I Genetic Determinants of Cancer Cell Survival in Tumor

Genetic Determinants of Cancer Cell Survival in Tumor Microenvironment Stresses

Melissa Marie Keenan

University Program in Genetics and Genomics Duke University

Date:______Approved:

______Jen-Tsan Ashley Chi, Supervisor

______Jack Keene

______James Koh

______Deborah M. Muoio

______Jeffrey Rathmell

Dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the University Program in Genetics and Genomics in the Graduate School of Duke University

2015

ABSTRACT

Genetic Determinants of Cancer Cell Survival in Tumor Microenvironment Stresses

Melissa Marie Keenan

University Program in Genetics and Genomics Duke University

Date:______Approved:

______Jen-Tsan Ashley Chi, Supervisor

______Jack Keene

______James Koh

______Deborah M. Muoio

______Jeffrey Rathmell

An abstract of a dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the University Program in Genetics and Genomics in the Graduate School of Duke University

2015

Abstract

In order to propagate a solid tumor, cancer cells must adapt to and survive under various tumor microenvironment (TME) stresses, such as hypoxia or lactic acidosis.

Additionally, cancer cells exposed to these stresses are more resistant to therapies, more likely to metastasize and often are worse for patient prognosis. While the presence of these stresses is generally negative for cancer patients, since these stresses are mostly unique to the TME, they also offer an opportunity to develop more selective therapeutics. If we achieve a better understanding of the adaptive mechanisms cancer cells employ to survive the TME stresses, then hopefully we, as a scientific community, can devise more effective cancer therapeutics specifically targeting cancer cells under stress. To systematically identify genes that modulate cancer cell survival under stresses, we performed shRNA screens under hypoxia or lactic acidosis. From these screens, we discovered that genetic depletion of acetyl-CoA carboxylase alpha (ACACA or ACC1) or

ATP citrate lyase (ACLY) protected cancer cells from hypoxia-induced apoptosis.

Furthermore, the loss of ACLY or ACC1 reduced the levels and activities of the oncogenic transcription factor ETV4. Silencing ETV4 also protected cells from hypoxia- induced apoptosis and led to remarkably similar transcriptional responses as with silenced ACLY or ACC1, including an anti-apoptotic program. Metabolomic analysis found that while α-ketoglutarate levels decrease under hypoxia in control cells, α-

ketoglutarate was paradoxically increased under hypoxia when ACC1 or ACLY were depleted. Supplementation with α-ketoglutarate rescued the hypoxia-induced apoptosis and recapitulated the decreased expression and activity of ETV4, likely via an epigenetic mechanism. Therefore, ACC1 and ACLY regulated the levels of ETV4 under hypoxia via increased α-ketoglutarate. These results reveal that the ACC1/ACLY-α-ketoglutarate-

ETV4 axis is a novel means by which metabolic states regulate transcriptional output for life vs. death decisions under hypoxia. Since many lipogenic inhibitors are under investigation as cancer therapeutics, our findings suggest that the use of these inhibitors will need to be carefully considered with respect to oncogenic drivers, tumor hypoxia, progression and dormancy. More broadly, our screen provides a framework for studying additional tumor cell stress-adaption mechanisms in the future.

