DNA METHYLATION PROFILING REVEALS NOVEL BIOMARKERS AND

IMPORTANT ROLES FOR DNA METHYLTRANSFERASES IN PROSTATE

CANCER

A DISSERTATION

SUBMITTED TO THE DEPARTMENT OF GENETICS

AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

Yuya Kobayashi

February 2011

http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/zx378tc0675

© 2011 by Yuya Kobayashi. All Rights Reserved.

Re-distributed by Stanford University under license with the author.

This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.

ii

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Richard Myers, Primary Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Gavin Sherlock, Co-Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

James Brooks

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Joseph Lipsick

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Hua Tang

Approved for the Stanford University Committee on Graduate Studies.

Patricia J. Gumport, Vice Provost Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.

iii

iv

ABSTRACT

Candidate gene based studies have identified a handful of aberrant CpG DNA

methylation events in prostate cancer (Brooks et al. 1998; Yegnasubramanian et al.

2004). However, large scale DNA methylation profiles have not been examined for

normal prostates or prostate tumors. Additionally, the mechanisms behind these DNA

methylation alterations are unknown. In this thesis, I describe the results of my efforts to

better understand these previously unexplored areas of biology.

For the study presented in this thesis, I quantitatively profiled 95 primary prostate tumors

and 86 healthy prostate tissue samples for their DNA methylation levels at 26,333 CpGs

representing 14,104 gene promoters by using the Illumina HumanMethylation27

platform. When the profiles of the prostate tissue samples were compared, I observed a

substantial number of tumor-specific DNA methylation alterations. A 2-class

Significance Analysis of this dataset revealed 5,912 CpG sites with increased DNA

methylation and 2,151 CpG sites with decreased DNA methylation in tumors (FDR <

0.8%). Prediction Analysis of this dataset identified 87 CpGs that are the most predictive

diagnostic methylation biomarkers of prostate cancer. By integrating available clinical

follow-up data, I also identified 69 prognostic DNA methylation alterations that correlate

with biochemical recurrence of the tumor.

v

To identify the mechanisms responsible for these genome-wide DNA methylation

alterations, I measured the gene expression levels of several DNA methyltransferases

(DNMTs) and their interacting proteins by TaqMan qPCR and observed increased

expression of DNMT3A2, DNMT3B, and EZH2 in tumors. Subsequent transient

transfection assays in cultured primary prostate cells revealed that DNMT3B1 and

DNMT3B2 overexpression resulted in increased methylation of a substantial subset of

CpG sites that also showed tumor-specific increased methylation.

vi

ACKNOWLEDGEMENTS

I would like to thank the following people and organizations for material, intellectual and

moral support:

All the patients who donated their prostates for the advancement of science.

Devin Absher, Kenny Day, Krista Stanton and Kevin Roberts for assistance with

experiments and analysis of the HumanMethylation27 data.

James Brooks and Zulfiqar Gulzar for collecting the samples and their guidance and

support.

Donna Peehl and Sarah Young for assistance with the cultured primary prostate

cells.

All the faculty, staff and student in the Stanford Genetics Department for creating

an intellectually stimulating and fun environment.

My thesis committee members Hua Tang, Joseph Lipsick and James Brooks for

helpful comments and discussion.

My advisors Rick Myers and Gavin Sherlock for outstanding guidance and

mentorship throughout my graduate career.

All my current and past lab-mates in the Myers and Sherlock Labs for their

friendship, support and insight, day-in and day-out.

My fantastic Stanford friends for all the great memories.

My family for their continued support and doing everything to open every door of

opportunity for me.

My wife Katie for her patience and love.

vii

TABLE OF CONTENTS

Section Page

CHAPTER 1: Introduction ......................................................................................... 1

CHAPTER 2: Methods ................................................................................................ 6

CHAPTER 3: DNA methylation profiles of normal prostates and prostate tumors

....................................................................................................... 13

CHAPTER 4: DNA Methyltransferases in prostate ................................................. 35

CHAPTER 5: Discussion ............................................................................................. 44

APPENDIX ................................................................................................................... 52

REFERENCES ............................................................................................................. 60

viii

List of Tables

Table 2.1: Primer sequences used in PyroMark assays ................................................. 11

Supplementary Table S1: Clinical information associated with prostate samples ..... 53

Supplementary Table S2: Diagnostic methylation markers of prostate cancer identified

by PAM ......................................................................................................................... 56

Supplementary Table S3: Prognostic methylation markers of prostate cancer identified

by SAM survival ........................................................................................................... 58

ix

List of Illustrations

Figure 3.1: Hierarchical clustering of prostate tissues by DNA methylation ............. 15

Figure 3.2: Normal vs Tumor unpaired 2-class SAM analysis of the 181 prostate

samples .......................................................................................................................... 16

Figure 3.3: Differentially methylated CpGs of prostate tumors ................................. 17

Figure 3.4: Normal vs Tumor paired 2-class SAM analysis of the 181 prostate samples

...................................................................................................... 19

Figure 3.5: Quantitative SAM analysis of 86 prostate samples based on age of patient at

the time of surgery ........................................................................................................ 19

Figure 3.6: 749-Age dependent CpGs ....................................................................... 20

Figure 3.7: GSTP1 CpG island hypermethylation in prostate tumors ....................... 22

Figure 3.8: APC proximal promoter hypermethylation in prostate tumors ............... 23

Figure 3.9: RASSF1 proximal promoter hypermethylation in prostate tumors .......... 24

Figure 3.10: Diagnostic markers of prostate cancer identified by PAM .................... 26

Figure 3.11: PyroMark validates HumanMethylation27 results ................................. 27

Figure 3.12: Comparison of neighboring CpGs by PyroMark .................................... 30

Figure 4.1: Expression of DNMTs and EZH2 correlates with global hypermethylation in

prostate tumors .............................................................................................................. 37

Figure 4.2: Overexpression of DNMTs and EZH2 results in increased methylation at a

subset of prostate tumor hypermethylation sites ........................................................... 40

1

CHAPTER 1

INTRODUCTION

1

Prostate cancer is the most commonly diagnosed malignancy and second leading cause of

cancer mortality for men in the United States with an estimated 217,730 new cases and

32,050 prostate cancer deaths projected for 2010 (Jemal et al. 2010). With nearly one in

six men diagnosed with this disease in their lifetime, it represents a $9.9 billion burden to

the U.S. healthcare system in 2006 (National Cancer Institute 2010).

After more than two decades of widespread serum prostate specific antigen (PSA)

testing, clinical prostate cancer has shifted to a predominantly localized disease.

However there are two key challenges in our current diagnostic and prognostic strategies

with regards to this disease. First, PSA testing is not specific to cancer – more common

and less serious conditions such as benign prostate hyperplasia and prostate infections

also increase PSA levels. For this reason, PSA testing has a false positive rate of greater

than 70% (P Finne et al. 2000). This high rate of false positives has not only been shown

to have psychological harms to patients (K Lin et al. 2008), but also leads to many

unnecessary prostate biopsies.

Second, despite the shift towards localized disease, two recent large-scale, randomized

trials of PSA screening suggest that prostate cancer is over-diagnosed and over-treated

(Andriole et al. 2009; Schröder et al. 2009). In the Andriole study, no difference in

prostate cancer mortality was observed between two groups of patients: cases comprised

of over 38,000 patients who received annual screening and an equal size control group

which received „usual care‟ outlined by their personal healthcare providers. In the

2

Schröder study, 182,000 men were either offered PSA screening every 4 years or were

offered no screening. A slight decrease in mortality rate was observed, but it was noted

that in order to save one life, 1,410 men would need to be screened and 49 patients would

need to be treated. While both studies observed higher levels of detection in the screened

group, this did not strongly correlate with reduced mortality.

This over treatment is most likely because many cancers that are detected are never

destined to progress. While some patients die of metastatic disease within 2 to 3 years of

diagnosis, with a 10-year survival rate of 91% and average age of onset of 72 years, most

patients live with a relatively indolent form of the disease and ultimately die of unrelated

causes. The broad range of clinical behavior is likely a reflection of the underlying

genomic diversity of the tumors (Taylor et al. 2010).

Previous studies of prostate tumors reported significant heterogeneity in the gene

expression profiles and genomic structural alterations including DNA copy number

changes and gene fusions involving the ETS family of transcription factors detectable in

approximately half of prostate tumors (Lapointe et al. 2004; Singh et al. 2002; Sboner et

al. 2010; Tomlins et al. 2008, 2005; King et al. 2009; Taylor et al. 2010; Robbins et al.

2011; Pflueger et al. 2011). However, exon sequencing of known oncogenes and tumor

suppressors has found few somatic mutations and the calculated background mutation

rate appears to be relatively low (Taylor et al. 2010). This suggests the presence of other

3

forms of genomic aberrations that contribute to the observed gene expression variations,

and in turn, the diversity in tumor behavior.

DNA methylation has long been suspected to play a role in tumorigenesis and cancer

progression in various tissue types (Lapeyre et al. 1981; Jones 1986; P W Laird and R

Jaenisch 1994, 1996; Samir K. Patra et al. 2002; Das and Singal 2004; Melanie Ehrlich 2002;

Manel Esteller and Herman 2002). Early studies in cancer epigenetics revealed an overall

reduction of 5-methylcytosine in various tumor genomes (Gama-Sosa et al. 1983; A P

Feinberg and Vogelstein 1983). In contrast, more recent studies identified many

hypermethylation events in CpG islands near known tumor suppressor transcriptional

start sites, which correlated with reduction in transcript levels (Brooks et al. 1998). Many

of these candidate gene-based approaches have led to discovery of potentially prognostic

DNA methylation events (Hannes M. Müller et al. 2003; Eun-Jung Kim et al. 2008).

However, recent advances in microarray and high-throughput massively-parallel

sequencing technologies have enabled investigators to study site-specific DNA

methylation events on a much broader scale. Recent studies of the DNA methylome in

colorectal cancer and glioblastomas have revealed valuable new insights into those

diseases, including the discovery of hundreds of affected genes previously not identified

(Irizarry et al. 2009; Noushmehr et al. 2010; The Cancer Genome Atlas. 2008).

In prostate cancer, hypermethylation of several tumor suppressor promoters has been well

documented. Most notably, the hypermethylation of the CpG island overlapping the

4

transcriptional start site of the GSTP1 gene has been associated with transcriptional

silencing and is described as the most common molecular alteration in prostate cancer

identified to date (X Lin et al. 2001; Woodson et al. 2008). Since GSTP1 promoter

methylation is very common and specific for prostate cancer, many investigators have

proposed using this methylation event as a diagnostic biomarker (Nakayama et al. 2004;

Cairns et al. 2001). A 2004 study looking at eight additional candidate-sites known to be

differentially methylated in other tumor types identified three gene promoters that were

specifically hypermethylated in prostate cancer (Jerónimo et al. 2004), APC, RASSF1

and CRBP1. Additionally, studies looking at DNA methyltransferases (DNMTs) and

DNMT-interacting proteins have suggested that dysregulation of these genes in prostate

cancer are responsible for the improper DNA methylation events in primary tumors and

cell lines (Hoffmann et al. 2007; Yaqinuddin et al. 2008; Ley et al. 2010; Brooks et al.

