TGFβ/SMAD4 Signaling and Altered Epigenetics Contribute to Increased Ovarian
Cancer Severity
Dissertation
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy
in the Graduate School of The Ohio State University
By
Daniel Edward Deatherage, B.A.
Molecular Cellular Developmental Biology
The Ohio State University
2011
Dissertation Committee:
Tim Huang, Ph.D., Advisor
Amanda Toland, Ph.D.
Victor Jin, Ph.D.
Huey-Jen Lin, Ph.D.
Copyright by
Daniel Edward Deatherage
2011
Abstract
Ovarian cancer is the eighth most common cancer and is the fifth most common cause of cancer related death among women. Early stage ovarian cancer is very responsive to treatments and more than 93% of patients diagnosed with early stage disease achieve a five year survival rate. By contrast less than 30% of patients who are diagnosed with late stage disease achieve a five year survival rate, yet more than 60% of all cases present as late stage.
Treatment options are typically surgery followed by a combination chemotherapy regiment of a platinum-based chemotherapeutic and a taxane derivative. While this treatment plan works well for early stage disease, recurrent and late stage disease are less responsive and secondary treatment options are not nearly as beneficial. Here we present work investigating the altered epigenetics and
TGFβ/SMAD4 signaling pathway in ovarian cancer in an effort to better understand the difference in disease severity at a molecular level.
We have identified a microRNA hsa-mir-9-3 which is epigenetically repressed by DNA methylation in a panel of primary ovarian cancer patients.
Quantitative analysis of DNA methylation in the CpG island which contains the hsa-mir-9-3 microRNA revealed significant hypermethylation in both patient samples and cell lines as compared to normal tissue samples. Although no
ii significant correlation with a clinical feature could be identified, functional studies reveal that the repression of this microRNA leads to increased proliferation rates as well as a decrease in apoptosis. Despite being unable to identify a specific target gene, we believe that the hypermethylation of the hsa-mir-9-3 locus serves as a novel biomarker for ovarian cancer.
TGFβ/SMAD4 signaling is commonly dysregulated in ovarian cancers while being a key growth inhibition signal for the ovarian surface epithelium during menstruation. Here we present a genome-wide profile of SMAD4 binding by ChIP-sequencing following TGFβ stimulation in the ovarian cancer cell line
A2780. While A2780 displays some altered signaling by having constitutive nuclear presence of SMAD4, TGFβ stimulation has previously been shown capable of inducing additional nuclear translocation and synthesis of SMAD4. We believe this to be the first truly genome-wide profiling of SMAD4 binding using next generation sequencing approaches in an ovarian cancer model.
Comparison of ChIP-seq results with previously reported ChIP-chip studies show dramatic biological and technical differences including more than 70% of all
SMAD4 binding loci being more than 10kb away from the nearest transcription start site. Gene expression analysis following TGFβ stimulation revealed a group of 318 genes whose expression changed following SMAD4 binding to the distal promoter of the gene. Of those 318 genes, a subset of them was used to predict patient survival in two independent previously published patient cohorts.
Together these results suggest that the loss of long distance SMAD4 gene
iii regulation following TGFβ stimulation may play a key role in ovarian carcinogenesis.
Additionally, we identified a novel biomarker, CLDN11, whose epigenetic repression is associated increased cisplatin resistance in a tissue culture model system. Examination of CLDN11 expression levels in a previously reported patient cohort revealed lower expression levels associated with increased tumor grade. Finally, loss of CLDN11 expression is associated with increased cellular motility.
In conclusion we have investigated and correlated several different epigenetic and signaling abnormalities associated with an increase in the severity of ovarian cancer while demonstrating the importance of recent technological advances in genome-wide methodologies. Together these results are likely to aid in both future discovery methods as well as patient prognosis and treatment.
iv
Dedication
To my wife whose love and reminders to eat and live while writing this showed me writing, living, eating, and loving can be multitasked.
v
Acknowledgements
My thanks go out to my advisor Dr. Tim Huang for his guidance support and motivation throughout my graduate studies. I will also be eternally appreciative of the time spent by my committee members, Dr. Huey-Jen Lin, Dr.
Amanda Toland, and Dr. Victor Jin. Support, insight, discussion, and challenges provided by them have made me a much better scientist.
It is important to acknowledge the contributions others have made to this work beyond recommendations and advice. Specifically Brian A. Kennedy was responsible for performing the computational work presented in Chapter 3. I can only hope that my interactions with him helped improve his own computational research as much as they helped improve my own bench-work. Additionally the cisplatin resistant A2780 Round 5 cells provided by Dr. Ken Nephew‘s group (Dr.
Dave Miller and Dr. Meng Li in particular) were greatly appreciated as they are central to the studies presented in Chapter 4. Finally, Pei-Yin Hsu for her assistance with performing the FACS analysis presented in Chapter 2 while I was preparing for my candidacy examination.
I also wish to extend my thanks for the numerous members, including those who have moved onto other positions, of the Huang lab, and the affiliated
vi labs of Dr. Huey-Jen Lin, Dr. Victor Jin, and Dr. Qianben Weng. Their daily conversations and assistance has been crucial to all the work presented herein.
Special thanks are given to Dr. Pearly Yan, Dr. Michael Chan, Dr. Greg Singer,
Dr. Dustin Potter, Dr. Alfred Cheng, Dr. Benjamin Rodriguez, Joseph Liu, Judy
Kuo, Dr. Yi-Wen Huang, Dr. Zhengang Peng, Dr. Tao Zuo, Dr. Yu-I Weng, Ta-
Ming Liu, Dr. Cenny Taslim, Dr. Shuying Sun, Sandya Liyanarachchi, Dr. Ya-Ting
Hsu, Michael Trimarchi, Huang-Kai Hsu, and Pei-Yin Hsu.
vii
Vita
2005-2011 Graduate Research Associate
The Ohio State University
2005 B.S. Biochemistry
University of Evansville
2008 Graduate Travel Award OSUMC Research Day
2005 Biology award for Outstanding Senior Thesis University of Evansville
Publications
Yeh, Kun-Tu; Che, Tze-Ho; Yang, Hui-Wen; Chou, Jian-Liang; Chen, Lin-Yu; Yeh, Chia-Ming; Chen, Yu-Hsin; Lin, Ru-Inn; Su, Her-Young; Chen, Gary C- W.; Deatherage, Daniel E.; Huang, Yi-Wen; Yan, Pearlly S.; Lin, Huey-Jen; Nephew, Kenneth P.; Huang, Tim H-M.; Lai, Hung-Cheng; Chan, Michael. Aberrant TGFβ/SMAD4 Signaling Contributes to Epigenetic Silencing of a Putative Tumor Suppressor, RunX1T1 in Ovarian Cancer. Epigenetics 6 2011.
Zuo, Tao; Liu, Ta-Ming; Lan, Xun; Weng, Yu-I; Shen, Rulong; Gu, Fei; Huang, Yi- Wen; Liyanarachchi, Sandya; Deatherage, Daniel E.; Hsu, Pei-Yin; Taslim, Cenny; Ramaswamy, Bhuvaneswari; Shapiro, Charles L.; Lin, Huey-Jen L.;
viii
Cheng, Alfred SL.; Jin, Victor; Huang, Tim H-M. Epigenetic Silencing Mediated through Activated PI3K/AKT Signaling in Breast Cancer. Cancer Research 71 1752-762 2011
Chou, Jian-Liang; Su, Her-Young; Chen, Lin-Yu; Liao, Yu-Ping; Hartman-Frey, Corinna; Lai, Yi-Hui; Yang, Hui-Wen; Deatherage, Daniel E., Kuo, Chieh-Ti; Huang, Yi-Wen; Yan, Pearlly S.; Hsiao, Shu-Huei; Tai, Chien-Kuo; Lin, Huey- Jen L.; Davuluri, Ramana V.; Chao, Tai-Kuang; Nephew, Kenneth P.; Huang, Tim H-M.; Lai, Hung-Cheng; Chang, Michael W-Y.. Promoter Hypermethylation of FBXO32, a novel TGF-β/SMAD4 Target Gene and Tumor Suppressor, is Associated with Poor Prognosis in Human Ovarian Cancer. Laboratory Investigation 90: 414-425, 2010.
Hsu, Pei-Yin; Hsu, Hang-Kai; Singer, Gregory A.C.; Yan, Pearlly S.; Rodriguez, Benjamin A.T.; Liu, Joseph C.; Weng, Yu-I; Deatherage, Daniel E.; Chen, Zhong, Pereira, Julia S.; Lopez, Ricardo; Russo, Jose; Wang, Qianben; Lamartiniere, Coral A.; Nephew, Kenneth P.; Huang, Tim H-M.. Estrogen- Mediated Epigenetic Repression of Large Chromosomal Regions through DNA Looping. Genome Research 20: 733-744, 2010.
Weng, Yu-I; Hsu, Pei-Yin; Liyanarachchi, Sandya; Liu, Joseph; Deatherage, Daniel E.; Huang, Yi-Wen; Zuo, Tao; Rodriguez, Benjamin; Lin, Ching-Hung; Cheng, Ann-Lii; Huang, Tim H-M.. Epigenetic Influences of Low-Dose Bisphenol A in Primary Human Breast Epithelial Cells. Toxicology and Applied Pharmacology 248: 111-121, 2010.
Deatherage, Daniel E.; Potter, Dustin; Yan, Pearlly S.; Huang, Tim-H.-M; Lin, Shili. Methylation Analysis by Microarray. Methods in Molecular Biology 556: 117-139, 2009
Yan, Pearlly S.; Potter, Dustin; Deatherage, Daniel E.; Huang, Tim H.-M; Lin, Shili. Differential Methylation Hybridization: Profiling DNA Methylation with a
ix
High-Density CpG Island Microarray. Methods in Molecular Biology 507 (Part III): 89-106, 2009.
Huang, Yi-Wen; Liu, Joseph C.; Deatherage, Daniel E.; Luro, Jingqin; Mutch, David G.; Goodfellow, Paul J.; Miller, David S.; Huang, Tim-H.M. Epigenetic Repression of microRNA-129-2 Leads to Overexpression of SOX4 Oncogene in Endometrial Cancer. Cancer Research 69: 9038-9046, 2009.
Hsu, Pei-Yin; Deatherage, Daniel E.; Rodriguez, Benjamin, A.T.; Liyanarachchi, Sandya; Weng, Yu-I; Zuo, Tao; Liu, Joseph; Cheng, Alfred S.L.; Huang, Tim H-M.. Xenoestrogen-Induced Epigenetic Repression of microRNA-9-3 in Breast Epithelial Cells. Cancer Research 69: 5936-5945, 2009
Lin, Huey-Jen L.; Zuo, Tao; Lin, Ching-Hung; Kuo, Chieh Ti; Liyanarachchi, Sandya; Sun, Shuying; Shen, Rulong; Deatherage, Daniel E.; Potter, Dustin; Asamoto, Lisa; Lin, Shili; Yan, Pearlly S.; Chen, Ann-Lii; Ostrowski, Michael C.; Huang, Tim H-M.. Breast Cancer-Associated Fibroblasts Confer AKT1- Mediated Epigenetic Silencing of Cystatin M in Epithelial Cells. Cancer Research 68: 10257-10266, 2008
Chan, Michael WY.; Huang, Yi-Wen; Hartman-Frey, Corinna; Kuo, Chieh-Ti; Deatherage, Daniel E.; Qin, Huaxia; Cheng, Alfred SL.; Yan, Pearlly S.; Davuluri, Ramana V.; Huang, Tim H-M.; Nephew, Kenneth P.; Lin, Huey-Jen L. Aberrant Transforming Growth Factor β1 Signaling and SMAD4 Nuclear Translocation Confer Epigenetic Repression of ADAM19 in Ovarian Cancer. Neoplasia 10: 908-919, 2008.
Edwards, Dale D.; Deatherage, Daniel E.; Ernsting, Brian R.. Random Amplified Polymorphic DNA Analysis of Kinship Within Host-Associated Populations of the Symbiotic Water Mite Unionicola foili (Acari: Unionicolidae). Experimental and Applied Acarology 34:67-77, 2004
x
Fields of Study
Major Field: Molecular Cellular Developmental Biology
xi
Table of Contents
Abstract ...... ii
Dedication ...... v
Acknowledgements...... vi
Vita ...... viii
Table of Contents...... xii
List of Tables ...... xx
List of Figures ...... xxi
Chapter 1 Introduction ...... 1
1.1 Ovarian Cancer ...... 1
1.2 Treatment ...... 2
1.3 Cisplatin and Carboplatin ...... 2
1.4 Methods of Cisplatin Resistance ...... 3
1.5 Paclitaxel and Docetaxel ...... 4
1.6 Chemoresistance ...... 5
1.7 Epigenetics ...... 6
xii
1.7.1 DNA Methylation ...... 7
1.7.2 Histone Modifications ...... 11
1.7.3 MicroRNA ...... 13
1.8 The TGFβ / SMAD Signaling Pathway ...... 14
1.9 Cancer Stem Cell Theory ...... 17
1.10 Biological Hypotheses ...... 18
Chapter 2 Cancer Specific Epigenetic Silencing of Hsa-mir-9-3 ...... 23
2.1 Introduction ...... 23
2.2 Results ...... 27
2.2.1 DMH analysis of primary ovarian cancer tumors identifies a
differentially hypermethylated region containing the hsa-mir-9-3
microRNA...... 27
2.2.2 COBRA analysis of a panel of ovarian cancer cell lines shows
methylation detected in DMH study of patient samples exists as a general
ovarian cancer marker...... 28
2.2.3 MassARRAY analysis of ovarian cancer cell lines reveals increased
levels of methylation in more aggressive cancer cell lines...... 28
2.2.4 Methylation of the hsa-mir-9-3 miRNA leads to a reversible epigenetic
silencing effect...... 29
xiii
2.2.5 Transfection of mature hsa-mir-9-3 into MCP2 cells shows loss of hsa-
mir-9-3 leads to increased proliferation rates and reduction in apoptotic
cells...... 30
2.2.6 MassARRAY analysis of patient samples confirms methylation patterns
observed in DMH study...... 31
2.2.7 Methylation of 85 tumor samples does not correlate with clinical
pathological factors...... 32
2.2.8 Hsa-mir-9-3 expression is varied more than fivefold among patients. ..33
2.2.9 FOXG1, a previously published target of hsa-mir-9 in brain tissue of
mice, is not repressed by hsa-mir-9-3 in human ovarian cancer cells. ....33
2.2.10 Hsa-mir-9-3 inhibition of KIF21A in MCP2 cells does not correlate with
a loss of KIF21A expression in ovarian cancer patients...... 34
2.2.11 Cancer specific methylation of hsa-mir-9-3 is not related to ovarian
cancer initiating cells...... 36
2.3 Discussion ...... 38
2.3.1 DMH: an Imperfect Breakthrough...... 38
2.3.2 Complexity of Epigenetic Regulation of a MicroRNA...... 41
2.3.3 Epigenetic Silencing of Hsa-mir-9-3...... 42
2.4 Materials and Methods ...... 44
2.4.1 Reagents ...... 44
xiv
2.4.2 Patient Tumor Samples ...... 45
2.4.3 Differential Methylation Hybridization (DMH) ...... 45
2.4.4 M-Score and Smudge Plots ...... 46
2.4.5 Cell Culture ...... 46
2.4.6 RNA Isolation and Reverse Transcription...... 48
2.4.7 Quantitative Reverse Transcription PCR (qRT-PCR)...... 48
2.4.8 Bisulfite Conversion ...... 48
2.4.9 Combined Bisulfite Restriction Analysis (COBRA) ...... 49
2.4.10 MassARRAY ...... 50
2.4.11 Clinical Correlations ...... 51
2.4.12 Western Blot ...... 52
2.4.13 Ovarian Cancer Initiating Cell (OCIC) Isolation ...... 52
Chapter 3 ChiP-seq Mapping of the TGFβ Pathway, Combined with Gene
Expression Profiling and In Silico Data Mining, Identifies Clinically Relevant
SMAD4 Target Genes in Ovarian Cancer ...... 81
3.1 Introduction ...... 81
3.2 Results ...... 84
3.2.1 Genome-wide SMAD4 occupancy defined by ChIP-seq technology. ...84
xv
3.2.2 Examination of SMAD4 occupancy prior to TGFβ stimulation reveals
distinct binding pattern in basal state...... 85
3.2.3 Examination of SMAD4 occupancy after TGFβ-stimulation reveals a
similar distribution of loci to the basal state albeit among drastically
different genes...... 86
3.2.4 Regulation of TGFβ-stimulated SMAD4 target gene expression in
A2780...... 88
3.2.5 SMAD4-dependent gene regulatory networks in TGFβ-induced ovarian
cancer cells...... 89
3.2.6 Gene signatures of selection and clinical outcome...... 92
3.3 Discussion ...... 93
3.4 Materials and Methods ...... 96
3.4.1 Cell Culture and TGFβ Stimulation ...... 96
3.4.2 Chromatin Immunoprecipitation and Massive Parallel Sequencing ...... 97
3.4.3 Gene Expression Profiling ...... 98
3.4.4 RT-qPCR and ChIP-qPCR ...... 98
3.4.5 Processing ChIP-Seq and Microarray Gene Expression Data ...... 99
3.4.6 Gene Regulatory Network Analysis ...... 100
3.4.7 Patient Cohorts ...... 101
xvi
Chapter 4 Hypermethylation of CLDN11 Associated with Cisplatin Resistance in
Ovarian Cancer and Loss of Expression Associated with Increased Tumor
Grade and Mobility...... 123
4.1 Introduction ...... 123
4.2 Results ...... 126
4.2.1 Differential methylation of A2780 R0 and A2780 R5 cells leads to
changes in expression...... 126
4.2.2 CLDN11 and SLC27A6 are repressed upon gain of cisplatin resistance
in a cell culture model...... 127
4.2.3 CLDN11 is methylated in a panel of ovarian cancer cell lines of varying
cisplatin resistance...... 128
4.2.4 CLDN11 is methylated in a set of ovarian cancer patients...... 130
4.2.5 DNA methylation of CLDN11 leads to epigenetic silencing of the gene
in ovarian cancer cell lines...... 131
4.2.6 CLDN11 expression is significantly lower in a group of ovarian cancer
patients as compared to a group of normal tissue samples...... 132
4.2.7 Lower CLDN11 expression was associated with increased tumor grade
as well as decreased survival times...... 133
4.2.8 Loss of CLDN11 expression is associated with increased levels of
motility...... 134
xvii
4.3 Discussion ...... 135
4.4 Materials and Methods ...... 138
4.4.1 Reagents ...... 138
4.4.2 Cell Culture ...... 138
4.4.3 RNA Isolation and Reverse Transcription...... 139
4.4.4 Quantitative Reverse Transcription PCR (qRT-PCR)...... 139
4.4.5 Bisulfite Conversion ...... 140
4.4.6 Combined Bisulfite Restriction Analysis (COBRA) ...... 140
4.4.7 Pyrosequencing ...... 141
4.4.8 Patient Cohorts ...... 142
Chapter 5 Discussion ...... 157
5.1 Multifaceted Mechanisms of Ovarian Carcinogenesis...... 157
5.2 Technological Advances as an Advancement of Science...... 160
5.3 DNA Methylation of the Hsa-mir-9-3 Locus Is Specifically Associated with
Ovarian Cancer...... 164
5.4 TGFβ/SMAD4 Signaling Regulates a Series of Genes Clinically Relevant to
Ovarian Cancer Outcome...... 167
xviii
5.5 Repression of the CLDN11 Gene through DNA Hypermethylation Is
Associated with Increased Cisplatin Resistance, Tumor Grade, and Mobility
in Ovarian Carcinogenesis...... 170
5.6 Concluding Remarks...... 172
References ...... 174
xix
List of Tables
Table 2.1 List of genes downregulated at least 1.8 fold in MCP2 cells following
transfection with synthetic hsa-mir-9, and their overlap with various target
prediction programs...... 79
Table 3.1 Details of SMAD4 ChIP-sequencing results...... 121
Table 3.2 List of 124 patients‘ tumor stage and median survival time for each
patient group determined based on Figure 3.11 hierarchical gene clustering
of TGFβ stimulated SMAD4 responsive genes...... 122
xx
List of Figures
Figure 2.1 Smudge plots of differential methylation hybridization microarrays for
the hsa-mir-9-3 locus...... 54
Figure 2.2 COBRA analyses of hsa-mir-9-3 methylation patterns in ovarian
cancer cell lines...... 56
Figure 2.3 Quantitative Sequenom MassARRAY analysis of hsa-mir-9-3 locus in
ovarian cancer cell lines...... 57
Figure 2.4 Re-expression of hsa-mir-9-3 following epigenetic drug treatment in
ovarian cancer cell lines...... 59
Figure 2.5 MCP2 cells display reduced cell proliferation rates upon exogenous
presence of synthetic hsa-mir-9...... 61
Figure 2.6 MCP2 cells display slight increases in both pro-apoptotic and
apoptotic cell populations by FACS analysis when treated with synthetic
hsa-mir-9...... 62
Figure 2.7 Quantitative Sequenom MassARRAY analysis of hsa-mir-9-3 locus in
primary ovarian cancer cell lines...... 63
Figure 2.8 Quantitative expression of hsa-mir-9-3 in a panel of ovarian cancer
patient tumors...... 66
xxi
Figure 2.9 Qualitative and quantitative western blot analysis of FOXG1 protein
level in the presence or absence of either synthetic hsa-mir-9 mimic, or
negative control...... 67
Figure 2.10 Quantitative expression of KIF21A in MCP2 cells under various
epigenetic drug treatments...... 69
Figure 2. 11 KIF21A expression level in MCP2 cells following synthetic knock-in
of Hsa-mir-9...... 70
Figure 2.12 Quantitative expression of KIF21A and hsa-mir-9-3 in a panel of 25
ovarian cancer patients...... 71
Figure 2.13 Comparison of KIF21A and hsa-mir-9-3 expression levels in primary
ovarian cancer tumors isolated from patients...... 73
Figure 2.14 Smudge plots of differential methylation hybridization microarrays for
the hsa-mir-9-3 locus in OCIC and bulk tumor samples...... 75
Figure 2.15 Quantitative Sequenom MassARRAY analysis of hsa-mir-9-3 locus in
12 sets of bulk primary tumors and subpopulation of OCIC samples...... 77
Figure 3.1 Distribution of SMAD4 binding loci relative to gene body. 102
Figure 3.2 Classification of SMAD4 binding patterns relative to opposite
condition. 103
Figure 3.3 ChIP-PCR of selected target genes identified by ChIP-Seq. 105
Figure 3.4 Heatmap of expression microarray results for TGF β stimulated and
unstimulated A2780 cells. 106
xxii
Figure 3.5 Overlap of expression microarray results with ChIP-sequencing
results. 107
Figure 3.6 Go annotation comparison of genes with an expression change,
SMAD4 binding, or both. 108
Figure 3.7 qRT-PCR of selected targets showing differential expression and
SMAD4 binding upon TGFβ stimulation. 109
Figure 3.8 Comparison of TGFβ/SMAD4 regulated genes in the A2780 ovarian
cancer cell line to previously reported genes in normal ovarian cell line and
skin cell line. 110
Figure 3.9 Comparison of the GO annotation of TGFβ/SMAD4 target genes in
three cell types. 111
Figure 3.10 TGFβ/SMAD4 gene regulatory networks of different tissue types. 114
Figure 3.11 Hierarchical clustering of patients based on expression of
TGFβ/SMAD4 target genes. 116
Figure 3. 12 Kaplan-Meier survival curve for patients displaying differential
expression of a 49 gene signature. 117
Figure 3.13 Hierarchical clustering based on expression of TGFβ/SMAD4 target
genes for additional patient data set. 119
Figure 3.14 Kaplan-Meier survival curve for patients with differential expression
of a 19 gene signature. 120
xxiii
Figure 4.1 DMH determined DNA methylation levels of genes over expressed in
either A2780 Round 0 or Round 5 cells...... 143
Figure 4.2 Quantitative real time PCR expression levels of 18 genes repressed in
A2780 Round 5 cells as compared to A2780 Round 0...... 144
Figure 4.3 DNA methylation analysis of CLDN11 and SLC27A6 in a panel of
ovarian cancer cell lines by COBRA...... 145
Figure 4.4 Pyrosequencing of CLDN11 locus in a panel of ovarian cancer cell
lines...... 147
Figure 4.5 Pyrosequencing of CLDN11 locus in a panel of 78 primary ovarian
tumors and 10 normal ovarian tissue samples...... 148
Figure 4.6 CLDN11 expression in a panel of ovarian cancer cell lines...... 150
Figure 4.7 Relative expression of CLDN11 in panel of 42 patients...... 151
Figure 4. 8 Kaplan-Meier survival curve for patients displaying differential
expression of CLDN11...... 153
Figure 4.9 CLDN11 siRNA knockdown is associated with increased motility by
wound healing assay...... 155
xxiv
Chapter 1
Introduction
1.1 Ovarian Cancer
Despite our advances in understanding ovarian cancer, and it being the eighth most common cancer, it remains the fifth most common cause of cancer related death among females and the number one gynecological related cancer cause of death (Howlader, et al., 1975-2008). While 93.5% of patients diagnosed with early stage ovarian cancer achieve five year survival, they only account for
15% of total diagnoses. Less than 28% of patients diagnosed with late stage disease are able to achieve five year survival while accounting for 62% of total diagnoses. As early stage ovarian cancer is often asymptomatic and confined to the ovary, there is little women can do beyond routine screenings to detect the disease at an early stage rather than wait till it spreads. The troubling fact remains that despite our advances in understanding the disease, early detection and surgical removal is the most critical prognostic determinant of outcome
(Mutch, 2002). This may be exacerbated by the fact that over 80% of tumors are eventually classified as chemo-resistant thereby drastically limiting treatment options (Agarwal & Kaye, 2003). The decrease in survival of late stage tumors
1 may be the result of the increased number of opportunities for individual cells to become chemo-resistant during treatment as later staged tumors have more cells, and are spread out through the abdominal cavity.
1.2 Treatment
While surgical debulking has long been a mainstay of patient treatment and more extensive debulking is a good prognostic indicator of patient outcome
(Mutch, 2002), subsequent chemotherapy treatments have evolved a lot since the use of melphalan was first shown to be an effective anti-cancer drug for use in the treatment of ovarian cancer in the 1970s (Agarwal & Kaye, 2003). Several studies have shown platinum based chemotherapy drugs to be effective at treating ovarian cancer when administered alone or in combination with a taxane compound.
1.3 Cisplatin and Carboplatin
Used since the 1980s as chemotherapeutic drugs for the treatment of a variety of different types of cancer, cis- diamminedichloroplatinum (II) (cisplatin and carboplatin) is thought to mediate cytotoxicity via the disruption of both transcription and replication by creating intra and inter strand DNA adducts
(Stordal & Davey, 2007). Platinum based treatments have been shown to be very effective at treating early stage ovarian cancer with the vast majority of patients
2 achieving five year survival (Ozols, 2006). Intercellular cisplatin levels represent a balance between cellular uptake and cellular efflux.
The limited knowledge we have of how cisplatin enters cells suggest two key mechanisms of uptake: passive diffusion and active transport (Wang &
Lippard, 2005). Passive diffusion is supported by studies showing linear rates of diffusion, inability to saturate uptake levels by increasing concentration, variation in extracellular PH having no effect, and addition of structural analogous not lowering cisplatin intracellular levels [(Hromas, North, & Burns, 1987); (Mann &
Andrews, 1991); (Binks & Dobrota, 1990)]. More recently active transport via copper transport proteins has been shown to represent a double edged sword for while CTR1 has been shown to increase intracellular levels of cisplatin, copper exporter ATP7b has been shown to be capable of removing cisplatin from the cell
[(Ishida, Lee, Thiele, & Herskowitz, 2002); (Komatsu, et al., 2000)].