Contents

Abstract ...... iv

List of Tables ...... xi

List of Figures ...... xii

Acknowledgements ...... xiv

1. Introduction ...... 1

1.1 Tumor Cells ...... 3

1.2 Stresses of the TME ...... 4

1.2.1 Glucose limitation ...... 5

1.2.2 Amino acid limitation ...... 7

1.2.3 Biophysical or mechanical stresses ...... 8

1.2.4 Lactic Acidosis ...... 10

1.2.5 Oxygen limitation (hypoxia) ...... 14

1.3 Functional genomics through RNA interference screens ...... 22

1.3.1 Introduction ...... 22

1.3.2 Designing an RNAi screen experiment ...... 22

1.3.3 Advantages and Disadvantages of RNAi screens ...... 26

1.3.4 Previous applications to cancer and the TME stresses ...... 29

1.4 Overview of Chapters ...... 33

2. Positive Selection Screen under Lactic Acidosis ...... 35

2.1 Introduction ...... 35

2.2 Methods ...... 36

2.2.1 Positive Selection Screen ...... 36

2.2.2 Cell culture, TME stress treatments and generation of stable shRNA cell lines ...... 38

2.2.3 Crystal violet staining ...... 39

2.2.4 Determination of cell number ...... 39

2.2.5 Flow cytometry ...... 39

2.2.6 Protein lysate collection and Western blots ...... 40

2.2.7 Quantitative real-time PCR ...... 40

2.3 Initial Results ...... 41

2.4 Validation of Results ...... 43

2.4.1 SEL1L ...... 45

2.4.2 RNF123 ...... 49

2.4.3 LIMD1 ...... 52

2.4.4 Other candidates of interest ...... 56

2.5 Future considerations ...... 57

3. Genome-wide Functional Genomic Screens under Hypoxia and Lactic Acidosis ...... 58

3.1 Introduction ...... 58

3.2 Methods ...... 59

3.2.1 Genome-wide pooled shRNA screen ...... 59

3.2.2 Cell culture, TME stress treatments and generation of stable shRNA cell lines ...... 62

3.2.3 Crystal violet staining ...... 63

vii

3.2.4 Determination of cell number ...... 63

3.2.5 Flow cytometry ...... 64

3.2.6 Protein lysate collection and Western blots ...... 64

3.2.7 Quantitative real-time PCR ...... 65

3.3 Analysis of genome-wide shRNA pooled screens ...... 65

3.3.1 Multiple Hairpin Analysis ...... 65

3.3.2 RIGER Analysis ...... 74

3.3.3 Integration of multiple hairpin changes with RIGER analysis ...... 83

3.3.4 Integration of gene expression data with multiple hairpin analysis ...... 88

3.4 Validation of Results ...... 92

3.4.1 ACACA ...... 94

3.4.2 ETV4 ...... 98

3.4.3 STK39 ...... 98

3.4.4 MLF1 ...... 104

3.4.5 SART1 ...... 108

3.5 Future considerations ...... 111

4. The Role of ACC1 in Hypoxic Cancer Cell Survival ...... 113

4.1 Introduction ...... 113

4.1.1 Lipogenic enzymes in cancer cell metabolism ...... 113

4.1.2 Lipogenesis under stress ...... 119

4.1.3 ETV4: A PEA3 family member ...... 120

4.1.3.1 An overview of the PEA3 Family ...... 120

viii

4.1.3.2 Regulation of ETV4 activity ...... 121

4.1.3.3 Role of ETV4 in normal physiology and cancer biology ...... 123

4.2 Methods ...... 124

4.2.1 Cell culture, TME stress treatments and generation of stable shRNA cell lines ...... 124

4.2.2 Crystal violet staining ...... 126

4.2.3 Determination of cell number ...... 126

4.2.4 Flow cytometry ...... 127

4.2.5 Protein lysate collection and Western blots ...... 127

4.2.6 Quantitative real-time PCR ...... 128

4.2.7 Microarrays and analysis ...... 128

4.2.8 Statistical Analysis ...... 129

4.2.9 Metabolomics profiling and analysis ...... 130

4.2.10 NAP+/NADPH measurements ...... 131

4.2.11 Chromatin Immunoprecipitation ...... 131

4.2.12 DNA methylation ...... 132

4.3 Investigating the specificity of the protective phenotype of ACC1 depletion ..... 133

4.4 Loss of ACC1 or ACLY did not protect cells from hypoxia-induced apoptosis through NADPH conservation to relieve oxidative stress ...... 137

4.5 Depletion of the transcription factor ETV4 led to similar hypoxia-protection and gene expression phenotypes as with the loss of ACC1 or ACLY ...... 141

4.6 Global metabolomics revealed that hypoxia-induced elevated α-ketoglutarate levels protected cells from apoptosis ...... 152

4.7 Hypoxia-induced increased α-ketoglutarate levels regulated ETV4, possibly through 2-oxoglutarate/Fe(II)-dependent dioxygenases ...... 157

