Genetic Variations Associated with Resistance to Doxorubicin and Paclitaxel in Breast Cancer

GENETIC VARIATIONS ASSOCIATED WITH RESISTANCE TO DOXORUBICIN AND PACLITAXEL IN BREAST CANCER

Irada Ibrahim-zada

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy

Department of Laboratory Medicine and Pathobiology University of Toronto

Genetic variations associated with resistance to doxorubicin and paclitaxel in breast cancer

Irada Ibrahim-zada

Doctor of Philosophy

Department of Laboratory Medicine and Pathobiology University of Toronto

2010

Abstract

Anthracycline- and taxane-based regimens have been the mainstay in treating breast cancer patients using chemotherapy. Yet, the genetic make-up of patients and their tumors may have a strong impact on tumor sensitivity to these agents and to treatment outcome. This study represents a new paradigm assimilating bioinformatic tools with in vitro model systems to discover novel genetic variations that may be associated with chemotherapy response in breast cancer. This innovative paradigm integrates drug response data for the NCI60 cell line panel with genome-wide Affymetrix SNP data in order to identify genetic variations associated with drug resistance.

This genome wide association study has led to the discovery of 59 candidate loci that may play critical roles in breast tumor sensitivity to doxorubicin and paclitaxel. 16 of them were mapped within well-characterized genes (three related to doxorubicin and 13 to paclitaxel).

Further in silico characterization and in vitro functional analysis validated their differential expression in resistant cancer cell lines treated with the drug of interest (over-expression of

RORA and DSG1, and under-expression of FRMD6, SGCD, SNTG1, LPHN2 and DCT).

iii

Interestingly, three and six genes associated with doxorubicin and paclitaxel resistance, respectively, are involved in the apoptotic process in cells. A constructed interactome suggested that there is cross-talk at the Nrf-2 oxidative stress pathway between genes associated with resistance to doxorubicin and paclitaxel.

This unique GWA approach serves as a proof-of-principle study and systematically investigates targets responsible for variable response to chemotherapy in breast tumor cells and possibly the tumors of breast cancer patients. Overall, the model discovered novel candidate genes that have not been previously associated with doxorubicin and paclitaxel cytotoxicity.

Future studies will be directed at illustrating a causative relationship between the observed genomic changes and drug resistance in breast cancer patients undergoing doxorubicin and paclitaxel chemotherapy.

To my parents:

My mom, Leylakhanim, who sacrificed her PhD candidacy for her family, for her children to accomplish it.

My dad, Tofig, who is my role model in my academic endeavors.

Explore. Dream. Discover. Mark Twain

Acknowledgments

To many people who showed support and inspired me during my graduate studies, I would like to express my sincere thanks.

I express my gratitude to my supervisor, Dr. Hilmi Ozcelik and my co-supervisor Dr. Kathleen I. Pritchard, for their guidance, advice and encouragement. I have learned so much, and without you, this would not have been possible. Your support was invaluable.

I would also like to thank my thesis committee advisors, Dr. Amadeo M. Parissenti (Sudbury, ON) and Dr. Steven Gallinger, who went above and beyond their duties throughout the project for their insight, support and collaboration.

I would like to extend my gratitude to people of Ozcelik lab who have been an essential part of my graduate life and have enriched my time in Canada. The special thanks go to Dr. Sevtap Savas and Dr. Hamdi Jarjanazi who introduced me to this research project and supported me as a part of a team. I would like to thank Susan Lau, Priscilla Chan, Lynda Doughty, and Joyce Keating for advice and support, for making my time in the lab a great learning experience and for providing a friendly atmosphere. I wish to thank summer research students, Andras Lindenmaier, Lawson Eng, and Melda Esendal who helped me to carry out my project. I am grateful to my fellow graduate students for their help and advice: Dr. George Charames, Dr. Sheron Perera, Dr. Miralem Mrkonjic, Stewart Cho, Ken Kron, and Eric Tram.

I am indepted to Dr. Amadeo Parissenti‟s laboratory and to many people from Sudbury Regional Cancer Centre for support and friendship, without which completion of this work would not have been possible. Most notably, I would like to extend my thanks to Dr. Stacey Santi, Dr. Julia Romeros, and Jane Vanderklift for all the help I received, for endless hours and support, for always being within reach when I needed guidance.

Finally, I would like to thank my family and friends for their love, support and unwavering confidence in me. To my Mom and Dad: your unconditional love, support and encouragement throughout the years have allowed me to be where I am today. You have been with me every step of the way, through good times and bad, since I left my home in Azerbaijan to study in

Canada. You instilled in me the confidence that I am capable of doing anything I put my mind to. To my brother, Orhan, thank you for encouragement and endless patience.

I would like to acknowledge the support of the Canadian Breast Cancer Foundation (2008-2011), the University of Toronto Open Fellowship Award (2008-2010), University Health Network Medical Staff Association, the Canadian Institutes of Health Research (MOP-89993), the Ontario Institute for Cancer Research (02May-0159), and the Northern Cancer Research Foundation.

vii

Table of Contents

Acknowledgments...... v

Table of Contents ...... vii

List of Tables ...... xi

List of Figures ...... xiii

Abbreviations ...... xv

Chapter 1 Introduction ...... 1

1 Introduction and Literature Review ...... 1

1.1 Chapters content...... 1

1.2 Breast cancer ...... 2

1.3 Molecular portraits of breast cancer ...... 4

1.4 A new treatment paradigm: pharmacogenomics ...... 6

1.5 Pharmacogenomics in breast cancer ...... 7

1.6 Single nucleotide polymorphism and Genome-wide Association Studies ...... 8

1.7 Cancer cell lines ...... 10

1.8 Hypothesis & Objectives: ...... 10

1.8.1 Hypothesis...... 10

1.8.2 Aims ...... 11

Chapter 2 Anthracyclines and Taxanes...... 13

Abstract ...... 13

2 Antracyclines and Taxanes ...... 14

2.1 Drug resistance in breast cancer ...... 14

viii

2.2 Doxorubicin: ...... 16

2.2.1 Biology of the potentially modulated target pathways ...... 16

2.2.2 Mechanisms of Resistance: ...... 17

2.3 Paclitaxel: ...... 20

2.3.1 Biology ...... 20

2.3.2 Mechanisms of resistance ...... 20

2.4 Genome-Wide Association Studies of Resistance to Doxorubicin and Paclitaxel ...... 24

2.4.1 Doxorubicin: ...... 24

2.4.2 Paclitaxel: ...... 31

2.5 Influence of the genome-wide approach on the genetic targets ...... 35

2.6 Summary ...... 36

Chapter 3 Genome-wide Association Studies of Doxorubicin Response in Cancer ...... 37

Abstract ...... 37

3 GWAS of doxorubicin response in cancer ...... 38

3.1 Introduction ...... 38

3.2 Materials and Methods ...... 39

3.2.1 GWA analysis: ...... 39

3.2.2 mRNA Expression Analyses...... 41

3.2.3 Statistical Analysis: ...... 46

3.2.4 Karyotypic Alterations ...... 46

3.3 Results ...... 47

3.3.1 Case-control design:...... 47

3.3.2 GWA Analysis: ...... 49

3.3.3 Haplotype Analysis: ...... 53

3.3.4 Gene Expression Analyses:...... 57

3.3.5 Karyotypic Alterations: ...... 67

3.4 Discussion ...... 70

3.4.1 Genetic Analyses: ...... 70

3.4.2 Characterization of Biologic Relevance ...... 71

3.5 Summary ...... 78

Chapter 4 Genome-wide Association Studies of Taxol Response in Cancer ...... 79

Abstract ...... 79

bstract ...... 80

4 Genome-wide Association Studies of Taxol Response in Cancer ...... 80

4.1 Introduction ...... 80

4.2 Materials and Methods ...... 81

4.2.1 GWA Analysis: ...... 81

4.2.2 In silico prediction: ...... 84

4.2.3 Results ...... 84

4.3 Discussion ...... 104

4.3.1 Genome-Wide Method and Application ...... 104

4.3.2 Genetic variations and Haplotypes ...... 104

4.3.3 Drug Mechanisms and Metabolism ...... 105

4.3.4 Previous and Novel Markers ...... 105

4.4 Conclusions ...... 109

Chapter 5 Candidate genes and pathways associated with resistance to doxorubicin and paclitaxel ...... 110

5 Cross-talk between doxorubicin and paclitaxel activated resistant pathways ...... 111

5.1 Introduction ...... 111

5.2 Methods: ...... 113

5.2.1 Pathway analysis ...... 113

5.2.2 Pilot Clinical Validation ...... 114

5.2.3 Statistical Analysis ...... 114

5.3 Results ...... 115

5.3.1 Pathway analysis ...... 115

5.3.2 Clinical Validation: ...... 121

5.4 Discussion: ...... 131

Chapter 6 Summary and Future Directions ...... 136

6 Summary and future Directions ...... 136

6.1 Conclusions ...... 136

6.2 Advantages and Limitations ...... 137

6.2.1 Advantages:...... 137

6.2.2 Limitations: ...... 139

6.3 Future directions ...... 140

References ...... 142

Appendix ...... 164

List of Tables

Table 1.1: Treatment options for Breast Cancer by Stage...... 3

Table 2.1: Key mechanisms of resistance to doxorubicin ...... 19

Table 2.2: Key mechanisms of resistance to paclitaxel ...... 23

Table 2.3: SNPs/genetic variations associated with doxorubicin resistance: ...... 27

Table 2.4: SNPs/polymorphisms associated with paclitaxel resistance: ...... 33

Table 3.1: Oligonucleotides for qRT-PCR ...... 46

Table 3.2: Case-control study design of doxorubicin sensitive and resistant NCI60 cell lines .... 50

Table 3.3: Statistically significant SNPs associated with doxorubicin response in the GWA study...... 52

Table 3.4: Haplotype association analysis of DSG-DSC gene family locus...... 55

Table 3.5: Haplotype association analysis of FRMD6 locus ...... 55

Table 3.6: Haplotype association analysis of RORA locus ...... 56

Table 3.7: mRNA expression of DSG1, RORA, and FRMD6 in 300 cancer cell lines panel from CaARRAY microarray database...... 57

Table 3.8: DSG1, FRMD6, and RORA expression in breast cancer cell lines...... 59

Table 3.9: FRMD6, DSG1, and RORA expression in breast tumors...... 62

Table 3.10: cDNA expression of DSG1, RORA, and FRMD6 in A2780, MES-SA, and MCF-7cc cell lines ...... 64

Table 3.11: Comparison of the cDNA levels of the DSG1, RORA and FRMD6 between WT and resistant cell lines on qRT-PCR ...... 65

xii

Table 3.12: Karyotype analysis within the six loci of interest (Chr. 7q21, 12p12, 14q21, 15q21, 17p12, 18q12) in NCI60 cell lines...... 68

Table 3.13: Translocations of the candidate loci in NCI60 cancer cell lines...... 69

Table 4.1: Case Control Design for sensitive and resistant cell lines at 10-6 M dose group...... 86

Table 4.2: Significant SNPs (FDR Correction Value < 0.005) associated with paclitaxel response at the lower dose group of 10-6 M along with statistical information...... 90

Table 4.3: Summary of case-control analysis of novel haplotypes associated with drug response at 10-6 dose...... 94

Table 4.4: Potential changes in transcription factor binding caused by SNP variants...... 99

Table 4.5: mRNA expression: Cases vs. Controls were analyzed by Mann-Whitney test using data on the NCI60 panel available from BioGPS system...... 102

Table 5.1: Top molecular and cellular functions (by DAVID Functional Annotation Tool) ..... 115

Table 5.2: Functional pathway annotation by DAVID: plasma membrane...... 119

Table 5.3: Results from ANOVA test for 4 Affymetrix U133A-plus 2.0 probes from GEO19615 database among 102 breast patients treated with anthracyclines ...... 122

Table 5.4: Kaplan-Meier survival analysis of FRMD6, RORA, and DSG1 in breast cancer patients from GEO19615 database...... 124

Table 5.5: Descriptive characteristics of the samples from MAQS-II dataset (E-GEOD-20194) ...... 127

Table 5.6: Comparison of gene expression in breast cancer samples treated with T/FAC or T/FEC from MAQS II dataset (E-GEOD-20194) ...... 128

xiii

List of Figures

Figure 3-1: RNA biointegrity results...... 43

Figure 3-2: Dissociation curve for DSG1 (A), RORA (B), FRMD6 (C), and S28 (D) in qRT-PCR (from SDS 2.2.2 software)...... 44

Figure 3-3: Amplification plots of DSG1, FRMD6, RORA, and S28 ...... 45

Figure 3-4: Doxorubicin dose response and identification of sensitive (cases) on the left and resistant (controls) cells on the right by Kernel density estimation...... 48

Figure 3-5: Haplotype blocks of Desmoglein-Desmocolin gene family (DSG1-4, DSC1-3)...... 53

Figure 3-6: LD blocks of Desmoglein-Desmocolin gene family (DSG1-4, DSC1-3), RORA and FRMD6 genes...... 54

Figure 3-7: Comparison of the cDNA levels of DSG1, RORA and FRMD6 between WT and resistant cell lines on qRT-PCR...... 66

Figure 3-8: Gene-gene interactions: DSG1, FRMD6, RORA with doxorubicin and epirubicin. Candidate genes are in red...... 74

Figure 3-9: Expression comparison of DSG1, RORA, and FRMD6 in normal and tumor breast tissues of 61 individuals ...... 77

Figure 4-1: SNP selection process ...... 83

Figure 4-2: Segregation of NCI60 cell line panel for the 10-6 M dose group of paclitaxel into sensitive and resistant groups...... 85

Figure 4-3: Results of Haploview analysis to identify haplotypes more significantly associated with drug response than the originally identified SNP...... 97

Figure 4-4: Interaction between our candidate genes with paclitaxel and other genes belonging to relevant pathways...... 100

xiv

Figure 4-5: mRNA expression in NCI60 cells in 8 genes on Affymetrix U133A chip as determined by Mann-Whitney test...... 101

Figure 5-1: Top pathways associated with the doxorubicin and paclitaxel resistance in NCI60 cell lines...... 116

Figure 5-2: Top pathways linked to the SNPs associated with doxorubicin and paclitaxel resistance...... 117

Figure 5-3: Survival plot for FRMD6, RORA, and DSG1 in the GEO19615 database ...... 125

Figure 5-4: Role of candidate genes associated with resistance to doxorubicin and paclitaxel in oxidative stress response...... 133

Abbreviations

ABC ATP-binding cassette ABI Applied Biosystems AC doxorubicin and cyclophosphamide AMISE Approximate Mean Integrated Squared Error ANOVA analysis of variance ATF artificial transcription factors ATP adenosine triphosphate BIG Breast International Group BTBD12 BTB (POZ) domain containing 12 CCDC26 coiled-coil domain containing 26 cCGH chromosomal CGH cDNA complementary DNA CFTR cystic fibrosis transmembrane conductance regulator CGH comparative genomic hybridization CK cytokeratin Cmax the maximum plasma concentration of the drug CMF cyclophosphamide, methotrexate, and fluorouracil CNV copy number variation CO2 carbon dioxide CT cycle threshold method DCIS ductal carcinoma in situ DCT DOPAchrome tautomerase DFS disease-free survival DNA deoxyribonucleic acid DOX doxorubicin DSG1 desmoglein 1 DTP Developmental Therapeutic Program

xvi

EDTA ethylenediaminetetraacetic acid EGFR-1 epidermal growth factor receptor-1 EORTC European Organisation for Research and Treatment of Cancer EPI Epirubicin ER estrogen receptor FAC fluorouracil, doxorubicin, and cyclophosphamide FDA Food and Drug Administration FDR-BH false discovery rate by Benjamini and Hochberg FEC fluorouracil, epirubicin, and cyclophosphamide FERM 4.1 protein, ezrin, radixin and moesin FISH fluorescence in situ hybridization FRMD6 FERM domain containing protein 6 GCRMA GC Robust Multi-array Average GEA Gene Expression Atlas GEO Gene Expression Omnibus GI50 50% growth inhibition GO Gene Ontology GRIK1 glutamate receptor, inotropic, kainate 1 GST glutathione transferase GWA genome-wide association GWAS genome-wide association studies HER2 human epidermal growth factor receptor 2 HWE Hardy-Weinberg equilibrium LCIS lobular carcinoma in situ LD linkage disequilibrium LOH loss of heterozygosity LPHN2 latrophilin 2 MAF minor allele frequency MAP microtubule-associated proteins MAQS-II the Microarray Quality Control Consortium project-II MAS5 Microarray Suite 5.0 statistical algorithm from Affymetrix

xvii

MDR multi-drug resistance Microarray in Node-Negative Disease May Avoid MINDACT Chemotherapy mRNA messenger RNA MRP multidrug resistance-associated protein M-FISH multiplex fluorescence in situ hybridization NADH nicotinamide adenine dinucleotide NCBI National Center for Biotechnology Information NCCN National Comprehensive Cancer Network NCI60 National Cancer Institute 60 cell lines panel Nrf2 nuclear factor erythroid 2-related factor 2 OR odds ratio OS overall survival pCR pathological complete response Pgp P glycoprotein PGRN Pharmacogenetics Research Network PR progesterone receptor Whole genome association analysis toolset developed by S. PLINK Purcell PTPRD protein tyrosine phosphatase receptor type D qRT-PCR quantitative real-time polymerase chain reaction RD residual disease RNA ribonucleic acid ROBO1 roundabout, axon guidance receptor, homolog 1 RORA retinoic acid receptor-related orphan receptor alpha ROS reactive oxygen species SEER Surveillance, Epidemiology, and End Results SGCD sarcoglycan, delta siRNA small interfering RNA SKY spectral karyotyping SNP single nucleotide polymorphism

xviii

SNTG1 Syntrophin, gamma 1 T/FAC paclitaxel, 5-fluorouracil, cyclophosphamide and doxorubicin T/FEC paclitaxel, 5-fluorouracil, cyclophosphamide and epirubicin TAILORx trial Trial Assigning Individualized Options for Treatment TBCI The Breast Cancer Intergroup UVB ultraviolet B WT wild-type ZNF607 zing finger protein 607

Chapter 1 Introduction

1 Introduction and Literature Review

1.1 Chapters content

This thesis consists of 6 chapters. Chapter 1 presents a review of breast cancer epidemiology, molecular classification, and treatment options and outcomes. It touches upon the field of pharmacogenomics and its application in cancer treatment. The application of genome-wide studies of genetic variations in cell-based models wraps up the discussion in this chapter. Finally, a hypothesis and aims of this study were presented. In Chapter 2, I review the mechanism of action of anthracyclines and taxanes using the example of doxorubicin and paclitaxel, respectively, related to the observed clinical phenotypes. Further, the results of previous genome- wide studies of resistance to both drugs are described. Chapter 3 describes the development and application of a genome-wide approach (GWA) to discover novel genetic variations associated with resistance to doxorubicin. This was achieved by implementing a novel case-control design on NCI60 cell lines treated with the drug of interest. An application of in silico bioinformatics tools indentified 3 candidate genes associated with a resistant phenotype. Further functional in vitro assessment was performed to validate these findings. In Chapter 4, the same approach is applied to the other drug of interest, paclitaxel. Additional bioinformatic tools are applied to validate these findings. Chapter 5 shows the integrative approach to publicly available databases in clinical validation of the in vivo and in vitro data. In addition, it describes the cross-talk between signaling pathways associated with paclitaxel and doxorubicin. In Chapter 6, an overall summary, strengths and limitations of the study, and discussion of future directions are presented.

1.2 Breast cancer

Breast cancer is the most common cancer and the second most common cause of death due to cancer in women worldwide. It is the most frequently diagnosed cancer in Canadian women. Almost 192,370 (27% of all cancer cases) new cases of breast cancer and 40,170 (15% of all cancer related deaths) breast cancer related deaths were predicted to occur in 2009 in United States. Almost 23,200 cases were expected to be diagnosed in Canada in 2010 constituting about 28% of all new cases of female cancer (1). There are approximately 2,533,193 women alive who have a history of breast cancer in USA. The lifetime risk of developing breast cancer is currently 1 in 9 women in Canada according to the Canadian Cancer Society (2, 3). One out of every 28 Canadian women are expected to die from breast cancer (3).

Breast cancer is a systemic disease with the potential for distant spread. Although early breast cancer can be curable with local or regional treatment, the presence of subclinical metastases in most cases requires systemic treatment (4). Optimal treatment includes hormonal therapy, chemotherapy, or both based on the extent of disease and tumor characteristics. Combination regimens are mainly used for adjuvant therapy because the benefits of combination treatment are greater by about 20% compared to single-agent therapy (5). A 5-year disease-free survival (DFS) and an overall survival (OS) rate for breast cancer patients with localized disease is 89% and 98%, respectively. However, 30% of women will progress to locally advanced or metastatic disease. The 5-year residual risks of recurrence are 7%, 11% and 13% in breast cancer patients with stage I, II, or III disease. The 5-year survival rate in metastatic disease is 26.7% (6-8).

The American Joint Committee on Cancer (AJCC) provides clinical and pathologic staging system based on the TNM classification where T refers to tumor, N to nodes, and M to metastasis. In terms of treatment, breast cancer is divided into (Table 1.1):

 The pure noninvasive carcinomas, which include lobular carcinoma in situ (LCIS) and ductal carcinoma in situ (DCIS) (stage 0);  Operable, local-regional invasive carcinoma with or without associated noninvasive carcinoma (clinical stage I, stage II, and some stage IIIA tumors);

 Inoperable local-regional invasive carcinoma with or without associated noninvasive carcinoma (clinical stage IIIB, stage IIIC, and some stage IIIA tumors);  Metastatic or recurrent carcinoma (stage IV) (9).

Table 1.1: Treatment options for Breast Cancer by Stage (10). BCS- breast conserving surgery, RT- radiation therapy. Cancer stage Classification Type Treatment 0 In situ Lobular carcinoma in No treatment or prophylaxis situ with tamoxifen Ductal carcinoma in situ Breast-conserving surgery (BCS) and radiation therapy (RT) I, IIa, IIb Early invasive BCS and RT IIIa, IIIb, IIIc Locally advanced Noninflammatory Chemotherapy, BCS and RT Inflammatory Chemotherapy, mastectomy, and RT IV Metastatic Initial or recurrent Chemotherapy, or RT or bisphosphonates, or palliative therapy

There is no single standard chemotherapy regimen equally efficient in advanced and metastatic breast cancer. The most frequently used combinations are fluorouracil, doxorubicin, and cyclophosphamide (FAC), fluorouracil, epirubicin, and cyclophosphamide (FEC), doxorubicin and cyclophosphamide (AC), and cyclophosphamide, methotrexate, and fluorouracil (CMF) (11). An addition of four cycles of paclitaxel to AC or of docetaxel to FEC improves disease- free survival (DFS) and overall survival rates in the adjuvant settings (12-14). Despite aggressive treatment, only 30% of women with locally advanced and inflammatory breast cancer remain cancer-free after 10 years (15). About 20% of women with early breast cancer will develop metastases. Fifty to 80 percent of women with metastatic disease refractory to hormonal treatment have a response to FAC, while 40 to 60 percent have a response to CMF. Meta-

4 analysis of randomized trials showed superiority of anthracycline-containing combinations to CMF. Combinations of taxanes and anthracyclines have response in 40 to 94 percent of women and complete remission in only 12 to 41 percent of women (16). The addition of taxanes to anthracycline-containing regimes improved overall survival from less than six months to 10-12 months and enhanced response rates from under 10% to 30-40%. However, around 90% of patients with metastatic disease will ultimately have a treatment failure as a result of resistance to anthracycline and taxanes- based metastatic regimens (6). Furthermore, despite the progress in adjuvant chemotherapy strategies, therapeutic results in breast cancer patients are far from ideal and need to be improved.

1.3 Molecular portraits of breast cancer

Breast cancer is a heterogeneous disease from both molecular and clinical standpoints. Its pathogenesis and prognosis are determined by tumor origin. Most breast cancers arise from epithelial cells lining ducts and lobules of the normal breast. There are many prognostic and predictive factors determining tumor behavior and selection of various therapies: tumor histology, clinical and pathological characteristics of the primary tumor, axillary node status, tumor hormone receptor content, ERBB2 status, presence or absence of detectable metastatic disease, patient co-morbid conditions, patient age, and menopausal status (9). Although the conventional TNM staging system remains important, it cannot be used to predict prognosis or response to treatment for breast cancer patients (17). With the development of gene microarray profiling, the following five main subtypes of breast cancer with distinct gene expression patterns have been identified by Perou et al. (18):  basal-like tumors,  luminal-A cancer (ER+ and low grade),  luminal-B cancer (ER+ and high grade),  human epithelial growth factor receptor 2 (HER2) overexpressing tumor (with high amplification of the ERBB2 gene),  normal breast-like or unclassified tumors

These groups demonstrate different levels of chemosensitivity, prognosis, and overall or relapse- free survival. The first group, basal-like breast carcinoma (8-20%), is characterized by

5 overexpression of cytokeratins (CKs) 5, 14 and 17, laminin, and fatty-acid-binding protein 7, and lack of expression of the estrogen receptor (ER), the progesterone receptor (PR) and HER-2/neu. This group is associated with the shortest disease-free (DFS) and overall survival (OS) (19). It also has a higher risk for brain and lung metastases (20). Anthracycline-based chemotherapy has been shown to be less effective in basal-like carcinomas. Another feature of tumors from this group is a dysfunction in the BRCA1 pathway possibly due to either gene promoter methylation or transcriptional inactivation.

Luminal-like breast carcinoma has been further divided into two subtypes (A and B) according to the levels of ER expression and tumor grade. Luminal type A (56-61%) is comprised of ER positive, PR+ and HER2- negative tumors with a low genomic grade similar to that of normal breast tissue. Luminal type B tumors (9-16%) are more aggressive, high grade tumors that may express EGFR-1, HER2, and cyclin E1. These correspond mainly to invasive lobular carcinomas(18).

The next subtype, the HER2-overexpressing breast carcinoma (8-16%), is further subdivided into two distinct subtypes: ER-negative and ER-positive. HER2-positive cases are less differentiated tumors with poor prognosis that are frequently associated with ductal carcinoma in situ (DCIS). Usually this subtype shows shorter time to progression and poorer overall survival. It also has a predilection for brain metastasis. Finally, the rest of tumors have been associated with normal breast-like type, or unclassified group. They usually have expression patterns similar to nonmalignant tissue. Not all tumors can be uniformly assigned to one of the described groups. Approximately 6-10% of breast carcinomas present with a mixture of highly proliferating basal cells and more differentiated luminal cells (21). Tumors in this group are close to basal-like carcinoma by its molecular pattern, but have slightly better prognosis. They are generally negative for all markers, including ER, PR, HER2, CK5 and EGFR. The so-called “triple negative breast carcinomas” fall into two abovementioned categories: the basal-like and unclassified tumors. Additional analysis of CK5/6 and EGFR markers after routine triple stain analysis (ER, PR, and HER2) provides a better diagnosis in triple-negative tumors. Despite the developed clinicopathologic parameters of tumors, there is a difference in prognosis within the groups. Indeed, the prognosis of CK14-positive basal-like breast carcinomas is better than that of patients with CK14-negative cases (19). Nevertheless, molecular profiling provides a

6 mechanism for association of gene expression patterns with the biological behavior and the clinical outcomes of patients.

1.4 A new treatment paradigm: pharmacogenomics

Pharmacogenomics is the study of molecular patterns in the human genome and their effect on the drug response (22). It reflects the expansion of knowledge by the Human Genome Project from pharmacogenetics, which was coined as a term in the 1950s (23). In the last decade, it has been rapidly expanded from an investigation of genes involved in pharmacokinetic and pharmacodynamic drug properties to contemporary genome-wide linkage analysis. There is mounting evidence supporting an early success in pharmacogenomics. Contemporary development of genomics, molecular pharmacology and bioinformatics lead to creation of the Pharmacogenetics Research Network (PGRN) that allows investigators to share resources, tools and approaches to individualized drug therapy (24). Targeted therapy-oriented research identified a biomarker assay targeting HER2/neu gene in breast cancer patients. Trastuzumab, lapatinib, and the anti-VEGF bevacizumab are successful examples of targeted therapeutics. There are many more therapy classes still under evaluation, including other anti-EGFR and anti- HER2 therapies as well as agents interfering with tyrosine kinases such as PI3K/AKT/mTOR and PARP inhibitors (25).

There are few established examples of pharmacogenomic studies successfully implemented into routine patient management. One of the first studies includes genotyping of TPMT variants for the prediction of 6-mercaptopurine-induced myelosuppression in patients with acute lymphoblastic leukemia. Determination of mutations in TMPT gene allows us to prevent severe drug toxicity due to low TMPT activity. This genetic polymorphism was highlighted as of clinical importance by the United States Food and Drug Administration (FDA) in 2003 (26). Other examples are CYP2D6 as a pharmacokinetic variant affecting effectiveness of tamoxifen, VKORC1 as a polymorphisms affecting pharmacodynamics of warfarin, and UTG1A1 as a polymorphisms associated with toxicity in patients treated with irinotecan (27).

Even though pharmacogenomics has application in broad different areas of medicine, a targeted approach is even more vital in oncology practice. Better efficacy and fewer side effects are two

7 major advantages of tailored therapy compared to standardized approaches. There were attempts to use interindividual variability in somatic tissues to predict response to conventional chemotherapy, although a few of them were successful. The genomic signatures obtained by using in vitro drug sensitivity data were combined with the independent sets of human tumors to pave the way for truly personalized medicine (28, 29). Genome-wide studies have become a new approach to the identification of genetic variations associated with drug response.

1.5 Pharmacogenomics in breast cancer

The genomic revolution led to the identification of either a single marker or a group of markers known as gene signatures or multigene classifiers. Several different strategies have been employed to create multigene predictors of response to antineoplastic agents across multiple independent data sets obtained from breast cancer patients, i.e. “top-down” and “bottom-up” genome-wide methods and candidate-gene approaches (30). In a clinical trial employing anthracycline-based chemotherapy, overexpression of the TOP2A gene in tumor was associated with higher response to anthracyclines. TOP2A amplification showed benefits in CEF over CMF therapy (31, 32) as well as in single-agent doxorubicin therapy. In the “bottom-up” approach, a 92-gene predictor set was developed to associate response with taxane treatment (30). In addition, several genetic signatures providing prognostic score were developed, including the Rotterdam 76-Gene set, Invasive gene signature, wound response indicator, the oncotype DXTM Recurrence ScoreTM, MammaPrint 70-Gene profile, Mammostrat, CellSearchTM. Only two of the gene sets, MammaPrint 70-Gene set and the 21-gene oncotype DXTM indicator are being validated in prospective clinical trials (33). In a prospective trial MINDACT (Microarray in Node-Negative Disease May Avoid Chemotherapy) the EORTC and the Breast International Group (BIG) are evaluating the prognostic value of the Mammaprint set whereas the Breast Cancer Intergroup (TBCI) are assessing OncotypeDX indicator in node-negative hormone receptor –positive patients in the prospective TAILORx (Trial Assigning Individualized Options for Treatment) trial. Although ongoing research studies target variable signaling pathways, those pathways haven‟t been implemented into standard practice and are currently undergoing evaluation in clinical trials. However, in theory at least detailed knowledge of genetic alterations would allow a more individualized treatment with lesser morbidity and mortality results. These

8 gene signatures represent a new approach to assessing the risk of recurrence and predetermining the response phenotype to chemotherapy.

1.6 Single nucleotide polymorphism and Genome-wide Association Studies

Single Nucleotide Polymorphism (SNP) is the most common type of DNA polymorphisms in the human genome. This type of genetic variation has a frequency of >1% in the general population which is much more common than genetic mutations. On average, SNPs are encountered once every 200 base pairs in the human genome. The contribution of any particular SNP to a variable disease phenotype depends on its location. SNPs that are located within the coding region of genes (intragenic) are classified based on their functional impact on the protein products to synonymous (with no alterations in amino acid codons) and non-synonymous (causing amino acid codon alterations). The rest of the SNPs (intergenic) are located within non-coding regions of the genome and mostly regarded as non-functional. They can, however, impact on gene regulatory sequences such as promoters, enhancers, and silencers. Although initially only non- synonymous SNPs were considered to be a “direct” target of genetic association studies, other SNPs have been recently implicated in susceptibility to human diseases or treatment response (34). This highlighted the implications of SNPs in promoters, introns, splice sites, intragenic and even synonymous SNPs (35). The initial idea of considering only „functional‟ SNPs has been extended to cover other SNPs in order to pick up regions that could contribute to complex diseases, namely cancer. SNPs along the same chromosome inherited as a combination of alleles form a haplotype block and are considered to be in linkage disequilibrium (LD). LD represents a measure of non-random associations between polymorphisms at different loci. The level of LD depends on genetic linkage, the rate of mutation, genetic drift and the rate of recombination. LD blocks and haplotypes represent additional tools to enhance the power of association studies. Advances in bioinformatics allow us to dramatically increase the contribution to the pool of potentially important genetic variations involved in cancer susceptibility and outcomes (36). Bioinformatics will prove to be crucial in detecting a pattern that could have diagnostic and prognostic implications in oncologic practice.

Genome-wide association (GWA) study is a high-throughput method to identify the most common genetic variations affecting the risk of development of a complex disease. This high- throughput approach allows identifying multiple candidate genes and pathways through genome analysis. The most commonly utilized approaches are large-scale genome sequencing, transcriptional and epigenetic profiling, analysis of DNA copy number, splicing aberrations and functional proteomics (37). GWAS is a more cost-efficient and powerful way to detect both rare and common variants compared with a candidate-gene approach. Current genetic technologies allow us to screen several hundred thousand SNPs in order to detect their association with the phenotype of interest. The statistical power of an association study is determined by the allele frequencies of variants and the size of their effect on the disease phenotype. SNPs with minor allele frequencies (MAF) of > 0.1 and odds ratios of > 1.3 provide adequate coverage of the human genome (38). Many SNPs are in strong LD, a non-random association of alleles at nearby loci, with other SNPs. The average LD range is 60-200kb in the general population. Although there have been an increasing number of SNPs available for full coverage of the genome, strong LD association can provide an adequate “coverage” of the region in an association study with as many as 30% of common SNPs not included in the study. The degree of LD is described in the statistical coefficient of determination, r2. As a rule of thumb, an r2 of 0.8 or greater is defined as “optimal” to provide comprehensive coverage of the region. The effect sizes of the predicted loci in GWA studies are reflected by odds ratios (OR). Interestingly, even though loci associated with individual cancer risk carries very modest power (average OR of 1.20) (39), loci associated with the pharmacogenomic effects often have much stronger effect (OR ranges from 1.95 to 25.0). This phenomenon is due to unrestrained rise in frequency of pharmacogenetic variants in the human population that remains unexposed to drugs evolutionary (40). One of the major caveats in GWA analyses is detection of false-positive and false-negative results. The general recommendation to overcome this problem is selection of variants with p-values of less than 10-6 in at least one dataset or performing a post-hoc correction for multiple comparisons and establishing replication studies to confirm the initial associations (38). Also, more than 80% of SNPs associated with the trait in GWAS are represented by intronic and intergenic SNPs (41). Despite the described problems, it seems that GWA studies have a high potential to define the genetic signatures associated with complex diseases.

