other mouse models of medulloblastoma. Collectively, these studies in both humans and mice have supported the importance of the Shh pathway in medulloblastoma pathogenesis, and provided compelling evidence that the Shh-driven subtype of medulloblastoma originates in either CGNPs of the EGL or NSCs that give rise to these cells.

1.2.2 WNT

Much like the molecular basis of Gorlin syndrome provided early clues that uncovered a broader role for the Shh pathway in medulloblastoma, the discovery of APC mutations in individuals with Turcot syndrome implicated the Wnt pathway as another developmental signaling pathway dysregulated in medulloblastoma32,33. Turcot syndrome (also known as brain tumour-polyposis syndrome Type I and II) is a familial cancer syndrome characterized by the association of a primary brain tumour (glioma or medulloblastoma) and multiple colorectal adenomas. Individuals with Type I syndrome harbor germline mutations in hMLH1 and are more likely to develop gliomas, whereas Type II syndrome is the result of APC mutations and is associated with medulloblastoma development. Mutation of APC has also been reported in ~3% of sporadic medulloblastomas, whereas activating mutations in -Catenin (CTNNB1), the central molecule of the Wnt pathway, are found in ~8% of sporadic cases12. Mutations in other components of the pathway (i.e. AXIN1) have also been identified in rare cases.

Similar to the mitogenic role Shh plays in modulating the expansion of CGNPs in the EGL during cerebellar development, Wnt signaling regulates proliferation of stem and progenitor cells in the fetal VZ, as well as the postnatal subventricular zone (SVZ). Targeted deletion of Wnt1 in the mouse results in embryonic lethality or death in very early postnatal life with mice exhibiting severe developmental abnormalities of the midbrain and cerebellum34,35. Similarly, disruption of Ctnnb1 in nestin-expressing cells of the developing mouse causes impaired morphogenesis of the midbrain and cerebellum36, further supporting a role for Wnt signaling in normal cerebellar development.

Multiple lines of both observational and experimental evidence suggest that the Wnt and Shh subtypes of medulloblastoma are mutually exclusive from one another and arise from distinct 7

cell types. From an immunohistochemical perspective, human tumours exhibiting activation of the Wnt pathway (i.e. -Catenin nucleopositivity) do not overlap with those showing evidence of Shh pathway activation (i.e. Gli1 and SFRP1 immunopositivity9; see Chapter 3 of this thesis). Secondly, from a histological perspective, the majority of medulloblastomas exhibiting Wnt pathway activation are of classic histology, while nodular/desmoplastic medulloblastomas are predominantly associated with Shh-driven cases37-39. Thirdly, from a clinical perspective, multiple recent independent clinical trials (i.e. PNET3 and SJMB-96) have established that - Catenin nucleopositivity is associated with an excellent prognosis compared to nucleonegative tumours, with five-year event-free survival rates of 89-100% noted in nucleopositive tumours versus rates of 60-70% in the remainder 9,10. Finally, as discussed in detail above, Shh-driven medulloblastomas most likely originate in CGNPs of the EGL or NSCs that give rise to these cells. Activation of the Wnt pathway in CGNPs through expression of constitutively active Ctnnb1 using either transgenic technology in vivo or viral transduction ex vivo does not alter the normal phenotype of these cells nor lead to tumour development12, suggesting Wnt-driven medulloblastomas arise from a distinct cell of origin.

1.2.3 NOTCH

Although not nearly as well characterized as the Wnt and Shh pathways in terms of its role in medulloblastoma development, the Notch signaling pathway has also been implicated in some medulloblastomas. Similar to the Shh pathway, Notch signaling is active in both germinal zones of the cerebellum, promoting proliferation and/or survival of neural stem and progenitor cell populations while inhibiting their differentiation. In medulloblastoma, Eberhart and colleagues have reported overexpression of NOTCH2 in some primary human cases; with rare amplifications of the NOTCH2 genomic locus also noted40. In contrast, NOTCH1 expression was shown to be undetectable in >50% of samples profiled. The same group demonstrated that NOTCH2 and NOTCH1 have antagonistic roles in medulloblastoma biology, with NOTCH2 upregulation capable of promoting cell proliferation, soft agar colony formation, and xenograft growth – all of which were inhibited by exogenous NOTCH1 expression. Interestingly, apparent cross-talk between the Notch and Shh pathways has been observed in murine medulloblastomas, as tumours derived from ND2:SmoA1 mice are characterized by elevated levels of Notch2 and the Notch target gene Hes530. These observations suggest that Shh and Notch pathway medulloblastomas may not represent independent entities. 8

1.2.4 HGF/MET

Another mitogenic pathway recently implicated in medulloblastoma is the hepatocyte growth factor (HGF)/Met signaling cascade. Li et al. were among the first to suggest this pathway in medulloblastoma pathogenesis, reporting that aberrant expression of c-Met in primary samples correlated with a poor clinical outcome41. The same study showed that treatment of medulloblastoma cell lines with HGF activates c-Met–dependent signaling, resulting in anchorage-independent growth, cell proliferation, cell cycle progression, and cell survival. Furthermore, using a xenograft model the authors showed that HGF provided medulloblastoma cells with a significant growth advantage and promoted an anaplastic phenotype. In a subsequent study, Dan Fults and colleagues used the RCAS/tv-a somatic cell gene transfer system to force HGF expression in murine nestin-expressing neural progenitors, reporting that ectopic HGF significantly enhanced Shh-dependent medulloblastoma formation42. This positive effect on tumourigenesis could be partially attenuated through systemic administration of a neutralizing antibody against HGF, suggesting that sustained HGF expression is required for the maintenance of these tumours. A recent report by Kongkham et al. provided further evidence for HGF/Met pathway dysregulation in medulloblastoma, identifying SPINT2, a negative regulator of the pathway, as a novel putative TSG that is epigenetically silenced in a large percentage of primary medulloblastomas and cell lines and capable of repressing medulloblastoma cell growth both in vitro and in vivo43.

1.3 Candidate Gene Approaches to Medulloblastoma: Lessons from Mice

There is little debating that the majority of candidate oncogenes and tumour suppressors thus far implicated in medulloblastoma pathobiology function as components of the developmental signaling pathways discussed above. Indeed, reports describing mutations in the Shh pathway have dominated the medulloblastoma literature in recent years and research on this pathway continues to account for an impressive amount of work in the field. However, several recent candidate gene approaches, many of which have employed transgenic mouse models (see Table 1 for a current list of genetically engineered mouse models of medulloblastoma), have provided insight into additional genes, pathways, and processes that contribute to normal cerebellar development and medulloblastoma aetiology. Although not comprehensive, a selection of some of the more recent and more informative candidate gene studies is described below. 9

1.3.1 MYCN

Over 25 years ago, the MYCN proto-oncogene was identified as a highly amplified component of homogeneously staining regions (HSRs) and double minute chromosomes (DMs) in human neuroblastomas and neuroblastoma cell lines that shared sequence homology with c-myc (i.e. MYC)44,45. Since its initial discovery, MYCN has been well characterized as a basic helix-loop- helix-zipper (bHLHZ) transcription factor that functions as a classical oncogene, particularly in tumours of the peripheral nervous system (PNS; i.e. neuroblastoma) and CNS (i.e. medulloblastoma)46. Although a role for MYCN deregulation in medulloblastoma biology has been repeatedly suggested in many genomic profiling studies (see below and Chapter 2 of this thesis), significant knowledge has been gained by studying Mycn function in the mouse. Indeed one of the most informative studies to date involved targeted disruption of Mycn in the developing cerebellum through generation of Nestin-cre;Mycnloxp/loxp mice47. These mice exhibit tremors, ataxia, and behavioral abnormalities owing to profound microencephaly with the most dramatic defects observed in the cerebellum and cerebral cortex. The cerebellum of Mycn null mice is virtually absent, with pronounced deficiencies observed in both cerebellar germinal zones, emphasizing the importance of Mycn in normal cerebellar development. In subsequent studies, work by multiple groups has confirmed an absolute requirement for Mycn activity downstream of Shh in mediating the proliferation of CGNPs and propagating Shh-driven medulloblastomas48-50 and forced Mycn expression in CGNPs derived from Ink4c-/-;Tp53-/- or preneoplastic cells isolated from Ptc+/- mice initiates their transformation in orthotopic mouse models51,52.

1.3.2 BMI1

BMI1 is a member of the Polycomb group of transcriptional repressors that is involved in the regulation of development, stem cell self-renewal, cell cycle, and senescence. Bmi1 was first identified as an oncogene capable of cooperating with c-Myc in a mouse model of B-cell lymphoma53 and BMI1 expression has since been shown to be deregulated in a variety of human malignancies, including medulloblastoma54 (see below and Chapter 2 of this thesis). Targeted deletion of Bmi1 in the mouse leads to skeletal and haematopoietic abnormalities, as well as significant neural defects including a pronounced reduction in the size of the cerebellum54,55. In a 2004 study by Sylvia Marino and colleagues, the cerebellar phenotype originally observed in 10 11

Bmi1 null mice was further characterized, reporting that Bmi1 is required for clonal expansion of CGNPs during cerebellar development and that Bmi1 expression is induced by Shh pathway activation in these cells54. Furthermore, this study identified frequent upregulation of BMI1 expression in primary human medulloblastomas and cell lines, suggesting it may function as a medulloblastoma oncogene. Although widespread deregulation of BMI1 expression has been observed in medulloblastoma54,56, the mechanism(s) responsible for BMI deregulation remain unclear. Future strategies that include targeted overexpression of Bmi1 in CGNPs or NSCs are necessary to confirm its role as a bona fide oncogene in medulloblastoma.

1.3.3 HIC1

HIC1 (HYPERMETHYLATED IN CANCER 1) is a POZ domain transcription factor frequently targeted for promoter hypermethylation and epigenetic silencing in an array of human cancers including medulloblastoma57. Importantly, the HIC1 gene maps to chromosome 17p13.3, the most frequently deleted locus in medulloblastoma (see Chapter 2 of this thesis). Heterozygous deletion of Hic1 (i.e. Hic1+/-) in the mouse results in the development of a variety of cancers (secondary to epigenetic silencing of the remaining wild-type Hic1 allele), but not medulloblastomas58. Recently, Watkins and colleagues investigated the tumour suppressor role of HIC1 in the context of medulloblastoma by generating Ptch1+/-;Hic1+/- heterozygous mice59. Compared to Ptch1+/- littermates, compound Ptch1+/-;Hic1+/- heterozygotes demonstrated more than a four-fold increase in medulloblastoma incidence. Of particular interest, the authors subsequently identified Atonal (Atoh1; also known as Math1), a proneural basic helix-loop-helix (bHLH) transcription factor that is required for CGNP proliferation, as a direct transcriptional target for repression by HIC1 in both medulloblastoma cell lines and CGNPs. Since Atoh1/Math1 is both positively regulated by and required for Shh signaling in CGNPs (see below), the authors proposed a novel mechanism by which PTCH1 and HIC1 cooperate to negatively regulate CGNP proliferation during cerebellar development. Although this study was indeed informative, it is worth noting that activated Shh signaling (i.e. secondary to PTCH1 mutation) and chromosome 17p deletion are mutually exclusive in human medulloblastoma37,38 (see Chapter 3 of this thesis), perhaps questioning the relevance of disrupting Hic1 on the Ptch1 background. Whether or not epigenetic silencing of HIC1 is more prevalent in Shh-driven tumours compared to non-Shh tumours requires further investigation. 12

1.3.4 ATOH1/MATH1

Atoh1/Math1 is an early marker of CGNPs that is required for their normal proliferation and subsequent differentiation and migration during cerebellar development60,61. Recently, in a parallel study to that described by Watkins and colleagues, Martine Roussel’s group described a novel mechanism by which bone morphogenic proteins (BMPs; i.e. BMP2 and BMP4 in this study) inhibit the proliferation and promote the differentiation of CGNPs by mediating Atoh1/Math1 turnover in these cells62. Although BMPs had been previously shown to antagonize Shh-dependent proliferation and induce CGNP differentiation63,64, this study confirmed that post-transcriptional downregulation of Atoh1/Math1 in CGNPs is required for BMP-mediated cell cycle exit as exogenous Atoh1/Math1 expression abrogated the effects of BMP treatment. Furthermore, the authors demonstrated anti-proliferative properties for BMPs in vivo by treating CGNP-like mouse medulloblastoma cells with BMP4 – an effect that was additive when combined with the Shh pathway inhibitor cyclopamine – suggesting a possible therapeutic role for BMP agonists in the treatment of medulloblastoma.

With independent reports by Watkins and Roussel converging on Atoh1/Math1 as a key molecule involved in the propagation of CGNPs during normal development and their transformed counterpart during medulloblastoma genesis, the possibility of Atoh1/Math1 as a novel therapeutic target in medulloblastoma has become an attractive hypothesis65,66. New evidence in this area has continued to emerge, as Flora et al. have further validated the importance of Atoh1/Math1 in both CGNP proliferation and medulloblastoma development in a very recent publication67. Using an elegant mouse model in which Atoh1/Math1 was deleted in the postnatal cerebellum, the authors showed that Atoh1/Math1 was required for transduction of Shh signaling in CGNPs and loss of Atoh1/Math1 expression prevented Shh-driven medulloblastoma formation. Critically, Gli2 was confirmed as a direct transcriptional target of Atoh1/Math1 in the postnatal cerebellum, providing a previously unrecognized link between Shh ligand and Gli2 expression. Taken together with the work described earlier, these findings provide further support for Atoh1/Math1 as a rationale target for treating medulloblastomas, especially those driven by aberrant Shh signaling. 13

1.3.5 TP53, Cell Cycle, & DNA Repair

Long recognized as the ‘Guardian of the Genome’, TP53 remains the most frequent target of mutation in human cancer and loss of normal TP53 function can have significant consequences as a result of its fundamental role in regulating cell cycle arrest, cell death, genomic stability, and cellular senescence68. Although not mutated at high frequency in sporadic medulloblastomas, TP53 germline mutations are the underlying cause of Li-Fraumeni syndrome, and Li-Fraumeni patients are predisposed to a variety of cancers including medulloblastoma69. The ramifications of TP53 loss in medulloblastoma are exemplified in Ptc+/-;Tp53-/- compound mutants which develop posterior fossa tumours with almost complete penetrance by 12 weeks, compared to the ~14% penetrance observed in Ptc heterozygous mutants (at ~6 months) and a complete lack of cerebellar tumours observed in Tp53 null mice26. These observations strongly imply a cooperative role between aberrant Shh signaling and Tp53 loss-of-function, a trend that has been recapitulated in several non-Shh-driven models of medulloblastoma (see below).

Normal cerebellar development requires a tightly orchestrated succession of cellular events involving a period of massive proliferation followed by terminal differentiation. As such, proper control of the cell division cycle including cell cycle exit is necessary to avoid aberrant proliferation that may progress to tumourigenesis. In 2000, Marino et al published an important study that strengthened the hypothesis that medulloblastomas arise from the malignant transformation of CGNPs, and implicated the Rb pathway as an important regulatory unit that may be targeted in medulloblastoma70. By generating Gfap-cre;Rbloxp/loxp;Tp53-/- and Gfap-cre; Rbloxp/loxp;Tp53loxp/loxp mice, the authors conditionally disrupted Rb in the context of either germline or somatically inactivated Tp53 pathways in astrocytes and CGNPs, resulting in the development of medulloblastomas but not glial tumours on both backgrounds. These tumours formed in the EGL and were Math1 positive, strongly implicating CGNPs as the cell of origin for these medulloblastomas. Moreover, since RB and related family members (i.e. P107 and P130) are critical regulators of the G1/S phase of the cell cycle, the significance of proper cell cycle control in preventing medulloblastoma genesis was suggested. In a later study, Marino and colleagues showed that compound loss of Rb/Tp53 delayed the normal differentiation of CGNPs, predisposing them to genomic instability that led to recurrent amplification of Shh target genes Mycn, Gli2, and Ptch2 as secondary hits promoting transformation71. 14

Work by Martine Roussel’s group over the better part of the last decade has further reinforced the importance of cell cycle control in CGNP biology, also showing that defects in the cell cycle machinery promote medulloblastoma in mice. By crossing Ink4c-/- mice with Tp53-/- animals to generate Ink4c-/-;Tp53-/- double knockouts, Zindy et al reported the development of a variety of tumours including medulloblastomas with relatively low penetrance72. Ink4c (also known as p18Ink4c) is a member of the Ink4 family of cell cycle inhibitors (CKIs) that specifically bind and impair the function of cyclin dependent kinases (CDKs), keeping the cell cycle in ‘check’ and promoting cell cycle exit as cells differentiate. Murine Ink4c is transiently expressed in CGNPs of the EGL, peaking at ~P10 and completely extinguished by P20 coinciding with cell cycle exit and differentiation of CGNPs73. The onset of medulloblastomas arising in the EGL secondary to Ink4c loss-of-function in this model (i.e. Ink4c-/-;Tp53-/-), as well as on a Ptc heterozygous background (i.e. Ptc+/-;Ink4c+/- and Ptc+/-;Ink4c-/-) as reported in a follow-up study by this group74, mirror the results of the Rb knockout study mentioned above and further stress the role of cell cycle control in normal cerebellar development. More recently, genetic disruption of another CKI, Kip1 (also known as p27Kip1), was shown to collaborate with loss of Ptc to increase medulloblastoma incidence and accelerate tumour onset73. Despite these compelling data resulting from the use of genetically engineered mouse models, genetic mutations in RB (RB1, RBL1, and RBL2) and INK4 (i.e. CDKN2A, CDKN2B, CDKN2C, and CDKN2D) families as well as other genes encoding CKIs are scarce in primary human medulloblastomas, although evidence supporting epigenetic inactivation of some of these genes has been reported74-77.

Much like the studies described above have implicated proper cell cycle control in preventing medulloblastoma, a number of mouse models have similarly suggested DNA repair enzymes as medulloblastoma tumour suppressors, highlighting a necessity for the maintenance of genomic stability in suppressing medulloblastoma. Multiple models generated by Peter McKinnon’s group and others have functionally explored the role of DNA repair in medulloblastoma genesis, including disruption of Lig4, Xrcc2, Xrcc4, or Brca2 in nestin-expressing cells (or ubiquitously in some cases) in the context of Tp53 deficiency78-81. Mice of these genotypes all develop highly penetrant medulloblastomas originating in the EGL that are characterized by activated Shh signaling. Moreover, loss-of-function of the above candidates, as well as Parp1 in a similar study82, resulted in tumour aneuploidy, translocations, and recurrent copy number aberrations that targeted genes known to be important in human medulloblastomas such as Ptch1, Mycn, and 15

CyclinD2. Once again, although the DNA repair machinery has not been shown to be mutated at high frequency in human medulloblastoma (i.e. rare germline BRCA2 mutations in patients with Fanconi anemia)83,84, these models have provided valuable proof-of-principle that such processes are likely involved in protecting CGNPs and NSCs from acquiring genetic insults that may promote their aberrant proliferation and subsequent transformation.

1.4 Genomics of Medulloblastoma

Genomics involves the study of genes and their function, typically in the context of an organism, a tissue, or a particular cell type. Cancer is a genomic disease that accounted for an estimated ~640,000 deaths in the United States and Canada in 200885,86. The goal of cancer genomics is to develop a comprehensive inventory of the full spectrum of mutations, whether inherited or acquired that contribute to tumourigenesis. Ultimately, through a better understanding of the cancer genome, targeted treatment options may be developed and implemented such that mortality due to cancer can be reduced in the future.

The human genome consists of ~3 billion base pairs of DNA and encodes an estimated ~24- 25,000 unique protein-coding genes87,88. During tumourigenesis, a variety of different types of somatic mutation arise at the level of the genome, which collectively provide a selective growth advantage to cells harboring these mutations and promote the onset of cancer. Some examples of somatic mutation present in the cancer genome include single base pair substitutions, insertions and deletions of DNA segments, structural rearrangements such as duplications, inversions, and translocations, as well as gene amplifications and deletions89. Estimates from recent genome- wide sequencing efforts suggest that a given cancer may contain anywhere from 40 to over 100 somatic mutations90-92. Since these numbers do not directly account for genes affected by structural changes and copy number aberrations, the actual number of genes targeted for mutation in a given tumour is likely even higher. Beyond the genome, deregulation of the epigenome, including hypermethylation of gene promoters and changes to the histone code, also contributes to cellular transformation93-96. Collectively, these abnormal genomic and epigenomic states present in a cancer cell aberrantly impact gene expression, leading to the disruption of normal cellular processes, including cell division. Comprehensive cancer genomics, therefore, includes studies at the level of the genome, epigenome, and transcriptome. 16

As discussed earlier, much of our current understanding of the molecular basis of medulloblastoma has been derived from insight into hereditary tumour syndromes97 and candidate gene approaches focused on developmental signaling pathways17,98,99. Studies of the PTCH1 gene in Gorlin syndrome and sporadic medulloblastomas, as well as knockout studies of it’s mouse homolog, Ptch1, have helped establish a role for aberrant Shh signaling in ~25-35% of medulloblastomas12,17. Similarly, identification of APC mutations in Turcot syndrome, and more frequent mutations of CTNNB1 in sporadic cases, have implicated the Wnt signaling cascade in ~10-15% of medulloblastoma patients12,17. Furthermore, patients with Li-Fraumeni syndrome have germline TP53 mutations, and develop a broad spectrum of cancer types, including medulloblastoma69,100.

Aside from what has been learned from the study of these familial tumour syndromes, the vast majority of additional oncogenes and TSGs implicated in medulloblastoma have been discovered from a priori candidate gene selection. Mutational screening has further implicated additional Shh (SUFU, SMO) and Wnt (AXIN1) pathway genes12,17. In addition, the NOTCH pathway is deregulated in a subset of human medulloblastomas and activated in certain mouse models12,17. Furthermore, candidate epigenetic approaches have identified hypermethylation of the promoter regions of known TSGs, HIC1, RASSF1A, CASP8, and others57. The relevance of several of these genes has been further validated in mouse models of medulloblastoma101-103. Although these single-gene and/or candidate gene studies have shed significant light on our understanding of medulloblastoma pathogenesis, the candidates identified to date likely represent only a small “piece” of the genomic puzzle responsible for the onset and progression of this pediatric tumour. Indeed, recent data from whole-genome sequencing projects of multiple tumour types implicates as many as 100 mutated genes per genome90-92. If such estimates prove applicable to the medulloblastoma genome, many candidates have yet to be identified.

1.4.1 Early cytogenetics and karyotyping of medulloblastoma

It has been over 20 years since Giemsa banding (G-banding) was first used to disclose chromosomal abnormalities in medulloblastoma (see Figure 2 for a timeline outlining the application of genomics to medulloblastoma). G-banding is a classical staining technique employed to visualize a cell’s karyotype, producing an alternating pattern of dark (heterochromatin) and light (euchromatin) bands along metaphase chromosomes104,105. Early 17

Figure 2. Evolution of genomic technologies and their application to the study of medulloblastoma. Timeline schematically depicts advances made in genomics over the last two decades, with dates reflecting application of the referenced technologies to studies of the medulloblastoma genome. Listed below each time period is the approximate resolution (base pairs, bp) of the techniques shown during a given period. Details of the respective genomic technologies are described in more detail throughout the text. 19

studies conducted at Duke University Medical Center and The Children’s Hospital of Philadelphia provided original and informative descriptions of the medulloblastoma karyotype106-108. Of significant interest, isochromosome 17q (i{17}q) was reported as the most frequent structural abnormality by both of these groups, and in at least a few cases, the only aberration observed. I{17}q is the most common isochromosome in human cancer109, generating a net loss of one copy of the majority of 17p and a net gain in one copy of 17q (Figure 3a). Frequent loss of heterozygosity (LOH) on chromosome 17p in medulloblastoma was independently confirmed by multiple groups in the early 1990s, typically through deletion mapping by restriction fragment length polymorphism (RFLP) analysis110-114. At present, cytogenetic aberrations affecting chromosome 17 remain the most common structural changes noted in medulloblastoma (Table 2)115-117, however, insight into the individual gene or combination of genes on this chromosome driving tumourigenesis has not significantly improved since these early findings.

Establishment and cytogenetic characterization of permanent medulloblastoma cell lines and xenografts in the late 80’s and early 90’s also provided initial insight into the prevalence of oncogene amplification in medulloblastoma. Amplification of the MYC locus on 8q24, often in the form of DMs, was reported in multiple cell lines and confirmed in primary tumours by several groups118-122. MYC family proto-oncogenes (MYC, MYCN, and MYCL1) remain among the most prevalent targets of gene amplification in medulloblastoma (Figure 3b)115,117.

The application of comparative genomic hybridization (CGH) to the cytogenetic characterization of medulloblastoma in the late 90’s resulted in a much greater appreciation of the degree of genomic imbalance present in this cancer. Using CGH to profile a panel of 27 primary medulloblastomas, Reardon and colleagues described frequent losses on chromosomes 10q, 11, 16q, 17p, and 8p, as well as recurrent gains on chromosomes 7 and 17q123. Several complementary follow-up studies based on a combination of G-banding, CGH, spectral karyotyping (SKY), and fluorescence in situ hybridization (FISH), confirmed these now well- recognized regions of genomic instability in medulloblastoma, and shed light on additional candidate regions124-131. An innovative report that involved members of our group, retrospectively analyzed a series of 19 primary medulloblastomas (in addition to 5 sPNETs) 20

Figure 3. Prominent genomic aberrations in medulloblastoma. (A) Isochromosome 17q (i{17}q) in medulloblastoma. Single nucleotide polymorphism (SNP) array copy number profile for a medulloblastoma patient with a characteristic i{17}q abnormality. I{17}q is the most common cytogenetic aberration in medulloblastoma, identified in ~30-40% of cases. This structural abnormality results in a net loss of one copy of chromosome 17p and net gain in one copy of 17q. Chromosome 17p loss and q gain are indicated in the copy number plot with green and red arrows, respectively. (B) MYC family amplification in medulloblastoma. SNP array copy number output showing focal, high-level amplification of MYC (8q24; upper panel), MYCN (2p24; middle panel), and MYCL1 (1p34; lower panel) in primary medulloblastomas. MYC family proto-oncogenes are collectively targeted for amplification in ~10% of primary medulloblastoma cases, more frequently than any other known oncogenes. 21

using classical G-banding,FISH, CGH, and SKY125. SKY is a multicolored FISH procedure that permits the identification of structural rearrangements and origins of marker chromosomes in the genome in a single experiment132. This “chromosome painting” technique is particularly useful for detecting structural aberrations lacking a net change in copy number, such as balanced translocations. The use of SKY in this study enabled the comprehensive identification of recurrent structural rearrangements in medulloblastoma, including those on chromosomes 7, 17, 3, 14, 10, and 22, something not possible through the use of G-banding or CGH alone.

Although these cumulative efforts provided the pediatric brain tumour community with relatively detailed summaries of the medulloblastoma karyotype, it wasn’t until the advent of new technologies capable of detecting copy number changes at a much higher resolution that novel candidate oncogenes and tumour suppressors in medulloblastoma could be more efficiently identified through the use of genomics. Over the past 5 years, novel, high-resolution (i.e. sub- megabase) genomic technologies have become available104,133,134. Applications of some of these technologies in studies of the medulloblastoma genome are discussed in detail below.

1.4.2 High-resolution genomics of medulloblastoma: digital karyotyping, array-CGH (aCGH), and single nucleotide polymorphism (SNP) genotyping arrays

High-resolution genomic profiling of medulloblastoma has recently implicated multiple candidate oncogenes that are recurrently amplified in this malignancy (Table 3). In 2005, two very similar but independent studies led by Greg Riggins and Hai Yan, used digital karyotyping to identify novel regions of copy number aberration in the medulloblastoma genome135,136. Digital karyotyping employs short sequence tags derived from specific genomic loci to provide a quantitative and relatively high-resolution profile of copy number aberrations throughout the genome137,138. Boon et al karyotyped 5 medulloblastoma cell lines by sequencing ~200,000 genomic tags per genome, identifying amplification of the OTX2 homeobox gene on chromosome 14q22 in the D487Med cell line135. Using quantitative PCR (qPCR), the authors confirmed recurrent amplification of OTX2 in both medulloblastoma cell lines (D425Med) and primary tumours. In the parallel study published by Di et al, OTX2 amplification was revealed in the D458Med cell line, also by digital karyotyping136. In addition, using data from serial analysis of gene expression (SAGE) libraries and qRT-PCR, OTX2 was shown to be specifically overexpressed in medulloblastomas, especially those of anaplastic histology, as compared to a wide 23

variety of other malignancies. Furthermore, inhibition of OTX2 expression by siRNA-mediated knockdown or all-trans-retinoic acid (ATRA) repressed medulloblastoma cell growth in vitro, suggesting OTX2 may represent an attractive target for therapy, particularly in medulloblastomas of the anaplastic subtype.

Among the first studies to apply aCGH technology to medulloblastomas, was a profile of 47 primary cases reported by Mendrzyk et al139. In this effort, typical cytogenetic abnormalities including gain of chromosomes 17q, 7, and 1q, as well as loss of 17p, 11p, 10q, and 8 were confirmed. Importantly, a minimal region of recurrent, high-level amplification targeting the CDK6 proto-oncogene on chromosome 7q21.2 was also identified. The author’s validated CDK6 copy number aberrations by FISH and went on to establish a negative correlation between moderate-high CDK6 protein expression and overall survival by measuring CDK6 status on a medulloblastoma tissue microarray (TMA).

FOXG1 is another candidate gene that has been implicated in medulloblastoma pathogenesis owing to its recurrent gain on 14q12 as revealed by aCGH analysis140. Adesina et al analyzed a small panel of medulloblastomas using a combination of conventional CGH (n=19) and aCGH (n=9) and reported gain of the FOXG1 locus in 6/9 cases in the test set and 55/59 cases in a validation series of tumours. Expression of FOXG1 was correlative with gene copy number and inversely correlated with p21 protein levels, a relationship that was strengthened in vitro as siRNA-mediated knockdown of FOXG1 in DAOY medulloblastoma cells resulted in increased p21 expression.

Amplifications of MYCL1, PDGFRA, and KIT, all proto-oncogenes not previously reported to be targeted in medulloblastoma, were also noted using aCGH technology141. Amplicons targeting these cancer genes have since been observed in our SNP array studies (see Chapter 2 of this thesis) and by others, suggesting they are relevant oncogenes in medulloblastoma115,142.

An earlier aCGH study of medulloblastoma analyzed a series of 16 primary cases and 3 medulloblastoma cell lines143. The authors noted a novel region of homozygous deletion on chromosome 6q23 in the DAOY cell line that targeted only 2 previously uncharacterized genes, both of which exhibited reduced expression in a large percentage of primary medulloblastomas analyzed. Our group has since validated this region of homozygous deletion in DAOY and 25

functionally confirmed L3MBTL3 as a putative medulloblastoma TSG mapping to this locus (see Chapter 2 of this thesis)115.

As detailed earlier, medulloblastomas can be histologically classified into five recognizable subtypes: classic, desmoplastic, anaplastic, large-cell, and MBEN12. Although there is considerable variability in terms of patient outcome between the different histological subtypes and histological staging has proven to be a less than ideal method of stratification, there remains a great deal of interest in defining their molecular basis. To gain an improved understanding of the genomics of desmoplastic medulloblastomas, Ehrbrecht et al. performed conventional CGH on a set of 22 sporadic cases of this subtype, followed by aCGH on a subset144. In their analysis, novel regions of amplification were reported on chromosomes 9p and 17q22-24, implicating candidate oncogenes in these regions. Notably, JMJD2C was suggested as a putative oncogene mapping to the amplified region found on 9p, another candidate we have since identified as recurrently amplified and over-expressed in an independent sample cohort and further shown to impact the state of methylation on histone lysines in CGNPs of the developing cerebellum (see Chapter 2 of this thesis)115.