5. Discussion ...... 169

5.1 Model of tumorigenic potential versus stress survival ...... 170

5.2 Implications of elevated α-ketoglutarate in hypoxia with inhibition of ACC1 or ACLY ...... 174

5.2.1 Source of elevated α-ketoglutarate in hypoxia with inhibition of ACC1 or ACLY ...... 174

5.2.2 α-ketoglutarate implications for stress response vs. oncogenesis ...... 176

5.2.3 Implications of α-ketoglutarate results with the “oncometabolite” 2- hydroxyglutarate (2-HG) ...... 177

5.3 Potential role of 2-OGDDs in hypoxia survival and implications for cancer therapies ...... 178

5.4 Implications for the role of ETV4 in cancer biology and therapeutics ...... 182

5.5 Implications for inhibitors of ACC1 and ACLY in cancer therapies ...... 184

5.6 Normoxic post-translational regulation of ETV4 by ACC1 or ACLY ...... 188

5.7 Conclusion ...... 194

Appendix A. Annotation of Multiple Hairpin Hits from genome-wide pooled shRNA screens...... 196

Appendix B. The potential gender bias of ETV4 mRNA up-regulation in TCGA datasets ...... 201

References ...... 205

Biography ...... 240

List of Tables

Table 1. Candidate hits from positive selection RNAi screen under LA in H1975 cells ... 42

Table 2. Experimental validation status and results of the 13 candidates from the Positive Selection Screen under LA ...... 44

Table 3. Overview of results from Multiple Hairpin Analysis ...... 69

Table 4. Genes that were Multiple Hairpin Hits (MHHs) in more than one category ...... 72

Table 5. Top 20 genes from RIGER analysis of the hypoxia survival condition ...... 79

Table 6. Top 20 genes from RIGER analysis of the hypoxia synthetic lethal condition .... 80

Table 7. Top 20 genes from RIGER analysis of the LA survival condition ...... 81

Table 8. Top 20 genes from RIGER analysis of the LA synthetic lethal condition ...... 82

Table 9. Candidate hits under hypoxia in both the multiple hairpin analysis and in the top 100 genes of the RIGER analysis, separated by gene knockdown phenotype ...... 84

Table 10. Candidate hits in LA in both the multiple hairpin analysis and in the top 100 genes in the RIGER analysis, separated by gene knockdown phenotype ...... 85

Table 11. Summary of validation work from genome-wide pooled screen ...... 93

Table 12. List of Candidates Identified in Multiple Hairpin Analysis ...... 196

Table 13. Summary of increased ETV4 mRNA expression in different cancer types, considering gender ...... 203

Table 14. Calculating significance of ETV4 mRNA upregulation in TCGA datasets by gender with Chi squared test ...... 204

List of Figures

Figure 1. Validation of SEL1L depletion as protective under LA...... 48

Figure 2. Validation of RNF123 depletion as protective under lactic acidosis...... 50

Figure 3. Validation of LIMD1 loss as protecitve under LA...... 53

Figure 4. Quality control analysis of shRNA microarray and screens...... 68

Figure 5. Overview of genome-wide shRNA screen strategy and results...... 86

Figure 6. Depletion of ACC1 protects cells from hypoxia-induced apoptosis...... 96

Figure 7. Depletion of ETV4 protects cells from hypoxia-induced apoptosis...... 97

Figure 8. Validation of STK39 depletion as synthetic sick under LA...... 103

Figure 9. Validation of MLF1 depletion as synthetic sick under LA...... 107

Figure 10. Validation of SART1 depletion as synthetic sick under LA...... 109

Figure 11. Model of enzymes and metabolites important for de novo lipogenesis...... 118

Figure 12. The depletion of ACC1, but not ACC2, protects cells specifically under a hypoxic stress...... 135

Figure 13. Effect of depleted ACC1 or ACLY on levels of HIF-1α...... 138

Figure 14. Depletion of ACLY protects from hypoxia-induced apoptosis...... 139

Figure 15. Loss of ACC1 does not protect cells from hypoxia-induced apoptosis through preservation of NADPH to relieve oxidative stress...... 140