1.7 Cancer cell lines

Cancer cell lines are important tools for extensive research in pharmacogenomics. They produce a controlled experimental system free of in vivo confounders with the ability to test numerous drugs with a narrow therapeutic index. They provide a cost-effective, unlimited resource of established cells that provide highly controlled environment to conduct pharmacogenomic studies (42). Although cell lines have numerous advantages, they may carry altered characteristics not present in the original tumor and the observed phenotype is often not fully reproducible in an in vivo system due to in vitro confounders. In addition, cells in vitro may not be able to regenerate pharmacokinetic features of metabolic enzymes such as CYP450 (43). The US National Cancer Institute (NCI) 60 cell line panel was developed in the late 1980s from the American Type Culture Collection repository. It represents the first cell-based model which consists of 59 cell lines (formerly 60) derived from cancers of nine different human organs, including the breast, colon, central nervous system, lymphocytes, skin melanocytes, ovary, prostate, and kidney (44). As a result of extensive studies in different aspects, detailed SNP genotyping, copy number variation (CNV), mRNA and microRNA expression, and proteomic data are publicly available for further research (43, 45, 46). Several studies successfully applied this information from NCI60 cell lines to examine the role of genetic variations in candidate genes in drug response in vitro (47-50). However, none of the previous studies attempted to identify polymorphisms on a genome-wide scale. In principle, the NCI60 panel of cancer cell lines represents an accessible framework to integrate and employ genomic signatures with pharmacological profiles in tumors.

1.8 Hypothesis & Objectives:

1.8.1 Hypothesis

I hypothesize that genetic variations are responsible for drug resistance to doxorubicin and paclitaxel in breast cancer patients. I hypothesize that the identification of new genetic signatures

11 would improve our approach to anthracycline- and taxane-based chemotherapy in breast cancer patients.

1.8.2 Aims

1.8.2.1 Aim 1: Discovery of Novel Genes and Genetic Variants involved in Resistance to Doxorubicin Treatment.

 To identify common and rare genetic profiles associated with doxorubicin response in NCI60 cell lines;  To carry out a haplotype analysis for more comprehensive investigation of the genetic association of the novel genes with resistance to doxorubicin;  To functionally validate the novel genes by measuring their expression using qRT-PCR in doxorubicin- and epirubicin-resistant cancer cell lines.

1.8.2.2 Aim 2: Discovery of Novel Genes and Genetic Variants involved in Resistance to Paclitaxel Treatment.

 To identify common and rare genetic profiles associated with paclitaxel response in NCI60 cell lines;  To carry out a haplotype analysis for more comprehensive investigation of the genetic association of the novel genes with resistance to paclitaxel;  To functional validate the association between the expression of the previously identified candidate genes and resistance to paclitaxel using paclitaxel resistant and sensitive members of the NCI60 cell panel.

1.8.2.3 Aim 3: Characterization of candidate genes in cancer cell lines.

 Characterization of the SNPs mapped within the candidate genes;  mRNA microarray expression;  Loss of heterozygosity (LOH) and comparative genomic hybridization (CGH) DNA copy number variation (CNV) data;

 Biological interpretation of the signaling pathways.

1.8.2.4 Aim 4: Discovery of the common signaling pathways associated with doxorubicin and paclitaxel- resistance in cancer and pilot validation of the findings in breast cancer patients

 To predict in silico the common mechanisms based on the Ingenuity system pathways.  To perform an exploratory validation of the differential gene expression in tumors from the breast cancer patients treated with doxorubicin and/or paclitaxel

Chapter 2 Anthracyclines and Taxanes

Abstract

In this chapter, I summarize the current understanding of the genetic factors modulating chemoresistance to doxorubicin and paclitaxel in breast cancer patients. I focus on currently available results from genome-wide association studies (GWAS) to compare them with knowledge previously obtained from conventional candidate gene approaches. I also discuss the application of various high-throughput methodologies in GWAS to identify genetic signatures associated with drug resistance. Overall, current studies showed that multigene signatures rather than a single gene are necessary to individualize management of breast cancer patients treated with anthracycline- and taxanes-based chemotherapy. To date, there is still no „gold standard‟ gene signature available that would allow us to prescreen and triage breast cancer patients about to begin to receive chemotherapy.

2 Antracyclines and Taxanes

2.1 Drug resistance in breast cancer

Drug resistance is an essential barrier to success in any therapeutic approach to breast cancer. Doxorubicin (DOX) and paclitaxel are widely used in multi-agent regimens in adjuvant therapy and as a single agent or in combinations in the metastatic setting; nevertheless, success rates even when using both agents are still low. The hazard ratios for taxanes vs. anthracyclines in single- agent trials was 1.19, p=0.011 for progression free survival with 38% and 33% response rate for taxanes and anthracyclines, respectively. Response rates in taxane-based combination trials were 57% (taxanes) vs. 46% (controls) with a hazard ratios of 0.92, p=0.031 in survival for taxanes compared with control arm (51). Although individually paclitaxel has been shown to be as effective as doxorubicin, combined application (doxorubicin with subsequent administration of paclitaxel) has a better overall survival rate. Primary and acquired drug resistance represents major challenges in the clinical efficacy of breast cancer treatment with these agents. Resistance to anthracyclines and taxanes has been defined as a recurrence within 6 to 12 months after adjuvant chemotherapeutic treatment, or tumor progression during therapy for metastatic breast cancer (52). Partly due to their well known structure and biology, and partly due to availability of these drugs on the market for many years, both represent very interesting targets to determine the fundamentals of variable response in patients. It is believed that variable response among patients is determined by genetic variations and peculiarities of each human genome. This belief forms a framework to develop a pharmacogenomic approach to cancer treatment.

Knowledge of the molecular basis of drug pharmacokinetics and pharmacodynamics formed the grounds for the initial approaches to pharmacogenomic studies of anthracyclines and taxanes, studies targeting candidate genes involved in the metabolism or transport of the drugs. The most widely studied mechanism of multi-drug resistance involves the overexpression of drug-efflux proteins. The multi-drug resistance (MDR) utilizing the adenosine triphosphate (ATP)-binding cassette transmembrane protein superfamily was one of the first phenomena associated with resistance to both doxorubicin and paclitaxel. The multiple drug resistance 1/P-glycoprotein

(MDR1/Pgp) (ABCB1), the multidrug resistance-associated protein (MRP) family members ABCC1, ABCC2, ABCC3, ABCC4, ABCC5, ABCC6, and the breast cancer resistance protein (BCRP), ABCG2, have been correlated with drug resistance (53). This has been extensively studied and led to the development of MDR inhibitors (54, 55). A meta-analysis estimated that 41.2% of all breast tumors express MDR1 Pgp protein(56). However, lack of clear-cut evidence in P-gp reversal studies to completely restore drug resistance by inhibition of P-gp demonstrated the need to look into different factors behind the multifactorial nature of the resistance (55). The advent of the genome-wide approach brought up the opportunity to undertake more comprehensive assessment of the genetic variations unbiased by prior knowledge in the pharmacokinetics and pharmacodynamics. This substantial change formed a large collection of new variants that would serve as tools for better treatment rationalization based on the observed genomic alterations. To identify the genomic targets, strong efforts have been put forth to study both cancer cell lines and patients‟ tumor samples. It has become a modern trend not only to refine the role of previously discovered genes in drug resistance, but also to provide insights into the novel genetic targets associated with drug resistance across different tumor types. In addition, over the last few years other approaches to genome-wide association studies have been developed. The most widely used are chromosomal comparative genomic hybridization (cCGH), single nucleotide polymorphisms (SNPs) array methods, and cDNA microarray studies.

It remains still unknown whether the observed genomic aberrations associated with resistance to chemotherapy are driven by selection of resistant populations of tumor cells (intrinsic) or by treatment-specific induction of mutations (acquired). In this review, we will focus on the genetic targets revealed by the genome-wide approach and their potential application as clinical biomarkers. For the purpose of this review, we only briefly cover genetic variants previously discovered by the candidate-gene approaches as this has already been extensively reviewed in the literature.

2.2 Doxorubicin:

2.2.1 Biology of the potentially modulated target pathways

Doxorubicin (DOX) (Adriamycin®; Bedford Laboratories, Bedford, OH) is one of the most widely used chemotherapy agents in breast cancer. DOX is an anthracycline antibiotic that has been used since its approval by National Cancer Institute (NCI) in 1974 as one of the essential anticancer drugs in the treatment of breast cancer, as well as soft tissue sarcomas, childhood solid tumours, and lymphomas (57-59). In a meta-analysis of adjuvant chemotherapy, the anthracycline-containing regimens were more effective in preventing recurrence and increasing survival in breast cancer patients (hazard ratio, 0.89; p=.001; risk of breast cancer death ratio, 0.84; p<.00001, two-tailed test) than CMF based therapies (11, 60). According to the National Comprehensive Cancer Network (NCCN), DOX is indicated as the preferred single agent as well as within the combination regimens for recurrent or metastatic breast cancer (61).

Anthracyclines exert their cytotoxic action by multiple mechanisms including inducing DNA damage through intercalation and binding to DNA, preventing DNA repair by inhibition of topoisomerase II and inducing apoptosis. The metabolism of anthracyclines also generates reactive oxygen species (ROS) that damage mitochondria ultimately leading to cell death (62).

DOX interferes with the cellular processes in mitochondria, disrupts cellular respiration, causes free radical formation, and activates the caspase cascades thereby leading to the induction of apoptosis in cells (63). DOX cellular uptake is mediated by family of solute carrier family proteins, namely SLC22A4 and SLC22A16 (64, 65). Even though they are still a matter of controversy, several mechanisms of action for DOX have been implicated. DOX causes growth arrest of the cells in G1 and G2 phases of the cell cycle, and eventually leads to apoptosis (63). The latter process is achieved by several mechanisms: DNA binding and alkylation, DNA cross- linking, interference with DNA unwinding, strand separation, and helicase activity (e.g. inhibition of topoisomerase II). Most of the anticancer agents either directly induce DNA damage or indirectly induce secondary stress-responsive signaling pathways to trigger apoptosis by activation of the intrinsic apoptotic pathway; some can simultaneously activate an extrinsic apoptotic pathway. DOX also interferes with a series of biological intracellular processes also

17 resulting in DNA damage. Furthermore, free radicals initiated by DOX generate DNA base damage that serves as an alternative mechanism of action of DOX (66). Overall, DOX interferes with cellular processes by activating both intrinsic and extrinsic apoptotic pathways. Interestingly, studies by Buchholz et al. showed that tumor response and clinical response to DOX treatment are associated with the degrees of apoptosis induction, which might serve as a predictive marker of long-term response (67).

The semi-synthetic derivative of DOX, epirubicin (EPI), which has been also actively used in the treatment of patients with breast cancer, in fact has similar mechanisms of action to DOX as well as similar though less severe dose-dependent adverse effects. In addition, EPI also showed a dose-dependent tumor responses with improved outcomes from intensive regimens compared to standard dose regimens in breast cancer patients (68). DOX and EPI are clinically approved by the Food and Drug Administration (FDA) drugs and are widely used in almost all stages and types of breast cancer, often as components of multiple-drug regimens in an attempt to overcome drug resistance. Currently, both DOX and EPI are widely used in the adjuvant therapy of women with breast cancer, and have been shown to be particularly effective in women with Her-2 overexpressing/amplified disease compared to non-anthracycline-based therapy (hazard ratio of relapse=0.71, p<0.001 in disease-free survival, and pooled hazard ratio of death=0.73, p<0.001 in overall survival)(69-71) and in women with the TOP2A gene alterations (31) . Due to the relatively cheap cost and well-known side effects of this old drug, DOX is still frequently used, particularly in the setting of metastatic disease. EPI however, can be given at 1.5 to double dose (in milligrams) to be as or more effective as DOX, because at the same dose it is only half as cardiotoxic. There is a window of opportunity for giving more EPI with less cardiotoxicity that has been exploited to give maximal doses of EPI particularly in the adjuvant setting (72, 73).

2.2.2 Mechanisms of Resistance:

The proposed list of mechanisms associated with resistance to doxorubicin is far from being completed. Several mechanisms of resistance have been implicated in doxorubicin resistance from in vitro studies (74-77):  decreased drug accumulation;  accelerated drug metabolism;

 lack of activation of the drug;  drug target protein alterations;  enhanced capacity of drug-induced DNA repair;  cellular suppression of apoptosis;  erbB-2/neu amplification.

Although separate mechanisms were elucidated from experimental work, the in vivo resistance phenotypes are often multifactorial. Roughly, they can be subdivided in the following groups:  The classic overexpression of P-gp multi-drug resistance (MDR) phenotype;  The overexpression of other drug transporters unrelated to P-gp MDR;  Changes in intracellular distribution of doxorubicin;  Alterations in gluthatione transferase (GST) content and drug metabolism and detoxification mechanisms;  Alterations in topoisomerase II expression;  Increased DNA repair (78).

The most important mechanisms were summarized in the following table (Table 2.1).

Table 2.1: Key mechanisms of resistance to doxorubicin

Resistance phenotype Mechanism of resistance Genes involved/targets

Classic MDR* decreased drug accumulation, increased efflux MDR1/P-gp, accompanied by a decrease in influx

Non-Pgp-Mediated MDR* decreased drug accumulation, increased efflux MRP accompanied by a decrease in influx

Changes in intracellular Vesicular drug compartmentalization concentration of doxorubicin accelerated drug metabolism;

lack of activation of the drug;

Gluthatione-S-transferase cellular suppression of apoptosis; Upregulation of GST alterations †

TOPO II or atypical MDR drug target proteins alterations; Topo II

cellular downregulation of TOPO II protein

DNA repair‡ enhanced capacity of drug-induced DNA P21WAF1/CIP1 repair: reversal of damage, excision and postreplication repair

Targeted agents erbB-2/neu amplification, VEGF inhibitors ERBB2, VEGF

*(78) †(79) ‡(80)

2.3 Paclitaxel:

2.3.1 Biology

Paclitaxel (Taxol, Bristol-Myers Squibb) was discovered in 1971 as an active constituent of the extract from the bark of the Pacific Yew Taxus brevifolia (81). It has been FDA approved for use in ovarian (1992), breast (1994), lung, head and neck, and many other types of cancers (82). It exerts its cytotoxic effect by stabilizing microtubules in the polymerized form and inducing apoptosis (83, 84). Typically, microtubules are formed by polymerization of heterodimers consisting of alpha and beta-tubulin subunits, and inhibit cell replication by disrupting normal mitotic spindle formation (85). This results in the induction of TP53 gene and modification of several protein kinases, consequently, leading to cell cycle arrest in the G2/M phase of the cell cycle and, eventually, induction of apoptosis or necrosis. Paclitaxel also induces the expression of Tumor Necrosis Factor (TNFa). Although it inhibits the disassembly of microtubules at subnanomolar concentrations, paclitaxel promotes increased formation of dysfunctional microtubules at higher clinical concentrations (86). Biologically active plasma concentrations of 0.05 to 0.1μmol per liter are usually associated with severe drug toxicity. Paclitaxel is eliminated from the body through metabolism by hepatic cytochrome P450 enzymes (CYP2C8, CYP3A4) and by biliary excretion (87). Wide variability in paclitaxel elimination by cancer patients (4- to 10-fold) contributes to the treatment outcomes (88).

Gehrmann et al. identified that the CYP1B1*3 polymorphism (Val432Leu) appears to affect progression-free survival in patients treated with doxorubicin and paclitaxel (87). A genotype with two leucine alleles confers a longer survival and has been associated with paclitaxel sensitivity in primary tumor cells. Scientists also speculate about the possible influence of this genotype on the sensitivity to doxorubicin, however, differences in this polymorphism amongst patients has not been statistically significant (87).

2.3.2 Mechanisms of resistance

The clinical utility of paclitaxel is also limited by drug resistance. Several mechanisms contributing to paclitaxel resistance has been suggested (89, 90):

 decreased intracellular drug accumulation (multi-drug resistance);  mutations in the target protein (tubulin);  alterations in the microtubule-associated proteins (MAPs), tau, stathmin and MAP4;  increased resistance to apoptosis;  erbB-2/neu amplification.

The most common genes associated with paclitaxel resistance include ABC transporters, genes involved in drug metabolism and genes controlling cell cycle progression. The differential levels of expression of multi-drug resistance genes (MDR1, MRP) as well as the role of polymorphisms in cytochrome P450 system genes and their effects on protein function have been extensively studied and reviewed previously (91). P-glycoprotein (P-gp)-mediated resistance is the most dominant notion in taxol resistance. This concept has been extensively covered in previous reviews (92, 93). P-gp is a protein product of the MDR1 gene that demonstrates over-expression in tumors after and during chemotherapy, although not uniformly. This supports the notion of selectivity and the acquired nature of resistance. However, a clear connection between elevated ABCB1 function and paclitaxel resistance has not been demonstrated in vivo as a variable response was found in attempts to restore paclitaxel sensitivity in cells by applying P-gp inhibitors. Also, in around 30-40% of cases paclitaxel is effective in patients with anthracycline resistant cancer. Given that P-gp confers resistance to a wide variety of chemotherapeutic drugs including both paclitaxel and doxorubicin, but anthracycline-resistant cases can still be efficiently treated with paclitaxel, this suggests that P-glycoprotein may not be as clinically relevant in taxane resistance as initially thought (85). Overexpression of ABCC2 (MRP2) and ABCC10 (MRP7) have been shown to confer resistance to paclitaxel in mouse models (94-96).

There are a group of studies focused on the correlation of p450 enzymes with paclitaxel resistance. These enzymes play a crucial role in the metabolism of paclitaxel in the liver. Gehrmann et al. observed correlation of CYP1B1*3 polymorphism with taxane resistance. A leu/leu variant was associated with increased sensitivity to paclitaxel in breast cancer patients (87). In another study of paclitaxel sensitivity in 10 ovarian cancer cell lines CYP2C8 and CYP3A4 were identified as putative predictive markers for efficacy of paclitaxel therapy. This study by Gehrmann et al. aimed to validate already known genetic variations using qRT-PCR. To

22 further explore drug sensitivity mechanisms Komatsu et al. used Codelink Uniset Human 20K I Bioarray with almost 2,000 probes and the same 10 ovarian cancer cell lines with or without paclitaxel treatment. This microarray study and subsequent qRT-PCR analysis shortlisted 3 genes as novel predictive biomarkers of paclitaxel response: TNFSF13B, IFIT3 and BTN3A2 (97). In addition, two different mechanisms contributing to the development of -tubulin–related resistance to paclitaxel have been reported: altered -tubulin isotype expression and mutations in class I -tubulin gene (85).

Besides the abovedescribed mechanisms of resistance derived from the candidate gene approach to paclitaxel sensitivity, there are many more associated mechanisms (Table 2.2).

Table 2.2: Key mechanisms of resistance to paclitaxel

Resistance type Mechanism of resistance Associated Genes

Multi-drug Decreased intracellular drug MDR1, MRP1, MXR resistance* accumulation

Mutations of the Tubulin –related modifications: cellular target  altered expression of -tubulin  b-tubulin classes I, II, III, IVa, protein* isoforms IVb, V and VI  post-translational modification of  Polyglutamylation, tubulin polyglycylation, phosphorylation,  point mutations in class I -tubulin acetylation tyrosination, and gene and -tubulin removal of the penultimate  alterations in microtubule-associated glutamate on subunits proteins (MAPs):  MAP4, stathmin, tau  competes with paclitaxel for microtubule binding erbB-2/neu Enhance sensitivity to paclitaxel; might ERBB2, co-amplification of TOPO amplification† be explained by co-amplification of other IIA, under-expression of tau genes

Increased resistance Disrupt phosphorylation Bcl-2, bax, caveolin to apoptosis†

* Galletti et al, 2007 (98); † Pusztai, 2007 (92)

2.4 Genome-Wide Association Studies of Resistance to Doxorubicin and Paclitaxel

Thus far, there are 9 GWA scans that have been carried out related to doxorubicin response in cancer with 5 being breast cancer studies (62, 99-101) and 4 GWA studies have been carried out on paclitaxel response (101-104). The GWA strategy, approach, and methods varied among these studies.

2.4.1 Doxorubicin:

Recent genome-wide studies have shed further light on possible mechanisms of drug resistance. There are only a few studies that have investigated the resistance to doxorubicin using a genome- wide approach.

The first published study using GWA to predict response to doxorubicin treatment was conducted in 2006 by Pierga J-Y et al. using comparative genome hybridization (cCGH) arrays (99). Microarray-based cCGH analysis in 44 pre-treatment samples from breast cancer patients suggested that there were no changes in DNA copy number after doxorubicin treatment; nevertheless, they identified amplification of the 11p15.2-11p15.5 chromosomal region after both neoadjuvant chemotherapy [doxorubicin and cyclophosphamide (AC)] and surgical intervention. Moreover, the comparison of pre-chemotherapy samples to post-surgical specimens further revealed that amplification in that region might be correlated with better prognosis in doxorubicin-based treatment regimens (50% vs. 18%, respectively), although patient outcomes were similar. In addition, there is a possible deletion in chromosomal region 13q31.1-13q34 associated with response to chemotherapy (Table 2.3). Unfortunately, association of this deletion with response was not confirmed after correction for multiple comparisons which is critical to reduce the false-positive results of GWA studies.

In another study based on CGH analysis of 23 cell lines with acquired drug resistance, several regions of gains and losses were identified (100). As reported by Yasui et al., only one cell line was doxorubicin-resistant (HT-29). The reported gains were in 6q16-22, 11p, 13q21-22, 14q21- 32, 18q12-21, and losses were in 7p21-22, 8q23-24, and 9q34. Fluorescence in situ hybridization

(FISH) within the identified regions with copy number variations demonstrated amplification of ABCD4 on chromosome 14q24.3 (3 DNA copies in resistance cells vs. 2 copies in parent cell lines) (100). Yasui et al. also demonstrated that alteration of gene-copy number in a few genes is capable of activating or down-regulating genes associated with drug resistance. An assessment of the expression of BCL2 and its variants in HT-29/ADM cells by qRT-PCR showed only slight overexpression of mRNA transcripts for BCL2-XL, and BCL2L2 (just under a 2-fold increase) and even lesser up-regulation in BCL2 and MCL1 genes. However, these findings did not reach significance as indicated by the authors (100).

The genome-wide screening of Saccharomyces cerevisae deletion strains conferring sensitivity to doxorubicin identified 71 genes associated with DNA repair, RNA metabolism, chromatin remodeling, amino acid metabolism, and heat shock response pathways. The major groups of genes were: DNA repair (RAD50, RAD51, RAD52, MRE11), chromatin remodeling (SNF/SWI complex, ACTR8), ribosomal protein (RPL37A), heat shock protein (Hsp70B), and superoxide dismutase (SOD1) (please refer to Table 2.3 for the full list) (62).

In a comparative study of doxorubicin-resistant MCF-7 cell lines to wild-type cells by cDNA microarrays analysis (102), Villeneuve et al. identified 25 genes with reproducibly higher expression in resistant cells and 4 genes that have lower expression. The validation by qRT-PCR and immunoblotting confirmed 9 of the 10 changes in gene expression and discovered five additional genes with altered expression. Authors also emphasized that the magnitude of change detected by cDNA microarray analysis was underestimated. 23 genes with altered expression were deemed to be MCF-7DOX specific (compared to MCF-7TAX cells) (Table 2.3).

Recently, another study was carried out on 115 breast carcinomas from women treated with anthracyclines (105). Predictive analysis of microarrays identified 114 probes encoding 75 known genes whose copy numbers were significantly different between responders and non- responders to anthracyclines. The only significantly associated genomic region with varying copy number associated with anthracycline response was chromosome 8q22 to which 12 different genes could be mapped (p < 2.1x10-9). In addition to the observed amplification of the region in SNP array analysis, it was observed that expression of these genes correlated with DNA copy number. Obtained results were confirmed by DNA interphase fluorescence in situ

26 hybridization (FISH). siRNA-based knock-down of expression of the12 candidate genes in the same chromosomal region in the BT459 breast cancer cell line demonstrated that blocked expression of two genes, YWHAZ and LAPTM4B, significantly increased sensitivity to anthracyclines. The further examination of 16 breast cancer cell lines confirmed a strong correlation between higher mRNA levels of LAPTM4B and resistance to anthracyclines (p< 0.00034). Higher expression levels of YWHAZ and LAPTM4B were also associated with poor outcome after adjuvant chemotherapy in breast cancer patients and shorter disease-free survival. In addition, above median expression of both genes was associated with the absence of a pathological complete response (pCR) and the presence of residual disease after a neoadjuvant (preoperative) epirubicin monotherapy (Table 2.3).

Table 2.3: SNPs/genetic variations associated with doxorubicin resistance: SNP/gene Chromosome Ratio in Referen Approach Source expression ce levels AP15 11p12-q12 1.66 (102) cDNA MCF-7 cell lines HLA-DRB4 6p21.3 2.25 microarray ABCB1 7q21.1 19.8 ZFP36L2 2p22.3-p21 1.533 TOMM20 1q42 1.58 MLL 11q23 1.52 MAP3K8 10p11.23 2.08 HLA-DQB1 6p21.3 1.52 MDH2 7cenq22 1.50 FDFT1 8p23.1 1.20 STX3A N/A 1.37 CRYAB 11q22.3-q23.1 1.73 CRYAA 21q22.3 1.46 LGALS3 14q21-q22 1.27 CLU 8q21-p22 1.63 SAT Xp22.1 1.53 POR 7q11.2 1.86 BACH1 21q22.11 2.66 MTR 1q43 1.66 GFI1B 9q34.13 2.18 EFNB3 17p13.1-p11.2 1.54 HIST1H2BC 6p21.3 1.84 MGC27165 14 1.69 NR4A1 12q13 1.64 MDK 11p11.2 1.56 PCNA 20pter-p12 0.623

TUBA1 2q36.1 0.667 TUBB4 16q24.3 0.654 FVT1 18q21.3 0.652 4q13.1 Gain in 9/21, (99) aCGN Frozen breast 43% profile core biopsies (99)(99) H-RAS, IGF2 11p15.2-p15.5 Gain in from 21 pts (99)(98) 14/21, 66% (97)(96) 12q13.3 Gain in (96)(95) 14/21, 66% (95)(95) 18p11.21 Gain in (95)(95) 12/21, 52% (95)(95) 19q13.2 Gain in 12/21, 52% SLITRK6, SLITRK5, 13q31.1- Loss in 24/44 (99) aCGN Frozen breast GRC5, GRC6, DCT, 13q32.2 profile core biopsies TGDS, SOX21, from 44 pts ABCC4, CLDN10, DZIP1, DNAJC3, UGCGL2, HS6ST3, HSP90AB6P, OXGR1, MBNL2, RAP2A 6q16-22 Gain (100) CGH HT-29 cell line 11p Gain

13q21-22 Gain 14q21-32 Gain 18q12-21 Gain 7p21-22 Loss 8q23-24 Loss 9q34 Loss

ABCD4 14q24.3 Amplified in FISH 3/5 YWHAZ 8q22 Amplified (105) cDNA BT549, MDA- LAPTM4B 8q22 Amplified microarray, MB-231, HCC38,

FISH, SNP CAMA-1 cell line array and 115 breast cancer patients RAD50 5q31.1 100-1,000 (62) Yeast extract fold from sensitivity Saccharomyces RAD51 15q15.1 cerevisiae RAD52 12p13-p12.2 RAD51L1 14q23-q24.2 MRE11 11q21 NBS1 8q21.3 (NBN) SNF/SWI complex 1p35.3 (ARID1A) TADA1L 1q24.1 ATP6V0C 16p13.3 DNAJA2 16q12.1 ZRF1 7q22.1 PGD 1p36.3-p36.13 RACK-1 (GNB2L1) 5q35.3 JMJD2C 9p24.1 10-100 fold (KDM4C) sensitivity AFG3L2 18p11 ACTR8 3p21.1 SC5DL 11q23.3 RPL37A 2q35

VPS36 13q14.3 HSP70B 1q23 (HSPA6) SOD1 21q22.1 SMARCD1 12q13-q14 2-10 fold sensitivity H2AFZ 4q24 SH3PXD2B 5q35.1 COX5A 15q24.1 PSAT1 9q21.2 TOP3 17p12-p11.2 SH3RF1 4q32.3-q33 AK2 1p34-p35.1 VPS28 8q24.3 SNF8 17q21.32 USF2 19q13 DDX17 22q13.1 RPL13 16q24.3/ 17p11.2 RPS9 19q13.4 RPLP1 15q22 ASF1 6q22.31 SLC25A32 8q22.3 PPM1A 14q23.1

CCS 11q13

2.4.2 Paclitaxel:

A paucity of genome-wide studies of paclitaxel resistance in breast cancer poses the main challenge in elucidating resistance mechanisms to the agent. Genome-wide screening of the 31 breast cancer cell lines in vitro for copy number variations (CNV) on Affymetrix arrays identified that basal-like cell lines are more sensitive to paclitaxel than luminal or HER2- amplified cells (101). A group of SNPs on chromosome 17 (17q21.21-17q21.23) were significantly associated with resistance to paclitaxel. Further analysis of the 17q21 region demonstrated amplification of the ABCC3 drug efflux pump in a subset of HER2-amplified and luminal tumors (101).

Blancafort et al. proposed a new and interesting strategy to study drug resistance based on the artificial transcription factors (ATFs) randomly assembled from three or six zinc finger domains, and small protein structural motifs (104). A genome-wide screening strategy with ATFs in HeLa cells was focused on the targets of regulators of the p53 tumor suppressor involved in resistance to paclitaxel. This study showed an up-regulation of p21WAF/CIP1, wig-1, and p14ARF in cells transduced with ATFs. Although some of the ATFs that induced drug resistance were associated with an up-regulation of the p14ARF-MDM2-p53 pathway, the response was not uniform (104). However, the most efficient ATF 3Zf-1-VP was shown to mediate both p53-dependent and - independent genes, i.e. -tubulin, P21WAF/CIP1, p14ARF and ephrin-A4. Given that P21WAF/CIP1 overexpression was previously correlated with ErbB-2 up-regulation and with resistance to taxanes, this supports the utility of application of ATFs for assessment of drug response (106, 107). Furthermore, the same group suggested application of ATF libraries in combination with siRNA technology to map drug resistance-related pathways. Although this group was claiming genome-wide screening by ATF libraries, their approach seems to be more oriented toward p53- associated pathways.

Another study of taxol-resistant MCF-7 cell lines by Hembruff et al. (103) using cDNA microarray analysis found 20 up-regulated and 12 down-regulated genes with at least 1.5-fold difference in expression compared to isogenic wild-type MCF-7 cells (Table 2.4). The

comparison of genetic changes in MCF-7TAX and MCF-7DOX cell lines uncovered 10 genes that were shared by both types of resistant cells (102). An additional verification by qPCR discovered three additional genes with reduced expression. Villeneuve et al. explains observed additional changes due to the increased sensitivity of qPCR or immunoblotting experiments compared with microarray experiments. Also, the comparison of cDNA microarray and qPCR methods led to the conclusion that observed changes in gene expression are most likely due to prolonged exposure to paclitaxel, i.e. acquired or selected variations.

Table 2.4: SNPs/polymorphisms associated with paclitaxel resistance: SNP/gene Chromosome Ratio in Reference Approach Source expression levels FURIN 15q26.1 2.276 (102) cDNA MCF-7 cell RDC1 2q37.3 6.172 microarray lines BCL2 18q21.3 2.065 KRT13 17q21-q21.2 1.503 FTH1 11q21-q13 1.401 RPL23A 17q11 1.541 RMP22 17p11.2 4.656 ITGB5 3q21.2 1.582 CD47 3q13.1q13.2 1.511 GADD45A 1p31.2-p31.1 1.50 CTCF 16q21-q22.3 1.659 ARR3 Xcen-q21 1.563

IF130 19p13.1 2.019 ABCB1 7q21.1 5.769 RFX5 1q21 1.502 FTL 19q13.3-q13.4 1.47 RPLP1 15q22 1.373 GSTM5 1p13.3 1.511 CXCR4 2q21 1.500 CXCR6 3p21 1.407 AKRIC2 10p15-p14 0.405 MT2A 16q13 0.603 S100P 16q13 0.603 BMP7 20q13 0.451 FDPS 1q21.3 0.710 TFF1 21q22.3 0.520

FXYD3 19q13.11 0.467 MCP 1q32 0.439 CAV1 7q31.1 0.484 MT1I 16q13 0.688 TOP2A 17q21-q22 0.708 CTSL 9q21-q22 0.624 IFI30 19p13.1 2.84 TUBB4 19p13.3 0.45 SAT Xp22.1 1.17 LGALS3 14q21-q22 0.49 MDH2 7q11.23 0.87 FDFT1 8p23.1-p22 0.36 STX3A 11q12.1 0.72x CRYAB 11q22.3-q23.1 0.87 p21WAF/CIP1 6p21.2 (104) ATFs HeLa cells (CDKN1A) wig-1 3q26.3-q27 (ZMAT3) p14ARF 9p21 (CDKN2A) 24 genes 17q21.31-33 4 copies (101) 100K SNP Breast cancer amplification, array and cell lines: 8/14 cell lines aCGH HCC-1419, analysis MDA-MB-453 MDA-MB-361, HCC-2218, UACC-812, ZR-75-30, BT474, EFM- 192A

ABCC3 17q21.31-33 >3.5 copies in (101) aCGH and 145 breast (208161_s_ 25% (out of 145) FISH tumors at) of HER2- analysis amplified, 11% of luminal; >2.2 amplified, 64 HER2- 16/64 (25%); 3-7 positive breast copies in 11% tumors

2.5 Influence of the genome-wide approach on the genetic targets

The potential input from genome-wide studies to the study of drug resistance is tremendous. Nevertheless, candidate-gene approach still has a great influence on the interpretation and further characterization of the genome-wide outcomes. For instance, in study of approximately 100 expressed transcripts corresponding to 24 genes associated with response to the drug, authors preferred to focus on drug transporter genes, which were the favorite targets of the era before GWA studies. Unfortunately, they offered no comments about other potential candidate loci from the study (101).

Pharmacogenetic studies with SNP and CNV chips have revealed a number of possible copy number variations or SNPs implicated in doxorubicin and paclitaxel resistance. Nevertheless, it seems that those results are still in the preliminary stage and further validation studies involving the signature genetic variations should be performed at the proteomic and metabolomic level to ensure that they can be used as clinical biomarkers. The other disadvantage of the performed studies is that although they share a common aim, the methods used are different. As such, only a minor portion of the genome-wide studies produce overlapping results. In addition, these studies have not yet been reproduced, and thereby, confirmed.

Genome-wide studies represent a good approach to open new venues for discovery and investigation of the drug resistance-associated loci. The addition of candidate gene-based

36 approaches for further validation of the targets discovered by GWS is the conclusive experimental approach.

2.6 Summary

Gene expression microarrays, CGH studies, and other high-throughput technologies have raised a new view on the genetic signatures associated with drug resistance. Genome-wide approaches changed the preconceived notion of putative predictive signatures by helping to remove biases based on prior discoveries related to drug resitance. For example, they helped demonstrate that not only multi-drug transporters and cytochrome P450 system proteins are associated with drug resistance, but there are also many other signaling pathways altering the intracellular level of proteins associated with resistance to paclitaxel and doxorubicin. Given the heterogeneity of breast tumors, attributing resistance to therapy to a single gene would be unlikely be correct.

Even though recent approaches led to the burst of discoveries of many genomic loci associated with drug resistance, there are likely many more genes yet to be discovered. However, due to the limited availability of reagents for poorly characterized proteins and mRNA transcripts as well as the lack of means to study intergenic SNPs, the pragmatic value of the discovered loci remains uncertain.