1.4 Methods of Cisplatin Resistance
Four different mechanisms (one for each of the key aspects of drug action) of cisplatin resistance have been identified in ovarian cancer. Cisplatin which is able to enter and remain inside the cell has been shown to be inactivated by sulfur-containing molecules such as glutathione [(Yang, Sharrocks,
& Whitmarsh, 2003); (Wang, Martindale, & Holbrook, 2000)]. Reduction of intracellular levels has been shown to be caused by variation of expression levels
3 of various copper transport proteins [(Ishida, Lee, Thiele, & Herskowitz, 2002);
(Komatsu, et al., 2000)]. While MRP2 has also been shown to be capable of causing increased efflux of cisplatin, it is believed to actually be transporting complexes of cisplatin and glutathione suggesting it may actually be exporting inactive cisplatin (Borst, Evers, Kool, & Wijnholds, 2000) which may explain the lack of clear association between increased multidrug resistance-associated protein 2 (MRP2) levels and clinical chemoresistance (Rabik & Dolan, 2007). In addition to affecting intracellular levels of active cisplatin, ovarian cancer cells have shown two unique ways of dealing with cisplatin created DNA adducts: increased activity of DNA repair mechanisms leading to lower accumulation of
DNA adducts (Lai, Ozols, Smyth, Young, & Hamilton, 1988), and increased tolerance to DNA adducts by disabling apoptotic pathways (Koberle, Masters,
Hartley, & Wood, 1999).
1.5 Paclitaxel and Docetaxel
Taxol [a taxane derivative isolated from the bark of the Taxus brevifolia tree (Wani, Taylor, Wall, Coggon, & McPhail, 1971)] was first evaluated for use in the treatment of many cancer types in the late 1980s during a phase I study and showed promising antitumor activity particularly for adenocarcinomas (Wiernik,
Schwartz, Strauman, Dutcher, Lipton, & Paietta, 1987). While only one patient with an ovarian adenocarcinoma was present in the phase I study, that patient
4 displayed a minor response to Taxol despite having been diagnosed as being cisplatin-resistant, and thus was further evaluated in several phase II clinical trials (Einzig, Wiernik, Sasloff, Runowicz, & Goldberg, 1992). Since then,
Paclitaxel and docetaxel in particular have been shown to be effective at treating ovarian cancer by interfering with proper microtubule dynamics leading to G2-M phase arrest and apoptosis (McGuire, et al., 1996). While they have been shown to have similar responses as platinum based therapies, initially they were more commonly used to treat tumors shown to be resistant to platinum based drugs which had previously been established as being effective in the treatment of ovarian cancer (Einzig, Wiernik, Sasloff, Runowicz, & Goldberg, 1992). As they operate in a fundamentally different manner than platinum drugs, and as resistance to platinum based drugs is not predictive of resistance to taxane based drugs, the combinatorial use of a platinum based and taxane based drug emerged as the standard treatment for ovarian cancer (McGuire, et al., 1996).
1.6 Chemoresistance
Despite some conflicting reports about advantages to concurrent use of platinum based drugs such as cisplatin and carboplatin with taxanes such as paclitaxel and docetaxel [(ICON Group, 2002); (Muggia, et al., 2000)], there is an overall trend showing combination therapy leads to longer progression-free survival (Sandercock, Parmar, Torri, & Qian, 2002). Unfortunately recent reports
5 state that while more than 90% of patients diagnosed with early stage disease are able to achieve complete remission, more than 50% of ovarian cancer patients who initially achieve complete remission with platinum and taxane combination treatment will eventually relapse, many with a chemoresistant form of the disease that drastically limits treatment options (Balch, Matei, Huang, &
Nephew, 2010). While numerous approaches are currently being investigated for how to limit chemoresistant relapse and treat chemoresistant tumors, the low or uncertain success of these studies has not allowed for a new standard of care to be developed. As epigenetic changes have been shown to contribute to chemoresistance, one avenue of research currently being examined is the use of epigenetic modifying drugs to re-sensitize cells to the platinum based drugs they have become resistant to in order to capitalize on the efficiency of those platinum based drugs as chemotherapeutic agents.
1.7 Epigenetics
While the field of genetics has focused on examining changes to the sequence of DNA which is capable of altering the function of a protein product, the field of epigenetics has focused on examining changes in all other heritable factors which can effect RNA transcription and protein translation [(Holliday &
Pugh, 1975); (Riggs, 1975)]. While there are approximately 25,000 to 30,000 genes in the human genome which can be mutated in any number of ways, the
6 expression of each gene can be modified in multiple epigenetic ways. Epigenetic changes have been shown to be involved in every type of cancer, and specific changes have been shown to significantly correlate with all tracked clinical features (Balch, Matei, Huang, & Nephew, 2010). While genetic studies have been ongoing for decades, there exists a growing consensus that epigenetic changes pose at least an equal, if not greater, role in carcinogenesis (Chan, et al., 2008).
Three key types of epigenetic markers are DNA methylation, microRNAs, and histone modifications. Intricate crosstalk between the pathways responsible for maintaining each of these markers is required for a normal epigenome as a change to one marker can influence the other markers. Dysregulation of each of these markers has been specifically shown to contribute to ovarian cancer progression (Jones & Baylin, 2007).
1.7.1 DNA Methylation
DNA methylation refers to the addition a methyl group to the 5‘ base of a cytosine base. In mammalian systems this has been shown to most commonly take place on a cytosine base when it is immediately followed by a guanine (so called CpG dinucleotides). The addition of these methyl groups is accomplished by a family of proteins called the DNA methyltransferases (DNMTs). Of these methyltransferases, evidence suggests that DNMT1 is primarily responsible for
7 maintaining DNA methylation in the genome during cellular division using the conserved strand of DNA as a template to dictate methylation of the new strand
(Kho, Baker, Layoon, & Smith, 1998). In a normal cell, DNA methylation is required for proper silencing of imprinted alleles and DNA transposons (Bestor,
1998).
Because the human genome is ~3.2 billion bases in length with 42% CG content, one would expect to find CpG dinucleotides in excess of 141 million
(4.41%) (Scarano, Iaccarino, Grippo, & Parisi, 1967). In actuality, fewer than 40 million CpG dinucleotides were found to exist in the human genome for less than a 1% representation. Long before the human genome was sequenced it was known that CpG dinucleotides were underrepresented in the genome as methylated cytosine residues are mutational hot spots (Coulondre, Miller,
Farabaugh, & Gilbert, 1978). These ~40million CpG sites are unevenly distributed throughout the genome, commonly clustering into CpG Islands which classically are defined as stretches of at least 500 bases which contain at least
55% GC content, and a ratio of observed CpG sites to expected CpG sites of more than 0.65 (Takai & Jones, 2002). Of the 27,800 CpG islands which exist in the human genome, more than 50% are located within promoter of a gene
(Davuluri, Grosse, & Zhang, 2001). Recently ―CpG island shores‖ have been defined as the regions flanking CpG islands (Irizarry, et al., 2009). By definition these regions must contain have a lower GC content and/or fewer overall CG dinucleotides.
8
It has been understood for many years that methylation of promoter CpG islands is sufficient to silence that gene (Jones & Laird, 1999). Recent studies have also shown that in some genes, methylation of CpG island shores is not only capable of affecting expression, but can be even more deterministic of expression than methylation of the CpG island itself (Irizarry, et al., 2009).
Additional work is needed to determine what if any specific factors determine the importance of the shore regions versus the body of the island. Information gained from the years of studying promoter CpG island methylation has shown that while some rare instances gene silencing can be caused by the methylation of cytosine bases in the recognition site of transcription factors affecting their ability to bind at their normal binding site, most often though it is the combination of DNA methylation, histone modifications, and nucleosomal remodeling which lead to the epigenetic silencing of a gene (Jones & Baylin, 2007).
Changes in DNA methylation have long been associated with cancer.
While in normal cells most promoter CpG islands are unmethylated and CpG sites in intergenic regions are heavily methylated, as a cell becomes cancerous, a global decrease in methylation is commonly observed while an increase in localized methylation in promoters of genes is observed [ (Esteller, 2008); (Jones
& Baylin, 2007)]. Methylation and demethylation events such as these have been reported to be associated with all types of clinical pathological features of cancer including survival, stage, progression, relapse, and chemoresistance (Balch,
Huang, Brown, & Nephew, 2004). While different events are responsible for
9 triggering changes in DNA methylation, two DNMTs (DNMT1 and DNMT3) are primarily responsible for enacting those changes. Unlike DNMT1, DNMT3 has shown no kinetic preference for methylating hemi-methylated or unmethylated
DNA and as such DNMT3 is thought to be primarily responsible for de novo methylation while DNMT1 is responsible for maintaining existing DNA methylation (Kho, Baker, Layoon, & Smith, 1998). As individuals age a global demethylation phenomenon has been noted for several tissue types (Richardson,
2003). It remains possible that this natural age related global demethylation phenomenon may make the elderly more susceptible to cancer and, indeed, more than 50% of ovarian malignancies occur in women over the age of 55
(Balch, Matei, Huang, & Nephew, 2010). Some specific signaling pathways such as RAS (Gazin, Wajapeyee, Gobeil, Virbasius, & Green, 2007) and TGFβ
[(Papageorgis, et al., 2010); (Kang, et al., 2005)] have been shown to influence
DNA methylation of specific genes.
The sequencing of the first human genome gave us a mostly accurate reference to compare DNA sequence against and design experiments, and additional sequencing efforts have provided greater accuracy while also identifying sequence polymorphisms among individuals (Venter, et al., 2001).
The first complete human methylomes (genome-wide determination of all methylation marks in a given cell type) have recently been sequenced (Lister, et al., 2009), but unlike the sequencing of the first genome, its value as a reference is somewhat questionable. Tissue specific methylation (Futscher, et al., 2002),
10 age related methylation loss (Richardson, 2003), response to environmental factors (Hsu, et al., 2009), and disease (Balch, Huang, Brown, & Nephew, 2004) all change DNA methylation patterns in a temporal/spatial manner meaning that while an individual may have a single genome, they may possess hundreds if not thousands of methylomes through their lifetime. It is thought that through the sequencing of multiple tissue samples from multiple individuals we will begin to determine what the normal state of methylation may be with some degree of certainty in particular tissues. From there we will begin to identify abnormalities which may contribute to cancer and other diseases. Coupling these methylomes with genome-wide studies of histone marks and microRNAs will enable the construction of a complete human epigenome and us to fully examine the interplay among the epigenetic forces giving insight into phenotype.
1.7.2 Histone Modifications
Histones are the proteins upon which DNA is coiled around in order to maintain structure, compact the DNA so it can fit inside a single cell, and ensure that transcription factors and machinery are able to access regions of DNA which are needed to be transcribed. Histones themselves assist in the expression or repression of genes by modifying the N-termini of core histones, particularly at lysine residues within H2A, H2B, H3, and H4 (Jones & Baylin, 2007). Genes which are not meant to be temporally or spatially transcribed lack acetylation
11 marks, and often are replaced by methyl groups via histone methyltransferases which is a mark of gene silencing (Cedar & Bergman, 2009). While acetylation is a marker for activation and methylation is a mark for repression makes for a good rule of thumb, the ‗histone code‘ hypothesis suggests that the situation is much more complex with each mark being capable of both activating and repressing transcription depending on which residue is actually modified, and the extent to which that residue is modified (i.e. mono-, di-, tri- methylated) (Campos &
Reignberg, 2009).
As with DNA methylation, maintenance of normal histone marks requires multiple proteins working in a highly regulated manner to ensure proper transcriptional control is maintained. In fact, DNA methylation itself has been shown to be partially responsible for coordinating the placement of repressive histone marks [(Jones, et al., 1998); (Nan, et al., 1998)]. DNA methylation and repressive histone marks are highly associated with each other and gene silencing. Though debate remains as to which event occurs first, it is likely that either event is able to come first depending on some complex set of circumstances. Just as with DNA methylation, the deregulation of this tightly controlled process has been associated with the silencing and activation of various tumor suppressors and oncogenes advancing the cancer process (Jones
& Baylin, 2007).
12
1.7.3 MicroRNA
Originally discovered in Caenorhabditis elegans, MicroRNAs are short 21-
23 nucleotide pieces of RNA which are not translated but rather interact with the
RNA induced silencing complex (RISC) to prevent complementary RNA transcripts from being translated (Lee, Feinbaum, & Ambros, 1993). To date more than 850 microRNAs have been discovered in the human genome, and several more are computationally predicted to exist (Friedman & Jones, 2009).
They are required for normal development, embryogenesis, and maintaining regulatory networks, but have also been shown to be dysregulated in cancer and other diseases [(Bartel, 2009); (Asli, Pitulescu, & Kessel, 2008)]. In cancer their dysregulation has been associated with a variety of clinical pathological features much like DNA methylation has.
MicroRNAs are able to repress translation via two distinct mechanisms: mRNA digestion and translational repression [(Bartel, 2009); (Selbach,
Schwanhäusser, Thierfelder, Fang, Khanin, & Rajewsky, 2008)]. Evidence suggests that the determining factor in which mechanism is used is directly related to the level at which the miRNA and transcript complement each other.
Instances of complete complementarity can lead to digestion of the transcript thereby enabling a single copy of the miRNA to silence many (potentially all) copies of the target transcripts (Du & Zamore, 2005). When a miRNA and mRNA complement each other to a lower degree the mRNA will not be digested, but rather the transcript will be translationally repressed. The exact mechanism of
13 translational repression is not fully understood, but recent studies have strongly implicated the inhibitory effect exerting itself at the initiation step [reviewed in
(Filipowicz, Bhattacharyya, & Sonenberg, 2008)]. These two distinct repressive pathways are therefore able to modulate protein function with two different severities: complete translational silencing by digesting the mRNA and incomplete inhibition by having copy number affects.
MicroRNAs are themselves subject to epigenetic regulation as they are initially transcribed as very long primary miRNA (pri-miRNA) transcripts (up to several 1000 bases) by RNA polymerase II before being processed to semi- mature 70-80 nucleotide precursor miRNA (pre-miRNA) which are then processed by a dsRNA-specific ribonuclease (Drosha) into the mature 21-23 nucleotide forms (Schickel, Boyerinas, Park, & Peter, 2008). Studies have shown than an individual miRNA may be capable of targeting hundreds of separate transcripts, further underscoring the danger of their deregulation [(Friedman &
Jones, 2009); (Selbach, Schwanhäusser, Thierfelder, Fang, Khanin, & Rajewsky,
2008)].
1.8 The TGFβ / SMAD Signaling Pathway
The transforming growth factor-β (TGFβ) signaling pathway is capable of modulating a variety of cellular functions including but not limited to: cellular proliferation, differentiation, and limiting epithelial repair during the menstrual
14 cycle (Massague, Blain, & Lo, 2000). It is capable of affecting so many different cellular functions in part because it uses the SMAD family of proteins as downstream transcription factors. SMAD proteins are notorious for their interaction with a wide variety of different cofactors including E2F1 (AN & Korc,
2005), Ap2 (Naso, Uitto, & Klement, 2003), PBX1 (Baily, Rave-Harel, McGillivray,
Coss, & Mellon, 2004), Oct1 (Oren, Torregroza, & Evans, 2005), and p300/CBP
(De Caestecker, et al., 2000). As SMAD transcription factors bind with very low affinity to their consensus binding site [so called ‗SMAD-Binding Elements‘
(SBEs)] (Dennler, Itoh, Vivien, ten Dijke, Huet, & Gauthier, 1998), it is actually the binding of the cofactors to their recognition sequences which dictates SMAD binding (Seoane, Le, Shen, Anderson, & Massague, 2004). As the TGFβ pathway is capable of affecting both cellular differentiation and proliferation, it is not surprising that it is often dysregulated in cancer (Derynck, Akhurst, &
Balmain, 2001). Interestingly, TGFβ signaling has been shown to confer both pro and anti cancer effects to tumors largely depending on the stage which they are in. Typically pro cancerous effects are reserved for late stage tumors, and anti cancerous effects occur mostly in early staged tumors. In fact, the TGFβ pathway has been shown to be dysregulated in numerous different cancer types including ovarian cancer with both pro and anti cancer properties [reviewed in (Blobe,
Schlemann, & Lodish, 2000)].
As TGFβ plays a growth inhibition signal for ovarian surface epithelial
(OSE) cells during the repair portion of the menstrual cycle, TGFβ exists near the
15 ovary where its dysregulation may play a role in ovarian cancer. In fact, dysregulation of TGFβ signaling has been shown to be associated with increased
OSE proliferation, positioning dysregulated TGFβ signaling as a key player in ovarian carcinogenesis [(Berchuck, et al., 1992); (Nilsson & Skinner, 2002)]. Loss of normal TGFβ signaling is further implicated in being involved in the cancerous process as more than 50% of deaths occur in postmenopausal women (Balch,
Matei, Huang, & Nephew, 2010).
Recently, the dysregulation TGFβ/SMAD signaling has been shown to be capable of influencing DNA methylation patterns in breast cancers (Papageorgis, et al., 2010). In their study they showed for the first time that high levels of TGFβ in the tumor microenvironment led to overactive SMAD signaling which was required to maintain a more metastatic phenotype implicating TGFβ signaling as providing an ―epigenetic memory‖ by controlling a specific DNA methylation pattern. While their study focused on novel pro-cancer epigenetic affects of TGFβ in a late stage (metastatic) breast cancer model, more work will be needed to confirm TGFβ‘s ability to maintain an ―epigenetic memory‖ is an intrinsic property.
If it is proven to be, normal cells may require proper TGFβ signaling for a normal epigenome and loss of signaling may promote early stage cancer progression.
16
1.9 Cancer Stem Cell Theory
Born out of attempts to determine the cells of origination of acute myeloid leukemia (AML), which presents as an accumulation of hematopoietic cells which are not capable of functional differentiation (Failkow, 1987), the cancer stem cell theory states that a subset of cells within a tumor are capable of both self renewal and differentiation in a manner which recreates the original tumor
(Bonnet & Dick, 1997). While it had been hypothesized for nearly three decades that AML was a disease originating in stem cells originally based on a subset of tumor cells displaying a different cell cycling pattern than most of the cells within the tumor [(Griffin & Laowenberg, 1986); (Park, Bergsagel, & McCulloch, 1971)], the creation of a NOD/SCID mouse model provided the first direct evidence that even a single CD34+ CD38- tumor cell could give rise to a AML tumor which contained CD34+ CD38+ cells (Bonnet & Dick, 1997). A consensus definition of cancer stem cells was recently developed as being ―cells within a tumor that possess the capacity to self-renew and to cause the heterogeneous lineages of cancer cells that comprise the tumor‖ (Clarke, et al., 2006).
Since the original discovery of cancer stem cells (CSCs) in AML, CSCs have also been identified in a number of additional solid tumors following their identification in breast cancer (Al-Hajj, Wicha, Benito-Hernandez, Morrison, &
Clarke, 2003). Visvader and Lindeman recently reviewed cancer stem cells in solid tumors (Visavader & Lindeman, 2008). Additionally though not reviewed by
Visvader and Lindeman, evidence of ovarian cancer stem cells has been
17 published (Zhang, et al., 2008). While cancer stem cells often make up a very small portion of total tumor volume, their ability to regenerate an entire tumor in mouse models show the importance of eradicating these cells during treatment.
Frighteningly, research has shown that a number of cancers actually enrich for cancer stem cells following radio or chemotherapy (Visavader & Lindeman,
2008).
It is interesting to consider such the theory in the context of chemoresistant and recurrent cancers. Specifically as chemotherapy treatments may not kill all cells in a tumor, the remaining cells may only be able to repopulate the tumor when the surviving cells are the cancer stem cells.
Additionally, as many chemotherapeutic drugs are known to make cells more prone to mutations, some of which may convey a survival advantage in the form of chemical resistance, the presence of these chemoresistance mutations in a stem cell may explain why recurrent tumors are often non-responsive to drugs that were originally used to treat them.
1.10 Biological Hypotheses
The first known record of cancer comes from an ancient Egyptian papyrus written more than 5,000 years ago (Hajdu, 2010). For all the time, money, and though that have been spent on trying to understand, detect, and cure this disease since then, new discoveries are still being made every year with no sign
18 of slowing down. With each new discovery more light is shed on cancer, yet it is rare that new discoveries completely discredit previous discoveries (for instance, the discovery of microRNAs involvement in cancer did not discredit the involvement of genetic mutation). I believe this underscores the complexity of the disease, and leads me to hypothesize that any cellular abnormality is capable of contributing to a cancerous state. Three main topics of research are presented here which I believe support this overall hypothesis in ovarian cancer, and I believe is indicative of cancer in general. While all abnormalities are capable of contributing to carcinogenesis, their contributions are not equal. Each abnormality has one or more specific physiological outcome which leads me to further hypothesize that abnormalities which are able to affect multiple physiological aspects simultaneously are the most dangerous. To this end, the work presented here has led to the generation of three additional distinct hypotheses directly related to each chapter each of which supports the concept that the identified abnormalities are affecting multiple processes and therefore very dangerous.
Chapter 2 focuses on the epigenetic silencing of the microRNA hsa-mir-9-
3 via DNA hypermethylation in both ovarian cancer cell lines and primary ovarian cancer patients. I hypothesize that the association of DNA hypermethylation at the hsa-mir-9-3 locus may serve as a useful biomarker for classifying cells as cancerous. Further, as the loss of hsa-mir-9 expression is associated with increased proliferation rates and reduction in apoptosis, ovarian cancer may
19 select for the loss of hsa-mir-9 expression, or the over expression of its target genes via other mechanisms. As it has been shown that an individual microRNA is capable of regulating the protein levels of hundreds of genes at the same time
(Selbach, Schwanhäusser, Thierfelder, Fang, Khanin, & Rajewsky, 2008), I further hypothesize that a single event (epigenetic repression of hsa-mir-9-3) is affecting multiple pathways to slight levels yet leading to an overall pro- cancerous state.
Chapter 3 focuses on the identification of TGFβ responsive SMAD4 targets in an ovarian cancer cell line via ChIP-sequencing and shows the association of these target genes with differential median survival times. As we compared the identified targets to previously identified targets in a non- tumorigenic immortalized ovarian surface epithelial cell line and noted exceptionally limited overlap, I hypothesize that dysregulation of the
TGFβ/SMAD4 pathway leads to an increased cancerous state due to SMAD4‘s ability to interact with a diverse range of other transcription factors to regulate gene expression. In this way a single aberration (dysregulated TGFβ/SMAD4 signaling) affects the expression of multiple genes which are then each able to affect multiple cellular processes and leading to the development of a more cancerous state.
Chapter 4 focuses on the epigenetic silencing of the CLDN11 gene in association with the acquisition of cisplatin resistance in a cell line model, and the
20 association of lower levels of CLDN11 expression with increased cellular mobility, and tumor grade. I hypothesize that the loss of CLDN11 expression leads to a net increase in cellular efflux of cisplatin and increased mobility leading to more aggressive ovarian tumors. As CLDN11 is associated with the tight junction complex, the loss of CLDN11 expression leads to an increase in surface area of each individual cell. While this provides no net benefit to passive efflux or influx of cisplatin from the individual cell, I believe that it leads to a net increase in cellular efflux as cisplatin molecules which leave the cell are less likely to be immediately taken back up by an adjacent cell as the cells are more spread out.
In this way, a single aberration (epigenetic silencing of CLDN11) is capable of regulating two distinct physiological processes: drug resistance, and increased cellular mobility.
While the contributions of aberrant DNA methylation, microRNA expression, dysregulated signaling pathways, and loss of cell adhesion molecules to the ovarian cancer process demonstrated within this text should not be diminished, in agreement with my overall hypothesis about cancer, I believe it is equally important to not underestimate the contributions of factors which I did not specifically study. With billions of dollars dedicated annually to the study of cancer, I believe if there was a simple cure or single causative factor, it would have been discovered by now. Ultimately while the term ―cancer‖ is used to represent a disease of abnormal cellular growth, it is highly possible that because of the heterogeneity of cells within a given tumor, each individual tumor needs to
21 be treated as a unique disease with unique intervention to cure. I hypothesize that when these types of patient specific treatments are developed, they will focus on many biomarkers to both diagnose and treat the disease with biomarkers capable of affecting multiple physiological processes being of increased importance. The identification of an epigenetically silenced microRNA hsa-mir-9-3, the identification of a group of TGFβ responsive SMAD4 target genes, and the identification of an epigenetically silenced gene, CLDN11, which contributes to cisplatin resistance could someday play a role in the personalized medicine treatment of ovarian cancer.
22
Chapter 2
Cancer Specific Epigenetic Silencing of Hsa-mir-9-3
2.1 Introduction
The epigenetic silencing of genes via DNA methylation has been shown to play a key role in the development of ovarian cancer (Balch, Matei, Huang, &
Nephew, 2010). Several specialized techniques have been developed to identify hypermethylated targets ab initio from a variety of source material as an alternative to whole genome bisulfite sequencing which is plagued by several problems not originally encountered by the human genome project and ultimately not cost or time effective. There are two main problems associated with whole genome bisulfite sequencing that are primarily based on the bisulfite conversion process. Converting non-methylated cytosine residues into thymine residues causes a loss in sequence complexity. When combined with shorter maximal read lengths (no more than 600 bases) which can be achieved because the bisulfite conversion process degrades DNA into smaller fragments, shotgun sequencing approaches and mapping become increasingly difficult. Ultimately these problems make a ―whole methylome‖ sequencing approach unviable using traditional sequencing approaches, and have spurred the development of
23 alternative methods to analyze the human methylome in different ways. One method used to interrogate the methylome is differential methylation hybridization
(DMH) which while able to overcome issues with whole genome bisulfite sequencing by relying on microarray technology focusing on a specific list of target locations (CpG islands), it does lose resolution as compared to a bisulfite sequencing approach.
DMH is a high-throughput microarray based methodology of studying the
CpG islands of the human genome. First developed in 1997 by our lab, it takes advantage of restriction enzymes whose recognition sequence contains a CG site, but their ability to digest the DNA is dependent on the methylation state of that CG site (Huang, Laux, Hamlin, Tran, Tran, & Lubahn, 1997). The method has been refined and improved [most notably by moving from gel based resolution, to clonal microarrays (Yan, Perry, Laux, Asare, Caldwell, & Huang,
2000), to commercial CpG island microarrays] over the years while being used to study DNA hypermethylation in a variety of cancers leading to the identification of several methylated biomarkers (Huang, Perrry, & Laux, 1999). The use of DMH has been recently reviewed in (Yan, Potter, Deatherage, Lin, & Huang, 2009).
Currently the G4492 microarray set from Agilent is used for the microarray analysis as it contains 244,000 probes covering all 27,800 annotated CpG islands.
24
In addition to studies showing the involvement of DNA methylation in cancer, the dysregulation of microRNAs has been implicated in a number of cancer systems (Calin, et al., 2002) including ovarian (Iorio, et al., 2007). As microRNAs are regulated by the same cellular machinery as conventional genes, it was a logical hypothesis that microRNAs could also be epigenetically regulated in a manner similar to genetic elements. A number of studies support this hypothesis by showing the hyper and hypo methylation of a microRNA and/or its promoter associated CpG island leading to changes in microRNA abundance
(Roman-Gomez, et al., 2009).
Two of the three microRNAs which encode for the mature hsa-mir-9 have recently been implicated in carcinogenesis. The microRNA hsa-mir-9-1 was recently shown to be significantly hypermethylated in a panel of 71 breast cancer tumors (Lehmann, et al., 2008). The microRNA hsa-mir-9-3 has also been implicated in a number of studies recently though with diverse functional involvement. Iorio et al. showed that in a hsa-mir-9-3 was frequently down regulated in breast cancers with high vascular invasion or lymph node metastasis thought the significance of this involvement was not established (Iorio, et al.,
2005). In addition to confirming frequent hypermethylation of hsa-mir-9-3 in breast cancer, Hsu et al. showed that exposure to xenoestrogens may play a role in the hypermethylation of hsa-mir-9-3 in breast cancer cells (Hsu, et al., 2009).
Neither study however was able to identify what gene was being repressed by hsa-mir-9. In a mouse brain development study FOXG1 was found to be the
25 target of the mouse homolog of hsa-mir-9 (which is completely identical to the human microRNA) in order to properly develop Cajal-Retzius cells in the medial pallium of developing telencephalon (Shibata, Kurokawa, Nakao, Ohmura, &
Aizawa, 2008).