Figure 16. Depletion of ACC1 or ACLY led to loss of ETV4 protein levels under hypoxia...... 143

Figure 17. Global transcriptional response to depletion of ACC1, ACLY or ETV4 is highly similar...... 147

Figure 18. PLEC and DUSP6 may be direct transcriptional targets of ETV4...... 149

xii

Figure 19. ACC1-altered genes likely represent both ETV4-dependent and -independent transcriptional targets...... 150

Figure 20. Metabolomics assay accurately reflects expected changes and the gene expression effects of different doses of α-KG...... 154

Figure 21. Loss of ACC1 or ACLY results in elevated levels of α-KG under hypoxia that protects from hypoxia-induced apoptosis...... 156

Figure 22. α–ketoglutarate affects ETV4 levels and activity possibly through histone demethylase 2-OGDDs...... 158

Figure 23. Effects of succinate or α–KG supplementation on survival, ETV4 regulation and epigenetic changes...... 161

Figure 24. Global and ETV4-specific histone methylation changes in shACC1 and shACLY cells...... 166

Figure 25. Model of how the ACLY-ACC1-KG-ETV4 axis protects from hypoxia-induced apoptosis...... 168

xiii

Acknowledgements

My dissertation is the culmination of work, advice and support from many people over the past six and a half years.

First, I would like to thank the Chi Lab. Ashley Chi has been a model mentor, always exemplifying outstanding hard work, pushing me to think critically and supporting my research and ideas. Jianli Wu has been a great lab manager and she was instrumental to completing both the positive and genome-wide screens. The rest of the

Chi lab members, past and current, have been wonderful people with which to share my days (and nights and weekends) and are some of the hardest working people I know.

My project would not have been the story it is without the help of many other people at Duke. So Young Kim and Beiyu Liu were essential to us being able to do this screen project and were extremely helpful during analysis of the screen and design of validation strategies. So Young supported and guided our story from the first to the last day. I am incredibly thankful for her patience and knowledgeable advice. Derek Cyr and

Joseph Lucas conducted the quantile normalization for the genome-wide screen.

Deborah Muoio’s guidance on the direction of the ACC1 story and assistance in designing the metabolomics experiments was critical. Olga Ilkayeva, Robert Stevens and

Dorothy Slentz executed the mass spectrometry experiments. Susan Murphy, Zhiqing

Huang and Carole Greiener performed the DNA methylation analysis. Xiaohu Tang and

xiv

Laura Tollini of the Chi Lab were critical in conducting experiments and advising the direction of research. Constructs for investigating the structure-function analyses for

LIMD1 were kindly provided as a gift from Tyson Sharp, now at the Barts Cancer

Institute in London. I greatly appreciate everyone’s scientific advice, technical skills and work for the completion of my PhD research.

I would like to thank my dissertation committee members for guiding me throughout this process: Ashley Chi, Deborah Muoio, Jack Keene, James Koh and Jeffrey

Rathmell. Their suggestions and advice were critical to my publication of a strong, novel scientific story and I appreciate their time, thoughts and valuable advice.

I would have been lost in graduate school if not for the support and endless question-answering by UPGG’s DGS, Allison Ashley-Koch, UPGG’s DGSAs Elizabeth

Labriola, Carolyn Weinbaum and Leslie Mavengere, and CMB’s DGSAs Jessica Rowland and Carol Richardson.

Personally, I would first like to thank my friends. I am thankful to have shared the ups and downs of graduate school, and the life that happened while we were here, with an exceptional group of talented, intelligent strong ladies: Katie Kretovich, Audrey

Bone, Jessica Robertson, Lauren Lohmer, Meghan Morrissey and Kyla Ost. Thank you to my college and high school friends who have been there for me from a far during graduate school, but have always been available for a weekend of fun when we did have the chance to see each other.

I would like to thank my new family, The Keenan’s. They have happily brought me into their family (#26!) and I am grateful for their love and patience when I talk about science. Especially to my parents-in-law Walter and Monica, for making the big trip out to Durham for my defense, thank you for being there.