There have been an impressive number of inquiries into the relationship between developmental signal transduction pathways and their role in medulloblastoma. Mutations in the Wnt, Shh, and NOTCH pathways have all been well described in the medulloblastoma literature12,17. Despite this, a comprehensive understanding of how specific genomic events contribute to aberrant signaling of these pathways has not been established. An important finding relevant to deregulated Wnt signaling in medulloblastoma was reported in 2006 in two independent but related studies37,145. Clifford et al profiled 19 primary medulloblastomas by aCGH, with a specific intent on genomically describing tumours with Wnt pathway activation (nuclear - Catenin; CTNNB1 or APC mutation)145. Interestingly, in both the initial cohort (n=19) and a validation series (n=32), single copy deletion of chromosome 6 (monosomy 6) was found exclusively in the Wnt pathway tumours. Identical findings were reported by Thompson et al, who consistently observed a correlation between the Wnt pathway signature (Wnt pathway expression; CTNNB1 mutation) and markedly reduced expression of genes mapping to chromosome 6 secondary to deletion37. Monosomy 6 is now widely accepted in the medulloblastoma community as a genomic marker of Wnt pathway tumours that is consistently associated with CTNNB1 mutation37,38,146. Critically, from a clinical perspective, monosomy 26

6/CTNNB1 mutation is among the most reliable genetic markers in medulloblastoma, correlating with a highly favorable prognosis9,117,145,146. Indeed, 100% of patients determined to belong to the “WNT” immunohistochemical category in the recent St Jude Medulloblastoma-96 (SJMB- 96) clinical trial were event-free at 5-years, compared to only 65% of patients in the “SHH” category9.

Very recently, Pfister and colleagues proposed a model for molecular risk stratification of pediatric medulloblastoma based on DNA copy number aberrations affecting chromosomes 6q, 17q, and MYC/MYCN loci117. Using aCGH, the authors initially profiled 80 primary medulloblastomas in an attempt to identify genomic aberrations of prognostic value, and found gain of chromosome 6q, amplification of MYC and MYCN, isolated gain of 17q, and i{17}q all to be associated with a poor clinical outcome. In contrast, loss of chromosome 6q was indicative of an excellent prognosis, consistent with the current literature145,146. Validation of these prognostic markers in a non-overlapping set of 260 primary cases by interphase FISH on a medulloblastoma TMA, Pfister et al. were able to establish an elegant staging system whereby patient outcome could be predicted based on the genomic status of only 4 markers (arranged from worst to best outcome): MYC/MYCN amplification, 6q gain, 17q gain, 6q/17q balanced, and 6q loss117.

Although several of the aCGH studies described above have been informative and enhanced our understanding of the medulloblastoma genome, most have profiled relatively modest sample cohorts (median sample size (n): ~24) using arrays that, although have been an improvement from classical CGH, have been of insufficient density and thus resolution (median resolution (Kb): ~500) to detect very focal genetic events. To address these caveats, we retrospectively collected an unprecedented cohort of 201 fresh-frozen primary medulloblastomas and 11 medulloblastoma cell lines and analyzed their genomes using high-resolution SNP genotyping arrays115. Results from this genomic profiling effort confirmed that deregulation of the histone code, particularly histone lysine methylation, contributes to the pathogenesis of medulloblastoma – see Chapter 2 of this thesis.

1.4.3 Medulloblastoma transcriptome profiling

Classically, strategies aimed at transcriptional profiling of cancer have involved the comparison of gene expression signatures obtained for normal and neoplastic tissues (Figure 4). In one of the earliest studies of medulloblastoma gene expression profiling, Michiels et al. used SAGE to 27

compare genes expressed in medulloblastoma to those in fetal brain (24.5 weeks)147. SAGE uses DNA sequencing technology to digitally quantify mRNA abundance by counting the frequency in which a short sequence tag (i.e. transcript) appears in a cDNA library148,149. SAGE technology has been shown to be very effective in quantifying gene expression and in the identification of novel genes/transcripts, as no a priori knowledge of the genome under study is required150-153. In the study by Michiels and colleagues, ~10,000 tags were sequenced in both medulloblastoma and fetal brain, with ~6000 unique genes identified in each transcriptome. Comparison of the SAGE data revealed 138 genes with significant differential expression between the two sources, including ZIC1 and OTX2, both showing significantly elevated expression in medulloblastoma that was confirmed by Northern blotting in multiple independent samples. As these genes are highly expressed in cerebellar germinal zones (i.e. EGL and SVZ), this study provided early clues into the origins of medulloblastoma.

The presence of disseminated disease at diagnosis is a strong, independent marker of poor prognosis for medulloblastoma patients, occurring in about one out of every three cases9,154-156. Understanding the molecular basis of metastatic medulloblastoma is therefore of extreme clinical importance. The first study to specifically compare metastatic (M+) to non-metastatic (M0) medulloblastoma at a gene expression level, was published by MacDonald and colleagues in 2001157. Twenty-three primary medulloblastomas designated as either M+ or M0 were analyzed using Affymetrix G110 cancer arrays, identifying 85 genes as differentially expressed between the two classes. Using a supervised class prediction algorithm, this 85-gene signature classified the M+ and M0 tumours with 72% accuracy. Of interest, PDGFR and members of the RAS/MAPK signaling cascade were reported as significantly upregulated in metastatic versus non-metastatic cases. Over-expression of PDGFR in metastatic disease was confirmed in an independent set of tumours by immunohistochemistry. In vitro assays performed in the DAOY medulloblastoma cell line showed that the PDGF ligand activated the RAS/MAPK pathway and promoted cell migration in this system, whereas neutralizing antibodies against PDGFR attenuated MAPK signaling and prevented ligand-mediated migration. In a follow-up study by Gilbertson and Clifford, the association between PDGFR over-expression and metastatic medulloblastoma was confirmed158, further supporting the validity of this pathway as a candidate for targeted therapy. 28

Figure 4. Transcriptional profiling of medulloblastoma. Heatmap showing gene expression data for 10 normal cerebellum samples (5 fetal, 5 adult) and 50 primary medulloblastomas analyzed using Affymetrix arrays. Genes exhibiting elevated expression are shown in red, whereas genes with reduced expression are depicted green. Unsupervised hierarchical clustering of samples using the most differentially expressed genes results in clear distinction between normal cerebellar samples and primary tumours. 30

Undoubtedly one of the most important papers in medulloblastoma genomics is the study by Pomeroy and colleagues, who surveyed the expression profiles of a large series of primary brain tumours and made at least 3 important findings of clinical significance159. Initially, 42 patient samples consisting of medulloblastomas (n=10), atypical teratoid rhabdoid tumours (AT/RTs; n=5), renal and extrarenal rhabdoid tumours (n=5), sPNETs (n=8), and non-embryonal brain tumours (malignant glioma; n=10), as well as normal human cerebella (n=4) were analyzed using Affymetrix HuGeneFl arrays containing ~6,000 known genes. Based on differentially expressed transcripts, the authors showed clear distinction between the different tumour types, establishing that histologically similar tumours such as medulloblastomas and sPNETs are molecularly distinct. The molecular distinction of medulloblastomas from sPNETs has important clinical implications, as, owing to their similar histology, these tumours were historically labeled under the same broad category of PNET and as a result treated with the same protocols160. In recent years, it has become evident that medulloblastomas and sPNETs are both molecularly and biologically distinct, with sPNETs typically exhibiting a worse prognosis141,161- 165.

Secondly, the authors compared the expression profiles in a set of 34 medulloblastomas of either classic (n=25) or desmoplastic (n=9) histology, revealing a significant degree of statistically significant differential expression between the two subtypes. Genes signifying desmoplastic medulloblastoma included PTCH, GLI, MYCN, and IGF2, all of which are now well-described targets of Shh signaling. Although a link between mutations in the Shh pathway and medulloblastoma pathogenesis had already been discovered, this study was among the first to report an association between aberrant Shh signaling and sporadic desmoplastic medulloblastoma.

A third key finding of this study stemmed from the authors’ use of gene expression data to predict the outcome of 60 medulloblastoma patients for whom clinical follow-up was available. Using a class prediction algorithm, an 8-gene classification model was generated that successfully predicted survival status for 47/60 patients profiled. Genes correlated with favorable outcome included markers of cerebellar differentiation (-NAP, NSCL1, and TRKC) as well as genes encoding components of the extracellular matrix (PLOD lysyl hydroxylase, collagen type VI, and elastin). In contrast, genes associated with poor outcome included those with a role in cell proliferation and metabolism (MYBL2, enolase 1, LDH, HMG1 (Y), 31

cytochrome C oxidase) and ribosomal protein-encoding genes. Much like the study by MacDonald et al. described earlier, this report demonstrates the utility of correlating gene expression profiles in medulloblastoma with a particular phenotype (i.e. favorable versus poor outcome) and provides rationale for the incorporation of similar molecular profiling strategies in the future diagnosis and treatment of medulloblastoma patients.

Following the Pomeroy study, a number of independent groups, including our own, have engaged in medulloblastoma transcriptome profiling efforts using a variety of technologies166-170. Boon et al used SAGE to analyze 20 primary medulloblastomas, identifying 30 transcripts exhibiting elevated expression in tumours compared to normal cerebellum and additional regions of the brain169. CD24, prolactin, TOP2A, MYCN, BARHL1, and the cancer-testis antigen (CTA) PRAME were all identified as over-expressed in this study. More recent array-based studies by our group have confirmed aberrant expression of CTAs from the MAGE and GAGE families in medulloblastoma cell lines and in some primary samples, suggesting these genes may be of importance in medulloblastoma166.

An earlier study by our group used suppression subtractive hybridization (SSH) to identify genes deregulated in both human and mouse (Ptch+/-) medulloblastoma compared to the normal, species-matched cerebellum170. In SSH, double-stranded cDNA libraries are first prepared from tester (i.e. medulloblastoma) and driver (i.e. normal cerebellum) RNA samples171. Heat denatured tester cDNA is subsequently digested, adapter-ligated, and then hybridized with the denatured driver cDNA to generate a subtracted cDNA library that is PCR amplified and cloned into a recipient plasmid for bacterial transformation and sequencing of clones for gene identification. In this effort, over 100 up-regulated cDNA fragments were identified in the human library, including ULIP, SOX4, Neuronatin, and the previously implicated BARHL1 and OTX2 genes. In addition, genes identified as upregulated in medulloblastomas from Ptch+/- mice included cyclinD2, thymopoietin, Musashi-1, and others.

Another informative expression profile of medulloblastoma was generated by Neben et al, whom analyzed ~4200 genes in 35 primary medulloblastomas in an attempt to identify candidates associated with patient outcome168. Based on mRNA levels, 54 genes were shown to be markers of poor outcome, and a subset (n=9) of these were further evaluated by immunohistochemistry in a non-overlapping set of 180 cases on a medulloblastoma TMA. Of these candidate genes, 32

STK15 positivity was identified as a negative prognostic marker of overall survival, whereas other putative markers implicated in the study (i.e. cyclin D1, stathmin 1) were not.

1.4.4 Molecular classification of medulloblastoma

Over the past decade, significant progress has been made in how we study the cancer genome. Indeed, gene expression profiling has proven to be an effective tool for the molecular classification of cancer, including brain tumours159,172-174. Following the studies of Macdonald and Pomeroy, Thompson et al. were the first to truly establish the existence of unique molecular subgroups of medulloblastoma using gene expression data37. Profiling a series of 46 primary medulloblastomas, the author’s performed unsupervised hierarchical clustering with the most informative genes in the dataset, identifying 5 molecular subgroups of medulloblastoma. By integrating immunohistochemistry, FISH, and mutational screening data generated from these samples, it was shown that molecular subgroups of medulloblastoma have specific genomic and genetic features. Importantly, this study was the first to demonstrate that WNT (i.e. monosomy 6, CTNNB1 mutation) and SHH tumours (i.e. PTCH1, SUFU mutation) are mutually exclusive.

A more recent study by Kool et al., performed a similar integrative genomics approach to further characterize molecular subgroups of medulloblastoma38. By combining array-based gene expression and copy number profiles for 52 primary cases, Kool and colleagues recapitulated the 5 molecular subgroups described by Thompson et al and correlated the different subgroups with specific genomic and clinical features. Importantly, the author’s furthered our knowledge of non-WNT/SHH tumours (subgroups A/B), showing that the three remaining subgroups (C/D/E) are closely related, and marked by elevated expression of neuronal differentiation (subgroups C/D) and retinal (subgroups D/E) genes. Furthermore, metastatic disease was shown to be more highly associated with subgroups C/D/E, providing further support for the potential stratification of patients based on molecular subgrouping.

In Chapters 3 and 4 of this thesis, we report 4 unique molecular subgroups of medulloblastoma that are genetically, transcriptionally, and clinically distinct. Moreover, we establish an efficient and universal classification scheme for medulloblastoma subgroups based on immunohistochemical staining with only 4 antibodies (Chapter 3). 33

1.4.5 Beyond protein-coding genes: microRNAs (miRNAs) in medulloblastoma

Over the course of the past five or so years, there has been a literal ‘explosion’ in the miRNA field, especially with respect to elucidating their role in human disease, in particular cancer175-177. MiRNAs are small, non-coding, single-stranded RNA molecules that post-transcriptionally regulate gene expression through their interaction with complementary sequences in the 3’ untranslated regions of target mRNAs178,179. Target mRNAs are either degraded or translationally repressed by specific miRNAs, depending on the degree of complementarity existing between the miRNA and its target (Figure 5). Despite an intense amount of investigation into the involvement of miRNAs in a variety of cancer types, knowledge of their role in medulloblastoma pathogenesis is still in its infancy. The few studies aimed at miRNA characterization in medulloblastoma reported to date are discussed below (Table 4).

In one of the initial global profiling efforts of miRNAs in medulloblastoma, Ferretti and colleagues performed Taqman qRT-PCR-based profiling of 248 miRNAs on a modest cohort of primary medulloblastomas (n=14), reporting consistent downregulation of miRNAs in tumours versus normal controls180. A total of 86 miRNAs were further evaluated in a larger series of tumours (n=34), with miR-9 and miR-125a, two downregulated candidates’ chosen for functional studies. Both miR-9 and miR-125a were induced by retinoic acid treatment of D283 medulloblastoma cells, an agent known to inhibit medulloblastoma cell proliferation. In addition, ectopic expression of miR-9 and miR-125a slowed medulloblastoma cell growth and impaired anchorage independence. Truncated TrkC (t-TrkC) was implicated as a target for post- transcriptional repression by both miR-9 and miR-125a in this study, providing a possible explanation for the observed cellular phenotypes. Importantly, 7/11 miRNAs reported as upregulated in medulloblastoma compared to controls corresponded to miR-17/92 and related paralogs (see below and Chapter 4 of this thesis). In a computational-based approach to predict miRNAs targeting a candidate mRNA, Pierson et al identified miR-124a as a negative regulator of the CDK6 proto-oncogene181, which is known to be amplified in medulloblastoma115,139. MiR-124 is the most abundant neuronal miRNA, playing a critical role in neurogenesis and neuronal differentiation182-184. Expression of miR-124a was shown to be significantly reduced in medulloblastomas compared to normal controls in this study, a trend similar to that observed in adult brain tumours185. Furthermore, ectopic miR-124a expression reduced CDK6 protein levels 34

Figure 5. Overview of miRNA biogenesis. MiRNAs are initially transcribed as long primary transcripts (pri-miRNAs) by RNA polymerase II, before being processed by Drosha in the nucleus to generate pre-miRNAs (~80 nt), which form a characteristic stem-loop conformation. Pre-miRNAs are exported from the nucleus into the cytoplasm where they are further processed by Dicer to form the mature miRNA (~22 nt). Mature miRNAs post-transcriptionally regulate target gene expression by forming hybrid complexes with target mRNAs in their 3’ untranslated region (UTR), and subsequently repress translation or promote degradation of the target mRNA, depending on the amount of complementarity that exists between the miRNA and its target(s). 36 37

in DAOY medulloblastoma cells and slowed the growth of both DAOY and D283 cell lines181. Using GLI1 status to discriminate Shh-driven medulloblastomas from non-Shh cases, Ferretti et al. stratified a series of 31 primary medulloblastomas into 2 classes (GLI1high and GLI1low) and performed Taqman-based expression profiling on a subset 250 miRNAs186. This approach identified a set of 34 miRNAs demonstrating significant differential expression between the 2 groups. Three candidates exhibiting reduced expression in GLI1high tumours, miR-125b, miR- 324-5p, and miR-326 were chosen for functional study based upon their predicted capacity to repress SMO and GLI1 transcripts. Indeed, all 3 candidates were shown to downregulate SMO levels when overexpressed in DAOY medulloblastoma cells186. Additionally, expression of these candidate miRNAs correlated with CGNP differentiation state in vitro and their exogenous expression reduced Shh-mediated proliferation and promoted neurite outgrowth in the same cell type. It is important to note that miR-17/92 was not reported as differentially expressed between GLI1high and GLI1low medulloblastomas in this study. This result may be explained by our observation of miR-17/92 aberrancy in both Shh and non-Shh-driven (i.e. Group C and WNT) tumours (see Chapter 4 of this thesis).187

Much like the Shh pathway, the NOTCH signal transduction cascade is known to be important in both Shh-mediated cerebellar development and in medulloblastoma pathogenesis 17,188. For example, the Hatton group has shown that Notch signaling maintains CGNP proliferation, and that Shh treatment of CGNPs induces expression of the Notch effector Hes1189. To investigate miRNAs negatively regulating the NOTCH pathway in the context of medulloblastomas, Garzia et al recently performed an in-silico analysis of miRBase to identify miRNAs predicted to target and repress HES1190. This candidate approach identified miR-199b-5p as capable of binding the 3’ UTR sequence of HES1 and repressing HES1 expression in medulloblastoma cell lines. In addition, stable expression of miR-199b-5p reduced medulloblastoma cell proliferation and anchorage-independent cell growth in vitro and impaired tumour formation in mouse xenograft models. Of potential clinical relevance, qRT-PCR analysis of miR-199b-5p in a panel of 61 medulloblastoma samples showed a statistically significant association between miR-199b-5p downregulation and metastasis. Although not directly addressed, the authors suggest the observed downregulation of miR-199b-5p in metastatic medulloblastomas may be due to epigenetic silencing. 38

In a very recent study led by Steve Clifford’s group, a novel region of amplification adjacent to the MYC locus on chromosome 8q24 was identified in two medulloblastoma cell lines using SNP genotyping arrays191. The 3 Mb minimal common region of this amplicon contained three coding genes (ZFAT1, LOC286094, KHDRBS3) and two miRNAs (miR-30b and miR-30d). The author’s went on to establish KHDRBS3, miR-30b, and miR-30d as the most pertinent targets, as these candidates were highly upregulated at the transcript level in cell lines harboring the amplicon, whereas the other genes mapping to the locus were not. Although the authors did not find amplification of these candidates in primary samples from their cohort, all were upregulated in a percentage of primary cases at the expression level compared to normal cerebellar controls, suggesting they may be transcriptionally regulated by alternative mechanisms, and may represent relevant targets in medulloblastoma.

As described in this thesis, we recently performed high-resolution SNP array profiling on a cohort of >200 medulloblastomas, which revealed recurrent, high-level amplification of the oncogenic miR-17/92 polycistron115,187. In Chapter 4, we establish miR-17/92 as a novel putative oncogene in medulloblastoma and provide evidence in support of miR-17/92 as a critical player in medulloblastoma pathogenesis.

Concurrent with our studies of the human medulloblastoma miRNAome discussed in Chapter 4, a complimentary effort aimed at identification of miRNAs deregulated in mouse models of medulloblastoma was reported192. Uziel et al performed unbiased next-generation (next-gen) sequencing to determine miRNA abundance in tumour cells isolated from spontaneous medulloblastomas of either Ink4c-/-;Ptch+/- or Ink4c-/-;Tp53-/- genotypes as compared to wild-type control cerebella (1 MO) and CGNPs (P6)192. This approach identified 26 miRNAs with elevated expression and 24 exhibiting reduced expression in the tumour models compared to the non-neoplastic controls. Among up-regulated miRNAs in murine medulloblastoma cells, miR- 17/92 and related paralogs accounted for a considerable number (9/26). The use of next-gen sequencing to quantify miRNA expression was powerful in this study, as it permitted discrimination between miR-17/92 and its highly related paralogs, miR-106a/363 and miR- 106b/25, and confirmed miR-17/92 as the most aberrantly expressed of the three clusters. Although miR-17/92 was shown to be upregulated in tumours from both Ink4c-/-;Ptch+/- and Ink4c-/-;Tp53-/- genotypes in this study, the author’s postulated miR-17/92 expression may be cooperating with Shh signaling. Indeed, profiling miR-17/92 expression by qRT-PCR in a 39

collection of previously characterized human medulloblastomas (5 Shh cases, 5 non-Shh cases)37, revealed significant over-expression of miR-17/92 in samples with a Shh signature.

To evaluate the oncogenic potential of the miR-17/92 cluster in primary CGNPs, the same group retrovirally expressed miR-17/92 in P6 CGNPs from both Ink4c-/-; Ptch+/- and Ink4c-/-; Tp53-/- mice and orthotopically transferred transduced cells into immunocompromised mice. Of interest, only cells derived from the Ptch+/- background developed medulloblastomas (9/9) in this model. These tumour cells were sensitive to cyclopamine, exhibited elevated Math1 and Gli1 expression, and lost expression of the wild-type Ptch allele, all indicative of active Shh signaling and suggestive of a collaborative relationship between miR-17/92 and Shh in these tumours.

1.4.6 Beyond genomics: the medulloblastoma epigenome

Until recently, the majority of cancer research efforts had focused on describing the genetic basis of cancer, studying everything from large cytogenetic aberrations to SNPs and mutations. However, over the past few years, there has been an ever-growing amount of literature linking the deregulation of epigenetics to malignancy93-96,193. Epigenetics is defined as mitotically heritable changes in gene expression that are not accompanied by modifications in primary DNA sequence. Epigenetic modifications include DNA methylation on cytosine residues, most often in the context of CpG dinucleotides, as well as post-translational modification of histone proteins, including methylation, acetylation, phosphorylation, and ubiquitination194,195. Hypermethylation of CpG islands located at the 5’ end of genes has been reported in most cancers and, either alone or in combination with genetic mechanisms (i.e. gene deletion, mutation), can contribute to TSG silencing. Although, a handful of known tumour suppressors have been confirmed to be silenced by promoter methylation in medulloblastoma using candidate gene approaches (i.e. HIC1, RASSFIA, CASP8)196-203, application of unbiased, whole-epigenome strategies to identify novel candidates have been scant to date, consisting of only the few published reports mentioned below.

Among the earliest studies to implicate aberrant promoter methylation in medulloblastoma on a global scale, was an effort led by Fruhwald and colleagues who employed the technique of Restriction Landmark Genomic Scanning (RLGS) to analyze DNA methylation patterns in 17 primary medulloblastomas and 5 medulloblastoma cell lines75. Using this method, the authors identified methylation in up to 1% of all CpG islands in primary tumours and 6% in 40

medulloblastoma cell lines. In addition, an association between hypermethylated sequences in medulloblastoma and poor prognosis was implied. Collectively, these findings provided early evidence that epigenetic events are likely to play a role in medulloblastoma pathogenesis.

In a study using microarray-based differential methylation hybridization (DMH), Waha et al identified hypermethylation of the secretory granule neuroendocrine protein 1 gene (SGNE1/7B2) in 16/23 (~70%) primary medulloblastomas and 7/8 (~87%) medulloblastoma cell lines investigated204. DMH involves a series of enzymatic digestions with methylation- insensitive followed by methylation-sensitive restriction enzymes, and uncut (methylated) fragments are PCR-amplified prior to hybridization to microarrays containing probes designed to interrogate CpG islands throughout the genome205. Expression of SGNE1 was determined to be downregulated in the majority of primary samples and cell lines compared to normal cerebellar controls, and SGNE1 transcription was restored in cell lines treated with the demethylating agent, 5-aza-2’-deoxycytidine (5-aza). Furthermore, re-expression of SGNE1 in the D283Med cell line resulted in growth suppression and reduced colony formation; suggesting SGNE1 may be a putative TSG in medulloblastoma.

An innovative report by Pfister and colleagues developed and applied a technique known as array-based profiling of reference-independent methylation status (Array-PRIMES) to globally survey DNA methylation patterns in the medulloblastoma genome206. Array-PRIMES compares two differentially digested (methylation-sensitive and methylation-specific) aliquots from the same sample genome by competitive hybridization to a CpG island microarray. The advantage of using test-versus-test as opposed to test-versus-control hybridization is that influences of tissue-specific methylation that may be present in the control sample, as well as genomic aberrations that may exist in the test sample and not in the control genome, are avoided. Using this methodology, Pfister et al showed a striking association between samples classified as either “low methylators” or “high methylators” and patient outcome, with the high methylator group exhibiting reduced overall survival. In addition, the GLI C2H2-type zinc-finger protein family member, ZIC2, was identified as a hypermethylated candidate by aPRIMES that was subsequently confirmed to be epigenetically silenced in a panel of primary medulloblastomas using a combination of pyrosequencing and qRT-PCR analysis. 41

In two technically similar yet independent genome-wide methylation studies conducted by Anderton and Kongkham, 5’-Aza-treated medulloblastoma cell lines were profiled on Affymetrix expression arrays in an effort to uncover novel TSGs silenced by aberrant promoter methylation43,207. In the report by Anderton et al, 3 medulloblastoma cell lines (D425Med, D283Med, and MED8A) were either left untreated or exposed to 5’-Aza and transcripts showing increased expression by microarray in response to the DNA methyltransferase inhibitor were investigated further by bioinformatically confirming the presence of a 5’ CpG island and assessing methylation status by bisulphite sequencing207. This approach, combined with gene expression analysis, identified COL1A2 as an epigenetically silenced candidate in medulloblastoma that is preferentially inactivated in non-desmoplastic and non-infant (>3 years) desmoplastic cases.

Kongkham et al performed a similar genome-wide 5’-Aza screen in a larger cohort of 9 medulloblastoma cell lines, but incorporated multiple additional criteria compared to the previous study when filtering identified candidate genes43. Genes demonstrating >2-fold upregulation in expression following 5’-Aza treatment, in at least 2 cell lines, with predicted CpG islands in the promoter region, and which were identified as targets of LOH based on SNP genotyping studies115, were selected for further analysis. Under these criteria, SPINT2, a negative regulator of the HGF/MET signaling pathway, was identified, exhibiting robust re- expression in 6/9 medulloblastoma cell lines profiled. The author’s confirmed downregulation of SPINT2 in a significant percentage of primary medulloblastomas (>2-fold in 41/56 samples) analyzed by qRT-PCR, and importantly, confirmed that aberrant promoter methylation (assessed by methylation-specific PCR) correlated with the observed reduction in gene expression in most cases. Stable re-expression of SPINT2 in medulloblastoma cell lines resulted in attenuation of the malignant phenotype, inhibiting cell proliferation, anchorage-independent growth in soft agar, and cell motility. Furthermore, orthotopic transplantation of D283 cells stably re- expressing SPINT2 into recipient nude mice significantly delayed time to mortality compared to empty vector control cells in an intracerebellar xenograft model. These data strongly implicate SPINT2 as a putative TSG in medulloblastoma and shed further light on the apparent role of aberrant HGF/MET signaling in medulloblastoma aetiology.

Collectively, these recent studies of the medulloblastoma epigenome have proven informative and further implicated epigenetic gene silencing as an important mechanism of TSG inactivation 42

in medulloblastoma. Future application of strategies that enrich for epigenetic modifications (i.e. methylation-dependent immunoprecipitation, MedIP; chromatin immunoprecipitation, ChIP) combined with high-resolution microarrays or next-gen sequencing technologies will likely lead to an improved appreciation of the role epigenetics plays in medulloblastoma.

1.5 Summary

Over the last two decades, a great deal of progress has been made in our understanding of the biology and genomics underlying medulloblastoma. Improvements in our knowledge of normal cerebellar development and the signal transduction pathways controlling this process have provided important insight into the molecular and cellular events involved in the malignant transformation that leads to medulloblastoma. Additionally, major clues into the origins of this disease have spawned from dissecting the aetiology of heritable tumour syndromes and been validated through the use of genetically engineered mouse models. Furthermore, genome-wide interrogation of the medulloblastoma genome, transcriptome, and epigenome has identified new genes and pathways that are mutated and aberrant in medulloblastoma, potentially implicating novel markers for future targeted therapies. Finally, the concept of molecular subtypes of medulloblastoma has recently arisen, providing hints into the different cellular origins of the disease and showing potential for improved patient stratification.

Despite these significant advances, much remains to be learned about this disease if more effective treatment options are to be developed in years to come. Indeed, the majority of what we currently know about the molecular basis of medulloblastoma is based on the Shh and Wnt pathways, which account for less than ~40% of cases, and conventional inhibitors of these pathways are only likely to be effective in certain contexts. The remaining ~60% of medulloblastomas remain very poorly understood and exhibit the worst prognosis. It is therefore critical that we develop a comprehensive understanding of the molecular basis of all medulloblastoma subtypes to allow for improved patient stratification and the development of rational targeted therapies for treating each of the individual subtypes in the future. 43

1.6 Hypothesis

The molecular biology underlying the majority of medulloblastomas remains poorly understood. Past studies of the medulloblastoma genome have been hampered by small sample sizes and the employment of low-resolution technologies. It is my hypothesis that application of modern high- resolution genomics to a large cohort of medulloblastomas will identify novel oncogenes and TSGs contributing to the pathogenesis of medulloblastoma and provide new insight into the genomics of medulloblastoma subtypes. 44

Chapter 2 High-Resolution Genomics of Medulloblastoma 2 * 2.1 Introduction

Over the course of the last two decades, we have gained considerable insight into the molecular genetics and genomics underlying medulloblastoma pathogenesis208. Frequent cytogenetic aberrations such as i{17}q and high-level amplification of MYC family proto-oncogenes are well-characterized in medulloblastoma, having been consistently reported since the early karyotyping studies of the late 1980s. In more recent years, an appreciation for the role of developmental signaling pathways has been adopted, with the Wnt and Shh cascades being the most prominent12,17. Despite these advances in the field, known genes and pathways currently account for less than 50% of all medulloblastomas. Candidate gene approaches in medulloblastoma have been predominantly focused on components of the Shh, p53, and Rb signaling pathways22,54,59,70,74. This approach has implicated, and in some cases confirmed, novel oncogenes and tumour suppressors, but has been biased towards a priori gene selection and proven to be of limited throughput. Unbiased, large-scale genomic studies aimed at defining the molecular basis of medulloblastoma have been lacking to date208. Moreover, the majority of the genomic efforts thus far undertaken have profiled very modest sample cohorts on platforms incapable of detecting focal copy number aberrations (CNAs). As a result, advances in our understanding of the genes and pathways that contribute to the initiation, maintenance, and progression of medulloblastoma has lagged, especially when compared to recent progress made in the adult brain tumour community91,209,210.

* Work in this Thesis Chapter contributed to the following publications: Northcott PA, Nakahara Y, Wu X, et al: Multiple recurrent genetic events converge on control of histone lysine methylation in medulloblastoma. Nat Genet 41:465-72, 2009. Bhatia B, Northcott PA, Hambardzumyan D, et al: Tuberous sclerosis complex suppression in cerebellar development and medulloblastoma: separate regulation of mammalian target of rapamycin activity and p27 Kip1 localization. Cancer Res 69:7224-34, 2009. Adamson DC, Shi Q, Wortham M, et al: OTX2 Is Critical for the Maintenance and Progression of Shh-Independent Medulloblastomas. Cancer Res 70:181-91, 2009. Nakahara Y, Northcott PA, Li M, et al: Genetic and epigenetic inactivation of Kruppel-like factor 4 in medulloblastoma. Neoplasia 12: 20-27, 2010. 45

In this Chapter, we have used human genomics to identify novel genes and pathways that drive medulloblastoma pathogenesis and potentially account for medulloblastomas of unknown aetiology115. Through high-resolution copy number analysis of >200 medulloblastomas, we have performed the most comprehensive genomic profiling of medulloblastoma to date. Importantly, this effort has led to the identification and characterization of a novel family of genes encoding histone-modifying enzymes that are recurrently targeted by CNAs in medulloblastoma, suggesting that deregulation of the histone code contributes to medulloblastoma biology.

2.2 Materials and Methods

2.2.1 Medulloblastoma tumour specimens

We obtained all tumour specimens in accordance with the Research Ethics Board at the Hospital for Sick Children (Toronto, Canada). A total of 201 primary medulloblastomas were obtained as surgically resected, fresh-frozen samples. We obtained tumour specimens from the Co-operative Human Tissue Network (Columbus, OH), the Brain Tumour Tissue Bank (London, Canada) and from our collaborators.