Chapter 3 Genome-wide Association Studies of Doxorubicin Response in Cancer

Abstract

Doxorubicin is a widely applicable chemotherapeutic agent in cancer treatment. We previously developed a new approach to the genome-wide detection of the genetic variations associated with drug resistance. In this study, we applied the same model to perform a genome-wide association (GWA) study of single nucleotide polymorphisms (SNPs) previously obtained from Developmental Therapeutics Program (DTP) using Affymetrix 125K SNP Chips and the NCI60 panel of cell lines. This approach identified three intragenic SNPs associated with doxorubicin GI50 values. To confirm the obtained association results, we compared mRNA expression in the drug-resistant tumor cell lines and breast tumors. RORA, FRMD6 and DSG1 genes were associated with variations in phenotype in doxorubicin-resistant A2780, MES-SA, and MCF-7 cells compared to isogenic wild-type cells. Significantly variable expression in RORA and FRMD6 was observed in human breast carcinoma tissues when compared to normal mammary glands. These findings suggest that application of the in silico GWA approach using the NCI60 cell line panel data followed by functional validation can provide insights into genetic variations responsible for resistance to chemotherapeutic agents.

3 GWAS of doxorubicin response in cancer

3.1 Introduction

Despite advanced diagnostic technologies and modern treatment protocols for early stage breast cancer, about 30% of women diagnosed with breast cancer still die from the disease. Traditional adjuvant clinical trials have been focused on the chemotherapy regimens intended for the general patient population (108). Variation in patients‟ response to chemotherapy as a consequence of environmental factors and genetic make-up is a critical challenge in modern oncology. Emerging evidence suggests that an individualized therapeutic approach based on the genetic background of the host and of tumors will improve the efficacy and diminish the adverse effects of the therapy (109).

Inter-individual variability is predetermined, in part, by genetic variations within the human genome, mostly attributed to single nucleotide polymorphisms (SNPs). SNPs can impact on the expression, intrinsic properties, and function of gene products, thus resulting in various phenotypes that could be associated with drug resistance. Hence, identification of SNPs that have influence on the outcomes of therapy may help to establish the most appropriate choices and doses for chemotherapeutic agents used to treat a specific cancer patient (110). Today, advancement in high-throughput genotyping technologies enables a thorough investigation of genome-wide genetic patterns and the discovery of candidate loci associated with efficacy of chemotherapeutic agents (111).

Doxorubicin (DOX) (Adriamycin®; Bedford Laboratories, Bedford, OH), an anthracycline antibiotic, has been one of the most effective anticancer drugs used to treat a variety of cancers since its NCI approval in 1974 (57-59). In a meta-analysis, anthracycline- containing regimens have been shown to be more effective in preventing recurrence and increasing survival in breast cancer patients than other regimens (11, 112). According to the National Comprehensive Cancer Network (NCCN), DOX is indicated as the preferred single agent as well as part of combination regimens for recurrent or metastatic breast cancer, and in multi-drug combinations for treatment of invasive breast cancer using adjuvant chemotherapy (61).

Although it is still a matter of controversy, several mechanisms of action have been implicated for DOX. It stimulates both intrinsic and extrinsic apoptosis, generates reactive oxygen intermediates, inhibits topoisomerase II, intercalates with DNA leading to DNA damage, and activates several signal transduction pathways causing growth arrest of cells in the G1 and G2 phases of the cell cycle (the only cytostatic action of DOX) through elevated levels of p53 (63). Free radicals produced during DOX metabolism by oxidoreductases such as cytochrome P450, NADH dehydrogenase, and xanthine oxidase also generate DNA base damage that serves as an alternative mechanism of action of DOX (66). Interestingly, studies by Buchholz et al. showed that both tumor and clinical response to DOX treatment are associated with a degree of apoptosis induction, which might serve as a functional marker of DOX response (67).

The NCI60 cell line panel, representing tumor cell lines from nine different tissue types, has been a valuable resource to test the activity of thousands of compounds for their antineoplastic capacity (44, 113). Besides drug sensitivity data, the NCI60 panel of cell lines has been also genotyped for over 124,000 SNPs using the Affymetrix 125K SNP arrays (44, 113, 114). Both the drug response and genetic data on this panel are publicly and freely accessible at the DTP website. In a recent study, using drug response and genetic data from the NCI60 panel of cell lines and from patients tumors, we have developed a methodology for the discovery of genes and genetic variants associated with response to various drugs (50). Herein, we applied this methodology to conduct GWA studies of genetic variations associated with the altered response of tumor cells to DOX.

3.2 Materials and Methods

3.2.1 GWA analysis:

3.2.1.1 Case-Control Design

In this study, we divided the NCI60 population into sensitive (case) and resistant (control) groups using kernel density estimator, a non-parametric distribution estimator of DOX-response (GI50) data. Kernel density estimator is a method that assesses data points with no fixed structure, i.e. does not depend on the end points of the bins. It smoothens the building blocks and provides better picture of clusters, also known as modes of distribution, observed in the

40 population. The density of the data determines the optimal bandwidth, an estimate of an approximation of the true value to maintain smoothness of the data. Details of the methodology are described in Jarjanazi et al.(50) (see Appendix). GI50 values (the concentrations of the drug required to inhibit growth by 50%) for 56 NCI60 cell lines treated with DOX were obtained from the publicly available Developmental Therapeutic Program (DTP) website -4 (http://dtp.nci.nih.gov). The log10 of the GI50 value for DOX for each cell line treated with a 10 M dose was retrieved and normalized to obtain a mean of zero and a standard deviation of one. The visual antimode was used as a cut-off value for defining sensitive (cases, n=24) and resistant (controls, n=32) cells. We have aligned the available genotype data with the corresponding cell lines in both sensitive and resistant categories, and prepared the data for GWA analysis using PLINK (http://pngu.mgh.harvard.edu/purcell/plink/) (115).

3.2.1.2 Single-Marker Association Analysis

Out of 124,000 SNPs available on Affymetrix arrays, 79,622 were selected for the case-control study after removing SNPs with a minor allele frequency (MAF) of <0.02 and those where genotyping data was not complete in 75% of the cell lines studied. Using PLINK, we performed the GWA analysis of single SNP markers where we have compared the MAF between sensitive and resistant genotypes by a chi-square test. Results were adjusted for correction using the False Discovery Rate by Benjamini and Hochberg (FDR-BH), which has been shown to be more appropriate in tightly linked SNP studies (116-118).

3.2.1.3 Haplotype Analysis

Haplotype analysis for the candidate regions was done using Haploview (http://www.broad.mit.edu/mpg/ haploview/index.php) (119). SNPs were tested for deviation from Hardy-Weinberg Equilibrium (HWE) (117). In each genomic region, SNPs were allocated to blocks based on a linkage disequilibrium (LD) of r2>0.8. Haplotype analyses between blocks were performed for all SNPs in these regions.

3.2.2 mRNA Expression Analyses

3.2.2.1 Genome-wide expression in cancer cell lines and breast tumor samples

We utilized the open source CaARRAY microarray data management system (provided by GlaxoSmithKline) to investigate the mRNA expression levels of DSG1, RORA and FRMD6 (genes that from the above analysis had SNPs that correlated with altered response to doxorubicin). This database contains genomic profiling data for over 300 cancer cell lines generated using Affymetrix GeneChip U133 Plus 2.0 arrays (120). A detailed description of genome-wide microarray studies in those 300 cell lines can be obtained from the website (https://cabig.nci.nih.gov/caArray_GSKdata/). The MAS5 algorithm and constitutively expressed human genes were used as references in every cell line (GAPDH, STAT1, ACTB, TBXA2R, LOC100008588). In addition, we have used the Gene Expression Atlas 1.1.3 Build r8960 (GEA), an enriched database of meta-analysis based on the condition–specific gene expression patterns, to perform a broader review of the putative candidate genes (http://www.ebi.ac.uk/gxa/). Using publicly available data from the ArrayExpress, we compared the expression of FRMD6, DSG1, and RORA between normal and cancer tissues (121). For these purposes, we analyzed microarray gene expression data from breast tissues assessed by a pathologist in 61 patients (41 with breast carcinoma, and 20 samples with normal breast tissue taken from healthy individuals undergoing mammoplasty or from surrounding healthy breast tissue of cancer patients) (122, 123). The statistical analysis was performed using SPSS 13.0 software, employing a 2 independent samples t-test. All results were tested for the equality of variance (Levin‟s test). A p value < 0.05 was accepted as significant.

3.2.2.2 mRNA expression Analyses in Doxorubicin-resistant Cancer Cell Lines

Three genes (DSG1, FRMD6, and RORA) identified as having SNP variations correlated with resistance to DOX in NCI60 cell lines were assessed for expression by quantitative real-time PCR (qRT-PCR) in a variety of paired isogenic doxorubicin-sensitive and doxorubicin-resistant cell lines as described previously by Hembruff et al.(124). They included MCF-7 cell lines

42 selected for survival in increasing concentrations of doxorubicin or its sister compound epirubicin (EPI) until the maximally tolerated dose was reached (MCF-7DOX-2 and MCF-7EPI cells, respectively). Expression was compared to cells “selected” in the absence of drug (MCF-

7CC cells). Other cell line pairs used in the assessment included wild type (MES-SA) and doxorubicin-resistant (MES-SADx5) uterine sarcoma cells, and wild type (A2780) and doxorubicin-resistant (A2780DOX) human ovarian carcinoma cell lines. The above cell lines were provided by Dr. A. Parissenti of the Sudbury Regional Hospital and Laurentian University. All cell lines were cultured in specific cell culture media supplemented with 10% fetal bovine serum (McCoy‟s for MES-SA cells and RPMI 1640 for all other cell lines). Cells were incubated at 37C in a humidified 5% CO2-95% air atmosphere and passaged using trypsin- EDTA. To retain the drug-resistant characteristics of the cell lines, cells were cultured in their respective drugs at their maximally tolerated selection doses. The medium was renewed every three days. The target gene was expressed relative to a reference gene (S28) amplified with it. All studies were done in triplicate.

3.2.2.3 RNA and cDNA preparation, and qRT-PCR

RNA was isolated from the cell lines described above using an RNAeasyTM mini kit (QIAGEN, Hilden, Germany) according to the manufacturer‟s protocol, with on-column DNAse treatment to ensure absence of genomic DNA in the samples. RNA was isolated from three separate flasks for each cell line. The concentration and quality of each RNA preparation was determined using an Agilent 2100 Bioanalyzer (Agilent Biotechnologies, Palo Alto, CA) following the manufacturer‟s instructions (Figure 3-1). A two-step reverse transcriptase reaction was used to convert RNA to cDNA, as described in Villeneuve DJ. et al. (102). qRT-PCR analyses were performed on an ABI PRISM 7500 Sequence Detection System (Applied Biosystems, Foster City, CA, USA) using a standard protocol (124). Amplification products were detected with the DNA-binding dye SYBR green in a final concentration of 200 μM. Each cycle was initiated with denaturation at 95°C for 10 minutes, followed by 40 cycles at 95°C for 10 seconds, at a primer- specific annealing temperature for 15 seconds, and at 72°C for 30 seconds. The purity of the amplification product was then assessed using a standard DNA melt curve. S28 was used as the reference gene in all experiments. A standard curve was generated for both the gene of interest and the reference gene (Figure 3-2, Figure 3-3).

A B

Figure 3-1: RNA biointegrity results (confirmed on the Agilent 2100 Bioanalyzer). All cells were analyzed in triplicates and has been labeled correspondingly as T1, T2, and T3. Panel

(A) shows RNA extracted from MES-SA, MES-SADx5 and MCF-7cc cells. Panel (B) shows

RNA extracted from A2780, A2780ADR, and MCF-7Dox12. Panel (C) shows results from

MCF-7EPI, MCF-7, and MCF-7Dox cells.

A B

C D

Figure 3-2: Dissociation curve for DSG1 (A), RORA (B), FRMD6 (C), and S28 (D) in qRT- PCR (from SDS 2.2.2 software). The top curve represents dissociation curve for the gene, the bottom curve represents control blank template. Analysis was done in triplicate.

Figure 3-3: Amplification plots of DSG1, FRMD6, RORA, and S28 (from SDS 2.2.2 software)

3.2.2.4 Primers:

The primer pairs (forward and reverse) for the qRT-PCR reactions were chosen so that they would produce an amplicon limited to exons only. Thus, the primers amplified cDNA but not genomic DNA. Oligonucleotides were designed for the experimental RORA, DSG1, FRMD6 genes, and the S28 (reference gene) (Table 3.1). The design of primers was carried out using PerlPrimer 1.1.17 platform (125). The uniqueness of the primer binding sites was confirmed using BLASTN 2.2.22+ at the NCBI human genome assembly (Genome Build 36.3) (126).

Table 3.1: Oligonucleotides for qRT-PCR

Gene Forward primer Reverse primer Amplicon length (bp)

DSG1 5‟-ATG GGA TTC TTG GTC TTA GG-3‟ 5‟- AGG TGG TAT TTG TGG TAT GAC- 3‟ 188

RORA 5‟-TTT CCC TAC TGT TCG TTC AC-3‟ 5‟- GCA AAC TCC ACC ACA TAC TG-3‟ 253

FRMD6 5‟-GTT TGT TGT GTA AAG TCC GT-3‟ 5‟-GTT TAG GGT GAA GAA GAT ATG G-3‟ 117

S28 5‟-TCC ATC ATC CGC AAT GTA AAA G-3‟ 5‟- GCT TCT CGC TCT GAC TCC AAA-3‟ 156

3.2.3 Statistical Analysis:

The comparative cycle threshold (CT) method (2-ΔΔCT) was used to calculate the amount of the cDNA of interest in the drug-resistant cell line relative to its wild-type parent (127). A non- parametric Mann-Whitney test of independent samples was applied to analyze the variability between drug-sensitive and drug-resistant samples. All statistical analyses were performed on SPSS 13.0. A 2-tailed p value of <0.05 was accepted as a significant result.

3.2.4 Karyotypic Alterations

Numerical changes and structural abnormalities of chromosomes corresponding to the identified genomic loci in the NCI60 cell line panel was previously assessed by Spectral Karyotyping

(SKY), Multiplex Fluorescence In Situ Hybridization (M-FISH), or Comparative Genomic Hybridization (CGH) techniques. The data was obtained from NCI and NCBI's SKY/M-FISH and CGH Database (2001) at http://www.ncbi.nlm.nih.gov/ sky/skyweb.cgi (128). A total of six loci of interest (7q21, 12p12, 14q21, 15q21, 17p12, 18q12) were compared on CNV among 55 cell lines. Statistical analysis by t-test of 2 independent samples was performed to compare the variations in those loci between sensitive and drug-resistant cell lines using SPSS 13.0.

3.3 Results

3.3.1 Case-control design:

A total of 56 NCI60 cell lines were analyzed to allocate cell lines into several clusters, modes of distribution, with distinct response to DOX treatment. The analysis of the standardized GI50 values (performed on SAS 9.1 with a non-parametric distribution test, density kernel estimation, with an optimal bandwidth of 0.2729 and an AMISE of 0.0197) showed two modes in the distribution. Accordingly, based on a cutoff value of zero (antimode), 24 cell lines were classified as doxorubicin-sensitive and 32 as doxorubicin-resistant (Figure 3-4). The Affymetrix 125K data, which interrogates over 124,000 SNP alleles spaced with a median intermarker distance of 8.5 kilobases (kb) were also downloaded from the DTP website (129). The Affymetrix arrays contain evenly spread SNPs across the genome mapped both within and between genes, and are well-suited for genome-wide association and fine mapping studies.

Figure 3-4: Doxorubicin dose response and identification of sensitive (cases) on the left and resistant (controls) cells on the right by Kernel density estimation. Analyzed on SAS 9.1 by non- parametric Kernel density estimation test). Red line represents an antimode; blue lines- distribution o the cells; shaded area corresponds to the number of cell in each bin. Antimode=0 (the cutoff point selected as the point of separation between 2 modes of distribution); 24 cells are sensitive; 32 cells are resistant. Cells were analyzed after normalization. X axis, normalized values; y axis, percentage of the cells within each bin.

3.3.2 GWA Analysis:

Single marker association analysis of 79,622 eligible SNP markers in 32 DOX-resistant and 24 DOX-sensitive cell lines had shown that 9.4% (7,459/79,622) and 0.06% (50/79,622) of SNPs were significantly different between sensitive and resistant cell lines at p values of <0.05 and <0.00005, respectively. Two breast cancer cell lines (T-47D, MCF-7) were found in the sensitive group, whereas six other cell lines (BT-549, MDA-N, MDA-MB-435, HS-578T, 231/ATCC, NCI/ADR-RES) were among the resistant group (Table 3.2). FDR multiple correction analysis demonstrated that eleven SNPs on 7q21, 12p12, 14q21, 14q22, 15q22, 17p.12, 17p13 and 18q12 showed statistically significant associations with DOX resistance (Table 3.3). Among these, four were found to be located within the coding regions of four genes: Desmoglein 1 (DSG1) at 18q12, retinoic acid receptor-related orphan receptor  (RORA) at 15q22, FERM domain- containing protein 6 (FRMD6) at 14q22.1, and the hypothetical gene, LOC388335 (transmembrane protein 220, TMEM220), at 17p13.1. Remaining seven SNPs were located in the intergenic region.

Table 3.2: Case-control study design of doxorubicin sensitive and resistant NCI60 cell lines

Sensitive-Cases Resistant-Controls

Cell type Cell name Normalized Cell type Cell name Normalized value value

Leukemia MOLT-4 -1.6 NSCL HOP-92 -0.2

NSCL NCI-H460 -1.6 CNS SF-268 -0.2

Leukemia SR -1.5 Leukemia K-562 -0.1

Breast T-47D -1.4 Melanoma SK-MEL-5 0.0

Leukemia RPMI-8226 -1.3 CNS SF-539 0.0

CNS SNB-75 -1.1 Colon HT29 0.0

Leukemia CCRF-CEM -1.0 Ovarian OVCAR-8 0.1

CNS SNB-19 -0.9 Ovarian IGROV1 0.1

Renal RXF 393 -0.8 Melanoma SK-MEL-2 0.3

CNS U251 -0.8 Breast HS 578T 0.3

Melanoma LOX IMVI -0.8 Melanoma UACC-257 0.3

Breast MCF7 -0.7 Colon KM12 0.3

NSCL A549/ATCC -0.7 Breast MDA-N 0.4

Prostate DU-145 -0.7 Melanoma UACC-62 0.4

Renal ACHN -0.6 Melanoma SK-MEL-28 0.4

Leukemia HL-60(TB) -0.6 Colon COLO 205 0.4

Renal SN12C -0.5 Melanoma MDA-MB-435 0.4

Colon SW-620 -0.4 NSCL EKVX 0.4

Colon HCT-116 -0.4 Colon HCC-2998 0.4

NSCL HOP-62 -0.4 Ovarian SK-OV-3 0.4

Renal 786-0 -0.4 Renal UO-31 0.5

CNS SF-295 -0.4 NSCL NCI-H322M 0.6

Renal A498 -0.3 Breast BT-549 0.7

Melanoma M14 -0.3 Ovarian OVCAR-3 0.7

Prostate PC-3 0.7

Ovarian OVCAR-5 0.8

Breast MDA-MB-231 0.8

Ovarian OVCAR-4 1.1

Renal CAKI-1 1.2

Renal TK-10 1.5

Ovarian NCI/ADR-RES 1.8

Colon HCT-15 4.7

Table 3.3: Statistically significant SNPs associated with doxorubicin response in the GWA study (analyzed on PLINK with p<0.05 after correction by FDR accepted as significant). Abbreviations: CHR, chromosome; Chr location, location on the chromosome according to the Genome Build 36.3; SNP (Affy marker), Affymetrix number for SNP; RS#, rs number for SNP; MAF, Minor Allele Frequency; OR, odds ratio; CI95, lower and upper limits within 95% confidence interval; UNADJ p value, nominal p value before test for multiple correction; FDR_BH, post-hoc test by FDR. CHR Chr location SNP (Affy RS# Genes Alleles MAF CHI- OR CI 95 UNADJ FDR_BH (build 36.3) marker) SQUARE P-value 7 96807099 2608235 rs953687 A/G 0.42 23.43 0.1 0.04, 0.26 1.29E-06 0.02 12 20318086 574098 rs890759 A/G 0.33 24.08 0.09 0.03, 0.25 9.26E-07 0.02 14 51103362 845857 rs726529 FRMD6 C/T 0.23 19.27 11.8 3.31, 42.15 1.14E-05 0.03 14 47307382 843236 n/a 19.16 16.5 3.60, 75.8 1.20E-05 0.03 14 48227998 843306 rs1986410 A/G 0.21 18.51 15.17 3.37, 68.35 1.69E-05 0.04 15 59011402 957347 rs1482049 RORA A/G 0.2 19.92 0.03 0.004, 0.26 8.08E-06 0.03 17 10837696 1089881 n/a 22.23 10.2 3.53, 29.5 2.42E-06 0.03 17 10558528 1089850 rs440655 TMEM220 A/C 0.4 19.56 9.11 3.13, 26.49 9.73E-06 0.03 17 14998961 1095434 rs1122073 A/G 0.29 17.85 0.15 0.06, 0.37 2.38E-05 0.05 18 27164613 1177502 rs2199301 DSG1 A/G 0.49 20.64 7.57 3.02, 18.96 5.54E-06 0.03 18 27026908 1177280 rs1365296 C/G 0.38 19.4 16.25 3.58, 73.67 1.06E-05 0.03

3.3.3 Haplotype Analysis:

Haplotype association analyses of sensitive and resistant cell lines were performed using the SNPs in linkage disequilibrium for the three known candidate gene loci. Since two of the eleven statistically significant SNPs (rs2199301 and rs1365296) identified in our study were in the DSG-DSC loci, haplotype analyses of a 500kb region on 18q12 (covering DSG1, DSG2, DSG3, DSG4, and DSC1, DSC2, and DSC3) were performed (Figure 3-5). Based on 40 SNPs with r2>0.8, a total of three LD blocks [haplotype block1: GTCTCAG (=10.19, p=0.001) and CCTCTTA (=17.15, p=3.45E-05), and haplotype block3: GTT (=8.254, p=0.004)] were constructed from this region, where block 1 contained both candidate SNPs (Figure 3-6). Haplotype analysis of RORA included 66 SNPs from a region of 732kb on 15q22. Six LD blocks with r2>0.8 were identified, where block 4-CCAAGC (=4.89, p=0.027) was significantly associated with DOX response in this gene. The analyses of 12 SNPs covering 114 kb of FRMD6 on 14q22.1 identified statistically significant associations with drug response [block1- AAG (= 5.867, p=0.0154) and GCA (= 17.754, p=2.51E-05)] in a single block (Table 3.5, Table 3.6).

Figure 3-5: Haplotype blocks of Desmoglein-Desmocolin gene family (DSG1-4, DSC1-3).

b c

Figure 3-6: LD blocks of Desmoglein-Desmocolin gene family (DSG1-4, DSC1-3), RORA and FRMD6 genes. A) Haplotype analysis of DSG-DSC gene family at 18q12 loci: 3 LD blocks were constructed, 2 of them are significant. B) Haplotype analysis of RORA: 6 LD blocks were constructed, 1 was significant (shown). C) Haplotype analysis of FRMD6 region: 1 LD block was constructed and significant. Each square displays the amount of LD between a pair of markers. The strength of LD is given by the numeric annotation and the intensity of the color of a box: red- the strongest, white- no LD.

Table 3.4: Haplotype association analysis of DSG-DSC gene family locus. (* denotes significant LD blocks) Block Haplotype Block Haplotype Case, Control Chi Square P Value number Frequency Frequencies Block 1 GTCTCAG 0.344 0.209, 0.515 10.193 0.001* CCTCTTA 0.227 0.403, 0.044 17.152 3.45E-05* GTTCTTA 0.222 0.251, 0.181 0.695 0.405 GTTTCTG 0.076 0.036, 0.140 3.589 0.058 GTCTCTG 0.022 0.002, 0.048 2.398 0.122 CCCTCAG 0.02 0.020, 0.022 0.003 0.961 CCTCTTG 0.019 0.000, 0.000 0.003 0.957 CCTTCAG 0.019 0.036, 0.000 1.603 0.205 GTTCTAG 0.018 0.017, 0.023 0.04 0.841 CTTCTTA 0.013 0.006, 0.001 0.153 0.695 Block 2 TGG 0.768 0.721, 0.818 1.267 0.260 CCA 0.221 0.278, 0.159 1.962 0.161 TCG 0.011 0.001, 0.023 1.074 0.3 Block 3 TAG 0.526 0.599, 0.434 2.858 0.091 GTT 0.285 0.164, 0.415 8.254 0.004* GAT 0.18 0.236, 0.129 1.938 0.164

Table 3.5: Haplotype association analysis of FRMD6 locus (*denotes significant haplotypes) Haplotype Block Haplotype Frequency Case, Control Frequencies Chi Square P Value Block 1 AAG* 0.497 0.403, 0.640 5.867 0.0154 GCA* 0.293 0.433, 0.065 17.754 2.51x 10-5 AAA 0.191 0.161, 0.273 1.973 0.1601 ACA 0.019 0.003, 0.022 0.834 0.3612

Table 3.6: Haplotype association analysis of RORA locus (*denotes significant haplotypes) Haplotype Block Haplotype Freq Case, Control Frequencies Chi Square P Value Block 1 ACC 0.587 0.570, 0.626 0.339 0.560 ACG 0.246 0.239, 0.243 0.002 0.966 GTG 0.167 0.190, 0.130 0.675 0.411 Block 2 TGC 0.845 0.800, 0.870 0.894 0.344 CAT 0.155 0.200, 0.130 0.894 0.344 Block 3 CG 0.714 0.683, 0.713 0.108 0.745 TA 0.276 0.316, 0.262 0.35 0.554 Block 4 CCAAGC 0.335 0.310, 0.395 0.831 0.362 CCGAGC 0.222 0.223, 0.214 0.013 0.909 CTGAGC* 0.202 0.283, 0.108 4.898 0.027 GCAGAT 0.144 0.117, 0.160 0.412 0.521 GCAAAT 0.07 0.033, 0.101 2.038 0.153 GCAAGT 0.016 0.017, 0.014 0.008 0.929 Block 5 AA 0.914 0.967, 0.913 1.402 0.236 GG 0.086 0.033, 0.087 1.402 0.236 Block 6 GC 0.613 0.643, 0.643 0 1 AT 0.387 0.357, 0.357 0 1

3.3.4 Gene Expression Analyses:

3.3.4.1 Differential Expression in 300 cell lines

The mRNA expression analysis of four transcripts, two from RORA and one each from DSG1 and FRMD6, were previously investigated in 300 tumor cell lines and were available at the CaARRAY microarray database (Table 3.7). Our analysis of these data indicated that overall, DSG1 and RORA were found to represent poorly-expressed genes, whereas medium expression was observed for FRMD6. The expression was altered in all four transcripts belonging to RORA, DSG1, and FRMD6 in the context of 300 cell lines.

Table 3.7: mRNA expression of DSG1, RORA, and FRMD6 in 300 cancer cell lines panel from CaARRAY microarray database. *Only expression values above the detection threshold have been included in the study. Cells have been grouped by the tissue type, and the mean expression signal has been obtained

Average expression by tissue (detection signal) with p<0.05* Tissue type DSG1 # cell lines RORA # cell lines FRMD6 # cell lines Bladder 149.3 2 57.9 2 1422.2 10 Bone n/a 0 42.5 3 1703.8 3 Brain n/a 0 33.9 2 374.9 6 Breast 76.0 1 75.5 9 451.5 17 Central Nervous n/a 0 56.8 6 951.0 11 System Cervix Uteri n/a 0 47.8 1 804.5 7 Colon n/a 0 125.8 9 262.7 12 Connective Tissue n/a 0 87.2 2 1743.8 3 Esophagus n/a 0 n/a 0 1015.5 3 Eye n/a 0 59.3 1 89.1 1 Heme and lymph 43.1 5 204.8 31 190.2 16

Kidney n/a 0 60.7 1 813.8 6 Liver n/a 0 58.0 3 682.2 8 Lung 13.0 2 64.0 30 570.0 64 Muscle n/a 0 83.6 2 300.7 4 Ovary n/a 0 51.8 1 763.3 3 Pancreas n/a 0 11.7 1 416.4 9 Pharynx n/a 0 n/a 0 1368.4 2 Placenta n/a 0 n/a 0 92.6 3 Prostate 46.6 1 27.3 1 708.7 3 Sarcoma n/a 0 n/a 0 2915.4 2 Skin n/a 0 63.0 3 1060.6 11 Stomach n/a 0 n/a 0 205.0 2 Synovial Membrane n/a 0 60.4 1 440.9 1 Thyroid gland n/a 0 45.3 2 702.1 4 Uterus n/a 0 49.0 2 321.1 7 Vulva n/a 0 42.8 1 2779.5 3 Overall mean 65.6 11 64.0 114 1004.1 221

3.3.4.2 Expression in breast versus other cancer cell lines among the panel of 300 cell lines

In addition, the expression values for these genes in individual breast cancer cell lines were compared to the mean expression of the 300 cell lines using the data in the CaARRAY microarray database. Unfortunately, expression of DSG1, RORA, and FRMD6 was not above the detection threshold level in all cell lines. They were detectable only in one, 9 and 17 of 45 breast cancer cell lines, respectively (Table 3.8). Nevertheless, we found that DSG1 (0.2 standard deviation (SD) above the mean) and RORA (0.3 SD above the mean) had a slightly higher expression level in breast cancer cell lines, whereas FRMD6 had lower expression (0.55 SD below mean).

Table 3.8: DSG1, FRMD6, and RORA expression in breast cancer cell lines *(p value reflects the level of the background noise and indicates the whether the gene expression was above the detection threshold on the genome-wide scale).

Gene name RORA (210426 ) FRMD6 (225481) DSG1 (206642) (probeset ID)

Detection Detection Experiment Detection Experiment Experiment Source name number Signal P value number Signal P value number Signal P value

BT20 182965 21.2295 0.219482 182965 434.795 0.0002441 182965 2.44677 0.932373

BT474 85193 33.9827 0.0561523 85191 121.754 0.0019531 85191 2.3484 0.888428

DU4475 172126 135.374 0.0041504 172126 166.427 0.0002441 172126 11.1891 0.466064

EFM19 165488 13.3248 0.19458 172129 132.505 0.0012207 172129 4.62991 0.80542

HCC1143 182980 9.51671 0.19458 170301 1350.77 0.0002441 170301 1.56125 0.850342

HCC1395 170301 50.5562 0.0185547 209551 1372.91 0.0002441 209551 3.42012 0.850342

HCC1599 209551 11.2129 0.398926 172135 195.435 0.0002441 172135 2.16859 0.80542

HCC1937 172135 49.7536 0.0041504 95936 452.432 0.0002441 95936 3.59534 0.780518

HCC1954 95936 42.1928 0.0561523 74903 645.864 0.0002441 74903 10.337 0.780518

HCC2157 74903 30.7828 0.334473 172138 37.0058 0.0952148 172138 1.05106 0.981445

HCC2218 172138 213.938 0.0019531 182977 209.439 0.0002441 182977 5.26698 0.828613

HCC38 182977 46.7165 0.0952148 91457 725.952 0.0002441 91457 75.9615 0.0141602

KPL1 172150 22.7886 0.27417 172150 220.662 0.0002441 172150 1.69712 0.904785

MCF7 127704 12.2952 0.0302734 127703 372.244 0.0002441 127703 2.42738 0.870361

MDAMB175VII 209563 46.7423 0.0080566 209563 378.45 0.0019531 209563 6.26103 0.780518

MDAMB453 63383 66.4183 0.0185547 63383 394.542 0.0002441 63383 7.4979 0.5

MT3 172162 24.5753 0.0302734 172162 1.57164 0.696289 172162 2.86447 0.919434

UACC812 168808 27.7521 0.246094 168808 114.896 0.0141602 168808 2.55629 0.943848

UACC893 168813 79.8495 0.0302734 168811 385.488 0.0007324 168811 2.06918 0.943848

3.3.4.3 Expression in breast tumor vs. normal tissue

To ensure that the observed variable expression in DSG1, RORA, and FRMD6 was not a cell line-specific phenomenon, we looked at their expression in breast tumor samples. Although our genes of interest were not among the top variably expressed genes in the microarray study of 60 breast samples, all three genes have shown differential expression between normal and breast tissue (Table 3.9). RORA (Affymetrix transcript# 210426 and 210479) and FRMD6 were found to be significantly over-expressed in breast tumors compared to normal breast tissue (p=1.534x10-5, 0.039, and 0.009, respectively). DSG1 was up-regulated in breast tumors, but results were not significant (p = 0.340).

Table 3.9: FRMD6, DSG1, and RORA expression in breast tumors. *data obtained from the ArrayExpress, E-TABM-276; tumor tissues were compared to normal breast tissues from healthy individuals. †Affymetrix GeneChip U133 Plus 2.0 was used for probe sets. ‡ Data was tested for the equality of variances by Levin‟s test.

95% Confidence Interval of Group Statistics the Difference

Std. Std. Error Sig. (2- Mean Disease State N Mean Deviation Mean t‡ tailed) Difference Lower Upper

FRMD6 Normal 20 380.800 260.716 58.2979 -2.706 0.009 -292.000 -507.900 -76.10 225481† Breast tumor 41 672.800 445.581 69.5880

DSG1 Normal 20 15.945 17.439 3.8995 -0.962 0.340 -13.153 -40.508 14.203 206642_at Breast tumor 41 29.098 59.676 9.3198

RORA Normal 20 39.380 17.665 3.9499 -4.713 1.534x 10-5 -33.740 -48.064 -19.415 210426_x_at Breast tumor 41 73.120 38.227 5.9701

RORA normal 20 56.215 28.972 6.4782 -2.114 0.039 -23.346 -45.447 -1.245 210479_s_at Breast tumor 41 79.561 44.947 7.0196

3.3.4.4 mRNA expression in the resistant cancer cell lines:

To assess the relationship between the expression of DSG1, RORA or FRMD6 and DOX sensitivity, we compared transcript levels of the three genes between isogenic doxorubicin- sensitive and doxorubicin-resistant tumour cell lines by qRT-PCR using the comparative CT method. This analysis showed that RORA was significantly up-regulated in DOX-resistant uterine sarcoma cells (MES-SADx5) and EPI-resistant breast cancer cells (MCF-7EPI) compared to their wild-type, drug-sensitive counterparts (p= 4.114x 10-5, and 3.147x10-2, respectively). RORA was also found to be up-regulated in the other doxorubicin-resistant cell lines, although the observed difference was not statistically significant (Table 3.10,Table 3.11). Moreover, A2780 and MES-SA cell lines showed no amplification of the gene in the wild type (WT) cells suggesting an absence of RORA gene expression in these doxorubicin-sensitive cells (Figure

3-7). DSG1 showed a similar up-regulation in doxorubicin-resistant MES-SADx5, MCF-7DOX and

MCF-7EPI cell lines (p= 4.114E-05, 0.297, and 0.008, respectively). The only exception was -ΔΔCT A2780DOX cells, where DSG expression was only slightly down-regulated (2 was 1.18 cDNA copy numbers). The opposite pattern of gene expression in doxorubicin resistance was observed for FRMD6, which was remarkably under-expressed in A2780ADR, MES-SADx5, MCF-

7DOX, and MCF-7EPI cells compared to isogenic doxorubicin-sensitive counterparts (p<0.05). The non-parametric analysis confirmed the significance of the findings for all three genes in MES- SA- and MCF-7 cell lines as well as in A2780 cell lines for FRMD6. Although some of the findings in certain cell lines proved not to be significant, they followed the same trend for each particular gene as in other cell lines.