Thank you to my family, The Kirkpatrick’s. I know I would not have finished my

PhD without their support. Thank you Kelly, Sharon and Dad for listening whenever I talk about my science, whether it was good or bad. Thank you to my Dad and Sharon for inspiring me to be a scientist, but reminding me that life is more important than my experiments. Thank you Kelly for being the best sister anyone could ever have. You always impress me and I’m just trying to keep up with you!

Thank you to my Mom, who even though she couldn’t be here to watch me finish, I know she is proud of me and still watching out for me. She always knew the right thing to say to help me get through the good and bad days. I hope that through doing cancer biology research I will help someone else get to spend longer with his or her mom one day.

Finally, thank you to my immediate family, my dog, Liam, and my husband,

Sean. Even though he’ll never know, Liam has been one of my best friends since the day we brought him home. His furry face and happiness to see me always make me smile and made the end of even the hardest days better. To Sean, my husband, thank you from the bottom of my heart. I know I could not have done this without you. Thank you for

xvi

always supporting my goal to get this PhD, for giving up weekend trips, and being ok with long days in lab. Thank you for cooking dinner, listening and learning about my project and always telling me I could do it. I am so thankful for your love and cannot wait for our lives together.

xvii

1. Introduction

One major division in cancer biology is the separation of solid tumors and non- solid tumor disease. The majority of cancer types are solid tumors. As a solid tumor develops, a number of changes occur within it and the immediately surrounding tissue.

These changes define a unique area in and around the proliferating cancer cells as the

“tumor microenvironment” or TME.

The tumor microenvironment includes both cellular and non-cellular components. Major cellular components surrounding the tumor cells in the TME are infiltrating immune cells, fibroblasts, and the pericytes and endothelial cells of blood vessels. A vast set of literature shows that the influence of non-tumor cells in the TME can be either pro-tumorigenic (immunosuppressive) or anti-tumorigenic (cytotoxic or immune-stimulatory) depending on the cell type and context. Non-cellular components include the lymph and interstitial fluid system that supplies nutrients and removes wastes, the supporting stromal and extracellular tissue and various “stresses” that develop in this microenvironment. The stroma can influence tumor development, maintenance, growth and progression (Farrow et al., 2008; Kharaishvili et al., 2014;

Pickup et al., 2014; Rucki and Zheng, 2014). The composition, modifications and density of the extracellular matrix also influences each “hallmark of cancer” (Pickup et al., 2014;

Hanahan and Weinberg, 2000 and 2011). An additional non-cellular component includes

the various chemical changes in the microenvironment, termed “stresses”, which distinguish the TME from normal tissue. We define the TME “stresses” as the physiological changes that develop inside a growing tumor, which are mostly unique from normal tissue, that create barriers to cell growth and thus represent selective pressures to cancer and other cells’ survival in the TME (Gatenby and Gillies, 2008).

These “stresses” are present throughout the TME and thus influence all of its cellular and non-cellular components. The presence, abundance and importance of all of these

TME components to the growth of the tumor fluctuate during tumor progression.

Together, components of the tumor microenvironment create an intricately connected and interdependent environment in which cancer cells form a tumor.

An improved understanding of the complex interactions and adaptive mechanisms of cancer cells in the TME will allow scientists and clinicians to better treat solid tumors. This Chapter provides an introduction to the topics relevant throughout the dissertation and its design. I will begin with a discussion of the tumor cells themselves, then describe the TME stresses and their implications to cancer biology and finish with a discussion of the functional genomics approach I have taken to better understand cancer cell adaptation to TME stresses.

1.1 Tumor Cells

The “War on Cancer” was initiated in 1971. Since then, thousands of researchers worldwide have sought to better understand and treat this disease. “Cancer” describes hundreds of diseases, partially classified by the cancer cells that populate each tumor.