2.2.2 100K and 500K GeneChip Mapping arrays

Medulloblastoma samples were processed and hybridized to Affymetrix SNP arrays at The Centre for Applied Genomics (TCAG) at the Hospital for Sick Children. We genotyped genomic DNA samples isolated from primary medulloblastomas and cell lines using the Affymetrix 50K Hind 240 and 50K Xba 240, or the 250K Nsp and 250K Sty GeneChip Mapping arrays as directed by the manufacturer. Briefly, 250 ng of DNA was digested with HindIII, XbaI, NspI or StyI (NEB), adaptor-ligated and PCR-amplified using a single primer with AmpliTaq Gold (Applied Biosystems). Amplified PCR products were pooled, concentrated and fragmented with DNase I. Products were subsequently labeled, denatured, and hybridized overnight to the respective arrays. Arrays were washed using an Affymetrix fluidics station and scanned using the GeneChip Scanner 3000. We generated CEL files using the Affymetrix GeneChip Operating Software (GCOS) 3.0.

2.2.3 SNP array data processing

Affymetrix CEL files were extracted using the Affymetrix Data Transfer Tool (version 1.1.0). For SNP genotyping, we used the BRLMM Analysis Tool (version 1.0) for individual array 46

platforms using default parameters. Copy number and LOH analyses were performed using both dChip 2006 and CNAG 2.0211,212. In dChip, we normalized arrays by invariant set normalization and computed signal intensities using PM/MM model-based expression. Raw copy number data was computed using 100 normal control samples as a reference (provided by S.W.S.) and inferred copy numbers were predicted using the hidden Markov model (HMM). We carried out LOH analysis using the HMM considering haplotype method, removing haplotypes consistent with 10% of reference samples. In CNAG, nonself analysis was done automatically with the same reference samples as above using a maximum of ten reference samples of the same sex per analysis. Inferred copy number changes and LOH were predicted using the HMM with default parameters.

To identify homozygous deletions, we used the following criteria: (i) 3 contiguous SNPs, (ii) size range 1 Kb–10 Mb and (iii) mean dChip/CNAG HMM copy number 0 or mean dChip raw copy number 0.4. To identify amplifications, we used the following criteria: (i) 5 contiguous SNPs, (ii) size range 10 Kb–10 Mb and (iii) mean dChip/CNAG HMM copy number 5. Recurrent, focal single copy losses were reported using the following criteria: (i) 3 samples with overlapping interstitial loss (CNAG HMM copy number = 1) and (ii) size range of individual losses 10 Kb–5 Mb.

To exclude abnormalities associated with known segmental duplications (LCRs), we compared amplifications and deletions to the LCRs detected in the Human Genome Segmental Duplication Database. Similarly, all amplifications and deletions were compared to known characterized structural variants through comparison with known copy number variants (CNVs) using the Database of Genomic Variants (February 2007)213. Regions of genomic gain or loss overlapping with known CNVs were eliminated.

To identify regions of statistical significance, raw copy number data was first segmented using GLAD (Gain and Loss Analysis of DNA)214 and probable CNVs were eliminated on the basis of their overlap with known, recurrent (2 samples) CNVs described in the HapMap and POPGEN control populations and/or the Ontario control population. We then analyzed filtered segmented copy number data with GISTIC (Genomic Identification of Significant Targets In Cancer)215,216 in GenePattern using default parameters. 47

2.2.4 Fluorescence in situ hybridization (FISH)

FISH for JMJD2 family members was carried out on a medulloblastoma tissue microarray as previously published37. BACs used for probes included RP11-1082E7 (JMJD2C, 9p24.1), RP11-235C23 (9q31.2 control), RP11-3214K1 (JMJD2B, 19p13.3), RP11-927F22 (19q13.32 control), RP11-105H7 (19q13.32 control), RP11-5C19 (JMJD2A, 1p34.1), RP11-54H19 (1q22 control) and RP11-336K24 (1q22 control).

2.2.5 Chromatin immunoprecipitation (ChIP)

ChIP of modified histones was done using the Chromatin Immunoprecipitation (ChIP) Assay Kit (Millipore) according to the manufacturer’s instructions. Briefly, 106 medulloblastoma cells were fixed in culture medium with 1% (vol/vol) formaldehyde (VWR International) at 37˚C for 10 min, washed twice on ice with cold PBS and collected by centrifugation at 2,000 rpm for 4 min at 4˚C. Cells were lysed in SDS lysis buffer for 10 min on ice, sonicated and cleared by centrifugation at 13,000 rpm for 10 min at 4˚C. Cell supernatants were diluted in ChIP dilution buffer and pre-cleared for 30 min with Protein A Agarose/Salmon Sperm DNA slurry at 4˚C before immunprecipitation with appropriate antibodies overnight at 4˚C. Immune complexes were captured by incubation with Protein A Agarose/Salmon Sperm DNA slurry for 1 h at 4˚C, before 5 min washes with low salt buffer, high salt buffer, and LiCl buffer, and two 5 min washes with TE buffer. Histone complexes were eluted from primary antibodies by two successive 15-min incubations with elution buffer (1% SDS, 0.1M NaHCO3). Histone-DNA cross-links were reversed by addition of 5M NaCl and heating at 65˚C for 4 h, followed by proteinase K digestion for one hour at 45˚C. DNA was then recovered by phenol/chloroform extraction and ethanol precipitation with glycogen. Resulting DNA pellets were washed once with 70% ethanol and resuspended in TE buffer. To assess the levels of H3K9me2 at promoters of candidate genes, we carried out end-point PCR reactions using primers targeting the promoter regions of MYC, TK1, and CDC25A.

2.2.6 Cell lines and cell culture

We purchased all media and reagents for culturing mammalian cells lines from Wisent unless otherwise stated. Medulloblastoma cell lines ONS76, UW228 and MED8A were grown as a monolayer in DMEM supplemented with 10% FBS. DAOY and D283 cell lines were grown as monolayers in AMEM with 10% FBS. The RES256 cell line was grown adherent in DMEM/F12 48 media supplemented with 2% FBS. D425, D458 and D556 were grown as suspension cultures in IMEM with 20% FBS, 10 mM HEPES and 0.225% sodium bicarbonate. MHH-MED-1 and D341 were grown in suspension in DMEM and AMEM with 10% FBS, respectively. For generation of DAOY and D283 stable cell lines, we selected stable transfectants in G418 at a concentration of 0.5 mg/ml and 2.0 mg/ml, respectively. For routine passaging of stable cell lines, we maintained cells in 0.2 mg/ml G418. NIH3T3 and Eco-Pheonix cells were grown in DMEM supplemented with 10% FBS. We cultured 293E cells in DMEM supplemented with 10% FBS and 300 μg/ml G418. All media were supplemented with a 1 antibiotic/antimycotic solution. CGNPs were cultured in NB-B27 (Neurobasal medium; Invitrogen) supplemented with

B27 (Invitrogen), 2 mM L-glutamine (Invitrogen), 1 mM Na-pyruvate (Invitrogen) and penicillin/streptomycin.

2.2.7 Plasmids For cDNA cloning of mammalian expression constructs, we carried out PCR using Platinum Hi- Fidelity Taq Polymerase (Invitrogen). All additional reagents for PCR were obtained from Invitrogen. Restriction enzymes and DNA ligase (Quick Ligation Kit) were purchased from New England Biolabs (NEB). To generate the L3MBTL3 expression plasmid, we amplified a 627-bp, HA (hemagglutinin) epitope-tagged fragment, corresponding to the 5’ end of mouse l3mbtl3 (NM_032438; 5’-HA-L3MBTL3) by PCR using a cDNA template (Image: 30356672) and appropriate HA-L3MBTL3 sense and antisense primers. The PCR product was cloned into the pCR 2.1-TOPO TA cloning vector (Invitrogen) according to the manufacturer’s instructions. pcDNA3.1(+) (Invitrogen) and the 5’-HA-L3MBTL3 cDNA were digested with NheI and BamHI, gel-purified and ligated. The resulting 5’-HA-L3MBTL3 cDNA and the l3mbtl3 library plasmid were digested with BamHI and NotI, gel-purified and ligated to create the final pcDNA3.1-HA-L3MBTL3 construct. The pcDNA3.1-GFP plasmid was constructed by digestion of pCAT3Blbp-GFP with NcoI and XbaI, and ligation of the GFP fragment into blunted EcoRI and XbaI sites of pcDNA3.1(+). We generated a C-terminal HA-tagged pcDNA3.1-SAMD3-HA construct by PCR amplification of the SAMD3 cDNA (BC029851) using SAMD3-HA sense and antisense primers. The PCR product was ligated into pcDNA3.1(-) (Invitrogen) at NotI and EcoRI sites. The pcDNA3.1-TMEM200A-HA plasmid was constructed by PCR amplification of the TMEM200A cDNA (BC044246) using TMEM200A-HA sense and antisense primers. The PCR product was ligated into pcDNA3.1(-) at NotI and EcoRI sites. The pBabe-HA-JMJD2C 49

construct (gift from K. Helin, University of Copenhagen) has been previously described217. To generate pWZL-HA-JMJD2C-IRES-GFP, the JMJD2C cDNA was subcloned from pBabe-HA- JMJD2C by digestion with NaeI and SalI and directional ligation into a blunted EcoRI site and SalI of pWZL-IRES-GFP (gift of A. M. Kenney, Memorial-Sloan Kettering). All constructs used in this study were fully sequenced (TCAG) in both directions before their use.

2.2.8 Antibodies For immunoblotting, mouse antibody to HA (anti-HA; F7; Santa Cruz Biotechnology), mouse anti-p27 KIP 1 (DCS-72.F6; Abcam) and mouse anti-b-actin (AC-74; Sigma-Aldrich) were used. Goat anti-mouse and goat anti-rabbit HRP-conjugated secondary antibodies were purchased from BioRad. For immunoprecipitation of HA-tagged recombinant proteins, we used rabbit anti-HA (Y-11; Santa Cruz Biotechnology). For ChIP experiments, we used ChIP-grade mouse anti- H3K9me2 (ab1220; Abcam) and normal mouse control IgG (sc-2025; Santa Cruz Biotechnology) antibodies; for indirect immunofluorescence, we used mouse anti-HA (HA.11; Covance), mouse anti-GFP (MAB3580; Chemicon) and rabbit anti-H3K9me2 (9753; Cell Signaling Technologies). Goat anti-mouse Alexa-488 and goat anti-rabbit Alexa-594 secondary antibodies were acquired from Molecular Probes. Antibodies used in immunohistochemistry included mouse anti-GLP (EHMT1; B0422; R & D Systems), ChIP-grade mouse anti-H3K9me2 (ab1220; Abcam), mouse anti-Histone H3K9me3 (ab6001; Abcam), ChIP-grade rabbit anti-Histone H3K9me1 (ab9045; Abcam), mouse anti-p27 KIP 1 (DCS-72.F6; Abcam) and rabbit anti-JMJD2C (A300-885A; Bethyl Laboratories).

2.2.9 Processing of tumour samples and cell lines Fresh-frozen medulloblastoma specimens were stored at –80 °C before being processed for extraction of nucleic acid. Tissue samples were pulverized under liquid nitrogen using a mortar and pestle and partitioned for extraction of genomic DNA and total RNA. For genomic DNA isolation, we subjected approximately 25–50 mg of crushed tissue to digestion using SDS/Proteinase K (Roche) for 3 h at 50 °C. Homogenates were extracted three times with buffer-saturated phenol (Invitrogen) before precipitation of DNA with two volumes of anhydrous ethanol and 10% (vol/vol) 10 M ammonium acetate. Precipitated DNA was washed three times with 70% ethanol and resuspended in reduced EDTA-TE (10 mM Tris, 0.1 mM EDTA; pH = 8.0). We quantitated samples by Nanodrop and assessed the integrity of isolated DNA by agarose gel electrophoresis before submission for SNP array. For RNA, crushed tumour specimens were 50

resuspended in 1 ml Trizol (Invitrogen), passed through a 20-gauge needle 5–10 times, and incubated at room temperature for 10 min. Following addition of 0.2 ml chloroform (Sigma- Aldrich), samples were mixed briefly and cleared by centrifugation at 12,000g for 15 min at 4 °C. Total RNA was precipitated by addition of 0.5 ml isopropanol, incubation at room temperature for 10 min, and subsequent centrifugation at 12,000g for 10 min at 4 °C. RNA pellets were washed once with 75% ethanol and dissolved in DEPC water (Invitrogen). Nanodrop-quantitated samples were evaluated by Bioanalyzer before any downstream analysis.

2.2.10 Real-time quantitative PCR (qPCR) CNAs detected by SNP array analysis were validated by real-time qPCR using the Ct method.

For each experiment, a minimum of two diploid loci (as predicted by SNP array analysis) were amplified and used for normalization, and a normal brain sample was included as a calibration template. Reactions were carried out in triplicate using Platinum SYBR Green qPCR SuperMix UDG (Invitrogen) with 25 ng of template per reaction as suggested by the supplier using a Chromo4 Real-Time System (Bio-Rad). We designed all primers using Primer3 and sequences are available upon request. For candidate gene expression profiling, we subjected 2 μg of total RNA to random hexamer-primed reverse transcription using SuperScript III (Invitrogen) according to the manufacturer’s instructions. PCR reactions were done as described above using 10 ng of cDNA per reaction with b-actin as a reference gene for data normalization. Normal fetal and adult cerebellum RNA samples (Biochain) served as controls for calibration. Primers were designed with PerlPrimer software. To compare levels of EHMT1 expression between 9q diploid and 9q monosomic samples, a two-sample Wilcoxon test was applied. For SMYD4, JMJD2C, JMJD2B, MYST3, and BMI1, a Wilcoxon signed rank test was used, and P values were calculated using exact methods where ties in ranks were not present.

2.2.11 Cell proliferation assays Cellular proliferation assays were done using the CellTiter 96 Aqueous One Solution Proliferation Assay (Promega) according to the manufacturer’s instructions. Briefly, 1 103 cells were seeded in triplicate on 96-well plates and colorimetric readings (absorbance of 490 nm) were taken at 24, 48, 72 and 96 h time points following 2 h incubations with proliferation reagent. To qualitatively assess cell growth in culture, we seeded 5 103 cells in triplicate on 10- cm dishes and cultured cells for 7 d before crystal violet staining. Cells were washed twice with 51

ice-cold PBS, followed by fixation with ice-cold methanol for 10 min on ice. Fixed cells were stained at room temperature with crystal violet (Sigma-Aldrich) solution (0.5% crystal violet

(w/v), 25% methanol), rinsed 2–3 times with Milli-Q H2O, and allowed to dry at room temperature.

2.2.12 Cell cycle analysis and cell viability assays Cell cycle analysis was done by propidium iodide (PI) staining of medulloblastoma cell lines and by measuring DNA content using a Becton Dickinson FACScan. Briefly, asynchronous cells were grown on 10-cm dishes and collected by centrifugation at 400g for 5 min at 4 °C. Cell pellets were resuspended in staining media (1 Hank's balanced salt solution (HBSS), 10 mM

HEPES-NaOH, pH 7.2, 2% calf serum (vol/vol), 10 mM NaN3) and fixed by adding ice-cold 80% ethanol, vortexing briefly, and incubating for 30 min to overnight at 4 °C. Fixed cells were collected by centrifugation at 400g for 5 min at 4 °C and pellets resuspended and incubated in 2 mg/ml RNAase A solution (Qiagen) for 5 min at room temperature. Staining was done by adding PI solution (HBSS, 0.1 mg/ml propidium iodide, 0.6% NP-40 (vol/vol)) and incubating for 30 min at room temperature protected from light. Cells were collected by a final spin, resuspended in staining media, and filtered into polystyrene tubes fitted with a cell-strainer cap. We collected data using CELLFIT software. To assess cell viability in DAOY stable cell lines, we used the Annexin V-FITC Apoptosis Detection Kit I (BD Pharmingen) according to the manufacturer’s instructions. Briefly, 48 h before analysis, medulloblastoma cells were seeded at a density of 5 105 cells in 10 cm dishes. Asynchronous cells were trypsinized, washed twice with ice-cold PBS, and resuspended in 1 binding buffer at a concentration of 1 106 cells/ml. A fraction (~1 105) of cell solution was then mixed with Annexin V-FITC and PI, vortexed gently, and incubated for 15 min at room temperature in the dark. Stained cells were then diluted in 1 binding buffer and analyzed by flow cytometry using a FACScan.

2.2.13 Retrovirus production To generate ecotropic retroviruses, Eco-Phoenix cells were seeded at a density of 5.5 106 cells in 10-cm dishes 24 h before transfection with 8 μg of Eco-Phi helper plasmid and 8 μg of viral construct using Fugene6 (Roche) as directed by the supplier. Twenty-four hours post- transfection, virus producer cells were washed once with PBS and then grown in serum-free DMEM at 32 °C. Viral supernatants were collected successively for 2 d, pooled, filtered through a 0.45-μm filter, flash-frozen, and stored at –80 °C. Amphotropic viruses were produced in 293E 52

cells (provided by A. M. Kenney, Memorial-Sloan Kettering) by transfecting cells with 10 μg each of vsv-g, gag-pol, and viral construct as described above. Supernatants were collected and stored as above.

2.2.14 Cerebellar granule neuron precursor (CGNP) isolation and transduction CGNPs were isolated from P5-P7 mouse cerebella by gradient centrifugation through Percoll. For immunofluorescence microscopy experiments, 7.5 105 freshly isolated cells were plated on poly-ornithine (Sigma-Aldrich) coated coverslips (Fisher Scientific) in 24-well dishes and treated with recombinant SHH protein (provided by A. M. Kenney, Memorial-Sloan Kettering) at a final concentration of 3–5 μg/ml. Three hours post-isolation, conditioned NB-B27 media was removed and stored, as cells were incubated with viral supernatant supplemented with 8 μg/ml polybrene (Sigma-Aldrich) for 3 h at 32 °C. Following viral transduction, conditioned media was replaced and cells were cultured for an additional 48–96 h at 37 °C. Transduction of NIH3T3 and medulloblastoma cells was done in a similar manner in their normal growth media.

2.2.15 Immunofluorescence and immunohistochemistry For immunofluorescence experiments, cells grown on coverslips in 24-well plates were fixed with 4% (w/vol) paraformaldehyde (PFA; Sigma-Aldrich) in PBS for 15 min at room temperature. Cells were washed twice with PBS, permeabilized with 1% Triton X-100 (Sigma- Aldrich) in PBS for 5 min, and washed three times with PBS, before blocking for 1 h in 5% (vol/vol) normal goat serum (NGS; Sigma-Aldrich) in PBS-T (PBS, 0.1% Triton X-100). Blocked cells were washed once with PBS, and incubated in primary antibody diluted in 2.5% NGS/PBS-T for 1 h at room temperature. Stained cells were washed three times and incubated with fluorophore-conjugated secondary antibodies in PBS-T for 1 h at room temperature, before three final washes with PBS and mounting of coverslips on glass slides with VECTASHIELD containing DAPI (Vector Laboratories). We used a Quorum spinning disk confocal microscope for microscopy and acquired images with Volocity 4.4. Immunohistochemistry for EHMT1 and H3K9me2 on primary tumours was done using two independent medulloblastoma TMAs (Johns Hopkin’s University and St. Jude Children’s Research Hospital). Staining for EHMT1, H3K9me2, H3K9me1, H3K9me3, p27, and JMJD2C in the developing mouse cerebellum was performed at the Department of Pathology at Toronto General Hospital. 53

2.2.16 Immunoblotting Medulloblastoma cells cultured on 10-cm dishes were washed twice with ice-cold PBS and lysed on ice with RIPA (150mM NaCl, 1% (vol/vol) NP-40, 0.5% (w/vol) DOC, 0.1% (w/vol) SDS, 50mM Tris-Cl, pH 8.0) for 10 min. Cell lysates were cleared by centrifugation at 13,000 rpm for 10 min at 4 °C. For preparation of nuclear extracts, cells were washed with ice-cold PBS and collected by centrifugation at 1,000 rpm for 5 min at 4 °C. The cytoplasmic fraction was removed by incubating cells in buffer A (10mM KCl, 1.5mM MgCl2, 0.5mM DTT, 10mM HEPES-KOH, pH 7.9) with protease inhibitors (Complete Mini; Roche) on ice for 10 min. Nuclei were pelleted and proteins extracted in RIPA buffer with inhibitors for 30 min at 4 °C with rotation. Nuclear extracts were then prepared by removing chromatin pellets after centrifugation at 13,000 rpm for 10 min at 4 °C. For immunoprecipitation of epitope-tagged proteins, nuclear lysates were incubated with primary antibodies overnight at 4 °C. Immune complexes were captured by incubation with Protein A Sepharose beads (Sigma-Aldrich) for 1 h at 4 °C and washed 3–4 times in wash buffer (PBS, 0.1% (vol/vol) NP-40) before SDS-PAGE and immunoblotting. Briefly, proteins were separated by SDS-PAGE, transferred to PVDF, and blocked in 5% (w/vol) skim milk in TBS-T (TBS, 0.1% (w/vol) Tween 20) for 1 h at room temperature. Blots were incubated with primary antibodies overnight at 4 °C, washed three times in TBS-T, and probed for 1 h at room temperature with the appropriate HRP-conjugated secondary antibodies diluted in TBS-T. Blots were then subjected to enhanced chemiluminescence (Western Lightning Chemiluminescent Reagent Plus; PerkinElmer) and exposed to X-ray film (Bioflex; InterScience).

2.3 Results

2.3.1 The medulloblastoma genome

Over the past 5 years, novel, high-resolution (i.e. sub-megabase) genomic technologies have become available for studying copy number states in both normal and diseased genomes104,133,134. Although multiple array-based studies of the medulloblastoma genome have recently been reported (reviewed in 208), the majority have profiled relatively modest sample cohorts (median sample size (n): ~24) using arrays that, although have been an improvement from classical CGH, have been of insufficient density and thus resolution (median resolution (Kb): ~500) to detect very focal genetic events and thus making it very difficult to discriminate ‘driver’ versus ‘passenger’ mutations. To address these issues and delineate CNAs that may target novel 54

oncogenes and TSGs in medulloblastoma, we retrospectively collected an unprecedented cohort of 201 fresh-frozen primary medulloblastomas and 11 medulloblastoma cell lines and analyzed their genomes using high-resolution SNP genotyping arrays115. These oligonucleotide arrays consist of 25mer probes designed to detect the genotype (i.e. A or B allele) of known SNPs at loci distributed across the genome218-221. The median intermarker distance (i.e. resolution) of probes on the 100K and 500K arrays employed in this study is 8.5 Kb and 2.5 Kb, respectively, which is at least an order of magnitude higher in terms of resolution than any previous array- based study of the medulloblastoma genome (Figure 1).

Visual overview of our medulloblastoma copy number data revealed several known recurrent regions of large-scale gain and loss, including gains on chromosomes 1q, 7, and 17q and losses on chromosomes 8, 9q, 10q, 11, 16q, 17p, and X (Figure 2a-c, Figure 3a-x). The most common structural aberration reported in medulloblastoma is i{17}q, which results in the net loss of one copy of 17p and a net gain in one copy of 17q116,117. In our dataset, we identified i{17}q in ~28% (59/212) of cases (Figure 4a,b), consistent with what has been reported in the literature.

Probe-level data generated using SNP arrays can be simultaneously used to infer both SNP copy number as well as SNP genotype. SNP genotypes mapping to a given chromosome or chromosomal region can be further analyzed to predict loci which have undergone LOH211,222. Furthermore, integration of LOH and copy number data for a given locus can discriminate LOH events occurring in the context of deletion from those occurring as a consequence of recombination (i.e. uniparental disomy; UPD). Analysis of SNP genotypes in our dataset identified multiple areas of consistent LOH, including chromosomes 6, 8, 9, 10, 11, 16q, 17, and X (Figure 5a,b). Of interest, our data demonstrates that mechanisms leading to LOH in medulloblastoma are chromosome-specific, showing monosomy, UPD, or a mixture of the two mechanisms (Figure 5b). For example, LOH on chromosome 16q occurred almost exclusively in the setting of 16q deletion, whereas LOH reported on 9q and 10q were the result of either deletion or UPD. The underlying biology driving clonal selection secondary to these extremely large regions of gain and loss is unknown, due to the difficulty of distinguishing ‘drivers’ from ‘passengers’. 55

2.3.2 Recurent targeting of known oncogenes and TSGs in medulloblastoma

A major strength of high-resolution SNP platforms is their ability to identify extremely focal regions of amplification and deletion. After eliminating known copy number variants (CNVs), manual curation of our dataset revealed 139 amplifications and 61 homozygous deletions targeting at least one RefSeq gene. To reduce potential biases that may be associated with manual annotation of CNAs, we employed the bioinformatic tool GISTIC (Genomic Identification of Significant Targets in Cancer) to identify significantly amplified and deleted loci in medulloblastoma (Figure 6a,b). The GISTIC algorithm aims to identify regions of the genome that are aberrant more often than would be expected by chance, with greater weight given to high- amplitude events (i.e. high-level amplifications or homozygous deletions) that are less likely to represent random aberrations and, thus, more likely to drive tumourigenesis215,216. High-level amplifications targeting several known medulloblastoma oncogenes, including, MYC, MYCN, OTX2, TERT, PDGFRA, and CDK6 were identified in our analysis (Table 1, Figure 6a). In addition, GISTIC identified significant regions of deletion including known TSGs, LRP1B, CDKN2A, PTEN, and CDH1, as well as novel events targeting genes involved in histone lysine

methylation including EHMT1, L3MBTL3, and SMYD4 (Figure 6b; discussed in detail below).

Despite the relatively high number of amplicons identified, only 12 regions of recurrent amplification were noted (Table 2). The MYC family (i.e. MYC, MYCN, and MYCL1) of proto- oncogenes is well characterized for their role in medulloblastoma pathogenesis117,119,121,122. Amplicons targeting MYC, MYCN, and MYCL1 were cumulatively found in ~9% (19/212) of samples (Figure 7a, Table 1), consistent with the 5-10% that has been previously reported for this gene family.

Aberrant activation of the Shh signaling pathway is found in ~25-30% of medulloblastomas37,38,187, with the majority of cases attributed to either loss-of-function mutations incurred in PTCH1 and SUFU, or activating mutations in SMO12,17. Of interest, we report novel, rare amplification of downstream Shh effectors, GLI1 and GLI2 (Figure 7b, Table 1). These amplicons have important implications for potential targeted therapies, as most agents currently employed to inhibit the Shh pathway are directed much further upstream and thus would likely be ineffective in cases such as these harboring GLI family amplification223,224. 56

Figure 1. Array-based profiling of medulloblastoma: summary of the literature. Scatter plot detailing reports in the literature that have profiled the medulloblastoma genome using array-based technologies. Each publication is represented as a coloured circle, with position along the y-axis determined by the number of samples analyzed, and position along the x-axis determined by the approximate median resolution of the array platform(s) employed in the study. 57

Figure 2. The medulloblastoma genome. A. Global view of regions of gain and loss across the genome in a series of 123 non-overlapping medulloblastomas genotyped on the Affymetrix 500K SNP array platform. Output from GenePattern SNP Viewer. Regions of gain are red, regions of loss are blue. B. Global view of regions of gain and loss across the genome in a series of 89 non-overlapping medulloblastomas genotyped on the Affymetrix 100K SNP array platform. Output from GenePattern SNP Viewer. Regions of gain are red, regions of loss are blue. C. Summary plot demonstrating the frequency of regions of gain and loss in the medulloblastoma genome (500K SNP array). Output from dChipSNP. Recurrent losses are observed on chromosomes 6, 8, 9q, 10q, 11, 16q, 17p, and X. Recurrent gains are observed on chromosomes 1q, 7, and 17q. 58

Figure 3. Gains and losses in the medulloblastoma genome. Chromosomal views showing regions of genomic gain (red) and loss (green) along each chromosome, for tumours analyzed on either the 100K or 500K Affymetrix SNP array platforms. For the X chromosome, tumours from males and females are displayed separately. Copy number output is from CNAG. 60

Figure 4. CNAs on chromosome 17 in medulloblastoma. A, B. Loss of chromosome 17p concurrent with gain of chromosome 17q (i{17}q) was observed in ~28% of medulloblastomas profiled on the 100K (A) and 500K (B) SNP array platforms. This is consistent with the known rate of occurrence of i{17}q in medulloblastoma. Copy number output from GenePattern SNP Viewer. 62

Figure 5. LOH in the medulloblastoma genome. A. Genome-wide depiction of regions of inferred LOH (blue) and regions of inferred retention of heterozygosity (yellow) in a series of 123 medulloblastomas (500K SNP array). Output is from dChipSNP. B. Bar graph exhibiting that some chromosomes always achieve LOH through deletion, other chromosomes have predominantly copy-number neutral LOH (UPD), and some demonstrate both mechanisms. 64

OTX2 is a member of a well-conserved family of bicoid-like homeodomain containing transcription factors that play important roles in embryo patterning, brain regionalization, and lineage specification225. The OTX2 gene has been previously shown to be amplified and overexpressed in medulloblastoma by multiple independent groups135,136. Through manual inspection of recurrent CNAs in our dataset, we identified a locus on chromosome 14q22 as the most frequent target of focal (1 Mb) gain/amplification detectable with the current platforms (21/212, ~10%; Figure 8a). Examination of the minimal common region gained in these samples using the UCSC Genome Browser (NCBI Build 35, hg17) revealed that the single gene, OTX2, was completely located within this minimal common region of chromosomal gain (Figure 8a)226. A subset of these CNAs targeting OTX2 were validated by quantitative PCR using genomic template (Figure 8b,c). The highly specific gain of OTX2 copy number in a large proportion of medulloblastomas implies that OTX2 gain is a driver event that confers a selective advantage during clonal expansion of this neoplasm.

Additional recurrent CNAs targeting known oncogenes not previously reported in medulloblastoma were also identified, including miR-17/92 and MYST3 (Table 1). miR-17/92 is an oncogenic miRNA cluster on 13q31 characterized in detail in Chapter 4 of this thesis and MYST3 maps to 8p11 and encodes a histone lysine acetyltransferase discussed further below.

Analysis of 61 homozygous deletions identified 20 known candidate TSGs, including, CDKN2A/2B/2C, LRP1B, ERBB4, MAP2K4, KLF4, and TSC1 (Table 3). It is noteworthy that several of these genes (i.e. CDKN2A/2B/2C, LRP1B) were deleted only in medulloblastoma cell lines (Figure 9a,b), questioning their relevance in primary malignancy. Of 20 regions of recurrent focal hemizygous deletion targeting a single gene, 6/20 (30%) targeted known TSGs (Table 4).