1998). These findings suggest that DNA methylation alteration is a common event in

prostate cancer. However, discovery of specific sites of alterations have been limited by

practical limitations of the throughput of quantitative DNA methylation assays that were

available at the time.

More recently, Kron et al. reported the DNA methylation profiles of 20 prostate tumors at

CpG islands across the genome using a human CpG island microarray. However, this

study did not include the profiles of normal prostate tissues and was only able to make

comparisons between the prostate tumors and six cases of age-matched lymphocytes

(Kron et al. 2009). While they were able to identify sites that were methylated or

5

unmethylated in prostate cancer, they could not examine the change in methylation states

due to their study design.

To identify DNA methylation alterations that occur in prostate cancer, I quantitatively

profiled 95 primary prostate tumors and 86 healthy prostate tissues for their DNA

methylation levels at 26,333 CpG sites in 14,104 gene promoters using the Illumina

HumanMethylation27 microarray platform. By applying existing microarray analysis

tools, I identified the differentially methylated CpGs and explored subsets of them that

accurately distinguished tumor and normal prostate tissues. Furthermore, I then

integrated available clinical data to discover novel prognostic markers of aggressive

tumors. Finally, I investigated the DNMT protein family, as well as their interacting

partners, for their role in the alteration of DNA methylation in prostate cancer.

6

CHAPTER 2

METHODS

7

Sample collection and preparation

All prostate samples used for this study were collected at the Stanford University Medical

Center between 1999 and 2007 with patient‟s informed consent under an IRB-approved

protocol. Multiple tissue samples were harvested from each prostate, flash frozen and

stored at -80°C. Sections of each prostate tissue sample were evaluated by a

genitourinary pathologist. The tumor and non-tumor areas were marked and

contaminating tissues were trimmed away from the block as described previously

(Lapointe et al. 2004). Tumor samples in which at least 90% of the epithelial cells were

cancerous, and non-tumor samples having no observable tumor epithelium, were selected

for extraction of DNA and RNA. Clinical information associated with prostate samples

included in the analysis is summarized in Supplemental Table S1.

Primary prostate cell culture and transfection assays

A primary culture of human prostatic epithelial cells (E-PZ-231) was established from

benign tissue of the peripheral zone of the prostate of a 56 year-old man who underwent

radical prostatectomy to treat prostate cancer. Using previously described methods

(Peehl 2002), primary cultures were serially passaged. When tertiary passage cells were

about 50% confluent, they were fed Complete PFMR- 4A medium (Peehl 2002) without

gentamycin until they reached ~85% confluency. Cells in each 60-mm, collagen-coated

dish were then transfected with 10 µg of plasmid DNA using Lipofectamine 2000

(Invitrogen) according to the manufacturer‟s instructions. After 48 hours, cells from

three 60-mm dishes per condition were dissociated with TrypLE Express (Invitrogen),

8

centrifuged, and snap-frozen in liquid nitrogen. These cell pellets were then used for

DNA isolation.

Nucleic acid isolation

DNA and RNA were isolated from tissue samples or cell cultures using Qiagen AllPrep

DNA/RNA mini kit (Qiagen) following the manufacturer‟s protocol, with the exception

of the RNA from primary prostate cell cultures. This RNA was isolated with Trizol

Reagent (Invitrogen) according to the manufacturer‟s instructions.

Sodium bisulfite conversion

Sodium bisulfite conversion of genomic DNA was performed using the EZ-96 DNA

Methylation Kit (Deep-Well format) (ZymoResearch). The conversion was completed

using the alternative incubation protocol for Illumina Infinium Methylation Assay, as

described by the manufacturer.

Methylation analysis by Illumina Infinium HumanMethylation27

Five hundred ng of sodium bisulfite-converted genomic DNA from patient samples or

cultured cells were assayed by Infinium HumanMethylaton27, RevB Beadchip Kits

(Illumina). The assay was performed using the protocol as described by the

manufacturer.

9

Beta score calculations, quality filtering and batch normalization

HumanMethylation27 array results were initially extracted and analyzed using Illumina

BeadStudio software with the Methylation Module v3.2. Beta scores were calculated

manually using values exported from BeadStudio. For each probe intensity value, I

subtracted the median negative background control probe value based on the color

channel. The beta score was calculated using the background subtracted intensity values

as: β = IntensityMethylated / (IntensityMethylated + IntensityUnmethylated). Any negative beta

scores were converted to a zero. Any beta scores with an associated detection p-value of

greater than 0.01 were converted to "missing values". To correct for any array-by-array

variation, I imputed all missing values using KNN (Troyanskaya et al. 2001), then

performed normalization using the ComBat R-package (W. Evan Johnson et al. 2006).

All previously imputed values were converted back to "missing values" for subsequent

analyses.

To remove CpG probes with potentially problematic hybridization, I performed BLAT on

all 27,578 probe sequences against the GRCh27/hg19 build of the human genome. One

thousand and twenty eight probes showed questionable mapping and therefore were

removed from analysis. I also identified 217 probes that included a SNP of greater than

3% minor allele frequency within 15 bp of the assayed CpG. These probes were also

rejected with consideration to potential variation in probe hybridization due to the

common SNP.

10

Clustering

Prior to each hierarchical clustering, the beta scores were mean centered. Hierarchical

clustering of the arrays was done using the software Cluster 3.0 with Average Linkage.

Because these datasets were too large to cluster the genes by Cluster 3.0, gene clustering

was done using XCluster, available through the Stanford Microarray Database (Sherlock

et al. 2001), using non-centered Pearson Correlation to perform the hierarchical

clustering.

Significance Analysis of Microarray (SAM)

Each SAM was performed as described in the software manual. The data were analyzed

using the latest version of SAM available at the time of this manuscript preparation,

which was version 3.09c. SAM was implemented using R version 2.10.0.

Prediction Analysis of Microarray (PAM)

Prior to PAM, the CpGs were sorted by standard deviation across all tumors and normals.

To improve statistical power, only CpGs which had a standard deviation of 0.04 or

greater were analyzed. PAM was performed as described in the software manual. The

data were analyzed using the latest version of PAM available at the time of this

manuscript preparation, which was version 2.11. PAM was implemented using R version

2.10.0. Based on visual examination of the training errors and the cross-validation

results, I set the shrinkage threshold to 10.5.

11

PyroMark assays

PyroMark assays were performed at the Stanford Protein and Nucleic Acid Facility using

the manufacturer‟s recommended protocol (Qiagen). For each target region, 3 primers

were used: a forward and reverse PCR primer and a sequencing primer. Primer

sequences are listed in Table 2.1

Target CpG Promoter Primer Sequence

cg19790294 CYBA Forward 5'-GTTTTTGAGTTTTTTTAGGGTTTTTTAAATT-3'

cg19790294 CYBA Reverse 5'-CCTTCACACCTTATCCTACTATTA-3'

cg19790294 CYBA Sequencing 5'-GATAAGTGTTTTTGTTTAATGT-3'

cg04448487 GDAP1L1 Forward 5'-TTTTTATTTTTGTAGGGAGTTTGA-3'

cg04448487 GDAP1L1 Reverse 5'-CTCTCTCTCCCCCAACATCACATA-3'

cg04448487 GDAP1L1 Sequencing 5'-AGGGAGTTTGATATTGAG-3'

cg02879662 HIF3A Forward 5'-GGGGTTTTTTTTTTGGAGATTT-3'

cg02879662 HIF3A Reverse 5'-CACCCCTACAATCCCTAA-3'

cg02879662 HIF3A Sequencing 5'-TTGGATTGTTGGGGG-3'

cg19853760 LGALS1 Forward 5'-TGAGGGGGGGTAGTAGTT-3'

cg19853760 LGALS1 Reverse 5'-ATCCCCACACTCACACAAA-3'

cg19853760 LGALS1 Sequencing 5'-TGATTTGTAATTGGTTGAAT-3'

cg04622802 LOC387758 Forward 5'-GGGTAATAGAGTTAGTATTTTGTTAG-3'

cg04622802 LOC387758 Reverse 5'-CCCAACAAACTTCATATAACTCTACA-3'

cg04622802 LOC387758 Sequencing 5'-AGTTAGTATTTTGTTAGGGT-3'

cg21096399 MCAM Forward 5'-TAGGTTTTTGGTTTGGGAAG-3'

cg21096399 MCAM Reverse 5'-AATCCCCTAAAAACTACATTAACT-3'

cg21096399 MCAM Sequencing 5'-GGGTAGTGATAGGTGT-3'

cg24340926 RAB33A Forward 5'-GGGTTTTTTTTATTGGTTAGTTAAAT-3'

cg24340926 RAB33A Reverse 5'-AACCCCAACATCCCCTTATCACA-3'

cg24340926 RAB33A Sequencing 5'-TTTTTATTGGTTAGTTAAATATAAT-3'

cg13102585 RPIP8 Forward 5'-GGGGATGGTTATGGAAGG-3'

cg13102585 RPIP8 Reverse 5'-ACAACCCCAAAACCATAATAATCT-3'

cg13102585 RPIP8 Sequencing 5'-GGTTATGGAAGGGTTGA-3'

cg22862656 SCGB2A2 Forward 5'-GGAATAAATAGAGTAAGGTTGGGTGTT-3'

cg22862656 SCGB2A2 Reverse 5'-ACCCCCAACATAAAAACCATCAACAACTTC-3'

cg22862656 SCGB2A2 Sequencing 5'-AGGTTGGGTGTTTATTTTTATA-3'

Table 2.1 Primer sequences used in PyroMark assays.

12

TaqMan gene expression assay

Expression levels of genes encoding several DNMT and DNMT-interacting proteins, as

well as beta-2-microglobulin as an endogenous control, were measured in 10 normal and

36 tumor samples by TaqMan Gene Expression Assay. I used the following Applied

Biosystems inventoried assays with FAM/MGD labeled probes (Assay ID in

parentheses): DNMT1 (Hs00945900_g1), DNMT3A (Hs00173377_m1), DNMT3A2

(Hs00601097_m1), DNMT3B (Hs01003405_m1), DNMT3L (Hs01081364_m1), EZH2

(Hs01016789_m1) and the Human B2M (beta-2-microglobulin) Endogenous Control.

Twenty five ng of cDNA were assayed in triplicate for each target, using the protocol as

described by the manufacturer, on the ABI PRISM 7900HT instrument. The results were

analyzed using the ABI SDS 2.4 and ABI RQ Manager 1.2.1 software. Briefly, the

average CT and delta-CT were calculated for each DNMT and EZH2. By integrating the

average CT value from the B2M CT, I calculated the delta-delta-CT. All sample delta-

delta-CT values were normalized to that of a tumor sample PC625T to generate an RQ

value. To present the RQ value as a positive value, 5 was added to each RQ value.

Expression vectors

The pcDNA3/Myc- EZH2 construct was a generous gift from A. Chinnaiyan (Okano et

al. 1999). The pcDNA3/Myc-DNMT3A, pcDNA3/Myc-DNMT3A2, pcDNA3/Myc-

DNMT3B1, pcDNA3/Myc-DNMT3B2 and pcDNA3/Myc-DNMT3B3 constructs were a

generous gift from A. Riggs (Chen et al. 2005).