Table 3.10: cDNA expression of DSG1, RORA, and FRMD6 in A2780, MES-SA, and MCF- 7cc cell lines

*Mean fold change have been determined by the comparative cycle threshold method (2-ΔΔCT), all values have been calibrated against the reference gene, S28, amplified with each candidate gene)

Cell Line Mean Fold change

DSG1 RORA FRMD6

A2780-ADR 1.18 9.979 0.824

MES-SA-Dx5 5972.93 3545.47 0.484

MCF-7cc-Dox 5.301 5.584 0.992

MCF-7cc-Epi 14.59 62.502 0.986

Table 3.11: Comparison of the cDNA levels of the DSG1, RORA and FRMD6 between WT and resistant cell lines on qRT-PCR * Analysis has been done based on the CT values of each gene normalized to the S28 control gene (as determined by Mann-Whitney test); † after exclusion of the experiments from the MCF- 7cc-DOX12 (2).

A2780 MES-SA MCF7cc-DOX† MCF7cc-EPI

DSG1

Mann-Whitney U 19.000 0 28.000 11.000

Z -1.898 -3.576 -1.104 -2.605

Exact Sig. [2*(1-tailed Sig.)] 0.063 4.114x10-5 0.297 0.008

FRMD6

Mann-Whitney U 17.000 14.000 26.000 9.000

Z -2.075 -2.340 -1.280 -2.782

Exact Sig. [2*(1-tailed Sig.)] 0.040 0.019 0.222 0.004

RORA

Mann-Whitney U 27.000 0.000 38.000 16.000

Z -1.192 -3.576 -0.221 -2.163

Exact Sig. [2*(1-tailed Sig.)] 0.258 4.114 x10-5 0.863 0.0315

Figure 3-7: Comparison of the cDNA levels of DSG1, RORA and FRMD6 between WT and resistant cell lines on qRT-PCR.

(Plotted on the SPSS 13.0, y-axis corresponds to the CT values of the gene expression in each cell line; A2780, MCF7cc, and MES-SA are the WT, whereas A2780-ADR, MCF7cc-Dox12, MCF7cc-EPI, MES-SA-Dx5 are the resistant cell lines). Asterisk labels significant difference in the gene expression between WT and resistant cell lines.

3.3.5 Karyotypic Alterations:

Eleven candidate SNPs were mapped to six genomic regions. Those six loci of interest (7q21, 12p12, 14q21, 15q21, 17p12, 18q12) were compared for copy number variation and chromosomal translocations among 55 members of the NCI60 panel of cell lines (Table 3.12). This analysis showed an average absolute number of allele copies above 2 in the majority of the cells with the range of means from 2.42 to 3.92 copies per loci. This finding may suggests an increase in copy number for the genes located in the same region compared to normal cells. An additional comparative study of the copy number variations between doxorubicin-sensitive and doxorubicin-resistant cell lines at these loci showed that there was no statistically significant difference in copy number between these two groups (

Table 3.13).

Table 3.12: Karyotype analysis within the six loci of interest (Chr. 7q21, 12p12, 14q21, 15q21, 17p12, 18q12) in NCI60 cell lines. * Cell lines have been divided into two groups according to the initial case-control design; † Average copy number variations for resistant and sensitive NCI60 cell lines; ‡ t-test of 2 independent samples on SPSS. Group Statistics t-test for Equality of Means‡ Mean 95% Confidence Interval (ploidy Std. Std. Error Sig. (2- of the Difference Cell response* N* number)† Deviation Mean t tailed) Lower Upper Chr 7q21 resistant 31 3.35 1.539 0.276 -1.333 0.189 -1.409 0.285 sensitive 24 3.92 1.558 0.318 Chr 12p12 resistant 31 2.94 1.031 0.185 0.361 0.720 -0.467 0.671 sensitive 24 2.83 1.049 0.214 Chr 14q21 resistant 31 3.10 1.012 0.182 0.897 0.374 -0.327 0.854 sensitive 24 2.83 1.129 0.231 chr 15q21 resistant 31 3.23 1.431 0.257 0.166 0.869 -0.654 0.772 sensitive 24 3.17 1.204 0.246 Chr 17p12 resistant 31 2.81 0.873 0.157 0.573 0.570 -0.351 0.630 sensitive 24 2.67 0.917 0.187 Chr 18q12 resistant 31 2.42 1.025 0.184 -0.726 0.472 -0.775 0.364 sensitive 24 2.63 1.056 0.215

Table 3.13: Translocations of the candidate loci in NCI60 cancer cell lines. NCI60 tumor type Phenotype Leukemia NSCL Breast Colon CNS Melanoma Prostate Renal Ovary unknown Total Percent # of resistant 1 4 5 6 2 5 1 3 6 1 34 100 cell Cell lines lines sensitive 6 3 2 4 4 2 1 7 0 0 29 100 Chr resistant 0 1 2 0 0 1 0 1 1 0 6 17.65 Translocations 7q21 sensitive 1 1 1 0 0 1 0 0 0 0 4 13.79 Chr resistant 0 2 0 1 0 0 0 0 2 0 5 14.71 Translocations 12p12 sensitive 0 0 0 1 0 0 0 1 0 0 2 6.90 Chr resistant 0 2 1 2 1 2 0 1 4 1 14 41.18 Translocations 14q21 sensitive 0 2 2 0 1 0 1 1 0 0 7 24.14 chr resistant 1 2 1 1 0 1 1 1 1 0 9 26.47 Translocations 15q21 sensitive 1 1 2 0 2 1 1 0 0 0 8 27.59 Chr resistant 0 2 1 0 0 1 0 0 1 0 5 14.71 Translocations 17p12 sensitive 0 0 0 0 1 0 0 0 0 0 1 3.45 Chr resistant 0 0 0 0 0 0 1 0 1 0 2 5.88 Translocations 18q12 sensitive 0 0 1 0 0 0 0 1 0 0 2 6.90

3.4 Discussion

3.4.1 Genetic Analyses:

GWA studies can provide us with important information on genetic variations associated with drug response in tumor cells that can be used to establish predictive gene sets that, if validated in the tumors of cancer patients, could be used as biomarkers to help optimize the management of cancer patients with chemotherapy (130). In this study, we performed GWA analysis using the genotypes of DOX-sensitive and DOX-resistant NCI60 cell lines to determine genetic variations associated with response to DOX. Previously published studies suggest that treatment of cancer cells with 1M DOX can be accepted as a clinically relevant acute dose, whereas higher cumulative doses with longer exposure to 50nM have been referred to as a sub-lethal concentration (131). The maximally tolerated clinical DOX concentration (Cmax) in the plasma of cancer patients is 3-10 M however, cellular DOX levels are usually 30-100 times higher than that achievable in the plasma of cancer patients (10-5 - 10-4 M) (132). Based on this knowledge, we have utilized the NCI60 cell line drug response data from the DTP website obtained using a DOX concentration of 10-4 M to reveal genetic abnormalities leading to drug resistance.

In this study, we have identified four SNPs mapped within FRMD6, RORA, TMEM220 and DSG1 that are associated with differential response to DOX. Seven other SNPs associated with DOX response were located in the intergenic regions. Interestingly, some of the identified SNPs mapped in the vicinity of each other. For example, SNPs AFFY#843236 and #843306 (rs1986410) were located in the vicinity of rs726529, which maps within the FRMD6 gene. Another SNP (AFFY#1177280) (rs1365296) was located between the DSC1 and DSG1 genes in the chr18q12 region, which is known to contain the desmosomal cadherin family of genes. In addition, two more SNPs were located on Chr17 [AFFY#1089881, and #1095434 (rs1122073)] proximal to TMEM220. Although at this time we cannot provide more information about the relevance of the intergenic SNPs in DOX response, there are several precedents in other genome-wide studies that identified variants associated with drug response in intergenic regions (133, 134).

Provided that SNPs used for the study represent markers evenly spread throughout the genome, the abovementioned observations further support the demonstrated association of the identified regions with resistance to doxorubicin.

Haplotype analysis of the three well-characterized gene loci further enhanced an observation from the single-marker association study. It can be inferred from Figure 3-5 that association is driven by SNPs in 2 LD blocks at 18q12 and 1 LD block at each of 15q22 and 14q22.1, which are in LD with neighboring SNPs. Hence, those areas may be a candidate region or in a LD with the true region associated with DOX resistance. Based on our results, the genetic basis of those regions are not known yet and warrant further fine-mapping analysis and investigation.

In this study, apart from the fact that the overall number of DNA copies is higher, no significant correlation was detected between CNV and resistance to DOX. These findings are in line with previous genome-wide cCGH study on breast cancer patients by Pierga J-Y et al, which demonstrated no significant correlations between clinical response to anthracyclines and DNA copy number in 44 patients (99). Lack of difference between CNV in candidate loci allows us to speculate that the observed difference in the gene expression might be due to the changes in the activity of specific transcriptional regulators.

3.4.2 Characterization of Biologic Relevance

3.4.2.1 DOX, Candidate Genes and Apoptosis

DOX affects both extrinsic and intrinsic apoptotic signaling pathways, both of which mediate cell death via a family of cysteine proteases (caspases) (135-140). It affects: a) the extrinsic apoptotic pathway by upregulating FAS/L expression, leading to its trimerization with FAS receptor, and b) intrinsic apoptosis through the release of cytochrome c and formation of intracellular reactive oxygen species (ROS) in mitochondria by a DOX metabolite. In turn, DOX activates CASP9, CASP8 and CASP2, the triggers of cell apoptosis (141, 142). Interestingly, all three candidate genes were also

72 found to be involved in cellular processes which regulate cellular apoptosis, and thus likely interfere with the anti-tumor activity of DOX (108).

DSG1, a calcium-binding transmembrane glycoprotein, is a major member of the desmosomal family of adhesion molecules, including DSG1, DSG2, DSG3, DSC1, DSC2, and DSC3, all clustered on chromosome 18q12. DSG1 participates in cell-cell adhesion and apoptotic processes, and serves as an essential molecule for cell morphogenesis and positioning. By enhancing cell-cell adhesion, DSG1 activity may interfere with DOX entry into tumor cells through its effects on “cell packing density” (143, 144). Early studies have shown better penetration and cell invasiveness of DOX into solid tumor tissues with less cohesively structured cell cultures (i.e. lacking both desmosomes and cadherin). An in vitro study of multicellular tumor spheroids and a monolayer model of MCF7 breast cancer cell lines showed increased DOX resistance and greater survival in the cells with typical tight junctions and desmosomes compared to those lacking such cohesions (144, 145).

In addition, DSG1 is directly cleaved by CASP3, the main apoptotic effector caspase, which has been shown to lead to decreased expression and dismantling of desmosomes during keratinocyte apoptosis (Figure 3-8) (146). The same authors have also shown that siRNA-mediated knockdown of DSG1 is associated with the inhibition of apoptosis in the cells (146). In line with these findings, our GWA study is the first to show an association between DSG1 SNPs and altered expression and DOX resistance.

RORA, or NRF1, is a constitutive transcription factor that belongs to the nuclear receptor super-family and spans a common fragile site (CFS) FRA15A that has been highly conserved during evolution. RORA plays an important role in gene transcription and cell growth regulation. It is an effective ROS scavenger in the early mitochondrial stage (147, 148), it positively regulates transcription from the RNA polymerase II promoter, and it up-regulates p21 and N-myc expression through binding to the ROR- response element, leading to cell growth inhibition (149); finally, it is down-regulated in cancer-derived cell lines and primary tumors and up-regulated in conditions of cellular stress (hypoxia, H202). It also regulates DNA damage checkpoint proteins (150-152)

(Figure 3-8). Since ROS formation is a key component of DOX action (by activation of the JNK/c-Jun pathway), over-expression of RORA would be expected to prevent ROS- induced apoptosis in cells, thereby causing drug resistance (58, 150, 153).Thus, underexpression of RORA is most likely to favor apoptotic processes and to potentiate the effect of DOX.

FRMD6 is a member of the 4.1 super-family of tumor suppressor genes, which includes the closely related proteins ezrin, radixin and moesin (ERM), merlin, talin and protein-tyrosine phosphatases. These proteins maintain the link between transmembrane proteins and the sub-membrane cytoskeleton (154). FRMD6 is distributed both in the cytoplasm and the plasma membrane, which contains an active form of protein. Levels of protein, activated by phosphorylation or tyrosine-kinase pathway, however, vary depending on cell-to-cell contact, and on the cell cycle stage (155). FRMD6 also seems to be involved in apoptosis, carcinogenesis and metastasis through the FERM domain (156, 157) (Figure 3-8). Therefore, functional studies also indicate a link of all three genes identified in GWA study as associated with DOX resistance at the cellular level.

Cell Membrane

Nucleus

Figure 3-8: Gene-gene interactions: DSG1, FRMD6, RORA with doxorubicin and epirubicin. Candidate genes are in red.

3.4.2.2 mRNA Expression Analyses

3.4.2.2.1 DSG1 mRNA expression

Over-expression of DSG1, a desmosomal cadherin, would be expected to enhance cell adhesion and facilitate development of resistance to DOX by increasing cell packing density and reducing diffusion of DOX in tumor cells (143). In malignant cells, loss of adhesion leads to matrix detachment, decreased levels of p53 and genomic instability which contribute to cell survival, cancer growth and therapy resistance (158).

Our results have shown that DSG1 is over-expressed (6-fold, p<0.05) in breast tumor compared to normal breast tissue (122), which suggests that DSG1 alterations may play a role in breast cancer development. We also observed over-expression of DSG1 in DOX- and EPI- resistant cancer cell lines (MCF-7DOX, MCF-7EPI and MES-SADx5) compared to isogenic drug- sensitive parental cells. We can postulate that upon activation of apoptosis, DSG1 is rapidly depleted by cleavage from caspase3 which perturbs intercellular adhesion and would lead to compensatory overexpression of DSG1 in tumor cells (146).

3.4.2.2.2 RORA mRNA expression

Overall, RORA has been found to be universally expressed in normal human breast, liver, brain, ovary and prostate tissues (147). RORA‟s up-regulation in cancer cells hinders apoptotic processes and counteracts the effect of many pro-apoptotic drugs (147). It has also been reported that there is a slightly increased mRNA level of RORA in breast cancer cells compared to normal breast tissue, and that RORA transcriptional activity negatively affects MCF-7 cell proliferation (150, 151).

In vitro measurement of relative mRNA expression demonstrated a two-fold overexpression of RORA in breast carcinoma cell lines relative to normal breast tissue (p=1.49E-6) (122). In addition, we show in the current study an up-regulation of RORA in DOX-resistant

MES-SA cells and EPI-resistant MCF-7EPI cells (p<0.05). Since RORA is a known, effective scavenger of free radicals, increased expression of RORA may block the DNA damage

76 induced by DOX through its generation of ROS. This would mute the apoptotic effect of DOX on tumor cells, thereby inducing drug resistance (150).

3.4.2.2.3 FRMD6 mRNA expression

It has been shown that FRMD6 is down-regulated 6-fold by the N-myc gene in SHEP2 and SHPE21N neuroblastoma cell lines (159). At the same time, it is known that the N-myc gene is predominantly over-expressed in only embryonic tissues and growing/developing tissues. DOX increases the induction of cMYC (160, 161) and enhances the activity of TERT (telomerase reverse transcriptase) which directly regulates FRMD6 expression (131). In addition, RNAi profiling of c-MYC target genes in breast cancer cell lines showed c-MYC- dependent over-expression of FRMD6 in MCF-7 and MDA-MB-231 cells, and c-MYC underexpression in BT-474 cell lines (ArrayExpress, access Number: E-GEOD-5823) (162) (Figure 3-9). Hence, the lack of this tumor suppressor gene may prevent tumor cells from committing to apoptosis, the main mechanism of action of doxorubicin in the cancer cells (163).

Over-expression of FRMD6 (2.3-fold) has been observed in breast tumors compared with normal breast tissue (p=3.36E-6) (122). In addition, we have observed a down-regulation of

FRMD6 expression in DOX-resistant MES-SADx5, and EPI-resistant MCF-7EPI cells (p<0.05).

Interestingly, up-regulation of RORA expression promotes an activation of the N-myc gene that directly down-regulates the FRMD6 gene (162). As a nuclear receptor, RORA forms a DNA-bound complex with its cofactor and the co-activator EP-300 (162). This induces a cascade of events: EP-300 reduces transcription of the c-MYC gene which negatively affects transcription of N-myc. The low level of transcribed c-MYC gene causes up-regulation of n- MYC, and hence, down-regulation of FRMD6 (164). Consequently, we can postulate that the synergistic interplay of up-regulated RORA and down-regulated FRMD6 would foster the development of resistance to DOX in cancer cells.

There are limitations to the interpretation of the experimental findings described in this study. For example, additional molecular alterations or groups of genetic variations may play critical roles in DOX response. A direct causal relationship between the presence of specific SNPs, gene expression, and drug resistance remains to be established. Nevertheless, this study

77 represents a new paradigm of assimilating bioinformatic tools with in vitro model systems to identify and confirm novel variations in gene structure and expression associated with response to DOX. As a result, we were able to reinforce our hypotheses and in silico findings with experimental data from cells lines and tumor samples. Our rationale has been based on the recent notion of utilizing publicly available databases with world-wide access and contributions to discover the potential candidate genetic variants associated with the resistance to DOX. Although the genome-wide model leads the way to establish a correlation between susceptibility loci and the therapeutic agent, it cannot as of yet indicate the causative relationship between them. Overall, the results of our study support the concept that genes affecting various apoptotic signaling pathways may be key in resistance to anthracyclines.

Figure 3-9: Expression comparison of DSG1, RORA, and FRMD6 in normal and tumor breast tissues of 61 individuals: (Accession #: E-TABM-276 data) (analyzed by t-test on SPSS 13.0). Y-axis corresponds to the Signal detected on gene expression, and X-axis is categorizing data according to disease state); asterisk indicates p<0.05

3.5 Summary

Our results show that genome-wide analysis is a powerful approach for the localization of genes associated with various phenotypes including chemotherapy drug response. This unique approach that does not have precedents in the past brings up the opportunity to reveal novel genetic profiles responsible for variable phenotypic appearances among patients and can serve as a proof-of-principle study. DSG1, RORA, and FRMD6 are pro-apoptotic genes that appear to play an important role in doxorubicin response in a variety of tumor cell lines. Future studies will be directed at illustrating a causative relationship between the above-described changes in gene expression and drug resistance in vitro and in tumors of breast cancer patients undergoing doxorubicin chemotherapy.

Chapter 4 Genome-wide Association Studies of Taxol Response in Cancer

Abstract

Paclitaxel is a microtubule-stabilizing drug that has been used against a variety of cancers. Despite aggressive treatment regimes using paclitaxel, response rates have varied from 20-86%. Genetic variability among patients opens avenues to identify novel genetic markers associated with paclitaxel response. Here, we used the GI50 response data on the NCI60 panel, available from the Developmental Therapeutic Program, to identify novel candidate markers in the form of SNPs associated with paclitaxel response. Using the previously described case-control design for genome-wide approach in single marker association, we defined sensitive and resistant cell line groups. Next, we used PLINK software to perform a GWAS of paclitaxel response with the NCI60 panel‟s SNP- genotype data on the Affymetrix 125k SNP-chip. 43 SNPs were found significantly associated with drug response after adjustment by FDR (p< 0.005 after correction) with 11 belonging to protein-coding genes (CFTR, ROBO1, PTPRD, BTBD12, CCDC26, DCT, SNTG1, SGCD, LPHN2, GRIK1, ZNF607). Using the bioinformatic tools: Ingenuity and FastSNP, we have proposed a model demonstrating the interaction among transcription factors: β-catenin, p53 and their interacting proteins along with microtubular interactions in the cellular response to paclitaxel. Analysis of mRNA expression data from BioGPS found significantly increased expression of DCT and SGCD in sensitive cell lines (p < 0.05). Using Haploview, haplotypes in GRIK1, ROBO1, SGCD, LPHN2, and PTPRD were found more significantly associated with paclitaxel response than their individual SNPs, highlighting the role of SNP-SNP interactions. These genetic variants represent potential powerful biomarkers for predicting paclitaxel response among patients.

80 bstract

4 Genome-wide Association Studies of Taxol Response in Cancer

4.1 Introduction

Paclitaxel (TaxolTM), a member of the taxane family, has been used to treat cancer since its approval by the Food and Drug Association (FDA) in 1992. It has commonly been used to treat ovarian and breast cancers, but has also been used to treat head and neck cancers, lung cancers, esophageal cancers, testicular cancers and sarcomas (90, 165-167).

Like other drugs of the taxane family, paclitaxel binds to microtubules to arrest cell growth and cause apoptosis (90). However, unlike other drugs that target microtubules and cause instability, paclitaxel stabilizes microtubules leading to the disruption of mitosis (G2/M cell-cycle arrest) and eventually undergo apoptosis through so called caspase-independent mitotic death (166, 168). There is also a microtubule- independent pathway leading to activation of intrinsic apoptotic pathways via caspase activation (90). TP53, a control switch of cell-cycle arrest, apoptosis and cell senescence, has been implicated in paclitaxel sensitivity, but still remains a matter of controversy as it seems to be activated by paclitaxel indirectly through downregulation of microtubules-stabilizing protein, MAP4 (90, 166). In addition, the PI3K/Akt and Raf-1 kinase pathways have also been implicated in paclitaxel induced microtubule-independent apoptosis (166).

Despite aggressive treatment regimens using paclitaxel, response rates are unsatisfactory and vary among patients. The overall response rate observed from clinical studies of breast cancer patients treated with paclitaxel varied from 21 to 86% with a median survival time from 11.8 to 22 months (166). Although dose-dense schedules usually have a better drug response, they can lead to additional adverse effects and increased toxicity. In particular, peripheral neuropathy, myelotoxicity, and severe neutropenia are often the dose-limiting factors (166).

Variation and unsatisfactory efficacy in drug response are not limited to cancer treatments, but include the general use of drug therapy (169). Although, the observed variation can be accounted for by differences in environmental factors such as age, weight or compliance, this variation can also be attributed to differences in the genetic make-up of patients (169). Genetic variation can occur in genes involved in the drug‟s pharmacokinetics, the drug‟s cellular target or other signaling pathway proteins downstream of the target (169). Genetic variability can include SNPs, micro-satellites or copy number variation (169).

In this study, we have expanded upon a previous approach by our group to identify novel genetic variants that are associated with a therapeutic response to paclitaxel (50). We use the NCI60 cell line panel as a model to simulate a patient population. The NCI60 panel has been commonly used to screen novel therapeutic agents and also has had its genetic make-up thoroughly characterized (44). Here, we use the drug screening results for paclitaxel and genomic data on the NCI60 panel, both available from DTP, to perform a GWAS of drug response with genomic variation in the form of SNPs. In addition, bioinformatic tools are applied to further characterize the role of these variations in drug response. In turn, these novel variants can serve as potential biomarkers to help to identify patients that will best respond to paclitaxel treatment and improve personalized medicine.

4.2 Materials and Methods

4.2.1 GWA Analysis:

4.2.1.1 Case-Control Design

The cellular response data for the NCI60 panel to paclitaxel treatment was acquired through DTP (http://dtp.nci.nih.gov/). Specifically, we obtained the GI50 response data at the 10-6 M dose (effective concentration that arrests cellular growth in 50% of the cells) for the NCI60 panel. The GI50 value is estimated from growth measurements taken at the 10-6M dose and 4 additional serial dilutions of this dose and appropriate controls as described at the DTP website (http://dtp.nci.nih.gov/branches/btb/ivclsp.html). The log10

82 of the GI50 values was normalized with a mean of 0 and standard deviation of 1. A non- parametric kernel estimation on the normalized GI50 values was performed by SAS 9.1 to define sensitive and resistant cell lines as described in the previous chapter (50). The cut-off value of 1.2 was determined by the visual antimode to define sensitive (cases, n=50) and resistant (controls, n=8) cells within the multimodal population.

4.2.1.2 Single-Marker Association Analysis

A total of 118, 409 SNPs were genotyped on the Affymetrix 125K Chip with median intermarker distance of 8.5 kilobases for the NCI60 panel (114). As a result of the quality control measures 20, 514 SNPs were removed for missing more than 25% of the genotype data, 20,176 SNPs were removed for having a minor allele frequency (MAF) of less than 2%, leaving 79,622 SNPs for further analysis (Figure 4-1). Drug response data for the cell lines from the DTP website were matched to the genomic data available from the SNP chip. Fifty-eight NCI60 cell lines (50 sensitive and 8 resistant) with genomic data available were used in the GWAS using PLINK (http://pngu.mgh.harvard.edu/~purcell/plink/) (115). We carried out single-marker association on a PLINK by chi-square test. Results were adjusted by post-hoc FDR-BH test. SNPs with a p value less than 0.005 were selected as significant. Fine-mapping of the identified SNPs to a genomic region was based on dbSNP v.36.3 (http://www.ncbi.nlm.nih.gov/projects/SNP/). Gene description was obtained from an online database, GeneCard (http://www.genecards.org).

Figure 4-1: SNP selection process

TOTAL: 118, 409 SNPs

20, 514 SNPs excluded ( were missing >25% of their genotype)

20,176 SNPs excluded (MAF<2%)

SELECTED for ANALYSIS: 79,622 SNP

4.2.1.3 Haplotype Analysis

To identify whether haplotype blocks that included our originally identified SNPs were more informative and significantly associated with paclitaxel response than the candidate SNP, a case-control haplotype analysis was performed using the cell line SNP data and the Haploview software (http://www.broad.mit.edu/haploview/haploview) (119). For each gene identified, we inputted the remaining SNPs belonging to the same gene and their associated genotype data from the Affymetrix 125K chip to determine if any haplotypes were more significantly associated with drug response (in terms of unadjusted p values). Pairwise SNP comparisons of markers that were greater than 500kb and cell lines that were missing 50% of their data were excluded from this analysis. SNPs that were missing 75% of their genotypes, or had a MAF less than 0.1% were also excluded from this analysis. Only LD blocks with r2 >0.8 were selected for analysis.

4.2.2 In silico prediction:

4.2.2.1 Functional Prediction

Functional prediction was performed for each candidate SNP to determine the potential effect it has on the corresponding gene product. This was conducted using online software, FastSNP (http://fastsnp.ibms.sinica.edu.tw/)(170). This tool looks at the risk of having the specific SNP variant (i.e. intronic enhancers, missense mutations) as well as predicts changes at putative transcription factor binding sites.

4.2.2.2 Interactome Analysis

To reveal an interaction of involved signaling pathways between identified candidate genes, their protein products and paclitaxel, we performed analysis on Ingenuity, a tool to model and analyze complex biological systems with a comprehensive online search algorithm (www.ingenuity.com).

4.2.2.3 mRNA Expression Analysis and Clinical Validation

In order to determine if changes in the expression of our identified genes will affect paclitaxel sensitivity, we obtained the mRNA expression data in NCI60 cell lines from BioGPS, the Gene Portal Hub (http://biogps.gnf.org) (171). mRNA expression values from Affymetrix U133A GeneChip were summarized using GCRMA algorithms based on the fluorescence intensity for each probe on the microarray. Cell lines were allocated to the sensitive (cases) and resistant (controls) groups according to the same design used for GWA SNP analysis. A Mann-Whitney test was used to determine if the expression level differences between the two groups were significant (p < 0.05).

4.2.3 Results

4.2.3.1 GWAS

The analysis of 58 NCI60 cell lines showed bimodal distribution after density kernel estimation with an optimal bandwidth of 0.2729, an AMISE of 0.0197 and a visual

85 antimode of 1.2 (Figure 4-2). Specific listing of cell lines belonging to sensitive (n=50, z<1.2) and resistant (n=8, z>1.2) categories can be found in Table 4.1.

Sensitive Resistant

Figure 4-2: Segregation of NCI60 cell line panel for the 10-6 M dose group of paclitaxel into sensitive and resistant groups. Analyzed on SAS 9.1 by non-parametric Kernel density estimation test). Green line represents an antimode; red line- distribution of the cells; empty bins correspond to the number of cell in each bin. Antimode=1.2 (the cutoff point selected as the point of separation between 2 modes of distribution); 50 cells are sensitive; 8 cells are resistant. Cells were analyzed after normalization. X axis, normalized values; y axis, percentage of the cells within each bin.

Table 4.1: Case Control Design for sensitive and resistant cell lines at 10-6 M dose group. This graph generated by SAS9.1 software shows the distribution of the NCI60 panel‟s GI50 response to paclitaxel after a non-parametric kernel estimation was applied. Sensitive- Cases Resistant-Controls Log SNP mRNA Log SNP mRNA Cell Type Cell Name Value Normal data data Cell Type Cell Name Value Normal data data Breast MDAN -8.513 -1.67261 N N NSCL EKVX -6.624 1.49829 Y Y Breast MDAMB435 -8.406 -1.493 Y Y NSCL HOP92 -6.504 1.69972 Y Y CNS SNB75 -8.321 -1.35032 Y Y Renal CAKI1 -6.399 1.87597 Y Y Colon HT29 -8.27 -1.26471 Y N Renal UO31 -6.335 1.9834 Y Y Leukemia RPMI8226 -8.24 -1.21435 Y Y Colon HCT15 -6.32 2.00858 Y Y Breast HS578T -8.212 -1.16735 Y Y NSCL LXFL529 -6.277 2.08076 N N Colon HCT116 -8.199 -1.14553 Y Y Ovarian OVCAR4 -6.23 2.15966 Y Y Leukemia HL60(TB) -8.124 -1.01963 Y N Renal ACHN -6.179 2.24527 Y Y Colon KM12 -8.055 -0.90381 Y Y Ovarian NCI/ADRR -6.05 2.46181 Y Y

NSCL NCIH522 -8.043 -0.88367 Y N Leukemia K562 -8.042 -0.88199 Y Y Breast MCF7 -8.041 -0.88031 Y Y Melanoma LOXIMVI -7.997 -0.80645 Y Y CNS SF539 -7.991 -0.79638 Y Y Colon COLO205 -7.958 -0.74098 Y Y NSCL NCIH460 -7.91 -0.66041 Y Y

Leukemia CCRFCEM -7.901 -0.6453 Y Y Melanoma SKMEL5 -7.872 -0.59662 Y Y Ovarian IGROV1 -7.87 -0.59327 Y Y Prostate TSUPRI -7.855 -0.56809 N N Colon SW620 -7.833 -0.53116 Y Y Colon HCC2998 -7.832 -0.52948 Y Y Leukemia SR -7.826 -0.51941 Y Y Ovarian OVCAR3 -7.82 -0.50933 Y Y Ovarian OVCAR8 -7.802 -0.47912 Y Y CNS U251 -7.801 -0.47744 Y Y Leukemia MOLT4 -7.755 -0.40023 Y Y

Prostate PC3 -7.754 -0.39855 Y N

Renal SN12C -7.728 -0.3549 Y Y

Prostate DU145 -7.666 -0.25083 Y N

Melanoma M14 -7.658 -0.2374 Y Y

NSCL NCIH23 -7.655 -0.23236 Y Y

CNS SF268 -7.625 -0.18201 Y Y

NSCL A549/ATC -7.611 -0.15851 Y Y

Ovarian SKOV3 -7.593 -0.12829 Y Y

Melanoma UACC62 -7.592 -0.12661 Y Y

NSCL NCIH322M -7.582 -0.10983 Y Y

Melanoma SKMEL28 -7.575 -0.09807 Y Y

Melanoma MALME3M -7.535 -0.03093 Y Y Breast BT549 -7.523 -0.01079 Y N Renal 7860 -7.462 0.09161 Y Y CNS SNB19 -7.429 0.147 Y Y NSCL HOP62 -7.41 0.1789 Y Y Ovarian OVCAR5 -7.392 0.20911 Y Y NSCL NCIH226 -7.375 0.23765 Y Y Renal RXF393 -7.336 0.30311 N Y Melanoma SKMEL2 -7.295 0.37194 Y Y Melanoma UACC257 -7.29 0.38033 Y Y Breast T47D -7.288 0.38369 Y N

Breast MDAMB231 -7.261 0.42901 Y Y

CNS SF295 -7.16 0.59855 Y Y TK10 -6.919 1.00309 Y Y Renal Renal A498 -6.903 1.02995 Y Y

A total of 43 SNPs were found to be significantly associated with paclitaxel response in a single-marker GWAS (FDR correction p< 0.005). The significantly associated SNPs were found to map to a total of eleven protein coding genes in the order of significance (Table 4.2):  Coiled-coil domain containing 26 (CCDC26) (p=3.81x10-5);  Syntrophin, gamma 1 (SNTG1) (p=6.78x10-5);  Glutamate receptor, ionotropic, kainate 1 (GRIK1) (p=0.0013);  Dopachrome tautomerase (DCT) (p=0.002081);  BTB (POZ) domain containing 12 (BTBD12) (p=0.003);  Sarcoglycan, delta (SGCD) (p=0.003);  Roundabout, axon guidance receptor, homolog 1 (ROBO1) (p=0.003);  Protein tyrosine phosphatase, receptor type, D (PTPRD) (p=0.004);  Cystic fibrosis transmembrane conductance regulator (CFTR) (p=0.0046);  Zinc finger protein 607 (ZNF607) (p=0.004);  Latrophilin 2 (LPHN2) (p=0.005).

Descriptions of the identified genes are also presented in Table 4.2. For SNPs variants which were only present in one group, PLINK were unable to produce odds ratios and confidence intervals.

Table 4.2: Significant SNPs (FDR Correction Value < 0.005) associated with paclitaxel response at the lower dose group of 10-6 M along with statistical information. Significant SNPs were traced back to their genes using dbSNP Build 36.3 in order to identify which genes they belonged to. (CHR = chromosome loci, Marker = SNP ID number on Affymetrix 125k chip, RefSNP_ID = SNP rs number, A1 = Variant allele showing difference, A2 = Alternate variant allele, F_A = frequency of A1 in resistant cell lines, F_U = frequency of A1 in sensitive cell lines, Unadj = unadjusted p value, Bonf = Bonferroni correction value, FDR_BH = FDR correction value, ChiSQ = Cochran-Mantel-Haenszel (CMH) statistic, OR = CMH odds ratio, CI= confidence interval, with Lower Bound 95% on confidence interval for CMH odds ratio and Upper Bound 95% on confidence interval for CMH odds ratio). CMH Statistics that were unavailable due to the allele not being present in either the resistant or sensitive group are listed as NA.