Among recurrent deletions in medulloblastoma, we identified a locus on 9q34 that includes the TSC1 tumour suppressor as one of the most frequent targets of focal loss in our dataset (Figure 10a,b)227. Individuals with germline mutations of TSC1 (or TSC2) develop tuberous sclerosis (i.e. tuberous sclerosis complex; TSC), an inherited neurocutaneous multisystem disorder in which afflicted individuals are predisposed to developing benign tumours throughout the body228. As depicted in Figure 10a, we identified one focal homozygous deletion and six focal hemizygous deletions of TSC1. Importantly, the minimal common region shared between these focal 65

Figure 6. Identification of statistically significant targets of gain and loss in medulloblastoma using GISTIC. A. GISTIC output shows significant regions of amplification and gain from non-overlapping cohorts of medulloblastomas analyzed on the 100K (left) and 500K (right) SNP array platforms. Significant regions are labeled by cytoband and notable genes are identified. B. GISTIC output shows significant regions of homozygous and hemizygous deletion from non-overlapping cohorts of medulloblastomas analyzed on the 100K (left) and 500K (right) SNP array platforms. Significant regions are labeled by cytoband and notable genes are identified. 67

68 69

Figure 7. Amplification of known oncogenes in medulloblastoma. A. Amplification of MYC family oncogenes in medulloblastoma in our cohort of 212 medulloblastomas. Output from dChipSNP. Vertical bars denote MYC family gene loci. B. Rare amplifications of GLI2 and GLI1 in medulloblastoma, downstream effectors of Shh signaling. Output from dChipSNP. Vertical bars denote the GLI1 and GLI2 loci. 70

Figure 8. Recurrent targeting of OTX2 in medulloblastoma. A. Integral view of amplifications and focal gains targeting OTX2 on chromosome 14q22.3 in medulloblastoma. 100K and 500K copy number output from CNAG depicts samples with gain in red above each chromosome ideogram. Two samples profiled on the 100K SNP array (D425 and D458) harbor high-level amplifications at this locus (shown in bright red). The minimal common region of amplification/gain in these samples is shown as output from the UCSC Genome Browser (NCBI Build 35), highlighting OTX2 as the sole gene targeted. B. Real-time PCR (qPCR) validation of OTX2 copy number data. A subset of medulloblastoma samples with inferred amplification (n=2), focal gain (n=6), or neutral copy number (n=4) by SNP array were queried for OTX2 copy number status by qPCR. Y-axis indicates OTX2 gene copy number calibrated to a normal diploid genome. C. Boxplots summarizing OTX2 copy number data determined by qPCR in (B). Y-axis indicates OTX2 gene copy number calibrated to a normal diploid genome. 72

Figure 9. Homozygous deletion of CDKN2A/CDKN2B and LRP1B is limited to medulloblastoma cell lines. A. Copy number output from dChip for 79 primary medulloblastomas and 10 medulloblastoma cell lines profiled on the 100K SNP array platform reveals recurrent homozygous deletion of the CDKN2A/CDKN2B locus on 9p21 in medulloblastoma cell lines (DAOY, MED8A, and ONS76). In total (100K and 500K), 4/11 cell lines analyzed showed homozygous deletion of this locus, whereas 0/201 primaries harbored the deletion. B. Chromosome 2q copy number output for the same samples in (A) showing recurrent homozygous deletion of LRP1B on 2q22 in medulloblastoma cell lines (MED8A, UW228, and ONS76). Similar to the data shown in (A), 4/11 cell lines analyzed exhibited homozygous deletion of this gene, which was not found in any primary samples (0/201). 75

Figure 10. Recurrent focal deletion of TSC1 in medulloblastoma. A. Integral view of recurrent losses on chromosome 9q34.13 targeting TSC1 in medulloblastoma. 100K and 500K copy number output from CNAG depicts samples with hemizygous loss in green below each chromosome ideogram. One medulloblastoma in the 500K dataset harbors a focal homozygous deletion (red box) on 9q34. The minimal common region of loss (boundary determined by the homozygous deletion) is shown as an output from the UCSC Genome Browser (NCBI Build 35) with the TSC1 gene highlighted. B. Deletions targeting TSC1 are statistically significant. GISTIC output for chromosome 9 copy number data (500K SNP array; n=123 samples), showing chromosome 9q is a significant region of loss in medulloblastoma. The sharp peak at 9q34.13 encompasses the TSC1 locus and pinpoints a minimal region of highly significant loss on 9q in the dataset (q-value=10-9.6; G- score=0.16). C. Downregulation of TSC1 in medulloblastomas with TSC1 deletion. qRT-PCR analysis of TSC1 expression in a series of medulloblastomas with either normal diploid TSC1 status (red bars) or TSC1 deletion (blue bars). Normal fetal (5 donors) and adult (5 donors) human cerebellum samples were included as controls. TSC1 expression was lower in tumour samples harboring TSC1 deletions compared to those retaining both copies of the gene (two-sample Wilcoxon test, p<0.05). Notably, no expression is observed in the medulloblastoma harboring homozygous deletion of TSC1 (MB-183). 78

(ranging from ~125 Kb to ~1.5 Mb) deletions consists of only 3 Refseq genes: TSC1, C9orf9, and C9orf98 (Figure 10a). This area of recurrent deletion on 9q was identified by GISTIC as statistically significant (Figure 10b). To investigate whether TSC1 deletion results in reduced mRNA expression levels, we performed qRT-PCR on a subset of tumour samples, in the context of either diploid TSC1 copy number or TSC1 deletion. Importantly, loss of TSC1 appeared to correlate with reduced TSC1 expression (Figure 10c; P < 0.05). The observed reduction in TSC1 expression in samples with TSC1 deletion suggests it may be haploinsufficient in medulloblastoma. Since individuals with TSC do not develop medulloblastoma, TSC1 loss-of- function secondary to deletion is more likely to be a progression event227.

2.3.3 Multiple recurrent genetic events converge on control of histone lysine methylation in medulloblastoma

Of 61 homozygous deletions identified in our dataset, only six (~10%) were recurrent (Table 5). Historically, many well-known TSGs were identified through mapping recurrent focal homozygous deletions229-232. Of the six recurrent homozygous deletions in the present study, only one targeted the coding region of a single RefSeq gene in primary tumours (Table 5). These recurrent, focal homozygous deletions of EHMT1 on chromosome 9q34.3 were verified in two primary medulloblastomas (Figure 11a,b); these somatic deletions were not present in matched constitutional DNA (Figure 11c). EHMT1 is a SET domain containing histone lysine methyltransferase that dimethylates H3K9, a predominantly repressive chromatin mark (Table 6)233,234. In addition, EHMT1 has recently been implicated as the critical target of the 9q subtelomeric deletion syndrome in which affected individuals present with a variety of clinical phenotypes, including severe mental retardation, hypotonia, brachycephaly, facial dysmorphology, heart defects, and behavioural problems235,236. Expression of EHMT1 transcript was decreased at least two-fold in 10% of medulloblastoma samples, particularly tumours monosomic at the EHMT1 locus (Figure 11d). Immunohistochemical staining for EHMT1 protein expression, and H3K9 dimethylation (H3K9me2) on a TMA consisting of 64 non- overlapping human medulloblastomas demonstrated that EHMT1 staining was absent in 16/64 (25%) cases, and 26/64 tumours (~41%) were void of nuclear staining for H3K9me2 (Figure 11e). Critically, of 16 tumours with no EHMT1 staining, 15/16 (~94%) also stained negative for H3K9me2 (P=0.0024), consistent with a model in which loss of EHMT1 leads to hypomethylation of H3K9 in medulloblastoma. 79

Figure 11. Recurrent homozygous deletions target the histone lysine methyltransferase, EHMT1, in medulloblastoma. A. Output from dChipSNP demonstrates focal homozygous deletion limited to EHMT1 in MB-118 and MB-149. MB-101 is diploid for chromosome 9q, whereas MB-160, -163, -165, -166, -184, -198 and -224 exhibit hemizygous deletion of 9q. B. GISTIC output for chromosome 9 (500K SNP array) demonstrates a significant region of focal loss on 9q34 at the EHMT1 locus. C. Real-time genomic PCR at the EHMT1 locus confirms somatic homozygous deletion in tumour samples, but not in matched constitutional DNA. D. qRT-PCR for EHMT1 shows significantly decreased expression of EHMT1 in samples with monosomy 9q, as opposed to tumours with diploid chromosome 9q (Two sample Wilcoxon test, p=0.0002468). E. Immunohistochemical staining for EHMT1 expression and H3K9 dimethylation was performed on a 64-tumour human medulloblastoma TMA. Staining was graded from 0-3 as illustrated. Percentage of tumours in each category is noted below each category. 82

Three additional homozygous deletions targeted genes with known or predicted roles in histone lysine methylation (total: 5/61 (~8%) homozygous deletions). We found focal homozygous and hemizygous deletions of the Polycomb genes SCML2, L3MBTL3, and L3MBTL2 in a subset of medulloblastomas (Table 6, Figure 12a,b). The MBT (malignant brain tumour) domains of these Polycomb proteins functions to bind and interpret the degree of histone lysine methylation, particularly H3K9237-241. A very focal intragenic homozygous deletion targeted SMYD4, a histone lysine methyltransferase at 17p13.3, the region most frequently deleted in medulloblastoma (Table 6, Figure 12c). Recently, using a retroviral gene trapping assay, SMYD4 was identified as a putative TSG in breast cancer, capable of inhibiting breast cancer cell growth in vitro and transcriptionally repressing expression of the PDGFRA oncogene242. Importantly, re-analysis of published data on 244 acute lymphoblastic leukemias and 371 lung adenocarcinomas genotyped on similar platforms as employed here, did not reveal any homozygous deletions of EHMT1, SCML2, L3MBTL3, or SMYD4 suggesting these events may be medulloblastoma-specific216,243.

Whereas EHMT1 and SMYD4 methylate histone lysine moieties, Jumonji family proteins function as histone lysine demethylases217,244,245. Our SNP array analysis detected amplification and focal gain of JMJD2B (Table 6, Figure 12d). Subsequent FISH experiments on a TMA of 88 non-overlapping medulloblastomas revealed recurrent amplifications (copy number 5-10) of JMJD2B, as well as the known oncogene JMJD2C (Table 6, Figure 12e)217. These enzymes demethylate H3K9, and perhaps H3K36244,246. Similarly, we found recurrent amplification of the histone lysine acetyltransferase MYST3 (Table 6, Figure 12f), which we predict to result in H3K9 hypomethylation, as H3K9 methylation by EHMT1 is blocked by H3K9 acetylation247.

qRT-PCR-based expression profiling of these candidates with known or predicted roles in histone lysine modification demonstrated their dysregulation in a significant percentage of primary medulloblastomas. SMYD4 exhibited a minimum of two-fold decreased expression in 30% (12/40) of medulloblastomas as compared to normal controls (Figure 13a). In contrast, we found increased expression of JMJD2C or JMJD2B in 15% and 7.5% of medulloblastomas, respectively (Figure 13b,c). Similarly, MYST3 showed a minimum of two-fold upregulation in ~28% (11/40) of tumours (Figure 13d). Analogous to the focal genetic events described in Table 6, we predict that these expression patterns should result in hypomethylation of H3K9. 84

Figure 12. CNAs target genes controlling histone lysine methylation in medulloblastoma. A. Large region of homozygous deletion on chromosome Xp22.13 includes SCML2, a Polycomb group transcriptional repressor. B. Homozygous deletion of L3MBTL3, a Polycomb group transcriptional repressor mapping to 6q22.33-q23.1, in the DAOY medulloblastoma cell line. C. Focal, intragenic homozygous deletion targeting SMYD4, a histone lysine methyltransferase, on 17p13.3. 85

Figure 12. D. High-level amplification of chromosome 19p13.3 includes JMJD2B, a histone lysine demethylase. E. Interphase FISH on paraffin embedded tissues on a medulloblastoma TMA demonstrates amplification of JMJD2C (green) at 9p24.1 as opposed to a control probe (red) at 9q31.2 in a representative medulloblastoma sample. F. Recurrent high-level amplification of MYST3 on chromosome 8p11.21. 86

Figure 13. Transcriptional deregulation of genes controlling histone lysine methylation in medulloblastoma. A. qRT- PCR of SMYD4 demonstrates greater than two-fold decreased expression in 30% of medulloblastomas as compared to a normal adult cerebellar control. Wilcoxon signed rank test, P=2.596e-06. B. qRT-PCR of JMJD2C shows greater than two-fold increased expression in 15% of medulloblastomas as compared to normal fetal cerebellum. Wilcoxon signed rank test, P=0.1351. C. qRT-PCR of JMJD2B shows greater than two-fold increased expression in 7.5% of medulloblastomas as compared to normal fetal cerebellum. Wilcoxon signed rank test, P=0.034. D. qRT-PCR of MYST3 demonstrates greater than two-fold increased expression in 28% of medulloblastomas as compared to normal fetal cerebellum. Wilcoxon signed rank test, P=0.2683. E. qRT-PCR of BMI1 shows greater than two-fold increased expression in >80 % of medulloblastomas as compared to normal fetal cerebellum. Wilcoxon signed rank test, P= 9.095e-13. 88

While individually uncommon, collectively these focal genetic events targeting genes that control histone lysine methylation were found in ~19% of medulloblastomas (Table 6). CNAs affecting chromatin genes listed in Table 6 were mutually exclusive in our dataset (P=2.2x10-16), suggesting that they have a common function. Published literature and our results (see below) provide strong links between H3K9 methylation and EHMT1, L3MBLT3, L3MBLT2, SCML2, JMJD2C, JMJD2B, and MYST317,217,233,247-250. Lack of nuclear staining for H3K9me2 was observed in 41% of medulloblastomas (Figure 11e). Thus, proper control of the histone code, particularly methylation at H3K9 is important in the pathogenesis of some medulloblastomas. There was no enrichment for gender, age group, or histological subtype in the medulloblastomas with CNAs in chromatin genes analyzed by SNP array, or the non-overlapping group analyzed on the TMA.

EHMT1 is part of the E2F6 complex that preferentially occupies MYC and E2F dependent

promoters of cells in G0 rather than in G1, suggesting that this complex contributes to silencing in quiescent cells233. There is strong causative evidence for E2F and MYC dependent transcription in the pathogenesis of human and murine medulloblastoma (Figure 7a)70,251, and in the proliferation and differentiation of neuronal progenitors in the cerebellar EGL, a putative cell of origin for a significant percentage of medulloblastomas47,49,252. Other members of the E2F6 complex include HP1, and MBT domain-containing Polycomb proteins (L3MBTL)233. The MBT (malignant brain tumour) domain is so named as drosophila l(3)mbt mutants have failure of neuronal differentiation, and develop invasive, malignant neuronal neoplasms in the larval brain, reminiscent of medulloblastoma253. The Polycomb protein BMI-1, which is overexpressed in the majority of medulloblastomas (Figure 13e), binds both E2F6, and L3MBTL3254. Bmi1-/- mice have a hypoplastic cerebellum54, and knockdown of BMI1 decreases growth of medulloblastoma in vitro and in vivo255.

The medulloblastoma cell line DAOY harbors a homozygous deletion on chromosome 6 that disrupts three genes: L3MBTL3, SAMD3, and TMEM200A (Table 6, Figure 12b, Figure 14a). This region of chromosome 6 was identified by GISTIC as a significant region of loss in medulloblastoma (Figure 14b). If loss of expression of L3MBTL3 provided a clonal advantage, we hypothesized that re-expression of L3MBTL3, but not SAMD3 or TMEM200A should attenuate the malignant phenotype. Stable re-expression of L3MBTL3 in DAOY results in 89

Figure 14. Re-expression of L3MBTL3 in the DAOY medulloblastoma cell line. A. Output from the UCSC Genome Browser illustrating a homozygous deletion on chromosome 6 that encompasses L3MBTL3, SAMD3, and TMEM200A. Inferred copy number data from 100K SNP array analysis of DAOY was uploaded to the UCSC Genome Browser and is shown in red. B. GISTIC output for chromosome 6 (100K SNP array) identifies a statistically significant region of extremely focal loss that includes L3MBTL3. C. MTS assay of DAOY stable cell lines. Two independent stable transfectants of the DAOY medulloblastoma cell line expressing L3MBTL3 are growth-inhibited as compared to DAOY empty vector, and SAMD3-, TMEM200A-, and GFP-expressing controls. D. Colony-forming assay of DAOY stable cell lines. Five thousand DAOY cells transfected with either L3MBTL3 or controls were seeded and grown for 7 days. There is greatly reduced growth of the cells re- expressing L3MBTL3 as compared to empty vector control. E. MTS assay of D283 stable cell lines. Overexpression of L3MBTL3 has minimal effect on the growth rate of the D283 medulloblastoma cell line. F. Increased expression of the cell cycle arrest protein p27Kip1 is observed in DAOY transfectants re-expressing L3MBTL3 as compared to empty vector control. G. No significant difference in the extent of Annexin V labeling is observed in DAOY cells re-expressing L3MBTL3 compared to controls. H. Flow cytometry analysis of DAOY cells transfected with L3MBTL3 shows a significantly decreased percentage of cells in G1 as compared to empty vector control. There is also accumulation of cells in S phase of the cell cycle in L3MBTL3 transfectants, as would be predicted in cells with decreased transcription from E2F-dependent promoters. I. ChIP followed by end-point PCR demonstrates that DAOY-L3MBTL3 transfectants show increased levels of H3K9 dimethylation in the promotor regions of the E2F6 target genes MYC, CDC25A, and TK1 as compared to controls. 91

considerably decreased proliferation as compared to controls when assessed by MTS assay and crystal violet staining (Figure 14c,d). Overexpression of L3MBTL3 in the medulloblastoma cell line D283 resulted in only a minor phenotypic change by MTS assay (Figure 14e). The critical cell cycle arrest protein p27Kip1 is transcriptionally repressed by Mycn in cerebellar EGL cells256. L3MBTL3 expressing DAOY cells exhibit elevated protein levels of p27Kip1 compared to controls (Figure 14g), as may be anticipated with blockade of MYC-dependant transcriptional repression. There was no significant difference in the incidence of apoptosis in the L3MBTL3 transfectants (Figure 14g). Flow cytometry of L3MBTL3 expressing-DAOY cells showed a marked reduction in the percentage of cells in G1 phase of the cell cycle as compared to controls (49% versus 66%, respectively), and accumulation of cells in S phase (33% versus 21%, respectively), consistent with the known cell cycle effects of ectopic E2F6 expression (Figure 14h)257. Knockdown of Drosophila L3mbt results in diminished H3K9 dimethylation of E2F responsive promotors250. ChIP experiments show that L3MBTL3 transfected DAOY cells have increased levels of H3K9 dimethylation in the promotors of three genes known to be targeted by the E2F6 complex (Figure 14i)233. This demonstrates that re-expression of L3MBTL3 can attenuate the malignant phenotype of a medulloblastoma cell line, alter the state of H3K9 methylation in known targets of the E2F6 complex, and, along with our genetic data, support a critical role for histone lysine methylation in the pathogenesis of medulloblastoma.

Proliferating neural progenitor cells in the outer EGL exit the cell cycle and migrate initially to the inner layer of the EGL, to subsequently form the post-mitotic, differentiated neurons of the IGL12,19. EHMT1 is expressed in the developing murine cerebellum at the height of EGL cell proliferation (P7) (Figure 15a,b). H3K9me2 staining in the EGL is seen predominantly in the inner, post-mitotic layer of the EGL, and not in the progenitor cells of the outer EGL (Figure 15c). Furthermore, staining for H3K9me2 co-localizes with the cell cycle arrest protein p27Kip1, (Figure 15d), demonstrating that H3K9 dimethylation occurs primarily in post-mitotic cells. While the P7 EGL has exuberant H3K9me2, there is no significant staining for H3K9me1, and only rare mitotic cells in the EGL stain for H3K9me3 (Figure 15c,e,f). Viral transduction of JMJD2C, but not EGFP in cerebellar EGL cells resulted in high levels of toxicity, as well as diminished levels of H3K9 dimethylation of cerebellar P7 EGL cells in vitro (Figure 15g). Viral transduction of NIH3T3 cells and the medulloblastoma cell line UW228 92

Figure 15. H3K9 in the developing cerebellum. A. H&E staining of the P7 murine cerebellum. The external layer of the EGL (ext-EGL), the internal layer of the EGL (int-EGL), the molecular layer (ML), and the subarachnoid space (SAS). Original magnification 400X, scale bar=50 microns (mu). B. EHMT1 staining of an adjacent section of the cerebellar EGL. Progenitor cells of the EGL are a putative cell of origin in medulloblastoma. C. H3K9me2 is noted to be more extensive in the inner, post- mitotic layer of the cerebellum, with very little staining observed in the outer, highly proliferative layer of the EGL. D. Expression of the cell cycle arrest protein p27Kip1 co-localizes with H3K9me2 in the inner EGL. E. Monomethylation of H3K9 (H3K9me1) is not observed by immunohistochemistry in the P7 cerebellum. F. Rare immunohistochemical staining for H3K9 trimethylation (H3K9me3) is observed in a small subset of mitotic cells of the P7 EGL. G. Retroviral infection of P7 EGL cells with WZL-GFP shows high efficiency of transduction (infection rate >50%), but only rare cells infected with WZL-HA-GASC1 (JMJD2C) could be identified (infection rate <1%). EGL cells expressing HA-JMJD2C have decreased levels of H3K9 dimethylation. H. Viral infection of NIH3T3 cells shows high levels of transduction for both WZL-GFP and WZL-HA-GASC1. I. Viral infection of the medulloblastoma cell line UW228 shows high levels of transduction for both WZL-GFP and WZL- GASC1. 94 was not accompanied by the same levels of toxicity, suggesting that toxicity is cell type specific (Figure 15h,i). Consistent with this, treatment of cerebellar EGL cells with the histone deacetylase inhibitor Trichostatin A results in H3K9 hyperacetylation, H3K9 hypomethlation, and high levels of cell death258. While the role of H3K9 methylation in cerebellar EGL differentiation has not been directly experimentally addressed, P19 cells cannot undergo retinoic acid induced terminal neuronal differentiation in the absence of HP1, the effector of H3K9 methylation258. These data, the known role of H3K9 methylation in embryonic stem cell differentiation, and the known defects in stem/progenitor cell compartments in L3mbtl3, Jmjd2c, Myst3, and Bmi1 mutant mice are consistent with a model in which proliferative cells in the outer EGL undergo methylation of H3K9 at the time of cell cycle exit, resulting in repression of genes that promote a progenitor cell phenotype (Figure 16a,b)259. We hypothesize that failure of physiological H3K9 methylation secondary to loss of E2F6 complex members, or erasure/blockade of H3K9 methylation results in failure of transcriptional silencing, and promotes cellular transformation in the cerebellar EGL (Figure 16c-e).

2.4 Discussion

This Chapter details how we have utilized genomics to profile and characterize the genome of a large series of human medulloblastomas. This effort disclosed the vast majority of previously well-annotated CNAs in medulloblastoma and validated the preliminary findings of multiple past and concurrent studies conducted by other groups. Importantly, our genomics strategy uncovered a novel family of genes regulating histone lysine modifications that are recurrently targeted in medulloblastoma. Moreover, our data suggest that modulation of histone lysine methylation state, particular H3K9me2, may be an important event during normal cerebellar development and deregulation of this process may contribute to malignant transformation. Indeed the observation of very focal homozygous and hemizygous deletions affecting genes regulating histone lysine modification implies that these events provided a selective clonal advantage to cells harboring these aberrations during the course of tumourigenesis.

Secondly, although observational at this stage, global loss of H3K9me2 observed in a large fraction of medulloblastomas was highly correlative with EHMT1 status, suggesting that loss of EHMT1 leads to a subsequent reduction in H3K9me2 levels that contributes to the malignant phenotype. Mice null for Glp, the mouse homolog of EHMT1, die embryonically and mouse 95

Figure 16. Model: role of H3K9 methylation in EGL development and medulloblastoma. A. Cartoon of two nucleosomes flanking a genomic locus that contains promoters with E2F and MYC binding sites. H3K9 is predominantly acetylated in cycling progenitor cells of the EGL, and therefore the chromatin is in an open configuration allowing the transcriptional machinery access with resultant gene expression from E2F- and MYC- dependant promotors. B. At the time of differentiation, the E2F6 complex is recruited to the locus, with resultant dimethylation of H3K9, followed by recruitment of HP1-gamma, a closed conformation of the chromatin, and exclusion of the transcriptional machinery. C. Failure to correctly express EHMT1 in progenitor cells results in the failure to dimethylate H3K9, and failure to repress the E2F- and MYC- dependant promoters. E2F and MYC targets are critical in expansion of the EGL, and for the formation of medulloblastoma in multiple animal models. D. Similarly, overexpression of JMJD2 family members results in demethylation of H3K9, failed recruitment of HP1-gamma, and de-repression of E2F- and MYC-dependant promoters. E. Overexpression of MYST3 results in inappropriate acetylation of H3K9, thereby blocking the ability of EHMT1 to dimethylate H3K9, and maintaining expression from E2F- and MYC- responsive promoters. 97

embryo fibroblasts (MEFs) isolated from these mice exhibit a dramatic reduction in H3K9me2 levels, indicating that Glp is the major histone lysine methyltransferase activity in the developing mouse. An identical phenotype is observed in G9a knockout mice, a histone lysine methyltrasferase that functions as an obligate heterodimer of Glp (EHMT1). It is clear from these mouse models that loss of Glp/G9a activity results in loss of H3K9me2, in agreement with what we have observed in medulloblastoma. This significant loss of H3K9me2 in medulloblastoma is also very consistent with observations reported by Andrew Feinberg’s group during the time our current study was in press. Using a ChIP-on-chip approach that specifically investigated H3K9me2 status across the genome, Wen et al confirmed that differentiated tissues harbor large, highly conserved H3K9-modified regions (up to 4.9 Mb in length) termed LOCKs (large organized chromatin K9 modifications), that are differentiation-specific, covering only approximately 4% of the genome in undifferentiated mouse embryonic stem (ES) cells, compared to 31% in differentiated ES cells, approximately 46% in liver, and 10% in brain. Importantly, these LOCKs were shown to require the activity of G9a. Finally, LOCKs were found to be substantially diminished in two cancer cell lines investigated (HCT116 and HeLa), agreeing with what we have reported in primary medulloblastomas. Although unclear at the moment, the collective results of these studies are consistent with a model in which stem/progenitor cell populations exist in a state of low H3K9me2 levels as they undergo clonal expansion and self-renewal, which is followed by a dramatic increase in the levels of H3K9me2 as these cells become post-mitotic and terminally differentiate. Furthermore, during tumourigenesis our data implies that this transition from low to high H3K9me2 is inhibited, which in turn prevents the differentiation process and permits aberrant cell division. To functionally validate these hypotheses and determine the consequence(s) of EHMT1 loss and H3K9me2 deregulation in the context of medulloblastoma, targeted deletion of Glp (EHMT1) in the developing cerebellum (i.e. CGNPs) is necessary. This strategy will permit an assessment of EHMT1’s role in cerebellar development and ideally it’s contribution to medulloblastoma pathogenesis.

Aside from EHMT1, we also identified at least 7 additional genes with a known or predicted role in modulating the post-translational modification of histones that were targeted by CNAs in medulloblastoma. L3MBTL3 re-expression in the DAOY medulloblastoma cell line provided proof-of-principle that this candidate is a putative TSG in medulloblastoma. Conversely, 98

attempts at over-expression studies of JMJD2C in primary CGNPs proved difficult and yielded little data, as achieving successful transgene expression in these cells was poor, despite having achieved highly efficient transduction of established medulloblastoma cell lines. With the exception of L3MBTL3, none of the candidates identified in primary medulloblastomas appeared deregulated in medulloblastoma cell lines, making the latter inappropriate for functional studies. To intelligently assess candidates such as L3MBTL3, SMYD4, and JMDJ2C in the context of medulloblastoma, additional genetically engineered mouse models are required.

In summary, our data highlight the genetic heterogeneity of medulloblastoma, and support an emerging theme in the literature wherein mutations of an individual gene are uncommon, but multiple rare genetic events converge on a single common pathway260. While prior successful targeted therapies for cancer focused on a single mutated gene, therapies based on pathway inhibition may be necessary and effective for some cancers261. Our results link genetic events in brain cancer with epigenetic control of gene expression and strengthen the link between improper control of the histone code and cancer. The recent identification of small molecules targeting histone lysine methylation levels262, in addition to our genetic and functional data, suggest that manipulating H3K9 methylation should be explored as a targeted therapy for medulloblastoma. 99

Chapter 3 Molecular Classification of Medulloblastoma 3 * 3.1 Introduction

Although the overall survival rates for patients with medulloblastoma have improved in recent years, the mortality rate remains significant, with survivors often suffering from neurological, endocrinological, and social sequelae as a result of current treatment options. Completely resected tumours from patients >3 years of age with no leptomeningeal dissemination at diagnosis are classified as standard-risk, whereas all others are considered high-risk. This stratification scheme does not adequately account for the prognostic variability that exists among patients ascribed to either of these risk groups. Rationale molecular-based classification methods for medulloblastoma are vital to permit accurate patient stratification, improved clinical trial design, and the future development of molecularly targeted therapies.

Published molecular markers of prognostic value in medulloblastoma include: nuclear -catenin, ERBB2, TP53, and TRKC immunopositivity263-265. Cytogenetic and genetic events including chromosome 17 aberrations, CTNNB1 mutation/monosomy 6, and MYC family amplification are informative predictors of patient outcome2,10,117,145,266-268. Recently, a molecular risk stratification approach based on multicolor FISH of MYC and MYCN loci, as well as chromosomal alterations on 17q and 6q proved more effective than conventional clinical criteria at predicting patient survival117. While statistically robust, the generalization of this technique could be limited by the geographic availability of high quality FISH laboratories.

In this Chapter, we have integrated genome-wide DNA copy number and mRNA expression profiles from a large cohort of primary medulloblastomas to generate a novel molecular classification scheme that reliably predicts patient prognosis. Since this is an immunohistochemistry-based approach, we expect that it will have widespread utility in clinical settings around the world, leading to significantly improved patient stratification.

* Work in this Thesis Chapter contributed to the following publications: Northcott PA, Korshunov A, Witt H, et al: Medulloblastoma comprises four distinct molecular variants. J Clin Oncol, In Press, 2010. 100

3.2 Materials & Methods

3.2.1 Tumour specimens and genomic datasets

All tumour specimens were obtained in accordance with the Research Ethics Board at the Hospital for Sick Children (Toronto, Canada) as described 115 and the NN Burdenko Neurosurgical Institute (Moscow, Russia)117. Copy number and expression array data were generated and analyzed as described 115,187. Content and construction of the medulloblastoma TMAs have been described previously 115,117,168,269.

3.2.2 Biostatistics and bioinformatics

Medulloblastoma samples were classified into molecular subgroups using TM4 Microarray Software Suite (MeV v4.4) by unsupervised hierarchical clustering (HCL) using the Pearson Correlation metric (average linkage) and bootstrapping analysis of high standard deviation (SD) genes. Subgroup-specific ‘signature’ genes were identified by a multivariate permutation test restricted on the proportion of false discoveries and one-way ANOVA. Principal Component Analysis (PCA) of gene expression data was performed using Partek Genomics Suite. Prediction analysis of microarrays (PAM) was carried out as described270, using the current dataset as a “training” dataset and the dataset reported by Kool et al (GEO accession: GSE 10327)38 as a “test” dataset. Pathway analysis was performed using Ingenuity Pathway Analysis (IPA, v7.5). Gene Set Enrichment Analysis (GSEA, v2.0), Non-negative matrix factorization (NMF), and Subclass Mapping (SubMap) were carried out as described271-274.

3.2.3 Immunohistochemistry

Antibodies against the following antigens were used: -catenin (BD Transduction Laboratories; 1: 100), DKK1 (Abnova; 1:100), GLI1 (Millipore; 1:5000), SFRP1 (Abcam; 1:2000), NPR3 (Abcam; 1:200), and KCNA1 (Abcam; 1:2000). TMA staining was performed, evaluated, and scored as published168. 101

3.3 Results

3.3.1 Transcriptional profiling identifies four non-overlapping subgroups of medulloblastoma with distinct demographics

Unsupervised HCL of medulloblastoma expression data identified four unique sample clusters: WNT, SHH, Group C, and Group D (Figure 1a). Support tree analysis of the clustering data reveals very high confidence for these four subgroups (96%; Figure 2a-c). The degree of separation among the four subgroups was further established by PCA (Figure 1b). Desmoplastic medulloblastomas were found predominantly in the SHH subgroup (P = 0.0023), but also found in Groups C and D (Figure 1a) 275. LCA medulloblastomas were found in SHH, Group C, and Group D. Known targets of the Wnt pathway (WIF1, DKK1, DKK2), and Shh pathway (HHIP, SFRP1, MYCN) all show clear, statistically significant differential expression in their respective subgroups (Figure 1a). Group C and D tumours both express high levels of the known medulloblastoma oncogenes OTX2135,136,226 and FOXG1B140. Group C and WNT tumours, but not SHH or Group D tumours have high levels of MYC expression (P = 8.88E-13). SHH, but not Group C or Group D tumours have high levels of MYCN expression (P = 3.321E-10), making Group D the only subgroup lacking elevated MYC family expression. As expected, monosomy 6 and 9q deletion were restricted to WNT and SHH tumours, respectively (Figure 1a) 10,37,38,145,146. Isochromosome 17q is restricted to Groups C and D (P<0.0001). MYC amplification (n=2/103) is limited to Group C, whereas MYCN amplicons (n=3/103) are distributed among SHH, Group C, and Group D.