13

CHAPTER 3

DNA METHYLATION PROFILES OF NORMAL

PROSTATES AND PROSTATE TUMORS

14

Prostate DNA methylation profiles

To explore the prostate DNA methylome, I profiled 95 primary prostate tumors and 86

normal prostate tissues, including 70 matched pairs, using the Illumina

HumanMethylation27 microarrays. This platform assays 27,578 CpG sites, all but 600 of

which are in the proximal promoter regions of 14,475 transcription start sites. After

batch correcting and quality filtering the data, I was able to determine quantitative

methylation status (beta scores) for 26,333 CpG sites in 14,104 promoters. To investigate

the similarities and differences of the DNA methylation profiles of the normal samples

and tumor samples, as well as their heterogeneity, I performed unsupervised hierarchical

clustering on the entire dataset (Figure 3.1). When the data were clustered by sample, I

observed two main clusters – one comprised almost entirely of normal samples (77/88)

and the other comprised almost entirely of the tumor samples (67/71). The branch

lengths in the normal sample cluster were generally shorter than the branch lengths in the

tumor sample cluster, indicating more heterogeneity in methylation profiles among the

tumor samples. Twenty-two of the samples did not fall into either of the two main

clusters and formed long off-shooting branches or small clusters. Eighteen of these were

tumor samples, further indicative of the heterogeneous nature of the tumor DNA

methylome. By visual inspection, the majority of the samples showed relatively little

methylation change between the tumor and normal clusters (Figure 3.1), and most of

these invariable CpG sites showed low levels of methylation in both normal and tumor

samples. However, there were distinct CpG clusters with methylation patterns that

distinguished the normal or tumor sample clusters, and, strikingly, a large number of CpG

sites showed increased methylation in the tumor cluster compared to the normal cluster.

15

Figure 3.1: Hierarchical clustering of prostate tissues by DNA methylation.

Unsupervised hierarchical clustering of 181 prostate tissues and 26,333 CpGs, by sample

and by CpG. Red branches represent tumor samples and blue branches represent normal

samples. Red pixels represent high DNA methylation while green pixels represent low

DNA methylation. As indicated in the Methods section, the beta scores were mean

centered prior to clustering.

16

To identify the CpG sites with statistically different DNA methylation status between

normal prostate tissues and tumors, I performed a two-class Significance Analysis of

Microarrays (SAM) (Tusher et al. 2001). As I had matched normal tissues for only 70 of

the 95 tumors used in this study, I conducted the SAM analysis as unpaired. The analysis

identified 5,912 CpG sites hypermethylated in tumors compared to normal tissues, and

2,151 CpG sites hypomethylated at FDR < 0.8% (Figure 3.2). This corresponds to 4,224

and 1,792 promoters, respectively. I performed hierarchical clustering on all samples

based on these 8,063 differentially methylated CpG sites (Figure 3.3). Of the 11,116

gene promoters represented by two or more CpG sites, only 223 had opposite methylation

effects (i.e., at least one hypermethylated CpG and at least one hypomethylated CpG).

When the distances from transcriptional start sites were compared in these 224 promoters

with opposite methylation effects, I saw enrichment for hypermethylated CpGs in the -

100 bp to +800 bp range, whereas I saw enrichment for the hypomethylated CpGs in the -

700 bp to -200 bp range. Thus, overall, nearly one third (8,063/26,333) of assayed

promoter CpGs had a statistically significant change in DNA methylation, with most of

those showing an increase in methylation. Interestingly, 43% (6,015/14,104) of all gene

promoters assayed had at least one CpG with a tumor-specific methylation change.

17

Figure 3.2: Normal vs Tumor unpaired 2-class SAM analysis of the 181 prostate

samples. False discovery rate of 0.78% resulted in 8,063 differentially methylated CpGs

including 5,912 hypermethylated CpGs (red) and 2,151 hypomethylated CpGs (green).

Figure 3.3: Differentially methylated CpGs of prostate tumors. Unsupervised

hierarchical clustering of 181 prostate tissues based on the 5,912 and 2,151 CpG sites

hypermethylated and hypomethylated in prostate tumors, respectively. Red branches

represent tumor samples and blue branches represent normal samples. Red pixels

represent high DNA methylation while green pixels represent low DNA methylation.

18

As the SAM analysis was unpaired, there is the possibility that inter-individual DNA

methylation differences could be a possible confounding factor. The relative

homogeneity of the normal samples, indicated by the short branch lengths and tight

clustering compared to the tumor samples, suggested that this was likely not a problem,

but it was still a concern I chose to address. I thus also conducted a two-class paired

SAM analysis on only the 70 matched sample pairs, identifying 5,556 hypermethylated

and 2,185 hypomethylated CpGs at a similar significance cutoff (FDR > 0.84%) (Figure

3.4). This paired analysis identified only 306 novel, differentially methylated CpGs that

were not discovered in the unpaired analysis. Furthermore, when the list of differentially

methylated CpGs was ranked by significance, most of the CpGs uniquely identified in the

paired analysis (278/306) fell in the bottom 25%. These data indicate that the prostate

tumor methylation signal is strong enough to overcome any inter-individual methylation

differences for the vast majority of CpGs assayed on this platform. For all subsequent

analyses, I used the unpaired SAM results to include all samples.

While the impact of inter-individual methylation differences was minimal for the purpose

of distinguishing tumor and normal samples, it was clear that such differences were

present. I chose to explore this phenomenon in the normal prostate samples using

available patient data. I performed a Quantitative SAM analysis to identify CpGs that

showed differential methylation relative to the age of the patient at the time of surgery

(range: 44-74 yrs). This analysis revealed 749 CpG sites that showed increasing

methylation with age, while no CpG showed decreasing methylation with age (FDR <

4.82%) (Figure 3.5 and Figure 3.6). When I repeated the age-dependent Quantitative

19

SAM analysis with tumor samples, no CpG showed correlation with age, despite the fact

that the majority of these came from the same patients as the normal samples. This

strongly suggests that the age-dependent DNA methylation pattern that I observe is

overridden by the alterations that occur in cancer.

Figure 3.4: Normal vs Tumor paired 2-class SAM analysis of the 181 prostate

samples. False discovery rate of 0.84% resulted in 7,741 differentially methylated CpGs

including 5,556 hypermethylated CpGs (red) and 2,185 hypomethylated CpGs (green).

Figure 3.5: Quantitative SAM analysis of 86 prostate samples based on age of

patient at the time of surgery. False discovery rate of 4.82% resulted in 749

differentially methylated CpGs, all hypermethylated (red).

20

Figure 3.6: 749-Age dependent CpGs. 749 age-dependent CpGs, clustered by CpG,

patients ordered by age.

Patient Age

44 74

21

Diagnostic methylation markers

Among the CpG sites that I found to be differentially methylated in tumor versus normal

prostate tissues by SAM, and shown clustered in Figure 3.3, were several sites that had

been previously characterized in prostate tumors, most notably several CpG sites near or

within the GSTP1 gene. Hypermethylation of the CpG island overlapping the

transcriptional start site of the GSTP1 gene has been associated with transcriptional

silencing and is described as the most common molecular alteration in prostate cancer

identified to date (X Lin et al. 2001; Woodson et al. 2008). Since GSTP1 promoter

methylation is very common and specific for prostate cancer, many investigators have

proposed using this methylation event as a diagnostic biomarker for prostate cancer

(Nakayama et al. 2004; Cairns et al. 2001). The HumanMethylation27 arrays contain

seven CpG sites in the GSTP1 promoter. Five of these sites showed significantly

increased DNA methylation in tumors, four of which are located in the promoter CpG

island that had been previously characterized as a site of hypermethylation in prostate

cancer (Brooks et al. 1998), while the fifth lies 88 bp downstream of the annotated CpG

island boundary (red circles in Figure 3.7A). The two remaining CpGs showed either no

differential methylation (gray circle in Figure 3.7A) or slight but statistically significant

hypomethylation (green circle in Figure 3.7A); both lie further upstream of the

transcriptional start site, outside of the promoter CpG island. Our data not only confirm

the previously described hypermethylation of the GSTP1 promoter CpG island, but also

show that CpG DNA methylation alteration is highly context dependent even within a

single promoter.

22

Figure 3.7: GSTP1 CpG island hypermethylation in prostate tumors. (A) Diagram

of the GSTP1 gene. Blue boxes represent the RefSeq annotation of the GSTP1 gene. The

green box represents a CpG island calculated by UCSC Genome Browser. Circles are

CpG sites assayed by HumanMethylation27: red circles represent probes that were

identified to be hypermethylated in prostate tumors by 2-class SAM, the green circle

represents a probe that was hypomethylated, and the gray circle represents a probe that

showed no significant change. The numbers below the circles indicate the relative

distance in base pairs from the predicted TSS. (B) Heatmap depicts DNA methylation

pattern of the 7 probes near GSTP1. The dendrogram is based on the hierarchical

clustering from Figure 2. Red branches represent tumor samples and blue branches

represent normal samples. Coordinates are based on NCBI36/hg18 human genome

assembly.

A

-831 -566 -89 +206 +543 +721 +976

23

In addition to GSTP1, I also examined our data specifically for methylation changes in

the promoters of APC and RASSF1, which have also been previously shown to have

hypermethylation in prostate cancer (Jerónimo et al. 2004) and were represented by

multiple probes on the HumanMethylation27 array. With APC, all six CpG sites

represented on the array showed hypermethylation in tumors, located 122 bp upstream to

488 bp downstream of the TSS (Figure 3.8). With RASSF1, three CpGs sites were

probed, located 58 bp upstream to 176 bp downstream of the TSS and within a CpG

island boundary; all three were hypermethylated (Figure 3.9). However, five of the six

probes located more than 2 kb downstream of the TSS in a second CpG island did not

show differential methylation.

A

+15 +131

-53

-122 +214 +488

24

Figur 3.8: APC proximal promoter hypermethylation in prostate tumors. (A)

Diagram of the APC gene. Blue boxes represent the RefSeq annotation of the APC gene.

There are no CpG islands in this window, calculated by the UCSC Genome Browser.

Circles are CpG sites assayed by HumanMethylation27: red circles represent probes that

were identified to be hypermethylated in prostate tumors by 2-class SAM. The numbers

above and below the circles indicate the relative distance in base pairs from the predicted

TSS. (B) Heatmap depicts DNA methylation pattern of the 6 probes near APC. The

dendrogram is based on the hierarchical clustering from Figure 2. Red branches

represent tumor samples and blue branches represent normal samples. Coordinates are

based on NCBI36/hg18 human genome assembly.

A

+176

-46

-58 +2908

+2693 +3115 +3697

+3438 +4063

25

Figure 3.9: RASSF1 proximal promoter hypermethylation in prostate tumors. (A)

Diagram of the RASSF1 gene. Blue boxes represent the RefSeq annotation of the

RASSF1 gene. Green boxes represent the CpG islands calculated by UCSC Genome

Browser. Circles are CpG sites assayed by HumanMethylation27: red circles represent

probes that were identified to be hypermethylated in prostate tumors by 2-class SAM and

the gray circles represent probes that showed no significant change. The numbers above

and below the circles indicate the relative distance in basepairs from the predicted TSS.