CHR Marker RefSNP_ID Name A1 F_A F_U A2 UNADJ BONF FDR_BH CHISQ OR CI 3.741; 1p31.1 91699 rs371363 LPHN2 A 0.5 0.0714 G 2.58E-06 0.2051 0.00477 22.11 13 45.17 3.975; 2q14 1413956 rs1898705 A 0.571 0.0851 G 1.76E-06 0.1405 0.003902 22.84 14.33 51.69 5.296; 2q14.3 1421343 C 0.438 0.0313 A 1.32E-07 0.01052 0.001169 27.84 24.11 109.8 2q31.2 1481588 T 0.25 0 C 1.02E-06 0.08144 0.003192 23.88 NA NA 4.721; 2q33.1 1503580 A 0.5 0.0375 G 1.26E-06 0.1006 0.003354 23.48 25.67 139.5

2q35 1523870 rs6739040 A 0.286 0 G 1.39E-06 0.1104 0.003451 23.3 NA NA 4.902; 2q36.3 1538316 T 0.375 0.0213 C 4.74E-07 0.03773 0.002331 25.37 27.6 155.4 3p24.3 1743877 C 0.286 0 A 2.32E-07 0.0185 0.001542 26.74 NA NA 3p14.2 1785132 G 0.333 0.0106 C 6.86E-07 0.05459 0.002481 24.66 46.5 4.628;

467.2 4.428; 3p12.3 1805457 rs1032966 ROBO1 A 0.333 0.0111 G 1.20E-06 0.09528 0.003354 23.58 44.5 447.3 5.136; 3q26.1 1876142 rs2404571 G 0.667 0.0854 A 2.81E-07 0.02238 0.001721 26.38 21.43 89.41 4.224; 4p15.3 1929743 rs685064 C 0.313 0.0114 T 2.01E-06 0.1602 0.004162 22.58 39.55 370.3 5.274; 5q23.2 2254477 rs959300 G 0.714 0.1111 A 9.98E-08 0.007946 0.001135 28.38 20 75.84 4.06; 5q33.3 2292501 rs7715464 SGCD A 0.563 0.0851 G 1.04E-06 0.08298 0.003192 23.85 13.82 47.05 4.714; 5q33.3 2294131 A 0.313 0.0102 G 5.15E-07 0.04098 0.002331 25.21 44.09 412.4 4.587; 6p25.1 2325528 rs2073042 A 0.786 0.1622 G 1.12E-06 0.08914 0.003301 23.71 18.94 78.25 7.85; 6p22.3 2343860 A 0.917 0.141 C 5.87E-09 0.000467 0.000156 33.88 67 571.8 4.227; 6q14 2403189 A 0.333 0.0116 G 2.09E-06 0.1665 0.004162 22.51 42.5 427.3 4.542; 6q22.1 2440993 rs594930 C 0.4 0.0217 T 1.38E-06 0.1098 0.003451 23.31 30 198.2

4.093; 7q31.2 2626491 rs213988 CFTR G 0.438 0.0444 A 2.03E-06 0.1616 0.004162 22.57 16.72 68.31 8p12 2715222 G 0.25 0 A 1.72E-06 0.1373 0.003902 22.88 NA NA

5.829; 8q11.22 2727769 rs2385525 C 0.429 0.0217 T 7.93E-08 0.006315 0.001052 28.82 33.75 195.4 6.591; 8q11.22 2727770 rs2132528 G 0.5 0.0256 A 3.77E-08 0.003005 0.00063 30.26 38 219.1 4.165; 8q11.22 2727954 T 0.583 0.0778 G 1.75E-06 0.1392 0.003902 22.85 16.6 66.17 8.841; 8q11.22 2728154 rs318885 SNTG1 G 0.6 0.025 T 1.70E-09 0.000136 6.78E-05 36.29 58.5 387.1 8.573; 8q24.21 2806750 rs2568409 CCDC26 G 0.643 0.0455 T 4.79E-10 3.81E-05 3.81E-05 38.76 37.8 166.7 9p23 2835786 G 0.25 0 A 6.07E-07 0.04836 0.002418 24.89 NA NA 4.311; 9p23 2836025 rs7470838 PTPRD T 0.857 0.2245 C 1.47E-06 0.1168 0.003538 23.19 20.73 99.66 4.143; 9p23 2836880 T 0.714 0.1413 C 1.26E-06 0.1005 0.003354 23.48 15.19 55.72 9p22.2 2845463 G 0.25 0 A 7.88E-07 0.06275 0.002728 24.39 NA NA 9p21.2 2856599 rs2060439 G 0.25 0 A 6.07E-07 0.04836 0.002418 24.89 NA NA 5.814; 9q21.33 2897567 G 0.6 0.0488 A 1.24E-07 0.009894 0.001169 27.95 29.25 147.2 4.912; 11q14.1 494182 rs7927911 A 0.429 0.0256 C 8.37E-07 0.06661 0.002775 24.27 28.5 165.4 12q21 629187 G 0.25 0 A 3.61E-07 0.02874 0.002053 25.89 NA NA 4.781; 13q13.3 709113 rs7335400 A 0.563 0.0714 C 1.58E-07 0.01257 0.001251 27.49 16.71 58.43 13q21.33 748668 G 0.857 0.1628 A 3.96E-08 0.003151 0.00063 30.17 30.86 6.213;

153.2 4.812; 13q32.1 779846 rs727299 DCT T 0.313 0.01 C 3.92E-07 0.03122 0.002081 25.73 45 420.8 4.616; 16p13.3 997458 rs714181 BTBD12 A 0.313 0.0104 G 6.76E-07 0.05379 0.002481 24.68 43.18 404 4.257; 16p13.12 1011445 rs251919 C 0.438 0.0385 T 2.41E-06 0.1918 0.004566 22.24 19.44 88.81 4.476; 16q23.3 1071765 G 0.571 0.0745 C 5.27E-07 0.04196 0.002331 25.16 16.57 61.35 4.227; 18p11.22 1162169 rs3975417 T 0.333 0.0116 C 2.09E-06 0.1665 0.004162 22.51 42.5 427.3 4.058; 19q13.12 1266548 rs958305 ZNF607 C 0.5 0.061 T 2.18E-06 0.1738 0.004239 22.43 15.4 58.44 5.504; 21q21.3 1647581 rs457531 GRIK1 T 0.5 0.0349 C 1.73E-07 0.01377 0.001251 27.31 27.67 139.1

Haplotype analyses showed significant haplotypes in a total of five genes [LPHN2

(TTGAGCATCATCTCCCC, pSNP = 2.58E-06 vs. phaplotype = 2.71E-08), PTPRD

(TGGATCCCGT, pSNP = 1.47E-06 vs phaplotype = 4.62E-10), GRIK1 (GT, pSNP = 1.73E-07 vs.

phaplotype = 1.01E-07), ROBO1 (AGGT, pSNP = 1.2E-06 vs. phaplotype = 7.91E-08), and SGCD

(GAC, pSNP = 1.04E-06 vs. phaplotype = 4.12E-07)] (Table 4.3). A figurative illustration of their relative positions among SNPs within the linkage disequilibrium (LD) blocks in the gene is shown in (Figure 4-3).

Table 4.3: Summary of case-control analysis of novel haplotypes associated with drug response at 10-6 dose. This table summarizes the novel haplotypes that were found more significantly associated with drug response based on the case-control breakdown than the originally identified SNP.

Haplotype Case, Control Ratio Case, Control Chi Haplotype Freq Counts Frequencies Square P Value

GRIK1

GC 0.901 9.7 : 6.3, 94.8 : 5.2 0.606, 0.948 18.139 2.05E-05

GT 0.073 6.3 : 9.7, 2.1 : 97.9 0.394, 0.021 28.362 1.01E-07

AT 0.016 0.0 : 16.0, 1.9 : 98.1 0.000, 0.019 0.311 0.5772

ROBO1

GGGT 0.628 4.9 : 7.1, 64.2 : 33.8 0.411, 0.655 2.734 0.0982

GATC 0.189 1.0 : 11.0, 19.8 : 78.2 0.083, 0.202 0.992 0.3193

GGTC 0.131 2.0 : 10.0, 12.4 : 85.6 0.167, 0.127 0.157 0.6918

AGGT 0.043 4.1 : 7.9, 0.6 : 97.4 0.339, 0.006 28.828 7.91E-08

SGCD

GAC 0.629 1.0 : 15.0, 72.0 : 28.0 0.061, 0.720 25.637 4.12E-07

GGC 0.181 5.0 : 11.0, 16.0 : 84.0 0.312, 0.160 2.154 0.1422

AAA 0.109 5.0 : 11.0, 7.7 : 92.3 0.311, 0.077 7.755 0.0054

AAC 0.043 4.0 : 12.0, 0.9 : 99.1 0.251, 0.009 19.847 8.39E-06

GAA 0.038 1.0 : 15.0, 3.4 : 96.6 0.064, 0.034 0.344 0.5576

PTPRD

CGGATCCCGC 0.391 0.0 : 16.0, 45.4 : 54.6 0.000, 0.454 11.931 6.00E-04

TGGATCCCGC 0.14 5.2 : 10.8, 11.0 : 89.0 0.325, 0.110 5.27 0.0217

TGGATCCCGT 0.095 8.3 : 7.7, 2.7 : 97.3 0.518, 0.027 38.832 4.62E-10

CGGGTCCCGC 0.091 0.0 : 16.0, 10.5 : 89.5 0.000, 0.105 1.857 0.173

CGGATACCGC 0.078 0.0 : 16.0, 9.0 : 91.0 0.000, 0.090 1.561 0.2115

CGGATCCTGT 0.043 0.0 : 16.0, 5.0 : 95.0 0.000, 0.050 0.836 0.3605

CGGATCCCGT 0.033 0.5 : 15.5, 3.3 : 96.7 0.032, 0.033 0 0.9864

CGGGTCCCGT 0.018 2.0 : 14.0, 0.0 : 100.0 0.125, 0.000 12.451 4.00E-04

TGAGCCCCGC 0.017 0.0 : 16.0, 2.0 : 98.0 0.000, 0.020 0.326 0.5683

TGGATACCGC 0.017 0.0 : 16.0, 2.0 : 98.0 0.000, 0.020 0.326 0.5683

TGGATCCTGC 0.017 0.0 : 16.0, 2.0 : 98.0 0.000, 0.020 0.326 0.5683

LPHN2

TCGAACACCGTGCCCCT 0.14 2.0 : 12.0, 13.1 : 80.9 0.143, 0.139 0.001 0.9701

TTGAGCATCATCTCCCC 0.12 8.0 : 6.0, 5.0 : 89.0 0.571, 0.053 30.902 2.71E-08

TTGAACACCGTGCCCTC 0.119 0.0 : 14.0, 12.9 : 81.1 0.000, 0.137 2.174 0.1403

TTAGGTGTCGCGCCCTC 0.111 3.0 : 11.0, 9.0 : 85.0 0.214, 0.095 1.753 0.1855

TTGAGCATCGCGTCCCT 0.087 0.0 : 14.0, 9.4 : 84.5 0.000, 0.100 1.541 0.2144

TTGAACGTAGTCTCCCC 0.064 1.0 : 13.0, 6.0 : 88.0 0.071, 0.063 0.013 0.9082

TTGAACACCGTGCCCCT 0.056 0.0 : 14.0, 6.1 : 87.9 0.000, 0.064 0.955 0.3284

TTGAGCATCGCGCCCTC 0.042 0.0 : 14.0, 4.5 : 89.5 0.000, 0.048 0.7 0.4029

TTAGGTGTCGCGCCCCT 0.028 0.0 : 14.0, 3.0 : 90.9 0.000, 0.032 0.467 0.4945

TTAGGTGTCGCCTCCCC 0.028 0.0 : 14.0, 3.0 : 91.0 0.000, 0.032 0.46 0.4978

TTGAGCATCGCGCTCTC 0.028 0.0 : 14.0, 3.0 : 91.0 0.000, 0.032 0.46 0.4978

TCGAACACCGTGCCCTC 0.026 0.0 : 14.0, 2.8 : 91.2 0.000, 0.030 0.425 0.5146

TTGAACACCGTGTCCTC 0.02 0.0 : 14.0, 2.1 : 91.8 0.000, 0.023 0.326 0.5682

TTGAACACCGTCTCCCC 0.019 0.0 : 14.0, 2.0 : 91.9 0.000, 0.022 0.311 0.577

TTGAGCACCGTGCCTTC 0.019 0.0 : 14.0, 2.0 : 92.0 0.000, 0.021 0.304 0.5817

TTGAGCATCGTGCCCCT 0.011 0.0 : 14.0, 1.2 : 92.8 0.000, 0.012 0.174 0.6768

TTGAACACCGCGTCCCT 0.01 0.0 : 14.0, 1.1 : 92.9 0.000, 0.012 0.168 0.6823

TTGAGCATCGCGTCCTT 0.01 0.0 : 14.0, 1.1 : 92.9 0.000, 0.012 0.166 0.6838

Figure 4-3: Results of Haploview analysis to identify haplotypes more significantly associated with drug response than the originally identified SNP. Solid lines represent SNPs that are part of the haplotype from the SNP block whereas dashed lines represent SNPs that are not part of the haplotype. Each square displays the amount of LD between a pair of markers. The strength of LD is given by the numeric annotation and the intensity of the color of a box: black- the strongest, white- no LD.

4.2.3.2 In silico prediction:

Among the 11 significant SNPs found to be linked to protein coding genes, 8 could be analyzed using FastSNP, while no data was available on the other three (PTPRD, CCDC26, SNTG1). Five of these eight SNPs were found to be located within the intronic sequences acting as potential intronic enhancers (GRIK1, DCT, SGCD, CFTR, and PTPRD). Their predicted changes in transcription factor binding are listed in Table 4.4. The ZNF607 and BTBD12 variants were found to cause conservative missense mutations with benign effects on protein structure. SNPs belonging to ROBO1, and LPHN2 were located within intronic sequences, but were not predicted to have a known function.

Using Ingenuity, we have identified other key proteins and pathways that may be involved in forming the cellular response to paclitaxel (Figure 4-4). Some of our newly identified candidates were found to be involved in classical tumor suppressor pathways [TP53, CTNNB1 (β-catenin)], growth pathways (MAPK, ERBB2), and microRNA regulation. However, some of the genes identified were not related to any of the aforementioned pathways, identified genes or to paclitaxel response (CCDC26 and BTBD12). From this figure, we can classify our identified genes into three groups: genes interacting with the p53 and β-catenin axis and paclitaxel response, genes that interact with microtubules to alter paclitaxel response, and other genes that do not belong to either group.

Table 4.4: Potential changes in transcription factor binding caused by SNP variants. This table lists the changes in transcription factor binding predicted by FastSNP for the identified putative intronic enhancers and their associated scores when predicting the likeliness of binding with the specific variant. N/A represents no transcription factor binding. Gene Variant 1 Variant 2 (SNP number) Allele Transcription Score Allele Transcription Score Factor Factor DCT (rs727299) A CdxA 92.9 G N/A N/A Sox-5 89.5 SGCD (rs7715464) G N/A N/A A C/EBPb 88.1 CdxA 85.0 GRIK1 (rs457531) T Sox-5 100.0 C deltaE 88.7 SRY 88.3 CFTR C C/EBP 86.9 T GATA-3 87.5 (rs213988) CDP CR 85.4 GATA-X 86.5 GATA-1 85.9 PTPRD A Sox-5 92.2 G entry 0 (rs7470838) HFH-2 90.1 SRY 90.0 HNF-3b 88.4

100

Figure 4-4: Interaction between our candidate genes with paclitaxel and other genes belonging to relevant pathways. (Red =candidate genes). Directed arrows indicates either changes in expression, activation or phosphorylation status of the target gene by the gene behind the arrow, while solid lines indicate associations or protein-protein interactions between genes on opposite ends of the line. Green lines represent direct links with p53, while red lines represent direct links to β-catenin.

101

4.2.3.3 Gene Expression Studies

Using mRNA expression data as measured using the Affymetrix U133A 2.0 Array, we were able to investigate the differential expression in 8 of our genes over 20 probes. 15 probes across 5 genes (CFTR, DCT, GRIK1, SGCD, and STNG1) showed a significant decrease in mRNA expression in cell lines with resistant phenotype compared to sensitive cases (Table 4.5, Figure 4-5).

Figure 4-5: mRNA expression in NCI60 cells in 8 genes on Affymetrix U133A chip (only significant results shown, p<0.05) as determined by Mann-Whitney test. Green bars represent expression in sensitive cell lines, orange bars – resistant cell lines. Error bars shows 95%confidence interval. Several Affymetrix probes has been labeled in numerical order (1, 2, 3). Y axis shows mean mRNA expression, x axis corresponds to the gene probes.

102

Table 4.5: mRNA expression: Cases vs. Controls were analyzed by Mann-Whitney test using data on the NCI60 panel available from BioGPS system. Genes which did not have any available probes to measure their expression are not listed. Expression differences between sensitive and resistant cell lines were considered to be significant if p < 0.05. Gene Std. (Affymetrix Error Mean difference p-value (Mann- N of cell Probe) Phenotype Mean (resistant- sensitive) Whitney test) lines

CFTR sensitive 10.61 -12.43 0.027* 46 (215702_s_at) resistant 0.18 8 CFTR sensitive 5.87 -8.20 0.008* 46 (215703_at) resistant 0.65 8 CFTR sensitive 14.45 -17.38 0.006* 46 (205043_at) resistant 0.04 8 CFTR sensitive 0.22 -0.87 0.011* 46 (217026_at) resistant 0.26 8 DCT sensitive 764.46 -1884.67 0.005* 46 (205337_at) resistant 0.13 8 DCT sensitive 13.03 -18.58 0.003* 46 (216513_at) resistant 0.16 8 DCT sensitive 508.07 -1085.37 0.004* 46 (216512_s_at) resistant 0.02 8 DCT sensitive 500.48 -1090.40 0.008* 46 (205338_s_at) resistant 0.02 8 GRIK1 sensitive 0.36 -1.31 0.022* 46 (207242_s_at) resistant 0.36 8 GRIK1 sensitive 0.21 -0.81 0.006* 46 (214611_at) resistant 0.19 8 LPHN2 sensitive 16.13 83.57 0.641 46 (206953_s_at) resistant 65.17 8 PTPRD sensitive 0.13 4.78 0.711 46 (205712_at) resistant 4.61 8

103

PTPRD sensitive 0.19 2.68 0.747 46 (213362_at) resistant 2.97 8 PTPRD sensitive 0.55 5.58 0.711 46 (214043_at) resistant 5.95 8 SGCD sensitive 3.91 -16.46 0.002* 46 (210329_s_at) resistant 1.76 8 SGCD sensitive 0.76 -1.71 0.002* 46 (214492_at) resistant 0.07 8 SGCD sensitive 1.13 -1.87 0.008* 46 (213543_at) resistant 0.13 8 SGCD sensitive 0.42 -1.61 0.004* 46 (210330_at) resistant 0.37 8 ROBO1 sensitive 98.67 11.74 0.526 46 (213194_at) resistant 134.71 8 SNTG1 sensitive 0.33 -0.88 0.013* 46 (220405_at) resistant 0.20 8

104

4.3 Discussion

4.3.1 Genome-Wide Method and Application

In this study, we have built upon a previous methodology developed by our group using the NCI60 cell line panel to identify genetic markers associated with drug response (50). Similar approaches have been later applied by a different group to perifosine (172). By applying this methodology to the entire human genome, we were able to identify novel variants that were not originally thought to be related to the drug‟s resistance. Although many studies have looked into the candidate-gene based or cDNA changes associated with paclitaxel response, this is the first genome-wide association study done for SNPs associated with paclitaxel response (173-176). We believe that SNPs serve as a tool to detect changes in genetic targets that are not detectable at the mRNA level.

One of the main features of this study is the application of clinically relevant doses of paclitaxel to determine drug resistance in cell lines. Although the concentration applied during the determination of the GI50 values (10-6 M) was above the therapeutic concentrations normally administered, the actual drug concentration at GI50% (Table 4.1) ranged between 0.104 nM and 7.96 µM. Provided that the therapeutic concentration range is within 100nM to 10 µM, these concentrations fall within this range (177). Furthermore, this is below the maximum tolerated dose, 9.02 µM (178). Under this dose, our GWAS identified 43 significantly associated SNPs with 11 of them belonged to protein-coding genes: ROBO1, SGCD, SNTG1, CCDC26, DCT, BTBD12, ZNF607, GRIK1, CFTR, LPHN2, and PTPRD. These are collectively involved in pathways belonging to cellular stress regulators, cell growth, cell death, and/or paclitaxel response as shown by Ingenuity.

4.3.2 Genetic variations and Haplotypes

Using FastSNP, the majority of these SNPs were found to lie in intronic enhancer regions of their respective genes (SGCD, DCT, GRIK1, and CFTR). All of them were associated with altered mRNA expression, particularly, with a significant increase in mRNA expression levels among sensitive cell lines (Table 4.5). It is also possible that they may alter transcription factor binding which in turn affects downstream genes (170). Similarly, it is unclear how the remaining

105 intronic variants: ROBO1, LPHN2 and KIAA0427 may affect drug response or alter the gene product. For BTBD12 and ZNF607, the predicted effect of conservative missense mutations would not have led to major changes in gene function. Lack of changes in expression level further underscores the importance of using genetic polymorphisms to identify novel genes for drug treatment in order to avoid overlooking of genes in other approaches. However, these predicted changes are still preliminary and need further validation.

Haplotypes are the informative way to detect SNPs linked to each other and working in concert as response predictors in the same gene or between different genes (179-182). A total of 5 genes had such haplotypes in our study. Further evidence for SNP-SNP interactions in determining drug response comes from different LD blocks that were identified in a number of our genes (data not shown). These LD blocks contain SNPs that were significantly linked to our original candidate SNP from the same gene. These SNPs may affect the gene product in a similar way, however would remain unidentified otherwise. Thus, combinations of SNPs may determine drug efficacy in a more powerful way enabling stratification of patients into different response groups and may be better biomarkers in predicting drug response.

4.3.3 Drug Mechanisms and Metabolism

Upon entering the body, paclitaxel is quickly bound by plasma proteins, which maintain the levels of unbound (active) paclitaxel in the plasma (90, 183). Upon reaching its cellular target, paclitaxel diffuses through the cell membrane and binds to a hydrophobic pocket found on β- tubulin (90). The major mechanisms for metabolism and clearance of paclitaxel are through oxidative metabolism involved with the CYP450 group of enzymes (173). SNPs found in the genes coding drug metabolizing enzymes (CYP2C8, CYP3A4/5 and CYP2C19), and drug transport proteins [P-glycoprotein (MDR1) and SLCO1B3] have an impact on paclitaxel clearance (167, 173, 176). These pharmacokinetic variants were unlikely to be identified in our in vitro model as pharmacokinetics plays a larger role in drug response in vivo.

4.3.4 Previous and Novel Markers

Our model has demonstrated the close interaction among β-catenin, p53 and their interacting proteins in the cellular response to paclitaxel. None of our novel candidates were previously associated with paclitaxel response, however many of them were found to interact with

106 previously linked genes and pathways. Our candidates can be classified to interact with paclitaxel either through the β-catenin and p53 transcriptional regulatory axis, microtubule interactions, or through other potential mechanisms.

4.3.4.1 β-catenin and p53 axis and paclitaxel response

Expression and localization of β-catenin is mainly regulated by the canonical Wnt signaling pathway. β-catenin plays a role in cell growth/morphogenesis, cell polarity and cell adhesion (184). Importantly, point mutations in β-catenin, APC, and AXIN have also been documented in cancer, underscoring the involvement of abnormal activation of this pathway in human tumors (185-187).

β-catenin induces expression of p53 and related candidates, which in turn regulates its transcriptional activity (188, 189). The tumor suppressor p53 is one of the most commonly mutated genes in cancer and has been implicated in paclitaxel chemosensitivity (90). Paclitaxel increases p53 expression, phosphorylation and its nuclear accumulation leading to the arrest of cell growth and apoptosis (190-192). Among the genes activated by β-catenin are such genes as Cyclin D1 (CCND1), Survivin (BIRC5) that were previously found required for paclitaxel response, and MYC, which also have crucial roles in cell growth, proliferation and differentiation (90, 187). MYC induced by β-catenin may also induce expression of p53, and p53 up-regulates CDKN and RBL2, resulting in growth arrest (193-195). It is believed that the BCL2 phosphorylation by RAF-1 kinase constitutes the major signaling pathway in paclitaxel-induced apoptosis (196). Both ERRB2 and Src increase tyrosine phosphorylation of β-catenin affecting its interactions with several proteins including p53, E-cadherin (CDH1) and PTPRD (197-201). For example, Src increases the binding of CDH1 and phosphorylated β-catenin (198). p53 also increases the expression of CDH1, which in turn also binds with β-catenin and decreases its transcriptional activity (197). PTPRA and PTPRD are receptor tyrosine phosphatases, which both directly interact with SRC (202, 203). p53 decreases expression of PTPRA which interacts with PTPRD (204, 205). PTPRD also dephosphorylates Src, which can alter RAF-1 kinase activation (202, 206).

107

4.3.4.2 Microtubule Interactions and paclitaxel response

Paclitaxel arrests cell growth and causes apoptosis as a microtubule stabilizing agent. Paclitaxel can only interact with β-tubulin when it is present as assembled tubulin (90). This leads to antiproliferative activity and can be caused by nanomolar concentrations of paclitaxel (90). Interaction of paclitaxel with microtubules is affected when cells that lack p53 have either increased point mutations in tubulin, which weaken drug-tubulin interactions, or altered expression of MAP4 protein (90, 166). At concentrations less than 10nM, microtubule dynamics are suppressed without affecting microtubule mass, while at higher concentrations, microtubule masses are formed (90). Continued increase in concentrations lead to G2/M cell cycle arrest and apoptosis (90). The effects observed by this magnitude of paclitaxel concentration are consistent with the magnitude of the GI50 values obtained from DTP.

Variants of our several novel candidates may also affect drug response through alterations in microtubular interactions. LPHN2 is a G-protein coupled receptor that interacts with tubulin through binding with G proteins, and with β-catenin through MAGI2 (207-209). LPNH2, which has been associated with breast cancer, causes the mass release of neurotransmitters involved in the microtubular transport upon binding to one of its ligands, α- Latrotoxin, a potent black-widow spider toxin (210). GRIK1 is a kainite receptor used in interneuronal communication and whose expression is altered by β-catenin (211, 212). GRIK1 interacts with microtubules in order to get transported to dendrites (213). Also, GRIK1 expression is regulated by ATN1, which also regulates STMN1 that reduces cell binding of paclitaxel (214, 215). Another one of our candidates, which serves as a neuronal receptor, is ROBO1. When activated by its ligand Slit, ROBO1 directly interacts with microtubules and specifically the beta 4 tubulin (TUBB4) subunit to regulate their dynamics (216-218). In addition, p53 increases expression of ROBO1 by direct interaction with it (219).

CFTR, an ABC chloride transporter channel, is required for proper microtubule distribution, whereby, mutant CFTR aggregates, as in cystic fibrosis, lead to microtubule disorganization (220). CFTR also interacts with CANX and DNAJA1 which establishes its link with DCT and ZNF607, respectively (221-224). β-catenin trans-activates DCT, which is used in the synthesis of melanin and is associated with chemoresistance to DNA-damaging agents such as radiation (225, 226). However, no correlation between DCT expression levels and paclitaxel

108 resistance has been observed in previous studies (225, 227). An observed highly significant association of the intragenic SNP within DCT is most likely due to the melanoma lines allocated to the sensitive group as DCT expression is a marker for melanomas (228). ZNF607, a zinc finger protein involved in transcriptional regulation, interacts with β-catenin through a developmental protein, MyoD family inhibitor, MDFI (223, 229). ZNF607 also interacts with CFTR via ATXN1 and DNAJA1 (223, 224). ATXN1 interacts with microtubules to alter cell morphology during neuronal development, while MDFI is important for myogenesis and microtubules play an important role in this process (218, 229, 230).

4.3.4.3 Candidates that do not interact with paclitaxel through known p53, β-catenin or microtubule interactions

SGCD is a member of a group of proteins that form the sarcoglycan complex (SGC) which is part of a larger complex called the dystrophin associated glycoprotein complex (DAGC) (231). The beta isoform of sarcoglycan, SGCB, part of the sarcoglycan complex core along with SGCD, co-localizes with tubulin during the M phase of the cell cycle (232). Given that paclitaxel also interacts with tubulin during M phase, SGCD polymorphisms may alter paclitaxel response (90, 176, 232). Another component of the DAGC is syntrophins whose family includes another candidate that we have identified, SNTG1 (231). β2-syntrophin, another member of this family, is associated with microtubules via PDZ domain on the microtubule associated proteins, MAST205 and SAST (233). SNTG1 shares this PDZ domain and may similarly interact with microtubules, affecting paclitaxel response (234). KIAA0427 has been classified by Gene Ontology as playing a role in RNA catabolism, protein binding, and translational initiation. It is commonly targeted with β-catenin, by microRNA, MIRN328. It is commonly mutated in breast and colorectal cancers and found phosphorylated during cell division (235, 236). Its expression increases through mutant p53 transactivation (237). Further studies are required to understand its role in paclitaxel response.

BTBD12 is a relatively uncharacterized gene, but recently it has been found to recruit different parts of DNA repair pathways used for double stranded breaks (238, 239). Some of paclitaxel‟s downstream effects include DNA damage and repression of various DNA repair genes leading to cytotoxicity (240). The missense mutation caused by our variant could reduce BTBD12‟s ability to repair DNA leading to increased paclitaxel susceptibility. CCDC26 is a

109 retinoic acid (RA) signaling modulator that has been associated with gliomas from the genome- wide studies (241, 242). Common to both retinoic acid signaling and paclitaxel is BCL2, which is a key player in paclitaxel mechanism. Observations on the effect of retinoic acid on BCL2 have varied, but in the breast tumor cell line, MCF-7, RA decreases BCL2 expression (243). Hence, reduced retinoic acid signaling modulation can decrease BCL2 expression, leading to chemosensitivity in breast tumor cells.

4.4 Conclusions

Our study has taken advantage of genotypic data and its integration with drug response data available for the NCI60 panel. We realize that this panel may not be an accurate representation of a patient population as cells in vivo may behave differently from cells in vitro. Furthermore, we are using a cell-based pharmacological approach to identify novel SNPs associated with drug response instead of a systems-based approach. However, this has helped to simplify the consequences of pharmacokinetic parameters and in vivo effects to focus on the drug target specifically. We have identified a number of novel variants belonging to genes that are associated with variation in drug response. By using a series of bioinformatic and literature investigations, we have been able to infer how these variations might give rise to the variation in paclitaxel response observed in the patient population. Many of these novel variants may not have been detected if only changes in mRNA expression level were investigated as only five of our candidates showed significant changes in expression levels between sensitive and resistant cell lines. These genetic variations represent promising biomarkers to predict paclitaxel resistance in cancer patients and can serve as targets for future functional and mechanistic studies on paclitaxel response.

Chapter 5 Candidate genes and pathways associated with resistance to doxorubicin and paclitaxel

In this chapter, to build upon knowledge about genes and molecular pathways involved in the resistant phenotypes in cancer treatment, we used an integrative approach to cluster identified pathways with overlapping genes likely to constitute common biological processes in resistance to doxorubicin and paclitaxel. We constructed a gene-based interactome of signal transduction pathways involved in the resistance to both chemotherapeutic agents. Pathway enrichment analysis of the 10,438 SNPs significantly associated with resistance to doxorubicin and paclitaxel in Genome-Wide Analysis (5240 and 5198 SNPs, respectively) was used to identify the top biological modules involved in the cross-talk. Based on those findings, we investigated the biological role of candidate genes in these pathways. In addition, the role of candidate genes was further validated in breast cancer patients treated with doxorubicin and paclitaxel. This has been achieved by comparing the expression level of genes in patients‟ samples and further correlating it with the clinical outcomes (either pathological complete response or survival rate). Consequently, the NRF2-associated oxidative stress pathway was a key network in resistance to both drugs. Eleven out of 16 genes are actively involved in this pathway either by interacting with reactive oxygen species or through other genes. FRMD6, CFTR, SGCD, LPHN2 and SNTG genes demonstrated an association with better outcomes in breast cancer patients treated with doxorubicin and paclitaxel.

110

111

5 Cross-talk between doxorubicin and paclitaxel activated resistant pathways

5.1 Introduction

It is generally believed that signal transduction pathways act as a coordinated response between sets of independent proteins to produce particular phenotypic responses. Previous findings have showed the importance of candidate genes discovered from GWAS in drug resistance.

Although paclitaxel and doxorubicin do not cause cross-resistance and have a different mechanism of action, there are many common intracellular end-targets of their cytotoxic effects. Taxol showed an increased expression of human superoxide dismutase 2 (SOD2) mRNA as a result of a protein-kinase C delta-dependent effect (244). It can induce lipopolysaccharide (LPS)-like effects enhancing induction of nitric oxide synthase and production of nitric oxide (245-247). Taxol also downregulates tumor necrosis factor receptors (TNF) and inhibits synthesis of TNF (244). Doxorubicin, in its turn, produces oxidative damage after it is reduced to semiquinone radicals by NADPH oxidase. It is mediated by the doxorubicin-iron complex (248, 249). In previous reports, scientists established that high nuclear factor-(erythroid derived-2)-like 2 (NFE2L2) activity and increased level of GSH-synthesizing enzyme were correlated with doxorubicin resistance in A2780DR cell lines (250).

In light of considerable overlap between downstream pathways of the genes identified from our GWAS as involved in the resistance to doxorubicin and paclitaxel, we hypothesized that common biological modules may underlie resistance to both drugs of interest.

Oxidative stress is a hallmark of many antineoplastic agents and cancer cells. Usually, it occurs as a result of imbalance between the production of reactive oxygen species (ROS) and the cellular antioxidant capacity. Even though it can arise from endogenously released intermediates of metabolic processes, it is most commonly caused by exogenous factors. Free radicals play an important role in the host defense reaction and as an intracellular mediator in cell proliferation, differentiation and gene expression. They are commonly represented by the reactive oxygen and nitrogen species, nitric oxide, peroxide, OH-, O2-, and peroxynitrite (ONOO-). Reactive oxygen species are not always damaging and are an integral part of the response to pathogens.

112

Mitochondrial respiration constitutively produces ROS to balance every metabolism (251). A variety of stress factors can induce oxidative stress pathways intracellularly. Many antioxidant genes and enzymes are regulated by NFE2L2, a key transcriptional activator that binds to the antioxidant response element (ARE) in the enhancer region. This regulation determines cell response to oxidative distress and its extent such as production of glutathione S-transferases (GSTs) and NADPH quinine oxidoreductase. ROS targets Nrf2-Kelch-like ECH associating protein 1 (Keap1) complex which leads to the dissociation of NFE2L2 with subsequent translocation to the nuclei (252). Increased level of the antioxidant defense in cancer cells was observed previously in resistant tumors.

Under normal conditions, an ability to defend against excessive generation of ROS is one of the tumor-suppressing mechanisms of p53. However, the actual role of p53 and oxidative stress still remains a matter of controversy due to the complexity of its nature (251). The first clear connection between p53 and oxidative stress has been shown by Polyak K. et al. (253). p53 up- regulates pro-oxidant enzymes that lead to the generation of ROS in mitochondria from an electron transport chain (ETC) and activation of apoptosis. In addition, p53 also increases ROS production as a suppressor of antioxidant genes. An imbalanced induction of pro-oxidant genes and antioxidant enzymes is most probably responsible for an increased generation of free radicals and p53-induced oxidative stress (251). More in-depth investigations indicated other mechanisms of generation of ROS by p53 such as control of transcription of genes involved in mitochondrial respiration, glycolysis, and the pentose phosphate shunt. There was also discovered an existence of a so-called redox-active regulation loop between the basal levels of cellular ROS and p53, including thioredoxin (TRX) reductase (TRP), and redox factor-1(Ref-1) (254). Sayan et al. discovered a positive feedback loop based on a transcription-independent (i.e. without activation of gene transcription of protein synthesis) apoptotic function of p53 as a direct target of cleavage by caspases. An accumulation of p53 induces caspase activation which in turn leads to p53 cleavage and relocation of the cleaved derivatives of p53 into mitochondria (255). Approximately 2% of cellular p53 traffics to mitochondria under oxidative stress conditions augmenting mitochondrial membrane depolarization and reactive oxygen generation (255). Its rapid accumulation in mitochondria precedes cytochrome c release, ROS formation, and caspase- 3 activation. Conceivably, it represents an efficient extranuclear apoptotic mechanism. Furthermore, in vivo validation of these findings showed that redirection of p53 to mitochondria

113 induces apoptosis in p53-deficient states (256). However, there is a fine line that defines a role of p53 in oxidative stress pathways: low dose of p53 carries an antioxidant role, whereas the high dose of p53 that is observed in severe stress enhances production of ROS formation and apoptosis (257).