We conducted a DMH study of 24 ovarian cancer tumors and identified the microRNA hsa-mir-9-3 as being hypermethylated to some degree in all tumor samples. Further quantitative studies were able to show the average methylation level of 18 of 85 (21%) tumor samples and all ovarian cancer cell lines as being hypermethylated in a 683 bp region surrounding the hsa-mir-9-3 gene relative to eight normal samples. Though we were unable to validate a specific target of the microRNA in ovarian cancer, and unable to correlate its repression with a specific clinical pathological feature, we were able to show that 10 of 12 ovarian cancer initiating cells showed no significant increase in methylation as compared to the bulk tumor from which they were isolated. This leads us to believe that while the methylation of hsa-mir-9-3 is clearly associated with ovarian cancer, it does not appear to be a driving force of the cancer.
26
2.2 Results
2.2.1 DMH analysis of primary ovarian cancer tumors identifies a differentially hypermethylated region containing the hsa-mir-9-3 microRNA.
DMH analysis was conducted on 24 primary ovarian cancer samples isolated from primary tumor samples. A 1656 bp CpG island on chromosome 15 with no associated annotated genes was identified as displaying a significant M- score for ovarian cancer patient samples. Interestingly while no genes were associated with the island, embedded in the middle of the CpG island resides the microRNA hsa-mir-9-3. Smudge plot visualization of the M-score analysis and a genomic map showing the placement of hsa-mir-9-3 is shown in Figure 2.1.
Interestingly all 24 tumor samples were methylated though their pattern of methylation was able to be classified into three different categories: core methylation (n=8, 33%) in which the center of the CpG island containing the microRNA was methylated while the edges of the CpG island were largely unmethylated; flanking methylation (n=10, 42%) in which one or both of the edges of the CpG island were methylated while the region containing the microRNA and immediately downstream of it were unmethylated; and a completely methylation (n=6, 25%) region which displayed some degree of methylation throughout the entire CpG island.
27
2.2.2 COBRA analysis of a panel of ovarian cancer cell lines shows methylation detected in DMH study of patient samples exists as a general ovarian cancer marker.
In order to verify the methylation detected by DMH, COBRA primers were designed for a 351bp region including the transcription start site of the mir-9-3 miRNA. A panel of ovarian cancer cell lines was initially analyzed rather than using the limited resources available with a primary patient sample. Differing levels of methylation were detected among the ovarian cancer cell lines while an
Immortalized Ovarian Surface Epithelial cell line (IOSE) displayed very low levels of methylation suggesting that the methylation detected may in fact be cancer specific (see Figure 2.2).
2.2.3 MassARRAY analysis of ovarian cancer cell lines reveals increased levels of methylation in more aggressive cancer cell lines.
As COBRA analysis is only a qualitative assay which is useful for validating the presence of methylation at any one of several CpG sites, it lacks quantitative capabilities to determine individual CpG sites, and to what extent they are methylated within a population. We therefore performed the
MassARRAY analysis from Sequenom to gain a quantitative measure of methylation for the majority of CpG sites using modified COBRA primers covering the same 351bp region as well as an adjacent 132bp region in the
28 promoter of the miRNA (see Figure 2.3). Of the six cell lines analyzed, only
A2780 did not display a significantly increased level of average methylation as compared to the average methylation and variance of eight samples with a nearly normal phenotype with a 90% confidence level. The average methylation of
A2780 was within 1% of being considered significant by the same criteria. The eight samples with a nearly normal phenotype include an immortalized non- tumorigenic ovarian surface epithelial cell line (IOSE) and seven normal ovarian surface epithelial (NOSE) samples isolated from non-cancerous patients.
Additionally, 54 of 90 (60%) of all CpG sites measured were significantly hypermethylated by the same criteria. The higher levels of methylation detected in the faster growing CP70, MCP2, and MCP3 lines compared with the slower growing A2780 cell line and the normal cell line (IOSE) suggested that the methylation of hsa-mir-9-3 may be involved in more advanced ovarian cancers.
2.2.4 Methylation of the hsa-mir-9-3 miRNA leads to a reversible epigenetic silencing effect.
While DNA methylation is associated with the silencing of genes (and several studies have shown the same phenomenon for the expression of microRNAs) (Iorio, et al., 2007), even the complete methylation of some regions has been shown to be insufficient for the silencing of genes. To determine if there was a functional consequence of the DNA methylation, the expression of mir-9-3
29 was determined by qRT-PCR in the panel of ovarian cancer cell lines (see Figure
2.4). The varied expression levels among the cell lines along with the varied levels of methylation dictated the study of the response to epigenetic drug treatments to verify that the difference in expression was being influenced by the
DNA methylation. Cells were treated with 0.5μM 5‘-aza-2‘-deoxycytidine (5- azaDC) for 72 hours, trichostatin A (TSA) for 12 hours, or a combination of both treatments to check for synergistic effects. While the least methylated cell line
(A2780) showed very little response to any of the treatments, the combination therapy showed significant upregulation for CP70, MCP2, MCP3, and SKOV3.
The lack of restored expression in the heavily methylated HeyC2 line upon synergistic treatment, suggested a more permanent form of silencing or possibly additional secondary expression effects being created by TSA treatment. Taken together, these results confirmed that epigenetic mechanisms were involved in the silencing of hsa-mir-9-3 in ovarian cancer cell lines.
2.2.5 Transfection of mature hsa-mir-9-3 into MCP2 cells shows loss of hsa- mir-9-3 leads to increased proliferation rates and reduction in apoptotic cells.
Having shown that DNA methylation led to the silencing of hsa-mir-9-3, we set out to determine if the loss of expression had a functional consequence within the cell. To do so, we conducted a knock-in experiment with a synthetic mimic of
30 hsa-mir-9 (the mature form of hsa-mir-9-3) into the ovarian cancer cell line MCP2 as it displayed the largest re-expression of hsa-mir-9-3 following epigenetic drug treatment. Six days after transfection with the miRNA mimic, a significantly lower number of cells were detected as compared to their untreated counterpart and nonspecific miRNA (see Figure 2.5). Further, FACS sorting of MCP2 cells revealed an increase in pro-apoptotic cells as compared to parental cells and a negative control (see Figure 2.6). Taken together these results suggested that the loss of hsa-mir-9-3 expression allowed cells to proliferate at an increased rate making them more dangerous than those cells which were able to maintain expression.
2.2.6 MassARRAY analysis of patient samples confirms methylation patterns observed in DMH study.
As cell lines are by definition subjected to prolonged exposure to tissue culture effects which have been shown to increase DNA methylation levels
(Smiraglia, Rush, Fruhwald, Dai, & Held, 2001), it is common for them to display increased levels of DNA methylation which is not observed in primary tumors, and thus not clinically relevant to the disease. The methylation of hsa-mir-9-3 was therefore evaluated by quantitative Sequenom MassARRAY analysis in a panel of 85 primary ovarian tumors and seven primary Normal Ovarian Surface
Epithelial cells (NOSE cells). While the region displayed little to no methylation in
31 all eight normal samples (includes IOSE), the average methylation level in 18 primary tumors (21%) and 366 of 1247 (29%) measured CpG sites displayed a significantly elevated average methylation level as compared to the methylation and variation of eight normal samples at a 90% confidence level (see Figure 2.7).
Despite the primary tumors displaying lower levels of methylation as compared to the panel of ovarian cancer cell lines investigated, it was noteworthy that a similar pattern of methylation was observed with highest levels of methylation of hsa-mir-9-3 being downstream of the microRNA itself, and only a few samples displaying high levels of methylation in the region immediately upstream of the
TSS. Overall the significantly increased levels of methylation in patient samples as compared to normal cells we observed led us to believe that the methylation of hsa-mir-9-3 was involved in the ovarian cancer process.
2.2.7 Methylation of 85 tumor samples does not correlate with clinical pathological factors.
As clinical pathological data was available for the 85 tumor samples we analyzed by Sequenom analysis, we investigated if the methylation level of the hsa-mir-9-3 microRNA was capable of serving any predictive role relating to outcome or severity of disease. No significant correlation was detected between
DNA methylation level and any of the clinical features including: progression free survival, overall survival, tumor type, chemotherapy response, or age. The
32 significantly lower levels of methylation observed among the seven normal NOSE samples still suggested that methylation of the hsa-mir-9-3 microRNA could serve as an ovarian cancer biomarker.
2.2.8 Hsa-mir-9-3 expression is varied more than fivefold among patients.
As the genomic DNA isolated for bisulfite conversion and MassARRAY analysis was isolated previously without RNA isolation, it was impossible for us to determine the expression of hsa-mir-9-3 in those patient samples. Expression of mir9-3 was therefore determined from freshly isolated RNA from 25 random frozen tumor samples collected from OSU. Relative expression was varied over more than fivefold among the 12 samples which had detectable levels of expression (see Figure 2.8).
2.2.9 FOXG1, a previously published target of hsa-mir-9 in brain tissue of mice, is not repressed by hsa-mir-9-3 in human ovarian cancer cells.
Literature reports of FOXG1 protein levels being reduced by hsa-mir-9 expression in mice brain tissue (Shibata, Kurokawa, Nakao, Ohmura, & Aizawa,
2008) led us to examine if we could detect a similar reduction in expression following transfection with the hsa-mir-9 mimic. As previous studies showed no change in mRNA levels of FOXG1 and as our own microarray analysis following hsa-mir-9 transfection showed no change in FOXG1 expression after three days
33
(-1.02 fold difference between treated and untreated) we attempted to verify a reduction in FOXG1 protein rather than mRNA. Western blot analysis with cy5 and cy3 linked secondary antibodies showed no repression of FOXG1 following transfection with synthetic hsa-mir-9 mimic either qualitatively or quantitatively
(see Figure 2.9).
2.2.10 Hsa-mir-9-3 inhibition of KIF21A in MCP2 cells does not correlate with a loss of KIF21A expression in ovarian cancer patients.
Predicting potential targets of a given microRNA has become a key area of interest for several labs which have produced several different prediction algorithms. A search for predicted targets among three programs: PicTar (Krek, et al., 2005), TargetScan (Lewis, Burge, & Bartel, 2005), and miRanda (John,
Enright, Aravin, Tuschl, Sander, & Marks, 2004) revealed 914, 936, and 5,238 potential targets respectively. The numerous (nearly 4000 unique genes some with different splice variants) predicted potential targets of hsa-mir-9-3 among the various target prediction resources, made screening all the potential targets unreasonable. While some microRNAs are known to modulate functional gene expression at the mRNA level by interfering with translation, many microRNA are able to regulate expression by causing the degradation of their target mRNA
(Selbach, Schwanhäusser, Thierfelder, Fang, Khanin, & Rajewsky, 2008). We therefore attempted to limit our list of potential target genes by transfecting a hsa-
34 mir-9 mimic into a cell line and comparing total mRNA expression against the untreated cell line. As MCP2 exhibited the greatest induction of hsa-mir-9-3 expression following epigenetic treatment, it was used for the knock-in and microarray analysis. A total of 35 genes were downregulated by at least 1.8 fold.
Of those genes, seven were predicted as a potential target of hsa-mir-9 by
PicTar, TargetScan, or miRanda with only one gene, KIF21A, being a predicted target by all three programs (see Table 2.1). As Oncomine data analysis revealed KIF21A had previously been reported to be upregulated in several cancer types we attempted to determine if hsa-mir-9-3 was responsible for regulating KIF21A expression.
Initial tests of KIF21A expression following the epigenetic treatments which had previously been shown to induce the expression of hsa-mir-9-3 revealed a significant decrease in KIF21A (see Figure 2.10). While such a finding is suggestive of hsa-mir-9‘s ability to decrease KIF21A expression, there are numerous secondary effects that could be playing a role in repressing KIF21A following epigenetic treatment. In order to verify that it was in fact hsa-mir-9 repressing KIF21A, we transfected a hsa-mir-9 mimic into MCP2 cells and measured KIF21A expression after six days coinciding with the decrease in proliferation we had previously observed (see Figure 2.11). After six days of exposure to the mir-9 mimic, KIF21A was 40% less than in MCP2 cells treated with a nonspecific microRNA mimic suggesting that KIF21A was an in vitro target of hsa-mir-9.
35
To test hsa-mir-9-3‘s ability to modulate the expression of KIF21A in vivo,
KIF21A expression was measured and plotted alongside hsa-mir-9-3 expression in the both the 12 patients expressing hsa-mir-9-3, and the 13 patients lacking hsa-mir-9-3 (see Figure 2.12 A, and B respectively). Of the 23 patients with detectable levels of KIF21A, the lowest four values corresponded to samples with no detectable level of hsa-mir-9-3 while the two samples which had no detectable level of KIF21A also had no detectable level of hsa-mir-9-3. When KIF21A expression was plotted against hsa-mir-9-3 expression for either the samples with hsa-mir-9-3 expression or all samples, the trend was for higher levels of hsa- mir-9-3 to associate with higher levels of KIF21A albeit with poor R2 values (0.25 and 0.006 for hsa-mir-9-3 expressing and all samples respectively) (see Figure
2.13). This surprising association of increasing levels of hsa-mir-9-3 expression correlating with increasing levels of KIF21A in patient samples led us to believe that while hsa-mir-9-3 may be capable of regulating KIF21A in a cell culture model, it was not playing a key role in actual tumors.
2.2.11 Cancer specific methylation of hsa-mir-9-3 is not related to ovarian cancer initiating cells.
The inability to correlate the methylation of hsa-mir-9-3 with any clinical- pathological data coupled with the inability to identify the mRNA target of the microRNA led us to examine if there was enriched levels of methylation in
36 ovarian cancer initiating cells (OCIC). Comparing six isolated OCIC samples with bulk tumor from the same patient sample revealed no significant differences in methylation by DMH (see Figure 2.14). Surprisingly, no dramatic increase in methylation was detected in any of the samples, and only sample 1 and 66 seemed to indicate any increase at all. As DMH lacks quantitative capabilities and only analyses a few CpG dinucleotides in a given region, we decided to examine the methylation of 12 OCIC samples and corresponding bulk tumors with the quantitative MassARRAY assay from Sequenom (see Figure 2.15).
Increased levels of methylation were detected in only 35 of 162 (21.6%) CpG dinucleotides detected in both bulk and OCIC samples and only two of 12 average methylation levels (~17%) were increased by at a 90% confidence level relative to the variation detected among the seven nose samples and the IOSE cell line previously analyzed.
An additional comparison of the 340 total CpG units with detected levels of methylation to the seven NOSE and IOSE samples revealed that 152 (~45%) and 13 of 24 (~54%) average methylation values were hypermethylated at a 90% confidence level. It is worthy of noting that the increased incidence in this group of patient samples as compared to the group of patients analyzed previously, which displayed ~29% of CpG units as being hypermethylated as compared to eight normal samples, is not likely caused by culture artifacts as similar numbers of hypermethylated sties were observed for both the non-cultured bulk samples
(80 of 169 for ~47%) and the cultured OCIC subpopulation (72 of 171 for ~42%).
37
Taken together, these results led us to believe that while the methylation of hsa-mir-9-3 is a clear biomarker for a cancerous state, the methylation is not enriched in a cancer stem cell like population and therefore is not capable of driving cancer progression while remaining a small population.
2.3 Discussion
2.3.1 DMH: an Imperfect Breakthrough.
DMH allows for the simultaneous study of more than 27,000 CpG islands in the human genome without requiring an additional preexisting selection of targets of interest. This approach has herein proven useful for identifying the methylation of microRNA hsa-mir-9-3 as a nonspecific ovarian cancer biomarker.
Without DMH (or similar methodology) screening an isolated CpG island with no known gene in its vicinity would likely have taken a backseat to studying more interesting CpG islands associated with genes known to be involved in carcinogenesis in other systems, or simply other traditional genes. While the use of DMH methodology is a breakthrough in that it allows for discovery of aberrantly methylated targets whose study otherwise would be unlikely, it is not without flaws. These flaws include: being limited to the probes placed on the array corresponding to specific areas of interest (in this case CpG islands) rather than the entire genome, being a microarray based technology complete with
38 difficulties in analysis and sample to sample variation, and lacking a quantitative analysis of the CpG residues which it does interrogate.
At the time of its creation, CpG island methylation had already become well established as being important for the epigenetic silencing of genes reviewed in (Jones & Laird, 1999) so once it was determined that there was no method of applying the ‗gold standard‘ of bisulfite sequencing to a genomic level, limiting analysis to these regions was highly logical. Unfortunately, the design of the DMH method could not account for the discovery of ―CpG island shore‖ methylation which was recently described (Irizarry, et al., 2009). ―CpG island shores‖ have been defined as being a small cluster of CpG dinucleotides either upstream or downstream of CpG islands which lack classification as a traditional
CpG island by one or more of the classic variables. In their initial work, and in several studies since, the authors have implicated so called ―shore‖ regions of methylation as not only being capable of modulating genetic expression, but also in some cases being more predictive of expression changes than island methylation. While the probe set on the most current DMH microarray some probes localized just off the edge of a CpG island, the nature of M-score analysis looking at a sliding window of study centered on a given base makes it difficult to achieve a high score even with high levels of methylation present. While future iterations of probe design for the microarray or changes to the M-score analysis may be capable of correcting for such issues, the rise of next generation
39 sequencing approaches have made such modifications unpractical for future study.
The need for the creation of DMH (and other methods like it) was born out of the assumption that it was neither time nor cost effected to be able to apply bisulfite sequencing to the entire genome in a time where next generation sequencing approaches were still a decade away from commercial application.
With the advances made in next generation sequencing over the last five years, the cost/time effectiveness of whole genome bisulfite sequencing has begun to be possible. Arabidopsis thaliana is a common model system for studies in plant biology for which epigenetic studies are also common (Lister, et al., 2008). In the years after DMH was developed, an immunoprecipitation of methylated cytosine method [methylcytosine immunoprecipitation (mCIP)] were employed to provide a global methylome map at a 35 bp resolution of Arabidopsis thaliana based on a genome tiling array covering 97% of the 120 Mb (Zhang, et al., 2006). Similar problems to provide a true single base pair resolution of methylation existed in
Arabidopsis despite being more than 4% the size of the human genome. The smaller size did allow for a true single base pair resolution shotgun bisulfite sequencing using next generation sequencing methods in 2008 (Lister, et al.,
2009). While the larger size of the human genome still requires additional advances so as to lower cost, the prospect of such advances not only seem reasonable, but also probable in the coming decade.
40
A shotgun bisulfite sequencing approach h in humans (while now possible, is still expensive) will eventually solve the remaining problems associated with the DMH analysis, but advances are already being made with (global DNA methylation methods on next gen platforms). Eventually the nature of next generation sequencing will allow for a quantitative measure of not only which residues are methylated for a given sample, but also the level of methylation of each of those residues without the need for follow-up study with additional methods such as the MassARRAY analysis from Sequenom.
2.3.2 Complexity of Epigenetic Regulation of a MicroRNA.
The expression of microRNAs, or as in this case the lack of expression of microRNAs, has exceptionally complex outcomes. The ability to regulate a number of different target genes in any given system means that dysregulation of a single microRNA may carry the same functional outcome as the dysregulation of numerous individual genes via a single event. Evidence of such microRNA dysregulation events have recently been reported to influence the expression of several hundred genes simultaneously at a low level (Selbach, Schwanhäusser,
Thierfelder, Fang, Khanin, & Rajewsky, 2008).
While previous studies have implicated hsa-mir-9 as being involved in breast cancer, we believe this to be the first evidence of hsa-mir-9-3 repression being involved in ovarian cancer. While attempts to identify either a correlation
41 with clinical pathological feature or a specific target gene have been unsuccessful, it remains possible that looking at additional patient cohorts or analyzing a larger region with newer more precise methods may prove capable of associating the methylation of a particular region with a given clinical feature.
Identifying a specific target may be accomplished by screening additional targets in a manner such as KIF21A was analyzed.
Despite the lack of a clear target, or clear pathological feature, evidence of increased levels of apoptosis in MCP2 cells following the transfection of synthetic hsa-mir-9 suggests that hsa-mir-9 may be targeting an anti-apoptosis gene, making hsa-mir-9-3 a pro-apoptosis gene. The hypermethylation of hsa-mir-9-3 is therefore capable of reducing miRNA levels leading to less cellular apoptosis which is of obvious benefit to highly proliferative cancer cells. While we were able to show epigenetic therapies were capable of restoring hsa-mir-9-3 and such therapies have been shown to be of therapeutic benefit in several studies (Balch,
Matei, Huang, & Nephew, 2010), the specific use of a synthetic hsa-mir-9 synthetic may also be a future therapeutic intervention method to increase apoptosis.
2.3.3 Epigenetic Silencing of Hsa-mir-9-3.
The microRNA hsa-mir-9-3 was identified as being methylated by differential methylation hybridization (DMH) in a panel of 24 tumors isolated from
42 ovarian cancer patients (see Figure 2.1). After confirming similar methylation in a panel of six ovarian cancer cell lines (see Figure 2.2 and Figure 2.3), and that the hypermethylation was responsible for epigenetically silencing hsa-mir-9-3 expression (see Figure 2.4), we qualitatively measured the methylation level of
85 primary ovarian cancer tumors using Sequenom‘s MassARRAY technology
(see Figure 2.7). In total 18 of the 85 tumors (21%) and 366 of 1247 (29%) individual CpG units were hypermethylated as compared to the variance in methylation detected in seven NOSE samples and the IOSE cell line.
Examination of the previously reported hsa-mir-9 target gene, FOXG1 revealed no direct repression (see Figure 2.9). While the in silico predicted target gene KIF21A was shown to be repressed by hsa-mir-9 in vitro (see Figure 2.10 and Figure 2.11), we were unable to confirm these findings in a panel of 25 ovarian cancer patients (see Figure 2.12 and Figure 2.13). Functionally, we provide evidence that hsa-mir-9-3 is involved in cellular proliferation and apoptosis, and that loss of expression is associated with increased proliferation and increased apoptosis (see Figure 2.5 and Figure 2.6).
Finally we were able to show that the methylation of hsa-mir-9-3 is not associated with Ovarian Cancer Initiating Cells (OCIC) as compared to bulk tumors as only two of 12 (~17%) matched pairs showed an increase in average methylation level for the OCIC samples (see Figure 2.14, and Figure 2.15).
Importantly as a similar number of CpG units were hypermethylated in OCIC
43
(42%) and bulk tumor samples (47%), we believe tissue culture effects were not contributing to increased methylation levels of the OCIC which underwent limited cell culture despite the cohort displaying an increase in the number of hypermethylated CpG units as compared to another cohort (29%).
Taken together these results show that while the hypermethylation of hsa- mir-9-3 is not directly related to a specific clinical factor, more than 20% of patients displaying a significant increase in average methylation as compared to normal samples suggests that the hypermethylation of the hsa-mir-9-3 locus may be a useful non-specific ovarian cancer biomarker. The examination of additional patient samples with more detailed clinical information, or identification of the specific target or targets of hsa-mir-9 may allow for correlation to specific clinical factors such as survival time.
2.4 Materials and Methods
2.4.1 Reagents
5-aza-2‘-deoxycytidine (DAC) and Trichostatin A (TSA) were purchased from Sigma (St. Louis, MO). Synthetic MicroRNA mimics for both hsa-mir-9 as well as negative controls were purchased from Thermo Scientific Dharmacon
(Chicago, IL).
44
2.4.2 Patient Tumor Samples
Tumor tissue samples were collected by the Ohio State University Tissue
Procurement service based on standard protocols approved by the Institutional
Review Board.
2.4.3 Differential Methylation Hybridization (DMH)
The DMH protocol has been developed and optimized in our lab and has previously been published (Deatherage, Potter, Yan, Huang, & Lin, 2009).
Briefly, 500ng of genomic DNA from 24 patients and six pairings of ovarian cancer initiating cells and corresponding bulk tumors was sonicated to a more manageable length of approximately 500bp. DNA adapters were ligated onto the ends of the sonicated DNA ends for subsequent low cycle PCR against the universal known sequence of the adapter. Before PCR, a sequential digestion with a combination of the methylation sensitive restriction enzymes (HpaII and
HinPI due to the difference in their restriction sites: C^CGG and G^CGC respectively), was conducted in order to not only limit false methylation signal caused by incomplete digestion, but also increase the number of methylated CG sites needed to produce a final signal for many fragments. As methylation of the
CG site within their respective recognition sites stops the enzymes from being able to digest, methylated fragments remain intact with the adapters on each end while unmethylated fragments are digested separating the adapters. Following
45
PCR, the DNA was labeled and subjected to microarray analysis on a chip containing all annotated CpG islands (27,800) found in the UCSC Genome
Browser.
2.4.4 M-Score and Smudge Plots
In order to visualize differences in regional methylation events based on individual probe intensities, a kernel smoothing function termed ―Methylation
Score‖ (M-Score) was used. Briefly, normalized log ratios were used to rank probes based on their percentile among all probes, and probes in the top and bottom 25th percentile were counted. The M-Score for a given region was determined by subtracting the number of probes in the bottom 25th percentile from the number of probes in the top 25th percentile before dividing by the total number probes within 1kb of the location of interest. Smudge plots were created using the M-Score values calculated for each probe on the microarray associated with a given CpG island, with specific values correlating to a white-red color gradient for the CpG island.
2.4.5 Cell Culture
A2780, CP70, MCP2, and MCP3 cells were cultured in RPMI-1640
(Invitrogen, Carlsbad, CA) supplemented with 10% fetal bovine serum (FBS)
(Invitrogen) and 50 U/ml penicillin-streptomycin (Invitrogen). Immortalized
46
Ovarian Surface Epithelial (IOSE) cells were cultured in a mixture of 50% media
199 and 50% MCDB 105 (Sigma, St. Louis, MO) supplemented with 10% FBS
(Invitrogen), 400 ng/ml hydrocortisone (Sigma), 10 ng/ml epidermal growth factor
(EGF) (Invitrogen), and 50 U/ml penicillin-streptomycin (Invitrogen). HeyC2 cells were cultured in DMEM (Invitrogen) supplemented with 5% FBS (Invitrogen),
0.1mM nonessential amino acid (Invitrogen), 2mM L-Glutamine(Invitrogen), 0.01
M of HEPES (Invitrogen), and 50 U/ml penicillin-streptomycin (Invitrogen).
SKOV3 cells were cultured in McCoy‘s 5A (Invitrogen) Supplemented with 10%
FBS (Invitrogen), 0.1mM nonessential amino acid (Invitrogen), 2mM L-
Glutamine(Invitrogen), 0.01 M of HEPES, and 50 U/ml penicillin-streptomycin
o (Invitrogen). All cells were cultured in a 37 5% CO2 incubator.
Epigenetic treatments were carried out with a DNA demethylating drug [5- aza-2‘deoxycytidine (DAC)], a histone deacetylase inhibitor [Trichostatin A
(TSA)], and a combination of both. For DAC treatment, cells were treated with
0.5 μM DAC for four days with fresh medium and DAC added every 24 hours before RNA isolation. For TSA treatment, cells were treated with 0.5 μM TSA for
12 hours prior to RNA isolation. For combination treatment, immediately upon completion of four days of DAC treatment cells were subjected to TSA treatment in fresh media lacking DAC for 12 hours before RNA was isolated.
47
2.4.6 RNA Isolation and Reverse Transcription
Total RNA was isolated before or after epigenetic drug treatment via standard TRIzol (Invitrogen) protocol. Subsequent to DNase I (Invitrogen) treatment, cDNA was generated from total RNA with Superscript III (Invitrogen) with a specific primers for either the hsa-mir-9-3 precursor, or a control microRNA (U6) as previously performed (Jiang, Lee, Gusev, & Schmittgen,
2005). For mRNA expression studies, an 80:20 mixture of random hexamers and oligo dT primers were used.
2.4.7 Quantitative Reverse Transcription PCR (qRT-PCR)
All reactions were conducted in triplicate with a melting curve analysis for each reaction on an ABI 7500 using Power SYBR Green master mix (Applied
Biosystems, Carlsbad, CA). The ΔΔCt method of gene expression analysis was used for determining relative levels of hsa-mir-9-3 microRNA expression by comparing it to U6 which has been described previously as being of suitable use as a stand-in for a housekeeping gene (Jiang, Lee, Gusev, & Schmittgen, 2005).