(B) Heatmap depicts DNA methylation pattern of the 9 probes near RASSF1. The

dendrogram is based on the hierarchical clustering from Figure 2. Red branches

represent tumor samples and blue branches represent normal samples. Coordinates are

based on NCBI36/hg18 human genome assembly.

While hierarchical clustering of samples using the most differentially methylated CpG

sites (the set shown in Figure 3.3) was able to distinguish most tumors from normal

tissues, the classification was not perfect, as indicated by the inclusion of normal tissue

samples within the tumor cluster and vice versa. To identify CpG sites that could best

predict either the tumor state or the normal state, I performed a Prediction Analysis of

Microarrays (PAM), to perform sample classification (Robert Tibshirani et al. 2002).

This analysis generated a list of 87 predictive CpG sites, most of which had increased

methylation in the tumor samples (83/87), and represented 82 gene promoters total

(Figure 3.10). The CYBA, GSTP1, KLK10, PPT2 and CXCL1 promoters each had two

CpGs represented in this list. Notably, in this ranked list of 87 predictive methylation

alterations, the GSTP1 hypermethylation was ranked 57th (Supplementary Table S2).

Thus I have identified 56 molecular events, most of which had not been previously

characterized, that are better markers of prostate cancer than is GSTP1.

26

Figure 3.10: Diagnostic markers of prostate cancer identified by PAM.

Unsupervised hierarchical clustering of 181 prostate samples based on the 87 diagnostic

CpG sites identified by PAM. Red branches represent tumor samples and blue branches

represent normal samples. Red pixels represent high DNA methylation while green

pixels represent low DNA methylation.

Validation by PyroMark sequencing

To validate the Prediction Analysis results and to validate the Illumina

HumanMethylation27 platform, I designed PyroMark assays for a subset of our

predictive CpGs. This pyrosequencing method provides quantitative percent methylation

for regions up to about 120 bp in length. I designed primers that would enable us to

assay nine diagnostic CpGs from the promoters of the RAB33A, HIF3A, GDAP1L1,

27

MCAM, LGALS1, RPIP8, CYBA, SCGB2A2, and LOC387758 genes, which were

selected from the top 40 most diagnostic CpGs identified in the PAM analysis. I selected

DNA from 12 matched tumor-normal pairs, previously assayed by HumanMethylation27,

treated them with bisulfite and performed DNA sequencing. When the percent

methylation readout from PyroMark and HumanMethylation27 of the same loci were

compared, I observed a striking correlation between the two platforms at all nine CpGs

(r2 values from 0.89 to 0.98) (Figure 3.11). While both PyroMark and

HumanMethylation27 both rely on bisulfite conversion, the subsequent chemistry

involved in quantifying methylation levels is substantially different. The strong

correlation between the two platforms not only indicated a high level of accuracy in

methylation quantification by both platforms, but also validated differential methylation

at the nine predictive CpG sites.

C

28

29

Figure 3.11: PyroMark validates HumanMethylation27 results. PyroMark

sequencing results compared to HumanMethylation27 beta scores at 9 diagnostic CpGs

identified by PAM. Blue circles are normal samples and red circles are tumor samples.

Y-axis: fraction methylation calculated from PyroMark. X-axis: fraction methylation

calculated from HumanMethylation27 (beta scores). Black line: linear regression. (A)

CYBA (cg19790294). (B) GDAP1L1 (cg04448487). (C) HIF3A (cg02879662). (D)

LGLS1 (cg19853760). (E) LOC387758 (cg04622802). (F) MCAM (cg21096399). (G)

RPIP8 (cg13102585). (H) RAB33A (cg24340926). (I) SCGB2A2 (cg22862656).

Because the PyroMark results provided sequence lengths of up to 120 bp, I was able to

explore the methylation states of additional CpGs near each of the nine predictive CpGs.

Each region contained 4 to 13 additional CpGs that could be measured by the PyroMark

assay. In most cases, neighboring CpGs within a region had similar methylation levels

(Figure 3.12A – 3.12I). However, there were several notable exceptions where I

observed dramatically different levels of methylation within these short regions. In the

MCAM region (Figure 3.11F), one CpG had 20% methylation while another CpG only

85 bp away had 90% methylation. Similarly, the RPIP8 region (Figure 3.12G), had a

CpG with 10% methylation and another CpG 54 bp away had as high as 80%

methylation. In the most extreme case, two SCGB2A2 region CpGs, only 2 bp apart, had

30

a difference in methylation of 40% (Figure 3.12I). It is clear from these plots that the

differences in methylation between tumor samples and normal samples were variable

within a single region. Even though I selected these nine regions because a single CpG in

each was among the most differentially methylated in the HumanMethylation27, if

another neighboring CpG in the same region had been assayed on the array, it may have

not been called substantially different between tumor and normal tissues. The most

striking example of this is in the RPIP8 region (Figure 3.12G), where a CpG 72 bp away

from the CpG assayed on HumanMethylation27 showed no difference between tumor

samples and normal samples. All the assayed CpGs were screened against dbSNP (build

130) for possible influence on the PyroMark readout by sequence variation. No SNPs

were found in these regions that could confound the PyroMark methylation

quantification.

31

32

33

Figure 3.12: Comparison of neighboring CpGs by PyroMark. PyroMark sequencing

results comparing neighboring CpGs of the 9 diagnostic CpGs identified by PAM. Each

diamond represents a CpG methylation level for an individual sample. Lines connect

CpGs from each sample. Blue lines are normal samples, red lines are tumor samples. Y-

axis: fraction methylation calculated from PyroMark. X-axis: relative coordinates in

basepairs. Box indicates CpG assayed by HumanMethylation27. (A) CYBA

(cg19790294). (B) GDAP1L1 (cg04448487). (C) HIF3A (cg02879662). (D) LGLS1

(cg19853760). (E) LOC387758 (cg04622802). (F) MCAM (cg21096399). (G) RPIP8

(cg13102585). (H) RAB33A (cg24340926). (I) SCGB2A2 (cg22862656).

34

Prognostic methylation markers

To explore tumor heterogeneity, I compared the methylation profiles of the 86 tumors

with respect to Gleason grade and time to biochemical recurrence (defined as serum PSA

> 0.07 ng/mL after surgery) of the donors. Gleason grade is a powerful predictor of

treatment failure, tumor progression and death from prostate cancer, and biochemical

recurrence has also been correlated with prostate cancer-specific mortality (Freedland et

al. 2005). I conducted a multiclass SAM in an effort to identify methylation events that

distinguished tumors of different Gleason grades, but was unable to identify such events.

Next, I conducted a SAM survival analysis with the time to biochemical recurrence as the

survival variable. With a false discovery rate of 26.7%, I identified six CpGs that showed

greater methylation in tumors from men who had shorter time to recurrence and 63 CpGs

that showed lower methylation in patients with shorter time to recurrence (Supplementary

Table S3). This strong bias towards lower methylation in aggressive tumors was striking

as I observed a bias for CpG sites with increased methylation in the tumor/normal

comparison. While I was only able to identify a small number of CpGs whose

methylation state correlated with time to recurrence, I noted that several of these CpG

sites are in the proximal promoter genes of known cancer-related genes, including 3

CpGs near MAGE gene family members which encode for strictly tumor-specific

antigens (Chomez et al. 2001) and 4 CpGs near WT1, a transcription factor gene

associated with Wilm's tumor.

35

CHAPTER 4

DNA METHYLTRANSFERASES IN PROSTATE

36

Correlation of tumor hypermethylation with DNA methyltransferase expression

With nearly one third of assayed CpGs showing changes in DNA methylation between

tumor and normal samples, I hypothesized that one or more of the DNA

methyltransferases (DNMTs), or a protein that interacts with a DNMT, had altered

activity, possibly due to changes in transcript abundance, in prostate tumors. Such

alterations in activity could in turn lead to global DNA methylation changes. To test this

hypothesis, I selected RNA from 10 of the normal and 36 of the tumor samples, and

measured the transcript abundance of DNMT1, DNMT3A, DNMT3A2, DNMT3B,

DNMT3L and EZH2 using the TaqMan Gene Expression assay. These genes comprise

the known maintenance methyltransferase (DNMT1) (Chuang et al. 1997), all known

methyltransferases with de novo capability [DNMT1 (Estève et al. 2005), DNMT3A

(Okano et al. 1999), DNMT3B (Okano et al. 1999)], and two interacting proteins thought

to target methyltransferases to specific genomic regions [DNMT3L (El-Maarri et al.

2009) and EZH2 (Okano et al. 1999)]. In addition, I uniquely assayed DNMT3A and its

alternative promoter variant DNMT3A2 by using transcript-specific primers and probes.

While several splice variants of DNMT3B have been characterized, I was unable to

design variant-specific primers and probes for them, so I instead designed primers and

probes to the common region of all DNMT3B variants. I did not observe detectable levels

of DNMT3L transcript abundance from either tumor or normal samples (data not shown).

When the transcript levels of the remaining genes were compared between normal and

tumor samples with a two-tailed t-test, three showed significant changes: DNMT3A2 (P =

0.0013), DNMT3B (P = 0.024) and EZH2 (P = 0.026), while DNMT1 and DNMT3A did

not (Figure 4.1A).

37

38

Figure 4.1: Expression of DNMTs and EZH2 correlates with global

hypermethylation in prostate tumors. Comparison of transcript levels of DNMTs and

EZH2 measured by TaqMan qPCR with the average DNA methylation levels of CpG

sites that are hypermethylated in prostate tumors. Blue circles are normal samples and

red circles are tumor samples. P-value was calculated by linear regression analysis. Y-

axis: average DNA methylation levels (beta score). X-axis: relative gene expression

levels [log2(RQ)]. Black line: linear regression. (A) DNMT1 expression. (B) DNMT3A

expression. (C) DNMT3A2 expression. (D) DNMT3B expression. (E) EZH2 expression.

(F) Comparison of DNMT and EZH2 transcript levels between normal tissues (blue) and

tumors (red). Significant differences are indicated by asterisks; P values were calculated

by t-test. Standard errors are depicted by error bars. Y-axis: relative gene expression

levels [log2(RQ)].

I compared the expression values for these five genes to global DNA methylation levels.

Specifically, I plotted the mean percent methylation of all 5,912 hypermethylated CpG

sites against relative expression of each methyltransferase or interacting protein, and

calculated regression and the goodness-of-fit of the regression for each sample. Again,

DNMT3A2 (r2

= 0.272, P = 0.0031), DNMT3B (r2

= 0.197, P = 0.0056) and EZH2 (r2

=

0.211, P = 0.0037) all showed significant correlation between expression and global

hypermethylation, while DNMT1 and DNMT3A did not (Figure 4.1B – 4.1F). The

correlation between DNMT3A2, DNMT3B and EZH2 expression and global

hypermethylation, in conjunction with the observed over-expression of the same genes in

tumors, suggests a possible causal role in the global methylation changes seen in prostate

tumor.