To explore the hypothesis, we have looked at the cross-talk of the downstream pathways associated with resistance to doxorubicin and paclitaxel. In addition, we performed a pilot clinical validation based on the publicly available microarray studies from patients‟ tumor samples.

5.2 Methods:

5.2.1 Pathway analysis

To test the similarities among regulatory pathways and genes associated with decreased response to doxorubicin and paclitaxel, we applied an integrative bioinformatic approach. Initially, we aimed to identify if there is any interacting pathways associated with resistance to doxorubicin and paclitaxel. Subsequently, the next aim was to functionally annotate candidate genes previously described in chapters 3&4 with regard to the revealed pathways. DAVID Functional Annotation Tool (http://david.abcc.ncifcrf.gov/) (258, 259) was used to annotate candidate genes by GO terms. Ingenuity Pathway Analysis (IPA) software (Ingenuity Systems, www.ingenuity.com) was used to perform both tasks. In the Ingenuity Pathway Analysis (IPA) software gene/SNPs sets can be functionally annotated and biologically relevant pathways may be identified. A pathway Core Analysis of 10,438 SNPs was done in the context of biological processes, pathways and molecular networks. Overrepresented biological pathways were used for the subsequent analysis of substantial interactions among candidate genes and their cross- talking partners. To obtain updated functional annotation of candidate genes (DSG1, RORA, FRDM6, GRIK1, ROBO1, CFTR, PTPRD, KIAA0427, BTBD12, CCDC26, DCT, SNTG1, SGCD, LPHN2, and ZNF607) and the corresponding pathways, we integrated previously published information. The goal of the annotation was to establish the connections if any between genes associated with the response to doxorubicin and paclitaxel.

114

5.2.2 Pilot Clinical Validation

Exploratory clinical validation was generated from published microarray datasets from Array Express (http://www.ebi.ac.uk). Whole-genome mRNA expression data in 115 primary breast tumors (74 adriamycin-containing regimes and 28 none), and in 230 breast patients with 278 tumor samples [232 paclitaxel, 5-fluorouracil, cyclophosphamide and doxorubicin (T/FAC) and 40 paclitaxel, 5-fluorouracil, cyclophosphamide and epirubicin (T/FEC)] were collected and analyzed from Array Express and Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo) [accession numbers: GSE19615 (105) and GSE20194]. Expression levels of candidate genes in tumors from women with and without recurrence were compared. GSE19615 were used to perform time-to-recurrence analysis and ANOVA analysis for three genes (FRMD6, RORA, and DSG1) associated with doxorubicin resistance. GSE20194 dataset was used to analyze association with treatment response in 11 genes (RORA, LPHN2, PTPRD, FRMD6, GRIK1, ROBO1, DSG1, SNTG1, SGCD, CFTR, and DCT).

5.2.3 Statistical Analysis

Statistical analyses were carried out using SPSS v. 13.0 (SPSS< Chicago, IL, USA). We applied independent 2-samples t-test to assess the association between candidate gene expression (continuous variables) and primary outcome in breast tumors from MAQS-II dataset. Outcomes with a p value < 0.05 were considered statistically significant. Analysis of variance (ANOVA) model was used to compare single-gene signatures between tumors from patients with and without recurrence with ER, PR, and HER2-receptor status as co-variables.

Survival analysis: Kaplan-Meier survival analysis was performed to estimate the significance of the survival difference between randomized treatment groups with the log-rank test. The alpha level was 0.05. In order to analyze the value of a particular gene, the cohort was divided into two groups based on the median expression of the gene. These two groups were then compared by a Kaplan-Meier plot on SPSS 13.0 with the stratification based on a treatment class. In addition, the results were adjusted by ER and HER2 status.

115

5.3 Results

5.3.1 Pathway analysis

Several pathways have been found to be significantly associated with resistance to doxorubicin and paclitaxel in the pool of 10,438 SNPs mapped to gene networks. Identified overrepresented pathways showed profound interactions with the following major canonical pathways that are, (1) PPAR /RXR, RAR activation, (2) Anti-apoptosis and p53 signaling, and (3) TGF- signaling (Figure 5-1). Analysis of molecular and cellular functions of the mapped SNPs identified 5 top functions (Table 5.1), including (1) cell morphology, (2) cell-to-cell signaling and interaction, (3) cellular movement, (4) cellular development, (5) cellular assembly and organization. An association analysis of molecules by pathways demonstrated strong relationship with (1) estrogen-calcium-breast cancer, (2) cancer pathways, and (3) NRF2-mediated Oxidative Stress Response (p<0.05) (Figure 5-2).

Table 5.1: Top molecular and cellular functions (by DAVID Functional Annotation Tool)

Name p-value # Molecules

Cell Morphology 1.83E-09 - 1.03E-02 188

Cell-to-Cell Signaling and Interaction 2.08E-07 - 1.05E-02 239

Cellular Movement 1.31E-06 - 1.06E-02 244

Cellular development 1.56E-06 - 8.99E-03 217

Cellular Assembly and Organization 4.30E-06 - 1.10E-02 182

116

Figure 5-1: Top pathways associated with the doxorubicin and paclitaxel resistance in NCI60 cell lines. Blue bars indicate pathway significance; orange squares indicate the ratio of the number of significant SNPs to the total number of SNPs in the pathway.

117

Figure 5-2: Top pathways linked to the SNPs associated with doxorubicin and paclitaxel resistance. Pathway analysis was done on Ingenuity based on the SNPs significantly associated with resistance to doxorubicin and paclitaxel. Numbers on the right indicate number of genes in the pathway. Shaded grey area corresponds to the number of SNPs overlapping with dataset in the analysis. Orange squares labels p-value.

118

A functional pathway analysis (by GO terms) of 16 candidate genes by DAVID Functional Annotation Tool (http://david.abcc.ncifcrf.gov/) (258, 259) identified plasma membrane location as the top of the list of associations in 9 genes (Table 5.2). Eight of them had both extracellular and cytoplasmic domains (Table 5.2). Six genes (GRIK1, PTPRD, DSG1, CFTR, LPHN2, ROBO1) (258) were involved in transmembrane transport. Pathway analysis shows convergence of 11 out of 16 candidate genes (DSG1, RORA, FRMD6, ROBO1, SNTG1, PTPRD, CFTR, SGCD, LPHN2, GRIK1, and DCT) on the NRF2-mediated Oxidative Stress response pathway. Redox signaling is an essential process in integration of growth factors and cell anchoring mechanisms (260). Further exploration of the mechanistic role of candidate genes in the oxidative stress pathway was done by means of construction of a biological interactome. Several genes were in direct interaction with ROS through the genes in the oxidative stress response network, including RORA, FRMD6, PTPRD, CFTR, and ROBO1. The rest of the genes were observed in the indirect interaction with ROS and oxidative stress pathways, namely DSG1, GRIK1, LPHN2, SNTG1, DCT and SGCD. Among all candidate genes, RORA, CFTR, and FRMD6 had the most numerous connection points with the members of oxidative stress network. In addition, several genes showed interaction not only with ROS but also with the NRF2 (NF- E2-related factor) thereby potentially influencing the NRF2-mediated oxidative stress pathway (RORA, FRMD6, DCT, TP53, CTNNB1, and CFTR). RORA, which has the most extensive connections with the Nrf2-oxidative stress pathway and ROS, forms a protein-protein complex with its p300, a co-factor that transactivates Nrf2 itself (164). CFTR is associated with HSPA9, heatshock protein 70k protein 9 (224) up-regulated by Nrf2 in stress (261).

It should be noted, however, that this in silico analysis may not represent the actual causal relationship between genes. Nevertheless, despite its limitations, this preliminary exploratory analysis suggests a connection of several well-known canonical pathways and classical tumor suppressor genes.

119

Table 5.2: Functional pathway annotation by DAVID (Enrichment Score: 1.9333685013839974): plasma membrane. Candidate genes identified from our GWAS as associated with response to doxorubicin- or paclitaxel-based treatment were evaluated for functional annotation.

Term Fold Count % P Value Genes Enrichment Bonferroni SLC2A9,PTPRD,FRMD6,GRIK1,ROBO1,DSG1, 9 56.25 GO:0044459 1.97E-04 SNTG1,SGCD,CFTR 4.01683 0.016607 LPHN2,DCT,SLC2A9,PTPRD,GRIK1,ROBO1, 9 56.25 Cytoplasmic 5.13E-04 DSG1,SGCD,CFTR 3.645631 0.048518 LPHN2,SLC2A9,PTPRD,GRIK1,ROBO1,DSG1, 8 50 Extracellular 9.18E-04 SGCD,CFTR 4.012353 0.085202 LPHN2,SLC2A9,PTPRD,FRMD6,GRIK1,ROBO1, 10 62.5 GO:0005886 0.001493 DSG1,SNTG1,SGCD,CFTR 2.603206 0.119285

N-linked LPHN2,DCT,SLC2A9,PTPRD,GRIK1, 9 56.25 glycosylation 0.002111 ROBO1,DSG1,SGCD,CFTR 2.975623 0.185305 Transmembrane LPHN2,DCT,SLC2A9,PTPRD,GRIK1, 9 56.25 region 0.006778 ROBO1,DSG1,SGCD,CFTR 2.500162 0.48299 LPHN2,DCT,SLC2A9,PTPRD,GRIK1, 9 56.25 Glycoprotien 0.010735 ROBO1,DSG1,SGCD,CFTR 2.430467 0.514759

120

Transmembane LPHN2,DCT,SLC2A9,PTPRD,GRIK1, 9 56.25 0.025482 ROBO1,DSG1,SGCD,CFTR 2.108238 0.82261 Membrane LPHN2,DCT,SLC2A9,PTPRD,FRMD6, 10 62.5 0.030526 GRIK1,ROBO1,DSG1,SGCD,CFTR 1.877328 0.874705

5 31.25 Receptor 0.033449 LPHN2,PTPRD,GRIK1,ROBO1,RORA 3.667919 0.897656

6 37.5 Signal peptide 0.055897 LPHN2,DCT,PTPRD,GRIK1,ROBO1,DSG1 2.510528 0.996225

LPHN2,DCT,SLC2A9,PTPRD,GRIK1,

9 56.25 GO:0016021 0.070306 ROBO1,DSG1,SGCD,CFTR 1.670583 0.997963

LPHN2,DCT,SLC2A9,PTPRD,GRIK1,

9 56.25 GO:0031224 0.085763 ROBO1,DSG1,SGCD,CFTR 1.613323 0.99951

6 37.5 Signal 0.106118 LPHN2,DCT,PTPRD,GRIK1,ROBO1,DSG1 2.142445 0.999456

3 18.75 GO:0007166 0.505132 LPHN2,PTPRD,GRIK1 1.682029 1

121

5.3.2 Clinical Validation:

Several clinical datasets with open-access from the ArrayExpress data repository were analyzed for validation of findings: 1. GSE19615 – 115 samples from breast cancer patients treated with doxorubicin (105) 2. GSE20194- 278 samples from breast cancer patients in MAQS-II dataset on T/FAC or T/FEC regimes

We performed our analysis on 115 breast cancer patients from GSE19615 meeting our criteria. The mean follow-up time in anthracycline-treated group (74 pts) was 60.66±20.965, and 58.68 ±22.527 in non-treated patients (41 pts); 38, 26 and 35 of 115 patients were estrogen-, HER2-, and PR-positive, respectively. Overall, 19/115 had metastasis or died from disease. ANOVA analysis showed that higher expression levels of FRMD6 is significantly correlated with a better response to doxorubicin-based treatment in breast cancer patients (Table 5.3). FRMD6 had a lower expression level in tumors from patients with recurrence compared to those without. RORA and DSG1 demonstrated a higher level of expression although the observed difference was not significant. In ANOVA analysis of 3 genes (RORA, FRMD6, and DSG1) only FRMD6 showed a significant difference between doxorubicin-treated and non-treated patients‟ samples (p=0.005) (Table 5.3). Survival analysis of doxorubicin-associated genes showed significant correlation of time-to-recurrence with FRMD6 gene expression (p=0.046) (Table 5.4, Figure 5-3).

122

Table 5.3: Results from ANOVA test for 4 Affymetrix U133A-plus 2.0 probes from GEO19615 database among 102 breast patients treated with anthracyclines

RORA RORA Gene FRMD6 DSG1 (206642_at) (210426_x_at)

Mean difference between treated and -177.194* 3.14556 3.41565 8.50026 non-treated patients

P value 0.005* 0.999 0.982 0.759

Lower -313.9138 -54.89312 -20.35739 -14.05 Bound 95% Confidence Interval Upper -40.47432 61.18424 27.18869 31.0509 bound

Mean 337.34967 14.63644 105.44122 102.335 With distant STD 188.11002 9.134143 55.108097 44.6756 recurrence N 9 9 9 9

Mean 427.63858 26.52411 90.47094 87.7503 Antracycline- No STD 221.48867 72.028591 41.863298 41.0096 based recurrence N 65 65 65 65

Mean 416.6575 25.07831 92.29165 89.5242

TOTAL STD 218.56291 67.623471 43.514945 41.4272

N 74 74 74 74

Mean 368.06125 100.08575 87.195 82.8353 With distant Non-treated recurrence STD 239.62221 180.18238 27.634995 30.3089

123

N 4 4 4 4

Mean 631.48325 8.90725 89.15617 80.722 No STD 295.0054 3.972926 33.04666 29.9849 recurrence N 24 24 24 24

Mean 593.85154 21.93275 88.876 81.0239

TOTAL STD 298.8755 68.384425 31.86903 29.4709

N 28 28 28 28

124

Table 5.4: Kaplan-Meier survival analysis of FRMD6, RORA, and DSG1 in breast cancer patients from GEO19615 database. Gene Group Low-dose: total High-dose: total Low dose: High dose: LogRank, Mantel- Breslow (Generalized N, (N of events) N, (N of events) Mean ± SE Mean ± SE Cox Wilcoxon) Chi- P Chi- P square square FRMD6 Anthracyclines 38 (2) 36 (7) 83.763 ± 74.361 ± 3.234 0.072 3.967 0.046* -based 2.228 4.660 None 20 (2) 21 (3) 76.050 68.714 ± ±4.692 5.535 RORA Anthracyclines 38 (4) 36 (5) 80.947 ± 77.472 ± .297 .586 .028 .867 (206642_at) -based 3.363 3.975 None 20 (4) 21 (1) 67.750 ± 80.048 ± 6.853 2.881 RORA 39 (3) 35 (6) 81.282 75.829 1.399 .237 1.705 .192 (210426_x_ ±2.642 ±4.539 at) 19 (2) 22 (3) 74.684 71.773 ±5.564 ±4.999 DSG1 38 (4) 36 (5) 79.263 78.222 1.330 .249 .853 .356 ±3.217 ±4.073 20 (1) 21 (4) 78.950 ± 67.762 ± 3.947 6.007

125

Figure 5-3: Survival plot for FRMD6, RORA, and DSG1 in the GEO19615 database: a) in anthracycline-treated breast cancer patients; b) in non-treated breast cancer patients. 1-high expression (blue), 2- low expression (green). Expression groups were defined by creating a binary variable based on the mean of gene expression.

126

We understand a limitation of our approach is using the mean to create a binary variable for high- and low-expressing gene groups. Although categorical variables can be used in Cox proportional hazard models, no survival curves for the high and low-dose groups can be drawn in this case.

A publicly available dataset from The MicroArray Quality Control (MAQC) Consortium project- II provided to ArrayExpress repository by the University of Texas M.D. Anderson Cancer Center (MDACC) Breast Cancer Pharmacogenomic Program was used for the second validation (262). Human breast cancer samples from 230 stage I-III patients were used for RNA extraction and gene expression profiling using Affymetrix U133A microarrays (Table 5.5). Patients received 6 months of neoadjuvant chemotherapy described in Methods. Response was categorized as a pathological complete response (pCR) or residual invasive cancer (RD). Gene expression of pre- treated samples was correlated with primary outcomes (pCR or RD). T-test analysis of 13 genes in 272 samples from the MAQS-II dataset divided by the response identified that underexpression of two transcripts of the SGCD gene were significantly associated with complete pathological response (p=0.026 and 0.001). An association of expression levels with response among the same 13 genes in 232 patients treated with T/FAC revealed LPHN2 (p=0.043, overexpression) and 2 transcripts of SGCD again (p=0.022, and 0.004, underexpression). Addition of receptor status (ER, PR, and HER2- receptors) as a 3rd co-variable added additional genes into the picture. As such, in case of ER-negative, and triple-negative (ER-, PR-, and HER2-) tumors, lower expression level of three genes were significantly associated with better primary outcomes (CFTR, SGCD, and SNTG1) (p value ranges from 0.012 to 0.047) (Table 5.6). The favorable primary outcome in patients with HER2-negative status was significantly associated with higher expression of LPHN2 and lower expression of SGCD genes (p=0.013, and 0.005, respectively). Additionally, low expression of two more genes (ROBO1and DCT) demonstrated borderline association with favorable outcomes (p<0.1), which might be improved with larger sample size.

127

Table 5.5: Descriptive characteristics of the samples from MAQS-II dataset (E-GEOD- 20194)

N of cPR N of RD Total N of samples

56 216 All samples 272

52 180 Patients treated with T/FAC 232

3 21 ER-positive and HER2-positive 24

46 ER-negative and HER2-negative 27 73

25 40 Triple-negative: ER-, PR-, HER2- 65

46 62 ER-negative 108

10 154 ER-positive 164

22 37 HER2-positive 59

34 179 HER2-negative 213

128

Table 5.6: Comparison of gene expression in breast cancer samples treated with T/FAC or T/FEC from MAQS II dataset (E-GEOD- 20194) (as compared by t-test)(* denotes significant results)

Patients treated ER-positive and ER-negative and Triple-negative: ER-, All samples with PFAC HER2-positive HER2-negative PR-, HER2- ER-negative HER2-negative

Gene/Affy Mean, P- Mean, Mean, Mean, Mean, U133A probe P-value SE value SE P-value SE P-value SE P-value Mean, SE P-value SE P-value Mean, SE

CFTR 0.042, 0.031, 0.041, 0.016, 0.08, 0.192, -0.081, 215702_s_at 0.837 0.203 0.887 0.219 0.968 1.002 0.963 0.354 0.835 0.381 0.475 0.268 0.75 0.253

CFTR -0.191, -0.212, 0.752, -0.675, -0.738, -0.488, -0.473, 215703_at 0.469 0.263 0.442 0.276 0.496 1.086 0.127 0.437 0.113 0.46 0.146 0.333 0.152 0.329

CFTR* -0.367, -0.352, -1.572, -0.62, -0.686, -0.499, -0.354, 217026_at 0.065 0.198 0.096 0.21 0.022* 0.64 0.045* 0.304 0.038* 0.324 0.043* 0.244 0.163 0.253

CFTR* -0.231, -0.242, -0.216, -0.634, -0.677, -0.527, -0.23, 205043_at 0.102 0.14 0.099 0.145 0.513 0.324 0.018* 0.262 0.021* 0.285 0.017* 0.218 0.266 0.207

LPHN2* 0.377, 0.494, 0.757, 0.108, 0.113, -0.24, 0.691, 206953_s_at 0.091 0.223 0.043* 0.243 0.356 0.803 0.708 0.289 0.716 0.309 0.361 0.262 0.013* 0.276

129

SGCD* -0.459, -0.508, -0.4, -1.035, -1.113, -0.774, -0.581, 210329_s_at 0.026* 0.203 0.022* 0.218 0.663 0.903 0.009* 0.383 0.008* 0.406 0.012* 0.301 0.05* 0.294

SGCD* -0.274, -0.234, -1.619, -0.579, -0.731, -0.272, -0.389, 210330_at 0.199 0.213 0.314 0.232 0.101 0.946 0.112 0.359 0.042* 0.352 0.346 0.287 0.129 0.256

SGCD * -0.732, -0.703, -0.297, -0.852, -0.719, -0.714, -0.806, 213543_at 0.001* 0.226 0.004* 0.239 0.759 0.957 0.015* 0.341 0.047* 0.354 0.012* 0.278 0.005* 0.283

SNTG1* -0.112, -0.01, -0.191, -0.485, -0.558, -0.342, -0.164, 220405_at 0.469 0.154 0.949 0.163 0.731 0.549 0.043* 0.236 0.028* 0.247 0.079 0.193 0.382 0.187

PTPRD -0.149, -0.094, 0.108, -0.237, -0.291, -0.277, -0.081, 205712_at 0.436 0.191 0.647 0.204 0.881 0.71 0.459 0.318 0.399 0.343 0.3 0.265 0.729 0.233

PTPRD -0.125, -0.143, 0.016, -0.092, -0.191, -0.089, -0.139, 214043_at 0.589 0.231 0.561 0.246 0.985 0.871 0.828 0.421 0.669 0.446 0.773 0.309 0.637 0.295

GRIK1 -0.031, -0.074, -0.107, 0.018, 0.003, 0.082, -0.067, 207242_s_at 0.884 0.214 0.745 0.227 0.882 0.711 0.959 0.341 0.994 0.365 0.765 0.272 0.805 0.269

ROBO1 0.257, 0.221, -0.535, -0.154, -0.259, -0.281, 0.597, 213194_at 0.268 0.231 0.381 0.252 0.669 1.234 0.709 0.409 0.543 0.422 0.349 0.299 0.07 0.321

130

DCT -0.218, -0.139, -0.959, -0.44, -0.522, -0.183, -0.344, 216512_s_at 0.238 0.185 0.491 0.201 0.053 0.469 0.204 0.343 0.141 0.35 0.487 0.262 0.153 0.24

DCT -0.127, -0.133, -0.984, -0.265, -0.165, -0.132, -0.139, 216513_at 0.551 0.213 0.565 0.232 0.328 0.984 0.445 0.345 0.649 0.361 0.638 0.28 0.597 0.262

131

5.4 Discussion:

In this study, we aimed to create a map of genetic interactions involved in resistance to doxorubicin and paclitaxel. It has been achieved by data mining pathways and previously submitted expression datasets. While it would be ideal to identify genetic variations associated with drug resistance in vivo, the current era of genomics has improved the predictive ability of methods based on knowledge of existing genetic interactions between pathways. To demonstrate this feature, we subsequently used the top identified pathway, the NRF2 oxidative stress pathway, to link to the target genes associated with doxorubicin or paclitaxel resistance from GWAS.

One of the most prominent players involved in the oxidative stress pathway is RORA. It has been shown that RORA protein decreases the activation of NFB in PAC1 cells by induction of a ROR response element in the promoter of IkappaB alpha (IB ), an inhibitor of NFkB (263). In addition, it also prevents expression of TNFa-induced IL-6, IL-8, and COX-2 genes. A previous study revealed cross-talk between PPARs, RAR, RXR and RORs (264). Galson D.L. et al. identified that RORA is specifically bound to the response element of the Epo promoter and the enhancer, which represents an ideal model system expressed in response to the hypoxia forming a DNA-protein complex (265). Other studies showed that TGFB1 protein increases expression of RORA mRNA in mouse Th17cells (266, 267). In addition, Yang et al demonstrated that RORA expression is dependent on signal transducer and activator of transcription 3 (STAT3) protein.

Another gene, FRMD6 seems to have a very interesting connection to the family of YWHAB and YWHAZ proteins (268), tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation proteins. It has direct protein-protein interaction with 14-3-3 proteins (YWHA family), which are capable of creating pro-apoptotic associations (269). In the genome-wide microarray analysis Lemberger et al. showed that cAMP responsive element binding protein (CREB) is necessary for expression of FRMD6 in the striatum of mouse via binding at the promoter as a transcription factor (270). Decrease of CREB protein abundance by ROS leads to the abrogation of protective mechanisms in the cell, which contributes to the pathogenesis of heart failure, a well-known dose-limiting toxicity of doxorubicin (271, 272). In previous experiments, we observed

132 underexpression of FRMD6 in resistant cancer cells which might be partially due to an increase of oxidative stress after treatment with doxorubicin.

PTPRD showed an interesting interaction with other PTPR family genes as well as with Src. It was demonstrated that PTPRD protein increases tyrosine dephosphorylation of Src protein (273), a non-receptor protein tyrosine kinase that is activated by intracellular ROS (274). Pretreatment with antioxidants in oxaliplatin-treated cells prevents ROS-induced Src activation. Early studies of response to molecular markers showed that the Src kinase family inhibitor, dasatinib, significantly inhibits invasion and fosters cell apoptosis in ovarian cancer (274). There is rising evidence that Src is affected by ROS and, at the same time, affects oxidative regulation in cells thereby influencing cell adhesion, membrane receptor cross-talk and tumor progression and metastasis (275). Also, ligand-bound activation of G-protein-coupled receptors leads to an increase in activation of Src. A redox-mediated activation of Src leads to the suppression of anoikis and promotes detachment-induced metastatic spread as was shown in lung adenocarcinoma and prostate carcinoma cells (276, 277). Therefore, we can speculate that Src dephosphorylation by PTPRD leads to activation of the former, and prevents the cell from entering into apoptosis. Consequently, it might serve as a potential resistance mechanism in case of overexpression of PTPRD. At the same time, lesser amounts of ROS decrease production of Src and potentiate the apoptotic process in cells.

ROBO1, a roundabout homolog1, is another prominent gene from the interactome study. It is up- regulated by p53 and BRCA1, and modulates neuronal, leukocytic, and endothelial migration (219). ROBO1 is a member of the neural cell adhesion family. Deletion resulting in a truncated protein product of this gene was previously associated with breast cancer (278). Although, ROBO1 did not show direct interaction with free radical molecules, it is over-expressed by p53 which enhances ROS production as part of its apoptotic mechanism under stress conditions.

Two of the foregoing genes (FRMD6, and ROBO1) were identified in connection with YWHAZ (and YWHAB), which were associated with drug resistance to doxorubicin in breast cancer patients (105). In addition, several other genes associated with taxol were indirectly related to YWHAZ through Src and CTNNB1 (ROBO1, DCT, PTPRD), and CFTR through heat shock 27kD protein 1 (HSPB1) (Figure 5-4).

133

Figure 5-4: Role of candidate genes associated with resistance to doxorubicin and paclitaxel in oxidative stress response. Green nodes and lines identify genes associated with paclitaxel response, purple nodes and lines identify genes associated with doxorubicin response, yellow stands for the key players in apoptosis of the cell, red arrows identify connections with reactive oxygen species (ROS).

Cell Membrane

Nucleus

134

CFTR is the cystic fibrosis transmembrane conductance regulator protein that has been previously associated with cystic fibrosis disease but not with cancer growth pathways. It is a cAMP-regulated channel that belongs to the ATP-binding cassette (ABC) transporter superfamily and serves as an organic anion efflux channel permeable to inorganic ions (Cl-) and glutathione (GSH). However, previous studies (279) demonstrated that expression of the CFTR gene is decreased by tumor necrosis alpha (TNFa) and interferon gamma, and enhanced by the NFB gene (279). At the same time, a recent study of the pathogenesis of cystic fibrosis associated this gene with a high production of ROS on the respiratory epithelial surface by NADPH oxidases (NOX/DUOX) family (280). Moreover, experimental inhibition of CFTR expression in normal cells led to a decrease in Nrf-2 activity (281) and increase in H202 production, which is in its turn a potent activator of NF-B (282). CFTR mutation leads to a decrease in GSH which consequently activates NFB with IL-1b and TNFa. Cell death ensues due to those alterations in stress-sensitive pathways (283).

Latrophillin 2 (LPHN2) is a member of the latrophillin subfamily of G-protein coupled receptors (GPCR) that is important in cell adhesion and signal transduction. It might be involved in the calcium-independent regulation of exocytosis. Previous studies in breast tumor cell lines correlated higher levels of LPHN2 product in those tumors compared to normal breast tissues (284, 285) which authors believe suggests a potential role of the gene in carcinogenesis.

SGCD (sarcoglycan) is a part of the dystrophin-glycoprotein complex that interacts with filamin 2 (FLN2), a protein that is involved in actin reorganization, survival and maintenance of membrane integrity.

In melanoma WM35 cells, overexpression of DOPAchrome tautomerase (DCT) protein activates Erk1/2 protein (225), which in its turn increases activation of NRF2 protein in cytoplasm (286). Consequently, this would lead to activation of antioxidant enzymes and counteract the action of free radical-generating anticancer drugs. Indeed, previous experiments with radioresistant cells subjected to UVB treatment showed overexpression of DCT (225).

In exploratory analyses, we attempted to identify molecular signatures associated with differences in treatment outcomes and correlate them with estrogen or HER2 receptor status (conventional predictors of pathological CR). Also, we performed pilot correlation of gene

135 expression with survival outcomes. There is widespread consensus that molecular tumor profiles have a great influence on the differences in response to treatment in breast cancer (287). Triple- negative tumors are associated with a poorer outcome compared to ER- and PR- positive tumors. In our study, we identified 3 genes that have been significantly associated with better outcome in triple-negative and HER2-negative tumors.

The dynamic pattern of survival curves in anthracycline-treated breast cancer patients identified that overexpression of RORA and underexpression of FRMD6 shows a trend towards better prognosis in this population. It is not clear yet to what extent the observed increase in SGCD, CFTR, SNTG, and LPHN2 mRNA levels correlate with recurrence rates in breast tumors treated with paclitaxel and doxorubicin. We postulate that deregulation of those genes is at the heart of their associations with recurrent disease. Unfortunately, due to the absence of a probe set for FMRD6 gene in the Affymetrix U133A Chip we could not validate variations of this gene in breast cancer patients treated with both paclitaxel and doxorubicin. However, based on the results of the rest of our studies FRMD6 has very good potential to be a strong marker of drug resistance in breast tumors.

We may speculate that probably the genetic variations discovered in our studies represent acquired mutations that develop in cancer tissues during chemotherapy. It is tempting to assume that these genes are playing a causative role in determining sensitivity to doxorubicin- and paclitaxel- containing therapy in breast cancer. However, these findings should probably be treated more as preliminary results that require further in vivo validation of mechanistic insights into a functional role. We may also assume that some of the candidate genes may serve as downstream effectors of the key biological events involved in the drug resistance. Future studies determining the potential of CFTR, SNTG, SGCD1, LPHN2, and FRMD6 as predictive markers for breast cancer treatment with doxorubicin and paclitaxel should be performed.

Chapter 6 Summary and Future Directions

6 Summary and future Directions

6.1 Conclusions

The research described in this thesis focused on the development of a new genome-wide approach to detect genetic signatures involved in drug resistance in breast cancer. The underlying concept behind this research project was the notion that identification of new genetic signatures has the potential to improve clinical outcomes of anthracycline- and taxanes-based chemotherapy in breast cancer patients. The research background, rationale, hypothesis and aims are described in Chapter I. Detailed review of the current state of research in the field is presented in Chapter II. In Chapter III, a genome-wide study of doxorubicin resistance in cancer cell lines is described. Chapter III also presents the results of the in silico characterization of three candidate genes (DSG1, RORA, and FRMD6). An in vitro validation of the gene expression in doxorubicin and epirubicin-resistant cell lines concluded this chapter. This study showed an association of 14q21, 15q21, and 18q12 regions with doxorubicin resistance. Further evaluation of the genes in these regions demonstrated that underexpression of FRMD6 and overexpression of DSG1 and RORA is significantly associated with a resistant phenotype in anthracycline-treated cancer cell lines. All three genes were shown to be involved in pro-apoptotic mechanisms in the cell which suggests the importance of apoptotic signaling pathways in the development of resistance to anthracyclines. Chapter IV describes similar methodology applied to detect genetic variations associated with paclitaxel resistance. 11 protein-coding genes (CFTR, ROBO1, PTPRD, BTBD12, CCDC26, DCT, SNTG1, SGCD, LPHN2, GRIK1, and ZNF607) were associated with a resistant phenotype in paclitaxel-treated cancer cell lines. Similar bioinformatic methods have been applied to characterize them, and validate their expression in cancer cell lines. Interestingly, five genes out of 13 showed differential mRNA expression levels between resistant and sensitive NCI60 cell lines, namely CFTR, GRIK1, DCT, SGCD, and SNTG1. Further investigation of the 136

137 downstream signaling pathways involved with candidate genes identified that p53, -catenin, and microtubule interacting pathways represent key players associated with resistance to paclitaxel. Chapter V presents an original approach to the organization of multiple association signals identified in two GWA studies according to underlying biological processes into one canonical signaling cascade that is likely to contribute to resistance to both doxorubicin and paclitaxel. This study is based on the detailed analysis of common pathways in 16 candidate genes and 10,438 SNPs involved in the resistance to both doxorubicin and anthracyclines. An Nrf2-oxidative stress pathway was identified as the conventional process with the most gene hits and associations in it. Discussion includes an in-depth analysis of the interacting genes and oxidative stress with putative mechanisms of resistance to both drugs. Finally, this chapter touches upon the pilot validation results in breast tumors from patients treated with anthracyclines- and/or paclitaxel containing regimens (Array Express, GSE19615, and GSE20194). The analysis in anthracycline-treated breast tumors identified that underexpression of FRMD6 gene is significantly associated with recurrent disease and poorer survival. A similar analysis in anthracyclines- and taxane-treated breast cancer patients showed evidence that underexpression of four out nine genes (CFTR, SNTG1, SGCD, and LPHN) was associated with significantly better outcomes after treatment. Additionally, these findings further validate the results obtained from in vitro and in silico analysis in cancer cell lines. Overall, four genes (FRMD6, CFTR, SGCD, and SNTG) were significantly associated with a resistant phenotype across all our studies.

6.2 Advantages and Limitations

6.2.1 Advantages:

Attempting to decipher a multigene signature to predict response to treatment in cancer patients is a challenging task. A multitude of genetic variations provide the basis for variable response to antineoplastic drugs. We have demonstrated a potential methodology based on cancer cell response to define genes associated with resistant phenotypes. o The most significant advantage of the presented methodology is that it allows us to detect genetic signatures to any anticancer agent used in chemotherapy. Since drug

138

resistance and low-efficiency rates in chemotherapeutic treatment are universal problems in the majority of tumors, this approach may potentially be effective in other tumor types and treatment regimens. o This novel methodology targets a large population of cancer patients whose prognosis may be improved by individualized therapy. o However, the study was based on the response in cancer cells which are known to undergo mutations during cell passage and propagation. Thus, their genotypes more closely resemble metastatic and recurrent tumor tissues rather than primary tumors. o Efficient combination of in silico and in vitro methods allows us to have an in-depth look at the combination of several drugs and diseases subject to availability of data for each of them separately. o Although doxorubicin and paclitaxel have different mechanism of actions, we identified the Nrf-2 oxidative stress pathway as a common biological process activated on the intracellular level in both cases. o Thus, the oxidative stress pathway and antioxidants are crucial for maintaining the defense mechanism and controlling increased/decreased sensitivity of cancer cells to anthracyclines- and taxanes-containing therapy. Although the use of antioxidants as supplementary agents in chemotherapy is an exciting future prospect to improve clinical outcomes, future studies should examine their effect in vitro prior to clinical application. o Although the various cellular and molecular events described above are important, it is critical to consider the role of cell-cell interactions and transcellular processes. It has been observed in our studies that 9 out of 16 genes were predominantly of cellular plasma membrane location. o Apparent lack of expression of certain genes in available datasets could plausibly be a genetic difference between primary and metastatic tumors, although other mechanisms may also account for this effect. o Our results show that hormone receptor status (ER-negative and triple-negative tumors) affects genes involved in resistance to anthracyclines and taxanes in breast cancer which is in agreement with previous literature reports.