2.4.8 Bisulfite Conversion
500ng of genomic DNA was subjected to bisulfite modification with the EZ
DNA Methylation Kit (Zymo Research, Irvine, CA) with three noteworthy differences from the manufacturer‘s protocol. Briefly, 500 ng of DNA in a volume
48 of 45 μl of water are snap frozen on dry ice before being rapidly thawed at 60o three times in order to fragment the DNA to allow the CT conversion reagent better access to the DNA. Conversion took place over the course of approximately 15.5 hours with cycling of: 50o two hours, 95o 15 seconds, 50o four hours, 95o 15 seconds, 50o 9.5 hours. Final elution of bisulfite converted DNA off the column was accomplished by two separate incubations of DNA grade water heated to 55o for five minutes in 50 μl aliquots. Our lab has observed these described modifications to achieve both a better conversion rate as well as increase the total amount of bisulfite converted DNA albeit at a slightly lower concentration.
2.4.9 Combined Bisulfite Restriction Analysis (COBRA)
PCR primers were designed against stretches of DNA within the hsa-mir-
9-3 region which lack CpG dinucleotides yet are rich with cytosines which were converted to uracil during the bisulfite conversion. Primers were designed in this manner so that when PCR amplification was performed, the DNA was enriched for fragments which were successfully converted. Amplified product was split into two equal aliquots with one aliquot being subjected to digestion with a restriction enzyme whose recognition sequence is only present if the original genomic DNA sequence was methylated, while the other aliquot was mock treated with a reaction mixture lacking restriction enzyme. Post digestion PCR products were
49 resolved on a 3% agarose gel with small DNA fragments in the restriction enzyme aliquot representing qualitative methylation of the locus.
2.4.10 MassARRAY
Quantitative methylation values for all samples were measured as technical triplicates using the T-cleavage kit for the MassARRAY platform
(Sequenom, San Diego, CA), and required a relative standard deviation of less than 10% to be considered for correlation with clinical data. Briefly a T7 promoter tag was added to the PCR amplified onto the flanking region of each of the regions analyzed by COBRA analysis. In Vitro transcription off the T7 promoter was carried out in the presence of a mixture of DNA (C) and RNA (A, G, and U) nucleotides before being treated with RNase A. The presence of a DNA C base
(which is not recognized by RNase A) allows for digestion only after Uracil bases leading to a set of fragments with a mass specific to their nucleotide composition
(Coolen, Statham, Gardiner-Garden, & Clark, 2007). A matrix-assisted laser desorption/ionization time of flight (MALDI-TOF) mass spectrometer in conjuncture with Sequenom‘s EpiTYPER analysis software package was used to determine quantitative methylation levels as a measure of the ratio of fragments with a base composition corresponding to both methylated and unmethylated fragments.
50
Studies in our lab and others have shown that while the Sequenom instrument is capable of recording methylation levels to be accurate to within 5% based on mixing experiments (Coolen, Statham, Gardiner-Garden, & Clark,
2007), the presence of PCR bias among methylated DNA can play a large role in accurately representing the level of methylation (Shen, Guo, Chen, Ahmed, &
Issa, 2007). As a means of correcting for PCR bias in our target locations, a standard curve of known quantities of CpG Methylase treated methylated DNA
(Millipore, Billerica, MA) and blood DNA were run alongside the samples of interest. A linear regression based on the standard curve and applied to the samples. As the standard curve values were able to give values both above
100% methylated, and lower than 0% methylated, a linear transformation was applied to all samples such that all samples were given a value of between 0 and
1.
2.4.11 Clinical Correlations
Patient data for the 85 patient tumors examined was analyzed for correlation with the calculated quantitative methylation values using Mann-
Whitney U tests in the statistical software SPSS (version 10.0; SPSS Inc.,
Chicago, IL). A p value of less than 0.05 was considered to be statistically significant.
51
2.4.12 Western Blot
Whole cell protein extracts were harvested as previously described (Long
& Nephew, 2006). Protein concentrations were quantified via the DC protein assay (BioRad, Hercules, CA). 30 μg of protein were separated on a SDS-PAGE gel and transferred to polyvinylidene difluoride (PVDF) membrane. Membranes were probed with a rabbit polyclonal FOXG1 antibody (Abcam), or goat polyclonal β-actin (Santa Cruz Biotechnology) as a loading control. A horseradish peroxidase conjugated secondary antibodies (KPL, Inc., Gaithersburg, MD) were used to detect primary antibody binding and visualized with the SuperSignal
West Pico chemiluminescent substrate (Peirce Biotechnology, Rockford, IL).
Quantitative protein levels were determined using Cy3/Cy5 conjugated secondary antibody (GE healthcare Picscataway NJ) on a Typhoon 9410
Variable Mode Imager (Amersham Biosciences, Uppsala Sweden).
2.4.13 Ovarian Cancer Initiating Cell (OCIC) Isolation
Several of the tumor samples obtained from the tissue procurement office at the Ohio State University were used to isolate Ovarian Cancer Initiating Cells
(OCIC). Briefly, tumors were mechanically minced into small pieces with scissors before overnight digestion at 37C 5% CO2 in DMEM/F12 medium (Invitrogen) supplemented with 300 U/mL of collagenase (Invitrogen) and hyaluronidase
(Calbiochem). Digested cell mixes were centrifuged at 700RPM for four minutes.
52
The supernatant containing both stromal and fibroblast cells was discarded.
Epithelial cell pellet was washed twice in DMEM/F12 before being cultured overnight on Ultra Low Attachment plates (Corning) in cancer stem cell medium:
DMEM/F12 supplemented with 5μg/mL insulin (Sigma), 20 ng/mL human recombinant epidermal growth factor (Invitrogen), 10 ng/mL basic fibroblast growth factor (Invitrogen), and .4% bovine serum albumin. A five minute, 37o C, incubation with 1mL trypsin (Invitrogen) and filtration through a 40-μm cell strainer was used to obtain single cells. Cells were allowed to grow until a reasonable number of attachment independent spheres were obtained. Cells were sorted on a FACS Aria II for the presence of CD44, CD117, and doubly positive cells. Genomic DNA was isolated from doubly positive cells using the
QIAmp DNA mini kit (Qiagen, Valencia, CA).
53
Figure 2.1 Smudge plots of differential methylation hybridization microarrays for the hsa-mir-9-3 locus.
54
Figure 2.1 Smudge plots of differential methylation hybridization microarrays for the hsa-mir-9-3 locus. White-red smudge plot generated based on M-score analysis of the hsa-mir-9-3 locus for 24 primary ovarian cancer tumors. DMH probe locations for the CpG island (green) are shown in red with the Hsa-mir-9-3 microRNA shown as a blue rectangle with an arrow denoting direction of transcription. All objects are drawn to scale. 24 primary tumors classified into three subcategories based on location of methylation regardless of methylation intensity.
55
Full Size Digested
Figure 2.2 COBRA analyses of hsa-mir-9-3 methylation patterns in ovarian cancer cell lines.
Figure 2.2 COBRA analyses of hsa-mir-9-3 methylation patterns in ovarian cancer cell lines. 351 bp region of the hsa-mir-9-3 region was PCR amplified in a 20 μl reaction before 10μl aliquots were subjected to TaqI or control digestion lacking restriction enzyme. A total of four TaqI sites were present in the amplicon.
Digested products were visualized on a 3% agarose gel with ethidium bromide staining. Both loss of intensity for full sized band and presence additional of lower molecular weight bands was judged to represent qualitative methylation at some combination of the four restriction sites. Full size and digested product locations denoted on the left hand axis.
56
Figure 2.3 Quantitative Sequenom MassARRAY analysis of hsa-mir-9-3 locus in ovarian cancer cell lines.
Figure 2.3 Quantitative Sequenom MassARRAY analysis of hsa-mir-9-3 locus in ovarian cancer cell lines. The 351 bp region analyzed by COBRA, as well as an adjacent region of 132 bp immediately upstream of it, were each PCR amplified and subjected to quantitative methylation analysis on Sequenom‘s
MassARRAY platform. Each amplicon contained six and nine ‗CpG units‘ respectively with between one and four CpG dinucleotides on each CpG unit for a total of 30 CpG dinucleotides. As CpG units are generated based on RNase A digested products, the methylation of individual CpG dinucleotides cannot be distinguished from one another and as such only the average methylation level of all CpG dinucleotides on the unit is reported. Braces are used to denote CpG units containing more than one CpG dinucleotide (vertical lines). Each colored circle represents the level of methylation rounded to the nearest percent as determined by the adjusted methylation value determined based on raw values
57 read on the EpiTYPER 1.0 software at the time of MALDI-TOF analysis and standard curve adjustment and linear transformations.
58
Figure 2.4 Re-expression of hsa-mir-9-3 following epigenetic drug treatment in ovarian cancer cell lines.
Figure 2.4 Re-expression of hsa-mir-9-3 following epigenetic drug treatment in ovarian cancer cell lines. RNA isolated from six ovarian cancer cell lines as well as a normal immortalized ovarian surface epithelial cell line were measured for the expression of hsa-mir-9-3 as well as a control microRNA
U6. RNA was isolated from cells treated with 0.5 μM 5-aza-2‘deoxycytidine
(5aza) for four days with a once a day media and drug change, or 0.5 μM
Trichostatin A (TSA) for 12 hours. For the combination treatment, TSA treatment was performed immediately upon the end of 5aza treatment. Bars represent the average of triplicate measurements Vertical lines indicate relative standard
59 deviation while * were used to indicate significant re-expression of hsa-mir-9-3 (p
< 0.05).
60
160 Untreated Control MCP-2 Negative Control
140 Synthetic Hsa-miR-9 ) 4 120
100
80
60
40 Cell NumberCell (*10 20
0 1 2 3 4 5 6 Time (day)
Figure 2.5 MCP2 cells display reduced cell proliferation rates upon exogenous presence of synthetic hsa-mir-9.
Figure 2.5 MCP2 cells display reduced cell proliferation rates upon exogenous presence of synthetic hsa-mir-9. Cells transfected with hsa-mir-9, negative control microRNA, or nothing were seeded into individual dishes and counted by crystal violet exclusion assay. Values represent the average of two independent experiments with triplicate seeding and measurement for each day.
A student‘s t-test was performed to achieve a p value of 0.003 at day six for hsa- mir-9 as compared to untreated control. Vertical bars represent standard deviation.
61
Figure 2.6 MCP2 cells display slight increases in both pro-apoptotic and apoptotic cell populations by FACS analysis when treated with synthetic hsa-mir-9.
Figure 2.6 MCP2 cells display slight increases in both pro-apoptotic and apoptotic cell populations by FACS analysis when treated with synthetic hsa-mir-9. Cells were transfected with hsa-mir-9, negative control microRNA, or nothing and allowed to grow for six days based on cell proliferation assay. Cells were submitted to FACS analysis. Bars represent the portion of cells in each of the listed states.
62
Figure 2.7 Quantitative Sequenom MassARRAY analysis of hsa-mir-9-3 locus in primary ovarian cancer cell lines.
63
Figure 2.7 Quantitative Sequenom MassARRAY analysis of hsa-mir-9-3 locus in primary ovarian cancer cell lines. The two regions of the hsa-mir-9-3 locus, which were previously studied in the ovarian cancer cell line panel, were each PCR amplified in a patient cohort of 85 primary ovarian cancer tumors and eight normal tissue samples. All samples were subjected to quantitative methylation analysis on Sequenom‘s MassARRAY platform. Each amplicon contained six and nine ‗CpG units‘ respectively each with between one and four
CpG dinucleotides and 30 total CpG dinucleotides. As CpG units are generated based on RNase A digested products, the methylation of individual CpG dinucleotides cannot be distinguished from one another and as such only the average methylation level of all CpG dinucleotides on the unit is reported.
Brackets are used to denote CpG units containing more than one CpG dinucleotide (vertical lines). Each colored circle represents the level of methylation rounded to the nearest percent as determined by the adjusted methylation value determined based on raw values read on the EpiTYPER 1.0 software at the time of MALDI-TOF analysis and standard curve adjustment and linear transformations. 18 of the 85 patient samples were hypermethylated relative to the average variance in seven NOSE samples and the IOSE cell line at the 90% confidence level. Additionally 366 of 1247 individual CpG units were determined to be hypermethylated by the same criteria. Circles with an X rather than a color denote CpG units which were unable to be analyzed due to one or
64 more factors including poor amplification, potential non-CpG methylation, or potential SNPs.
65
Figure 2.8 Quantitative expression of hsa-mir-9-3 in a panel of ovarian cancer patient tumors.
Figure 2.8 Quantitative expression of hsa-mir-9-3 in a panel of ovarian cancer patient tumors. Quantitative SYBR green PCR was used to determine expression levels of the hsa-mir-9-3 microRNA and U6, an endogenous microRNA control. Bars represent the average hsa-mir-9-3 expression value relative to the U6 microRNA in two independent reverse transcription amplifications each analyzed in triplicate. Vertical bars represent relative standard deviations of each sample.
66
Figure 2.9 Qualitative and quantitative western blot analysis of FOXG1 protein level in the presence or absence of either synthetic hsa-mir-9 mimic, or negative control.
Figure 2.9 Qualitative and quantitative western blot analysis of FOXG1 protein level in the presence or absence of either synthetic hsa-mir-9 mimic, or negative control. Protein was isolated from untreated MCP2 cells before transfection of microRNA mimic or negative control and after six days of transfection. Top portion of figure shows fluorescence of Cy3/Cy5 conjugated secondary antibodies on PVDF membrane. Bottom portion shows the relative
67 density of FOXG1 relative to β Actin after six days of transfection as compared to relative density observed for untreated cells at day 0.
68
Figure 2.10 Quantitative expression of KIF21A in MCP2 cells under various epigenetic drug treatments.
Figure 2.10 Quantitative expression of KIF21A in MCP2 cells under various epigenetic drug treatments. MCP2 cells were treated with 0.5μM 5‘-aza-2‘- deoxycytidine for three days, and/or 0.5 μM Trichostatin A before TRIzol RNA extraction for quantitative SYBR green PCR of KIF21A and GAPDH, an internal control. Expression values represent KIF21A expression relative to an untreated
RNA sample.
69
Figure 2. 11 KIF21A expression level in MCP2 cells following synthetic knock-in of Hsa-mir-9.
Figure 2.11 KIF21A expression level in MCP2 cells following synthetic knock-in of hsa-mir-9. RNA isolated from MCP2 cells before and after six days of transfection with synthetic hsa-mir-9 was subjected to reverse transcription and quantitative SYBR green real time PCR. Bars represent replicate KIF21A expression values relative to GAPDH each measured in triplicate with vertical lines representing relative standard deviation values. * denotes significant repression with a p value of less than 0.05.
70
A.
B.
Figure 2.12 Quantitative expression of KIF21A and hsa-mir-9-3 in a panel of 25 ovarian cancer patients.
71
Figure 2.12 Quantitative expression of KIF21A and hsa-mir-9-3 in a panel of
25 ovarian cancer patients. RNA was isolated from a panel of a panel of 25 ovarian cancer patients and plotted relative to the patient with the lowest detected level of expression for each product. Patient 55 had the lowest detectable level of KIF21A while patient 41 had the lowest detectable level of hsa-mir-9-3, and as such all expression values were plotted relative to their expression level. The 13 patients lacking detectable levels of hsa-mir-9-3 were plotted relative to increasing levels of KIF21A (A) while the 12 patients with detectable levels of hsa-mir-9-3 were plotted relative to increasing levels of hsa- mir-9-3 (B). Each bar represents amplification from two independent reverse transcription reactions performed in triplicate with vertical lines indicating relative standard deviation.
72
A.
B.
Figure 2.13 Comparison of KIF21A and hsa-mir-9-3 expression levels in primary ovarian cancer tumors isolated from patients.
73
Figure 2.13 Comparison of KIF21A and hsa-mir-9-3 expression levels in primary ovarian cancer tumors isolated from patients. Relative expression values of hsa-mir-9-3 were plotted along the X axis while relative KIF21A expression values were plotted along the Y axis. A linear trendline was fitted to the data points to show the overall trend of KIF21A expression relative to hsa- mir-9-3 levels. R2 value for the displayed trendline is displayed on the graph.
Figure A shows all 25 patients regardless of if hsa-mir-9-3 was detected. Figure
B excludes those patients which lacked detectable levels of hsa-mir-9-3.
74
Figure 2.14 Smudge plots of differential methylation hybridization microarrays for the hsa-mir-9-3 locus in OCIC and bulk tumor samples.
Figure 2.14 Smudge plots of differential methylation hybridization microarrays for the hsa-mir-9-3 locus in OCIC and bulk tumor samples.
White-red smudge plot generated based on DMH M-score analysis of the hsa- mir-9-3 locus for six sets of DNA isolated from an ovarian cancer initiating cells population isolated from primary ovarian cancer tumors, and DNA from matched bulk tumor. DMH probe locations for the CpG island (green) are shown in red
75 with the Hsa-mir-9-3 microRNA shown as a blue rectangle with an arrow denoting direction of transcription. All objects are drawn to scale.
76
Figure 2.15 Quantitative Sequenom MassARRAY analysis of hsa-mir-9-3 locus in 12 sets of bulk primary tumors and subpopulation of OCIC samples.
Figure 2.15 Quantitative Sequenom MassARRAY analysis of hsa-mir-9-3 locus in 12 sets of bulk primary tumors and subpopulation of OCIC samples. The two regions of the hsa-mir-9-3 locus which were previously studied in the ovarian cancer cell line panel and primary patient samples, were each PCR amplified in a set of 12 primary tumors with both bulk DNA and DNA from a subpopulation of cancer initiating cells and subjected to quantitative methylation analysis on Sequenom‘s MassARRAY platform. Each amplicon
77 contained six and nine ‗CpG units‘ respectively each with between one and four
CpG dinucleotides and 30 total CpG dinucleotides. As CpG units are generated based on RNase A digested products, the methylation of individual CpG dinucleotides cannot be distinguished from one another and as such only the average methylation level of all CpG dinucleotides on the unit is reported. Braces are used to denote CpG units containing more than one CpG dinucleotide
(vertical lines). Each colored circle represents the level of methylation rounded to the nearest percent as determined by the adjusted methylation value determined based on raw values read on the EpiTYPER 1.0 software at the time of MALDI-
TOF analysis and standard curve adjustment and linear transformations. Circles with an X rather than a color denote CpG units which were unable to be analyzed due to one or more factors including poor amplification, potential non-CpG methylation, or potential SNPs.
78
Gene Fold PicTar TargetScan miRanda KIAA1546 -2.68 Yes BTG4 -2.26 C9orf100 -2.17 CENPA -2.10 CDCA3 -2.04 CDKN3 -2.01 NASP -2.01 PLK1 -2.01 CDC20 -1.98 TOP2A -1.98 KIF21A -1.97 Yes Yes Yes ASPM -1.96 AURKA -1.95 CDC25C -1.95 TAF15 -1.95 TROAP -1.93 BIRC5 -1.92 DLG7 -1.91 GTSE1 -1.91 UBE2C -1.88 CDKN3 -1.87 PARP12 -1.87 CCNA2 -1.85 CENPE -1.85 Yes FGF7 -1.85 HAND1 -1.85 HMMR -1.85 Yes MKI67 -1.85 KIF2C -1.84 IGHA1 -1.83 IGHA2 -1.83 NUSAP1 -1.82 CDCA8 -1.81 Yes HMGA2 -1.81 Yes Yes PIF1 -1.80 Yes Table 2.1 List of genes downregulated at least 1.8 fold in MCP2 cells following transfection with synthetic hsa-mir-9, and their overlap with various target prediction programs.
79
Table 2.1 List of genes downregulated at least 1.8 fold in MCP2 cells following transfection with synthetic hsa-mir-9, and their overlap with various target prediction programs.
80
Chapter 3
ChIP-seq Mapping of the TGFβ Pathway, Combined with Gene Expression
Profiling and In Silico Data Mining, Identifies Clinically Relevant SMAD4
Target Genes in Ovarian Cancer
3.1 Introduction
The transforming growth factor-β (TGFβ) signaling pathway plays an important role in controlling proliferation, differentiation, and other cellular processes including the growth of ovarian surface epithelial cell (OSE)
[(Berchuck, et al., 1992); (Wong & Leung, 2007)]. Dysregulation of TGFβ signaling is frequently observed in epithelial ovarian cancer (EOC) and may be crucial to EOC development [(Derynck, Akhurst, & Balmain, 2001); (Nilsson &
Skinner, 2002)]. The effects of TGFβ are mediated by three TGFβ ligands —
TGFβ1, TGFβ2 and TGFβ3, acting through TGFβ type 1 and type 2 receptors
[(Heldin, Miyazono, & Dijke, 1997); (Shi & Massague, 2003); (Feng & Derynck,
2005)]. TGFBR2 is the specific receptor for TGFβ ligands. The functional receptor complex regulates the activation of downstream SMAD and non SMAD pathways (Derynck & Zhang, 2003). The phosphorylated type 1 receptor recruits and phosphorylates receptor-regulated Smads (R-Smads). Of the five R-Smads
81 in mammals, the TGFBR2–ALK5 complex activates SMAD2 and SMAD3, whereas the TGFBR2–ALK1 complex activates SMAD1, SMAD5, and SMAD8
(Miyazawa, Shinozaki, Hara, Furuya, & Miyazono, 2002). Activated R-Smads form heteromeric complexes with the common partner SMAD (co-SMAD; SMAD4 in mammals) and translocate into the nucleus (Shi & Massague, 2003). As the affinity of the activated SMAD complex for the SMAD-binding element is insufficient to support association with endogenous promoters of target genes,
SMAD complexes must associate with other DNA binding transcription factors to regulate expression. Numerous studies have shown that various families of transcription factors, such as the forkhead, homeobox, zinc finger, LEF1, Ets, and basic helix–loop–helix (bHLH) families, can serve as SMAD4 partner proteins to achieve high affinity and selectivity for target promoters with the appropriate binding elements [(Koinuma, et al., 2009); (Koinuma, Tsutsumi,
Kamimura, Imamura, Aburatani, & Miyazono, 2009); (Qin, et al., 2009);
(Ikushima, et al., 2008); (Gomis, et al., 2006)].
The A2780 human epithelial ovarian cancer cell line is sensitive to cis- diamminedichloroplatinum (II) (cisplatin), one of the platinum-type agents
(carboplatin or cisplatin) used in the treatment of ovarian cancer. In addition to serving as a useful model for studying drug-sensitive disease, A2780 cells only display partial TGFβ dysregulation, indicated by retaining a modest increase in
SMAD4 expression and translocation of existing SMAD4 from the cytoplasm to the nucleus following TGFβ stimulation (Chan, et al., 2008). Thus, this cell line is
82 also an appropriate model system for carrying out genome-wide mapping of
SMAD4 target genes and identifying deregulated TGFβ /SMAD4- genes and pathways implicated in ovarian cancer patients.
Recent comparisons of ChIP-seq (chromatin immunoprecipitation- sequencing) to array-based approaches clearly demonstrated that ChIP-seq technology yielded higher resolution, greater depth, and improved mapping accuracy of transcription factor binding and histone modifications on a genome- wide scale [(Barski, et al., 2007); (Johnson, Mortazavi, Myers, & Wold, 2007);
(Mikkelsen, et al., 2007)]. In the current study, we used ChIP-seq technology to study TGFβ/SMAD4 regulation in the platinum-sensitive A2780 ovarian cancer cell line. We profiled SMAD4 binding loci following stimulation with TGFβ. Using computational approaches, we have investigated the SMAD4 binding pattern and compared it with the SMAD4 binding pattern of both a normal immortalized ovarian surface epithelial cell (IOSE) from our previous study (Qin, et al., 2009) and human keratinocytes (HaCaT) from Koinuma et al (Koinuma, Tsutsumi,
Kamimura, Imamura, Aburatani, & Miyazono, 2009). Further, we generated
TGFβ/SMAD4-regulated gene signatures and utilized an in silico mining approach to correlate the identified signatures with clinical outcome data from two publicly available ovarian cancer patient cohorts. Our integrative approach revealed significant associations of TGFβ/SMAD4 regulatory networks with both progression free and overall survival in ovarian cancer patients. By identifying thousands of SMAD4 binding loci as well as regulated genes, our data provide
83 both a new resource for studying the mechanism underlying dysregulated TGFβ signaling in ovarian cancer cells as well as potential prognostic biomarkers for future ovarian cancer translational research.
3.2 Results
3.2.1 Genome-wide SMAD4 occupancy defined by ChIP-seq technology.
Our previous studies [(Qin, et al., 2009); (Li, et al., 2009); (Chan, et al.,
2008); (Chou, et al., 2010)] and others [(Wong & Leung, 2007); (Nilsson &
Skinner, 2002); (Yamada, Baldwin, & Karlan, 1999); (Baldwin, Tran, & Karlan,
2003); (Tanaka, Kobayashi, Suzuki, Kanayama, & Terao, 2004)] have tried to establish and characterize the molecular mechanisms of dysregulated TGFβ- mediated signaling in ovarian cancer cells and acquired cisplatin-resistant ovarian cancer cells. In order to further elucidate the details of the underlying mechanisms, we used ChIP-seq technology to identify the genomic locations bound by SMAD4 in A2780 cells before and after TGFβ stimulation.
Using ChIP-seq, all samples were initially sequenced to generate a set of raw reads (each with a length of 36 bp) from Illumina GAIIx system ranging from
~43 million to ~51 million reads per sample. After mapping to UCSC Human
HG18 assembly, a set of ~26 million and ~32 million mapped reads with unique genomic locations were obtained for unstimulated A2780 and TGFβ-stimulated
84
A2780 respectively (see Table 3.1). We then applied our peak-calling detection program, BELT, [(Lan, Bonneville, Apostolos, Wang, & Jin, 2011); (Frietze, X,
Jin, & Farnham, 2010)] to identify the binding loci of SMAD4 in these two conditions. Briefly, our BELT program uses a percentile scoring method to determine the enrichment threshold value for each of the top percentiles from all binding regions, followed by identifying the number of binding loci at each level.
In order to determine the significance of each percentile, a set of randomly simulated reads is used as a background to estimate the false discovery rate
(FDR). Our ChIP-seq data confirmed multiple SMAD4 binding loci previously identified in different tissues and cell types including Gadd45A, CTGF, JAG1,
LEMD3 (Gomis, et al., 2006), MYC (Lim & Hoffmann, 2006), EDN1, RYBP, DST, and BCAT1 (Koinuma, Tsutsumi, Kamimura, Imamura, Aburatani, & Miyazono,
2009).
3.2.2 Examination of SMAD4 occupancy prior to TGFβ stimulation reveals distinct binding pattern in basal state.
We identified 2,009 SMAD4 binding loci in the basal (unstimulated) condition in the A2780 cell line (see Table 3.1). We found that 1,499 (74.6%) loci were located within +/-100 kb of a known RefSeq gene (Kent, et al., 2002).
Surprisingly, only small portion (267 of 1499, 13.3%) were within the promoter region (+/-8kb), of a gene while the majority of binding loci were either 10 kb
85 upstream of the 5‘TSS or 10 kb downstream 3‘TTS (see Figure 3.1 – red line).
This unbiased whole genome-wide location analysis suggested that many other previous genome-wide studies based on promoter ChIP-chip technology
[(Koinuma, Tsutsumi, Kamimura, Imamura, Aburatani, & Miyazono, 2009); (Qin, et al., 2009); (Fei, et al., 2010)] may only identify subsets of SMAD4 target genes.
3.2.3 Examination of SMAD4 occupancy after TGFβ-stimulation reveals a similar distribution of loci to the basal state albeit among drastically different genes.
Upon stimulation with TGFβ, 2,362 SMAD4 binding loci were identified.
Overall, the distribution of the location of SMAD4 binding loci after TGFβ stimulation is very similar to the one before stimulation (see Table 3.1 and Figure
3.1 – black line). However, the binding patterns between two conditions (before and after TGFβ stimulation) are dramatically different (see Figure 3.2). We first removed the binding loci located far away from any known RefSeq genes (+/-100 kb) and then classified them (1,723 loci for stimulated and 1,499 loci for unstimulated) into four different binding patterns: 1) Basal Binding -- two binding loci are associated with same gene and within one kb distance of each other (i.e. unchanged binding); 2) Shift Binding – two binding loci are associated with same gene in both conditions, but they are more than one kb apart from one another;
86
3) Stimulated Only Binding – a binding loci associated with a gene only in the stimulated condition; 4) Unstimulated Only Binding – a binding loci associated with a gene only in the unstimulated condition. Based on the above classification, we determined that 74.2% (1,279 of 1,723) and 73.5% (1,102 of 1,499) of the binding loci were in the Stimulated Only Binding and Unstimulated Only Binding categories respectively. While 24.8% (429 of 1,723) and 25.5% (382 of 1,499) binding loci were classified into the Shift Binding category for the stimulated and the unstimulated condition respectively, only 15 binding loci in each condition
(0.9% and 1.0% respectively) fell into the Basal Binding category. Our genomic mapping results showed that TGFβ stimulation of ovarian cancer cells may alter the landscape of SMAD4 binding patterns.