Cancer cells are defined as hyper-proliferative cells that have lost their ability to properly respond to normal cellular signals to arrest proliferation. Spontaneous, genetic or environmental causes can initiate cancer, and the relative importance of these factors depends on the particular disease. Cancer cells are highly heterogeneous, varying temporally and spatially within one tumor and across different tumors of the same disease (Meacham and Morrison, 2013; Marusyk et al., 2012). Despite this heterogeneity,

Hanahan and Weinberg outlined “Hallmarks of Cancer” that are likely common to all cancer cells: limitless replication potential, self-sustaining growth signals, insensitivity to antigrowth signals, evasion of cell death pathways, sustained angiogenesis, reprogrammed cellular metabolism, immune evasion and tissue invasion and metastasis. Through genomic instability and tumor-promoting inflammation, these hallmarks of cancer drive normal cells to a state of malignancy (Hanahan and Weinberg,

2001 and 2011). Based on an overwhelming amount of scientific literature dedicated to understanding cancer biology, it is clear that these Hallmarks and the overall progression of the disease are influenced by the complex interactions between the cancer

cells and the environment in which they reside. A unique feature of this TME in which the tumor develops are TME “stresses”, which will be discussed in detail in the next

Section of this introduction.

1.2 Stresses of the TME

We define the TME “stresses” as the physiological changes that develop inside a growing tumor that are barriers to cell growth and so represent selective pressures to cancer cells’ survival in the TME (Gatenby and Gillies, 2008). While some cancer cells may undergo cell cycle arrest or cell death as a result of these TME stresses, the stresses also select for a sub-population of more aggressive tumor cells (Yun et al., 2009; Neri and Supuran, 2011; Wilson and Hay, 2011; Dhani et al., 2015). Since proliferating tumor cells expand beyond available nutrient supplies, as well as the physical space of the normal tissue, these stresses can be biochemical or biophysical in nature. As proliferating tumor cells expand, a lack of available nutrients and oxygen arises due to an increased distance from blood vessels (Gatenby and Gillies, 2008). Therefore, nutrient limitation is a major stress in the TME. As tumor cells gain further mutations, and in order to cope with the limited nutrients, they stimulate new blood vessels to form in a process called angiogenesis. However, these new blood vessels are often of poor structural soundness and organization (Carmeliet and Jain, 2000; Baluk et al., 2005).

Therefore, due to both a lack of vessels and their ineffectiveness, there is limited nutrient availability, oxygen perfusion and waste removal in the TME (Carmeliet and Jain, 2000).

While TME stresses influence all cell types and components of the TME, a vast majority of research focuses on the response of cancer cells to the presence of these stresses.

The rationale of my dissertation work is that in order to generate a tumor, cancer cells must be able to first adapt to, then survive in, and eventually proliferate under metabolically unfavorable conditions. As these stresses are mostly unique to tumors, targeting cancer cells specifically under stress offers a significant therapeutic window

(Wilson and Hay, 2011; Neri and Supuran, 2011). Therefore, if we can better understand how cancer cells adapt to stress, we could devise more effective and selective therapeutics in the long run. This rest of this Section will describe the causes of, cancer cells’ responses to and clinical and therapeutic implications of the major TME stresses.

1.2.1 Glucose limitation

Tumor cells’ metabolism is significantly altered compared to normal cells to meet the increased demand for nutrients to support their excess bioenergetic and biosynthetic needs (Vander Heiden et al., 2009). In response to these increased needs for growth and proliferation, cancer cells display increased uptake and metabolism of glucose to fuel their major metabolic pathways. At the same time as there is increased demand, the lack of available glucose due to tumors’ outgrowth of the blood supply further accelerates

glucose deprivation as a major TME stress. Glucose deprivation can lead to a number of complex physiological responses including altered signaling pathways (Spitz et al., 2000;

El Mjiyad et al., 2011), changes in the cell cycle (Jones et al., 2005), altered sensitivity to drugs (Menendez et al., 2012), and hypersensitivity to cell death (Spitz et al., 2000; Shim et al., 2008). Recently, cancer metabolism research has made a resurgence, in part due to the recognition that many oncogenes, tumor suppressors and signaling-activated transcription factors can regulate metabolic processes (Shaw and Cantley, 2012). In particular, HIF-1α-, RAS- and c-MYC-driven cancers all upregulate glucose metabolism

(Semenza, 2013; Dang, 2011; Li and Simon, 2013; White, 2013). Glucose deprivation contributes to several negative prognostic features, such as the selection for RAS mutations (Yun et al., 2009), expression of pro-angiogenic genes (Marjon et al., 2004) and induction of genes associated with metastasis or drug resistance (Banerjee et al., 2014;

Ledoux et al., 2003).