Demographic comparison reveals notable differences between the subgroups. SHH-driven tumours are most common in infants (3 years) and adults (16 years) (P < 0.0001; Figure 1c,d). Group C tumours peak in childhood (3-10 years), but are completely absent in individuals over the age of 10 years (P = 0.0184). Group D and WNT tumours show a more distributed age of onset (median = 9-10 years), ranging from infancy to adulthood (Figure 1c,d). The published gender ratio for medulloblastoma is ~1.5:1, male:female, and is 1.55:1 in our cohort. Notably, ~70% of males belong to Group C/D compared to only ~42% of females (Figure 1e; P = 0.0142). Conversely, ~47% of females are classified as SHH tumours versus only ~24% of males (P = 0.0312). The incidence of WNT tumours in female patients (~11%) is nearly double that of males (~6%). 102

103

Figure 1. Transcriptional profiling of medulloblastoma identifies four non-overlapping molecular subgroups with distinct demographics. A. Unsupervised HCL of Human 1.0 Exon array expression data from 103 primary medulloblastomas using 1450 high SD genes. Clinical features (age group, gender, and histology) for the 103 samples included in the study are shown below the dendrogram. Age Group: infants (3 years; blue), children (4-15 years; green), adults (16 years; red), unknown (black). Gender: males (blue), females (pink). Histology: classic (white), desmoplastic (grey), LCA (orange), MBEN (brown), unknown (black). Statistical significance for the different clinical features was determined using Chi-square (age group) and Fisher’s exact test (gender, histology). *P-value determined by comparing gender prevalence in WNT/SHH tumours versus Group C/D tumours using Fisher’s exact test. **P-value corresponds to over-representation of desmoplastic tumours in the SHH subgroup as determined using Fisher’s exact test. The heatmap below the dendrogram shows the expression profile for 10 genes well characterized in medulloblastoma and demonstrates their significant pattern of differential expression among the four subgroups. Statistical significance of differential gene expression was determined using one-way ANOVA. Common genomic aberrations known to occur in medulloblastoma are shown below the heatmap. Blue boxes indicate loss/deletion, red boxes indicate gain/amplification, and white boxes denote balanced copy number state for the specified genomic aberration. B. PCA of the primary medulloblastomas described in (A) using the same 1450 high SD genes employed in clustering. Individual samples are represented as colored spheres (blue = WNT, red = SHH, yellow = Group C, green = group D) and ellipsoids represent 2 SD of the data distribution for each subgroup. C. Age at diagnosis distribution for each of the four medulloblastoma subgroups. D. Pie charts showing the frequency (%) of the four molecular subgroups among three patient age categories: “infants” (3 years), “children” (4-15 years), and “adults” (16 years). E. Pie charts showing the frequency (%) of medulloblastoma subgroups by gender. 104

105

106

107

Figure 2. Unsupervised HCL of medulloblastomas consistently identifies four distinct subgroups. A. Bootstrap analysis of 103 primary medulloblastomas using 1450 high SD genes. Number of genes employed in clustering was determined empirically by selecting the minimal number of high SD genes that showed the highest support in subsequent bootstrap analysis. Samples were clustered using the Pearson Correlation metric (average linkage) and support tree analysis denotes the percentage of times a given node was supported following 1000 iterations. B. Enlarged support tree result from HCL shown in (A). There is strong support (96%) for the discrimination of four medulloblastoma subgroups. C. Bootstrap analysis of medulloblastomas described in (A) and (B) using 1300, 1400, and 1500 high SD genes. Solid colored line shown above the dendrogram indicates the medulloblastoma subgroup affiliation (blue = WNT, red = SHH, yellow = Group C, green = Group D) as determined in (A). 108

3.3.2 Molecular classification of medulloblastoma by transcriptional profiling

An 84-gene classifier of medulloblastoma subgroup ‘signature’ genes derived from the current dataset using PAM270 was applied to a published dataset of 62 medulloblastomas reported by Kool et al38, separating this cohort into the current four subgroups (Figure 3a, Figure 4a-c). Comparison of the 20 most highly differentially expressed genes discriminating Groups C and D in both datasets demonstrates that 38/40 genes are concordantly differentially expressed (P < 0.05, Figure 3b). Application of NMF273, an unsupervised bioinformatic tool for determining the number of independent classes within an expression dataset, strongly supports the existence of four medulloblastoma subgroups (Figure 3c, Figure 5a-c). Direct comparison of clustering results obtained by HCL and NMF confirmed strong concordance between the independent analyses (Rand index = 0.9309, adjusted Rand index = 0.8287)276. Another bioinformatic algorithm, SubMap,274 suggests Kool subgroups C and D both correspond to our Group D (P<0.01, Figure 3d). In the Kool dataset, adult tumours (16+ years) were devoid of Group C cases. Notably, there was a very high rate of M+ patients in Group C in this dataset (75%; P = 0.0039, Figure 3e).

IPA identified the top canonical signaling pathways over-represented in medulloblastoma subgroups (Table 1). As expected, Wnt signaling was enriched in the WNT group tumours (P = 0.00069), and Shh signaling in the SHH subgroup (P = 0.00058). However, Wnt pathway genes were also enriched in SHH and Group C, but not Group D. Both WNT and SHH group tumours had an over-representation of genes involved in axonal guidance. Both Group C (phototransduction and glutamate signaling) and Group D (semaphorin, cAMP, G-protein coupled receptors, and -adrenergic signaling) were characterized by an over-representation of pathways involved in neuronal development. GSEA271 comparison of genes discriminating Group C and Group D demonstrated that genes up-regulated in Group C positively correlate with genes associated with medulloblastoma treatment failure identified by Pomeroy et al,159 and with genes associated with elevated MYC levels (Figure 6). 109

110

Figure 3. Molecular classification of medulloblastoma by transcriptional profiling. A. Left panel. Supervised analysis of the four medulloblastoma subgroups showing the most differentially expressed ‘signature’ genes that discriminate the subgroups. The top 25 ‘signature’ genes for each subgroup are shown. Right panel. PAM was used to identify a gene signature that could robustly classify our “training” dataset into the four molecular subgroups and predict the subgroup affiliation of samples in a “test” dataset of 62 medulloblastomas published by Kool et al38. The expression profile of ‘signature’ genes identified in the “test” dataset is shown for the predicted subgroups of the Kool dataset. B. Expression heatmap showing 40 significant genes discriminating Group C and Group D medulloblastomas in our cohort (Left panel) and the same 40 genes described in Group C and D tumours from the Kool dataset (Right panel). C. Consensus NMF analysis of our medulloblastoma cohort provides strong statistical support for the existence of four medulloblastoma subgroups. Agreement between HCL and NMF clustering data was supported by calculation of the Rand index (Rand index = 0.9309, adjusted Rand index = 0.8287). D. SubMap analysis comparing the four medulloblastoma subgroups identified in the current study to the five subgroups previously reported by Kool et al. SubMap supports the existence of four medulloblastoma subgroups: (WNT = Kool-A, SHH = Kool-B, Group C = Kool-E, Group D = Kool-C/Kool-D). E. Incidence of metastasis in predicted medulloblastoma subgroups of the Kool dataset. Subgroup affiliation was predicted for the Kool samples using PAM as described in (A), and patient metastatic status was then plotted for each of the predicted subgroups. Significance was assessed by Fisher’s exact test. 111

112

Figure 4. Classification of medulloblastoma using PAM. A. Cross-tabulation of true (rows) versus predicted (columns) classes for the PAM fit. In the “training” dataset, all WNT, SHH, and Group D tumours were classified correctly by PAM, whereas 2 Group C samples were erroneously classified as Group D (cross-validation mis-classification rate ~2%). B. Composition of the PAM classifier. Shrunken centroids for the 84 genes employed as the PAM classifier are depicted for each of the 4 molecular subgroups. C. Correlation matrix showing the performance of the PAM classifier on the “test” dataset of Kool et al. Rows indicate sample classification according to Kool et al, and columns represent the subgroup assignment predicted by PAM. All Kool62-“A” and Kool62-“B” tumours were accurately classified as WNT and SHH tumours, respectively. Kool62- “C/D” tumours were classified as Group D medulloblastoma, whereas Kool62-“E” group samples were predominantly classified as Group C (9/11) and rarely as Group D (2/11) tumours. 113

114

Figure 5. Consensus NMF analysis of medulloblastoma expression data. A-C. Consensus NMF was performed on our expression dataset to determine the most robust number of classes (k) in medulloblastoma. Values of k ranging from 2-8 were tested using 1000-4000 high SD genes. Results generated from gene inputs of 1000 (A), 2000 (B), and 3000 (C) high SD genes are shown. 115

116

117

Figure 6. Gene set enrichment analysis (GSEA) of medulloblastoma subgroups: overlap with chemical and genetic perturbation gene sets. Enrichment plots generated in GSEA showing significant overlap between chemical and genetic perturbation (CGP) gene sets from the Molecular Signatures Database (MSigDB; Broad Institute) and gene signatures defining medulloblastoma subgroups. GSEA was performed by contrasting each medulloblastoma subgroup to the rest (with the exception of Group C vs Group D) to identify overlap with CGP gene sets. Plots for select gene sets that positively correlate with a particular medulloblastoma subgroup are shown. 118

3.3.3 Subgroup-specific genetic events in the medulloblastoma genome

With the exception of i{17}q, which is reported in ~30-50% of cases115,117, most of the known chromosomal changes in medulloblastoma occur at low frequency. To determine if certain genetic aberrations were more prominent when accounting for subgroups, we manually catalogued all chromosomal changes in our dataset in a subgroup-specific manner (Figure 7a,b; Figure 8). Not surprisingly, monosomy 6 was found exclusively in WNT tumours 37,38,145,187, and 9q loss was found only in SHH tumours38.

Multiple novel regions of genomic aberration were identified in SHH tumours, including chromosome 9p gain (P = 0.0005), which often co-occurred with 9q loss (i.e. i{9}p), as well as gains of 3q (P = 0.0015), 20q (P = 0.0331), and 21q (P = 0.0124). Chromosome 10q loss was primarily limited to SHH and Group C tumours (P = 0.0034). Events significantly over- represented in Group C included 1q gain (P = 0.0004), distal 5q loss (P = 0.0038), and 16q loss (P = 0.0055). Multiple events were over-represented in both Group C and D compared to WNT/SHH tumours, including gain of chromosomes 17q (P = 0.0014) and 18 (P = 0.0136), i(17)q (P < 0.0001), and loss of 11p (P = 0.0005). Despite the commonalities between Group C and D, these subgroups could be discriminated by several genetic events that are significantly more common in Group C, including 1q gain (P = 0.0021), loss of distal 5q (P = 0.0689), and 10q loss (P = 0.0168; Figure 7c). Isochromosome 17q was significantly more prominent in Group D (23/35; 65.7%) than Group C (7/27; 25.9%; P = 0.0024), and loss of the X chromosome occurs more commonly in Group D females (P = 0.0002; Figure 7b).

The bioinformatic tool GISTIC215 delineated statistically significant, subgroup-specific genetic changes in an unbiased manner. GISTIC confirmed all of the large genomic abnormalities identified by previous manual curation of the dataset, and revealed several novel subgroup- specific amplifications and deletions (Figure 9a,b). GISTIC identified gain of chromosome 2 (q- value = 0.2412-0.0019) and deletion of chromosome 14 (q-value = 0.7534-0.0017) as significant in the SHH subgroup. Focal MYC amplification on 8q24 was identified exclusively in Group C tumours (q-value = 0.0011) (Figure 9a). Loss of chromosome 8p loss was highly significant in Group C (q-value = 0.5086-0.0005), whereas Group D tumours exhibited significant loss on both 8p and 8q (q-value = 0.1662-0.0010) (Figure 9b). 119

120

Figure 7. Subgroup-specific genetic events in the medulloblastoma genome. A. Distribution of statistically significant, subgroup-specific CNAs identified by manual curation of our medulloblastoma series. Significance was determined by Fisher’s exact test. Blue boxes indicate loss/deletion, red boxes indicate gain/amplification, and white boxes denote balanced copy number state for the specified genomic aberration. B. Frequency of subgroup-specific CNAs within the four medulloblastoma subgroups. C. Comparison of statistically significant CNAs over-represented in Group C and Group D medulloblastomas. Significance was determined by Fisher’s exact test. 121

122

Figure 8. Copy number summary plots for medulloblastoma subgroups. Summary plots showing the frequency of gains (red) and losses (blue) in the medulloblastoma genome for the four different subgroups. 500K SNP array data was analyzed in dChip and frequency of gains (HMM copy number 2.75) and losses (HMM copy number 1.25) observed on each chromosome are shown. Subgroup-specific copy number changes are discussed in detail in the text. WNT (n=5), SHH (n=26), Group C (n=21), Group D (n=29). 123

Figure 9. GISTIC analysis of medulloblastoma subgroups identifies statistically significant subgroup-specific CNAs. A, B. GISTIC analysis of the four medulloblastoma subgroups represented in our data series. Significance plots show regions of statistically significant gains/amplifications (A) and losses/deletions (B) in the medulloblastoma subgroups. Yellow arrows indicate copy number aberrations restricted to a single subgroup and green arrows mark regions that are significant in multiple subgroups. Loci marking select candidate genes are shown for reference. 124

3.3.4 Molecular heterogeneity in SHH-driven medulloblastoma

Excessive Shh signaling is implicated in the pathogenesis of human and mouse medulloblastoma22,25,37,38,187,277,278. Despite consistent upregulation of Shh targets including HHIP, SFRP1, and MYCN, there is considerable inter-tumoural genetic heterogeneity in this subgroup (Figure 7a). Unsupervised HCL of SHH cases (n=33) identified three different groups of SHH-driven medulloblastomas (Figure 10a). Interestingly, all adult SHH tumours cluster together. PCA also demonstrates the distinction between adult and infant SHH medulloblastomas (Figure 10b). Overlay of chromosome 9q and 10q loss status on the clustering data, demonstrates that 10q loss is absent in adult SHH medulloblastomas (Figure 10a).

3.3.5 Validation of medulloblastoma subgroups by immunohistochemistry

We selected highly expressed, subgroup-specific ‘signature’ genes based on our microarray expression dataset (Figure 3a, Figure 11a) and the availability of high quality commercial antibodies. Staining two separate medulloblastoma TMAs using these antibodies, as well as - catenin and GLI1, demonstrated robust staining in a subgroup-specific manner (Figure 11b). Remarkably, 288/294 (~98%) tumour samples stained positive for a single marker protein, a failure rate of only ~2.1% (P < 0.0001; Figure 11c). Demographics for the TMA patients validate the results from our microarray dataset (Figure 11d-f, Figure 1c-e), confirming that SHH tumours occur primarily in infants and adults (P < 0.0001). Group C tumours were largely confined to childhood (P < 0.0001) with two exceptions, both aged 18 yrs (Figure 11d,e). WNT group tumours were almost three times more common in females than males (17% vs 6%; P = 0.0046; Figure 11f). Metastases were significantly over-represented in Group C (46.5%; P = 0.0006), followed by Group D (29.7%), similar to the Kool dataset (Figure 11g, Figure 3e). Overall survival in the DFKZ cohort (n=236; Figure 12a), the JHU cohort (n=50; Figure 12b), and both cohorts combined (n=287; Figure 12c) demonstrates that Group C tumours have the worst prognosis (logrank P < 0.001). Replotting the overall survival curves, dividing each subgroup into M0 and M+ sub-subgroups shows that Group C patients exhibit a dismal prognosis regardless of M stage (Figure 12d). A multivariate analysis evaluating age, M stage, histology, extent of resection, and subgroup (WNT/SHH/Group C/Group D) revealed that only 125

Figure 10. Molecular heterogeneity in SHH-driven medulloblastomas. A. Unsupervised HCL of SHH-driven medulloblastomas (n=33) using 1350 high SD genes. Age Group: “infants” (blue), “children” (green), and “adults” (red). Chromosome 9q and 10q copy number status are included below the clustering dendrogram (blue=loss, white=balanced). B. PCA of infant (blue) and adult (red) SHH-driven medulloblastomas using the same 1350 genes included in (A). 126

Figure 11. Validation of medulloblastoma subgroups by immunohistochemistry. A. Differential expression of four selected genes (DKK1, SFRP1, NPR3, and KCNA1) in the four medulloblstoma subgroups as determined by Exon array expression profiling. B. Representative immunohistochemistry for -Catenin, DKK1, SFRP1, GLI1, NPR3, and KCNA1 on a medulloblastoma TMA. C. Results obtained from immunohistochemical staining of two independent medulloblastoma TMAs (DFKZ and JHU) consisting of 294 non-overlapping primary cases. Pie chart demonstrates that 288/294 (~98%) of tumours stained positive for a single marker (DKK1, SFRP1, NPR3, or KCNA1), 1/294 (~0.3%) stained positive for multiple markers, and 5/294 (~1.7%) did not stain for any of the four markers. Significance was determined using Fisher’s exact test. D. Age at diagnosis distribution for each of the four medulloblastoma subgroups as determined by immunohistochemical staining of the medulloblastoma TMAs. E. Pie charts showing the frequency (%) of the four molecular subgroups based on TMA staining among three patient age categories: “infants” (3 years), “children” (4-15 years), and “adults” (16 years). F. Pie charts showing the frequency (%) of medulloblastoma subgroups by gender based on TMA staining. G. Incidence of metastases by subgroup as determined by TMA staining. Significance was assessed by Fisher’s exact test. 127

128

Figure 12. Prognostic significance of medulloblastoma subgroups. A. Kaplan-Meier analysis showing overall survival probability for patients on the DFKZ medulloblastoma TMA (n=236) separated by subgroup. B. Kaplan-Meier analysis showing overall survival probability for patients on the JHU TMA (n=50) separated by subgroup. C. Combined overall survival probability for both the DFKZ and JHU TMAs (n=287) separated by subgroup. D. Combined overall survival probability for both TMAs showing medulloblastoma subgroups separated by metastatic status. E. Kaplan-Meier analysis discriminating LCA (green) from non-LCA (red) histology in Group C medulloblastomas present on the TMAs. 129

LCA histology and Group C subgroup were prognostic (Figure 13a,b). Indeed, Group C tumours with LCA histology have the worst prognosis (Figure 12e).

3.4 Discussion

Prior attempts to subgroup medulloblastoma using genomics identified five or six subgroups37,38. Both studies clearly show the WNT group of tumours clustering alone, while SHH driven tumours did not always segregate as a single subgroup37. Both the current and Kool datasets show that although WNT tumours are classic medulloblastomas and most desmoplastic medulloblastomas are SHH tumours, the WNT and SHH subgroups always cluster together, suggesting that they are biologically similar. Although desmoplastic tumours were most commonly seen in the SHH group, they were also found in Group C and Group D, supporting the known difficulty in making a histological diagnosis of nodular desmoplastic medulloblastoma275. We observed LCA tumours in SHH, Group C, and Group D tumours, and others have described the LCA phenotype in WNT tumours10. Interestingly, children with a WNT subgroup LCA medulloblastoma have been reported to have a good prognosis10.

SHH tumours were seen in both infants and adults, whereas WNT and Group D tumours are distributed across all age groups. Group C tumours peak in childhood, are very rarely seen in older teenagers, and are never seen in adults. The paucity of Group C tumours and predominance of Group D tumours seen in adolescence may account for distinct clinical course and pattern of relapse that has been described in adolescents with medulloblastoma 7. The SJMB-96 study showed that the sex ratio for high-risk cases of medulloblastoma was 3.9:1 (M:F), and several reports have suggested that females with medulloblastoma have a better outcome than males7-9. We believe this is attributable to a higher female incidence of WNT tumours, and perhaps SHH tumours. Indeed the improved survival among females is only seen among older children and adults (where WNT tumours are more prevalent) highly supporting a WNT subgroup prevalence as an explanation for the improved outcome in females8. Prior clinical trials studying medulloblastoma have tended to divide patients by age group, with infants often being studied separately. Our results suggest that infants with medulloblastoma should be broken down by subgroup and that infants with Group C/D tumours may account for the children who respond poorly to current therapies279. This is consistent with identified prognostic factors of desmoplasia (good) and metastatic status (poor), which might be identifying SHH group and 130

131

Figure 13. Survival probability of subgroups and histological subtypes of medulloblastoma. A. Overall survival (OS) and progression-free survival (PFS) probabilities for medulloblastoma subgroups based on staining of the two non-overlapping TMAs. B. OS and PFS for histological subtypes of medulloblastoma based on the medulloblastomas present in the TMA cohorts. 132

Group C tumours, respectively279.

Analysis of 62 medulloblastomas by Kool identified three non-SHH/WNT medulloblastoma subgroups, while our analysis of 103 medulloblastomas revealed only two non-SHH/WNT subgroups. The statistical support for three non-SHH/WNT subgroups as reported by Kool was either 21% or 41%, depending on the number of genes employed in support tree analysis 38. In the current manuscript, both HCL and PCA demonstrate that Group C is distinct from Group D with strong statistical support. NMF supports the existence of four subgroups of medulloblastoma. The initial obvious difference between Group C and Group D is that while MYC is highly expressed in Group C and WNT tumours, and MYCN is highly expressed in SHH group tumours, neither MYC nor MYCN is highly expressed in Group D tumours. Additionally, our data shows that Group C and Group D have different demographics, rate of metastases, genetic profiles, and clinical outcomes. Our immunohistochemical analysis of medulloblastoma TMAs revealed that only 1/294 medulloblastomas stained positive for both Group C and Group D signature genes demonstrating the lack of overlap between the two groups. NMF and SubMap analysis also support the existence of two non-SHH/WNT groups of medulloblastoma. Taken together, all of these data highly support the existence of two non-SHH/WNT subgroups. We have chosen to call these two groups Group C (childhood tumours with high MYC levels, frequent CSF dissemination, and a crummy prognosis) and Group D (distributed demographics).

Our novel four antibody approach to subgroup medulloblastoma should be broadly applicable across the globe as the technique does not require RNA, FISH, PCR, microarray technology, or any other techniques from molecular biology, but rather uses immunohistochemistry on archival specimens which is in routine use in most neuropathology laboratories around the world. Furthermore, as marker staining is quite robust, we hope that there will be little inter-observer variability.

Analysis of overall survival demonstrates a marked reduction in survival for children with Group C medulloblastoma regardless of metastatic stage. This suggests that some children with ‘average-risk’ medulloblastoma who subsequently recur likely belong to Group C. Indeed, the failure of metastatic stage to predict prognosis in our multivariate analysis suggests that the published association between poor prognosis and metastatic stage may be attributable to the high rate of metastases in Group C tumours. 133

Historically, small round blue cell tumours of the cerebellum have been grouped under the rubric of medulloblastoma. Recently, AT/RTs were shown to represent a distinct subgroup of blue cell tumours based on different histopathology and, subsequently, a lack of expression of the hSNF5/INI1 TSG280,281. Our data suggests that the histological entity of medulloblastoma comprises four diseases that are demographically, clinically, transcriptionally, and genetically distinct. These four non-overlapping types of medulloblastoma can be distinguished through the application of four commercial antibodies on formalin-fixed, paraffin-embedded tumour material. Future clinical trials should prospectively validate this immunohistochemical approach. Most importantly, due to the non-overlapping character of the four types of medulloblastoma, we suggest that targeted therapies may need to be developed against each subtype individually as they are biologically distinct. 134

Chapter 4 Identification of Oncogenic miRNAs in Medulloblastoma 4 * 4.1 Introduction MiRNAs belong to a recently described class of small regulatory RNAs that coordinate important cellular processes, including differentiation, proliferation and apoptosis. The discovery of miRNAs as key post-transcriptional regulators of the cell’s transcriptome has led to an intense amount of research aimed at the characterization of these molecules both in normal development and in human disease, particularly cancer. Indeed, in the past five years, there has been irrefutable evidence in support of miRNAs as a new class of oncogenes and tumour suppressors175-177,282,283. To date, only a handful of studies have explored the role of miRNAs in medulloblastoma pathogenesis284. Given that medulloblastoma is an embryonal tumour exemplifying failure of a normal developmental process, it is conceivable that deregulation of miRNA expression contributes to failed cerebellar development and thus medulloblastoma. In our high-resolution copy number profiling described in Chapters 2 and 3, we identified high-level amplification of miR-17/92, a polycistronic cluster of highly conserved miRNAs that have been shown to contribute to tumour development in both human and murine cancers285,286. MiR-17/92 is located on chromosome 13 in humans (chr 14 in mice); paralogous clusters also exist including miR- 106a/363 and miR-106b/25285,286. At the time of this study, a role for miR-17/92 in medulloblastomas and cerebellar development had not been described.

As mentioned throughout this thesis, CGNPs are propable cells-of-origin for the subset of medulloblastomas characterized by aberrant activation of the Shh signaling pathway. CGNPs undergo rapid Shh-dependent expansion perinatally in mice and humans, and excessive Shh pathway activity promotes medulloblastoma17,19,287. In this Chapter, we demonstrate that miR-

* Work in this Thesis Chapter contributed to the following publications: Northcott PA, Fernandez LA, Hagan JP, et al: The miR-17/92 polycistron is up-regulated in sonic hedgehog-driven medulloblastomas and induced by N-myc in sonic hedgehog-treated cerebellar neural precursors. Cancer Res 69:3249-55, 2009. 135

17/92 is amplified and overexpressed in medulloblastoma, particularly in the medulloblastoma subgroup driven by Shh signaling187. In addition, we show that the miR-17/92 cluster is a target of Shh signaling through N-myc activity in CGNPs. Over-expression of miR-17/92 synergized with exogenous Shh in promoting CGNP proliferation and was able to drive proliferation in the absence of Shh signaling. These findings suggest that miR-17/92 is an essential component of the Shh mitogenic signaling apparatus in CGNPs, and that its upregulation downstream of aberrantly activated Shh contributes to medulloblastoma.

4.2 Materials and Methods

4.2.1 Medulloblastoma tumour specimens

4.2.2 100K and 500K GeneChip Mapping arrays

Medulloblastoma samples were processed and hybridized to Affymetrix Genechip Human Mapping 100K and 500K Array sets at TCAG at the Hospital for Sick Children (Toronto, Canada). SNP arrays were processed and the data analyzed for copy number changes as described in Chapter 2 of this thesis115.

4.2.3 MicroRNA arrays

MicroRNA expression profiling of primary human medulloblastomas was performed on Trizol- extracted (Invitrogen) total RNA using The Ohio State University Comprehensive Cancer Center Version 3.0 microRNA microarray. The resulting raw data was quantile normalized and 288 thresholded to a value of 4.5 (log2) as described previously . Using only probes designed against known human mature microRNAs, statistical analyses were performed using BRB- ArrayTools developed by Dr. Richard Simon and the BRB-ArrayTools Development Team. Reported fold-changes reflect ratios of geometric means and p-values for informative probes were calculated using the class comparison and class prediction tools. Murine microRNA profiling was carried out using total RNA isolated with the mirVana extraction kit (Ambion) and 136

hybridization to LC Sciences (Houston, TX) miRNA microarrays. These arrays detect miRNA transcripts corresponding to 599 mature miRNAs contained in the Sanger miRBase Release MiRMouse 11.0. Raw data were median normalized across arrays and miRNAs whose expression displayed greater than two-fold differences across triplicate samples when comparing vehicle-treated to Shh-treated samples were validated as described below.

4.2.4 TaqMan miRNA assays

Expression of MYCN and MYC were quantified relative to ACTB using Platinum SYBR Green qPCR SuperMix UDG (Invitrogen). Primer sequences are available upon request. Taqman microRNA assays (Applied Biosystems) were used to quantify mature miRNA expression as previously described289. Samples were analyzed using StepOnePlus and ABI Prism 7900HT Sequence Detection Systems (Applied Biosystems). Statistical significance of qRT-PCR data was determined using two-sample Wilcoxon and Student’s T-tests.

4.2.5 Affymetrix exon arrays

Affymetrix exon array profiling of medulloblastoma samples was performed as described in Chapter 3 of this thesis. Medulloblastoma samples were classified into molecular subgroups using TM4 Microarray Software Suite (MeV) by unsupervised HCL using Pearson Correlation and bootstrapping analysis of 1300 high SD genes. Clustering trees showed at least 97% statistical support for the segregation of samples into four unique subgroups. Significant genes representative of the individual subgroups were identified using T-tests with standard Bonferroni correction.

4.2.6 Fluorescence In Situ Hybridization (FISH)

FISH for hsa-mir-17/92 was carried out on a medulloblastoma tissue microarray as previously published37. BAC clones used in the analysis included PR11-97P7 (hsa-mir-17/92; 13q31.3), as well as RP11-936K15 and RP11-539J14 (13q12.11) as adjacent controls.

4.2.7 Primary CGNP cultures

Harvest of neural precursors from neonates and preparation of cerebella from wild-type and mutant mice for histological analysis were carried out in compliance with the Memorial Sloan- Kettering Institutional animal care and use committee guidelines. CGNP cultures were 137

generated and treated with SHH (R&D Systems) as previously described49. The miR-17-19b cassette (gift of Jason Huse and Eric Holland, Memorial Sloan-Kettering Cancer Center) was cloned into the retroviral vector pWzl-ires-GFP. Retrovirus production and infection of CGNPs and subsequent immunostaining were carried out as previously described49. Stained cells were visualized with a Leica DM5000B microscope and images taken using Leica FW400 software. For quantification, TIFF images of 4 random fields were taken for each experimental group using the 20X objective and quantified using Volocity.

4.3 Results

4.3.1 The miR-17/92 cluster is recurrently amplified in medulloblastoma

As described in detail in Chapter 2 of this thesis, we profiled 201 primary human medulloblastomas using Affymetrix SNP arrays to delineate recurrent CNAs contributing to medulloblastoma pathogenesis115. This effort identified two medulloblastomas with recurrent, focal, high-level amplification on chromosome 13q31.3, sharing a minimal common region that spans ~1.82 Mb (Figure 1a). Only two genes mapped to this amplified locus – GPC5, a cell surface proteoglycan that bears heparan sulfate, and C13orf25, a non-coding host gene for the miR-17/92 polycistron (NCBI Build 36.1; Figure 1b)290, an oncogenic miRNA cluster well- characterized in haematological malignancies and other solid tumours285,291,292. Since a role for GPC5 had not been previously implicated in cancer, we chose to focus our attention on miR- 17/92 as the more likely target of the amplicon. Amplification of this miRNA cluster had not been previously reported in medulloblastoma. Further examination of the tumours harboring miR-17/92 CNAs revealed amplification of MYCN and GLI2 (Figure 1a). We subsequently carried out FISH on a TMA to determine the incidence of miR-17/92 amplification in a non- overlapping series of 80 medulloblastomas. Low-level amplification of miR-17/92 was identified in ~6% (5/80) of cases (Figure 1c). Taken together, these results suggest that miR- 17/92 may function as an oncogene in a subset of medulloblastomas.

4.3.2 miR-17/92 is overexpressed in human and murine medulloblastomas

To gain an appreciation for the extent of miRNA deregulation in medulloblastoma, we performed a genome-wide survey of 548 mature miRNAs in a series of 90 primary human medulloblastomas and 10 normal human cerebella (CB; 5 fetal, 5 adult). Unsupervised HCL of 138

samples and differentially expressed miRNAs in the dataset could easily discriminate medulloblastomas from normal CB samples (Figure 2a). Notably, components of miR-17/92 including miR-18a, miR-19b, and miR-20a, as well as the paralogous miR-106a (miR-106a/363 cluster) clustered together, were expressed at low levels in the normal CB, and showed considerably higher levels of expression in the majority of medulloblastomas (Figure 2a). Statistical comparison of miRNA profiles for medulloblastomas versus normal CB samples revealed consistent overexpression of miR-17/92 and its related paralogs (miR-106a/363 and miR-106b/25) in medulloblastoma (Figure 2b left panel, c). Re-analysis of the data after removing the 22 probesets which detect components of miR-17/92 or paralogous clusters, shows that there are relatively few remaining overexpressed miRNAs in medulloblastoma compared to normal CB (Figure 2b right panel). As miR-17/92 was overexpressed in a large percentage of human medulloblastomas compared to normal CB, we next examined its expression in murine models of medulloblastoma. Medulloblastomas derived from ND2:SmoA1 and Ptc+/- mice, two well-described mouse models of the disease25,30, showed marked over-expression of the miR- 17/92 cluster compared to cerebellum from age-matched tumour-free littermates (Figure 2d).