Further, in order to verify that TGFβ stimulation resulted in the binding changes we observed in the ChIP-seq data, we randomly chose a set of 10 targets identified by our analysis and performed ChIP-qPCR using DNA isolated from an immunoprecipitation that was distinct from the DNA used for ChIP-seq.
Our ChIP-qPCR validations not only confirmed the targets identified in the ChIP- seq data but also demonstrated that the activated exogenous TGFβ signaling indeed induced SMAD4 binding into one of four classified binding patterns (see
Figure 3.3).
87
3.2.4 Regulation of TGFβ-stimulated SMAD4 target gene expression in
A2780.
Next, we performed gene expression microarrays to determine the expression status for SMAD4 target genes after TGFβ stimulation. A2780 mRNA from three independent biological replicates of both before and after three hours of TGFβ stimulation was prepared and assayed on Affymetrix U133 Plus 2
Platform. Overall, 3,191 genes were identified as being significantly up or down- regulated after TGFβ stimulation with at least a 0.5 Log2-fold change in expression and a p value of less than 0.1 (see Figure 3.4). After examining the correlation with 1,443 TGFβ-stimulated SMAD4 target genes (corresponding to
1,723 SMAD4 binding loci in the stimulated condition), a majority (2,873 of 3,191) of genes with differential expression in A2780 surprisingly lacked SMAD4 binding loci, where 318 genes had at least one SMAD4 binding loci and showed at least a 0.5 Log2-fold expression change after three hours of TGFβ stimulation (see
Figure 3.5).
Gene ontology analysis showed that the differentially expressed genes with SMAD4 binding loci were significantly enriched for genes involved with cell part morphogenesis and developmental proteins (see Figure 3.6 - Gene &
Binding Locus), in line with the previous studies in different cell types [(Qin, et al.,
2009); (Fei, et al., 2010)]. We also found that SMAD4 binding associated genes lacking differential expression were enriched for genes with an EGF-like domain
(and EGF polymorphic variants) suggesting that different signaling pathways may
88 mediate SMAD4 functions other than TGFβ signaling (see Figure 3.6 - Binding
Locus only). The largest set of genes (those with differential expression, but
SMAD4 binding loci) was involved in immune functions and proteinaceous extracellular matrix (see Figure 3.6 - Gene Only).
To further confirm differential expressed SMAD4 targeted genes resulting from TGFβ stimulation, we randomly chose a set of 18 targets identified by our analysis and performed RT-qPCR. More than 70% (13 of 18) genes were validated by RT-qPCR as shown in Figure 3.7 (the five genes which did not agree with the microarray results were not shown).
3.2.5 SMAD4-dependent gene regulatory networks in TGFβ-induced ovarian cancer cells.
Our previous study (Qin, et al., 2009) and a study from Koinuma et al
(Koinuma, Tsutsumi, Kamimura, Imamura, Aburatani, & Miyazono, 2009) have identified a set of 150 TGFβ stimulated SMAD4 target genes in IOSE (an immortalized ovarian surface epithelial cell line) and a set of 92 TGFβ stimulated
SMAD4 target genes in HaCaT (an immortalized keratinocyte cell line). It was not surprising to find limited overlap of only six of 150 in IOSE, six of 92 in HaCaT, and one for all three studies in common with the 318 SMAD4 target genes in this study (see Figure 3.8) as only one, A2780, is a cancer cell line and the other two are non-tumorigenic immortalized cell lines. Other possibilities for such low
89 overlapping rates include: limited targets identified using promoter array (ChIP- promoter-chip) as they required pre-selection of loci to interrogate as compared to unbiased ChIP-Seq, and the use of different SMAD4 antibodies in the different studies. GO analysis (Huang, Sherman, & Lempicki, 2009) also showed target genes in HaCaT and IOSE were primarily involved in regulation of cell proliferation (or anti-apoptosis) and development process (muscle development), which were different from target genes in A2780 (see Figure 3.9).
To further compare the difference of the TGFβ-stimulated SMAD4- dependent gene regulatory information between these three cell types, we applied a computational analytical approach we previously developed (Gu, et al.,
2010) to build the SMAD4-dependent regulated networks in HaCaT, IOSE, and
A2780 respectively (see Figure 3.10). Briefly, our computational analytical approach started with ChIP based datasets and gene expression data. Each
SMAD4 binding locus was matched to known a RefSeq gene ID which were then examined for differential gene expression. A set of differentially expressed
SMAD4 target genes after TGFβ stimulation were further used for finding the most significant transcription factor (TF) binding partners by ChIPMotifs (Jin,
Spostolos, Nagisetty, & Farnham, 2009) or ChIPModudles (Jin, Rabinovich,
Squazzo, Green, & Farnham, 2006), which were used as Hub TFs. The Hub TF- gene connection was determined by scanning the Hub TFs‘ position weight matrixes (PWMs) in all binding loci and a permutation test was used to test the
90 reliability of each connection of the network. The resulted regulatory network was visualized by Cytoscape (Shannon, et al., 2003).
We identified six Hub TFs, GFI1, NR3C1, SOX17, STAT4, ZNF354C, and
TCF8 from 318 SMAD4-dependent target genes in A2780 cells, while four Hub
TFs, LEF1 (TCF), ELK1, COUPTF (NR2F5), and E2F, were identified in IOSE cells by our previous study using a similar approach (CART model) (Qin, et al.,
2009). Our computational analytical approach also identified three Hub TFs,
E2F1, SP1, and USF for 92 SMAD4-dependent target genes in HaCaT cells, which was very similar to the TF motifs identified from the Koinuma et al. study
(Koinuma, Tsutsumi, Kamimura, Imamura, Aburatani, & Miyazono, 2009). The top motif reported in their study, AP1, was missed in our results due to using an advanced classification algorithm in our ChIPModules (Jin, Rabinovich, Squazzo,
Green, & Farnham, 2006) and being able to eliminate those TF motifs which are also enriched in random sets. Interestingly, we also found one Hub TF E2F
(E2F1) was common between the two normal cells, but not in common with
A2780 cells. Together with GO function analysis, our results indicated that E2F may act as a major SMAD4 co-transcription factor partner in mediating cell proliferation in normal cells but lost in carcinoma cells. The resultant gene regulatory networks (GRN) for all three cells are shown in Figure 3.10. Overall, our gene regulatory network analysis strongly indicates that TGFβ stimulates a different SMAD4-dependent regulatory mechanism in ovarian cancer cells
91 compared to normal cells, i.e., the SMAD4 regulation network has become
―rewired‖ in ovarian cancer cells.
3.2.6 Gene signatures of selection and clinical outcome.
One of the promising potential applications of genome-wide ‗omics studies using cell line systems is identification of gene signatures that can provide better prognostic information compared with standard clinical and pathological parameters [(Van De Vijver, et al., 2002); (Chibon, et al., 2010)]. To address the relationship of TGFβ stimulated SMAD4-dependent target genes and clinical outcome of ovarian cancer patients, we examined the 307 target genes identified in A2780 cells in this study, which were not identified in previous studies of normal cells, in two different clinical ovarian cancer cohort studies that had reported survival data [(Bild, et al., 2006); (Lu, et al., 2004)]. For each data set, we first classified the patients into different sub-groups based on their gene signatures, and then correlated the data with the patient survival information. For the 153 patient cohort from Bild et al (Bild, et al., 2006), we were able to use 187 of the 307 genes identified in the gene expression dataset to apply the hierarchical clustering method and classify the genes into four gene clusters
(GCs) (see Figure 3.11). For each of the four gene clusters, we further clustered the 153 samples into four patient groups (PGs), and correlated the PGs with their survival information. We found a signature of a subset of 49 genes that was able
92 to predict a significant survival correlation for 62 of the patients with a p-value of less than 0.05 (see Figure 3.12 A, and B). Specifically PG4 (25 patients) displayed a poor median survival of 31 months compared to PG3 (37 patients), with a median survival of 63 months. Due to limited pathological information available for this patient cohort, we were not able to significantly correlate our gene signatures with other clinical outcomes. However, a notably high percentage of stage IV patients clustered into PG3 while all stage IC and two stage IIC patients clustered into PG4, despite a similar number of stage IIIC patients in each (Table 3.2), perhaps indicating that TGFβ/SMAD4 regulated genes could be potentially used to classify a subtype of ovarian cancer patients.
When we applied the same in silico mining approach to the second patient cohort from Lu et al (Lu, et al., 2004), (comprised of 42 patient samples and samples isolated from five normal people), the results showed that a gene signature of 19 of the 307 genes predicted better survival rates for PG4 and those without disease than the other PGs with a p value of 0.0078 (see Figure
3.13 and Figure 3.14).
3.3 Discussion
We have for the first time applied ChIP-seq technology to whole-genome- wide mapping of TGFβ-stimulated, SMAD4-dependent regulated genes in an ovarian cancer cell line (A2780). Our data show that compared to the basal state
93
(no TGFβ stimulation), a majority of SMAD4 binding loci are either newly bound to chromatin (74.2%) or shifted bound (24.8%) upon TGFβ stimulation, suggesting TGFβ stimulated cancer cells may alter the landscape of SMAD4 binding patterns. Further, our GO analysis revealed striking similarities between the top 10 GO categories for 1,443 and 1,316 SMAD4 target genes in stimulated and unstimulated conditions (data not shown). However, 318 differentially expressed genes, containing at least one stimulated SMAD4 binding loci, were significantly enriched for more specific GO terms, such as cell part morphogenesis and developmental proteins. This result indicates that SMAD4 may regulate a very specific set of target genes in response to TGFβ signaling, in order to facilitate specific functions in that cell type through this specific signaling pathway. Indeed, GO analysis for SMAD4 target genes without gene expression level changes after TGFβ stimulation found one of the enriched gene categories is ‗EGF like signaling‘, providing further evidence that other signaling pathways may modulate SMAD4-dependent regulated genes in ovarian cancer.
Similar to other findings for transcription factors, including estrogen receptor alpha (ERα) [(Carroll, et al., 2006); (Welboren, et al., 2009); (Fullwood, et al., 2009)], androgen receptor (AR) (Wang, et al., 2009), and peroxisome proliferator-activated receptor (PPAR) (Nielsen, et al., 2008), we observed that a majority (>70%) of SMAD4 binding loci located more than eight kb away from
5‘TSS of a known RefSeq gene. This might suggest the SMAD4 binding loci come in close proximity to the promoter through chromosome looping upon
94
TGFβ stimulation. Interestingly, our de novo motif analysis also identified a
SMAD-like motif in a set of 5-distal binding loci but not in a set of 5‘-promoter loci
(data not shown). Our genome-wide location analysis also pinpoints the importance of whole-genome-wide sequencing technologies, as we showed many binding loci are far away from the 5‘TSS of a known gene and therefore a promoter-array technology may miss many target binding loci of a transcription factor. Future studies will focus on conducting ChIP-3C-qPCR to confirm whether these distal binding loci are indeed related to these particular genes, potentially uncovering the underlying mechanism of TGFβ/SMAD4 mediated gene regulation.
One important aspect of this study is the use of in silico mining of publicly available patient cohort data to identify a subset of TGFβ/SMAD4 target genes as a gene signature for predicting clinical (survival) outcomes. As far as we know, this is the first study to attempt to use TGFβ signaling responsive SMAD4 regulated genes to classify ovarian cancer patients into different sub-types of patient groups, as well as predict poor survival from good survival populations with statistical significance (see Figure 3.12 A, B and Figure 3.14). Thus, combining ChIP-seq identified binding loci, gene expression profiling, and an in silico mining of patient cohorts may provide a powerful approach for identifying potential gene signatures with biological and clinical importance.
95
In conclusion, our study provides a comprehensive genome-wide map of thousands of TGFβ/SMAD4 targets in an ovarian cancer cell line, which could further be used for studying SMAD4 functions in tumorigenesis. To our knowledge, this is the first study to link TGFβ/SMAD4 regulated genes to clinical information on ovarian cancer patient survival and identify potential gene signatures for prognosis in ovarian cancer.
3.4 Materials and Methods
3.4.1 Cell Culture and TGFβ Stimulation
A2780 cells were cultured in RPMI 1640 (Invitrogen, Carlsbad, CA)
o supplemented with 10% fetal bovine serum in a 37 5%CO2 incubator. Prior to
TGFβ stimulation, cells were split at ~70% confluency and inspected daily. For
ChIP, 80% confluent cells were optimally stimulated with 10ng/ml recombinant
TGFβ1 (Sigma, St. Louis, MO) for one hour prior to formaldehyde cross-linking while expression analysis was performed after three hours of stimulation with
10ng/ml TGFβ1.
96
3.4.2 Chromatin Immunoprecipitation and Massive Parallel Sequencing
Chromatin immunoprecipitation (ChIP) was performed as previously described [(Cheng, et al., 2006); (Lee, Johnstone, & Young, 2006)] with some note worthy changes. Briefly, cells were rinsed with room temperature PBS before being cross-linked in a 1% formaldehyde solution. Cells were then harvested and homogenized in the presence of protease inhibitors before DNA was sonicated. Magnetic Dynal beads (Invitrogen) combined with a mixture of antibodies [20% SMAD4 #9515 (Cell Signaling Technology, Danvers, MA) and
80% SMAD4 DCS-46 (Santa Cruz Biotechnology, Santa Cruz, CA)] were used to pull down SMAD4 overnight. Purified DNA was used to detect fold enrichment by
SYBR Green qRT-PCR.
Sequencing libraries were generated for massive parallel sequencing using standard methods. Briefly, 500ng of pulldown DNA was subjected to end repair, terminal adenylation, and adapter ligation before fragments ranging from
~175-250 were isolated from a 2% E-gel (Invitrogen,). Subsequent to a standardized 12 cycle PCR, DNA quality was evaluated on a DNA 1000
Bioanalyzer chip (Agilent Technologies, Santa Clara, CA) before being submitted for sequencing on an Illumina GAII.
97
3.4.3 Gene Expression Profiling
Total RNA was extracted from cells using TRIzol (Invitrogen) for microarray analysis on an Affymetrix HGU133 Plus 2 arrays (which includes more than 55,000 probes corresponding to ~23,000 human genes on the chip) using standard hybridization and scanning protocols provided by Affymetrix
(Santa Clara, CA). RNA for each sample (TGFβ-stimulated and unstimulated conditions) was isolated in biological triplicate.
3.4.4 RT-qPCR and ChIP-qPCR
Total RNA was extracted from TGFβ stimulated and unstimulated cells using TRIzol (Invitrogen). cDNA was then generated from one µg of isolated RNA through reverse transcription with a mix of oligo-dT and random hexamers in the presence of SuperScript III (invitrogen). A Step One Plus instrument from Applied
Biosystems was used in conjuncture with SYBR Green reagents (Applied
Biosystems, Carlsbad, CA) to detect levels of gene expression. The ΔΔCt method of gene expression analysis was used to determine the relative level of expression as compared to the internal GAPDH control. A student‘s t-test was used to determine statistical significance of differential expression between biological triplicates.
Our SMAD4 ChIP-seq results were confirmed via ChIP-qRT-PCR. Briefly primers of approximately 150 bp were constructed to cover regions where were
98 sequenced in the ChIP-seq experiment, and amplified with SYBR Green (Applied
Biosystems). DNA was quantified on a nanodrop ND3300 (Thermo Scientific,
Waltham, MA) using a Quant-iT Picogreen dsDNA Assay Kit (Invitrogen), and equal quantities of DNA were used as template. Comparison of SMAD4 and IgG pulldowns for both stimulated and unstimulated samples are reported as the average value of triplicate measurements a student‘s t-test was used to determine statistical significance.
3.4.5 Processing ChIP-Seq and Microarray Gene Expression Data
A standard procedure for extracting image files, mapping the reads onto human genome, and filtering the mapped reads to unique reads was followed with the Solexa 1.6 pipeline. A total of three lanes of sequencing were run for
TGFβ stimulated and unstimulated samples. The reads from these four lanes were combined in to a single data set. Both samples in the combined data set were processed using BELT [(Lan, Bonneville, Apostolos, Wang, & Jin, 2011);
(Frietze, X, Jin, & Farnham, 2010)] developed in our laboratory with a 300 nucleotide bin size at an acceptance threshold of 0.996 vs. an input sample. For all datasets, we narrowed the peaks with 500bp in length.
The microarray expression data was normalized using the standard protocol for the MAS5 algorithm implemented by Affymetrix in R, and a student's t-test was performed to determine the significance of the difference between the
99 sets of biological triplicates for the stimulated and untreated samples.
Significance was liberally defined as p < 0.10, and a differential fold chance was defined as Log2-fold change > 0.50.
3.4.6 Gene Regulatory Network Analysis
We apply our computational analytical approach developed in our laboratory (Gu, et al., 2010), which includes a de novo method ChIPModule (Jin,
Rabinovich, Squazzo, Green, & Farnham, 2006) to identify the Hub TFs for 318
TGFβ/SMAD4 genes in A2780 cells and 92 TGFβ/SMAD4 genes in HaCaT cells respectively. The Hub TFs for 150 TGFβ/SMAD4 genes in IOSE cells were from our previous study (Qin, et al., 2009) which used a similar machine learning approach, CART model (Breiman, Friedman, Olshen, & Stone, 1984). The gene regulatory networks were constructed by scanning the binding loci of each gene using the position weight matrix (PWM) of Hub TFs. The topology and visualization of the resulted hierarchal network is built by Cytoscape (Shannon, et al., 2003), where blue nodes represent Hub TFs, while red and green nodes correspond to Up and Down regulated genes respectively. The significance of the network is statistically tested by a permutation test to determine the probability of each edge of the network under random circumstances.
100
3.4.7 Patient Cohorts
The patient cohorts including gene expression and survival information for patients used in this work were from two previous studies, Bild et al (Bild, et al.,
2006) and Lu et al (Lu, et al., 2004), which are publicly available. All gene expression data were previously normalized and were directly used in our study
(153 patients in Bild et al from GSE3149 and 42 patients for Lu et al provided in their Supplemental Tables).
For all hierarchical cluster analyses, log expression values of each gene were mean centered, and genes and tumors were clustered by using Pearson correlation and average linkage (MatLab). The Kaplan–Meier estimate was used to compute survival curves using logrank test, and the p-value of the likelihood- ratio test was used to assess statistical significance. The survival curves were generated using Prism 5 (GraphPad Software).
101
Figure 3.1 Distribution of SMAD4 binding loci relative to gene body.
Figure 3.1 Distribution of SMAD4 binding loci relative to gene body.
Distribution of SMAD4 binding loci for both stimulated and unstimulated conditions based on their relative position to the transcription start site of the closest known RefSeq gene. Peaks mapping to a gene body are plotted as a percent of gene body. Horizontal scale is different for each section as listed.
102
Figure 3.2 Classification of SMAD4 binding patterns relative to opposite condition.
Figure 3.2 Classification of SMAD4 binding patterns relative to opposite condition. SMAD4 binding loci were classified into one of four categories based on SMAD4 binding near the same location in the opposite treatment. SMAD4 binding loci which had no detected binding associated with the same RefSeq gene in the opposite treatment are designated ―Only‖ and represented in yellow for either the stimulated condition (left) or unstimulated condition (right). SMAD4 binding loci detected on the same gene in each condition within 1,000 bp of each other are designated ―Basal‖ and represented in blue for both treatment conditions. Similarly, ―Shift‖ binding loci (shown in red) represent loci which have binding associated with the same gene in the opposite treatment condition, but the loci are more than 1,000 bp apart. Differences in the number of ―Shift‖ binding
103 loci for each condition represent situations of unequal number of binding loci associated with a single gene.
104
Figure 3.3 ChIP-PCR of selected target genes identified by ChIP-Seq.
Figure 3.3 ChIP-PCR of selected target genes identified by ChIP-Seq. Bars represent abundance of DNA present following SMAD4 ChIP pull down as compared to DNA present following pull down with non-specific IgG antibody as determined by quantitative SYBR Green PCR. U and S used to represent regions of the gene with associated binding in the Unstimulated and Stimulated conditions of SLC40A1 and PARK2 respectively. Vertical bars represent relative standard deviation among reactions run in triplicate. * represents a t-test p-value of less than 0.05 and denotes significant enrichment relative to IgG control.
105
Figure 3.4 Heatmap of expression microarray results for TGF β stimulated and unstimulated A2780 cells.
Figure 3.4 Heatmap of expression microarray results for TGF β stimulated and unstimulated A2780 cells. A heatmap of the expression fold changes for genes between the unstimulated and the TGFβ stimulated condition, showing three group of genes, up-regulated, no change, and down-regulated. Up and down regulated genes are defined as having a Log2 fold change of greater than
0.5 or less than -0.5 respectively.
106
Figure 3.5 Overlap of expression microarray results with ChIP-sequencing results.
Figure 3.5 Overlap of expression microarray results with ChIP-sequencing results. A comparison between the genes with SMAD4 binding loci (1443) in the
TGFβ stimulated condition with all genes showing differential expression (3193), showing three different groups, those with differential expression and no SMAD4 binding loci, those with no differential expression and a SMAD4 binding loci and those with both.
107
Figure 3.6 Go annotation comparison of genes with an expression change, SMAD4 binding, or both.
Figure 3.6 GO annotation comparison of genes with an expression change,
SMAD4 binding, or both. GO annotations for the three different groups of genes shown in the Venn diagram of Figure 3.5. Increased length of blue bar denotes decreased significance of association.
108
Figure 3.7 qRT-PCR of selected targets showing differential expression and SMAD4 binding upon TGFβ stimulation.
Figure 3.7 qRT-PCR of selected targets showing differential expression and
SMAD4 binding upon TGFβ stimulation. RNA expression level as determined by qRT-PCR relative to GAPDH expression levels. Bars represent average expression values of biological triplicates performed in triplicate (n=9). Vertical bars represent relative standard deviation while * represents a t-test p-value of less than 0.05 and denotes significant difference in expression between the unstimulated and stimulated conditions.
109
Figure 3.8 Comparison of TGFβ/SMAD4 regulated genes in the A2780 ovarian cancer cell line to previously reported genes in normal ovarian cell line and skin cell line.
Figure 3.8 Comparison of TGFβ/SMAD4 regulated genes in the A2780 ovarian cancer cell line to previously reported genes in normal ovarian cell line and skin cell line. A Venn diagram showing the number of genes regulated by TGFβ/SMAD4 in three different cell types and which genes were regulated in multiple cell types. Green denotes the A2780 ovarian cancer cell line, immortalized ovarian surface epithelial cell is shown in red, and a normal skin cell line is shown in blue.
110
Figure 3.9 Comparison of the GO annotation of TGFβ/SMAD4 target genes in three cell types.
Figure 3.9 Comparison of the GO annotation of TGFβ/SMAD4 target genes in three cell types. GO annotations genes regulated by TGFβ/SMAD4 in the three different cell types (normal/cancerous ovary, and skin). Increased length of blue bar denotes increased fold enrichment of each annotation relative to random enrichment of all annotation while p values are listed to denote relative significance of association.
111
A.
Figure 3.10 (continued)
112
Figure 3.10 (continued)
B.
Figure 3.10 (continued)
113
Figure 3.10 (continued)
C.
Figure 3.10 TGFβ/SMAD4 gene regulatory networks of different tissue types.
Figure 3.10 TGFβ/SMAD4 gene regulatory networks of different tissue types. Cytoscape images of genes regulated by SMAD4 upon stimulation with
TGFβ for three different cell types. A. represents regulatory network of A2780 cells and work done for the current study, B. represents regulatory network
114 determined based on the work of (Koinuma, Tsutsumi, Kamimura, Imamura,
Aburatani, & Miyazono, 2009) in skin cells, and C. represents regulatory network determined for IOSE cells previously (Qin, et al., 2009) in our laboratory.
115
Figure 3.11 Hierarchical clustering of patients based on expression of TGFβ/SMAD4 target genes.
Figure 3.11 Hierarchical clustering of patients based on expression of
TGFβ/SMAD4 target genes. Hierarchical clustering of 187 of the 307 genes uniquely regulated in A2780 cells following TGFβ stimulation into four gene clusters (designated G1, G2, G3, and G4). The vertical axis represents the gene clusters (187 genes) and the horizontal axis stands for diverse samples (153 patients).
116
A.
B.
Figure 3. 12 Kaplan-Meier survival curve for patients displaying differential expression of a 49 gene signature.
117
Figure 3.12 Kaplan-Meier survival curve for patients displaying differential expression of a 49 gene signature. The expression of the 49 genes associated with the G2 gene cluster identified in Figure 3.11 was used to cluster 153 patients into one of four patient groups (PG1, PG2, PG3, and PG4). The horizontal axis denotes survival time in months while the vertical axis represents percent of surviving patients for each patient group. Survival curve for all patient groups is shown in Figure A. Survival of patient groups 3 and 4 is shown in more detail in along with the corresponding log-rank test p-value of 0.0471 in Figure B.
118
Figure 3.13 Hierarchical clustering based on expression of TGFβ/SMAD4 target genes for additional patient data set.
Figure 3.13 Hierarchical clustering based on expression of TGFβ/SMAD4 target genes for additional patient data set. The same methods as described for Figure 3.11 were performed for a separate patient data set with similar results. In this case, 278 of the 307 genes which were regulated by SMAD4 following TGFβ stimulation were clustered into four gene clusters as denoted
119 along the left axis. As previously, the vertical axis represents gene clusters and the horizontal axis represents individual patient samples (n=42).
Figure 3.14 Kaplan-Meier survival curve for patients with differential expression of a 19 gene signature.
Figure 3.14 Kaplan-Meier survival curve for patients with differential expression of a 19 gene signature. The expression of 19 genes associated with the G4 gene cluster identified in Figure 3.13 was used to cluster 42 patients into one of four patient groups as denoted by different colored survival curves.
The horizontal axis denotes survival time in months while the vertical axis denotes percent of surviving patients. The logrank test P value of 0.0078 is displayed to denote significance.
120
Raw Mapped Belt Identified Loci Near Condition Reads Reads Binding Loci Gene Stimulated ~51 Million ~32 Million 2,362 1,723 Unstimulated ~43 Million ~26 Million 2,009 1,499 Table 3.1 Details of SMAD4 ChIP-sequencing results.
Table 3.1 Details of SMAD4 Chip-sequencing results. Raw reads represent
36 bp single end reads. Mapped reads represent reads which mapped uniquely to the human genome. Loci near gene represent belt determined binding loci which are located within 100kb of the nearest gene.
121
Tumor Stage Median Patient Survival Group Unstaged months IC (3) IIC (4) IIIA (3) IIIB (5) IIIC (91) IV (17) (1)
PG1 (29) 37 0 1 0 0 24 4 0
PG2 (33) 23 0 1 0 1 27 4 0
PG3 (37) 31 0 0 3 2 24 7 1
PG4 (25) 63 3 2 0 2 16 2 0
Table 3.2 List of 124 patients’ tumor stage and median survival time for each patient group determined based on Figure 3.11 hierarchical gene clustering of TGFβ stimulated SMAD4 responsive genes.
Table 3.2 List of 124 patients’ tumor stage and median survival time for
each patient group determined based on Figure 3.11 hierarchical gene
clustering of TGFβ stimulated SMAD4 responsive genes. While the entire
data set contained 153 patients, survival information was only available for 124 of
those patients. Patients without survival information were excluded from analysis.
The number listed in parentheses for each tumor stage denotes the number of
patients with that stage of tumor in the entire patient cohort, and the number
listed in parentheses for each patient group represents the number of patients
associated with that patient group.
122
Chapter 4
Hypermethylation of CLDN11 Associated with Cisplatin Resistance in
Ovarian Cancer and Loss of Expression Associated with Increased Tumor
Grade and Mobility.
4.1 Introduction
As with many cancers, early detection of ovarian cancer has been consistently associated with the best clinical outcomes (Agarwal & Kaye, 2003).