139

6.2.2 Limitations:

Several possible limitations could be considered for our clinical validation studies: o We have linked NCI60 cancer cell lines treated with the drug of interest to the genotypes to develop our strategy. Although, it is the most promising available in vitro study model, we should always take into considerations that it might not be directly applicable to the human tumor tissues as cancer cell lines undergo changes with cell passage and propagation. To minimize this discrepancy, we ran a number of validation tests in other cell lines and on tumor tissues from patients. o It would have been much more conclusive to assess gene expression and SNP frequencies in a much larger patient populations. Also, single-agent therapy is no longer actively in use in breast cancer. Hence, it is hard to differentiate from potential effects on the genetic variations caused by other drugs that could overlap with the doxorubicin action in patients. o Incomplete annotation of the human genome and lack of knowledge of the function of many human genes limits the genome-wide pathway analysis as many genes cannot be assigned to known pathways. o Even though detailed genome-wide mapping of SNPs is widely available, the function of susceptibility loci in intergenic regions still represents a scientific puzzle that needs to be solved. o Although there are plenty of curated different databases, lack of uniformity among those public resources may substantially impact on the outcomes of the study. To minimize this effect, we manually selected and confirmed data obtained from these resources. o Finally, another possible limitation to our study could be that there is limited publicly available data. As well, there is much more to learn about genes and their potential function than is currently known. To derive more definitive mechanisms associating discovered candidate genes with drug resistance further experimental evidence is necessary.

140

6.3 Future directions

Several future studies are contemplated. o Our analysis is arguably the unique integration of pharmacogenetics and functional genomics in the resistance of anthracyclines and taxanes in breast cancer yielding a series of candidate genes, pathways and mechanisms. Future studies may include these prime targets for future hypothesis driven studies focused on deep re- sequencing, biological validation in cell culture and transgenic animal work on these individual candidate genes. o Second, our work provides integrated evidence prioritizing a number of genes as candidate pharmacogenomic biomarkers in breast cancer. It is still unclear whether observed alterations in gene expression were truly related to the causation of the resistant phenotype in tumor tissues or were instead related to medication effects developed after treatment. To distinguish the causative role of the observed alterations, one can possibly verify the outcomes on the different set of cell lines, for example, lymphoblastoid cell lines. These cell lines were derived from the individual lymphoblastoid cells carrying the Epstein-Barr Virus (EBV) and have non-malignant properties. This model has been extensively used to study drug sensitivity in several types of cancer. The International HapMap project and the 1000 Genomes Project (http://www.1000genomes.org) helped to build a catalog of human genetic variations on more than 4 million SNPs, copy number variation (CNV), and gene expression data which would serve as a useful tool to validate the nature of the observed variations from our study. However, this model system has certain limitations due to possible alterations of the cellular processes to certain drugs, i.e. doxorubicin, in response to the Epstein-Barr Virus (288, 289). o Third, the use of GWAS data in conjunction with gene expression data as part of an integrative approach in pharmacogenomics, followed by pathway-enrichment analysis can serve as a stepping stone for unraveling the novel genetic signatures involved in the resistance to chemotherapy. An in-depth analysis of the identified pathway in terms of combination of genes and gene expression patterns may need to be carried out in order to lead to major re-evaluations of our therapeutic approaches in breast cancer.

141 o In this study we combined genome-wide association data, expression profiling and linkage to predict candidate genes and pathways for further research. One might also consider proteomic studies to identify protein products differentially expressed between resistant and sensitive tumors in the future. o Furthermore, our study focused on breast cancer patients for validation, but future validation experiments should also include other types of cancer that have been represented on the NCI60 cell panel such as lung, colon, melanoma, brain and hematologic malignancies. o Finally, we produced promising results from the pilot clinical validation. Although, there was a convergence of outcomes in gene expression in four candidate genes from GWA study, further clinical validation on a larger sample size would be beneficial in order to identify and confirm findings. Such a model can be represented by complimenting it with a large-sample size study on breast cancer patients treated with anthracyclines and taxanes or by a meta-analysis of several similar studies.

142

References

1. Canadian Cancer Society. Canadian Cancer Statistics 2010April 2010: Available from: (http://www.cancer.ca/Canada- wide/About%20cancer/Cancer%20statistics/~/media/CCS/Canada%20wide/Files%20List/Englis h%20files%20heading/PDF%20-%20Policy%20-%20Canadian%20Cancer%20Statistics%20- %20English/Canadian%20Cancer%20Statistics%202010%20-%20English.ashx). .

2. Horner MJ RL, Krapcho M, Neyman N, Aminou R, Howlader N, Altekruse SF, Feuer EJ, Huang L, Mariotto A, Miller BA, Lewis DR, Eisner MP, Stinchcomb DG, Edwards BK (eds),. SEER Cancer Statistics Review. (http://seer.cancer.gov/csr/1975_2006/). [Based on November 2008 SEER data submission] Bethesda, MD: National Cancer Institute; 1975-2006 [updated February 2009]; Available from: http://seer.cancer.gov/csr/1975_2006/.

3. Canadian Cancer Society. Breast Cancer Statistics2010: Available from: http://www.cancer.ca/Canada- wide/About%20cancer/Cancer%20statistics/Stats%20at%20a%20glance/Breast%20cancer.aspx? sc_lang=en.

4. Hortobagyi GN. Treatment of breast cancer. N Engl J Med. 1998 Oct 1;339(14):974-84.

5. Hortobagyi GN, Buzdar AU. Current status of adjuvant systemic therapy for primary breast cancer: progress and controversy. CA Cancer J Clin. 1995 Jul-Aug;45(4):199-226.

6. Moreno-Aspitia A, Perez EA. Anthracycline- and/or taxane-resistant breast cancer: results of a literature review to determine the clinical challenges and current treatment trends. Clin Ther. 2009 Aug;31(8):1619-40.

7. Ries L, Melbert D, Krapcho M, Stinchcomb D, Howlader N, Horner M, et al. SEER Cancer Statistics Review, 1975-20052008: Available from: http://seer.cancer.gov/csr/1975_2005.

8. Jemal A, Siegel R, Ward E, Hao Y, Xu J, Thun MJ. Cancer statistics, 2009. CA Cancer J Clin. 2009 Jul-Aug;59(4):225-49.

9. National Comprehensive Cancer Network. NCCN Practice Guidelines in Oncology- v.1.2010: Breast Cancer. http://www.nccn.org/clinical.asp. February 2010 [cited 2010 February]; Available from: http://www.nccn.org/professionals/physician_gls/PDF/breast.pdf.

10. Maughan KL, Lutterbie MA, Ham PS. Treatment of breast cancer. Am Fam Physician. 2010 Jun 1;81(11):1339-46.

11. Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials. Lancet. 2005 May 14- 20;365(9472):1687-717.

143

12. Pater JL, Pritchard KI. Optimal chemotherapy for women with breast cancer: the plot thickens. J Clin Oncol. 2003 Mar 15;21(6):963-4.

13. Henderson IC, Berry DA, Demetri GD, Cirrincione CT, Goldstein LJ, Martino S, et al. Improved outcomes from adding sequential Paclitaxel but not from escalating Doxorubicin dose in an adjuvant chemotherapy regimen for patients with node-positive primary breast cancer. J Clin Oncol. 2003 Mar 15;21(6):976-83.

14. Roche H, Fumoleau P, Spielmann M, Canon JL, Delozier T, Serin D, et al. Sequential adjuvant epirubicin-based and docetaxel chemotherapy for node-positive breast cancer patients: the FNCLCC PACS 01 Trial. J Clin Oncol. 2006 Dec 20;24(36):5664-71.

15. Hortobagyi GN. Multidisciplinary management of advanced primary and metastatic breast cancer. Cancer. 1994 Jul 1;74(1 Suppl):416-23.

16. Hortobagyi GN, Holmes FA. Optimal dosing of paclitaxel and doxorubicin in metastatic breast cancer. Semin Oncol. 1997 Feb;24(1 Suppl 3):S4-7.

17. Burke HB. Outcome prediction and the future of the TNM staging system. J Natl Cancer Inst. 2004 Oct 6;96(19):1408-9.

18. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, et al. Molecular portraits of human breast tumours. Nature. 2000 Aug 17;406(6797):747-52.

19. Haupt B, Ro JY, Schwartz MR. Basal-like breast carcinoma: a phenotypically distinct entity. Arch Pathol Lab Med. Jan;134(1):130-3.

20. Raica M, Jung I, Cimpean AM, Suciu C, Muresan AM. From conventional pathologic diagnosis to the molecular classification of breast carcinoma: are we ready for the change? Rom J Morphol Embryol. 2009;50(1):5-13.

21. Rakha EA, Reis-Filho JS, Ellis IO. Impact of basal-like breast carcinoma determination for a more specific therapy. Pathobiology. 2008;75(2):95-103.

22. Roden DM, Altman RB, Benowitz NL, Flockhart DA, Giacomini KM, Johnson JA, et al. Pharmacogenomics: challenges and opportunities. Ann Intern Med. 2006 Nov 21;145(10):749- 57.

23. Kalow W. Pharmacogenomics: Historical Perspective and Current Status. In: Innocenti F, ebrary Inc., editors. Pharmacogenomics methods and protocols. Totowa, NJ: Humana Press Inc.; 2005. p. x, 214 p.

24. Giacomini KM, Brett CM, Altman RB, Benowitz NL, Dolan ME, Flockhart DA, et al. The pharmacogenetics research network: from SNP discovery to clinical drug response. Clin Pharmacol Ther. 2007 Mar;81(3):328-45.

25. Di Cosimo S, Baselga J. Targeted therapies in breast cancer: where are we now? Eur J Cancer. 2008 Dec;44(18):2781-90.

144

26. Weinshilboum RM. Pharmacogenomics: catechol O-methyltransferase to thiopurine S- methyltransferase. Cell Mol Neurobiol. 2006 Jul-Aug;26(4-6):539-61.

27. Coate L, Cuffe S, Horgan A, Hung RJ, Christiani D, Liu G. Germline genetic variation, cancer outcome, and pharmacogenetics. J Clin Oncol. Sep 10;28(26):4029-37.

28. Ginsburg GS, Willard HF. Genomic and personalized medicine: foundations and applications. Transl Res. 2009 Dec;154(6):277-87.

29. Overdevest JB, Theodorescu D, Lee JK. Utilizing the molecular gateway: the path to personalized cancer management. Clin Chem. 2009 Apr;55(4):684-97.

30. Sotiriou C, Pusztai L. Gene-expression signatures in breast cancer. N Engl J Med. 2009 Feb 19;360(8):790-800.

31. O'Malley FP, Chia S, Tu D, Shepherd LE, Levine MN, Bramwell VH, et al. Topoisomerase II alpha and responsiveness of breast cancer to adjuvant chemotherapy. J Natl Cancer Inst. 2009 May 6;101(9):644-50.

32. Pritchard KI, Messersmith H, Elavathil L, Trudeau M, O'Malley F, Dhesy-Thind B. HER-2 and topoisomerase II as predictors of response to chemotherapy. J Clin Oncol. 2008 Feb 10;26(5):736-44.

33. Pusztai L. Current status of prognostic profiling in breast cancer. Oncologist. 2008 Apr;13(4):350-60.

34. Chorley BN, Wang X, Campbell MR, Pittman GS, Noureddine MA, Bell DA. Discovery and verification of functional single nucleotide polymorphisms in regulatory genomic regions: current and developing technologies. Mutat Res. 2008 Jul-Aug;659(1-2):147-57.

35. Crawford DC, Nickerson DA. Definition and clinical importance of haplotypes. Annu Rev Med. 2005;56:303-20.

36. Erichsen HC, Chanock SJ. SNPs in cancer research and treatment. Br J Cancer. 2004 Feb 23;90(4):747-51.

37. Gonzalez-Angulo AM, Hennessy BT, Mills GB. Future of personalized medicine in oncology: a systems biology approach. J Clin Oncol. 2010 Jun 1;28(16):2777-83.

38. Wang WY, Barratt BJ, Clayton DG, Todd JA. Genome-wide association studies: theoretical and practical concerns. Nat Rev Genet. 2005 Feb;6(2):109-18.

39. Ioannidis JP, Castaldi P, Evangelou E. A compendium of genome-wide associations for cancer: critical synopsis and reappraisal. J Natl Cancer Inst. 2010 Jun 16;102(12):846-58.

40. Cirulli ET, Goldstein DB. Uncovering the roles of rare variants in common disease through whole-genome sequencing. Nat Rev Genet. 2010 Jun;11(6):415-25.

145

41. Manolio TA. Genomewide association studies and assessment of the risk of disease. N Engl J Med. 2010 Jul 8;363(2):166-76.

42. Wang L, Weinshilboum RM. Pharmacogenomics: candidate gene identification, functional validation and mechanisms. Hum Mol Genet. 2008 Oct 15;17(R2):R174-9.

43. Welsh M, Mangravite L, Medina MW, Tantisira K, Zhang W, Huang RS, et al. Pharmacogenomic discovery using cell-based models. Pharmacol Rev. 2009 Dec;61(4):413-29.

44. Shoemaker RH. The NCI60 human tumour cell line anticancer drug screen. Nat Rev Cancer. 2006 Oct;6(10):813-23.

45. CellMiner, build 1.0. http://discover.nci.nih.gov/cellminer [database on the Internet]. Available from: http://discover.nci.nih.gov/cellminer.

46. Shankavaram UT, Varma S, Kane D, Sunshine M, Chary KK, Reinhold WC, et al. CellMiner: a relational database and query tool for the NCI-60 cancer cell lines. BMC Genomics. 2009;10:277.

47. Le Morvan V, Bellott R, Moisan F, Mathoulin-Pelissier S, Bonnet J, Robert J. Relationships between genetic polymorphisms and anticancer drug cytotoxicity vis-a-vis the NCI-60 panel. Pharmacogenomics. 2006 Sep;7(6):843-52.

48. Puyo S, Le Morvan V, Robert J. Impact of EGFR gene polymorphisms on anticancer drug cytotoxicity in vitro. Mol Diagn Ther. 2008;12(4):225-34.

49. Sasaki S, Kobunai T, Kitayama J, Nagawa H. DNA methylation and sensitivity to antimetabolites in cancer cell lines. Oncol Rep. 2008 Feb;19(2):407-12.

50. Jarjanazi H, Kiefer J, Savas S, Briollais L, Tuzmen S, Pabalan N, et al. Discovery of genetic profiles impacting response to chemotherapy: application to gemcitabine. Hum Mutat. 2008 Apr;29(4):461-7.

51. Piccart-Gebhart MJ, Burzykowski T, Buyse M, Sledge G, Carmichael J, Luck HJ, et al. Taxanes alone or in combination with anthracyclines as first-line therapy of patients with metastatic breast cancer. J Clin Oncol. 2008 Apr 20;26(12):1980-6.

52. Pivot X, Asmar L, Buzdar AU, Valero V, Hortobagyi G. A unified definition of clinical anthracycline resistance breast cancer. Br J Cancer. 2000 Feb;82(3):529-34.

53. Lal S, Mahajan A, Chen WN, Chowbay B. Pharmacogenetics of target genes across doxorubicin disposition pathway: a review. Curr Drug Metab. 2010 Jan;11(1):115-28.

54. O'Driscoll L, Clynes M. Biomarkers and multiple drug resistance in breast cancer. Curr Cancer Drug Targets. 2006 Aug;6(5):365-84.

55. O'Driscoll L, Clynes M. Molecular markers of multiple drug resistance in breast cancer. Chemotherapy. 2006;52(3):125-9.

146

56. Trock BJ, Leonessa F, Clarke R. Multidrug resistance in breast cancer: a meta-analysis of MDR1/gp170 expression and its possible functional significance. J Natl Cancer Inst. 1997 Jul 2;89(13):917-31.

57. Takemura G, Fujiwara H. Doxorubicin-induced cardiomyopathy from the cardiotoxic mechanisms to management. Prog Cardiovasc Dis. 2007 Mar-Apr;49(5):330-52.

58. Minotti G, Menna P, Salvatorelli E, Cairo G, Gianni L. Anthracyclines: molecular advances and pharmacologic developments in antitumor activity and cardiotoxicity. Pharmacol Rev. 2004 Jun;56(2):185-229.

59. Cutts SM, Nudelman A, Rephaeli A, Phillips DR. The power and potential of doxorubicin-DNA adducts. IUBMB Life. 2005 Feb;57(2):73-81.

60. Gluck S. Adjuvant chemotherapy for early breast cancer: optimal use of epirubicin. Oncologist. 2005 Nov-Dec;10(10):780-91.

61. NCCN. Clinical Practice Guidelines in Oncology. National Comprehensive Cancer Network; 2006; Available from: http://www.nccn.org/professionals/physician_gls/PDF/breast.pdf.

62. Xia L, Jaafar L, Cashikar A, Flores-Rozas H. Identification of genes required for protection from doxorubicin by a genome-wide screen in Saccharomyces cerevisiae. Cancer Res. 2007 Dec 1;67(23):11411-8.

63. Gewirtz DA. A critical evaluation of the mechanisms of action proposed for the antitumor effects of the anthracycline antibiotics adriamycin and daunorubicin. Biochem Pharmacol. 1999 Apr 1;57(7):727-41.

64. Lal S, Wong ZW, Jada SR, Xiang X, Chen Shu X, Ang PC, et al. Novel SLC22A16 polymorphisms and influence on doxorubicin pharmacokinetics in Asian breast cancer patients. Pharmacogenomics. 2007 Jun;8(6):567-75.

65. Okabe M, Szakacs G, Reimers MA, Suzuki T, Hall MD, Abe T, et al. Profiling SLCO and SLC22 genes in the NCI-60 cancer cell lines to identify drug uptake transporters. Mol Cancer Ther. 2008 Sep;7(9):3081-91.

66. Gajewski E, Rao G, Nackerdien Z, Dizdaroglu M. Modification of DNA bases in mammalian chromatin by radiation-generated free radicals. Biochemistry. 1990 Aug 28;29(34):7876-82.

67. Buchholz TA, Davis DW, McConkey DJ, Symmans WF, Valero V, Jhingran A, et al. Chemotherapy-induced apoptosis and Bcl-2 levels correlate with breast cancer response to chemotherapy. Cancer J. 2003 Jan-Feb;9(1):33-41.

68. Coukell AJ, Faulds D. Epirubicin. An updated review of its pharmacodynamic and pharmacokinetic properties and therapeutic efficacy in the management of breast cancer. Drugs. 1997 Mar;53(3):453-82.

147

69. Pritchard KI, Shepherd LE, O'Malley FP, Andrulis IL, Tu D, Bramwell VH, et al. HER2 and responsiveness of breast cancer to adjuvant chemotherapy. N Engl J Med. 2006 May 18;354(20):2103-11.

70. Pritchard KI, Messersmith H, Elavathil L, Trudeau M, O'Malley F, Dhesy-Thind B. Her- 2 and topoisomerase II as predictors of response to chemotherapy J Clin Oncol. [Abstract]. 2008;26:1-9.

71. Gennari A, Sormani MP, Pronzato P, Puntoni M, Colozza M, Pfeffer U, et al. HER2 status and efficacy of adjuvant anthracyclines in early breast cancer: a pooled analysis of randomized trials. J Natl Cancer Inst. 2008 Jan 2;100(1):14-20.

72. Levine MN, Pritchard KI, Bramwell VH, Shepherd LE, Tu D, Paul N. Randomized trial comparing cyclophosphamide, epirubicin, and fluorouracil with cyclophosphamide, methotrexate, and fluorouracil in premenopausal women with node-positive breast cancer: update of National Cancer Institute of Canada Clinical Trials Group Trial MA5. Br Cancer Res Treat. [Abstract]. 2002;76:S33-A17.

73. Levine MN, Pritchard KI, Bramwell VH, Shepherd LE, Tu D, Paul N. Randomized trial comparing cyclophosphamide, epirubicin, and fluorouracil with cyclophosphamide, methotrexate, and fluorouracil in premenopausal women with node-positive breast cancer: update of National Cancer Institute of Canada Clinical Trials Group Trial MA5. J Clin Oncol. 2005 Aug 1;23(22):5166-70.

74. Harrison DJ. Molecular mechanisms of drug resistance in tumours. J Pathol. 1995 Jan;175(1):7-12.

75. Tomida A, Tsuruo T. Drug resistance pathways as targets. In: Baguley B, Kerr D, editors. Anticancer drug development. 1 edition ed. San Diego: Academic Press; Oct-2001. p. 397.

76. Paul D, Cowan K. Drug Resistance in Breast Cancer. In: Bowcock AM, editor. Breast Cancer: Molecular Genetics, Pathogenesis, and Therapeutics. Totowa, New Jersey: Humana Press; 1999.

77. Tubbs R, Barlow WE, Budd GT, Swain E, Porter P, Gown A, et al. Outcome of patients with early-stage breast cancer treated with doxorubicin-based adjuvant chemotherapy as a function of HER2 and TOP2A status. J Clin Oncol. 2009 Aug 20;27(24):3881-6.

78. Nielsen D, Maare C, Skovsgaard T. Cellular resistance to anthracyclines. Gen Pharmacol. 1996 Mar;27(2):251-5.

79. Saleh EM, El-Awady RA, Abdel Alim MA, Abdel Wahab AH. Altered expression of proliferation-inducing and proliferation-inhibiting genes might contribute to acquired doxorubicin resistance in breast cancer cells. Cell Biochem Biophys. 2009;55(2):95-105.

80. Boatright KM, Salvesen GS. Mechanisms of caspase activation. Curr Opin Cell Biol. 2003 Dec;15(6):725-31.

148

81. Wani MC, Taylor HL, Wall ME, Coggon P, McPhail AT. Plant antitumor agents. VI. The isolation and structure of taxol, a novel antileukemic and antitumor agent from Taxus brevifolia. J Am Chem Soc. 1971 May 5;93(9):2325-7.

82. Gueritte F. General and recent aspects of the chemistry and structure-activity relationships of taxoids. Curr Pharm Des. 2001 Sep;7(13):1229-49.

83. Green H, Soderkvist P, Rosenberg P, Mirghani RA, Rymark P, Lundqvist EA, et al. Pharmacogenetic studies of Paclitaxel in the treatment of ovarian cancer. Basic Clin Pharmacol Toxicol. 2009 Feb;104(2):130-7.

84. Kwon WS, Rha SY, Jeung HC, Ahn JB, Jung JJ, Noh SH, et al. G-T haplotype (2677G>T/A and 3435C>T) of ABCB1 gene polymorphisms is associated with ethnic differences to paclitaxel sensitivity in cancer cells with different gene expression pattern. Cancer Lett. 2009 May 18;277(2):155-63.

85. Maeno K, Ito K, Hama Y, Shingu K, Kimura M, Sano M, et al. Mutation of the class I beta-tubulin gene does not predict response to paclitaxel for breast cancer. Cancer Lett. 2003 Jul 30;198(1):89-97.

86. Rowinsky EK, Donehower RC. Paclitaxel (taxol). N Engl J Med. 1995 Apr 13;332(15):1004-14.

87. Gehrmann M, Schmidt M, Brase JC, Roos P, Hengstler JG. Prediction of paclitaxel resistance in breast cancer: is CYP1B1*3 a new factor of influence? Pharmacogenomics. 2008 Jul;9(7):969-74.

88. Somlo G, Doroshow JH, Synold T, Longmate J, Reardon D, Chow W, et al. High-dose paclitaxel in combination with doxorubicin, cyclophosphamide and peripheral blood progenitor cell rescue in patients with high-risk primary and responding metastatic breast carcinoma: toxicity profile, relationship to paclitaxel pharmacokinetics and short-term outcome. Br J Cancer. 2001 Jun 15;84(12):1591-8.

89. Paul C. Drug resistance in Breast cancer. In: Bowcock AM, editor. Breast cancer : molecular genetics, pathogenesis, and therapeutics. Totowa, N.J.: Humana Press; 1999. p. xiii, 582 p.

90. McGrogan BT, Gilmartin B, Carney DN, McCann A. Taxanes, microtubules and chemoresistant breast cancer. Biochim Biophys Acta. 2008 Apr;1785(2):96-132.

91. Marsh S, Somlo G, Li X, Frankel P, King CR, Shannon WD, et al. Pharmacogenetic analysis of paclitaxel transport and metabolism genes in breast cancer. Pharmacogenomics J. 2007 Oct;7(5):362-5.

92. Pusztai L. Markers predicting clinical benefit in breast cancer from microtubule-targeting agents. Ann Oncol. 2007 Dec;18 Suppl 12:xii15-20.

93. Green H, Soderkvist P, Rosenberg P, Horvath G, Peterson C. ABCB1 G1199A polymorphism and ovarian cancer response to paclitaxel. J Pharm Sci. 2008 Jun;97(6):2045-8.

149

94. Liu Y, Peng H, Zhang JT. Expression profiling of ABC transporters in a drug-resistant breast cancer cell line using AmpArray. Mol Pharmacol. 2005 Aug;68(2):430-8.

95. Lagas JS, Vlaming ML, van Tellingen O, Wagenaar E, Jansen RS, Rosing H, et al. Multidrug resistance protein 2 is an important determinant of paclitaxel pharmacokinetics. Clin Cancer Res. 2006 Oct 15;12(20 Pt 1):6125-32.

96. Huisman MT, Chhatta AA, van Tellingen O, Beijnen JH, Schinkel AH. MRP2 (ABCC2) transports taxanes and confers paclitaxel resistance and both processes are stimulated by probenecid. Int J Cancer. 2005 Sep 20;116(5):824-9.

97. Komatsu M, Hiyama K, Tanimoto K, Yunokawa M, Otani K, Ohtaki M, et al. Prediction of individual response to platinum/paclitaxel combination using novel marker genes in ovarian cancers. Mol Cancer Ther. 2006 Mar;5(3):767-75.

98. Galletti E, Magnani M, Renzulli ML, Botta M. Paclitaxel and docetaxel resistance: molecular mechanisms and development of new generation taxanes. ChemMedChem. 2007 Jul;2(7):920-42.

99. Pierga JY, Reis-Filho JS, Cleator SJ, Dexter T, Mackay A, Simpson P, et al. Microarray- based comparative genomic hybridisation of breast cancer patients receiving neoadjuvant chemotherapy. Br J Cancer. 2007 Jan 29;96(2):341-51.

100. Yasui K, Mihara S, Zhao C, Okamoto H, Saito-Ohara F, Tomida A, et al. Alteration in copy numbers of genes as a mechanism for acquired drug resistance. Cancer Res. 2004 Feb 15;64(4):1403-10.

101. O'Brien C, Cavet G, Pandita A, Hu X, Haydu L, Mohan S, et al. Functional genomics identifies ABCC3 as a mediator of taxane resistance in HER2-amplified breast cancer. Cancer Res. 2008 Jul 1;68(13):5380-9.

102. Villeneuve DJ, Hembruff SL, Veitch Z, Cecchetto M, Dew WA, Parissenti AM. cDNA microarray analysis of isogenic paclitaxel- and doxorubicin-resistant breast tumor cell lines reveals distinct drug-specific genetic signatures of resistance. Breast Cancer Res Treat. 2006 Mar;96(1):17-39.

103. Hembruff SL, Villeneuve DJ, Parissenti AM. The optimization of quantitative reverse transcription PCR for verification of cDNA microarray data. Anal Biochem. 2005 Oct 15;345(2):237-49.

104. Blancafort P, Tschan MP, Bergquist S, Guthy D, Brachat A, Sheeter DA, et al. Modulation of drug resistance by artificial transcription factors. Mol Cancer Ther. 2008 Mar;7(3):688-97.

105. Li Y, Zou L, Li Q, Haibe-Kains B, Tian R, Desmedt C, et al. Amplification of LAPTM4B and YWHAZ contributes to chemotherapy resistance and recurrence of breast cancer. Nat Med. Feb;16(2):214-8.

150

106. Menendez JA, Mehmi I, Lupu R. Heregulin-triggered Her-2/neu signaling enhances nuclear accumulation of p21WAF1/CIP1 and protects breast cancer cells from cisplatin-induced genotoxic damage. Int J Oncol. 2005 Mar;26(3):649-59.

107. Yang W, Klos KS, Zhou X, Yao J, Yang Y, Smith TL, et al. ErbB2 overexpression in human breast carcinoma is correlated with p21Cip1 up-regulation and tyrosine-15 hyperphosphorylation of p34Cdc2: poor responsiveness to chemotherapy with cyclophoshamide methotrexate, and 5-fluorouracil is associated with Erb2 overexpression and with p21Cip1 overexpression. Cancer. 2003 Sep 15;98(6):1123-30.

108. Gonzalez-Angulo AM, Morales-Vasquez F, Hortobagyi GN. Overview of resistance to systemic therapy in patients with breast cancer. Adv Exp Med Biol. 2007;608:1-22.

109. Shastry BS. Pharmacogenetics and the concept of individualized medicine. Pharmacogenomics J. 2006 Jan-Feb;6(1):16-21.

110. Donnelly JG. Pharmacogenetics in cancer chemotherapy: balancing toxicity and response. Ther Drug Monit. 2004 Apr;26(2):231-5.

111. Lango H, Weedon MN. What will whole genome searches for susceptibility genes for common complex disease offer to clinical practice? J Intern Med. 2008 Jan;263(1):16-27.

112. Clarke M, Coates AS, Darby SC, Davies C, Gelber RD, Godwin J, et al. Adjuvant chemotherapy in oestrogen-receptor-poor breast cancer: patient-level meta-analysis of randomised trials. Lancet. 2008 Jan 5;371(9606):29-40.

113. Stinson SF, Alley MC, Kopp WC, Fiebig HH, Mullendore LA, Pittman AF, et al. Morphological and immunocytochemical characteristics of human tumor cell lines for use in a disease-oriented anticancer drug screen. Anticancer Res. 1992 Jul-Aug;12(4):1035-53.

114. Garraway LA, Widlund HR, Rubin MA, Getz G, Berger AJ, Ramaswamy S, et al. Integrative genomic analyses identify MITF as a lineage survival oncogene amplified in malignant melanoma. Nature. 2005 Jul 7;436(7047):117-22.

115. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender D, et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet. 2007 Sep;81(3):559-75.

116. Montana G. Statistical methods in genetics. Brief Bioinform. 2006 Sep;7(3):297-308.

117. Balding DJ. A tutorial on statistical methods for population association studies. Nat Rev Genet. 2006 Oct;7(10):781-91.

118. Logan BR, Rowe DB. An evaluation of thresholding techniques in fMRI analysis. Neuroimage. 2004 May;22(1):95-108.

119. Barrett JC, Fry B, Maller J, Daly MJ. Haploview: analysis and visualization of LD and haplotype maps. Bioinformatics. 2005 Jan 15;21(2):263-5.

151

120. McCusker JP, Phillips JA, Beltran AG, Finkelstein A, Krauthammer M. Semantic web data warehousing for caGrid. BMC Bioinformatics. 2009;10 Suppl 10:S2.

121. Parkinson H, Kapushesky M, Kolesnikov N, Rustici G, Shojatalab M, Abeygunawardena N, et al. ArrayExpress update--from an archive of functional genomics experiments to the atlas of gene expression. Nucleic Acids Res. 2009 Jan;37(Database issue):D868-72.

122. Cheng AS, Culhane AC, Chan MW, Venkataramu CR, Ehrich M, Nasir A, et al. Epithelial progeny of estrogen-exposed breast progenitor cells display a cancer-like methylome. Cancer Res. 2008 Mar 15;68(6):1786-96.

123. [cited 2008]; 04/12/2008:[transcription profiling, co-expression, disease state]. Available from: http://www.ebi.ac.uk/gxa/experiment/E-TABM-276.

124. Hembruff SL, Laberge ML, Villeneuve DJ, Guo B, Veitch Z, Cecchetto M, et al. Role of drug transporters and drug accumulation in the temporal acquisition of drug resistance. BMC Cancer. 2008;8:318.

125. Marshall OJ. PerlPrimer: cross-platform, graphical primer design for standard, bisulphite and real-time PCR. Bioinformatics. 2004 Oct 12;20(15):2471-2.

126. Zhang Z, Schwartz S, Wagner L, Miller W. A greedy algorithm for aligning DNA sequences. J Comput Biol. 2000 Feb-Apr;7(1-2):203-14.

127. Livak KJ, Schmittgen TD. Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods. 2001 Dec;25(4):402-8.

128. Knutsen T, Gobu V, Knaus R, Padilla-Nash H, Augustus M, Strausberg RL, et al. The interactive online SKY/M-FISH & CGH database and the Entrez cancer chromosomes search database: linkage of chromosomal aberrations with the genome sequence. Genes Chromosomes Cancer. 2005 Sep;44(1):52-64.

129. DTP. Developmental Therapeutic Program. Available from: http://dtp.nci.nih.gov/mtargets/download.html.

130. Gibbs JR, Singleton A. Application of genome-wide single nucleotide polymorphism typing: simple association and beyond. PLoS Genet. 2006 Oct 6;2(10):e150.

131. Elmore LW, Rehder CW, Di X, McChesney PA, Jackson-Cook CK, Gewirtz DA, et al. Adriamycin-induced senescence in breast tumor cells involves functional p53 and telomere dysfunction. J Biol Chem. 2002 Sep 20;277(38):35509-15.

132. Souid AK, Penefsky HS, Sadowitz PD, Toms B. Enhanced cellular respiration in cells exposed to doxorubicin. Mol Pharm. 2006 May-Jun;3(3):307-21.

133. Tenesa A, Farrington SM, Prendergast JG, Porteous ME, Walker M, Haq N, et al. Genome-wide association scan identifies a colorectal cancer susceptibility locus on 11q23 and replicates risk loci at 8q24 and 18q21. Nat Genet. 2008 May;40(5):631-7.

152

134. Gudmundsson J, Sulem P, Gudbjartsson DF, Blondal T, Gylfason A, Agnarsson BA, et al. Genome-wide association and replication studies identify four variants associated with prostate cancer susceptibility. Nat Genet. 2009 Oct;41(10):1122-6.

135. Zhang W, Kornblau SM, Kobayashi T, Gambel A, Claxton D, Deisseroth AB. High levels of constitutive WAF1/Cip1 protein are associated with chemoresistance in acute myelogenous leukemia. Clin Cancer Res. 1995 Sep;1(9):1051-7.

136. Cowell IG, Okorokov AL, Cutts SA, Padget K, Bell M, Milner J, et al. Human topoisomerase IIalpha and IIbeta interact with the C-terminal region of p53. Exp Cell Res. 2000 Feb 25;255(1):86-94.

137. Dunkern TR, Wedemeyer I, Baumgartner M, Fritz G, Kaina B. Resistance of p53 knockout cells to doxorubicin is related to reduced formation of DNA strand breaks rather than impaired apoptotic signaling. DNA Repair (Amst). 2003 Jan 2;2(1):49-60.

138. Green PS, Leeuwenburgh C. Mitochondrial dysfunction is an early indicator of doxorubicin-induced apoptosis. Biochim Biophys Acta. 2002 Oct 9;1588(1):94-101.