39

DNMT overexpression recapitulates hypermethylation events seen in prostate

tumors

To determine whether the increased transcript abundance of DNMT3A2, DNMT3B and

EZH2 in tumor cells has a causal role in the hypermethylation of a large number of

promoter CpGs, I expressed these genes from the CMV promoter in transient transfection

assays in primary cultures of normal prostatic epithelial cells. I used plasmids expressing

DNMT3A, DNMT3A2, DNMT3B1, DNMT3B2, and DNMT3B3, an EZH2-cDNA plasmid,

and a no-insert plasmid. I co-transfected each cDNA plasmid with the no-insert plasmid,

and independently with the EZH2 plasmid, and also included a mock no-insert plasmid

only transfection. I calculated the change in DNA methylation for each CpG between

each cDNA transfection and the mock transfection after 48 hours. I then plotted the ideal

cumulative distribution function of the DNA methylation level change at all 26,333 CpG

sites along with the empirical cumulative distribution function of just the changes at the

5,912 CpG sites hypermethylated in tumors (Figure 4.2A – 4.2K), and tested the

difference in the two distribution functions using the Kolmogorov-Smirnov (K-S) test. In

all eleven experimental transfections, the distribution of the 5,912 CpG sites was

significantly enriched compared to the null: DNMT3A (P = 6.0E-45), DNMT3A2 (P =

3.5E-62), DNMT3B1 (P = 1.2E-31), DNMT3B2 (P = 5.2E-39), DNMT3B3 (P = 4.6E-44),

EZH2 (P = 1.1E-59), DNMT3A+EZH2 (P = 7.8E-64), DNMT3A2+EZH2 (P = 9.8E-65),

DNMT3B1+EZH2 (P = 2.1E-29), DNMT3B2+EZH2 (P = 6.7E-42), DNMT3B3+EZH2 (P

= 2.5E-67). Consistent with our hypothesis, when the plots of the empirical cumulative

distribution functions were visually inspected, I observed that the low P-value of the K-S

test appeared to be driven more by the CpGs of increased methylation rather than CpGs

of decreased methylation in all eleven conditions.

40

41

42

Figure 4.2: Overexpression of DNMTs and EZH2 results in increased methylation

at a subset of prostate tumor hypermethylation sites. Ideal (black) and empirical (red)

cumulative distribution functions of change in DNA methylation after DNMT or EZH2

transfection into cultured normal prostate cells. The empirical distribution functions are

based on the 5,912 CpGs that were hypermethylated in prostate tumors, while the ideal

distribution functions are based on all 26,333 CpGs assayed on the array.

Overexpression of (A) DNMT3A, (B) DNMT3A2, (C) DNMT3B1, (D) DNMT3B2, (E)

DNMT3B3, (F) EZH2, (G) DNMT3A and EZH2, (H) DNMT3A2 and EZH2, (I)

DNMT3B1 and EZH2, (J) DNMT3B2 and EZH2, and (K) DNMT3B3 and EZH2.

To test specifically whether the list of 5,912 CpG sites was statistically enriched for

CpGs with substantially increased DNA methylation, I performed a series of chi-square

tests. Based on the distribution of CpG methylation levels in tumor and normal tissues at

these CpG sites, I set a cutoff value of 0.05. In other words, CpG sites where the

methylation increased by 5 percent or greater in the experimental transfection compared

to the mock transfection were considered to have substantially increased DNA

methylation. I calculated expected values based on the distribution of these CpGs with

substantially increased DNA methylation in the entire set of 26,333 CpGs. When chi-

square tests were performed, all eleven experimental conditions had very low p-values:

DNMT3A (P = 1.1E-45), DNMT3A2 (P = 1.7E-66), DNMT3B1 (P = 8.9E-127),

DNMT3B2 (P = 1.8E-157), DNMT3B3 (P = 6.6E-10), EZH2 (P = 9.4E-31),

DNMT3A+EZH2 (P = 1.5E-13), DNMT3A2+EZH2 (P = 1.1E-11), DNMT3B1+EZH2 (P

= 1.9E-185), DNMT3B2+EZH2 (P = 9.4E-107), DNMT3B3+EZH2 (P = 2.3E-68).

Again, DNMT3B1 and DNMT3B2, which are alternative splicing isoforms differing by

the presence of one exon, both in the presence and absence of EZH2 co-transfection,

showed the lowest P-values, all less than 1E-100. From these data, I conclude that our

list of 5,912 CpGs is indeed enriched for CpGs with substantially increased methylation

43

when DNMTs or EZH2 were overexpressed, with DNMT3B1 and DNMT3B2 appearing

to have the strongest impact on the DNA methylation levels at these sites.

Based on these data, I further investigated the altered DNA methylation in the

DNMT3B1 and DNMT3B2 overexpression experiments. Because these splice isoforms

differ by only one exon coding for 21 amino acids in a linker region (Sakai et al. 2004), I

suspected that they would share many targets. To identify the CpGs targeted by

DNMT3B1 and DNMT3B2 in prostate tumors, I examined the list of CpGs that were

hypermethylated in prostate tumors and in the overexpression experiments. Specifically,

I looked for overlaps in the list of CpGs with 5% or greater increase in methylation

compared to the mock in the DNMT3B1 (1267 CpGs), DNMT3B1+EZH2 (1322 CpGs),

DNMT3B2 (1261 CpGs), and DNMT3B2+EZH2 (1235 CpGs) overexpression

experiments. Four hundred and thirty eight CpGs were represented in all 4 lists and an

additional 425 CpGs were represented in 3 of the 4 lists. I performed two permutation

tests to determine the likelihood of our results. In the first permutation test, I generated 4

lists of CpGs (1267, 1322, 1261 and 1235 CpGs, respectively) drawn randomly from the

whole list of 26,333 CpGs and counted the number of incidences where there was an

overlap of 438 CpGs in all 4 lists. It was never observed in the 10,000 iterations. In our

second permutation test, I repeated the first permutation test but changed the criteria to

observing at least 863 CpGs overlapping in 3 of the 4 lists. This too was never observed

in 10,000 iterations. This provided further evidence that the differentially methylated

CpGs in the DNMT3B1 and DNMT3B2 overexpression experiments indeed significantly

deviated from random sampling, and are likely to be those that are specifically, directly

or indirectly, targeted by these methyltransferases.

44

CHAPTER 5

DISCUSSION

45

DNA methylation changes in prostate cancer and its potential as diagnostic and

prognostic biomarkers

Alterations in DNA methylation have been shown to play a role in tumorigenesis and

cancer progression in many malignancies, including prostate cancer. Until recently,

technical limitations have restricted these findings to either characterization of a handful

of candidate loci or of overall abundance of 5-methylcytosine in the genome. Here, I

present quantitative DNA methylation levels at more than 26,000 loci across 14,000 gene

promoters. Because I assayed 95 cancers and 86 normal prostate tissues in parallel at

CpGs specifically enriched at gene promoters, I was able to show that 43% of gene

promoters represented in our assay had a tumor-specific methylation change. In addition

to confirming methylation changes seen in previously published candidate loci studies, I

also identified thousands of novel changes, including a set of hypermethylated loci more

strongly predictive of prostate cancer than GSTP1. Our data show that DNA methylation

changes in prostate cancer occur on a broad scale, at many loci throughout the genome.

DNA methylation alteration has been observed in early cancers and precursor lesions

suggesting that methylation changes drive malignant initiation rather than tumor

progression (Baylin et al. 2001; Belinsky et al. 1998; Guerrero-Preston et al. 2009;

Brooks et al. 1998). Our observations are largely consistent with this hypothesis. If the

acquisition of DNA methylation alterations continues throughout tumor progression,

variation in methylation profiles should be observed in tumors of different histological

grades and clinical outcomes. Although I detected more heterogeneity among tumors

than among normal tissues, the vast majority of tumors fell in a single cluster and I did

46

not observe obvious subclassifications, though some tumor samples did cluster with

normal samples. I compared clinical outcomes of the donors of the tumors that clustered

with normal tissues against the donors of the other tumors but did not observe any

differences in Gleason grades or time-to-recurrence (data not show). However, from the

little inter-tumor heterogeneity that did exist, I identified several dozen DNA methylation

changes that correlated with patients' time-to-recurrence.

Both the diagnostic and prognostic markers identified in this study have the potential to

be clinically useful. Following the identification of GSTP1 hypermethylation in prostate

cancer, it was suggested that DNA methylation alterations in prostate cells can be

detected from patient urine, semen or blood (Goessl et al. 2001). DNA methylation,

unlike RNA transcript level changes or protein abundance, is a stable marker making

them an ideal as targets of clinical testing. Furthermore because we saw tumor-specific

acquisition and loss of methylation marks, diagnosis may be made simply by the presence

or absence of certain methylation events rather than looking for fold change in

abundance, and therefore is not particularly dependent on patients‟ background levels.

By developing urine, semen or blood-based test utilizing these DNA methylation changes

could potentially supplant PSA testing and/or reduce the number of biopsies performed.

Potentially more importantly, if a DNA methylation test can accurately predict the course

of prostate tumor progression, the number of unnecessary prostatectomies can be

drastically reduced. However, before either of these tests can be developed, it would be

necessary to repeat our findings in an independent replication sample set.

47

Mechanisms of DNA methylation alterations

The fact that I observed changes at a very specific subset of CpG sites across most

tumors, rather than a global DNA methylation deregulation or instability, suggests a

common mechanism among prostate cancers. This specificity in target sites was

particularly apparent in gene promoters assayed by multiple probes and by the PyroMark

assay (Supplemental Text S2). The case of GSTP1 illustrates this point well, where the

methylation changes were highly context dependent: only the CpG island overlapping the

transcriptional start site was hypermethylated. Based on these findings, I suspect that

cellular processes involved with targeted CpG methylation regulation are themselves

misregulated or altered in early tumor initiation. The most likely candidates are DNMTs

and DNMT-interacting proteins. In support of this hypothesis, I observed significant

correlations between the gene expression levels and levels of global hypermethylation for

several of these candidates. In vitro experiments in normal prostatic epithelial cells

confirmed that overexpression of DNMT3B1 and DNMT3B2 leads to the

hypermethylation of a subset of the prostate tumor-specific changes. These data, together

with previous observations, strongly suggests that dysregulation of DNMTs and possibly

DNMT- interacting proteins are among the earliest events in tumorigenesis.

While I did not address the mechanism for the observed decreased methylation of some

CpGs in tumors, there are four likely possibilities. First, there may be aberrations in the

maintenance DNA methyltransferase gene, DNMT1. Although I did not observe a

decrease in the DNMT1 transcript level, there may be translational dysregulation of this

gene or mutations that leads to decreased activity. A decrease in DNMT1 activity may

48

lead to improper maintenance and gradual loss of methylation with every DNA

replication. However, this would likely lead to a global loss rather than targeted loss at

particular CpGs, and therefore, is the least likely scenario. A second possibility is the

dysregulation of a direct or indirect DNA demethylase. While there have been a few

reports of such enzymes in mammalian cells, none has been conclusive and their

existence is still speculative (Bhutani et al. 2010; Iyer et al. 2009; Okada et al. 2010). A

third possibility is that the targeted hypomethylation may be the result of dysregulation of

an interacting protein of DNMT1 or the hypothetical DNA demethylase. With more than

twenty DNMT1-interacting proteins already identified, it is conceivable that one or more

of them are involved in DNMT1 targeting. Finally, there could be a chromatin level

rearrangement that is influencing DNA accessibility. It could be that an active

demethylase gains access to previously inaccessibly regions after chromatin remodeling.