4.3.3 miR-17/92 upregulation is associated with activated Shh signaling in human medulloblastoma

Aberrant activation of the Shh pathway through mutation of pathway members has been documented in 25-30% of medulloblastomas37,38. To classify the 90 medulloblastomas employed in miRNA expression profiling above into medulloblastoma subgroups, we performed unsupervised HCL using mRNA expression data generated from Affymetrix exon arrays as described in detail in Chapter 3 of this thesis. Unsupervised analysis using 1300 differentially expressed mRNAs segregated the tumours into the four characteristic molecular subgroups: WNT (blue), SHH (red), Group C (yellow), and Group D (green) (Figure 3a, b). As anticipated, these four subgroups were supported by their expression pattern and specific genomic features, including monosomy 6 (WNT), chromosome 9q loss (SHH), and i{17}q (Group C and Group D) (Figure 4a). Of particular interest, miR-17/92 was most highly expressed in the SHH subgroup, followed by Group C, and WNT medulloblastomas (Figure 4a, b, Figure 5a-f). Confirming previous reports159, we observed high MYCN expression in the SHH tumours, whereas MYC levels were most elevated in WNT and Group C tumours (Figure 4a, Figure 6a,b). MYC and MYCN have both been reported to transcriptionally regulate miR-17/92293,294. We compared 139

140

Figure 1. Recurrent amplification of the miR-17/92 locus in primary human medulloblastomas. A. Top panels: Focal, high-level amplification of miR-17/92 on chromosome 13q31.3 in 2 medulloblastomas. These samples also show amplification of MYCN and GLI2 (MB-184 and MB-7, respectively; lower panels). B. Representative FISH for a medulloblastoma with amplification of miR-17/92 using a bacterial artificial chromosome (BAC) clone mapping to the miR-17/92 locus labeled in red and a chromosome 13 centromere control probe labeled in green. 141

142

Figure 2. Overexpression of miR-17/92 in human and murine medulloblastomas. A. Heatmap showing that components of the miR-17/92 polycistron (yellow box) are expressed at low levels in the normal cerebellum, are expressed at elevated levels in medulloblastomas, and cluster together based on their common pattern of expression. Other miRNAs including miR-375, miR- 182, and miR-183 are also elevated in a subset of tumours. B. Scatterplot analysis showing that components of miR-17/92 and related paralogs (miR-106a and miR-106b clusters) are the most significantly upregulated miRNAs in human medulloblastoma compared to normal cerebellar samples (left panel). Mature miRNAs are depicted as circles with diameter determined by the ratio of expression for a given miRNA in the brain versus other tissues and the colors reflect the parametric p-values between the two classes (i.e normal versus tumour). Performing the same analysis between medulloblastomas and normal cerebellum without miR-17/92, miR-106a, and miR-106b clusters demonstrates that only a few miRNAs independent of these clusters are significantly upregulated in medulloblastoma (right panel). C. Summary of the top 20 most significantly overexpressed miRNAs in medulloblastoma compared to normal cerebellum samples. 11/20 of the significantly upregulated miRNAs correspond to components of miR-17/92 or related paralogs, providing strong evidence in support of miR-17/92 overexpression in medulloblastoma. D. Taqman qRT-PCR analysis of murine medulloblastomas from SmoA1 (upper panel) and Ptc+/- (lower panel) mice shows upregulation of the miR-17/92 cluster in tumours from both of these mouse models. Data for components of the miR-17/92 cluster (blue) or controls (red) is represented as the fold-change between medulloblastomas from either SmoA1 or Ptc+/- mice and age-matched control cerebella. 143

Figure 3. Unsupervised HCL of 90 medulloblastomas identifies four unique subgroups. A. Heatmap showing expression profile of 1300 high SD genes used for clustering. B. Support tree result for the clustering shown in A. Numbers indicate the statistical support for the nodes of the trees (%), based on resampling of the data by bootstrapping genes. 144

145

Figure 4. miR-17/92 is overexpressed in SHH-dependent medulloblastomas and tumours with elevated MYC family expression. A. Heatmap showing expression of MYC, MYCN, and the miR-17/92 cluster suggests a strong correlation between miR-17/92 overexpression and the SHH subgroup as defined by molecular classification of the 90 primary medulloblastomas used in miRNA profiling. Expression array analysis of protein-coding genes identifies four consistent subgroups: WNT (blue), SHH (red), Group C (yellow), and Group D (green). Identification of subgroup-specific expression ‘signatures’ was achieved by comparing each of the respective subgroups to the other three using T-test statistics. miR-17/92 expression is also elevated in the WNT and Group C subgroups, both of which exhibit higher MYC expression, compared to Group D which is characterized by lower MYC levels. B. Representative miRNAs of the four molecular subgroups described in A identified using T-test statistics. miR-17/92 expression is highest in the SHH subgroup and lowest in Group D. C. Division of the medulloblastoma series into higher MYC/MYCN (n=51) and lower MYC/MYCN (n=39) expressing groups identifies highly significant upregulation of miR- 17/92 and related paralogs in the higher MYC/MYCN expressing tumours. Scatterplot analysis shows significant, differentially expressed miRNAs between the two groups of medulloblastomas. D. qRT-PCR validation of the expression array data in a subset of medulloblastomas (n=30) confirming the consistent correlation between MYC/MYCN status and miR-17/92 expression. 146

Figure 5. miR-17/92 upregulation in specific medulloblastoma subgroups. A-F. Boxplots showing expression data as determined by miRNA array for components of the miR-17/92 polycistron in molecular subgroups of medulloblastoma. Expression of miR-17/92 is consistently highest in the SHH subgroup (red), followed by Group C (yellow) and WNT (blue) subgroup tumours. Group D (green) tumours are characterized by the lowest expression of the miRNA cluster. Expression values represent log2-transformed signal intensities from the normalized array data. 147

Figure 6. MYC family expression in specific medulloblastoma subgroups. A, B. Boxplots showing expression data as determined by exon array for MYCN (A) and MYC (B) in molecular subgroups of medulloblastoma. Expression values represent log2-transformed signal intensities from the normalized array data. 148

miR-17/92 expression between tumours with higher MYCN/MYC expression to tumours with lower expression to determine whether miR-17/92 regulation might also be myc-dependent in medulloblastoma. As shown in Figure 4c, components of miR-17/92 (miR-17, miR-20a, miR- 92a) and related paralogs (miR-106a, miR-20b, miR-25, miR-93) represented the majority of upregulated miRNAs in medulloblastomas with higher MYCN/MYC (n=51) expression as compared to lower expressing MYCN/MYC (n=39) tumours.

We carried out TaqMan miRNA assays to validate the correlation between MYCN/MYC and miR-17/92 expression observed on the array platforms. Samples were divided into 3 groups of 10 tumours each: higher MYCN, higher MYC, and lower MYCN/MYC and then qRT-PCR was performed for MYCN, MYC, miR-17, and miR-18. As predicted from the mRNA array data, the high MYCN (P=3.33E-08) and high MYC (P=3.33E-08) tumours did not overlap (Figure 4d, top and middle panels). Importantly, miR-17 (P=4.92E-05) and miR-18 (P=3.75E-05) were significantly upregulated in both the higher MYCN and higher MYC expressing groups as compared to the lower MYCN/MYC expressing group (Figure 4d, lower panel). These results provide strong evidence that upregulation of the miR-17/92 polycistron may be MYCN/MYC- dependent in medulloblastomas.

4.3.4 miR-17/92 is upregulated by Shh signaling in primary CGNP cultures

To determine whether the relationship between activated Shh signaling and miR-17/92 upregulation we observed in medulloblastomas reflected co-option of developmental programs, we cultured murine CGNPs with or without exogenous SHH (+/- cyclohexamide) for 24h, then performed array-based miRNA profiling. Of 599 mouse miRNAs assayed, 19 were significantly changed, 9 upregulated and 10 downregulated (Figure 7a). The miR-17/92 polycistron was upregulated in Shh-treated CGNPs, but not in the presence of cyclohexamide indicating that a new protein intermediate needs to be synthesized to regulate miR-17/92 expression.

Validation by qRT-PCR in Figure 7b, shows the six miRNAs within the miR-17/92 cluster were consistently up-regulated by Shh, which was abrogated by cyclohexamide (data not shown). These results indicate that the association between activated Shh signaling and miR-17/92 expression is conserved between normal Shh mitogenic activity in CGNPs and oncogenic Shh signaling in medulloblastoma. 149

150

Figure 7. SHH and N-myc drive the expression of miR-17/92 in cerebellar neural precursor cells resulting in mitosis. A. Table showing microRNAs whose expression was significantly altered in CGNPs by Shh treatement. The miR-17-92 cluster is upregulated, as is its paralog miR106a. B. qRT-PCR quantification to validate Shh-mediated upregulation of miR-17/92 members in CGNPs. C. Quantification of miR-17/92 expression in CGNPs transduced with retroviruses expressing N-myc, stabilized N-myc (CA), Gli1, or Gli2. N-myc, but not Gli, drives miR17-92 expression. D. Immunofluorescence staining for Ki67 showing that viral transduction of CGNPs with miR-17/92 increases Ki67 labeling of P7 cerebellar neural precursor cells, and that miR-17/92 can synergize with SHH. Graph shows results of automated quantification of Ki67 staining. 151

4.3.5 miR-17/92 induces proliferation of CGNPs downstream of N-myc

We have previously shown that N-myc is a downstream target of Shh whose induction is not protein synthesis dependent, and which can drive CGNP proliferation in the absence of Shh signaling49,295. We asked whether miR-17/92 was regulated by N-myc in CGNPs. We infected CGNPs with retroviruses carrying N-myc or the stabilized mutant N-mycT50A which can prolong CGNP proliferation in vitro50. N-myc transduction resulted in increased expression of the miR- 17/92 cluster, in the presence and absence of Shh (Figure 7c). In contrast, neither Gli1 nor Gli2 expression induced miR-17/92 in the absence of Shh; indeed, Gli1 and Gli2 suppressed Shh- mediated miR-17/92 expression. These results indicate that the Shh pathway effectors N-myc and Gli regulate different microRNA targets.

Since N-myc expression alone is sufficient to drive CGNP proliferation, we asked whether miR- 17/92 contributes to the N-myc-regulated proliferation program. We infected CGNPs with retroviruses expressing five of the six miRNAs within the miR-17/92 cluster (pWzl-miR-17- 19b)291. After 48 hours, we measured CGNP proliferation by quantifying Ki67 staining. Over- expression of the miR-17/92 cluster increased proliferation in Shh-treated cells (Figure 7d). miR-17/92 alone was able to maintain cell proliferation in the absence of Shh, albeit not at the same levels as Shh alone, suggesting that its expression does not recapitulate the complete Shh/N-myc proliferative response.

4.4 Discussion

In this Chapter we have documented the first report describing amplification of a miRNA in medulloblastoma. Recurrent amplification and widespread overexpression of miR-17/92 observed in this study strongly support a role for this miRNA cluster as a novel bona fide oncogene in medulloblastoma. Using a combination of high-resolution SNP arrays and FISH on medulloblastoma TMAs, we identified amplification or high-level gain of miR-17/92 in ~6% of primary tumours. This frequency is comparable with that observed for other well-annotated medulloblastoma oncogenes, such as MYC and MYCN that are targeted in 5-10% of cases reported in the literature. Perhaps equally impressive is the high percentage of primary medulloblastomas exhibiting aberrant expression compared to normal controls – upwards of 60% of cases analyzed showed greater than two-fold upregulation compared to normal adult cerebellum. At the time the current study was in press, a complimentary effort aimed at 152

identification of miRNAs deregulated in mouse models of medulloblastoma was reported192. Using next-gen sequencing, Uziel et al identified murine miR-17/92 as highly deregulated in medulloblastomas derived from both Ink4c-/-;Ptch+/- and Ink4c-/-;Tp53-/- mice, consistent with the results we had observed in the human malignancy.

The concept and clinical utility of classifying medulloblastomas into unique molecular subgroups was presented in detail in Chapter 3 of this thesis. When we re-analyzed our miRNA expression array data in a subgroup-specific manner in the current Chapter, the benefit of subgrouping medulloblastomas was further exemplified, as we showed the highest overexpression of miR-17/92 in the SHH subgroup, followed by Group C and WNT tumours. Importantly, similar findings were reported in the study by Uziel and colleagues, as they also showed miR-17/92 expression was preferentially upregulated in Shh-driven cases192. From a subgroup perspective, we noted that all subgroups but Group D – the only medulloblastoma subgroup not characterized by elevated MYC family expression – exhibit aberrant expression of miR-17/92. Indeed when tumours were classified into either high MYC/MYCN or low MYC/MYCN categories and miR-17/92 expression analyzed, a strong correlation between MYC/MYCN status and miR-17/92 was established. Since miR-17/92 is known to be transcriptionally-regulated by both MYC and MYCN, as previously published293,294 and further demonstrated for N-myc in this Chapter, the parallel association between MYC/MYCN status and miR-17/92 expression makes sense from a mechanistic standpoint. Collectively, these observations correlating miR-17/92 expression to its genomic status, to medulloblastoma subgroup, and to MYC family expression provide at least three molecular mechanisms accounting for the significant deregulation of miR-17/92 in medulloblastoma (Figure 8).

In summary, we have shown that high level amplification and overexpression of miR-17/92 are a hallmark of SHH-associated medulloblastomas in humans and in mice, and that miR-17/92 expression correlates with high levels of MYC family proto-oncogenes. We also show that in normally proliferating CGNPs, miR-17/92 is a Shh target whose expression is regulated by N- myc. Our finding that Shh regulates expression of an oncogenic microRNA provides additional insights as to the mechanisms through which Shh drives cell cycle progression. Our observation that miR-17/92 expression increases Shh-mediated CGNP proliferation provides insight into its role in human medulloblastoma, suggesting that high levels of miR-17/92 can provide cells with 153

Figure 8. Multiple mechanisms lead to deregulation of miR-17/92 in medulloblastoma. Cartoon depicting at least three modes of miR-17/92 upregulation in medulloblastoma. First, medulloblastomas can be discriminated into four unique molecular subgroups: WNT, SHH, Group C and Group D. miR-17/92 was found to be aberrantly expressed in three of these subgroups — SHH, Group C, and WNT — that exhibit either elevated N-Myc (SHH) or Myc (Group C, WNT) activity, and both of which transcriptionally upregulate miR-17/92. Secondly, both MYCN and MYC are genomically amplified in some medulloblastomas, and tumours harboring these amplicons express aberrant levels of miR-17/92, irrespective of medulloblastoma subgroup affiliation. Finally, we have reported amplification of miR-17/92 in medulloblastoma, providing a third mechanism leading to miR-17/92 aberrancy. 154 a selective growth advantage through an enhanced proliferative capacity. A role for miR17-92 in tumour cell survival may also be at play, as its targets identified in lymphoma include PTEN and the pro-apoptotic p53 target TP53INP1296,297. 155

Chapter 5 Conclusions & Future Directions 5 5.1 The Medulloblastoma Genome: Answered and Unanswered Questions

The primary goal of our large-scale copy number profiling study was to identify the “lowest hanging fruit” in the medulloblastoma genome – in other words, the most obvious and sensible unknown candidates targeted by CNAs that could potentially contribute to the pathogenesis of medulloblastoma. As we identified highly focal, single-gene recurrent homozygous and hemizygous deletions of EHMT1 and SMYD4, two histone lysine methyltransferases, as well as multiple additional CNAs targeting genes of related function, we decided to focus our attention on this topic. In Chapter 2, we have provided strong genetic and supporting functional evidence that deregulation of the histone code likely contributes to the pathogenesis of at least some medulloblastomas.

Although the application of high-resolution genomics to a large series of medulloblastomas in this study has proven informative and uncovered a novel family of genes targeted in this disease, this approach has generated many new questions regarding the medulloblastoma genome and still left some important queries from the past unanswered. Probably one of the most important and relevent questions in medulloblastoma genomics pertains to large areas of recurrent genomic gain and loss. For instance, the medulloblastoma community has inquisitively questioned the gene target(s) of i{17}q for literally decades. Additional anomalies such monosomy 6, chromosome 7 gain, 9q loss, 10q loss, and others have been suspected to harbor the critical oncogenes or TSGs that promote the onset of these aberrations, but most have remained ill- defined. The use of high-density SNP arrays in this study afforded us the opportunity to detect very focal CNAs including high-level amplifications and homozygous deletions that may be targeting the gene(s) driving the more notable chromosomal changes characteristic of the medulloblastoma karyotype. Indeed, several focal homozygous deletions mapping to ‘suspect’ areas in medulloblastoma were identified in this study – SMYD4 (17p13.3), MAP2K4 (17p12), L3MBTL3 (chr 6q), EHMT1 (chr 9q), and TSC1 (chr 9q) – potentially implicating these genes as the putative TSGs driving some of these larger chromosomal changes. Despite such 156

observations, the majority of these focal CNAs were rare in our dataset and few genes were targeted in more than one sample, making candidate prioritization a challenge. Ongoing and future copy number profiling of an increasingly larger cohort of primary medulloblastomas will be necessary to confirm candidate genes identified in the current study in additional samples. Moreover, given the possibility that rare homozygous deletions may pinpoint TSGs more commonly inactivated by point mutations or indels that are undetectable using the array-based technologies employed here, exon re-sequencing of candidate genes identified in this study in a large number of samples should also be considered in order to verify the frequency in which these genes are targeted genetically in medulloblastoma. Ideally, samples included in such sequencing analyses should be logically chosen (i.e. samples with LOH at a locus of interest, samples from a specific medulloblastoma subgroup, etc) in order to determine if a given candidate is more frequently mutated in a particular context and thus more likely to represent a ‘classical’ TSG acting as a ‘driver’ at these loci.

Integration of DNA copy number data with complementary genome-wide datasets including LOH, gene expression, and epigenetic (i.e. CpG methylation) profiling data will inevitably provide further insight into the identity of ‘driver’ genes involved in medulloblastoma pathogenesis. In this thesis, 103 primary medulloblastomas were profiled using both high- resolution SNP arrays and exon arrays, enabling preliminary integrative analyses as evidenced in our genomic characterization of medulloblastoma subgroups presented in Chapter 3. Future bioinformatic analyses focused on a gene-level integration of these datasets will be necessary to identify copy number-driven gene expression events in medulloblastoma and potentially pinpoint the gene(s) driving recurrent regions of both large and focal CNAs.

5.2 Characterization of Medulloblastoma Subgroups and the Continued Search for the “Medulloblast”

In Chapters 3 and 4 of this thesis, the concept and utility of medulloblastoma subgrouping was presented. Through application of gene expression profiling to the largest cohort of primary medulloblastoma samples published to date, we have described the existence of four molecularly and clinically distinct entities of medulloblastoma we have designated WNT, SHH, Group C, and Group D. Using multiple independent unsupervised class discovery approaches, we consistently produced four unique sample clusters (i.e. subgroups) in our dataset. This result is in contrast to 157

the five-six subgroups reported on smaller datasets in the previous literature37,38. The main difference between our four-subgroup classification scheme compared to the five subgroups reported by Kool et al in 2008, is the description of two non-Shh/Wnt-driven subgroups identified in our study versus the three such groups noted in the Kool publication. Although our data strongly suggests that there is some extent of intra-subgroup heterogeneity (i.e. three clusters of SHH tumours described in Chapter 3), the statistical support for further division of the four main subgroups was poor in our dataset (as it was in Kool’s study), and the utility of imposing additional divisions of the current four subgroups remains to be seen. Indeed extension of our four-subgroup classification system to a large series of nearly 300 non-overlapping medulloblastomas present on two independent TMAs showed robust performance, efficiently affiliating ~98% of samples to a single subgroup. More importantly, our system demonstrated significant prognostic capability, showing that Group C medulloblastoma patients have the worst prognosis.

Now that we have described the existence of four distinct medulloblastoma subgroups exhibiting distinct genetics, transcriptional profiles, and clinical characteristics, a major question that has emerged is “do these four subgroups represent four distinct diseases originating from four distinct cell types?” If so, which cell(s) give rise to the different subgroups? Although these are important questions from a developmental biology perspective, it is of equal importance that these questions are answered for diagnostic and therapeutic reasons. If indeed these four subgroups represent different diseases arising in different cell types of the cerebellum, it will be even more important to incorporate medulloblastoma subgroup affiliation into patient stratification and treatment protocols in the future, appropriatiely tailoring therapies to each of the medulloblastoma subgroups individually.

There are many lines of published evidence that have convincingly established CGNPs as the likely cell of origin for Shh-driven (i.e. SHH subgroup) medulloblastomas. In addition, it has become increasingly evident that WNT subgroup tumours in all likelihood do not come from the EGL and do not overlap with the SHH subgroup. At present, much less is known of where (i.e. cerebellar compartment or cell type) Group C and Group D medulloblastomas may originate. Using array-based transcriptional profiling, we have identified reliable molecular markers of the four medulloblastoma subgroups that can be extrapolated to reliably classify paraffin-embedded medulloblastoma samples. It is conceivable that this approach may be equally applicable to 158

strategies aimed at the identification of the cell or cell type(s) from which the different subgroups are originating. Future immunohistochemical studies of the developing cerebellum using the subgroup-specific markers described in Chapter 3 (or additional subgroup-specific markers not yet tested) may prove informative in the search for the cellular origin of medulloblastoma subgroups.

As described in detail in the introductory Chapter of this thesis, the vast majority of published mouse models of medulloblastoma recapitulate the SHH subgroup of the disease and models of the other subgroups are currently poorly represented. We and others have described genetic and genomic events that appear to be exclusive to a given medulloblastoma subgroup – CTNNB1 mutations are found only in WNT tumours, miR-17/92 is amplified in SHH tumours, i{17}q is restricted to Group C and Group D, etc. If appropriate cell type markers can be identified for each of the different medulloblastoma subgroups as mentioned above, rationally designed mouse models harboring subgroup-relevant mutations may be generated. By using a suitable promoter to drive transgene expression or Cre recombinase, subgroup-specific genetic/genomic events observed in the human tumours may be evaluated in a systematic manner. Furthermore, functional genomic screens such as the Sleeping Beauty transposon system298,299 may be performed for each of the subgroups. By appropriately modeling each of the four medulloblastoma subgroups in the mouse, we will gain a better understanding of the biology underlying these subgroups and ideally have a system available for testing novel therapeutics specific for each of the individual subgroups. Patients of different medulloblastoma subgroups exhibit disparate clinical outcomes and therefore should be stratified and treated differently. In order for such “personalized” treatments to become available for patients in the future, appropriate models that closely ressemble the human disease must be generated so novel therapeutics may be evaluated.

5.3 A Role for Small RNAs in Medulloblastoma Development

In Chapter 4 of this thesis we implicated miR-17/92 as a novel medulloblastoma oncogene – the first report of a miRNA acting as an oncogene in this malignancy. We showed that miR-17/92 is recurrently amplified in medulloblastoma and overexpressed in the majority of primary tumours compared to non-neoplastic cerebellum187. In addition, both Martine Roussel’s group and our own established that miR-17/92 is most highly upregulated in SHH-driven 159

medulloblastomas187,192, and we further showed that miR-17/92 expression is regulated by active Shh signaling through N-Myc in CGNPs. Despite these important findings, several questions pertaining to the role of miR-17/92 in both normal cerebellar development and medulloblastoma remain to be elucidated.

Since miRNAs function as post-transcriptional modulators of the cell’s gene expression program, it is of keen interest to understand the extent of mRNAs targeted by a given miRNA in order to appreciate the impact of the miRNA on various cellular processes. Moreover, by confirming the mRNA targets of a given miRNA, the consequences of deregulated miRNA expression can be put into context. To date, several targets of the oncogenic miR-17/92 polycistron have been validated, including E2F1, E2F2, and E2F3 – activating E2Fs that promote the transition from G1 to S phase of the cell cycle293,300,301. Although E2F inhibition is somewhat paradoxical if miR-17/92 is functioning as oncogene, it is believed that repression of E2Fs via miR-17/92 results in attenuation of E2F-mediated apoptosis, as elevated levels of E2F proteins (especially E2F1) can induce apoptosis. CDKN1A (p21), a negative regulator of the G1-S checkpoint, has also been confirmed as a target of miR-17 and related miRNAs (miR-20a, miR-106b, and miR- 93), providing another mechanism by which these miRNAs may deregulate the cell cycle302,303.

Additional genes regulating cell death including the proapoptotic gene BCL2L11/BIM and negative regulator of the PI 3-kinase pathway PTEN have been identified as miR-17/92 targets296,303-305. Furthermore, anti-angiogenic factors such as TSP1 and CTGF are also known to be inhibited by miR-17/92306. Collectively, these observations have implicated miR-17/92 as a modulator of cell proliferation, cell death, and angiogenesis, underscoring its potential contribution to multiple aspects of tumourigenesis285.

Currently there are no confirmed targets of miR-17/92 in the context of medulloblastoma187,192. Indeed, Uziel et al examined PTEN status in the context of ectopic miR-17/92 and found no change in PTEN protein levels192. In order to better comprehend the oncogenic effect(s) of aberrant miR-17/92 expression in medulloblastoma, strategies aimed at identifying its mRNA targets in this tumour are warranted. As a confirmed target of Shh signaling, it is conceivable that miR-17/92 post-transcriptionally represses genes negatively regulating the Shh pathway. Using over-expression and/or knockdown approaches combined with array-based expression 160 profiling, bona fide targets of miR-17/92 in medulloblastoma may be uncovered, leading to a more complete understanding of miR-17/92 function in medulloblastoma pathogenesis.

Although the work of Uziel et al and our own has strongly implicated miR-17/92 as a novel medulloblastoma oncogene acting downstream of activated Shh signaling, future strategies investigating miR-17/92 function through the use of genetically engineered mouse models will shed further light on its role in cerebellar development and medulloblastoma. For instance, the generation of a transgenic mouse model overexpressing the miR-17/92 polycistron in CGNPs of the developing cerebellum would be an ideal approach to further characterize its role in both cerebellar development and medulloblastoma. As a confirmed target of the Shh pathway, targeted miR-17/92 expression in CGNPs (i.e. Math1-expressing cells) where Shh signaling is robust would be intuitive. Similarly, targeted deletion of miR-17/92 in CGNPs using Cre/lox technology would likely prove very informative from a developmental perspective. Mice null for miR-17/92 succumb almost immediately after birth as a result of severe lung and heart defects305. CGNP-specific disruption of miR-17/92 would allow for a comprehensive assessement of miR-17/92’s contribution to the development of the cerebellum.

In our characterization of the medulloblastoma miRNAome described in Chapter 4, miR-17/92 was among the most obvious candidates that emerged – identified as recurrently amplified in primary tumours and exhibiting widespread upregulation in medulloblastomas compared to normal cerebellar controls – thus making this miRNA cluster a logical choice for follow-up studies. Aside from miR-17/92, our array-based profiling of miRNAs identified several additional candidate miRNAs and miRNA clusters as dysregulated in medulloblastoma. For instance, miR-128a and miR-128b were identified as among the most frequently downregulated miRNAs in medulloblastoma when compared to normal cerebellum. Recently, Godlewski et al described consistent down-regulation of miR-128 in glioblastoma and confirmed the BMI1 proto-oncogene as a target for repression by miR-128307. From a medulloblastoma subgroup perspective, the miR-452-224 cluster was found to be highly upregulated in the WNT subgroup of medulloblastomas. Mir-224 has been reported as aberrantly expressed in hepatocellular carcinomas308 and miR-452 over-expression has been noted in urothelial carcinomas309. The significance of the differential expression observed for these and other miRNAs in medulloblastoma as reported in Chapter 4 will require further investigation in future studies. Given the extensive role miRNAs have been shown to play in cancer in recent years, it is highly 161

probably that additional candidates beyond miR-17/92 such as those alluded to above also contribute to medulloblastoma genesis.

5.4 The Future of Medulloblastoma Genomics

Over the past few years, microarray technologies have significantly increased our understanding of the medulloblastoma genome, transcriptome, and to some extent, epigenome. Moving forward, array platforms will undoubtedly continue to be used in genome-wide profiling efforts of medulloblastoma, especially as the resolution and coverage of these methods continues to improve, and the cost of these screens remains affordable. However, recent breakthroughs in DNA sequencing technologies have had a profound impact on the genomics community, and their utility in medulloblastoma research is without question imminent.

Since the early 1990’s, the capillary-based Sanger method of DNA sequencing has been the mainstay for most applications in molecular biology, including the first drafts of the human genome published in 2000310,311. More recently, conventional sequencing has been successfully used in large-scale re-sequencing efforts, profiling anywhere from a few hundred genes to all known protein-coding genes in a single cancer genome90-92,209,312,313. Indeed, initial exon re- sequencing of the colorectal, breast, pancreatic, and glioblastoma multiforme genomes has revealed new genes and pathways involved in the pathogenesis of the respective cancer types90- 92,312. However, these studies relied on PCR-mediated amplification of literally hundreds of thousands of exons combined with an enormous workload of conventional sequencing, unrealistic tasks for the vast majority of the cancer genomics community.

Fortunately, a revolution in DNA sequencing technology has occurred over the last few years that is rapidly changing the field of cancer genomics314-317. Next-gen (aka “deep”) sequencing biochemistries now permit the parallel acquisition of up to tens of gigabases (Gb) of DNA sequence of variable “read” length in a single experiment. There are multiple next-gen sequencing options currently available – 454 (Roche), Solexa (Illumina), SOLiD (ABI)316,318 – each with its own strengths and weaknesses. The repertoire and capabilities of these platforms is continually improving. For example, the SOLiD 3 System, currently boasts >20 Gb of DNA sequence per run, compared to the ~750-1,000 bp generated using traditional Sanger sequencing. Although the cost/base ratio is significantly lower for next-gen technologies, the current cost of a single SOLiD 3 run is in the neighborhood of $15,000 (versus ~$5 for conventional sequencing), 162

making next-gen prohibitively expensive for a large percentage of the research community, especially when considering sequencing large numbers of patient samples.

Multiple “proof-of-principle” studies using next-gen technology have now been applied to the study of various aspects of both normal and cancer genomes. These published reports include array-based, targeted capture and next-gen sequencing of ~200,000 protein-coding exons in the human genome, allowing the specific identification of both common and rare sequence variants319,320. Unbiased, whole-genome next-gen sequencing has also been reported for multiple normal human genomes321-324, and recently, the acute myeloid leukemia (AML) genome from a single individual325. Producing nearly 100 billion bases of sequence, the AML study reported ten genes with acquired somatic mutations that were not present in the patient-matched genome from normal skin cells.

Perhaps equally impressive, next-gen sequencing has not been limited to studies aimed at identification of sequence variants and mutations318,326. Structural aberrations including both inter- and intrachromosomal rearrangements (i.e. inversions, inverted/tandem duplications, translocations) and CNAs (i.e. amplifications and deletions) have been identified in human cancer cells using next-gen sequencing, with improved specificity and sensitivity compared to array-based methods327,328. Whole-transcriptome (aka “RNA-Seq”) profiling has also been described using next-gen approaches, permitting quantification of transcript abundance (mRNA, miRNA, etc) and identification of novel genes and isoforms in an unbiased manner329-334. In contrast to array-based technologies used in gene expression analyses, RNA-Seq requires no a priori knowledge of the transcriptome under investigation, thus enabling full-transcriptome characterization and eliminates biases associated with array content. Similarly, extension of unbiased next-gen sequencing to studies of the mammalian epigenome have also been undertaken, including genome-wide reports of DNA methylomes335,336, mapping of histone modifications337,338, and detailing the locations of DNA-binding proteins339,340.

Collectively, these emerging next-gen sequencing-based approaches for studying the cancer genome hold great promise for comprehensive analyses of medulloblastoma. As the cost and bioinformatics involved in next-gen become more mainstream, next-gen-based profiling of the medulloblastoma genome, transcriptome, and epigenome will surely be the priority of several groups (Figure 1). Undoubtedly, these efforts will lead to a more complete understanding of the 163 genes and pathways involved in the initiation, maintenance, and progression of medulloblastoma. Additionally, as larger patient cohorts are collected and profiled using these advanced methods, more specific and reliable molecular classification of medulloblastoma will likely be made possible. Finally, correlation of genomic data with patient clinical data such as presence of metastatic disease and overall survival will undoubtedly be improved.