Cisplatin and carboplatin are the primary chemotherapeutic agents used in the treatment of primary ovarian cancer with more than 90% of patients entering remission if caught in the early stage. Additionally more than 80% of ovarian cancer patients have been shown to relapse after successful treatment with a chemoresistant form of ovarian cancer (Agarwal & Kaye, 2003). With remission being tragically short lived for a significant number of people, and with that remission being refractory to chemotherapeutic treatments, significant work is dedicated to finding new drug treatment options. An alternative approach to novel drug discovery to treat refractory cancers is finding ways to re-sensitize cells to the drugs which the cancer is refractory to. If cells could be re-sensitized to the platinum based drugs which enable most people to enter remission, those same
123 platinum based drugs could continue to be used with to achieve remission after patients relapse with refractory disease.
Studies have shown that epigenetics can be involved in nearly all aspects of cellular biology including being associated with drug resistance. Additional identification of the epigenetic abnormalities associated with platinum based drug resistance is of critical importance if tumors are to be re-sensitized to those drugs. While significant contributions have been made by using limited approaches targeting particular genes implicated as being of interest for various reasons, here we made use of the largely unbiased differential methylation hybridization (DMH) data available for a cell line culture model system (Li, et al.,
2009). DMH is a microarray based methodology of analyzing the methylation state of restriction endonuclease sites within CpG islands previously developed by our lab (Deatherage, Potter, Yan, Huang, & Lin, 2009). The model system itself is interesting in that it makes use a cisplatin resistant cell line subpopulation isolated from the cisplatin sensitive cell line A2780.
While other comparisons between cisplatin sensitive and resistant cell lines have been used to produce important findings, comparing different cell lines as a model system can be problematic in that the lines are isolated from different primary tumors with different genetic backgrounds making analysis more complex as differences between cell lines may or may not be caused by the difference in cisplatin sensitivities. The use of the cisplatin resistant A2780 cell
124 line allows for a greater degree of certainty that differences between the lines are specifically related to cisplatin resistance. Importantly, the ―round 5 cells‖ display a cisplatin 50% growth inhibition (GI50) resistance level of 35 μM which is consistent with doses used in other cisplatin resistant cell lines implying that differences associated with cisplatin resistance are not an artifact of high or low drug levels (Li, et al., 2009).
In 1998 Furuse et al. identified the first two members of the claudin gene family in studies of chicken liver which is known to be highly enriched for tight junctions with subsequent homologue studies identifying both proteins human tissue (Furuse, Fujita, Hiragi, Fujimoto, & Tsukita, 1998). To date the claudin gene family has been found to contain more than 20 different members which have been shown to be critical for proper formation of tight junctions in the paracellular barrier of epithelial layers (Rüffer & Gerke, 2004). CLDN11 has recently been shown to play a role in cell-cell adhesion and with gene silencing leading to increased motility and invasion in a cell culture model and epigenetic silencing through hypermethylation of the CLDN11 locus playing a role in gastric cancer (Agarwal, et al., 2009).
In this section we will discuss the findings of a gene, CLDN11, which was shown to be epigenetically downregulated via hypermethylation in cisplatin resistant A2780 round 5 cells as compared to parental cells. This finding was confirmed in a panel of ovarian cancer patients, the hypermethylation of the gene
125 was shown to be associated with a loss of expression. Hypermethylation of the
CLDN11 locus was also detected in a panel of 78 primary ovarian cancer tumors and repression was demonstrated to be of statistical significance in a patient cohort relative to survival with lower levels of CLDN11 being associated with an increased tumor grade. While somewhat counter intuitive, we hypothesize that the loss of this cell-cell adhesion molecule provides additional resistance to cisplatin by allowing for better efflux of cisplatin out of the cell.
4.2 Results
4.2.1 Differential methylation of A2780 R0 and A2780 R5 cells leads to changes in expression.
Cisplatin sensitive A2780 Round 0 (R0) cells and cisplatin resistant A2780
Round 5 (R5) cells were previously subjected to DMH in an effort to uncover
DNA methylation changes associated with increased cisplatin resistance (Li, et al., 2009). To focus on methylation events which had a functional expression consequence, the DMH results were compared to expression microarray results from the same study. Figure 4.1 shows a breakdown of the 1,936 genes which were over expressed in R0 cells, and 1,362 genes which were over expressed in round 5 cells and the methylation status as measured by DMH. We found 105 over expressed genes in R0 cells which were hypermethylated in R5, and 138
126 hypermethylated genes in R0 cells which were over expressed in R5 suggesting the acquisition or loss of methylation during the increase in cisplatin resistance was responsible for the change in expression. An additional 284 genes were found to be either over expressed while being hypermethylated or repressed while being hypomethylated. As the behavior of these genes suggests a more complex mechanism than strict DNA methylation leading to epigenetic gene silencing (such as secondary transcription factor involvement), they were not studied.
4.2.2 CLDN11 and SLC27A6 are repressed upon gain of cisplatin resistance in a cell culture model.
We chose to focus primarily on the 105 genes which became hypermethylated following cisplatin treatment and were silenced. Of these 105 genes we decided to test the expression of 18 genes we had additional interest in for both A2780 R0 and R5 cells to validate the microarray findings (see Figure
4.2). While six genes displayed a significant increase in expression, eight genes were shown to be significantly downregulated by RT-qPCR while a ninth gene displayed less than 60% expression in R5 cells as compared to R0 with a nearly significant p value of less than 0.051. Two genes, CLDN11, and SLC27A6 displayed a loss of more than 99% of R0‘s expression.
127
CLDN11 was of interest to us as it had most notably previously been shown to be hypermethylated in gastric cancer patients, and the loss of expression associated with said hypermethylation was correlated with increased levels of cell motility and invasiveness in a cell culture model yet it had not yet been investigated in ovarian cancer (Agarwal, et al., 2009). SLC27A6, a fatty acid transport protein (Gimeno, 2007), had not yet been previously studied in depth for an association with cancer of any type based on a literature search though some evidence of allele-specific DNA methylation being associated with the SLC27A6 gene in a mouse model (Schilling, Chartouni, & Rehli, 2009). As the seemingly unrelated copper transport system has been implicated in cisplatin uptake [(Ishida, Lee, Thiele, & Herskowitz, 2002); (Komatsu, et al., 2000)], it stands to reason that fatty acid transport may also play some role in cisplatin uptake. The additional knowledge that the expression of both these genes had been previously been shown to have significant expression changes in a subset of tumors from publicly available ovarian cancer patient datasets on Oncomine, led us to select these genes for further study.
4.2.3 CLDN11 is methylated in a panel of ovarian cancer cell lines of varying cisplatin resistance.
In order to verify the methylation increase detected by DMH, the qualitative DNA methylation assay COBRA was performed on the CLDN11 and
128
SLC27A6 loci as a precursor to quantitative methylation detection methods being performed. 851 bases of the CLDN11 promoter and 294 bp of the SLC27A6 promoter were analyzed for a panel of ovarian cancer cell lines which possess varying levels of cisplatin resistance. For CLDN11, a region spanning 453bp containing the transcription start site analyzed by COBRA showed greater methylation in R5 cells than in R0 cells while other cisplatin resistant lines showed low levels of varying methylation (see Figure 4.3). Additionally, an adjacent region covering 398bp in the gene body displayed methylation in the cancer cell lines as compared to normal cells, but R0 and R5 cells were qualitatively indistinguishable from one another based on the methylation level of the five AciI sites analyzed in this qualitative assay. Conversely, SLC27A6 displayed very little or no methylation in any cell line or normal controls. Previous studies have been an exceptionally strong indicator of very low levels of methylation being detected when quantitative methylation detection methods are employed. As such SLC27A6 was dropped from further study (see Figure 4.3).
In order to quantitatively determine the amount of methylation present at the CLDN11 locus, pyrosequencing was performed for the same regions as were analyzed by COBRA on the same cell lines (see Figure 4.4). Pyrosequencing results confirmed that the CLDN11 locus underwent a hypermethylation transformation during the selection of cisplatin resistant of round 5 cells. Overall, a 34% increase in average methylation was detected between R0 and R5 cells across both regions while 33 of 36 (92%) total CpG sites showed a significant
129 increase (with a 90% confidence level) in methylation for R5 cells as compared to
R0 based on the variation among a panel of 10 normal patient samples and an immortalized ovarian surface epithelial cell line (IOSE). In addition to all cell lines displaying similar yet varied levels of methylation, all were significantly increased as compared to a the group of normal patient samples and IOSE cells shown in
Figure 4.5.
4.2.4 CLDN11 is methylated in a set of ovarian cancer patients.
In order to determine the true relevance of CLDN11 in ovarian cancer, a set of 78 patients were analyzed for their level of methylation via pyrosequencing. In agreement with the cell line data, we found that 50 of 78
(70%) of the ovarian cancer patients displayed a significant increase in average methylation as compared to the variation in methylation among a group of 10 normal ovarian surface epithelial cells (NOSE) at a 90% confidence level (see
Figure 4.5). Additionally 818 of 2808 (29%) total CpG sites analyzed were hypermethylated as compared to the group of normal samples when analyzed by the same criteria. It was interesting that while some samples displayed an increase in methylation for the region immediately downstream of the transcription start site, the region further downstream and into the first exon displayed even higher levels of methylation corresponding with the region of heaviest methylation detected in R5 cells.
130
4.2.5 DNA methylation of CLDN11 leads to epigenetic silencing of the gene in ovarian cancer cell lines.
As a precursor to examining additional patient datasets for significant association of CLDN11 with a cancerous phenotype, the panel of ovarian cancer cell lines was examined for levels of CLDN11 expression to verify that the methylation detected was capable of silencing gene expression. Figure 4.6 shows that CLDN11 was expressed at lower levels in R5 cells than in R0 counterparts, as expected. In addition to confirming the original microarray work, it also showed that the increase in methylation was capable of causing the repression of CLDN11. MCP2, MCP3, and MCP8 all displayed significantly lower levels of expression as compared to A2780 R0 cells in agreement with their increase in methylation. CP70 actually displayed a significant increase in
CLDN11 expression as compared to A2780 R0 cells. Interestingly, several of the
CpG dinucleotides furthest downstream from the TSS displayed equal or slightly lower levels of methylation in CP70 as compared to R0 cells. Overall this finding suggests that DNA methylation plays a role in repressing CLDN11 expression among these cell lines and either implicates those residues which are least methylated in CP70 as being the most functionally relevant to gene expression, or suggests that CP70 cells are able to express CLDN11 by some other mechanism such as lacking repressive histone modifications.
131
4.2.6 CLDN11 expression is significantly lower in a group of ovarian cancer patients as compared to a group of normal tissue samples.
Patient cohorts with known global DNA methylation status are severely limited due in no small part to a lack of methods capable of quantitatively measuring global DNA methylation. Armed with the knowledge that DNA methylation was able to cause a reduction in CLDN11 expression in ovarian cancer cell lines we set out to examine CLDN11 expression in a previously reported patient cohort. Oncomine, a publicly available online repository of expression microarray results for patient data sets, was used to determine if repression of CLDN11 expression had previously been reported in other studies.
While the work originally performed by Lu et al. showed that the expression of as few as three genes was sufficient to classify tumor samples from normal tissue despite significant differences in tumor heterogeneity (Lu, et al., 2004), their bigger scientific contribution may be the publishing of the expression data of a group of ovarian cancer patients for use in future in silico analysis. Using their microarray data, we examined CLDN11 expression in the patient samples as compared to the non-cancerous samples (see Figure 4.7). While all three tumor grades displayed lower average expression levels of CLDN11 as compared to normal tissue samples, grades 2, 3, and 4 all were significantly lower. As only three grade 1 tumors existed in this data set, it is possible that by studying additional grade 1 tumors, the standard deviation may be lowered to a point where significance can be attached to their repression. It was also exceptionally
132 interesting that 36 of the 41 (88%) patients with known tumor grade displayed lower levels of CLDN11 expression than normal tissue samples.
In agreement with our methylation results, we saw that 88% of patients showed decreased levels of expression relative to noncancerous samples in their cohort as compared to 69% of patients which showed an increase in the average level of methylation in our own cohort (see Figure 4.7 and Figure 4.5). The higher percentage of loss of CLDN11 expression as compared to increases in DNA methylation suggests that the loss of CLDN11 expression may be functionally relevant and that tumor cells may evolve multiple methods of silencing CLDN11 in order to achieve a more cancerous outcome.
4.2.7 Lower CLDN11 expression was associated with increased tumor grade as well as decreased survival times.
As lower levels of CLDN11 expression corresponded to an increase in tumor grade (see Figure 4.7), we were interested to see if it may also correlate to a decrease in survival time. Disappointingly, Figure 4.8 showed that tumors with the lowest levels of CLDN11, regardless of grade, displayed slightly longer survival times when evaluated by a Kaplan-Meier survival curve. Taken together these results show that the loss of CLDN11 expression is not only associated with cisplatin resistance in a cell model via DNA methylation and that such DNA methylation is detectable in patient tumors, but that in at least one patient cohort,
133 loss of expression is associated with slightly longer survival times despite associating with increased tumor grade.
Unfortunately efforts to locate a publically available data set with known cisplatin sensitivities and clinical outcomes have proved futile. This would be a key piece of evidence as the current studies do not tie cisplatin resistance to actual ovarian cancer patients directly. Without this evidence it remains possible that loss of CLDN11 expression is associated with increased cisplatin in a patient samples, and therefore decreased survival times. It is however interesting to note that the 88% of patients in the patient cohort display a lower level of CLDN11 expression as compared to normal samples, which is only slightly higher than the published rate of 80% of patients who relapse with a cisplatin resistance form of ovarian cancer (Agarwal & Kaye, 2003). Similarly, 70% of patients analyzed in our own study had higher average methylation values as compared to normal samples.
4.2.8 Loss of CLDN11 expression is associated with increased levels of motility.
As the loss of CLDN11 expression has previously been reported to be involved in increased levels of motility and invasion, we attempted to compare the levels of motility in R0 cells treated with a CLDN11 siRNA to see their motility would more closely resemble R5 cells. A wound healing assay was performed for
134 three situations: A2780 round 0 cells, A2780 round 5 cells, and A2780 cells treated with a CLDN11 siRNA. Cells were seeded 24 hours before siRNA knock down was performed, and an additional 24 hours was given before cell plates were scratched with a pipette tip to simulate a wound being made (see Figure 4.9
A). Representative images taken in 24 hour increments of the wound healing assay are shown in Figure 4.9 B along with CLDN11 expression values at matched times (see Figure 4.9 C). In agreement with previous studies, by 48 hours post wounding untreated A2780 cells had very little invasion of the wound while both A2780 R0 cells treated with CLDN11 siRNA and A2780 R5 had significant closure of the wound.
4.3 Discussion
Cisplatin chemotherapy in combination with other chemotherapeutic agents such as taxanes along with surgery is the primary treatment option for the treatment of ovarian cancer [(McGuire, et al., 1996); (Ozols, 2005)]. While most patients initially respond to cisplatin treatment if it is caught in the early stage, more than 75% of patients already have stage 3 or 4 cancer lowering their five- year survival rate to 30.6% as compared to an overall rate of 46% (Ries, et al.,
1975-2005). Regardless of stage at time of diagnosis, the majority of ovarian cancer patients will represent with a recurrent cancer that is refractory to cisplatin treatment (Agarwal & Kaye, 2003). While new chemotherapeutic agents remain a
135 goal for ovarian cancer treatment, being able to re-sensitize ovarian cancer tumors to cisplatin would be a means to the same end. Evidence suggests that epigenetic abnormalities such as DNA methylation significantly contribute to the cisplatin resistant phenotype (Li, et al., 2009), and here we have reported the epigenetic silencing of a key component of the tight junction, CLDN11 which previously had been reported to be silenced in gastric cancer (Agarwal, et al.,
2009) to be silenced in ovarian cancer.
While we were able to show both the hypermethylation of the CLDN11 promoter (see Figure 4.5) and the loss of CLDN11 expression being associated with increased tumor grade (see Figure 4.7) as opposed to normal tissue.
Surprisingly the lowest levels of CLDN11 expression were associated with a slightly longer average survival time (41 versus 45 months). As we were unable to show CLDN11‘s direct involvement with cisplatin resistance in patient samples, it remains possible that additional patient data sets with known cisplatin sensitivities may show a decreased in survival time associated with decreased
CLDN11 expression and increased cisplatin resistance.
Showing that both the loss of CLDN11 expression and hypermethylation took place concurrently during the acquisition of cisplatin resistance in the A2780 ovarian cancer cell line model (see Figure 4.4 and Figure 4.6) seemingly implicates the loss of the tight junction complex being associated with cisplatin resistance in a counter intuitive manner. While the loss of the tight junction
136 complex leads to increased motility (see Figure 4.9 B) in a fairly straight forward manner of less cell-cell adhesion anchoring one cell to the next, it also increases cell surface area thereby exposing more of the cell to the media. As several studies have implicated one of the primary modes of cellular uptake of cisplatin as being passive diffusion [(Hromas, North, & Burns, 1987); (Mann & Andrews,
1991); (Binks & Dobrota, 1990)] an increase in cell surface area would seem to be associated with an increase in sensitivity rather than resistance. As a key method of cisplatin removal is also cellular efflux, an increase in cell surface area being associated with increased cisplatin resistance suggests that the increase in surface area more strongly favors efflux than uptake. A potential mechanism for the increase in cell surface area favoring efflux rather than uptake is that when cells are tightly associated with one another, cisplatin can only enter from the top of the cell where the drug is present, but can leave the cell in any direction. If the cisplatin molecules leave the cell through any direction other than the direction they entered, they either immediately enter another cell or enter a small space between cells from which they are likely to enter another nearby cell. When cells are more spread out due to the loss of cell-cell adhesion molecules efflux in any direction immediately removes the cisplatin from the cells rendering it inert until its reuptake by another cell. Alternatively the diminished levels of CLDN11 in three grade 1 tumors which are known to respond very well to cisplatin treatments may suggest that the loss of CLDN11 is not directly related to an
137 increase in cisplatin resistance, but with only three samples it is difficult to draw definitive conclusions.
4.4 Materials and Methods
4.4.1 Reagents
CLDN11 SMARTpool siRNA and a negative control were purchased through Thermo Scientific Dharmacon (Chicago, IL).
4.4.2 Cell Culture
A2780, CP70, MCP2, and MCP3 cells were cultured in RPMI-1640
(Invitrogen, Carlsbad, CA) supplemented with 10% fetal bovine serum (FBS)
(Invitrogen) and 50 U/ml penicillin-streptomycin (Invitrogen). Immortalized
Ovarian Surface Epithelial (IOSE) cells were cultured in a mixture of 50% media
199 and 50% MCDB 105 (Sigma, St. Louis, MO) supplemented with 10% FBS
(Invitrogen), 400 ng/ml hydrocortisone (Sigma), 10 ng/ml epidermal growth factor
(EGF) (Invitrogen), and 50 U/ml penicillin-streptomycin (Invitrogen). HeyC2 cells were cultured in DMEM (Invitrogen) supplemented with 5% FBS (Invitrogen),
0.1mM nonessential amino acid (Invitrogen), 2mM L-Glutamine(Invitrogen), 0.01
M of HEPES (Invitrogen), and 50 U/ml penicillin-streptomycin (Invitrogen).
138
SKOV3 cells were cultured in McCoy‘s 5A (Invitrogen) Supplemented with 10%
FBS (Invitrogen), 0.1mM nonessential amino acid (Invitrogen), 2mM L-
Glutamine(Invitrogen), 0.01 M of HEPES, and 50 U/ml penicillin-streptomycin
o (Invitrogen). All cells were cultured in a 37 5% CO2 incubator.
4.4.3 RNA Isolation and Reverse Transcription
Total RNA was isolated before or after epigenetic drug treatment via standard TRIzol (Invitrogen) protocol. Subsequent to DNase I (Invitrogen) treatment, cDNA was generated from one μg total RNA with Superscript III
(Invitrogen) and an 80:20 mixture of random hexamers and oligo dT primers
4.4.4 Quantitative Reverse Transcription PCR (qRT-PCR)
All reactions were conducted in triplicate with a melting curve analysis for each reaction on A Step One Plus instrument from Applied Biosystems was used in conjuncture with Power SYBR Green master mix (Applied Biosystems,
Carlsbad, CA). The ΔΔCt method of gene expression analysis was used to determine the relative level of expression as compared to the internal GAPDH control.
139
4.4.5 Bisulfite Conversion
500ng of genomic DNA was subjected to bisulfite modification with the EZ
DNA Methylation Kit (Zymo Research, Irvine, CA) with three noteworthy differences from the manufacturer‘s protocol. Briefly, 500 ng of DNA in a volume of 45 μl of water are snap frozen on dry ice before being rapidly thawed at 60o three times in order to fragment the DNA to allow the CT conversion reagent better access to the DNA. Conversion took place over the course of approximately 15.5 hours with cycling of: 50o two hours, 95o 15 seconds, 50o four hours, 95o 15 seconds, 50o 9.5 hours. Final elution of bisulfite converted DNA off the column was accomplished by two separate incubations of DNA grade water heated to 55o for 5 minutes in 50 μl aliquots. Our lab has observed these described modifications to achieve both a better conversion rate as well as increase the total amount of bisulfite converted DNA albeit at a slightly lower concentration.
4.4.6 Combined Bisulfite Restriction Analysis (COBRA)
PCR primers were designed against stretches of DNA within the CLDN11 promoter region which lack CpG dinucleotides yet are rich with cytosines which were converted to uracil during the bisulfite conversion so that PCR amplification is enriched for fragments of DNA which were successfully converted. Amplified product was split into two equal aliquots with one aliquot being subjected to
140 digestion with a restriction enzyme whose recognition sequence is only present if the original genomic DNA sequence was methylated, while the other aliquot was mock treated with a reaction mixture lacking restriction enzyme. Post digestion
PCR products were resolved on a 3% agarose gel with small DNA fragments in the restriction enzyme aliquot representing qualitative methylation of the locus.
Primers were designed against two regions of the CLDN11 gene and one region of the SLC47A6 gene based on a bisulfite converted genome and avoiding the need for redundant primers.
4.4.7 Pyrosequencing
Primers were designed against a bisulfite converted genome in regions lacking CpG dinucleotides so as to avoid PCR bias using the PyroMark Assay
Design 2.0 software. Regions were amplified using standard protocols before being subjected to pyrosequencing on a Qiagen PyroMark MD. Every reaction and CpG site was inspected on the Pyro Q-CpG 1.0.9 software for errors in sequence pattern or bisulfite conversion. Sequence pattern errors were resolved by analyzing their impact on surrounding CpG patterns and adjusting the dispensation pattern while samples which displayed errors in bisulfite conversion were reconverted and re analyzed. Samples which still displayed incomplete bisulfite conversion after a second bisulfite conversion were dropped from the study.
141
4.4.8 Patient Cohorts
The patient cohorts including gene expression and survival information for patients used in this work were previously published (Lu, et al., 2004). Gene expression data was previously normalized for all 42 patients and were directly used in our study. A logrank test was used to determine significance of survival differences
142
Figure 4.1 DMH determined DNA methylation levels of genes over expressed in either A2780 Round 0 or Round 5 cells.
Figure 4.1 DMH determined DNA methylation levels of genes over expressed in either A2780 Round 0 or Round 5 cells. Genes showing an expression level of at least 1.5 fold over the other cell type were examined for a change in methylation between the two different cisplatin sensitivities of at least
1.5 fold. Values for each category are shown on the Figure, and each section is drawn to scale.
143
Figure 4.2 Quantitative real time PCR expression levels of 18 genes repressed in A2780 Round 5 cells as compared to A2780 Round 0.
Figure 4.2 Quantitative real time PCR expression levels of 18 genes repressed in A2780 Round 5 cells as compared to A2780 Round 0. RNA isolated from A2780 round 0 and 5 cells was subjected to reverse transcription followed by quantitative SYBR green PCR. Bars represent average expression value of triplicates relative to the internal control GAPDH and round 0 expression level. Vertical lines indicate relative standard deviation while * denotes significant repression in round 5 cells a t-test p value of less than 0.05.
144
Full Size Digested
Full Size Digested
Full Size Digested
Figure 4.3 DNA methylation analysis of CLDN11 and SLC27A6 in a panel of ovarian cancer cell lines by COBRA.
Figure 4.3 DNA methylation analysis of CLDN11 and SLC27A6 in a panel of ovarian cancer cell lines by COBRA. Bisulfite converted DNA was PCR amplified for 45 cycles using primers designed against the converted genomic location which spanned regions lacking CG dinucleotides. Reactions were split into equal reactions with one being subjected to restriction endonuclease digestion and the other was subjected to mock digestion lacking restriction enzyme. The specific restriction enzyme used in each reaction is listed while the
+ symbols denote the lanes corresponding to the reactions that received enzyme
145 as opposed to mock treatment. Images represent 3% agarose gels with dark bands representing amplified DNA. Full size and digested product locations denoted on the left hand axis.
146
Figure 4.4 Pyrosequencing of CLDN11 locus in a panel of ovarian cancer cell lines.
Figure 4.4 Pyrosequencing of CLDN11 locus in a panel of ovarian cancer cell lines. Bisulfite converted DNA was amplified for two separate regions of the
CLDN11 locus and subjected to pyrosequencing. The CLDN11 gene is drawn in blue with black arrow representing the transcription start site and small vertical black lines represent CpG sites. Colored circles were used to represent the quantitative methylation levels rounded to the nearest percentage.
147
Figure 4.5 Pyrosequencing of CLDN11 locus in a panel of 78 primary ovarian tumors and 10 normal ovarian tissue samples.
148
Figure 4.5 Pyrosequencing of CLDN11 locus in a panel of 78 primary ovarian tumors and 10 normal ovarian tissue samples. Bisulfite converted
DNA was amplified for two separate regions of the CLDN11 locus and subjected to pyrosequencing. The CLDN11 gene is drawn in blue with black arrow representing the transcription start site and small vertical black lines represent
CpG sites. 50 of 78 samples were determined to be hypermethylated relative to the average methylation of 10 normal samples and their variance at the 90% confidence interval. 818 of 2808 individual CpG sites were determined to be hypermethylated by the same criteria. Colored circles were used to represent the quantitative methylation levels rounded to the nearest percentage.
149
Figure 4.6 CLDN11 expression in a panel of ovarian cancer cell lines.
Figure 4.6 CLDN11 expression in a panel of ovarian cancer cell lines.
CLDN11 mRNA levels were measured by quantitative reverse transcription PCR using SYBR Green reagents. Values represent the average of two independent triplicate measurements while vertical lines represent relative standard deviation.
* was used to denote significant repression with a p value of less than 0.05.
150
Figure 4.7 Relative expression of CLDN11 in panel of 42 patients.
Figure 4.7 Relative expression of CLDN11 in panel of 42 patients. CLDN11 expression values in four normal patients were averaged and used to calculate relative expression levels for all patient samples including normal tissue. Red bars indicate average level of expression for either normal tissue or each tumor grade while blue diamonds were used to denote the normalized expression values. T-tests were performed to analyze the statistical significance of the
151 average expression level for each tumor grade as compared to normal tissue.
Average expression for tumor grades 2, 3, and 4 all displayed a significant p value of less than 0.05 but were not denoted directly on the figure.
152
Figure 4. 8 Kaplan-Meier survival curve for patients displaying differential expression of CLDN11.
Figure 4.8 Kaplan-Meier survival curve for patients displaying differential expression of CLDN11. Expression of CLDN11 was used to sort patients into two groups before examining the groups for differential survival: High Expression
(red) and Low Expression (blue). The horizontal axis denotes survival time in months while the vertical axis denotes percent of surviving patients for each group. A log-rank test p-value of 0.0422 denotes the significance of higher expression of CLDN11 being associated with decreased survival time.
153
A.
B.
CLDN11
Figure 4.9 (continued)
154
Figure 4.9 (continued)
C.
Figure 4.9 CLDN11 siRNA knockdown is associated with increased motility by wound healing assay.
Figure 4.9 CLDN11 siRNA knockdown is associated with increased motility by wound healing assay. A timeline of cell treatments and manipulations is shown in Figure A. A2780 R0 cells, A2780 R0 cells treated for 24 hours with
SMARTpool CLDN11 siRNA , and A2780 R5 cells were scratched with p1000 pipette tip to simulate a wound (t=0). Representative pictures showing scratches for each cell line at t=0, t=24, and t=48 are shown in B. SYBR Green RT-qPCR was conducted on RNA isolated at t=0 and t=24 from untreated A2780 R0 cells,
CLDN11 siRNA treated A2780 R0 cells, and A780 R5 cells as shown in C.