Because of cancer cells’ dependence on glucose and glucose deprivation’s ability to induce malignant characteristics, targeting glucose metabolism is considered an attractive therapeutic target. However, treatments that prevent glucose uptake are challenged by excessive toxicities with non-metabolizable glucose analogs (Vander

Heiden, 2011). One clinical utility of tumors’ high glucose uptake is that it allows for visualization of tumors through FDG-PET scanning for diagnosis and treatment

considerations (Kelloff et al., 2005). While many potential avenues for targeting glucose metabolism remain, additional research reveals that targeting other metabolic pathways may also provide therapeutic windows.

1.2.2 Amino acid limitation

In contrast to the role of glucose as a carbon source, amino acids can serve as important sources of both carbon and nitrogen (Keenan and Chi, 2014). Essential and non-essential amino acids can be limited in the TME due to poor delivery from the vascular system (Tang et al., 2014). Cancer cell dependencies on particular amino acids offer novel opportunities for metabolic targeting that are distinct from the more traditionally targeted glucose metabolism.

Glutamine is the most abundant amino acid in blood, but glutamine availability can become limited in the TME. Cancer cells rely on glutamine to supply a number of metabolic needs, including IDH-mediated reductive carboxylation to replenishing TCA intermediates through anapleurosis and to support lipid synthesis (Wise et al., 2011;

Metallo et al., 2012). Major oncogenes such as MYC and RAS upregulate glutamine metabolism, thus highlighting the importance of this amino acid to cancer cell metabolism (Dang, 2011; Li and Simon, 2013; White, 2013).

Identifying and characterizing specific deprivations and dependencies of other amino acids has allowed clinicians to successfully target tumor cell metabolism in the

clinic. Acute lymphoblastic leukemia requires asparagine for survival and so L- asparaginase treatment is an effective treatment for these patients (Muller and Boos,

1998). Inhibiting arginine metabolism has proven successful in the treatment of leukemia, melanomas and hepatocellular carcinomas (Vander Heiden, 2011; Scott et al.,

2000; Delage et al., 2010). Therapeutic potential also exists for targeting enzymes and specific isoforms of serine, proline and glycine metabolism (Li and Simon, 2013). To best apply therapeutics to target the metabolic needs of cancer cells, more research will need to identify these dependencies, the genetic alterations they are associated with, the flux of these critical metabolites, and the particular enzymes that will be most amenable to inhibition (Vander Heiden, 2011).

1.2.3 Biophysical or mechanical stresses

The physical or mechanical TME stresses result from abnormal forces that develop within the tumor. While all cells experience forces during normal development and physiology, different tissues are accustomed to different “normal” forces, and in tumors these forces are greatly altered. Mechanical forces in the TME can be divided into fluid forces or solid-state forces (Ariffin et al., 2014), which can be further broken down into compressive, tensile, shearing and hydrostatic forces (Butcher et al., 2009). Fluid forces are the result of hyper-permeable blood vessels and poor lymph system drainage causing excess extracellular fluid. Solid-state forces are caused mostly through the

increased number of cells in a limited space, as well as alterations to the ECM that often result in a more expansive and stiff TME (Jain et al., 2014). Due to these different types of forces, the interstitial pressure present in tumors can be ten-fold higher than in normal tissues (Butcher et al., 2009). Some groups have designed 3D cell culture systems to precisely monitor, compose and manipulate forces on cells in order to understand their phenotypic influences on the cells (Butcher et al., 2009; Griffith and Swartz, 2006).