Twenty years of surveying the medulloblastoma genome has facilitated a detailed description of the medulloblastoma karyotype, led to the identification and validation of bona fide oncogenes and tumour suppressors, and implicated key signaling pathways and networks that are recurrently deregulated. Much of the recent progress made in this field owes to improvements in the technologies available for interrogation of the genome. In order to attain comprehensive appreciation of a cancer genome such as medulloblastoma, unbiased, high-resolution genome- wide interrogation must be undertaken, ideally using a combination of both complementary (i.e. microarrays and next-gen sequencing) and integrative (i.e. genome, transcriptome, and epigenome) technologies. In addition, an adequate sample size is critical to such studies if a full range of both common and rare genomic changes is to be captured. Large, coordinated multi- institutional consortiums such as The Cancer Genome Atlas (TCGA) are now applying this philosophy to study the genomes of brain (glioblastoma multiforme), lung (squamous carcinoma), and ovarian (serous cystadenocarcinoma) cancers209. As part of their pilot project, the TCGA is profiling large numbers of tumours of these origins to assess aberrations in DNA sequence (i.e. substitutions, indels), genomic copy number (i.e. amplifications, deletions), chromosomal composition (i.e. rearrangements), DNA methylation (i.e. hyper- and hypo- promoter methylation), and gene expression (i.e. aberrant expression, splicing). Results from the TCGA and other similar large-scale collaborative efforts using this type of broad approach to cancer genomics have recently proven to be a success90,91,209,313. In order to make future progress in the fine genomic mapping of medulloblastoma, analogous strategies are warranted. Large- scale, collaborative genomics projects will provide a more detailed characterization of this genome than ever before, and optimistically, many new candidates will be uncovered. These efforts should continue to improve our ability to diagnose, stratify, and treat medulloblastoma, eventually leading to decreased mortality and improved quality of life for our patients. 164

165

Figure 1. The future of medulloblastoma genomics. Cartoon illustrating how collaborative approaches that involve integration of genomics, transcriptome analysis, and epigenomics will be applied in the future characterization of medulloblastoma. For each category, some of the leading technologies of the present and future are listed, with applications relying on next-gen sequencing present in each. The union of these orthogonal genome/transcriptome/epigenome interrogation strategies will greatly accelerate the identification of novel candidate genes and pathways involved in medulloblastoma genesis, and permit follow-up functional validation studies, correlation with clinical variables, and improved molecular classification. 166

References

1. McNeil DE, Cote TR, Clegg L, et al: Incidence and trends in pediatric malignancies medulloblastoma/primitive neuroectodermal tumor: a SEER update. Surveillance Epidemiology and End Results. Med Pediatr Oncol 39:190-4, 2002

2. Gajjar A, Hernan R, Kocak M, et al: Clinical, histopathologic, and molecular markers of prognosis: toward a new disease risk stratification system for medulloblastoma. J Clin Oncol 22:984-93, 2004

3. Finlay JL, Erdreich-Epstein A, Packer RJ: Progress in the treatment of childhood brain tumors: no room for complacency. Pediatr Hematol Oncol 24:79-84, 2007

4. Ellison DW, Clifford SC, Gajjar A, et al: What's new in neuro-oncology? Recent advances in medulloblastoma. Eur J Paediatr Neurol 7:53-66, 2003

5. Crawford JR, MacDonald TJ, Packer RJ: Medulloblastoma in childhood: new biological advances. Lancet Neurol 6:1073-85, 2007

6. Packer RJ, Vezina G: Management of and prognosis with medulloblastoma: therapy at a crossroads. Arch Neurol 65:1419-24, 2008

7. Tabori U, Sung L, Hukin J, et al: Distinctive clinical course and pattern of relapse in adolescents with medulloblastoma. Int J Radiat Oncol Biol Phys 64:402-7, 2006

8. Curran EK, Sainani KL, Le GM, et al: Gender affects survival for medulloblastoma only in older children and adults: a study from the Surveillance Epidemiology and End Results Registry. Pediatr Blood Cancer 52:60-4, 2009

9. Gajjar A, Chintagumpala M, Ashley D, et al: Risk-adapted craniospinal radiotherapy followed by high-dose chemotherapy and stem-cell rescue in children with newly diagnosed medulloblastoma (St Jude Medulloblastoma-96): long-term results from a prospective, multicentre trial. Lancet Oncol 7:813-20, 2006

10. Ellison DW, Onilude OE, Lindsey JC, et al: beta-Catenin status predicts a favorable outcome in childhood medulloblastoma: the United Kingdom Children's Cancer Study Group Brain Tumour Committee. J Clin Oncol 23:7951-7, 2005

11. Louis DN, Ohgaki H, Wiestler OD, et al: The 2007 WHO classification of tumours of the central nervous system. Acta Neuropathol 114:97-109, 2007

12. Gilbertson RJ, Ellison DW: The origins of medulloblastoma subtypes. Annu Rev Pathol 3:341-65, 2008

13. Ellison D: Classifying the medulloblastoma: insights from morphology and molecular genetics. Neuropathol Appl Neurobiol 28:257-82, 2002 167

14. Eberhart CG, Kepner JL, Goldthwaite PT, et al: Histopathologic grading of medulloblastomas: a Pediatric Oncology Group study. Cancer 94:552-60, 2002

15. Garre ML, Cama A, Bagnasco F, et al: Medulloblastoma variants: age-dependent occurrence and relation to Gorlin syndrome--a new clinical perspective. Clin Cancer Res 15:2463-71, 2009

16. Grundy RG, Wilne SH, Robinson KJ, et al: Primary postoperative chemotherapy without radiotherapy for treatment of brain tumours other than ependymoma in children under 3 years: results of the first UKCCSG/SIOP CNS 9204 trial. Eur J Cancer 46:120-33

17. Marino S: Medulloblastoma: developmental mechanisms out of control. Trends Mol Med 11:17-22, 2005

18. Eberhart CG: In search of the medulloblast: neural stem cells and embryonal brain tumors. Neurosurg Clin N Am 18:59-69, viii-ix, 2007

19. Behesti H, Marino S: Cerebellar granule cells: insights into proliferation, differentiation, and role in medulloblastoma pathogenesis. Int J Biochem Cell Biol 41:435-45, 2009

20. Gailani MR, Bale SJ, Leffell DJ, et al: Developmental defects in Gorlin syndrome related to a putative tumor suppressor gene on chromosome 9. Cell 69:111-7, 1992

21. Hahn H, Wicking C, Zaphiropoulous PG, et al: Mutations of the human homolog of Drosophila patched in the nevoid basal cell carcinoma syndrome. Cell 85:841-51, 1996

22. Taylor MD, Liu L, Raffel C, et al: Mutations in SUFU predispose to medulloblastoma. Nat Genet 31:306-10, 2002

23. Yauch RL, Dijkgraaf GJ, Alicke B, et al: Smoothened mutation confers resistance to a Hedgehog pathway inhibitor in medulloblastoma. Science 326:572-4, 2009

24. Lam CW, Xie J, To KF, et al: A frequent activated smoothened mutation in sporadic basal cell carcinomas. Oncogene 18:833-6, 1999

25. Goodrich LV, Milenkovic L, Higgins KM, et al: Altered neural cell fates and medulloblastoma in mouse patched mutants. Science 277:1109-13, 1997

26. Wetmore C, Eberhart DE, Curran T: Loss of p53 but not ARF accelerates medulloblastoma in mice heterozygous for patched. Cancer Res 61:513-6, 2001

27. Oliver TG, Read TA, Kessler JD, et al: Loss of patched and disruption of granule cell development in a pre-neoplastic stage of medulloblastoma. Development 132:2425-39, 2005

28. Yang ZJ, Ellis T, Markant SL, et al: Medulloblastoma can be initiated by deletion of Patched in lineage-restricted progenitors or stem cells. Cancer Cell 14:135-45, 2008

29. Lee Y, Kawagoe R, Sasai K, et al: Loss of suppressor-of-fused function promotes tumorigenesis. Oncogene 26:6442-7, 2007 168

30. Hallahan AR, Pritchard JI, Hansen S, et al: The SmoA1 mouse model reveals that notch signaling is critical for the growth and survival of sonic hedgehog-induced medulloblastomas. Cancer Res 64:7794-800, 2004

31. Hatton BA, Villavicencio EH, Tsuchiya KD, et al: The Smo/Smo model: hedgehog-induced medulloblastoma with 90% incidence and leptomeningeal spread. Cancer Res 68:1768-76, 2008

32. Attard TM, Giglio P, Koppula S, et al: Brain tumors in individuals with familial adenomatous polyposis: a cancer registry experience and pooled case report analysis. Cancer 109:761-6, 2007

33. Hamilton SR, Liu B, Parsons RE, et al: The molecular basis of Turcot's syndrome. N Engl J Med 332:839-47, 1995

34. McMahon AP, Bradley A: The Wnt-1 (int-1) proto-oncogene is required for development of a large region of the mouse brain. Cell 62:1073-85, 1990

35. Thomas KR, Capecchi MR: Targeted disruption of the murine int-1 proto- oncogene resulting in severe abnormalities in midbrain and cerebellar development. Nature 346:847-50, 1990

36. Schuller U, Rowitch DH: Beta-catenin function is required for cerebellar morphogenesis. Brain Res 1140:161-9, 2007

37. Thompson MC, Fuller C, Hogg TL, et al: Genomics identifies medulloblastoma subgroups that are enriched for specific genetic alterations. J Clin Oncol 24:1924-31, 2006

38. Kool M, Koster J, Bunt J, et al: Integrated genomics identifies five medulloblastoma subtypes with distinct genetic profiles, pathway signatures and clinicopathological features. PLoS One 3:e3088, 2008

39. Pomeroy SL, Tamayo, P., Gaasenbeek, M., et al.: Prediction of central nervous system embryonal tumour outcome based on gene expression. Nature 415:436-442, 2002

40. Fan X, Mikolaenko I, Elhassan I, et al: Notch1 and notch2 have opposite effects on embryonal brain tumor growth. Cancer Res 64:7787-93, 2004

41. Li Y, Lal B, Kwon S, et al: The scatter factor/hepatocyte growth factor: c-met pathway in human embryonal central nervous system tumor malignancy. Cancer Res 65:9355- 62, 2005

42. Binning MJ, Niazi T, Pedone CA, et al: Hepatocyte growth factor and sonic Hedgehog expression in cerebellar neural progenitor cells costimulate medulloblastoma initiation and growth. Cancer Res 68:7838-45, 2008

43. Kongkham PN, Northcott PA, Ra YS, et al: An epigenetic genome-wide screen identifies SPINT2 as a novel tumor suppressor gene in pediatric medulloblastoma. Cancer Res 68:9945-53, 2008 169

44. Kohl NE, Kanda N, Schreck RR, et al: Transposition and amplification of oncogene-related sequences in human neuroblastomas. Cell 35:359-67, 1983

45. Schwab M, Alitalo K, Klempnauer KH, et al: Amplified DNA with limited homology to myc cellular oncogene is shared by human neuroblastoma cell lines and a neuroblastoma tumour. Nature 305:245-8, 1983

46. Schwab M: MYCN in neuronal tumours. Cancer Lett 204:179-87, 2004

47. Knoepfler PS, Cheng PF, Eisenman RN: N-myc is essential during neurogenesis for the rapid expansion of progenitor cell populations and the inhibition of neuronal differentiation. Genes Dev 16:2699-712, 2002

48. Hatton BA, Knoepfler PS, Kenney AM, et al: N-myc is an essential downstream effector of Shh signaling during both normal and neoplastic cerebellar growth. Cancer Res 66:8655-61, 2006

49. Kenney AM, Cole MD, Rowitch DH: Nmyc upregulation by sonic hedgehog signaling promotes proliferation in developing cerebellar granule neuron precursors. Development 130:15-28, 2003

50. Kenney AM, Widlund HR, Rowitch DH: Hedgehog and PI-3 kinase signaling converge on Nmyc1 to promote cell cycle progression in cerebellar neuronal precursors. Development 131:217-28, 2004

51. Zindy F, Uziel T, Ayrault O, et al: Genetic alterations in mouse medulloblastomas and generation of tumors de novo from primary cerebellar granule neuron precursors. Cancer Res 67:2676-84, 2007

52. Kessler JD, Hasegawa H, Brun SN, et al: N-myc alters the fate of preneoplastic cells in a mouse model of medulloblastoma. Genes Dev 23:157-70, 2009

53. Haupt Y, Bath ML, Harris AW, et al: bmi-1 transgene induces lymphomas and collaborates with myc in tumorigenesis. Oncogene 8:3161-4, 1993

54. Leung C, Lingbeek M, Shakhova O, et al: Bmi1 is essential for cerebellar development and is overexpressed in human medulloblastomas. Nature 428:337-41, 2004

55. van der Lugt NM, Domen J, Linders K, et al: Posterior transformation, neurological abnormalities, and severe hematopoietic defects in mice with a targeted deletion of the bmi-1 proto-oncogene. Genes Dev 8:757-69, 1994

56. Shakhova O, Leung C, Marino S: Bmi1 in development and tumorigenesis of the central nervous system. J Mol Med 83:596-600, 2005

57. Lindsey JC, Anderton JA, Lusher ME, et al: Epigenetic events in medulloblastoma development. Neurosurg Focus 19:E10, 2005 170

58. Chen WY, Zeng X, Carter MG, et al: Heterozygous disruption of Hic1 predisposes mice to a gender-dependent spectrum of malignant tumors. Nat Genet 33:197-202, 2003

59. Briggs KJ, Corcoran-Schwartz IM, Zhang W, et al: Cooperation between the Hic1 and Ptch1 tumor suppressors in medulloblastoma. Genes Dev 22:770-85, 2008

60. Ben-Arie N, Bellen HJ, Armstrong DL, et al: Math1 is essential for genesis of cerebellar granule neurons. Nature 390:169-72, 1997

61. Jensen P, Smeyne R, Goldowitz D: Analysis of cerebellar development in math1 null embryos and chimeras. J Neurosci 24:2202-11, 2004

62. Zhao H, Ayrault O, Zindy F, et al: Post-transcriptional down-regulation of Atoh1/Math1 by bone morphogenic proteins suppresses medulloblastoma development. Genes Dev 22:722-7, 2008

63. Angley C, Kumar M, Dinsio KJ, et al: Signaling by bone morphogenetic proteins and Smad1 modulates the postnatal differentiation of cerebellar cells. J Neurosci 23:260-8, 2003

64. Rios I, Alvarez-Rodriguez R, Marti E, et al: Bmp2 antagonizes sonic hedgehog- mediated proliferation of cerebellar granule neurones through Smad5 signalling. Development 131:3159-68, 2004

65. Briggs KJ, Eberhart CG, Watkins DN: Just say no to ATOH: how HIC1 methylation might predispose medulloblastoma to lineage addiction. Cancer Res 68:8654-6, 2008

66. Grimmer MR, Weiss WA: BMPs oppose Math1 in cerebellar development and in medulloblastoma. Genes Dev 22:693-9, 2008

67. Flora A, Klisch TJ, Schuster G, et al: Deletion of Atoh1 disrupts Sonic Hedgehog signaling in the developing cerebellum and prevents medulloblastoma. Science 326:1424-7, 2009

68. Levine AJ: p53, the cellular gatekeeper for growth and division. Cell 88:323-31, 1997

69. Malkin D, Li FP, Strong LC, et al: Germ line p53 mutations in a familial syndrome of breast cancer, sarcomas, and other neoplasms. Science 250:1233-8, 1990

70. Marino S, Vooijs M, van Der Gulden H, et al: Induction of medulloblastomas in p53-null mutant mice by somatic inactivation of Rb in the external granular layer cells of the cerebellum. Genes Dev 14:994-1004, 2000

71. Shakhova O, Leung C, van Montfort E, et al: Lack of Rb and p53 delays cerebellar development and predisposes to large cell anaplastic medulloblastoma through amplification of N-Myc and Ptch2. Cancer Res 66:5190-200, 2006

72. Zindy F, Nilsson LM, Nguyen L, et al: Hemangiosarcomas, medulloblastomas, and other tumors in Ink4c/p53-null mice. Cancer Res 63:5420-7, 2003 171

73. Ayrault O, Zindy F, Rehg J, et al: Two tumor suppressors, p27Kip1 and patched- 1, collaborate to prevent medulloblastoma. Mol Cancer Res 7:33-40, 2009

74. Uziel T, Zindy F, Xie S, et al: The tumor suppressors Ink4c and p53 collaborate independently with Patched to suppress medulloblastoma formation. Genes Dev 19:2656-67, 2005

75. Fruhwald MC, O'Dorisio MS, Dai Z, et al: Aberrant promoter methylation of previously unidentified target genes is a common abnormality in medulloblastomas--implications for tumor biology and potential clinical utility. Oncogene 20:5033-42, 2001

76. Frank AJ, Hernan R, Hollander A, et al: The TP53-ARF tumor suppressor pathway is frequently disrupted in large/cell anaplastic medulloblastoma. Brain Res Mol Brain Res 121:137-40, 2004

77. Ebinger M, Senf L, Wachowski O, et al: Promoter methylation pattern of caspase- 8, P16INK4A, MGMT, TIMP-3, and E-cadherin in medulloblastoma. Pathol Oncol Res 10:17- 21, 2004

78. Frappart PO, Lee Y, Russell HR, et al: Recurrent genomic alterations characterize medulloblastoma arising from DNA double-strand break repair deficiency. Proc Natl Acad Sci U S A 106:1880-5, 2009

79. Lee Y, McKinnon PJ: DNA ligase IV suppresses medulloblastoma formation. Cancer Res 62:6395-9, 2002

80. Frappart PO, Lee Y, Lamont J, et al: BRCA2 is required for neurogenesis and suppression of medulloblastoma. EMBO J 26:2732-42, 2007

81. Yan CT, Kaushal D, Murphy M, et al: XRCC4 suppresses medulloblastomas with recurrent translocations in p53-deficient mice. Proc Natl Acad Sci U S A 103:7378-83, 2006

82. Tong WM, Ohgaki H, Huang H, et al: Null mutation of DNA strand break- binding molecule poly(ADP-ribose) polymerase causes medulloblastomas in p53(-/-) mice. Am J Pathol 162:343-52, 2003

83. Hirsch B, Shimamura A, Moreau L, et al: Association of biallelic BRCA2/FANCD1 mutations with spontaneous chromosomal instability and solid tumors of childhood. Blood 103:2554-9, 2004

84. Dewire MD, Ellison DW, Patay Z, et al: Fanconi anemia and biallelic BRCA2 mutation diagnosed in a young child with an embryonal CNS tumor. Pediatr Blood Cancer 53:1140-2, 2009

85. Marrett LD, De P, Airia P, et al: Cancer in Canada in 2008. CMAJ 179:1163-70, 2008

86. Jemal A, Siegel R, Ward E, et al: Cancer statistics, 2008. CA Cancer J Clin 58:71- 96, 2008 172

87. Finishing the euchromatic sequence of the human genome. Nature 431:931-45, 2004

88. Pruitt KD, Tatusova T, Maglott DR: NCBI reference sequences (RefSeq): a curated non-redundant sequence database of genomes, transcripts and proteins. Nucleic Acids Res 35:D61-5, 2007

89. Stratton MR, Campbell PJ, Futreal PA: The cancer genome. Nature 458:719-24, 2009

90. Jones S, Zhang X, Parsons DW, et al: Core signaling pathways in human pancreatic cancers revealed by global genomic analyses. Science 321:1801-6, 2008

91. Parsons DW, Jones S, Zhang X, et al: An integrated genomic analysis of human glioblastoma multiforme. Science 321:1807-12, 2008

92. Wood LD, Parsons DW, Jones S, et al: The genomic landscapes of human breast and colorectal cancers. Science 318:1108-13, 2007

93. Brena RM, Costello JF: Genome-epigenome interactions in cancer. Hum Mol Genet 16 Spec No 1:R96-105, 2007

94. Ting AH, McGarvey KM, Baylin SB: The cancer epigenome--components and functional correlates. Genes Dev 20:3215-31, 2006

95. Jones PA, Baylin SB: The fundamental role of epigenetic events in cancer. Nat Rev Genet 3:415-28, 2002

96. Jones PA, Baylin SB: The epigenomics of cancer. Cell 128:683-92, 2007

97. Taylor MD, Mainprize TG, Rutka JT: Molecular insight into medulloblastoma and central nervous system primitive neuroectodermal tumor biology from hereditary syndromes: a review. Neurosurgery 47:888-901, 2000

98. Wechsler-Reya R, Scott MP: The developmental biology of brain tumors. Annu Rev Neurosci 24:385-428, 2001

99. Grimmer MR, Weiss WA: Childhood tumors of the nervous system as disorders of normal development. Curr Opin Pediatr 18:634-8, 2006

100. Srivastava S, Zou ZQ, Pirollo K, et al: Germ-line transmission of a mutated p53 gene in a cancer-prone family with Li-Fraumeni syndrome. Nature 348:747-9, 1990

101. Raffel C: Medulloblastoma: molecular genetics and animal models. Neoplasia 6:310-22, 2004

102. Piedimonte LR, Wailes IK, Weiner HL: Medulloblastoma: mouse models and novel targeted therapies based on the Sonic hedgehog pathway. Neurosurg Focus 19:E8, 2005 173

103. Fults DW: Modeling medulloblastoma with genetically engineered mice. Neurosurg Focus 19:E7, 2005

104. Speicher MR, Carter NP: The new cytogenetics: blurring the boundaries with molecular biology. Nat Rev Genet 6:782-92, 2005

105. Bayani J, Squire JA: Traditional banding of chromosomes for cytogenetic analysis. Curr Protoc Cell Biol Chapter 22:Unit 22 3, 2004

106. Bigner SH, Mark J, Friedman HS, et al: Structural chromosomal abnormalities in human medulloblastoma. Cancer Genet Cytogenet 30:91-101, 1988

107. Griffin CA, Hawkins AL, Packer RJ, et al: Chromosome abnormalities in pediatric brain tumors. Cancer Res 48:175-80, 1988

108. Biegel JA, Rorke LB, Packer RJ, et al: Isochromosome 17q in primitive neuroectodermal tumors of the central nervous system. Genes Chromosomes Cancer 1:139-47, 1989

109. Mertens F, Johansson B, Mitelman F: Isochromosomes in neoplasia. Genes Chromosomes Cancer 10:221-30, 1994

110. Raffel C, Gilles FE, Weinberg KI: Reduction to homozygosity and gene amplification in central nervous system primitive neuroectodermal tumors of childhood. Cancer Res 50:587-91, 1990

111. James CD, He J, Carlbom E, et al: Loss of genetic information in central nervous system tumors common to children and young adults. Genes Chromosomes Cancer 2:94-102, 1990

112. Cogen PH, Daneshvar L, Metzger AK, et al: Involvement of multiple chromosome 17p loci in medulloblastoma tumorigenesis. Am J Hum Genet 50:584-9, 1992

113. Cogen PH, Daneshvar L, Metzger AK, et al: Deletion mapping of the medulloblastoma locus on chromosome 17p. Genomics 8:279-85, 1990

114. Thomas GA, Raffel C: Loss of heterozygosity on 6q, 16q, and 17p in human central nervous system primitive neuroectodermal tumors. Cancer Res 51:639-43, 1991

115. Northcott PA, Nakahara Y, Wu X, et al: Multiple recurrent genetic events converge on control of histone lysine methylation in medulloblastoma. Nat Genet 41:465-72, 2009

116. Rickert CH, Paulus W: Comparative genomic hybridization in central and peripheral nervous system tumors of childhood and adolescence. J Neuropathol Exp Neurol 63:399-417, 2004

117. Pfister S, Remke M, Benner A, et al: Outcome prediction in pediatric medulloblastoma based on DNA copy-number aberrations of chromosomes 6q and 17q and the MYC and MYCN loci. J Clin Oncol 27:1627-36, 2009 174

118. Friedman HS, Burger PC, Bigner SH, et al: Phenotypic and genotypic analysis of a human medulloblastoma cell line and transplantable xenograft (D341 Med) demonstrating amplification of c-myc. Am J Pathol 130:472-84, 1988

119. Bigner SH, Friedman HS, Vogelstein B, et al: Amplification of the c-myc gene in human medulloblastoma cell lines and xenografts. Cancer Res 50:2347-50, 1990

120. Pietsch T, Scharmann T, Fonatsch C, et al: Characterization of five new cell lines derived from human primitive neuroectodermal tumors of the central nervous system. Cancer Res 54:3278-87, 1994

121. Badiali M, Pession A, Basso G, et al: N-myc and c-myc oncogenes amplification in medulloblastomas. Evidence of particularly aggressive behavior of a tumor with c-myc amplification. Tumori 77:118-21, 1991

122. Tomlinson FH, Jenkins RB, Scheithauer BW, et al: Aggressive medulloblastoma with high-level N-myc amplification. Mayo Clin Proc 69:359-65, 1994

123. Reardon DA, Michalkiewicz E, Boyett JM, et al: Extensive genomic abnormalities in childhood medulloblastoma by comparative genomic hybridization. Cancer Res 57:4042-7, 1997

124. Gilhuis HJ, Anderl KL, Boerman RH, et al: Comparative genomic hybridization of medulloblastomas and clinical relevance: eleven new cases and a review of the literature. Clin Neurol Neurosurg 102:203-209, 2000

125. Bayani J, Zielenska M, Marrano P, et al: Molecular cytogenetic analysis of medulloblastomas and supratentorial primitive neuroectodermal tumors by using conventional banding, comparative genomic hybridization, and spectral karyotyping. J Neurosurg 93:437-48, 2000

126. Nicholson J, Wickramasinghe C, Ross F, et al: Imbalances of chromosome 17 in medulloblastomas determined by comparative genomic hybridisation and fluorescence in situ hybridisation. Mol Pathol 53:313-9, 2000

127. Aldosari N, Wiltshire RN, Dutra A, et al: Comprehensive molecular cytogenetic investigation of chromosomal abnormalities in human medulloblastoma cell lines and xenograft. Neuro Oncol 4:75-85, 2002

128. Avet-Loiseau H, Venuat AM, Terrier-Lacombe MJ, et al: Comparative genomic hybridization detects many recurrent imbalances in central nervous system primitive neuroectodermal tumours in children. Br J Cancer 79:1843-7, 1999

129. Eberhart CG, Kratz JE, Schuster A, et al: Comparative genomic hybridization detects an increased number of chromosomal alterations in large cell/anaplastic medulloblastomas. Brain Pathol 12:36-44, 2002

130. Bigner SH, McLendon RE, Fuchs H, et al: Chromosomal characteristics of childhood brain tumors. Cancer Genet Cytogenet 97:125-34, 1997 175

131. Cohen N, Betts DR, Tavori U, et al: Karyotypic evolution pathways in medulloblastoma/primitive neuroectodermal tumor determined with a combination of spectral karyotyping, G-banding, and fluorescence in situ hybridization. Cancer Genet Cytogenet 149:44- 52, 2004

132. Bayani JM, Squire JA: Applications of SKY in cancer cytogenetics. Cancer Invest 20:373-86, 2002

133. Davies JJ, Wilson IM, Lam WL: Array CGH technologies and their applications to cancer genomes. Chromosome Res 13:237-48, 2005

134. Morozova O, Marra MA: From cytogenetics to next-generation sequencing technologies: advances in the detection of genome rearrangements in tumors. Biochem Cell Biol 86:81-91, 2008

135. Boon K, Eberhart CG, Riggins GJ: Genomic amplification of orthodenticle homologue 2 in medulloblastomas. Cancer Res 65:703-7, 2005

136. Di C, Liao S, Adamson DC, et al: Identification of OTX2 as a medulloblastoma oncogene whose product can be targeted by all-trans retinoic acid. Cancer Res 65:919-24, 2005

137. Wang TL, Maierhofer C, Speicher MR, et al: Digital karyotyping. Proc Natl Acad Sci U S A 99:16156-61, 2002

138. Leary RJ, Cummins J, Wang TL, et al: Digital karyotyping. Nat Protoc 2:1973- 86, 2007

139. Mendrzyk F, Radlwimmer B, Joos S, et al: Genomic and protein expression profiling identifies CDK6 as novel independent prognostic marker in medulloblastoma. J Clin Oncol 23:8853-62, 2005

140. Adesina AM, Nguyen Y, Mehta V, et al: FOXG1 dysregulation is a frequent event in medulloblastoma. J Neurooncol 85:111-22, 2007

141. McCabe MG, Ichimura K, Liu L, et al: High-resolution array-based comparative genomic hybridization of medulloblastomas and supratentorial primitive neuroectodermal tumors. J Neuropathol Exp Neurol 65:549-61, 2006

142. Lo KC, Rossi MR, Eberhart CG, et al: Genome wide copy number abnormalities in pediatric medulloblastomas as assessed by array comparative genome hybridization. Brain Pathol 17:282-96, 2007

143. Hui AB, Takano H, Lo KW, et al: Identification of a novel homozygous deletion region at 6q23.1 in medulloblastomas using high-resolution array comparative genomic hybridization analysis. Clin Cancer Res 11:4707-16, 2005

144. Ehrbrecht A, Muller U, Wolter M, et al: Comprehensive genomic analysis of desmoplastic medulloblastomas: identification of novel amplified genes and separate evaluation of the different histological components. J Pathol 208:554-63, 2006 176

145. Clifford SC, Lusher ME, Lindsey JC, et al: Wnt/Wingless pathway activation and chromosome 6 loss characterize a distinct molecular sub-group of medulloblastomas associated with a favorable prognosis. Cell Cycle 5:2666-70, 2006

146. Fattet S, Haberler C, Legoix P, et al: Beta-catenin status in paediatric medulloblastomas: correlation of immunohistochemical expression with mutational status, genetic profiles, and clinical characteristics. J Pathol 218:86-94, 2009

147. Michiels EM, Oussoren E, Van Groenigen M, et al: Genes differentially expressed in medulloblastoma and fetal brain. Physiol Genomics 1:83-91, 1999

148. Velculescu VE, Zhang L, Zhou W, et al: Characterization of the yeast transcriptome. Cell 88:243-51, 1997

149. Velculescu VE, Zhang L, Vogelstein B, et al: Serial analysis of gene expression. Science 270:484-7, 1995

150. Velculescu VE: Essay: Amersham Pharmacia Biotech & Science prize. Tantalizing transcriptomes--SAGE and its use in global gene expression analysis. Science 286:1491-2, 1999

151. van Ruissen F, Baas F: Serial analysis of gene expression (SAGE). Methods Mol Biol 383:41-66, 2007

152. Saha S, Sparks AB, Rago C, et al: Using the transcriptome to annotate the genome. Nat Biotechnol 20:508-12, 2002

153. Polyak K, Riggins GJ: Gene discovery using the serial analysis of gene expression technique: implications for cancer research. J Clin Oncol 19:2948-58, 2001

154. Zeltzer PM, Boyett JM, Finlay JL, et al: Metastasis stage, adjuvant treatment, and residual tumor are prognostic factors for medulloblastoma in children: conclusions from the Children's Cancer Group 921 randomized phase III study. J Clin Oncol 17:832-45, 1999

155. Taylor RE, Bailey CC, Robinson KJ, et al: Outcome for patients with metastatic (M2-3) medulloblastoma treated with SIOP/UKCCSG PNET-3 chemotherapy. Eur J Cancer 41:727-34, 2005

156. Kortmann RD, Kuhl J, Timmermann B, et al: Postoperative neoadjuvant chemotherapy before radiotherapy as compared to immediate radiotherapy followed by maintenance chemotherapy in the treatment of medulloblastoma in childhood: results of the German prospective randomized trial HIT '91. Int J Radiat Oncol Biol Phys 46:269-79, 2000

157. MacDonald TJ, Brown KM, LaFleur B, et al: Expression profiling of medulloblastoma: PDGFRA and the RAS/MAPK pathway as therapeutic targets for metastatic disease. Nat Genet 29:143-52, 2001

158. Gilbertson RJ, Clifford SC: PDGFRB is overexpressed in metastatic medulloblastoma. Nat Genet 35:197-8, 2003 177

159. Pomeroy SL, Tamayo P, Gaasenbeek M, et al: Prediction of central nervous system embryonal tumour outcome based on gene expression. Nature 415:436-42, 2002

160. Rorke LB: Classification and grading of childhood brain tumors. Overview and statement of the problem. Cancer 56:1848-9, 1985