Vertical lines indicate relative standard deviation while * denotes a significant
155 repression of CLDN11 expression relative to time untreated A2780 R0 cells at the same time with a p-value of less than 0.05.
156
Chapter 5
Discussion
5.1 Multifaceted Mechanisms of Ovarian Carcinogenesis.
The studies presented here represent three unique mechanisms of ovarian carcinogenesis with significant overlap among the epigenetic mechanisms. Knudson‘s two hit hypothesis states that for most genes, two mutations are required to inactivate a tumor suppressor gene, one on each of the allelic copies of the gene in a single cell (Knudson, 20001). Subsequently the hypothesis has expanded to the idea that alterations of any type, including but not limited to genetic mutations, which affect normal cellular physiology are capable of providing either of the ‗hits‘ to a given gene. Indeed, the major mechanisms of carcinogenesis studied here, DNA methylation (Esteller, et al.,
2001), disruption of normal microRNA levels (Calin & Croce, 2006), and aberrant activation of a signaling pathway (Maynard, Sikich, Lieberman, & LaMantia,
2001), have all been shown to be capable of producing such hits to a given gene or genes for a disease. While Knudson‘s hypothesis focuses on carcinogenesis from the prospective of a single gene, Hanahan and Weinberg, focusing on carcinogenesis from the prospective of a single cell, proposed that there are six
157 essential cellular processes which must be altered in order for a cell to become cancerous (Hanahan & Weinberg, 2000). Taken together this suggests that a dozen individual alterations may be required to occur in a single cell in order for it to become cancerous. By extension, when one further considers the amount of redundancy present in cells for a single process or pathway (noticeably observed in mouse knockout models which show functional yet abnormal development of a particular system upon knockout of a single gene), numerous hits to each of the six pathways described by Hanahan and Weinberg are likely to be required for the most devastating cancers. Indeed, better prognosis being associated with early detection may be caused simply by limiting the number of hits a cell acquires to each of the processes during tumorigenesis therefore making the tumor less adaptive to therapeutic approaches.
The repression of CLDN11 was shown to be significantly associated with worse patient survival among 75 patients in a previously published ovarian cancer cohort. As CLDN11 has previously been described to be part of the tight junction complex for cell-cell adhesion, its repression represents a terminal alteration which is incapable of altering the presence of additional cellular factors.
Its hypermethylation is a double ‗hit‘ in Knudson‘s hypothesis, and as it increases cell motility and invasion it satisfies one of Hanahan and Weinberg‘s six processes which must be disturbed for a cell to become cancerous. The mechanism of how the hypermethylation of CLDN11 is initiated and selected for remains unclear.
158
By contrast both the repression of hsa-mir-9-3 and the dysregulation of
TGFβ/SMAD4 signaling represent intermediate events which are capable of affecting the presence or absence of a multitude of different genes and gene products. In silico analysis of hsa-mir-9 for potential target genes by several different target prediction programs revealed nearly 4,000 unique genes plus additional splice variant forms of those genes. While it is obvious that all of these genes will not be verified as targets (as was the case with KIF21A in my own work), evidence has shown that a single microRNA is capable of repressing several hundred different genes in vivo albeit at minor levels (Selbach,
Schwanhäusser, Thierfelder, Fang, Khanin, & Rajewsky, 2008). The protein level of hundreds of genes being regulated as a result of aberrant microRNA expression suggests that such aberrant expression events may not lead to a tightly controlled physiological outcome or even every regulation event advancing the cancerous process, but rather overall lead to a pro-cancerous state. As such, it stands to reason that while the repression of hsa-mir-9-3 is associated with a cancerous state, the wide variety of processes that may be disturbed by that repression may both be limited in severity of effect and lack a coherent physiological outcome to be associated with the repression. Similar to the repression of hsa-mir-9-3 potentially affecting hundreds of genes, we showed
TGFβ/SMAD4 signaling was capable of regulating the expression of 318 genes and that the dysregulation of 187 of those genes correlated with a worse clinical outcome. It is likely that with a greater understanding of what gene products are
159 specifically being over expressed as a result of the repression of hsa-mir-9-3, a gene signature of a subset of those genes may be used to predict outcome in a set of patients as well.
5.2 Technological Advances as an Advancement of Science.
While past advances always seem to have been inevitable and with the promise of future advances appearing endless, the current state of technology always appears to be frustratingly prolonged. The heavy use of differential methylation hybridization (DMH) as a discovery tool for identifying both hsa-mir-
9-3 and CLDN11 as epigenetically repressed targets of ovarian cancer must serve as a reminder that the use of present technologies can be instrumental to making discoveries that may otherwise be years away, even if such reductionist approaches will eventually fall by the wayside because of their assumptions.
Next generation sequencing approaches have recently shown their ability to do what previously was impossible: sequence an entire human methylome. As the platforms first came online the Illumina Genome Analyzer II platform was used to sequence the methylome of Arabidopsis using a method referred to as MethylC- seq (Lister, et al., 2008) which served as a proof of principle for the same method to be used to sequence the first human methylome less than two years later
(Lister, et al., 2009). While it is now possible to sequence an entire human methylome, reductionist and enrichment approaches such as RRBS (Meissner,
160 et al., 2008), MeDIP-seq (Jacinto, Ballestar, & Esteller, 2008), and MBD-seq
(Serre, Lee, & Ting, 2010) are still likely to be used for the coming years for two reasons: the economic cost of sequencing more than one billion reads to achieve more than 14x genomic coverage, and current difficulties in processing such large amounts of complex data (Milosavljevic, 2010). I hypothesize that these current restrictions will be overcome and whole genome bisulfite sequencing will someday eventually be regarded as the gold standard for DNA methylation analysis.
Perhaps the most interesting finding that came out of sequencing the first human methylomes was the amount of non-CG methylation (methylation of cytosines which are not followed by a guanine residue) that was detected: nearly
25% of all methylation in embryonic stem cells, and the methylation was found to be largely removed in differentiated cells but to return upon induction of a pluripotent stem cell phenotype (Lister, et al., 2009). This finding is interesting for two reasons not mentioned in their original work. Most qualitative and quantitative assays are not designed to interrogate non-CpG methylation, and many are not capable of detecting it. Other studies have shown that cancer cells can adopt ‗stem cell features‘ [(Krivtsov, et al., 2006); (Takahashi & Yamanaka,
2006)] suggesting the same cells may also gain non-CpG methylation may play a critical role in carcinogenesis. Overall this suggests that finding ―CpG island shore‖ methylation is capable of causing epigenetic silencing (Irizarry, et al.,
2009) may be but the first evidence that while CpG island methylation is a key
161 contributor to epigenetic silencing, any methylated cytosine may be capable of exerting epigenetic repression effects on gene expression. Further, the belief that
CpG island methylation is critical for epigenetic repression may have stemmed from the sheer concentration cytosines in CpG islands. This appears to be potentially true for both development and carcinogenesis.
The first published human genome sequence contained sequence reads taken from a total of five individuals, and in addition to providing the first full consensus sequence, 2.1 million single-nucleotide polymorphisms (SNPs) were identified (Venter, et al., 2001). The initial of report of less than 1% of SNPs changing protein sequence has since proven to be accurate through significant resequencing efforts though the number of known SNPs now number nearly 10 million (The International HapMap Consortium, 2005). Despite not changing protein sequence, SNPs have been shown to be important as significant markers for several cancer types including breast (Easton, et al., 2007), and prostate
(Amundadottir, et al., 2006). As next generation sequencing platforms continue to improve both in data acquisition and lower cost, additional SNPs will likely be discovered and stronger correlations made to abnormal states. This helps underscore the importance of sequencing more human genomes after the initial genome was sequenced. The identification of more SNPs by sequencing additional individuals serves as a useful parallel for the importance of sequencing additional human methylomes. Human methylomes analysis will undoubtedly be more complex than SNP analysis for while SNPs represent variation among
162 individuals with at most four variants, tissue specific methylation and different percents of cells harboring said methylation dictate that a single individual will contain a nearly innumerable number of methylomes (Milosavljevic, 2010).
Additionally, just as genome-wide association studies were performed once a significant number of SNPs had been identified in order to identify associations of particular SNPs with diseased states [outlined in (Hirschhorn & Daly, 2005) and an early/recent review of associations for several cancer types (Easton &
Rosalind, 2008)], once multiple methylomes from many tumors of a single cancer type are sequenced associations of methylated loci with various cancer features such as survival and progression are likely to emerge.
In addition to being able to sequence an entire human methylome, the ability of next generation sequencing approaches to sequence material without having to make assumptions of what is important, allow us to make new discoveries we would have otherwise missed based on faulty assumptions. Such is the case with using ChIP-seq, as compared to microarray based ChIP-chip methodology, to determine SMAD4 binding patterns following TGFβ stimulation.
One of the most interesting findings of the TGFβ/SMAD4 signaling project was that more than 70% of all binding sites were detected at more than 10KB away from the nearest transcription start site, a finding which would be unlikely to have been discovered using microarray based methods as such methods typically focus specifically on the promoter region of genes. The same phenomenon has now been observed for several other transcription factors including estrogen
163 receptor alpha (ERα) [(Carroll, et al., 2006); (Welboren, et al., 2009); (Fullwood, et al., 2009)], androgen receptor (AR) (Wang, et al., 2009), and the peroxisome proliferator-activated receptor (PPAR) (Nielsen, et al., 2008).
Given the advances of next generation sequencing and the implications of several preliminary studies, it‘s easy to say that one of the next issues which will need to be addressed is a lack of computational methods needed to analyze the complex data sets that will be produced. This is particularly true of analyzing multiple tumor methylomes. Additionally the potentially revolutionary finding of large amounts of non-CpG methylation in embryonic stem cells, and induced pluripotent stem cells (Lister, et al., 2009) may mean that current experimental methods may need to be redesigned to interrogate the methylation of cytosine residues outside of CpG islands and potentially outside of CpG dinucleotide context for the purpose of ascertaining an individual methylation events contribution to the carcinogenesis process.
5.3 DNA Methylation of the Hsa-mir-9-3 Locus Is Specifically Associated with Ovarian Cancer.
Differential Methylation Hybridization (DMH) was used to profile more than
27,000 annotated CpG islands in a panel of 24 ovarian cancer tumors. The methylation of a CpG island on chromosome 15 was of particular interest to us as it contained the microRNA hsa-mir-9-3 embedded within it (see Figure 2.1). At
164 the time, the hsa-mir-9-3 locus had been implicated as being involved in tumor metastasis among a cohort of breast cancer patients (Iorio, et al., 2005), while the hsa-mir-9-1 microRNA had been shown to be significantly epigenetically repressed by DNA methylation in a panel of 71 breast tumors (Lehmann, et al.,
2008). Together this led us to believe that the mature hsa-mir-9 microRNA may be commonly epigenetically repressed through hypermethylation of the hsa-mir-
9-3 locus in ovarian cancer.
Subsequent quantitative Sequenom MassARRAY analysis of a larger panel of 85 tumors and eight normal samples revealed a specific association of the hypermethylation of hsa-mir-9-3 locus with the cancerous state but not with a specific clinical factor (see Figure 2.7). Analysis of the hsa-mir-9 microRNA in a cell line system was neither able to confirm its ability to regulate the protein level of FOXG1 (see Figure 2.9) as had previously been reported (Shibata, Kurokawa,
Nakao, Ohmura, & Aizawa, 2008) nor to validate its ability to regulate the KIF21A gene to any significant degree in primary patient samples (see Figure 2.12 A and
B).
Being unable to identify a specific target gene, we attempted to determine what the functional consequence of hsa-mir-9 expression was by introducing a synthetic hsa-mir-9 mimic and saw an increase in both pro-apoptotic and apoptotic cells (see Figure 2.6) and a decrease in total cell number (see Figure
2.5) in the MCP2 ovarian cancer cell line. Together this implicated that the loss of
165 the hsa-mir-9 microRNA through the epigenetic silencing of the hsa-mir-9-3 precursor leads to both an increase in proliferation and reduction in apoptosis.
The cancer stem cell theory [originally proposed in leukemia and multiple myeloma (Park, Bergsagel, & McCulloch, 1971) and proven in acute myeloid leukemia (Bonnet & Dick, 1997)] states that a small subpopulation of cells possessing stem cell like self renewal qualities is responsible for bulk tumor growth. As abnormalities, being more central to tumorigenesis, within such cells are of greater interest for study, and lacking a clinical correlation with methylation levels of hsa-mir-9-3 in bulk tumors, we sought to examine said methylation levels in ovarian cancer initiating cells (OCIC). A series of six sets of bulk tumor and a matched OCIC subpopulation of cells were subjected to DMH with no discernable enrichment of methylation among OCIC samples (see Figure 2.14).
Sequenom‘s quantitative MassARRAY analysis revealed no hypermethylation of the hsa-mir-9-3 locus in OCIC samples relative to matched bulk samples (see
Figure 2.15). While OCIC samples displayed increased levels of methylation for a significant portion of both average and individual CpG sites relative to 85 previously studied samples, bulk tumors also displayed a similar increase in methylation levels showing the patient cohort as a whole was more methylated than the previously studied cohort and that limited cell culture did not aberrantly inflate methylation levels.
166
In summary the average methylation of the hsa-mir-9-3 locus was identified as being significantly hypermethylated in 18 of 85 (21%) primary ovarian cancer tumors and ~30% of all CpG units in those tumors as compared to normal tissue samples. Unfortunately we were unable to identify any correlation with a specific clinical or pathological feature. Hsa-mir-9 expression was shown to be capable of repressing cellular growth and inducing both a pro- apoptotic and apoptotic state through and unknown mechanism not involving
KIF21A or FOXG1. Finally, hsa-mir-9-3 hypermethylation does not appear to be a marker for ovarian cancer initiating cells.
5.4 TGFβ/SMAD4 Signaling Regulates a Series of Genes Clinically Relevant to Ovarian Cancer Outcome.
Global SMAD4 binding patterns were analyzed by ChIP-sequencing in stimulated and unstimulated ovarian cancer cell line A2780 (see Figure 3.1) which despite constitutive presence of nuclear SMAD4 is still capable of additional nuclear translocation of cytoplasmic SMAD4 and additional production of SMAD4 upon TGFβ stimulation (Chan, et al., 2008). In an attempt to determine which of the more than 2,000 differential SMAD4 binding loci were responsible for regulating gene expression, gene expression was profiled using an expression microarray (see Figure 3.4). A total of 318 genes were identified as being both differentially expressed and to have SMAD4 binding associated
167 within the distal promoter of the same gene following TGFβ stimulation (see
Figure 3.5). The primary prognostic finding was that of the 318 differentially regulated genes, a gene signature of 187 genes was used to show a shorter median survival rate among a group of previously reported primary ovarian cancer tumors (Bild, et al., 2006) while a different gene signature of 19 genes was able to show differential survival in a panel of 42 patients (Lu, et al., 2004)
(see Figure 3.12 A, B; and Figure 3.14 respectively).
Comparison of the genes we identified to other previously published datasets of TGFβ responsive SMAD4 target genes [(Koinuma, Tsutsumi,
Kamimura, Imamura, Aburatani, & Miyazono, 2009); (Qin, et al., 2009)] revealed very little overlap (see Figure 3.8). Several reasons both technical and biological exist for the low level of overlap. Technical differences associated with previous studies being ChIP-chip rather than ChIP-seq led to more than 70% of all sites in the current study being in regions of the genome which were not analyzed in the previous studies. Additionally, one of the previously analyzed datasets was a normal skin cell line while the other cell line was a normal ovarian cell line so the potential for biological differences among the three studies logically exists.
Finally, each of the previously published papers report having used different antibodies than the mix of two antibodies which were used in our study potentially contributing to differences in specificity in binding of exposed epitopes.
Importantly, as little overlap was seen between the two previously analyzed normal cell lines strongly suggests that while the technical differences may have
168 concealed additional binding loci in the previous data sets, the biological variation among the tissue types is vast.
Together these findings highlight the ability of ChIP-seq methodologies to be integrated into existing data pipelines to identify clinically relevant genes which can be used both as biomarkers, and potentially serve as new therapeutic targets for treatment intervention. Such integration appears to be a key step in providing a more thorough picture of relevant cellular events as more than 70% of SMAD4‘s total binding sites were detected at more than 10kb away from the transcription start site of a nearby gene in agreement with several other recent studies of other transcription factors including estrogen receptor alpha (ERα)
[(Carroll, et al., 2006); (Welboren, et al., 2009); (Fullwood, et al., 2009)], androgen receptor (AR) (Wang, et al., 2009), and the peroxisome proliferator- activated receptor (PPAR) (Nielsen, et al., 2008). Together these findings suggest that the amount of long distance gene regulation through chromosomal looping may remain underestimated, and that current ‗genome-wide‘ binding studies performed on promoter microarrays may be lacking a significant amount of information regarding long distance binding.
169
5.5 Repression of the CLDN11 Gene through DNA Hypermethylation Is
Associated with Increased Cisplatin Resistance, Tumor Grade, and Mobility in Ovarian Carcinogenesis.
As cisplatin chemotherapy is the primary treatment option for ovarian cancer patients following debulking surgery, but the majority of patients eventually relapse with a cisplatin resistant cancer (Agarwal & Kaye, 2003) determining the molecular causes of cisplatin resistance is likely to play a direct role in improving patient treatment and outcome. While several tissue culture model systems exist for studying differences in cisplatin sensitivity, they often are based on tumors isolated from different patients at different times. Overall this adds a layer of complexity in genomic variability to any differences observed between the different cell lines. To overcome this issue Li and colleagues recently created model system for studying the acquisition of cisplatin resistance by treating a cisplatin sensitive cell line (A2780) with cisplatin to select for increasingly cisplatin resistant cells which they were gracious enough to provide us with for study (Li, et al., 2009).
DNA methylation levels were measured by differential methylation hybridization (DMH) for the parental cisplatin sensitive cells (A2780 R0) as well as the maximally cisplatin resistant cells (A2780 R5) while global mRNA expression levels were determined by expression microarray. Comparing differential methylation events with differentially expressed genes revealed a total of 243 genes which were hypermethylated in the same line as a loss of
170 expression of that gene was observed (see Figure 4.1). As we primarily focus on hypermethylation events we chose to focus on a subset of the 105 genes which were hypermethylated and repressed in A2780 R5 cells (see Figure 4.2).
CLDN11 emerged as an exceptionally repressed gene showing less than 1% of
A2780 R0 expression levels and being similarly repressed among additional cell lines while being hypermethylated in an number of cell lines (see Figure 4.4 and
Figure 4.6). The increase in DNA methylation was confirmed in a panel of 78 primary ovarian cancer samples (see Figure 4.5).
A previously published data set from Lu et al. (Lu, et al., 2004) was used to determine if the loss of CLDN11 expression was associated with a difference in survival. While lower levels of CLDN11 expression were associated with an increase in tumor grade and difference from a set of normal samples (see Figure
4.7), lower levels of CLDN11 were associated with a slight increase in average survival time of 41 to 45 months (see Figure 4.8). It remains possible that additional patient data sets with known cisplatin sensitivities and CLDN11 expression levels may show a decrease in survival time for lower levels of
CLDN11 expression.
In order to verify that the loss of CLDN11 expression was associated with an increase in mobility as had previously been reported (Agarwal, et al., 2009) we performed a cell mobility assay while knocking down CLDN11 with a
SMARTpool siRNA from Dharmacon (see Figure 4.9 B, C). A drastic increase in
171 wound closure was observed by 48 hours for A2780 R0 cells treated with the
CLDN11 siRNA as compared to untreated cells which more closely resembled closure of A2780 R5 cells.
Overall we have shown that loss of CLDN11 expression is associated with an increase in cellular mobility, an increase in tumor grade, and an increase in cisplatin resistance. Unfortunately we were unable to show the loss of CLDN11 expression to be associated with a decrease in survival time, though it remains possible that a larger data set or one with known recurrent cisplatin resistant tumors may.
5.6 Concluding Remarks.
In conclusion, we have shown three independent aberrations in ovarian cancer: the epigenetic silencing of the hsa-mir-9-3 microRNA; a group of ovarian cancer specific TGFβ responsive, SMAD4 target genes which contribute to increased ovarian cancer severity; and the identification of an epigenetically silenced claudin gene, CLDN11, whose expression is related to cisplatin resistance, cellular mobility, and tumor grade. Together these results may someday help with a personalized medicine approach which I hypothesize will be necessary for the treatment of ovarian cancer, and all cancer in general, due to the complex nature of the disease. The ability of these described aberrations to affect multiple physiological processes implicates them as being key factors to
172 help determine the appropriate course of treatment when combined with additional factors. By identifying specific biomarkers which, in combination with other factors, may be used to gain better insight into the severity of a particular tumor, doctors may someday be able to direct a course of treatment which will not only allow a patient to enter into remission, but do so without many of the negative side effects associated with cancer treatment.
173
References
Agarwal, R., & Kaye, S. (2003). Ovarian cancer: strategies for overcoming resistance to chemotherapy. Nature Reviews Cancer , 63 (1), 12-31.
Agarwal, R., Mori, Y., Cheng, Y., Jin, Z., Olaru, A., Hamilton, J., et al. (2009). Silencing of claudin-11 is associated with increased invasiveness of gastric cancer cells. Plos one , 4 (11), e8002.
Al-Hajj, M., Wicha, M., Benito-Hernandez, A., Morrison, S., & Clarke, M. (2003). Prospective identification of tumorgenic breast cancer cells. Proceedings of the National Academy of Sciences of the USA , 100, 3983-3988.
Amundadottir, L., Sulem, P., Gudmundsson, J., Helgason, A., Baker, A., Agnarsson, B., et al. (2006). A common variant associated with prostate cancer in European and African populations. Nature Genetics , 38 (6), 652-658.
AN, B., & Korc, M. (2005). Smad7 abrogates transforming growth factor-betaI- mediated growth inhibition in COLO-357 cells through functional inactivation of the retinoblastoma protein. Journal of Biological Chemistry , 280, 21858-21866.
Asli, N., Pitulescu, M., & Kessel, M. (2008). MicroRNAs in organogenesis and disease. Current Molecular Medicine , 8 (8), 698-710.
Baily, J., Rave-Harel, N., McGillivray, S., Coss, D., & Mellon, P. (2004). Activin regulation of the follicle-stimulating hormone beta-subunit gene involves smads and the TALE homeodomain proteins, Pbx1 and Prep 1. Molecular Endocrinology , 18, 1158-1170.
Balch, C., Huang, T., Brown, R., & Nephew, K. (2004). The epigenetics of ovarian cancer drug resistance and resensitization. American Journal Obstetric Gynecology , 191 (5), 1552-1572.
Balch, C., Matei, D., Huang, T., & Nephew, K. (2010). Role of epigenetics in ovarian and endometrial cancers. Epigenomics , 2 (3), 419-447.
174
Baldwin, R., Tran, H., & Karlan, B. (2003). Loss of c-myc repression coincides with ovarian cancer resistance to transforming growth factor beta growth arrest independt of transforming growth factor beta/smad signaling. Cancer Research , 63, 1413-1419.
Barski, A., Cuddapah, S., Cui, K., Roh, T., Schones, D., Wang, Z., et al. (2007). High-resolution profiling of histone methylation sin the human genome. Cell , 129, 823-837.
Bartel, D. (2009). MicroRNAs: target recognition and regulatory functions. Cell , 136 (2), 215-233.
Berchuck, A., Rodriguez, G., Olt, G., Whitaker, R., Boente, M., Arrick, B., et al. (1992). Regulation of growth of normal ovarian epithelial cells and ovarian cancer cell lines by transofrming growth factor-beta. American Journal Obstetric Gynecology , 166, 676-6842.
Bestor, T. (1998). The host defence function of genomic methylation patterns. Novartis Foundation Symposium , 214, 187-195.
Bild, A., Yao, G., Chang, J., Wang, Q., Potti, A., Chasse, D., et al. (2006). Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature , 439, 353-357.
Binks, S., & Dobrota, M. (1990). Kinetics and mechanism of uptake of platinum- based pharmaceuticals by the rat small intestine. Biochemical Pharmacology , 40 (6), 1329-1336.
Blobe, G., Schlemann, W., & Lodish, H. (2000). Role of transforming growth factor beta in human disease. New England Journal of Medicine , 342 (18), 1350- 1358.
Bonnet, D., & Dick, J. (1997). Human acute myeloid leukemia is organized as a hierarchy that originates from a primitive hematopoietic cell. Nature Medicine , 3 (7), 730-737.
Borst, P., Evers, R., Kool, M., & Wijnholds, J. (2000). A family of drug transporters: the multidrug resitance associated protiens. Journal of the National Cancer Institute , 92 (16), 1295-1302.
Breiman, L., Friedman, J., Olshen, R., & Stone, C. (1984). Classification and regression trees. Chapman & Hall New York NY .
175
Calin, G., & Croce, C. (2006). MicroRNA signatures in human cancers. Nature Reviews Cancer , 6 (11), 857-66.
Calin, G., Dumitru, C., Shimizu, M., Bichi, R., Zupo, S., Noch, E., et al. (2002). Frequent deletions and down-regulation of micro- RNA genes miR15 and miR16 at 13q14 in chronic lymphocytic leukemia. Proceedings of the National Academy of Sciences of the USA , 99 (24), 15524-15529.
Campos, E., & Reignberg, D. (2009). Histones: annotating chromatin. Annual Review Genetics , 43, 559-599.
Carroll, J., Meyer, C., Song, J., Li, W., Geistlinger, T., Eeckhoute, J., et al. (2006). Genome-wide analysis of estrogen receptor binding sites. Nature Genetics , 38 (11), 1289-1297.
Cedar, H., & Bergman, Y. (2009). Linking DNA methylation and histone modification: patterns and paradigms. Nature Reviews Genetics , 10 (5), 295- 304.
Chan, M., Huang, Y., Hartman-Frey, C., Kuo, C., Deatherage, D., Qin, H., et al. (2008). Aberrant transforming growth factor beta 1 signaling and SMAD4 nuclear translocation confer epigenetic repression of adam19 in ovarian cancer. Neoplasia , 10 (9), 908-919.
Chan, T., Glockner, S., Yi, J., Chen, W., Van Neste, L., Cope, L., et al. (2008). Convergence of mutation and epigenetic alterations identifies common genes in cancer that predict for poor prognosis. Plos Medicine , 5 (5), e114.
Cheng, A., Jin, V., Fan, M., Smith, L., Liyanarachchi, S., Yan, P., et al. (2006). Combinatorial analysis of transcription factor partners reveals recruitment of c- MYC to estrogen receptor-alpha responsive promoters. Mollecular Cell , 21, 393- 40.
Chibon, F., Lagarde, P., Salas, S., Perot, G., Brouste, V., Tirode, F., et al. (2010). Validated prediction of clinical outcome in sarcomas and multiple types of cancer on the basis of gene expression signature related to genome complexity. Nature Medicine , 16, 781-787.
Chou, J., Su, H., Chen, L., Liao, Y., Hartman-Frey, C., Lai, Y., et al. (2010). Promoter hypermethylation of FBXO32, a novel tgf-beta/smad4 target gen and tumor suppressor, is associated with poor prognosis in human ovarian cancer. Lab Investigation , 90, 414-425.
176
Clarke, M., Dick, J., Dirks, P., Eaves, C., Jamieson, C., Jones, D., et al. (2006). Cancer stem cells - perspectives on current status and future directions: aacr workshop on cancer stem cells. Cancer Research , 66 (19), 9339- 9344.
Coolen, M., Statham, A., Gardiner-Garden, M., & Clark, S. (2007). Genomic profiling of CpG methylation and allelic specificity using quantiative high- throughput mass spectrometry: critical evaulation and improvements. Nucleic Acids Research , 35 (18), e119.
Coulondre, C., Miller, J., Farabaugh, P., & Gilbert, W. (1978). Molecular basis of base substitution hotspots in Escherichia coli. Nature , 274 (5673), 775-780.
Davuluri, R., Grosse, I., & Zhang, M. (2001). Computational identification of promoters and first exons in the human genome. Nature Genetics , 29 (4), 412- 417.