To further interogate these possibilities, a better understanding of the DNA demethylation

is needed.

Conclusions and future directions

By approaching DNA methylation in cancer from a genomic perspective, I was able to

gain new insights into the underlying biology of prostate cancer, as well as discover

novel markers for more accurate diagnosis of the disease. However, our study was

limited in scale by technology and practicality: with only 26,000 assayed CpGs, mostly

biased towards gene promoters, it is likely that these results are not representative of the

28 million CpGs found in the human genome. The vast majority of sites targeted by the

Illumina HumanMethylation27 microarrays were chosen because of their proximity to an

49

annotated transcription start site and CpG islands. Because ectopic DNA methylation at

CpG islands has been thought to silence nearby genes (Bird 2002; Jones and Baylin

2002), this was a reasonable approach to select CpGs to be assayed. However this

platform assays only about half of all CpG islands. Even among these assayed CpG

islands, the PyroMark assay revealed some degree of variability in methylation levels

(Figure 3.11). This suggests that because of the low coverage of about 2 CpGs per

island, the methylation levels I observed is not representative of the whole CpG island.

This is particularly significant as recent studies have suggested that the edges of and

regions adjacent to CpG islands, or CpG island shores, may be the most functional region

in terms of gene regulation by DNA methylation (Irizarry et al. 2009). These

observations warrant a more detailed look at global DNA methylation patterns.

Since the time the experiments detailed here was completed, Illumina released the 2nd

generation of these arrays which covers approximately 450,000 CpGs, including those in

CpG islands, CpG island shores, miRNA promoter regions, intergenic regions and

additional gene promoters not covered by the HumanMethylation27, while maintainig

90% of the sites assayed by the HumanMethylation27 platform. This method would be

ideal for future studies intended to study CpG sites that are likely to have a functional

impact based on our current assumptions of DNA methylation. In addition to the

advantages of the targeted approach of the HumanMethylation27, this

HumanMethylation450 platform would assay additional regions and provide greater

depth of coverage at each target region.

50

Alternative to HumenMethylation450, researchers have developed several high-

throughput sequencing based methods for assaying genome wide DNA methylation

(Brunner et al. 2009; Meissner et al. 2005). In particular, the Reduce Representation

Bisulfite Sequencing (RRBS) method developed by Meissner et al. can provide

quantitative DNA methylation profile of several hundred thousand to a couple million

CpGs. While these sequencing-based approaches are more time consuming and

expensive, they provide DNA methylation information in a more agnostic way – that is,

these sites will not be limited to those hand-picked based on genome annotation.

However, that is not to say that these methods aren‟t without bias. By design, both

methods are biased towards CpG-rich regions of the genome and can only interrogate

regions bound by two MspI recognition sequence (CCGG).

While both HumanMethylation450 and RRBS would allow for a more thorough

exploration of the prostate genome, they have different advantages and disadvantages.

HumanMethylation450 is more likely to identify potentially functional methylation

changes but are limited to those sites selected by Illunina. RRBS allows for an agnostic

scan of the DNA methylation profiles, but are still restricted to specific regions of the

genome and is more expensive. For the immediate future, either of these approaches

could significantly improve the quality of the DNA methylation profiles I depicted in this

study and the appropriate method should be chosen based on these differences. However,

I speculate that as the cost of sequencing rapidly drops and third generation sequencing

technologies, including those that can directly measure DNA methylation, becomes more

widely available, researchers will soon be able to affordably perform whole genome

51

DNA methylation sequencing on large sample sets. This would dramatically improve our

understanding of DNA methylation alterations that occur in cancer.

Beyond DNA methylation, recent success in integrative analysis of copy-number

variation (CNV) and gene expression data highlights the great value in studying prostate

cancer from multiple perspectives (Taylor et al. 2010). Expanding such an integrative

analysis to include DNA methylation data along with gene expression and CNV data is

likely to lead to a better understanding of prostate cancer biology.

Finally, our quantitative methylation analyses revealed a wide spectrum of methylation

states (beta scores) rather than the expected binary states of “methylated” or

“unmethylated.” This was true in both tumor and normal tissues. This indicates cell

population heterogeneity, despite careful dissection of the sample by a pathologist based

on histological characteristics, at the molecular level. A careful investigation of this

population heterogeneity may lead to not only a more detailed picture of DNA

methylation alteration, but also a better understanding of tumor progression such as

mutually exclusive or obligate co-occurring events.

52

APENDIX

SUPPLEMENTARY TABLES

53

Supplementary Table S1: Clinical information associated with prostate samples.

PC# Age Pre-treatment PSA Path Gr

(Gleason) Months

followed-up Recurrence Days to

recurrence

15 43 5.12 3+3 90.2 None -

19 57 4.42 4+5 87.2 Biochemical 574

21 61 7.82 3+4 88.4 None -

22 53 4.64 3+3 85.3 None -

26 62 4.32 3+4 81.6 Biochemical 1809

37 50 3.2 4+3 85.3 Biochemical 1217

45 56 6.58 3+4 91.8 None -

47 51 9.92 3+4 64.7 Biochemical 1128

83 64 7.92 4+5 31.2 None -

84 71 8.48 3+4 80.9 None -

85 47 4.8 3+4 70 None -

86 65 2.1 3+4 69 None -

88 53 5.31 3+3 77.8 None -

92 69 4.51 3+4 74.5 None -

97 53 4.1 4+3 85.8 Biochemical 1834

99 73 6.26 3+4 83.6 None -

100 67 4.4 4+3 83.1 None -

103 61 5.08 4+4 87.5 Biochemical 177

111 57 44.46 3+4 61.4 None -

159 47 2.7 3+4 79.4 None -

163 65 4.2 4+5 62.7 None -

166 44 2.79 3+4 36.8 None -

167 58 6.68 3+4 54.9 None -

184 63 6.71 4+3 44.6 None -

185 55 6.03 3+4 45.2 None -

188 72 9.14 3+3 69.5 None -

190 72 6.03 3+4 58.1 None -

205 66 6.76 3+4 16.9 None -

223 62 5.1 3+4 74.9 Biochemical 1852

228 45 6.55 3+4 74.3 Biochemical 2013

229 48 6.04 3+4 54.7 None -

233 69 9.74 3+4 15.4 None -

237 66 4+5 26 Unknown -

242 56 10.92 3+3 79 None -

248 73 15.9 4+3 116 Biochemical 90

252 65 42 3+4 22.9 Biochemical 306

265 59 4.53 4+4 20.8 Biochemical 83

274 62 3.9 3+4 105.5 None -

283 70 10.8 3+4 104.1 None -

335 58 3.78 4+3 45.5 None -

336 65 5.96 3+4 22.2 None -

343 69 10.77 3+3 62.0 None -

54

348 63 2.16 3+4 73 None -

351 62 3.04 4+3 62.3 None -

352 51 3.62 4+3 66.1 None -

361 72 3.09 3+4 24.4 None -

362 67 2.78 3+3 44.8 None -

366 57 4.01 4+3 63.8 Biochemical 370

367 52 5.9 3+4 3 None -

370 56 10.76 4+4 62.7 Biochemical 475

375 61 4.29 3+4 16.5 None -

378 62 8.976 3+4 60.6 Biochemical 1455

389 64 7.45 4+5 40.8 None -

393 54 5.18 3+3 33.6 None -

398 72 12.86 3+4 53.7 Biochemical 92

405 64 15.44 3+4 40.1 Biochemical 75

423 68 8.13 3+3 49.0 None -

430 55 9.37 3+4 9.6 None -

448 63 7.1 3+4 35.6 None -

452 62 5.12 4+3 48.2 Biochemical 575

453 71 4.97 4+3 55.1 None -

455 56 4.5 3+4 49.1 None -

457 62 5.74 3+4 38.7 None -

463 70 4.11 4+3 52.1 None -

470 72 6.8 3+4 47.9 None -

473 66 4.64 3+4 52.8 None -

474 68 6.16 3+4 43.3 None -

477 58 4.62 3+4 11.9 None -

480 51 0.94 3+3 48.2 None -

482 64 2.84 3+3 45.9 None -

485 61 5.91 3+4 49.7 Biochemical 1491

488 66 9.61 4+3 28.9 None -

490 63 4.48 3+4 44 None -

491 64 6.21 3+3 9.3 None -

491 64 6.21 3+3 9.3 None -

494 65 2.38 3+4 41.4 None -

494 65 2.38 3+4 41.4 None -

498 50 4.68 3+3 50.7 None -

501 61 3.93 4+3 41.4 None -

527 61 7.66 3+3 40 None -

537 51 22.7 4+3 30.1 None -

538 69 10.7 3+4 33.3 Biochemical 746

540 46 6.2 3+4 18.3 None -

544 59 7 3+4 0 None -

547 65 10.1 4+5 30.7 Biochemical 96

551 48 4.8 3+3 22.8 None -

555 64 5.5 3+4 24.2 None -

563 63 9.8 3+4 5.6 None -

55

574 43 9.6 4+5 51 Biochemical 570

575 50 9.71 3+3 25.1 None -

579 48 4.7 3+3 22.6 None -

582 58 3.61 3+4 19.7 None -

593 50 7.56 3+3 8.8 None -

594 60 4.3 3+4 27.0 None -

599 49 8.9 3+4 24.5 None -

601 56 5.3 4+3 0 None -

604 54 3.8 3+4 0 None -

610 64 3.6 4+3 0 None -

616 59 4.45 4+3 1.4 None -

619 71 4.17 3+4 3.3 None -

621 60 3.13 3+4 16.8 None -

625 64 3 3+4 21.0 None -

626 54 3.28 3+4 1.4 None -

627 53 7.5 4+5 19.1 Biochemical 41

629 71 3.2 3+3 20.5 None -

634 70 14.91 4+5 11.7 None -

636 59 6.38 3+3 16.8 Biochemical 139

643 61 7.16 3+4 18.1 None -

645 53 3.3 3+3 15.4 None -

646 64 4.9 3+4 16.6 None -

648 64 8.4 4+4 0 None -

07 8392 66 21.2 3+5 12 None -

07 866 66 13.5 3+4 21 None -

07 8980 72 23.2 4+3 12 Biochemical 1

07 9957 53 4.4 3+4 12 None -

TB 1872 71 68 Biochemical 1

TB 2682 57 28.5 4+3 45 None -

TB1875 73 12 Biochemical

56

Supplementary Table S2: Diagnostic methylation markers of prostate cancer identified

by PAM.