139. Tao Z, Withers HG, Penefsky HS, Goodisman J, Souid AK. Inhibition of cellular respiration by doxorubicin. Chem Res Toxicol. 2006 Aug;19(8):1051-8.

140. Nagata S. Apoptotic DNA fragmentation. Exp Cell Res. 2000 Apr 10;256(1):12-8.

141. Ranney MK, Ahmed IS, Potts KR, Craven RJ. Multiple pathways regulating the anti- apoptotic protein clusterin in breast cancer. Biochim Biophys Acta. 2007 Sep;1772(9):1103-11.

142. Bachur NR, Gee MV, Friedman RD. Nuclear catalyzed antibiotic free radical formation. Cancer Res. 1982 Mar;42(3):1078-81.

143. Whittock NV, Bower C. Genetic evidence for a novel human desmosomal cadherin, desmoglein 4. J Invest Dermatol. 2003 Apr;120(4):523-30.

144. Grantab R, Sivananthan S, Tannock IF. The penetration of anticancer drugs through tumor tissue as a function of cellular adhesion and packing density of tumor cells. Cancer Res. 2006 Jan 15;66(2):1033-9.

145. dit Faute MA, Laurent L, Ploton D, Poupon MF, Jardillier JC, Bobichon H. Distinctive alterations of invasiveness, drug resistance and cell-cell organization in 3D-cultures of MCF-7, a human breast cancer cell line, and its multidrug resistant variant. Clin Exp Metastasis. 2002;19(2):161-8.

146. Dusek RL, Getsios S, Chen F, Park JK, Amargo EV, Cryns VL, et al. The differentiation- dependent desmosomal cadherin desmoglein 1 is a novel caspase-3 target that regulates apoptosis in keratinocytes. J Biol Chem. 2006 Feb 10;281(6):3614-24.

147. Boukhtouche F, Mariani J, Tedgui A. The "CholesteROR" protective pathway in the vascular system. Arterioscler Thromb Vasc Biol. 2004 Apr;24(4):637-43.

153

148. Katoh M. Comparative genomics on ROR1 and ROR2 orthologs. Oncol Rep. 2005 Nov;14(5):1381-4.

149. Dussault I, Giguere V. Differential regulation of the N-myc proto-oncogene by ROR alpha and RVR, two orphan members of the superfamily of nuclear hormone receptors. Mol Cell Biol. 1997 Apr;17(4):1860-7.

150. Zhu Y, McAvoy S, Kuhn R, Smith DI. RORA, a large common fragile site gene, is involved in cellular stress response. Oncogene. 2006 May 11;25(20):2901-8.

151. Dai J, Ram PT, Yuan L, Spriggs LL, Hill SM. Transcriptional repression of RORalpha activity in human breast cancer cells by melatonin. Mol Cell Endocrinol. 2001 May 15;176(1- 2):111-20.

152. Chauvet C, Bois-Joyeux B, Berra E, Pouyssegur J, Danan JL. The gene encoding human retinoic acid-receptor-related orphan receptor alpha is a target for hypoxia-inducible factor 1. Biochem J. 2004 Nov 15;384(Pt 1):79-85.

153. Boukhtouche F, Vodjdani G, Jarvis CI, Bakouche J, Staels B, Mallet J, et al. Human retinoic acid receptor-related orphan receptor alpha1 overexpression protects neurones against oxidative stress-induced apoptosis. J Neurochem. 2006 Mar;96(6):1778-89.

154. Bretscher A, Edwards K, Fehon RG. ERM proteins and merlin: integrators at the cell cortex. Nat Rev Mol Cell Biol. 2002 Aug;3(8):586-99.

155. Gunn-Moore FJ, Welsh GI, Herron LR, Brannigan F, Venkateswarlu K, Gillespie S, et al. A novel 4.1 ezrin radixin moesin (FERM)-containing protein, 'Willin'. FEBS Lett. 2005 Sep 12;579(22):5089-94.

156. Gautreau A, Poullet P, Louvard D, Arpin M. Ezrin, a plasma membrane-microfilament linker, signals cell survival through the phosphatidylinositol 3-kinase/Akt pathway. Proc Natl Acad Sci U S A. 1999 Jun 22;96(13):7300-5.

157. Li Q, Wu M, Wang H, Xu G, Zhu T, Zhang Y, et al. Ezrin silencing by small hairpin RNA reverses metastatic behaviors of human breast cancer cells. Cancer Lett. 2008 Mar 8;261(1):55-63.

158. Yu D, Hung M-c. Breast cancer chemosensitivity. New York

Austin, Tex.: Springer Science+Business Media ;

Landes Bioscience; 2007.

159. Valentijn LJ, Koppen A, van Asperen R, Root HA, Haneveld F, Versteeg R. Inhibition of a new differentiation pathway in neuroblastoma by copy number defects of N-myc, Cdc42, and nm23 genes. Cancer Res. 2005 Apr 15;65(8):3136-45.

160. He AW, Cory JG. Effect of doxorubicin on wild-type and deoxyadenosine-resistant mouse leukemia L1210 cells. Int J Oncol. 1999 May;14(5):891-5.

154

161. Fan S, Wang JA, Yuan RQ, Rockwell S, Andres J, Zlatapolskiy A, et al. Scatter factor protects epithelial and carcinoma cells against apoptosis induced by DNA-damaging agents. Oncogene. 1998 Jul 16;17(2):131-41.

162. Cappellen D, Schlange T, Bauer M, Maurer F, Hynes NE. Novel c-MYC target genes mediate differential effects on cell proliferation and migration. EMBO Rep. 2007 Jan;8(1):70-6.

163. Perego P, Corna E, De Cesare M, Gatti L, Polizzi D, Pratesi G, et al. Role of apoptosis and apoptosis-related genes in cellular response and antitumor efficacy of anthracyclines. Curr Med Chem. 2001 Jan;8(1):31-7.

164. Lau P, Bailey P, Dowhan DH, Muscat GE. Exogenous expression of a dominant negative RORalpha1 vector in muscle cells impairs differentiation: RORalpha1 directly interacts with p300 and myoD. Nucleic Acids Res. 1999 Jan 15;27(2):411-20.

165. Soma D, Kitayama J, Ishigami H, Kaisaki S, Nagawa H. Different tissue distribution of paclitaxel with intravenous and intraperitoneal administration. J Surg Res. 2009 Jul;155(1):142- 6.

166. Marchetti P, Urien S, Cappellini GA, Ronzino G, Ficorella C. Weekly administration of paclitaxel: theoretical and clinical basis. Crit Rev Oncol Hematol. 2002 Dec 27;44 Suppl:S3-13.

167. van Schaik RH. CYP450 pharmacogenetics for personalizing cancer therapy. Drug Resist Updat. 2008 Jun;11(3):77-98.

168. Kitagawa K, Niikura Y. Caspase-independent mitotic death (CIMD). Cell Cycle. 2008 Apr 15;7(8):1001-5.

169. Efferth T, Volm M. Pharmacogenetics for individualized cancer chemotherapy. Pharmacol Ther. 2005 Aug;107(2):155-76.

170. Yuan HY, Chiou JJ, Tseng WH, Liu CH, Liu CK, Lin YJ, et al. FASTSNP: an always up-to-date and extendable service for SNP function analysis and prioritization. Nucleic Acids Res. 2006 Jul 1;34(Web Server issue):W635-41.

171. Wu C, Orozco C, Boyer J, Leglise M, Goodale J, Batalov S, et al. BioGPS: an extensible and customizable portal for querying and organizing gene annotation resources. Genome Biol. 2009;10(11):R130.

172. Zhang W, Liu W, Poradosu E, Ratain MJ. Genome-wide identification of genetic determinants for the cytotoxicity of perifosine. Hum Genomics. 2008 Sep;3(1):53-70.

173. Spratlin J, Sawyer MB. Pharmacogenetics of paclitaxel metabolism. Crit Rev Oncol Hematol. 2007 Mar;61(3):222-9.

174. Green H. Pharmacogenomics of importance for paclitaxel chemotherapy. Pharmacogenomics. 2008 Jun;9(6):671-4.

155

175. Oshita F, Sekiyama A, Saito H, Yamada K, Noda K, Miyagi Y. Genome-wide cDNA microarray screening of genes related to the benefits of paclitaxel and irinotecan chemotherapy in patients with advanced non-small cell lung cancer. J Exp Ther Oncol. 2006;6(1):49-53.

176. Mielke S. Individualized pharmacotherapy with paclitaxel. Curr Opin Oncol. 2007 Nov;19(6):586-9.

177. Sheikh SH, Mulchandani A. Continuous-flow fluoro-immunosensor for paclitaxel measurement. Biosens Bioelectron. 2001 Dec;16(9-12):647-52.

178. Ohtsu T, Sasaki Y, Tamura T, Miyata Y, Nakanomyo H, Nishiwaki Y, et al. Clinical pharmacokinetics and pharmacodynamics of paclitaxel: a 3-hour infusion versus a 24-hour infusion. Clin Cancer Res. 1995 Jun;1(6):599-606.

179. Tang Y, Zhu J, Chen L, Zhang S, Lin J. Associations of matrix metalloproteinase-9 protein polymorphisms with lymph node metastasis but not invasion of gastric cancer. Clin Cancer Res. 2008 May 1;14(9):2870-7.

180. Greenwood PM, Lin MK, Sundararajan R, Fryxell KJ, Parasuraman R. Synergistic effects of genetic variation in nicotinic and muscarinic receptors on visual attention but not working memory. Proc Natl Acad Sci U S A. 2009 Mar 3;106(9):3633-8.

181. Kim JH, Kim H, Lee KY, Kang JW, Lee KH, Park SY, et al. Aryl hydrocarbon receptor gene polymorphisms affect lung cancer risk. Lung Cancer. 2007 Apr;56(1):9-15.

182. Zhao D, Sun T, Zhang X, Guo Y, Yu D, Yang M, et al. Role of CD14 promoter polymorphisms in Helicobacter pylori infection--related gastric carcinoma. Clin Cancer Res. 2007 Apr 15;13(8):2362-8.

183. Gelderblom H, Mross K, ten Tije AJ, Behringer D, Mielke S, van Zomeren DM, et al. Comparative pharmacokinetics of unbound paclitaxel during 1- and 3-hour infusions. J Clin Oncol. 2002 Jan 15;20(2):574-81.

184. Kimelman D, Xu W. beta-catenin destruction complex: insights and questions from a structural perspective. Oncogene. 2006 Dec 4;25(57):7482-91.

185. Klarmann GJ, Decker A, Farrar WL. Epigenetic gene silencing in the Wnt pathway in breast cancer. Epigenetics. 2008 Mar-Apr;3(2):59-63.

186. Tejpar S, Michils G, Denys H, Van Dam K, Nik SA, Jadidizadeh A, et al. Analysis of Wnt/Beta catenin signalling in desmoid tumors. Acta Gastroenterol Belg. 2005 Jan-Mar;68(1):5- 9.

187. Gehrke I, Gandhirajan RK, Kreuzer KA. Targeting the WNT/beta-catenin/TCF/LEF1 axis in solid and haematological cancers: Multiplicity of therapeutic options. Eur J Cancer. 2009 Nov;45(16):2759-67.

156

188. Damalas A, Ben-Ze'ev A, Simcha I, Shtutman M, Leal JF, Zhurinsky J, et al. Excess beta-catenin promotes accumulation of transcriptionally active p53. EMBO J. 1999 Jun 1;18(11):3054-63.

189. Sadot E, Geiger B, Oren M, Ben-Ze'ev A. Down-regulation of beta-catenin by activated p53. Mol Cell Biol. 2001 Oct;21(20):6768-81.

190. Bacus SS, Gudkov AV, Lowe M, Lyass L, Yung Y, Komarov AP, et al. Taxol-induced apoptosis depends on MAP kinase pathways (ERK and p38) and is independent of p53. Oncogene. 2001 Jan 11;20(2):147-55.

191. Giannakakou P, Nakano M, Nicolaou KC, O'Brate A, Yu J, Blagosklonny MV, et al. Enhanced microtubule-dependent trafficking and p53 nuclear accumulation by suppression of microtubule dynamics. Proc Natl Acad Sci U S A. 2002 Aug 6;99(16):10855-60.

192. Stewart ZA, Tang LJ, Pietenpol JA. Increased p53 phosphorylation after microtubule disruption is mediated in a microtubule inhibitor- and cell-specific manner. Oncogene. 2001 Jan 4;20(1):113-24.

193. Tu Y, Wu W, Wu T, Cao Z, Wilkins R, Toh BH, et al. Antiproliferative autoantigen CDA1 transcriptionally up-regulates p21(Waf1/Cip1) by activating p53 and MEK/ERK1/2 MAPK pathways. J Biol Chem. 2007 Apr 20;282(16):11722-31.

194. Grandori C, Cowley SM, James LP, Eisenman RN. The Myc/Max/Mad network and the transcriptional control of cell behavior. Annu Rev Cell Dev Biol. 2000;16:653-99.

195. Helmbold H, Komm N, Deppert W, Bohn W. Rb2/p130 is the dominating pocket protein in the p53-p21 DNA damage response pathway leading to senescence. Oncogene. 2009 Oct 1;28(39):3456-67.

196. Blagosklonny MV, Fojo T. Molecular effects of paclitaxel: myths and reality (a critical review). Int J Cancer. 1999 Oct 8;83(2):151-6.

197. Nawrocki-Raby B, Gilles C, Polette M, Martinella-Catusse C, Bonnet N, Puchelle E, et al. E-Cadherin mediates MMP down-regulation in highly invasive bronchial tumor cells. Am J Pathol. 2003 Aug;163(2):653-61.

198. Kemler R. From cadherins to catenins: cytoplasmic protein interactions and regulation of cell adhesion. Trends Genet. 1993 Sep;9(9):317-21.

199. Levina E, Oren M, Ben-Ze'ev A. Downregulation of beta-catenin by p53 involves changes in the rate of beta-catenin phosphorylation and Axin dynamics. Oncogene. 2004 May 27;23(25):4444-53.

200. Al Moustafa AE, Foulkes WD, Benlimame N, Wong A, Yen L, Bergeron J, et al. E6/E7 proteins of HPV type 16 and ErbB-2 cooperate to induce neoplastic transformation of primary normal oral epithelial cells. Oncogene. 2004 Jan 15;23(2):350-8.

157

201. Schroeder JA, Adriance MC, McConnell EJ, Thompson MC, Pockaj B, Gendler SJ. ErbB-beta-catenin complexes are associated with human infiltrating ductal breast and murine mammary tumor virus (MMTV)-Wnt-1 and MMTV-c-Neu transgenic carcinomas. J Biol Chem. 2002 Jun 21;277(25):22692-8.

202. Walton KM, Dixon JE. Protein tyrosine phosphatases. Annu Rev Biochem. 1993;62:101- 20.

203. Zheng XM, Resnick RJ, Shalloway D. A phosphotyrosine displacement mechanism for activation of Src by PTPalpha. EMBO J. 2000 Mar 1;19(5):964-78.

204. Blanchetot C, Tertoolen LG, Overvoorde J, den Hertog J. Intra- and intermolecular interactions between intracellular domains of receptor protein-tyrosine phosphatases. J Biol Chem. 2002 Dec 6;277(49):47263-9.

205. Janssens V, Van Hoof C, De Baere I, Merlevede W, Goris J. The phosphotyrosyl phosphatase activator gene is a novel p53 target gene. J Biol Chem. 2000 Jul 7;275(27):20488- 95.

206. Noren NK, Pasquale EB. Eph receptor-ephrin bidirectional signals that target Ras and Rho proteins. Cell Signal. 2004 Jun;16(6):655-66.

207. Rahman MA, Ashton AC, Meunier FA, Davletov BA, Dolly JO, Ushkaryov YA. Norepinephrine exocytosis stimulated by alpha-latrotoxin requires both external and stored Ca2+ and is mediated by latrophilin, G proteins and phospholipase C. Philos Trans R Soc Lond B Biol Sci. 1999 Feb 28;354(1381):379-86.

208. Tobaben S, Sudhof TC, Stahl B. The G protein-coupled receptor CL1 interacts directly with proteins of the Shank family. J Biol Chem. 2000 Nov 17;275(46):36204-10.

209. Nishimura W, Yao I, Iida J, Tanaka N, Hata Y. Interaction of synaptic scaffolding molecule and Beta -catenin. J Neurosci. 2002 Feb 1;22(3):757-65.

210. Matsushita H, Lelianova VG, Ushkaryov YA. The latrophilin family: multiply spliced G protein-coupled receptors with differential tissue distribution. FEBS Lett. 1999 Jan 29;443(3):348-52.

211. Jane DE, Lodge D, Collingridge GL. Kainate receptors: pharmacology, function and therapeutic potential. Neuropharmacology. 2009 Jan;56(1):90-113.

212. Morkel M, Huelsken J, Wakamiya M, Ding J, van de Wetering M, Clevers H, et al. Beta- catenin regulates Cripto- and Wnt3-dependent gene expression programs in mouse axis and mesoderm formation. Development. 2003 Dec;130(25):6283-94.

213. Kayadjanian N, Lee HS, Pina-Crespo J, Heinemann SF. Localization of glutamate receptors to distal dendrites depends on subunit composition and the kinesin motor protein KIF17. Mol Cell Neurosci. 2007 Feb;34(2):219-30.

158

214. Alli E, Yang JM, Ford JM, Hait WN. Reversal of stathmin-mediated resistance to paclitaxel and vinblastine in human breast carcinoma cells. Mol Pharmacol. 2007 May;71(5):1233-40.

215. Luthi-Carter R, Strand AD, Hanson SA, Kooperberg C, Schilling G, La Spada AR, et al. Polyglutamine and transcription: gene expression changes shared by DRPLA and Huntington's disease mouse models reveal context-independent effects. Hum Mol Genet. 2002 Aug 15;11(17):1927-37.

216. Nguyen-Ba-Charvet KT, Chedotal A. Role of Slit proteins in the vertebrate brain. J Physiol Paris. 2002 Jan-Mar;96(1-2):91-8.

217. Chang C, Yu TW, Bargmann CI, Tessier-Lavigne M. Inhibition of netrin-mediated axon attraction by a receptor protein tyrosine phosphatase. Science. 2004 Jul 2;305(5680):103-6.

218. Conde C, Caceres A. Microtubule assembly, organization and dynamics in axons and dendrites. Nat Rev Neurosci. 2009 May;10(5):319-32.

219. Ongusaha PP, Ouchi T, Kim KT, Nytko E, Kwak JC, Duda RB, et al. BRCA1 shifts p53- mediated cellular outcomes towards irreversible growth arrest. Oncogene. 2003 Jun 12;22(24):3749-58.

220. Hollande E, Salvador-Cartier C, Alvarez L, Fanjul M. Expression of a wild-type CFTR maintains the integrity of the biosynthetic/secretory pathway in human cystic fibrosis pancreatic duct cells. J Histochem Cytochem. 2005 Dec;53(12):1539-52.

221. Norez C, Noel S, Wilke M, Bijvelds M, Jorna H, Melin P, et al. Rescue of functional delF508-CFTR channels in cystic fibrosis epithelial cells by the alpha-glucosidase inhibitor miglustat. FEBS Lett. 2006 Apr 3;580(8):2081-6.

222. Toyofuku K, Wada I, Valencia JC, Kushimoto T, Ferrans VJ, Hearing VJ. Oculocutaneous albinism types 1 and 3 are ER retention diseases: mutation of tyrosinase or Tyrp1 can affect the processing of both mutant and wild-type proteins. FASEB J. 2001 Oct;15(12):2149-61.

223. Lim J, Hao T, Shaw C, Patel AJ, Szabo G, Rual JF, et al. A protein-protein interaction network for human inherited ataxias and disorders of Purkinje cell degeneration. Cell. 2006 May 19;125(4):801-14.

224. Wang X, Venable J, LaPointe P, Hutt DM, Koulov AV, Coppinger J, et al. Hsp90 cochaperone Aha1 downregulation rescues misfolding of CFTR in cystic fibrosis. Cell. 2006 Nov 17;127(4):803-15.

225. Pak BJ, Lee J, Thai BL, Fuchs SY, Shaked Y, Ronai Z, et al. Radiation resistance of human melanoma analysed by retroviral insertional mutagenesis reveals a possible role for dopachrome tautomerase. Oncogene. 2004 Jan 8;23(1):30-8.

159

226. Yasumoto K, Takeda K, Saito H, Watanabe K, Takahashi K, Shibahara S. Microphthalmia-associated transcription factor interacts with LEF-1, a mediator of Wnt signaling. EMBO J. 2002 Jun 3;21(11):2703-14.

227. Michard Q, Commo S, Belaidi JP, Alleaume AM, Michelet JF, Daronnat E, et al. TRP-2 specifically decreases WM35 cell sensitivity to oxidative stress. Free Radic Biol Med. 2008 Mar 15;44(6):1023-31.

228. Timar J, Meszaros L, Ladanyi A, Puskas LG, Raso E. Melanoma genomics reveals signatures of sensitivity to bio- and targeted therapies. Cell Immunol. 2006 Dec;244(2):154-7.

229. Pan W, Jia Y, Wang J, Tao D, Gan X, Tsiokas L, et al. Beta-catenin regulates myogenesis by relieving I-mfa-mediated suppression of myogenic regulatory factors in P19 cells. Proc Natl Acad Sci U S A. 2005 Nov 29;102(48):17378-83.

230. Tassin AM, Maro B, Bornens M. Fate of microtubule-organizing centers during myogenesis in vitro. J Cell Biol. 1985 Jan;100(1):35-46.

231. Liu LA, Engvall E. Sarcoglycan isoforms in skeletal muscle. J Biol Chem. 1999 Dec 31;274(53):38171-6.

232. Hashimoto R, Yamaguchi M. Dynamic changes in the subcellular localization of Drosophila beta-sarcoglycan during the cell cycle. Cell Struct Funct. 2006;31(2):173-80.

233. Lumeng C, Phelps S, Crawford GE, Walden PD, Barald K, Chamberlain JS. Interactions between beta 2-syntrophin and a family of microtubule-associated serine/threonine kinases. Nat Neurosci. 1999 Jul;2(7):611-7.

234. Hogan A, Shepherd L, Chabot J, Quenneville S, Prescott SM, Topham MK, et al. Interaction of gamma 1-syntrophin with diacylglycerol kinase-zeta. Regulation of nuclear localization by PDZ interactions. J Biol Chem. 2001 Jul 13;276(28):26526-33.

235. Dephoure N, Zhou C, Villen J, Beausoleil SA, Bakalarski CE, Elledge SJ, et al. A quantitative atlas of mitotic phosphorylation. Proc Natl Acad Sci U S A. 2008 Aug 5;105(31):10762-7.

236. Sjoblom T, Jones S, Wood LD, Parsons DW, Lin J, Barber TD, et al. The consensus coding sequences of human breast and colorectal cancers. Science. 2006 Oct 13;314(5797):268- 74.

237. Scian MJ, Stagliano KE, Deb D, Ellis MA, Carchman EH, Das A, et al. Tumor-derived p53 mutants induce oncogenesis by transactivating growth-promoting genes. Oncogene. 2004 May 27;23(25):4430-43.

238. Munoz IM, Hain K, Declais AC, Gardiner M, Toh GW, Sanchez-Pulido L, et al. Coordination of structure-specific nucleases by human SLX4/BTBD12 is required for DNA repair. Mol Cell. 2009 Jul 10;35(1):116-27.

160

239. Svendsen JM, Smogorzewska A, Sowa ME, O'Connell BC, Gygi SP, Elledge SJ, et al. Mammalian BTBD12/SLX4 assembles a Holliday junction resolvase and is required for DNA repair. Cell. 2009 Jul 10;138(1):63-77.

240. Swanton C, Nicke B, Schuett M, Eklund AC, Ng C, Li Q, et al. Chromosomal instability determines taxane response. Proc Natl Acad Sci U S A. 2009 May 26;106(21):8671-6.

241. Yin W, Rossin A, Clifford JL, Gronemeyer H. Co-resistance to retinoic acid and TRAIL by insertion mutagenesis into RAM. Oncogene. 2006 Jun 22;25(26):3735-44.

242. Shete S, Hosking FJ, Robertson LB, Dobbins SE, Sanson M, Malmer B, et al. Genome- wide association study identifies five susceptibility loci for glioma. Nat Genet. 2009 Aug;41(8):899-904.

243. Niu MY, Menard M, Reed JC, Krajewski S, Pratt MA. Ectopic expression of cyclin D1 amplifies a retinoic acid-induced mitochondrial death pathway in breast cancer cells. Oncogene. 2001 Jun 14;20(27):3506-18.

244. Das KC, Guo XL, White CW. Protein kinase Cdelta-dependent induction of manganese superoxide dismutase gene expression by microtubule-active anticancer drugs. J Biol Chem. 1998 Dec 18;273(51):34639-45.

245. Ding AH, Porteu F, Sanchez E, Nathan CF. Shared actions of endotoxin and taxol on TNF receptors and TNF release. Science. 1990 Apr 20;248(4953):370-2.

246. Manthey CL, Brandes ME, Perera PY, Vogel SN. Taxol increases steady-state levels of lipopolysaccharide-inducible genes and protein-tyrosine phosphorylation in murine macrophages. J Immunol. 1992 Oct 1;149(7):2459-65.

247. Kirikae F, Kirikae T, Qureshi N, Takayama K, Morrison DC, Nakano M. CD14 is not involved in Rhodobacter sphaeroides diphosphoryl lipid A inhibition of tumor necrosis factor alpha and nitric oxide induction by taxol in murine macrophages. Infect Immun. 1995 Feb;63(2):486-97.

248. Doroshow JH, Davies KJ. Redox cycling of anthracyclines by cardiac mitochondria. II. Formation of superoxide anion, hydrogen peroxide, and hydroxyl radical. J Biol Chem. 1986 Mar 5;261(7):3068-74.

249. Kostoryz EL, Yourtee DM. Oxidative mutagenesis of doxorubicin-Fe(III) complex. Mutat Res. 2001 Feb 20;490(2):131-9.

250. Shim GS, Manandhar S, Shin DH, Kim TH, Kwak MK. Acquisition of doxorubicin resistance in ovarian carcinoma cells accompanies activation of the NRF2 pathway. Free Radic Biol Med. 2009 Dec 1;47(11):1619-31.

251. Liu B, Chen Y, St Clair DK. ROS and p53: a versatile partnership. Free Radic Biol Med. 2008 Apr 15;44(8):1529-35.

161

252. Kobayashi M, Yamamoto M. Molecular mechanisms activating the Nrf2-Keap1 pathway of antioxidant gene regulation. Antioxid Redox Signal. 2005 Mar-Apr;7(3-4):385-94.

253. Polyak K, Xia Y, Zweier JL, Kinzler KW, Vogelstein B. A model for p53-induced apoptosis. Nature. 1997 Sep 18;389(6648):300-5.

254. Seemann S, Hainaut P. Roles of thioredoxin reductase 1 and APE/Ref-1 in the control of basal p53 stability and activity. Oncogene. 2005 Jun 2;24(24):3853-63.

255. Sayan BS, Sayan AE, Knight RA, Melino G, Cohen GM. p53 is cleaved by caspases generating fragments localizing to mitochondria. J Biol Chem. 2006 May 12;281(19):13566-73.

256. Marchenko ND, Zaika A, Moll UM. Death signal-induced localization of p53 protein to mitochondria. A potential role in apoptotic signaling. J Biol Chem. 2000 May 26;275(21):16202- 12.

257. Tomko RJ, Jr., Bansal P, Lazo JS. Airing out an antioxidant role for the tumor suppressor p53. Mol Interv. 2006 Feb;6(1):23-5, 2.

258. Dennis G, Jr., Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, et al. DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol. 2003;4(5):P3.

259. Huang da W, Sherman BT, Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc. 2009;4(1):44-57.

260. Chiarugi P, Fiaschi T. Redox signalling in anchorage-dependent cell growth. Cell Signal. 2007 Apr;19(4):672-82.

261. Rangasamy T, Cho CY, Thimmulappa RK, Zhen L, Srisuma SS, Kensler TW, et al. Genetic ablation of Nrf2 enhances susceptibility to cigarette smoke-induced emphysema in mice. J Clin Invest. 2004 Nov;114(9):1248-59.

262. Popovici V, Chen W, Gallas BG, Hatzis C, Shi W, Samuelson FW, et al. Effect of training-sample size and classification difficulty on the accuracy of genomic predictors. Breast Cancer Res.12(1):R5.

263. Delerive P, Monte D, Dubois G, Trottein F, Fruchart-Najib J, Mariani J, et al. The orphan nuclear receptor ROR alpha is a negative regulator of the inflammatory response. EMBO Rep. 2001 Jan;2(1):42-8.

264. Tomaru Y, Nakanishi M, Miura H, Kimura Y, Ohkawa H, Ohta Y, et al. Identification of an inter-transcription factor regulatory network in human hepatoma cells by Matrix RNAi. Nucleic Acids Res. 2009 Mar;37(4):1049-60.

265. Galson DL, Tsuchiya T, Tendler DS, Huang LE, Ren Y, Ogura T, et al. The orphan receptor hepatic nuclear factor 4 functions as a transcriptional activator for tissue-specific and hypoxia-specific erythropoietin gene expression and is antagonized by EAR3/COUP-TF1. Mol Cell Biol. 1995 Apr;15(4):2135-44.

162

266. Yang XO, Pappu BP, Nurieva R, Akimzhanov A, Kang HS, Chung Y, et al. T helper 17 lineage differentiation is programmed by orphan nuclear receptors ROR alpha and ROR gamma. Immunity. 2008 Jan;28(1):29-39.

267. Huber M, Heink S, Grothe H, Guralnik A, Reinhard K, Elflein K, et al. A Th17-like developmental process leads to CD8(+) Tc17 cells with reduced cytotoxic activity. Eur J Immunol. 2009 Jul;39(7):1716-25.

268. Novac O, Alvarez D, Pearson CE, Price GB, Zannis-Hadjopoulos M. The human cruciform-binding protein, CBP, is involved in DNA replication and associates in vivo with mammalian replication origins. J Biol Chem. 2002 Mar 29;277(13):11174-83.

269. Jin J, Smith FD, Stark C, Wells CD, Fawcett JP, Kulkarni S, et al. Proteomic, functional, and domain-based analysis of in vivo 14-3-3 binding proteins involved in cytoskeletal regulation and cellular organization. Curr Biol. 2004 Aug 24;14(16):1436-50.

270. Lemberger T, Parkitna JR, Chai M, Schutz G, Engblom D. CREB has a context- dependent role in activity-regulated transcription and maintains neuronal cholesterol homeostasis. FASEB J. 2008 Aug;22(8):2872-9.

271. Ozgen N, Guo J, Gertsberg Z, Danilo P, Jr., Rosen MR, Steinberg SF. Reactive oxygen species decrease cAMP response element binding protein expression in cardiomyocytes via a protein kinase D1-dependent mechanism that does not require Ser133 phosphorylation. Mol Pharmacol. 2009 Oct;76(4):896-902.

272. Lee B, Cao R, Choi YS, Cho HY, Rhee AD, Hah CK, et al. The CREB/CRE transcriptional pathway: protection against oxidative stress-mediated neuronal cell death. J Neurochem. 2009 Mar;108(5):1251-65.

273. Taravath S, Tonsgard JH. Cerebral malformations in Carpenter syndrome. Pediatr Neurol. 1993 May-Jun;9(3):230-4.

274. Kopetz S, Lesslie DP, Dallas NA, Park SI, Johnson M, Parikh NU, et al. Synergistic activity of the SRC family kinase inhibitor dasatinib and oxaliplatin in colon carcinoma cells is mediated by oxidative stress. Cancer Res. 2009 May 1;69(9):3842-9.

275. Guarino M. Src signaling in cancer invasion. J Cell Physiol. Apr;223(1):14-26.

276. Wei L, Yang Y, Zhang X, Yu Q. Altered regulation of Src upon cell detachment protects human lung adenocarcinoma cells from anoikis. Oncogene. 2004 Dec 2;23(56):9052-61.

277. Giannoni E, Fiaschi T, Ramponi G, Chiarugi P. Redox regulation of anoikis resistance of metastatic prostate cancer cells: key role for Src and EGFR-mediated pro-survival signals. Oncogene. 2009 May 21;28(20):2074-86.

278. Sundaresan V, Chung G, Heppell-Parton A, Xiong J, Grundy C, Roberts I, et al. Homozygous deletions at 3p12 in breast and lung cancer. Oncogene. 1998 Oct 1;17(13):1723-9.

163

279. Brouillard F, Bouthier M, Leclerc T, Clement A, Baudouin-Legros M, Edelman A. NF- kappa B mediates up-regulation of CFTR gene expression in Calu-3 cells by interleukin-1beta. J Biol Chem. 2001 Mar 23;276(12):9486-91.

280. Pongnimitprasert N, El-Benna J, Foglietti MJ, Gougerot-Pocidalo MA, Bernard M, Braut-Boucher F. Potential role of the "NADPH oxidases" (NOX/DUOX) family in cystic fibrosis. Ann Biol Clin (Paris). 2008 Nov-Dec;66(6):621-9.

281. Chen J, Kinter M, Shank S, Cotton C, Kelley TJ, Ziady AG. Dysfunction of Nrf-2 in CF epithelia leads to excess intracellular H2O2 and inflammatory cytokine production. PLoS One. 2008;3(10):e3367.

282. Hayashi T, Ueno Y, Okamoto T. Oxidoreductive regulation of nuclear factor kappa B. Involvement of a cellular reducing catalyst thioredoxin. J Biol Chem. 1993 May 25;268(15):11380-8.

283. Ntimbane T, Comte B, Mailhot G, Berthiaume Y, Poitout V, Prentki M, et al. Cystic Fibrosis-Related Diabetes: From CFTR Dysfunction to Oxidative Stress. Clin Biochem Rev. 2009 Nov;30(4):153-77.

284. White GR, Varley JM, Heighway J. Isolation and characterization of a human homologue of the latrophilin gene from a region of 1p31.1 implicated in breast cancer. Oncogene. 1998 Dec 31;17(26):3513-9.

285. White GR, Varley JM, Heighway J. Genomic structure and expression profile of LPHH1, a 7TM gene variably expressed in breast cancer cell lines. Biochim Biophys Acta. 2000 Apr 25;1491(1-3):75-92.

286. Papaiahgari S, Kleeberger SR, Cho HY, Kalvakolanu DV, Reddy SP. NADPH oxidase and ERK signaling regulates hyperoxia-induced Nrf2-ARE transcriptional response in pulmonary epithelial cells. J Biol Chem. 2004 Oct 1;279(40):42302-12.

287. Pusztai L, Cristofanilli M, Paik S. New generation of molecular prognostic and predictive tests for breast cancer. Semin Oncol. 2007 Apr;34(2 Suppl 3):S10-6.

288. Nilsson K, Ponten J. Classification and biological nature of established human hematopoietic cell lines. Int J Cancer. 1975 Feb 15;15(2):321-41.

289. Feng WH, Hong G, Delecluse HJ, Kenney SC. Lytic induction therapy for Epstein-Barr virus-positive B-cell lymphomas. J Virol. 2004 Feb;78(4):1893-902.

164

Appendix

This is a reprint of work done earlier in PhD training. This article is reprinted from:

Human Mutation 29(4), 461-467, 2008

165

166

167

168

169

170

171

172

173

174

175

176

177

178