De Caestecker, M., Yahata, T., Wang, D., Parks, W., Huang, S., Hills, C., et al. (2000). The smad4 activation domain (SAD) is a proline-rich, p300-dependent transcriptional activation domain. Journal of Biological Chemistry , 275, 2115- 2122.
Deatherage, D. E., Potter, D., Yan, P. S., Huang, T. H., & Lin, S. (2009). Methylation analysis by microarray. Methods in Molecular Biology , 556, 117-139.
Dennler, S., Itoh, S., Vivien, D., ten Dijke, P., Huet, S., & Gauthier, J. (1998). Direct binding of smad3 and smad4 to critical TGF beta-inducible elements in the promoter of human plasminogen activator inhibor-type 1 gene. EMBO Journal , 17, 3091-3100.
Derynck, R., & Zhang, Y. (2003). Smad-dependent and Smad-independent pathways in TGF-beta family signalling. Nature , 425, 577-584.
Derynck, R., Akhurst, R., & Balmain, A. (2001). TGF-beta signaling in tumor suppression and cancer progression. Nature Genetics , 29, 117-129.
Du, T., & Zamore, P. (2005). microPrimer: the biogenesis and function of microRNA. Development , 132, 4645-4652.
Easton, D., & Rosalind, A. (2008). Genome-wide association studies in cancer. Human molecular genetics , 17 (R2), R109-R115.
177
Easton, D., Pooley, K., Dunning, A., Pharoah, P., Thompson, D., Ballinger, D., et al. (2007). Genome-wide association study identifies novel breast cancer susceptibility loci. Nature , 447 (7148), 1087-93.
Einzig, A., Wiernik, P., Sasloff, J., Runowicz, C., & Goldberg, G. (1992). Phase II study and long-term follow-up of patients treated with taxol for advanced ovarian adenocarcinoma. Journal of Clinical Oncology , 10 (11), 1748-1753.
Esteller, M. (2008). Epigenetics in cancer. New England Journal of Medicine , 358 (11), 295-304.
Esteller, M., Fraga, M., Guo, M., Garcia-Foncillas, J., Hedenfalk, I., Godwin, A., et al. (2001). DNA methylation patterns in hereditary human cancers mimic sporadic tumorigenesis. Human Molecular Genetics , 10 (26), 3001-7.
Failkow, P. E. (1987). Clonal devellopment, stem-cell differentiation, and clincial remission in acute nonlymphocytic leukemia. New England Journal of Medicine , 317, 468-473.
Fei, T., Xia, K., Li, Z., Zhou, B., Zhou, S., Chen, H., et al. (2010). Genome wide mapping of SMAD target genes reveals the role of BMP signaling in embryonic stem cell fate determingation. Genome research , 20, 36-44.
Feng, X., & Derynck, R. (2005). Specificity and versatility in TGF-beta signaling through smads. Annual Reviews Cellular Developmental Biology , 21, 659-693.
Filipowicz, W., Bhattacharyya, S., & Sonenberg, N. (2008). Mechanisms of post- transcriptional regulation by microRNAs: are the answers in sight? Nature Reviews Genetics , 9, 102-114.
Friedman, J., & Jones, P. (2009). MicroRNAs: critical mediators of differentiation, development and disease. Swiss Medicine weekly , 139 (33), 466-472.
Frietze, S., X, L., Jin, V., & Farnham, P. (2010). Genomic targets of the KRAB and SCAN domain-containing zince finger protein. Journal of Biological Chemistry , 4, 1393-1403.
Fullwood, M., Liu, M., Pan, Y., Liu, J., Xu, H., Mohamed, Y., et al. (2009). An oestrogen-receptor alpha bound human chromatin interactome. Nature , 462 (7269), 58-64.
178
Furuse, M., Fujita, K., Hiragi, T., Fujimoto, K., & Tsukita, S. (1998). Claudin-1 and -2: novel integral membrane proteins localizing at tight junctions with no sequence similarity to occludin. Journal of Cell Biology , 141 (7), 1539-50.
Futscher, B., Oshiro, M., Wozniak, R., Holtan, N., Hanigan, C., Duan, H., et al. (2002). Role for DNA methylation in the control of cell type specific maspin expression. Nature Genetics , 31 (2), 175-179.
Gazin, C., Wajapeyee, N., Gobeil, S., Virbasius, C., & Green, M. (2007). An elaborate pathway required for Ras-mediated epigenetic silencing. Nature , 449 (7165), 1073-1078.
Gimeno, R. (2007). Fatty acid transport proteins. Current Oppinion in Lipidology , 18 (3), 271-276.
Gomis, R., Alarco, N., He, W., Wang, Q., Seoane, J., Lash, A., et al. (2006). A FoxO-Smad synexpression group in human keratinocytes. Proceedings of the National Academy of Sciences of the USA , 103, 12747-12752.
Griffin, J., & Laowenberg, B. (1986). Clonogenic cells in acute myeloblastic leukemia. Blood , 68, 1185-1195.
Gu, F., Hsu, H., Hsu, P., Wu, J., Ma, Y., Parvin, J., et al. (2010). Interference of hierarchal regulatory network of estrogen-dependent breast cancer through ChIP based data. BMC Systems Biology (4), 170.
Hajdu, S. I. (2010). A note from history: landmarks in history of cancer, part 1. Cancer , 117 (5), 1097-1102.
Hanahan, D., & Weinberg, R. (2000). The hallmarks of cancer. Cell , 100 (1), 57- 70.
Heldin, C., Miyazono, K., & Dijke, P. (1997). TGF-beta signaling from cell membrane to nucleus through smad protines. Nature , 390, 465-471.
Hirschhorn, J., & Daly, M. (2005). Genome-wide association studies for common diseases and complex traits. Nature Reviews Genetics , 6 (2), 95-108.
Holliday, R., & Pugh, J. (1975). DNA modification mechanisms and gene activity during development. Science , 187 (4173), 226-232.
179
Howlader, N., Noone, A., Krapcho, M., Neyman, N., Aminou, R., Waldron, W., et al. (1975-2008). SEER cancer statistics review. http://seer.cancer.gov/csr/1975_2008 , National Cancer Institute. Bethesda, MD.
Hromas, R., North, J., & Burns, C. (1987). Decreased cisplatin uptake by resistant L1210 leukemia cells. Cancer Letters , 36 (2), 197-201.
Hsu, P., Deatherage, D., Rodriguez, B., Liyanarachchi, S., Weng, Y., Zuo, T., et al. (2009). Xenoestrogen-induced epigenetic repression of microRNA-9-3 in breast epithelial cells. Cancer Research , 69 (14), 5936-5945.
Huang, D., Sherman, B., & Lempicki, R. (2009). Systematic and integrative analysis of large gene lists using DAVID bioinformatics resource. Nature Protocols , 4, 44-57.
Huang, T., Laux, D., Hamlin, B., Tran, P., Tran, H., & Lubahn, D. (1997). Identification of DNA methylation markers for human breast carcinomas using the methylation-sensitive restriction fingerprinting technique. Cancer Research , 57, 1030-1034.
Huang, T., Perrry, M., & Laux, D. (1999). Methylation profiling of CpG islands in human breast cancer cells. Human Molecular Genetics , 8 (3), 459-470.
ICON Group. (2002). Paxitaxel plus carboplatin versus standard chemotherapy with either single-agent carboplatin or cyclophosphamide, doxorubicin, and cisplatin in women with ovarian cancer: the ICON3 randomised trial. Lancet , 360 (9332), 505-515.
Ikushima, H., Komuro, A., Isogaya, K., Shinozaki, M., Hellman, U., Miyazawa, K., et al. (2008). An Id- like mollecule, HHM, is a synexpression group restricted regulator of TGF-beta signaling. EMBO Journal , 27, 2955-2965.
Iorio, M., Ferracin, M., Liu, C., Veronese, A., Spizzo, R., Sabbioni, S., et al. (2005). MicroRNA gene expression deregulation in human breast cancer. Cancer Research , 65 (16), 7065-7070.
Iorio, M., Viosone, R., Di Leva, G., Donati, V., Petrocca, F., Casalini, P., et al. (2007). MicroRNA signatures in human ovarian cancer. Cancer Research , 67 (18), 8699-8707.
180
Irizarry, R., Ladd-Acosta, C., Wen, B., Wu, Z., Montano, C., Onyango, P., et al. (2009). The human colon cancer methylome shows similar hypo- and hypermethylation at conserved tissue-specific CpG island shores. Nature Genetics , 41 (2), 178-86.
Ishida, S., Lee, J., Thiele, D., & Herskowitz, I. (2002). Uptake of the anticancer drug cisplatin mediated by the copper transporter Ctr1 in yeast and mammals. Proceedings of the National Academy of Sciences of the USA , 99 (22), 14298- 14302.
Jacinto, F., Ballestar, E., & Esteller, M. (2008). Methyl-DNA immunoprecipitation (MeDIP): hunting down the DNA methylome. Biotechniques , 44 (1), 35-43.
Jiang, J., Lee, E., Gusev, Y., & Schmittgen, T. (2005). Real-time expression profiling of microRNA precursors in human cancer cell lines. Nucleic Acids Research , 33 (17), 5394-5403.
Jin, V., Rabinovich, A., Squazzo, S., Green, R., & Farnham, P. (2006). A computational genomics approach to identify cis-regulatory modules for chromatin immunoprecipitation microarray data - a case study using E2F1 in cancersq. Genome Research , 16, 1585-1595.
Jin, V., Spostolos, J., Nagisetty, N., & Farnham, P. (2009). W-ChIPMotifs: a web application tool for de novo motif discovery from ChIP-based high-throughput data. Bioinformatics , 25, 3191-3193.
John, B., Enright, A., Aravin, A., Tuschl, T., Sander, C., & Marks, D. (2004). Human microRNA targets. Plos Biology , 2 (11), e363.
Johnson, D., Mortazavi, A., Myers, R., & Wold, B. (2007). Genome-wide mapping of in vivo protein-DNA interactions. Science , 316, 1497-1502.
Jones, P., & Baylin, S. (2007). The epigenomics of cancer. Cell , 128 (4), 683- 692.
Jones, P., & Laird, P. (1999). Cancer epigenetics comes of age. Nature Genetics , 21 (2), 163-167.
Jones, P., Veenstra, G., Wade, P., Vermaak, D., Kass, S., Landsberger, N., et al. (1998). Methylated DNA and MeCP2 recruit histone deacetylase to repress transcription. Nature Genetics , 19, 187-191.
181
Kang, Y., Tulley, S., Gupta, G., Sergaova, I., Chen, C., Manova, T. K., et al. (2005). Breast cancer bone metastasis mediated by the smad tumor suppressor pathway. Proceedings of the National Accademy of Sciences of the USA , 102 (39), 13909-13914.
Kent, W., Sugnet, C., Furey, T., Roskin, K., Pringle, T., Zahler, A., et al. (2002). The human gneome browser at UCSC. Genome Research , 12, 996-1006.
Kho, M., Baker, D., Layoon, A., & Smith, S. (1998). Stalling of Human DNA (Cytosine-5) Methyltransferase at Single Strand Conformers form a Site of Dynamic Mutation. Journal of Molecular Biology , 275 (1), 67-79.
Knudson, A. (20001). Two genetic hits (more or less) to cancer. Nature Reviews Cancer , 1 (2), 157-62.
Koberle, B., Masters, J., Hartley, J., & Wood, R. (1999). Defective repair of cisplatin-induced DNA damage caused by reduced XPA protein in testicular germ cell tumours. Current Biology , 9 (5), 273-276.
Koinuma, D., Tsutsumi, S., Kamimura, N., Imamura, T., Aburatani, H., & Miyazono, K. (2009). Promoter-wide analysis of smad4 binding sites in human epithelial cells. Cancer Science , 100 (11), 2133-2142.
Koinuma, D., Tsutsumi, S., Kamimura, N., Taniguchi, H., Miyazawa, K., Sunamura, M., et al. (2009). Chromatin immunoprecipitation on microarray analysis of smad 2/3 binding sites reveals roles of ETS1 and TFAP2A in transforming growth factor beta signaling. Mollecular Cell biology , 29, 172-186.
Komatsu, M., Sumizawa, T., Mutoh, M., Chen, Z., Terada, K., Furukawa, T., et al. (2000). Copper-transporting P-type adenosine triphosphatase (ATP7B) is associated with cisplatin resistance. Cancer Research , 60 (5), 1312-1316.
Krek, A., Grun, D., Poy, M., Wolf, R., Rosenberg, L., Epstein, E., et al. (2005). Combinatorial microRNA target predictions. Nature Genetics , 37, 495-500.
Krivtsov, A., Twomey, D., Feng, Z., Stubbs, M., Wang, Y., Faber, J., et al. (2006). Transformation from committed progenitor to leukaemia stem cell initiated by MLL-AF9. Nature , 442 (7104), 818-22.
Lai, G., Ozols, R., Smyth, F., Young, R., & Hamilton, T. (1988). Enhanced DNA repair and resistance to cisplatin in human ovarian cancer. Biochemical Pharmacology , 37 (24), 4597-4600.
182
Lan, X., Bonneville, R., Apostolos, J., Wang, W., & Jin, V. (2011). W-ChIPeaks: a comprehensive web application tool to process ChIP-chip and ChIP-seq data. Bioinformatics , 27, 428-430.
Lee, R., Feinbaum, R., & Ambros, V. (1993). The C. elegans heterochronic gene lin-4 encodes small RNAs with antisense complementarity to lin-14. Cell , 75 (5), 843-854.
Lee, T., Johnstone, S., & Young, R. (2006). Chromatin immunoprecipitation and microarray-based analysis of protein location. Nature Protocols , 1, 729-748.
Lehmann, U., Hasemeier, B., Christgen, M., Muller, M., Romermann, D., Langer, F., et al. (2008). Epigenetic inactivation of microRNA gene hsa-mir-9-1 in human breast cancer. Journal of Pathology , 214 (1), 17-24.
Lewis, B., Burge, C., & Bartel, D. (2005). Conserved seed pairing, often flanked by adenosines, indicates that thousands of human genes are microRNA targets. Cell , 120 (1), 15-20.
Li, M., Balch, C., Montgomery, J., Jeong, M., Chung, J., Yan, P., et al. (2009). Integrated analysis of DNA methylation and gene expression reveals specific signaling pathways associated with platinum resistance in ovarian cancer. BMC Medical Genomics , 2 (34).
Lim, S., & Hoffmann, F. (2006). Smad4 cooperates with lymphoid enhancer- binding facotr 1/T cell -specific factor to increase c-myc expression in the absence of TGF-signaling. Proceedings of the National Academy of Sciences , 103, 18580-18585.
Lister, R., O'Malley, R., Tonti-Filippini, J., Gregory, B., Berry, C., Millar, A., et al. (2008). Highly integrated single-base resolution maps of the epigenome in Arabidopsis. Cell , 133 (3), 523-36.
Lister, R., Pelizzola, M., Dowen, R., Hawkins, R., Hon, G., Tonti-Filippini, J., et al. (2009). Human DNA methylomes at base resolution show widespread epigenomic differences. Nature , 462 (7271), 315-322.
Long, X., & Nephew, K. (2006). Fulvestrant (ICI 182,780)-dependent interatcting proteins mediate immobilization and degradation of estrogen receptor-alpha. Journal of Biological Chemistry , 281, 9607-9615.
183
Lu, K., Patterson, A., Wang, L., Marquez, R., Atkinson, E., Baggerly, K., et al. (2004). Selection of potential markers for epithelial ovarian cancer with gene expression arrays and recursive descent partition analysis. Clinical Cancer Research , 10 (10), 3291-3300.
Mann, S., & Andrews, P. H. (1991). Modulation of cis- diamminedichloroplatinum(II) accumulation and sensitivity by forskolin and 3- isobutyl-1-methylxanthine in sensitive and resistant human ovarian carcinoma cells. International Journal of Cancer , 48 (6), 866-872.
Massague, J., Blain, S., & Lo, R. (2000). TGFbeta signaling in growth control, cnacer, and heritable disorders. Cell , 103, 295-309.
Maynard, T., Sikich, L., Lieberman, J., & LaMantia, A. (2001). Neural development, cell-cell signaling, and the "two-hit" hypothesis of schizophrenia. Schizophrenia Bulletin , 27 (3), 457-476.
McGuire, W., Hoskins, W., Brady, M., Kucera, P., Partridge, E., Look, K., et al. (1996). Cyclophosphamide and cisplatin compared with paclitaxel and cisplatin in patients with stage III and stage IV ovarian cancer. New England Journal of Medicine , 334 (1), 1-6.
Meissner, A., Mikkelsen, T., Gu, H., Wernig, M., Hanna, J., Sivachenko, A., et al. (2008). Genome-scale DNA methylation maps of pluripotent and differentiated cells. Nature , 454 (7205), 766-70.
Mikkelsen, T., Ku, M., Jaffe, D., Issac, B., Lieberman, E., Giannoukous, G., et al. (2007). Genome-wide maps of chromatin state in pluripotent and lineage- committed cells. Nature , 448, 553-560.
Milosavljevic, A. (2010). Putting epigenome comparison into practice. Nature biotechnology , 28 (10), 1053-56.
Miyazawa, K., Shinozaki, M., Hara, T., Furuya, T., & Miyazono, K. (2002). Two major smad pathways in tgf-beta superfamily signaling. Genes Cells , 7, 1191- 1204.
Muggia, F., Braly, P., Brady, M., Sutton, G., Niemann, T., Lentz, S., et al. (2000). Phase III randomized study of cisplatin versus paclitaxel versus cisplatin and paclitaxel in patients with suboptimal stage III or IV ovarian cancer: a gynecologic oncology group study. Journal of Clinical Oncology , 18 (1), 106-115.
184
Mutch, D. (2002). Surgical management of ovarian cancer. Seminars in Oncology , 1 (S1), 3-8.
Nan, X., Ng, H., Johnson, C., Laherty, C., Turner, B., Eisenman, R., et al. (1998). Transcriptional repression by the methyl-CpG-binding protein MeCP2 involves a histone deacetylase complex. Nature , 393, 386-389.
Naso, M., Uitto, J., & Klement, J. (2003). Transcriptional control of the mouse Col7a1 gene in keratinocytes: basal and transforming growth factor-beta regulated expression. Journal of Investigative Dermatology , 121, 1469-1478.
Nielsen, R., Pedersen, T., Hagenbeek, D., Moulos, P., Siersbaek, R., Megens, E., et al. (2008). Genome-wide profiling of PPARgamma RXR and RNA polymerase II occupancy reveals temporal activation of distinct metabolic pathways and changes in RXR dimer composition durring adipogenesis. Genes and Development , 22 (21), 2953-2967.
Nilsson, E., & Skinner, M. (2002). Role of transforming growth factor beta in ovarian surface epithelium biology and ovarian cancer. Reproductive biomedicine online , 5, 254-258.
Oren, T., Torregroza, I., & Evans, T. (2005). An Oct-1 binding site mediates activation of the gata2 promoter by BMP signaling. Nucleic Acids Research , 33, 4357-4367.
Ozols, R. (2006). Systemic therapy for ovarian cancer: current status and new treatments. Seminars in Oncology , 2 (S6), S3-S11.
Ozols, R. (2005). Treatment goals in ovarian cancer. International journal of Gynecological Cancer , 15 (supplement 1), 3-11.
Papageorgis, P., Lambert, A., Ozturk, S., Gao, F., Pan, H., Manne, U., et al. (2010). Smad signaling is required to maintain epigenetic silencing during breast cancer progression. Cancer Research , 70 (3), 968-978.
Park, C., Bergsagel, D., & McCulloch, E. (1971). Mouse myeloma tumor stem cells: a primary cell culture assay. Journal of the National Cancer Institute , 46 (2), 411-422.
Qin, H., Chan, M., Liyanarachchi, S., Balch, C., Potter, D., Souriraj, I., et al. (2009). An intergrative ChIP-chip and gene expression profiling to model smad regulatory modules. BMC Systems Biology , 3 (73).
185
Rabik, C., & Dolan, M. (2007). Molecular mechanisms of resitance and toxicity associated with platinating agents. Cancer Treatment Reviews , 33 (1), 9-23.
Richardson, B. (2003). Impact of again on DNA methylation. Ageing Research Reviews , 2 (3), 245-261.
Ries, L., Melbert, D., Krapcho, M., Stinchcomb, D., Howlader, N., Horner, M., et al. (1975-2005). SEER Cancer Statistics Review. http://seer.cancer.gov/csr/1975_2005/ , National Cancer Institute. Bethesda, MD.
Riggs, A. (1975). X inactivation, differentiation, and DNA methylation. Cytogenetics and Cell Genetics , 14 (1), 9-25.
Roman-Gomez, J., Agirre, X., Jimenez-Velasco, A., Arqueros, V., Vilas-Zornoza, A., Rodriquez-Otero, P., et al. (2009). Epigenetic regulation of MicroRNAs in acute lymphoblastic leukemia. Journal of Clinical Oncology , 27, 1316-1322.
Rüffer, C., & Gerke, V. (2004). The C-terminal cytoplasmic tail of claudins 1 and 5 but not its PDZ-binding motif is required for apical localization at epithelial and endothelial tight junctions. European Journal of Cell Biology , 83 (4), 135-44.
Sandercock, J., Parmar, M., Torri, V., & Qian, W. (2002). First-line treatment for advanced ovarian cancer: paclitaxel, platinum and the evidcne. British Journal of Cancer , 87 (8), 815-824.
Scarano, E., Iaccarino, M., Grippo, P., & Parisi, E. (1967). The heterogeneity of theymine methyl group origin in DNA pyrimidine isostichs of developing sea urchin embryos. Proceedings of the National Accademy of Sciences of the USA , 57 (5), 1394-1400.
Schickel, R., Boyerinas, B., Park, S., & Peter, M. (2008). MicroRNAs: key players in the immune system, differentiation, tumorigenesis, and cell death. Oncogene , 27 (45), 5959-5974.
Schilling, E., Chartouni, C., & Rehli, M. (2009). Allele-specific DNA methylation in mouse strains is mainly determined by cis-acting sequences. Genome Research , 19 (11), 2028-35.
Selbach, M., Schwanhäusser, B., Thierfelder, N., Fang, Z., Khanin, R., & Rajewsky, N. (2008). Widespread changes in protein synthesis induced by microRNAs. Nature , 455 (7209), 58-63.
186
Seoane, J., Le, H., Shen, L., Anderson, S., & Massague, J. (2004). Integration of smad and forkhead pathways in the control of neuroepithelial and glioblastoma cell proliferation. Cell , 117, 211-223.
Serre, D., Lee, B., & Ting, A. (2010). MBD-isolated genome sequencing provides a high-throughput and coprehensive survey of DNA methylation in the human genome. Nucleic Acids Research , 38 (2), 253-257.
Shannon, P., Markiel, A., Ozier, O., Baliga, N., Wang, J., Ramage, D., et al. (2003). Cytoscape: a software environment for integrated models of biomolecular interatcion networks. Genome Research , 13, 2498-504.
Shen, L., Guo, Y., Chen, X., Ahmed, S., & Issa, J. (2007). Optimizing annealing temperature overcomes bias in bisulfite PCR methylation analysis. Biotechniques , 42 (1), 48-58.
Shi, Y., & Massague, J. (2003). Mechanisms of TGF-beta signaling from cell membrane to the nucleus. Cell , 113, 685-700.
Shibata, M., Kurokawa, D., Nakao, H., Ohmura, T., & Aizawa, S. (2008). MicroRNA-9 modulates cajal-retzius cell differentiaion by suppressing foxg1 expression in mouse medial pallium. The Journal of Neuroscience , 28 (41), 10415-10421.
Smiraglia, D., Rush, L., Fruhwald, M., Dai, Z., & Held, W. (2001). Excessive CpG island hypermethylation in cancer cell lines versus primary human malignancies. Human Molecular Genetics , 10, 1413-1419.
Stordal, B., & Davey, M. (2007). Understanding cisplatin resistance using cellular models. IUBMB Life , 59 (11), 696-699.
Takahashi, K., & Yamanaka, S. (2006). Induction of pluripotent stem cells from mouse embryonic and adult fibroblast cultures by defined factors. Cell , 126 (4), 663-76.
Takai, D., & Jones, P. (2002). Comprehensive analysis of CpG islands in human chromosomes 21 and 22. Proceedings of the National Academy of Sciences of the USA , 99 (6), 3740-3745.
187
Tanaka, Y., Kobayashi, H., Suzuki, M., Kanayama, N., & Terao, T. (2004). Transforming grwoth factor-beta-1 dependent urokinase up-regulation and promotion of invasion are involved in Src-MAPK-dependent signaling in human ovarian cancer cells. Journal of Biological CHemistry , 279, 8567-8576.
The International HapMap Consortium. (2005). A haplotype map of the human genome. Nature , 437 (7063), 1299-1320.
Van De Vijver, m., He, Y., Van't Veer, L., DAI, H., Hart, A., Voskuil, D., et al. (2002). A gene-expression signature as a predictor of survial in breast cancer. New England Journal of Medicine , 347, 1999-2009.
Venter, J., MD, A., EW, M., PW, L., RJ, M., GG, S., et al. (2001). The sequence of the humang genome. Science , 291 (5507), 1304-51.
Visavader, J., & Lindeman, G. (2008). Cancer stem cells in solid tumours: accumulating evidence and unresolved questions. Nature Reviews Cancer , 8, 755-768.
Wang, D., & Lippard, S. (2005). Cellular processing of platinum anticancer drugs. Nature Reviews Drug Discovery , 4 (4), 307-320.
Wang, Q., W, L., Zhang, Y., Yuan, X., Xu, K., Yu, J., et al. (2009). Androgen receptor regulates a distinct transcription program in androgen-independent prostate cancer. Cell , 138 (2), 245-256.
Wang, X., Martindale, J., & Holbrook, N. (2000). Requirement for EFK activation in cisplatin-induced apoptosis. Journal of Biological Chemistry , 275 (50), 39435- 39443.
Wani, M., Taylor, H., Wall, M., Coggon, P., & McPhail, A. (1971). Plant antitumor agents. VI. The isolation and structure of taxol, a novel antileukemic and antitumor agent from Taxus brevifolia. Journal of the American Chemical Society , 93 (9), 2325-7.
Welboren, W., Van Driel, M., Janssen-Megens, E., Van Heeringen, S., Sweep, F., Span, P., et al. (2009). ChIP-Seq of ERα and RNA polymerase II defines genes differentially responding to ligands. The EMBO Journal , 28 (10), 1418- 1428.
188
Wiernik, P., Schwartz, E., Strauman, J., Dutcher, J., Lipton, R., & Paietta, E. (1987). Phase I clinical and pharmacokinetic study of taxol. Cancer Reasearch , 47 (9), 2486-2493.
Wong, A., & Leung, P. (2007). Role of endocrine and growth factors on the ovarian surface epithelium. Journa of Obstinetric Gynaecology , 33, 3-16.
Yamada, S., Baldwin, R., & Karlan, B. (1999). Ovarian carcinoma cell cultures are resistant to TGF-beta1-mediated growth inhibition despite expression of functional receptors. Gynecological Oncology , 75, 72-77.
Yan, P., Perry, M., Laux, D., Asare, A., Caldwell, C., & Huang, T. (2000). CpG island arrays: an application toward deciphering epigenetic signatures of Breast cancer. Clinical Cancer Research , 6, 1432-1438.
Yan, P., Potter, D., Deatherage, D., Lin, S., & Huang, T. (2009). Differential Methylation Hybridization: Profiling DNA methylation with a high-density CpG island microarray. In Tost, DNA methylation methods and protocols (pp. 89-106).
Yang, S., Sharrocks, A., & Whitmarsh, A. (2003). Transcriptional regulation by the MAP kinase signaling cascades. Gene , 320, 3-21.
Zhang, S., Balch, C., Chan, M., Lai, H., Matei, D., Schilder, J., et al. (2008). Identification and characterization of ovarian cancer-initiating cells from primary human tumors. Cancer Research , 68 (11), 4311-4320.
Zhang, X., Yazaki, J., Sundaresan, A., Cokus, S., Chan, S., Chen, H., et al. (2006). Genome-wide high-resolution mapping and functional analysis of DNA methylation in arabidopsis. Cell , 126 (6), 1189-201.
189