Rank CpG ID Gene Symbol

1 cg00489401 FLT4

2 cg10541755 EIF5A2

3 cg05270634 RND2

4 cg02879662 HIF3A

5 cg17231524 MGC39606

6 cg26537639 CYBA

7 cg22262168 MOBKL2B

8 cg14563260 EDG2

9 cg19790294 CYBA

10 cg07186138 APOBEC3C

11 cg14672994 FLJ20920

12 cg21096399 MCAM

13 cg15146752 EPHA2

14 cg24340926 RAB33A

15 cg20557104 B3GALT7

16 cg04622802 LOC387758

17 cg17965019 HIST1H3J

18 cg09300114 SLC16A5

19 cg08359956 LR8

20 cg10453365 RHCG

21 cg08924430 FLJ20032

22 cg13102585 RPIP8

23 cg00848728 DAB1

24 cg03085312 RARA

25 cg06428055 ELF4

26 cg04448487 GDAP1L1

27 cg09851465 C1orf87

28 cg08348496 HAPLN3

29 cg22862656 SCGB2A2

30 cg22319147 CDH5

31 cg27223047 FBN2

32 cg08965235 LTBP3

33 cg24715245 UCHL1

34 cg02254461 AXUD1

35 cg26025891 PSTPIP1

36 cg01683883 CMTM2

37 cg17606785 EFS

38 cg21307628 URB

39 cg18328334 TNS1

40 cg19853760 LGALS1

41 cg16232979 TPM4

42 cg23502772 MGC42105

57

43 cg04034767 GRASP

44 cg20083676 EDG3

45 cg21623671 ANXA6

46 cg12627583 AOX1

47 cg19713460 SYNGR1

48 cg19423196 MAT1A

49 cg22892110 MAPK15

50 cg12727795 PDGFRB

51 cg15835232 HLF

52 cg12100791 PYCARD

53 cg09704415 SPATA6

54 cg04337944 FBLN1

55 cg14360917 SP2

56 cg26420196 GAS6

57 cg04920951 GSTP1

58 cg27554782 CHRNB4

59 cg00727590 PLA2G3

60 cg14188232 ITGA11

61 cg18145505 GREM1

62 cg18711066 NFATC3

63 cg26124016 RARB

64 cg24512400 KLK10

65 cg15528736 FCGRT

66 cg01777397 RSNL2

67 cg03513363 DUSP15

68 cg21790626 ZNF154

69 cg02659086 GSTP1

70 cg00862041 GPRASP2

71 cg18552413 DARC

72 cg23499956 S100A16

73 cg17329164 PPT2

74 cg18006568 FLJ12056

75 cg14539231 EPSTI1

76 cg04273431 PRR3

77 cg15910208 KLK10

78 cg12585943 PPT2

79 cg15309006 LOC63928

80 cg17568996 NFAM1

81 cg24467291 RSN

82 cg02029926 CXCL1

83 cg20786074 EFEMP1

84 cg25806808 CXCL1

85 cg23092823 PODN

86 cg09099744 CDKN2A

87 cg25259754 FCRL3

58

Supplementary Table S3: Prognostic methylation markers of prostate cancer identified

by SAM survival.

Rank CpG ID Gene Symbol q-value (%)

1 cg01352108 KCNK4 0

2 cg24068372 LOC349136 0

3 cg20870559 OAS2 0

4 cg03734874 FLJ42486 0

5 cg03640944 KIAA1754 18.05041

6 cg02320454 GPR150 25.7863

7 cg17173423 MS4A3 15.04201

8 cg05047411 MAGEA8 15.04201

9 cg26164184 FCN2 15.04201

10 cg04645174 OR7A17 26.81402

11 cg05828624 REG1A 26.81402

12 cg21325760 MAGEL2 26.81402

13 cg20804821 GPR62 26.81402

14 cg03600318 SFTPD 26.81402

15 cg11061975 SIRPB2 26.81402

16 cg14620221 OR8B8 26.81402

17 cg13311440 CD48 26.81402

18 cg27504299 TCL1B 26.81402

19 cg03109316 ZNF80 26.81402

20 cg00918005 REG3G 26.81402

21 cg17836145 VNN2 26.81402

22 cg15457079 CPN1 26.81402

23 cg07688234 PFC 26.81402

24 cg22511262 WT1 26.81402

25 cg24169915 FLJ25773 26.81402

26 cg03833774 ZCCHC5 26.81402

27 cg20832020 VSIG9 26.81402

28 cg17338403 SLCO3A1 26.81402

29 cg01564343 TREML1 26.81402

30 cg22228134 GZMH 26.81402

31 cg22442090 GIMAP5 26.81402

32 cg01731341 FGF6 26.81402

33 cg19000186 CNGA1 26.81402

34 cg15711744 ANP32D 26.81402

35 cg03544379 OR7C2 26.81402

36 cg07443748 CESK1 26.81402

37 cg04353769 MS4A6A 26.81402

38 cg04014889 MAGEL2 26.81402

39 cg07379574 C19orf4 26.81402

40 cg10994126 PAPPA2 26.81402

41 cg03014957 DEFB118 26.81402

42 cg24012708 HDHD3 26.81402

59

43 cg13447818 FLG 26.81402

44 cg05222924 WT1 26.81402

45 cg18368125 TMED6 26.81402

46 cg19718882 WIT-1 26.81402

47 cg13097816 GPR35 26.81402

48 cg12237269 SLN 26.81402

49 cg19241311 DEFB123 26.81402

50 cg16777782 CDH13 26.81402

51 cg05248470 LILRB2 26.81402

52 cg16158220 REGL 26.81402

53 cg21353232 SEZ6L 26.81402

54 cg13482233 HEPH 26.81402

55 cg12234947 GNAT2 26.81402

56 cg15075718 MFRP 26.81402

57 cg01351032 CIITA 26.81402

58 cg01693350 WT1 26.81402

59 cg12878228 PRSS1 26.81402

60 cg06550629 GPR133 26.81402

61 cg01757745 C10orf93 26.81402

62 cg02813121 S100A12 26.81402

63 cg06806711 MS4A1 26.81402

64 cg13297249 FLJ38379 26.81402

65 cg01369413 UBQLN3 26.81402

66 cg09217923 TAAR2 26.81402

67 cg00690280 WFDC10B 26.81402

68 cg21742836 PPP4C 26.81402

69 cg24122922 C20orf39 26.81402

60

REFERENCES

61

Andriole GL, Grubb RL, Buys SS, Chia D, Church TR, Fouad MN, Gelmann EP, Kvale

PA, Reding DJ, Weissfeld JL, et al. 2009. Mortality Results from a Randomized

Prostate-Cancer Screening Trial. N Engl J Med 360: 1310-1319.

Baylin SB, Esteller M, Rountree MR, Bachman KE, Schuebel K, and Herman JG. 2001.

Aberrant patterns of DNA methylation, chromatin formation and gene expression

in cancer. Hum. Mol. Genet. 10: 687-692.

Belinsky SA, Nikula KJ, Palmisano WA, Michels R, Saccomanno G, Gabrielson E,

Baylin SB, and Herman JG. 1998. Aberrant methylation of p16INK4a is an early

event in lung cancer and a potential biomarker for early diagnosis. Proceedings of

the National Academy of Sciences of the United States of America 95: 11891 -

11896.

Bhutani N, Brady JJ, Damian M, Sacco A, Corbel SY, and Blau HM. 2010.

Reprogramming towards pluripotency requires AID-dependent DNA

demethylation. Nature 463: 1042-1047.

Bird A. 2002. DNA methylation patterns and epigenetic memory. Genes & Development

16: 6 -21.

Brooks JD, Weinstein M, Lin X, Sun Y, Pin SS, Bova GS, Epstein JI, Isaacs WB, and

Nelson WG. 1998. CG island methylation changes near the GSTP1 gene in

prostatic intraepithelial neoplasia. Cancer Epidemiology Biomarkers &

Prevention 7: 531 -536.

Brunner AL, Johnson DS, Kim SW, Valouev A, Reddy TE, Neff NF, Anton E, Medina

C, Nguyen L, Chiao E, et al. 2009. Distinct DNA methylation patterns

characterize differentiated human embryonic stem cells and developing human

fetal liver. Genome Res 19: 1044-1056.

Cairns P, Esteller M, Herman JG, Schoenberg M, Jeronimo C, Sanchez-Cespedes M,

Chow N, Grasso M, Wu L, Westra WB, et al. 2001. Molecular Detection of

Prostate Cancer in Urine by GSTP1 Hypermethylation. Clinical Cancer Research

7: 2727 -2730.

Chen Z, Mann JR, Hsieh C, Riggs AD, and Chédin F. 2005. Physical and functional

interactions between the human DNMT3L protein and members of the de novo

methyltransferase family. J. Cell. Biochem 95: 902-917.

Chomez P, De Backer O, Bertrand M, De Plaen E, Boon T, and Lucas S. 2001. An

overview of the MAGE gene family with the identification of all human members

of the family. Cancer Res 61: 5544-5551.

Chuang LS, Ian H, Koh T, Ng H, Xu G, and Li BF. 1997. Human DNA-(Cytosine-5)

Methyltransferase-PCNA Complex as a Target for p21WAF1. Science 277: 1996-

62

2000.

Das PM, and Singal R. 2004. DNA Methylation and Cancer. Journal of Clinical

Oncology 22: 4632 -4642.

Ehrlich M. 2002. DNA methylation in cancer: too much, but also too little. Oncogene 21:

5400-5413.

El-Maarri O, Kareta MS, Mikeska T, Becker T, Diaz-Lacava A, Junen J, Nusgen N,

Behne F, Wienker T, Waha A, et al. 2009. A systematic search for DNA

methyltransferase polymorphisms reveals a rare DNMT3L variant associated with

subtelomeric hypomethylation. Hum. Mol. Genet. 18: 1755-1768.

Esteller M, and Herman JG. 2002. Cancer as an epigenetic disease: DNA methylation and

chromatin alterations in human tumours. J. Pathol. 196: 1-7.

Estève P, Chin HG, and Pradhan S. 2005. Human maintenance DNA (cytosine-5)-

methyltransferase and p53 modulate expression of p53-repressed promoters.

Proceedings of the National Academy of Sciences of the United States of America

102: 1000-1005.

Feinberg AP, and Vogelstein B. 1983. Hypomethylation distinguishes genes of some

human cancers from their normal counterparts. Nature 301: 89-92.

Finne P, Finne R, Auvinen A, Juusela H, Aro J, Määttänen L, Hakama M, Rannikko S,

Tammela TL, and Stenman U. 2000. Predicting the outcome of prostate biopsy in

screen-positive men by a multilayer perceptron network. Urology 56: 418-422.

Freedland SJ, Humphreys EB, Mangold LA, Eisenberger M, Dorey FJ, Walsh PC, and

Partin AW. 2005. Risk of Prostate Cancer–Specific Mortality Following

Biochemical Recurrence After Radical Prostatectomy. JAMA: The Journal of the

American Medical Association 294: 433 -439.

Gama-Sosa MA, Slagel VA, Trewyn RW, Oxenhandler R, Kuo KC, Gehrke CW, and

Ehrlich M. 1983. The 5-methylcytosine content of DNA from human tumors.

Nucleic Acids Res 11: 6883-6894.

Goessl C, Müller M, Heicappell R, Krause H, Straub B, Schrader M, and Miller K. 2001.

DNA-based detection of prostate cancer in urine after prostatic massage. Urology

58: 335-338.

Guerrero-Preston R, Báez A, Blanco A, Berdasco M, Fraga M, and Esteller M. 2009.

Global DNA methylation: a common early event in oral cancer cases with

exposure to environmental carcinogens or viral agents. P R Health Sci J 28: 24-

29.