161. Li MH, Bouffet E, Hawkins CE, et al: Molecular genetics of supratentorial primitive neuroectodermal tumors and pineoblastoma. Neurosurg Focus 19:E3, 2005

162. Reddy AT, Janss AJ, Phillips PC, et al: Outcome for children with supratentorial primitive neuroectodermal tumors treated with surgery, radiation, and chemotherapy. Cancer 88:2189-93, 2000

163. Pfister S, Remke M, Toedt G, et al: Supratentorial primitive neuroectodermal tumors of the central nervous system frequently harbor deletions of the CDKN2A locus and other genomic aberrations distinct from medulloblastomas. Genes Chromosomes Cancer 46:839- 51, 2007

164. Cohen BH, Zeltzer PM, Boyett JM, et al: Prognostic factors and treatment results for supratentorial primitive neuroectodermal tumors in children using radiation and chemotherapy: a Childrens Cancer Group randomized trial. J Clin Oncol 13:1687-96, 1995

165. Timmermann B, Kortmann RD, Kuhl J, et al: Role of radiotherapy in the treatment of supratentorial primitive neuroectodermal tumors in childhood: results of the prospective German brain tumor trials HIT 88/89 and 91. J Clin Oncol 20:842-9, 2002

166. Kasuga C, Nakahara Y, Ueda S, et al: Expression of MAGE and GAGE genes in medulloblastoma and modulation of resistance to chemotherapy. Laboratory investigation. J Neurosurg Pediatr 1:305-13, 2008

167. Park PC, Taylor MD, Mainprize TG, et al: Transcriptional profiling of medulloblastoma in children. J Neurosurg 99:534-41, 2003

168. Neben K, Korshunov A, Benner A, et al: Microarray-based screening for molecular markers in medulloblastoma revealed STK15 as independent predictor for survival. Cancer Res 64:3103-11, 2004

169. Boon K, Edwards JB, Siu IM, et al: Comparison of medulloblastoma and normal neural transcriptomes identifies a restricted set of activated genes. Oncogene 22:7687-94, 2003

170. Yokota N, Mainprize TG, Taylor MD, et al: Identification of differentially expressed and developmentally regulated genes in medulloblastoma using suppression subtraction hybridization. Oncogene 23:3444-53, 2004

171. von Stein OD: Isolation of differentially expressed genes through subtractive suppression hybridization. Methods Mol Biol 175:263-78, 2001

172. Golub TR, Slonim DK, Tamayo P, et al: Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 286:531-7, 1999 178

173. Nutt CL, Mani DR, Betensky RA, et al: Gene expression-based classification of malignant gliomas correlates better with survival than histological classification. Cancer Res 63:1602-7, 2003

174. Mischel PS, Cloughesy TF, Nelson SF: DNA-microarray analysis of brain cancer: molecular classification for therapy. Nat Rev Neurosci 5:782-92, 2004

175. Calin GA, Croce CM: MicroRNA signatures in human cancers. Nat Rev Cancer 6:857-66, 2006

176. Hammond SM: MicroRNAs as oncogenes. Curr Opin Genet Dev 16:4-9, 2006

177. Esquela-Kerscher A, Slack FJ: Oncomirs - microRNAs with a role in cancer. Nat Rev Cancer 6:259-69, 2006

178. Ambros V: The functions of animal microRNAs. Nature 431:350-5, 2004

179. Bartel DP: MicroRNAs: target recognition and regulatory functions. Cell 136:215-33, 2009

180. Ferretti E, De Smaele E, Po A, et al: MicroRNA profiling in human medulloblastoma. Int J Cancer 124:568-77, 2009

181. Pierson J, Hostager B, Fan R, et al: Regulation of cyclin dependent kinase 6 by microRNA 124 in medulloblastoma. J Neurooncol 90:1-7, 2008

182. Cheng LC, Pastrana E, Tavazoie M, et al: miR-124 regulates adult neurogenesis in the subventricular zone stem cell niche. Nat Neurosci 12:399-408, 2009

183. Makeyev EV, Zhang J, Carrasco MA, et al: The MicroRNA miR-124 promotes neuronal differentiation by triggering brain-specific alternative pre-mRNA splicing. Mol Cell 27:435-48, 2007

184. Conaco C, Otto S, Han JJ, et al: Reciprocal actions of REST and a microRNA promote neuronal identity. Proc Natl Acad Sci U S A 103:2422-7, 2006

185. Silber J, Lim DA, Petritsch C, et al: miR-124 and miR-137 inhibit proliferation of glioblastoma multiforme cells and induce differentiation of brain tumor stem cells. BMC Med 6:14, 2008

186. Ferretti E, De Smaele E, Miele E, et al: Concerted microRNA control of Hedgehog signalling in cerebellar neuronal progenitor and tumour cells. EMBO J 27:2616-27, 2008

187. Northcott PA, Fernandez LA, Hagan JP, et al: The miR-17/92 polycistron is upregulated in sonic hedgehog-driven medulloblastomas and induced by N-myc in sonic hedgehog- treated cerebellar neural precursors. Cancer Res 69:3249-55, 2009

188. Fogarty MP, Kessler JD, Wechsler-Reya RJ: Morphing into cancer: the role of developmental signaling pathways in brain tumor formation. J Neurobiol 64:458-75, 2005 179

189. Solecki DJ, Liu XL, Tomoda T, et al: Activated Notch2 signaling inhibits differentiation of cerebellar granule neuron precursors by maintaining proliferation. Neuron 31:557-68., 2001

190. Garzia L, Andolfo I, Cusanelli E, et al: MicroRNA-199b-5p impairs cancer stem cells through negative regulation of HES1 in medulloblastoma. PLoS ONE 4:e4998, 2009

191. Lu Y, Ryan SL, Elliott DJ, et al: Amplification and overexpression of Hsa-miR- 30b, Hsa-miR-30d and KHDRBS3 at 8q24.22-q24.23 in medulloblastoma. PLoS One 4:e6159, 2009

192. Uziel T, Karginov FV, Xie S, et al: The miR-17~92 cluster collaborates with the Sonic Hedgehog pathway in medulloblastoma. Proc Natl Acad Sci U S A 106:2812-7, 2009

193. Esteller M: Epigenetics in cancer. N Engl J Med 358:1148-59, 2008

194. Bernstein BE, Meissner A, Lander ES: The mammalian epigenome. Cell 128:669- 81, 2007

195. Kouzarides T: Chromatin modifications and their function. Cell 128:693-705, 2007

196. Harada K, Toyooka S, Maitra A, et al: Aberrant promoter methylation and silencing of the RASSF1A gene in pediatric tumors and cell lines. Oncogene 21:4345-9, 2002

197. Lusher ME, Lindsey JC, Latif F, et al: Biallelic epigenetic inactivation of the RASSF1A tumor suppressor gene in medulloblastoma development. Cancer Res 62:5906-11, 2002

198. Horiguchi K, Tomizawa Y, Tosaka M, et al: Epigenetic inactivation of RASSF1A candidate tumor suppressor gene at 3p21.3 in brain tumors. Oncogene 22:7862-5, 2003

199. Rood BR, Zhang H, Weitman DM, et al: Hypermethylation of HIC-1 and 17p allelic loss in medulloblastoma. Cancer Res 62:3794-7, 2002

200. Waha A, Koch A, Meyer-Puttlitz B, et al: Epigenetic silencing of the HIC-1 gene in human medulloblastomas. J Neuropathol Exp Neurol 62:1192-201, 2003

201. Grotzer MA, Eggert A, Zuzak TJ, et al: Resistance to TRAIL-induced apoptosis in primitive neuroectodermal brain tumor cells correlates with a loss of caspase-8 expression. Oncogene 19:4604-10, 2000

202. Gonzalez-Gomez P, Bello MJ, Inda MM, et al: Deletion and aberrant CpG island methylation of Caspase 8 gene in medulloblastoma. Oncol Rep 12:663-6, 2004

203. Lindsey JC, Lusher ME, Anderton JA, et al: Identification of tumour-specific epigenetic events in medulloblastoma development by hypermethylation profiling. Carcinogenesis 25:661-8, 2004 180

204. Waha A, Koch A, Hartmann W, et al: SGNE1/7B2 is epigenetically altered and transcriptionally downregulated in human medulloblastomas. Oncogene 26:5662-8, 2007

205. Yan PS, Potter D, Deatherage DE, et al: Differential methylation hybridization: profiling DNA methylation with a high-density CpG island microarray. Methods Mol Biol 507:89-106, 2009

206. Pfister S, Schlaeger C, Mendrzyk F, et al: Array-based profiling of reference- independent methylation status (aPRIMES) identifies frequent promoter methylation and consecutive downregulation of ZIC2 in pediatric medulloblastoma. Nucleic Acids Res 35:e51, 2007

207. Anderton JA, Lindsey JC, Lusher ME, et al: Global analysis of the medulloblastoma epigenome identifies disease-subgroup-specific inactivation of COL1A2. Neuro Oncol 10:981-94, 2008

208. Northcott PA, Rutka JT, Taylor MD: Genomics of medulloblastoma: from Giemsa-banding to next-generation sequencing in 20 years. Neurosurg Focus 28:E6

209. Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature 455:1061-8, 2008

210. Yan H, Parsons DW, Jin G, et al: IDH1 and IDH2 mutations in gliomas. N Engl J Med 360:765-73, 2009

211. Lin M, Wei LJ, Sellers WR, et al: dChipSNP: significance curve and clustering of SNP-array-based loss-of-heterozygosity data. Bioinformatics 20:1233-40, 2004

212. Nannya Y, Sanada M, Nakazaki K, et al: A robust algorithm for copy number detection using high-density oligonucleotide single nucleotide polymorphism genotyping arrays. Cancer Res 65:6071-9, 2005

213. Iafrate AJ, Feuk L, Rivera MN, et al: Detection of large-scale variation in the human genome. Nat Genet 36:949-51, 2004

214. Hupe P, Stransky N, Thiery JP, et al: Analysis of array CGH data: from signal ratio to gain and loss of DNA regions. Bioinformatics 20:3413-22, 2004

215. Beroukhim R, Getz G, Nghiemphu L, et al: Assessing the significance of chromosomal aberrations in cancer: methodology and application to glioma. Proc Natl Acad Sci U S A 104:20007-12, 2007

216. Weir BA, Woo MS, Getz G, et al: Characterizing the cancer genome in lung adenocarcinoma. Nature 450:893-8, 2007

217. Cloos PA, Christensen J, Agger K, et al: The putative oncogene GASC1 demethylates tri- and dimethylated lysine 9 on histone H3. Nature 442:307-11, 2006

218. Matsuzaki H, Dong S, Loi H, et al: Genotyping over 100,000 SNPs on a pair of oligonucleotide arrays. Nat Methods 1:109-11, 2004 181

219. Huang J, Wei W, Zhang J, et al: Whole genome DNA copy number changes identified by high density oligonucleotide arrays. Hum Genomics 1:287-99, 2004

220. Matsuzaki H, Loi H, Dong S, et al: Parallel genotyping of over 10,000 SNPs using a one-primer assay on a high-density oligonucleotide array. Genome Res 14:414-25, 2004

221. LaFramboise T: Single nucleotide polymorphism arrays: a decade of biological, computational and technological advances. Nucleic Acids Res 37:4181-93, 2009

222. Beroukhim R, Lin M, Park Y, et al: Inferring loss-of-heterozygosity from unpaired tumors using high-density oligonucleotide SNP arrays. PLoS Comput Biol 2:e41, 2006

223. Romer J, Curran T: Targeting medulloblastoma: small-molecule inhibitors of the Sonic Hedgehog pathway as potential cancer therapeutics. Cancer Res 65:4975-8, 2005

224. Berman DM, Karhadkar SS, Hallahan AR, et al: Medulloblastoma growth inhibition by hedgehog pathway blockade. Science 297:1559-61, 2002

225. Simeone A, Puelles E, Acampora D: The Otx family. Curr Opin Genet Dev 12:409-15, 2002

226. Adamson DC, Shi Q, Wortham M, et al: OTX2 Is Critical for the Maintenance and Progression of Shh-Independent Medulloblastomas. Cancer Res 70:181-91, 2010

227. Bhatia B, Northcott PA, Hambardzumyan D, et al: Tuberous sclerosis complex suppression in cerebellar development and medulloblastoma: separate regulation of mammalian target of rapamycin activity and p27 Kip1 localization. Cancer Res 69:7224-34, 2009

228. Crino PB, Nathanson KL, Henske EP: The tuberous sclerosis complex. N Engl J Med 355:1345-56, 2006

229. Li J, Yen C, Liaw D, et al: PTEN, a putative protein tyrosine phosphatase gene mutated in human brain, breast, and prostate cancer. Science 275:1943-7, 1997

230. Guldberg P, thor Straten P, Birck A, et al: Disruption of the MMAC1/PTEN gene by deletion or mutation is a frequent event in malignant melanoma. Cancer Res 57:3660-3, 1997

231. T'Ang A, Varley JM, Chakraborty S, et al: Structural rearrangement of the retinoblastoma gene in human breast carcinoma. Science 242:263-6, 1988

232. Stratton MR, Williams S, Fisher C, et al: Structural alterations of the RB1 gene in human soft tissue tumours. Br J Cancer 60:202-5, 1989

233. Ogawa H, Ishiguro K, Gaubatz S, et al: A complex with chromatin modifiers that occupies E2F- and Myc-responsive genes in G0 cells. Science 296:1132-6, 2002

234. Tachibana M, Ueda J, Fukuda M, et al: Histone methyltransferases G9a and GLP form heteromeric complexes and are both crucial for methylation of euchromatin at H3-K9. Genes Dev 19:815-26, 2005 182

235. Kleefstra T, Brunner HG, Amiel J, et al: Loss-of-function mutations in euchromatin histone methyl transferase 1 (EHMT1) cause the 9q34 subtelomeric deletion syndrome. Am J Hum Genet 79:370-7, 2006

236. Kleefstra T, Smidt M, Banning MJ, et al: Disruption of the gene Euchromatin Histone Methyl Transferase1 (Eu-HMTase1) is associated with the 9q34 subtelomeric deletion syndrome. J Med Genet 42:299-306, 2005

237. Klymenko T, Papp B, Fischle W, et al: A Polycomb group protein complex with sequence-specific DNA-binding and selective methyl-lysine-binding activities. Genes Dev 20:1110-22, 2006

238. Kim J, Daniel J, Espejo A, et al: Tudor, MBT and chromo domains gauge the degree of lysine methylation. EMBO Rep 7:397-403, 2006

239. Trojer P, Li G, Sims RJ, 3rd, et al: L3MBTL1, a histone-methylation-dependent chromatin lock. Cell 129:915-28, 2007

240. Wu S, Trievel RC, Rice JC: Human SFMBT is a transcriptional repressor protein that selectively binds the N-terminal tail of histone H3. FEBS Lett 581:3289-96, 2007

241. Guo Y, Nady N, Qi C, et al: Methylation-state-specific recognition of histones by the MBT repeat protein L3MBTL2. Nucleic Acids Res 37:2204-10, 2009

242. Hu L, Zhu YT, Qi C, et al: Identification of Smyd4 as a potential tumor suppressor gene involved in breast cancer development. Cancer Res 69:4067-72, 2009

243. Mullighan CG, Goorha S, Radtke I, et al: Genome-wide analysis of genetic alterations in acute lymphoblastic leukaemia. Nature 446:758-64, 2007

244. Cloos PA, Christensen J, Agger K, et al: Erasing the methyl mark: histone demethylases at the center of cellular differentiation and disease. Genes Dev 22:1115-40, 2008

245. Agger K, Christensen J, Cloos PA, et al: The emerging functions of histone demethylases. Curr Opin Genet Dev 18:159-68, 2008

246. Shin S, Janknecht R: Diversity within the JMJD2 histone demethylase family. Biochem Biophys Res Commun 353:973-7, 2007

247. Shi Y, Sawada J, Sui G, et al: Coordinated histone modifications mediated by a CtBP co-repressor complex. Nature 422:735-8, 2003

248. Fodor BD, Kubicek S, Yonezawa M, et al: Jmjd2b antagonizes H3K9 trimethylation at pericentric heterochromatin in mammalian cells. Genes Dev 20:1557-62, 2006

249. Howe L, Auston D, Grant P, et al: Histone H3 specific acetyltransferases are essential for cell cycle progression. Genes Dev 15:3144-54, 2001 183

250. Lu J, Ruhf ML, Perrimon N, et al: A genome-wide RNA interference screen identifies putative chromatin regulators essential for E2F repression. Proc Natl Acad Sci U S A 104:9381-6, 2007

251. Olson MV, Johnson DG, Jiang H, et al: Transgenic E2F1 expression in the mouse brain induces a human-like bimodal pattern of tumors. Cancer Res 67:4005-9, 2007

252. Marino S, Hoogervoorst D, Brandner S, et al: Rb and p107 are required for normal cerebellar development and granule cell survival but not for Purkinje cell persistence. Development 130:3359-68, 2003

253. Gateff E, Loffler T, Wismar J: A temperature-sensitive brain tumor suppressor mutation of Drosophila melanogaster: developmental studies and molecular localization of the gene. Mech Dev 41:15-31, 1993

254. Trimarchi JM, Fairchild B, Wen J, et al: The E2F6 transcription factor is a component of the mammalian Bmi1-containing polycomb complex. Proc Natl Acad Sci U S A 98:1519-24, 2001

255. Wiederschain D, Chen L, Johnson B, et al: Contribution of polycomb homologues Bmi-1 and Mel-18 to medulloblastoma pathogenesis. Mol Cell Biol 27:4968-79, 2007

256. Zindy F, Knoepfler PS, Xie S, et al: N-Myc and the cyclin-dependent kinase inhibitors p18Ink4c and p27Kip1 coordinately regulate cerebellar development. Proc Natl Acad Sci U S A 103:11579-83, 2006

257. Cartwright P, Muller H, Wagener C, et al: E2F-6: a novel member of the E2F family is an inhibitor of E2F-dependent transcription. Oncogene 17:611-23, 1998

258. Panteleeva I, Boutillier S, See V, et al: HP1alpha guides neuronal fate by timing E2F-targeted genes silencing during terminal differentiation. EMBO J 26:3616-28, 2007

259. Thomas T, Corcoran LM, Gugasyan R, et al: Monocytic leukemia zinc finger protein is essential for the development of long-term reconstituting hematopoietic stem cells. Genes Dev 20:1175-86, 2006

260. Annunziata CM, Davis RE, Demchenko Y, et al: Frequent engagement of the classical and alternative NF-kappaB pathways by diverse genetic abnormalities in multiple myeloma. Cancer Cell 12:115-30, 2007

261. Romer JT, Kimura H, Magdaleno S, et al: Suppression of the Shh pathway using a small molecule inhibitor eliminates medulloblastoma in Ptc1(+/-)p53(-/-) mice. Cancer Cell 6:229-40, 2004

262. Kubicek S, O'Sullivan RJ, August EM, et al: Reversal of H3K9me2 by a small- molecule inhibitor for the G9a histone methyltransferase. Mol Cell 25:473-81, 2007

263. Ray A, Ho M, Ma J, et al: A clinicobiological model predicting survival in medulloblastoma. Clin Cancer Res 10:7613-20, 2004 184

264. Gilbertson RJ, Perry RH, Kelly PJ, et al: Prognostic significance of HER2 and HER4 coexpression in childhood medulloblastoma. Cancer Res 57:3272-80, 1997

265. Grotzer MA, Janss AJ, Fung K, et al: TrkC expression predicts good clinical outcome in primitive neuroectodermal brain tumors. J Clin Oncol 18:1027-35, 2000

266. Grotzer MA, Hogarty MD, Janss AJ, et al: MYC messenger RNA expression predicts survival outcome in childhood primitive neuroectodermal tumor/medulloblastoma. Clin Cancer Res 7:2425-33, 2001

267. Gilbertson R, Wickramasinghe C, Hernan R, et al: Clinical and molecular stratification of disease risk in medulloblastoma. Br J Cancer 85:705-12, 2001

268. Pan E, Pellarin M, Holmes E, et al: Isochromosome 17q is a negative prognostic factor in poor-risk childhood medulloblastoma patients. Clin Cancer Res 11:4733-40, 2005

269. Eberhart CG, Chaudhry A, Daniel RW, et al: Increased p53 immunopositivity in anaplastic medulloblastoma and supratentorial PNET is not caused by JC virus. BMC Cancer 5:19, 2005

270. Tibshirani R, Hastie T, Narasimhan B, et al: Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Natl Acad Sci U S A 99:6567-72, 2002

271. Subramanian A, Tamayo P, Mootha VK, et al: Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102:15545-50, 2005

272. Mootha VK, Lindgren CM, Eriksson KF, et al: PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet 34:267-73, 2003

273. Brunet JP, Tamayo P, Golub TR, et al: Metagenes and molecular pattern discovery using matrix factorization. Proc Natl Acad Sci U S A 101:4164-9, 2004

274. Hoshida Y, Brunet JP, Tamayo P, et al: Subclass mapping: identifying common subtypes in independent disease data sets. PLoS One 2:e1195, 2007

275. McManamy CS, Pears J, Weston CL, et al: Nodule formation and desmoplasia in medulloblastomas-defining the nodular/desmoplastic variant and its biological behavior. Brain Pathol 17:151-64, 2007

276. Hubert L, Arabie, P: Comparing partitions. Journal of classification 2:193-218, 1985

277. Vorechovsky I, Tingby O, Hartman M, et al: Somatic mutations in the human homologue of Drosophila patched in primitive neuroectodermal tumours. Oncogene 15:361-6, 1997

278. Wolter M, Reifenberger J, Sommer C, et al: Mutations in the human homologue of the Drosophila segment polarity gene patched (PTCH) in sporadic basal cell carcinomas of the 185 skin and primitive neuroectodermal tumors of the central nervous system. Cancer Res 57:2581-5, 1997

279. Rutkowski S, Gerber NU, von Hoff K, et al: Treatment of early childhood medulloblastoma by postoperative chemotherapy and deferred radiotherapy. Neuro Oncol 11:201-10, 2009

280. Judkins AR, Mauger J, Ht A, et al: Immunohistochemical analysis of hSNF5/INI1 in pediatric CNS neoplasms. Am J Surg Pathol 28:644-50, 2004

281. Biegel JA, Zhou JY, Rorke LB, et al: Germ-line and acquired mutations of INI1 in atypical teratoid and rhabdoid tumors. Cancer Res 59:74-9, 1999

282. Hammond SM: MicroRNAs as tumor suppressors. Nat Genet 39:582-3, 2007

283. Croce CM: Causes and consequences of microRNA dysregulation in cancer. Nat Rev Genet 10:704-14, 2009

284. Fernandez LA, Northcott PA, Taylor MD, et al: Normal and oncogenic roles for microRNAs in the developing brain. Cell Cycle 8:4049-54, 2009

285. Mendell JT: miRiad roles for the miR-17-92 cluster in development and disease. Cell 133:217-22, 2008

286. Petrocca F, Vecchione A, Croce CM: Emerging role of miR-106b-25/miR-17-92 clusters in the control of transforming growth factor beta signaling. Cancer Res 68:8191-4, 2008

287. Rubin JB, Rowitch DH: Medulloblastoma: a problem of developmental biology. Cancer Cell 2:7-8, 2002

288. Rao Y, Lee Y, Jarjoura D, et al: A comparison of normalization techniques for microRNA microarray data. Stat Appl Genet Mol Biol 7:Article22, 2008

289. Chen C, Ridzon DA, Broomer AJ, et al: Real-time quantification of microRNAs by stem-loop RT-PCR. Nucleic Acids Res 33:e179, 2005

290. Ota A, Tagawa H, Karnan S, et al: Identification and characterization of a novel gene, C13orf25, as a target for 13q31-q32 amplification in malignant lymphoma. Cancer Res 64:3087-95, 2004

291. He L, Thomson JM, Hemann MT, et al: A microRNA polycistron as a potential human oncogene. Nature 435:828-33, 2005

292. Hayashita Y, Osada H, Tatematsu Y, et al: A polycistronic microRNA cluster, miR-17-92, is overexpressed in human lung cancers and enhances cell proliferation. Cancer Res 65:9628-32, 2005

293. O'Donnell KA, Wentzel EA, Zeller KI, et al: c-Myc-regulated microRNAs modulate E2F1 expression. Nature 435:839-43, 2005 186

294. Schulte JH, Horn S, Otto T, et al: MYCN regulates oncogenic MicroRNAs in neuroblastoma. Int J Cancer 122:699-704, 2008

295. Oliver TG, Grasfeder LL, Carroll AL, et al: Transcriptional profiling of the Sonic hedgehog response: a critical role for N-myc in proliferation of neuronal precursors. Proc Natl Acad Sci U S A 100:7331-6, 2003

296. Xiao C, Srinivasan L, Calado DP, et al: Lymphoproliferative disease and autoimmunity in mice with increased miR-17-92 expression in lymphocytes. Nat Immunol 9:405-14, 2008

297. Inomata M, Tagawa H, Guo YM, et al: MicroRNA-17-92 down-regulates expression of distinct targets in different B-cell lymphoma subtypes. Blood 113:396-402, 2009

298. Dupuy AJ, Jenkins NA, Copeland NG: Sleeping beauty: a novel cancer gene discovery tool. Hum Mol Genet 15 Spec No 1:R75-9, 2006

299. Collier LS, Largaespada DA: Hopping around the tumor genome: transposons for cancer gene discovery. Cancer Res 65:9607-10, 2005

300. Sylvestre Y, De Guire V, Querido E, et al: An E2F/miR-20a autoregulatory feedback loop. J Biol Chem 282:2135-43, 2007

301. Woods K, Thomson JM, Hammond SM: Direct regulation of an oncogenic micro- RNA cluster by E2F transcription factors. J Biol Chem 282:2130-4, 2007

302. Ivanovska I, Ball AS, Diaz RL, et al: MicroRNAs in the miR-106b family regulate p21/CDKN1A and promote cell cycle progression. Mol Cell Biol 28:2167-74, 2008

303. Petrocca F, Visone R, Onelli MR, et al: E2F1-regulated microRNAs impair TGFbeta-dependent cell-cycle arrest and apoptosis in gastric cancer. Cancer Cell 13:272-86, 2008

304. Olive V, Bennett MJ, Walker JC, et al: miR-19 is a key oncogenic component of mir-17-92. Genes Dev 23:2839-49, 2009

305. Ventura A, Young AG, Winslow MM, et al: Targeted deletion reveals essential and overlapping functions of the miR-17 through 92 family of miRNA clusters. Cell 132:875-86, 2008

306. Dews M, Homayouni A, Yu D, et al: Augmentation of tumor angiogenesis by a Myc-activated microRNA cluster. Nat Genet 38:1060-5, 2006

307. Godlewski J, Nowicki MO, Bronisz A, et al: Targeting of the Bmi-1 oncogene/stem cell renewal factor by microRNA-128 inhibits glioma proliferation and self- renewal. Cancer Res 68:9125-30, 2008

308. Wang Y, Lee AT, Ma JZ, et al: Profiling microRNA expression in hepatocellular carcinoma reveals microRNA-224 up-regulation and apoptosis inhibitor-5 as a microRNA-224- specific target. J Biol Chem 283:13205-15, 2008 187

309. Veerla S, Lindgren D, Kvist A, et al: MiRNA expression in urothelial carcinomas: important roles of miR-10a, miR-222, miR-125b, miR-7 and miR-452 for tumor stage and metastasis, and frequent homozygous losses of miR-31. Int J Cancer 124:2236-42, 2009

310. Lander ES, Linton LM, Birren B, et al: Initial sequencing and analysis of the human genome. Nature 409:860-921, 2001

311. Venter JC, Adams MD, Myers EW, et al: The sequence of the human genome. Science 291:1304-51, 2001

312. Jones S, Hruban RH, Kamiyama M, et al: Exomic sequencing identifies PALB2 as a pancreatic cancer susceptibility gene. Science 324:217, 2009

313. Ding L, Getz G, Wheeler DA, et al: Somatic mutations affect key pathways in lung adenocarcinoma. Nature 455:1069-75, 2008

314. Rusk N, Kiermer V: Primer: Sequencing--the next generation. Nat Methods 5:15, 2008

315. Schuster SC: Next-generation sequencing transforms today's biology. Nat Methods 5:16-8, 2008

316. Shendure J, Ji H: Next-generation DNA sequencing. Nat Biotechnol 26:1135-45, 2008

317. Coombs A: The sequencing shakeup. Nat Biotechnol 26:1109-12, 2008

318. Ansorge WJ: Next-generation DNA sequencing techniques. N Biotechnol 25:195- 203, 2009

319. Hodges E, Xuan Z, Balija V, et al: Genome-wide in situ exon capture for selective resequencing. Nat Genet 39:1522-7, 2007

320. Ng SB, Turner EH, Robertson PD, et al: Targeted capture and massively parallel sequencing of 12 human exomes. Nature, 2009

321. Wheeler DA, Srinivasan M, Egholm M, et al: The complete genome of an individual by massively parallel DNA sequencing. Nature 452:872-6, 2008

322. Wang J, Wang W, Li R, et al: The diploid genome sequence of an Asian individual. Nature 456:60-5, 2008

323. Bentley DR, Balasubramanian S, Swerdlow HP, et al: Accurate whole human genome sequencing using reversible terminator chemistry. Nature 456:53-9, 2008

324. Kim JI, Ju YS, Park H, et al: A highly annotated whole-genome sequence of a Korean individual. Nature 460:1011-5, 2009

325. Ley TJ, Mardis ER, Ding L, et al: DNA sequencing of a cytogenetically normal acute myeloid leukaemia genome. Nature 456:66-72, 2008 188

326. Morozova O, Marra MA: Applications of next-generation sequencing technologies in functional genomics. Genomics 92:255-64, 2008

327. Campbell PJ, Stephens PJ, Pleasance ED, et al: Identification of somatically acquired rearrangements in cancer using genome-wide massively parallel paired-end sequencing. Nat Genet 40:722-9, 2008

328. Chiang DY, Getz G, Jaffe DB, et al: High-resolution mapping of copy-number alterations with massively parallel sequencing. Nat Methods 6:99-103, 2009

329. Mortazavi A, Williams BA, McCue K, et al: Mapping and quantifying mammalian transcriptomes by RNA-Seq. Nat Methods 5:621-8, 2008

330. Pan Q, Shai O, Lee LJ, et al: Deep surveying of alternative splicing complexity in the human transcriptome by high-throughput sequencing. Nat Genet 40:1413-5, 2008

331. Wang Z, Gerstein M, Snyder M: RNA-Seq: a revolutionary tool for transcriptomics. Nat Rev Genet 10:57-63, 2009

332. Wilhelm BT, Marguerat S, Watt S, et al: Dynamic repertoire of a eukaryotic transcriptome surveyed at single-nucleotide resolution. Nature 453:1239-43, 2008

333. Nagalakshmi U, Wang Z, Waern K, et al: The transcriptional landscape of the yeast genome defined by RNA sequencing. Science 320:1344-9, 2008

334. Morin RD, O'Connor MD, Griffith M, et al: Application of massively parallel sequencing to microRNA profiling and discovery in human embryonic stem cells. Genome Res 18:610-21, 2008

335. Meissner A, Mikkelsen TS, Gu H, et al: Genome-scale DNA methylation maps of pluripotent and differentiated cells. Nature 454:766-70, 2008

336. Down TA, Rakyan VK, Turner DJ, et al: A Bayesian deconvolution strategy for immunoprecipitation-based DNA methylome analysis. Nat Biotechnol 26:779-85, 2008

337. Barski A, Cuddapah S, Cui K, et al: High-resolution profiling of histone methylations in the human genome. Cell 129:823-37, 2007

338. Mikkelsen TS, Ku M, Jaffe DB, et al: Genome-wide maps of chromatin state in pluripotent and lineage-committed cells. Nature 448:553-60, 2007

339. Robertson G, Hirst M, Bainbridge M, et al: Genome-wide profiles of STAT1 DNA association using chromatin immunoprecipitation and massively parallel sequencing. Nat Methods 4:651-7, 2007

340. Johnson DS, Mortazavi A, Myers RM, et al: Genome-wide mapping of in vivo protein-DNA interactions. Science 316:1497-502, 2007