63

Hoffmann MJ, Engers R, Florl AR, Otte AP, Muller M, and Schulz WA. 2007.

Expression changes in EZH2, but not in BMI-1, SIRT1, DNMT1 or DNMT3B are

associated with DNA methylation changes in prostate cancer. Cancer Biol. Ther

6: 1403-1412.

Irizarry RA, Ladd-Acosta C, Wen B, Wu Z, Montano C, Onyango P, Cui H, Gabo K,

Rongione M, Webster M, et al. 2009. The human colon cancer methylome shows

similar hypo- and hypermethylation at conserved tissue-specific CpG island

shores. Nat Genet 41: 178-186.

Iyer LM, Tahiliani M, Rao A, and Aravind L. 2009. Prediction of novel families of

enzymes involved in oxidative and other complex modifications of bases in

nucleic acids. Cell Cycle 8: 1698-1710.

Jemal A, Siegel R, Xu J, and Ward E. 2010. Cancer Statistics, 2010. CA Cancer J Clin

60: 277-300.

Jerónimo C, Henrique R, Hoque MO, Mambo E, Ribeiro FR, Varzim G, Oliveira J,

Teixeira MR, Lopes C, and Sidransky D. 2004. A Quantitative Promoter

Methylation Profile of Prostate Cancer. Clinical Cancer Research 10: 8472-8478.

Johnson WE, Rabinovic A, and Li C. 2006. Adjusting batch effects in microarray

expression data using Empirical Bayes methods. Biostat kxj037.

Jones PA, and Baylin SB. 2002. The fundamental role of epigenetic events in cancer.

Nat. Rev. Genet 3: 415-428.

Jones PA. 1986. DNA Methylation and Cancer. Cancer Research 46: 461-466.

Kim E, Kim Y, Jeong P, Ha Y, Bae S, and Kim W. 2008. Methylation of the RUNX3

Promoter as a Potential Prognostic Marker for Bladder Tumor. The Journal of

Urology 180: 1141-1145.

King JC, Xu J, Wongvipat J, Hieronymus H, Carver BS, Leung DH, Taylor BS, Sander

C, Cardiff RD, Couto SS, et al. 2009. Cooperativity of TMPRSS2-ERG with PI3-

kinase pathway activation in prostate oncogenesis. Nat Genet 41: 524-526.

Kron K, Pethe V, Briollais L, Sadikovic B, Ozcelik H, Sunderji A, Venkateswaran V,

Pinthus J, Fleshner N, van der Kwast T, et al. 2009. Discovery of Novel

Hypermethylated Genes in Prostate Cancer Using Genomic CpG Island

Microarrays. PLoS ONE 4: e4830.

Laird PW, and Jaenisch R. 1994. DNA methylation and cancer. Hum. Mol. Genet 3 Spec

No: 1487-1495.

Laird PW, and Jaenisch R. 1996. The role of DNA methylation in cancer genetic and

64

epigenetics. Annu. Rev. Genet 30: 441-464.

Lapeyre JN, Walker MS, and Becker FF. 1981. DNA methylation and methylase levels in

normal and malignant mouse hepatic tissues. Carcinogenesis 2: 873-878.

Lapointe J, Li C, Higgins JP, van de Rijn M, Bair E, Montgomery K, Ferrari M, Egevad

L, Rayford W, Bergerheim U, et al. 2004. Gene expression profiling identifies

clinically relevant subtypes of prostate cancer. Proceedings of the National

Academy of Sciences of the United States of America 101: 811-816.

Ley TJ, Ding L, Walter MJ, McLellan MD, Lamprecht T, Larson DE, Kandoth C, Payton

JE, Baty J, Welch J, et al. 2010. DNMT3A Mutations in Acute Myeloid

Leukemia. N Engl J Med. http://www.ncbi.nlm.nih.gov/pubmed/21067377

(Accessed November 29, 2010).

Lin K, Lipsitz R, Miller T, and Janakiraman S. 2008. Benefits and Harms of Prostate-

Specific Antigen Screening for Prostate Cancer: An Evidence Update for the U.S.

Preventive Services Task Force. Annals of Internal Medicine 149: 192 -199.

Lin X, Tascilar M, Lee W, Vles W, Lee B, Veeraswamy R, Asgari K, Freije D, Van Rees

B, Gage W, et al. 2001. GSTP1 CpG island hypermethylation is responsible for

the absence of GSTP1 expression in human prostate cancer cells. American

Journal of Pathology 159: 1815-1826.

Meissner A, Gnirke A, Bell GW, Ramsahoye B, Lander ES, and Jaenisch R. 2005.

Reduced representation bisulfite sequencing for comparative high-resolution

DNA methylation analysis. Nucleic Acids Res 33: 5868-5877.

Müller HM, Widschwendter A, Fiegl H, Ivarsson L, Goebel G, Perkmann E, Marth C,

and Widschwendter M. 2003. DNA Methylation in Serum of Breast Cancer

Patients. Cancer Research 63: 7641-7645.

Nakayama M, Gonzalgo ML, Yegnasubramanian S, Lin X, De Marzo AM, and Nelson

WG. 2004. GSTP1 CpG island hypermethylation as a molecular biomarker for

prostate cancer. J. Cell. Biochem 91: 540-552.

National Cancer Institute. 2010. Cancer Trends Progress Report - 2009/2010 Update.

http://progressreport.cancer.gov/ (Accessed January 8, 2011).

Noushmehr H, Weisenberger DJ, Diefes K, Phillips HS, Pujara K, Berman BP, Pan F,

Pelloski CE, Sulman EP, Bhat KP, et al. 2010. Identification of a CpG Island

Methylator Phenotype that Defines a Distinct Subgroup of Glioma. Cancer Cell

17: 510-522.

Okada Y, Yamagata K, Hong K, Wakayama T, and Zhang Y. 2010. A role for the

elongator complex in zygotic paternal genome demethylation. Nature 463: 554-

65

558.

Okano M, Bell DW, Haber DA, and Li E. 1999. DNA methyltransferases Dnmt3a and

Dnmt3b are essential for de novo methylation and mammalian development. Cell

99: 247-257.

Patra SK, Patra A, Zhao H, and Dahiya R. 2002. DNA methyltransferase and

demethylase in human prostate cancer. Mol. Carcinog. 33: 163-171.

Peehl DM. 2002. Human Prostatic Epithelial Cells - Culture of Epithelial Cells, Second

Edition (eds R. I. Freshney and M. G. Freshney), John Wiley & Sons, Inc., New

York, USA.

http://onlinelibrary.wiley.com/doi/10.1002/0471221201.ch6/summary (Accessed

September 27, 2010).

Pflueger D, Terry S, Sboner A, Habegger L, Esgueva R, Lin P, Svensson MA,

Kitabayashi N, Moss BJ, MacDonald TY, et al. 2011. Discovery of non-ETS gene

fusions in human prostate cancer using next-generation RNA sequencing.

Genome Research 21: 56 -67.

Robbins CM, Tembe WA, Baker A, Sinari S, Moses TY, Beckstrom-Sternberg S,

Beckstrom-Sternberg J, Barrett M, Long J, Chinnaiyan A, et al. 2011. Copy

number and targeted mutational analysis reveals novel somatic events in

metastatic prostate tumors. Genome Research 21: 47 -55.

Sakai Y, Suetake I, Shinozaki F, Yamashina S, and Tajima S. 2004. Co-expression of de

novo DNA methyltransferases Dnmt3a2 and Dnmt3L in gonocytes of mouse

embryos. Gene Expression Patterns 5: 231-237.

Sboner A, Demichelis F, Calza S, Pawitan Y, Setlur S, Hoshida Y, Perner S, Adami H,

Fall K, Mucci L, et al. 2010. Molecular sampling of prostate cancer: a dilemma

for predicting disease progression. BMC Medical Genomics 3: 8.

Schröder FH, Hugosson J, Roobol MJ, Tammela TL, Ciatto S, Nelen V, Kwiatkowski M,

Lujan M, Lilja H, Zappa M, et al. 2009. Screening and Prostate-Cancer Mortality

in a Randomized European Study. N Engl J Med 360: 1320-1328.

Sherlock G, Hernandez-Boussard T, Kasarskis A, Binkley G, Matese JC, Dwight SS,

Kaloper M, Weng S, Jin H, Ball CA, et al. 2001. The Stanford Microarray

Database. Nucleic Acids Res 29: 152-155.

Singh D, Febbo P, Ross K, Jackson D, Manola J, Ladd C, Tamayo P, Renshaw A,

D'Amico A, Richie J, et al. 2002. Gene expression correlates of clinical prostate

cancer behavior. Cancer Cell 1: 203-209.

Taylor BS, Schultz N, Hieronymus H, Gopalan A, Xiao Y, Carver BS, Arora VK,

66

Kaushik P, Cerami E, Reva B, et al. 2010. Integrative Genomic Profiling of

Human Prostate Cancer. Cancer Cell 18: 11-22.

The Cancer Genome Atlas. 2008. Comprehensive genomic characterization defines

human glioblastoma genes and core pathways. Nature 455: 1061-1068.

Tibshirani R, Hastie T, Narasimhan B, and Chu G. 2002. Diagnosis of multiple cancer

types by shrunken centroids of gene expression. Proceedings of the National

Academy of Sciences of the United States of America 99: 6567-6572.

Tomlins SA, Rhodes DR, Perner S, Dhanasekaran SM, Mehra R, Sun X, Varambally S,

Cao X, Tchinda J, Kuefer R, et al. 2005. Recurrent fusion of TMPRSS2 and ETS

transcription factor genes in prostate cancer. Science 310: 644-648.

Tomlins SA, Rhodes DR, Yu J, Varambally S, Mehra R, Perner S, Demichelis F,

Helgeson BE, Laxman B, Morris DS, et al. 2008. The role of SPINK1 in ETS

rearrangement-negative prostate cancers. Cancer Cell 13: 519-528.

Troyanskaya O, Cantor M, Sherlock G, Brown P, Hastie T, Tibshirani R, Botstein D, and

Altman RB. 2001. Missing value estimation methods for DNA microarrays.

Bioinformatics 17: 520-525.

Tusher VG, Tibshirani R, and Chu G. 2001. Significance analysis of microarrays applied

to the ionizing radiation response. Proceedings of the National Academy of

Sciences of the United States of America 98: 5116-5121.

Woodson K, O'Reilly KJ, Hanson JC, Nelson D, Walk EL, and Tangrea JA. 2008. The

usefulness of the detection of GSTP1 methylation in urine as a biomarker in the

diagnosis of prostate cancer. J. Urol 179: 508-511; discussion 511-512.

Yaqinuddin A, Qureshi S, Qazi R, and Abbas F. 2008. Down-regulation of DNMT3b in

PC3 cells effects locus-specific DNA methylation, and represses cellular growth

and migration. Cancer Cell International 8: 13.

Yegnasubramanian S, Kowalski J, Gonzalgo ML, Zahurak M, Piantadosi S, Walsh PC,

Bova GS, De Marzo AM, Isaacs WB, and Nelson WG. 2004. Hypermethylation

of CpG Islands in Primary and Metastatic Human Prostate Cancer. Cancer

Research 64: 1975 -1986.