However, in general, the appreciation of and research efforts into the forces and pressures of the TME are limited.

The best understood effects of mechanical stresses are their negative outcomes on the efficacy of chemotherapies and increased motility of cancer cells. Due to altered forces and vasculature in the TME, therapeutic agents must rely on passive and inefficient diffusion throughout the tumor, which can be slow and only take place over short distances (Ariffin et al., 2014). In addition, a stiffer ECM can enhance gene expression programs for phenotypes of invasion and metastasis (Pickup et al., 2014).

Based on understanding these adverse effects, various approaches have been proposed to reduce intratumoral pressure by improving fluid circulation or by decreasing the presence of cells or structural components to “free up space”. These treatments include anti-angiogenics, vasodilatory agents, vascular-normalization, vascular-disrupting agents or “stress-alleviation treatment”, proteolytic enzymes, and physical

manipulations (Ariffin et al., 2014; Jain et al., 2014). These approaches can be classified as either stress-alleviating or vascular normalization. Stress-alleviation treatments may be more effective in cancer types where there is less angiogenesis and more ECM deposition, such as pancreatic, colon or sarcomas. On the other hand, “vascular normalization” methods may be more effective for cancers with hyper-permeable vessels and lower solid stress forces, such as subsets of glioblastomas, melanomas and ovarian carcinomas (Jain et al., 2014). It remains challenging to understand the specific causes of fluid and solid stress in individual tumors and integrate this information into treatment regimes. Although difficult to study in vitro or in vivo, a better understanding of this TME stress will likely result from the integration of physics, biology and engineering (Butcher et al., 2009).

1.2.4 Lactic Acidosis

In 1956, Otto Warburg first described transformed cells’ preferential use of cytosolic glycolysis rather than oxidative phosphorylation in the mitochondria for glucose catabolism (Warburg, 1956). This shift in metabolism is termed the Warburg effect or aerobic glycolysis. The Warburg effect occurs with or without oxygen, although hypoxia can stimulate it (see Section 1.2.5). Since cytosolic glycolysis results in lactate instead of CO2 as an end product, the Warburg effect is a major cause of the increased extracellular lactic acidosis in the TME. Also, overexpression of proton pumps can lead

to an increased ability to drive protons out of the intracellular space (Cardone et al.,

2005). Without functional lymph drainage systems, the lactic acid “end product” gets pumped out of the cell, but remains in the extracellular space rather than being removed

(Peppicelli et al., 2014). Therefore, LA can happen both dependent and independent of hypoxia, consistent with data showing their differential locations in tumors (Gulledge and Dewhirst, 1996; Helmlinger et al., 1997; Schornack and Gillies, 2003) and the differential transcriptional profiles they elicit (Chen et al., 2008; Chen et al., 2010; Tang et al., 2012). This information is critical to correctly informing therapeutic strategies designed to target tumor cells under different stresses.

The responses of cancer cells to LA are less thoroughly investigated than some other TME stresses. More studies investigated the effects of either lactate (lactosis) or acid (acidosis) alone. Transcriptional profiling of both HMECs and breast cancer cells under LA, acidosis, glucose deprivation or hypoxia gives us an overview of cellular responses to these stresses (Chen et al., 2008; Chen et al, 2010; Tang et al., 2012). These studies showed very few genes similarly regulated by hypoxia and LA (Chen et al.,

2008), but a greater overlap between the glucose deprivation and LA responses. The majority of the LA transcriptional response was driven by acidosis (Chen et al., 2010), so studies considering the effects of acidosis are highly relevant to LA. Both glucose deprivation and LA induced a “starvation response” that activated elements of the

amino acid response pathway. LA also activated GPCR signaling, antigen processing and signaling and glucose catabolism (Chen et al., 2010). Acidosis induced GPCR signaling in vascular endothelial cells and rat renal tube cells as well (Nowik et al., 2008;

Dong et al., 2013; Curthoys et al., 2007). Multiple GO groups were also repressed transcriptionally in breast cancer cells under LA, including RNA metabolism, cell cycle genes and proliferation genes (Chen et al., 2008 and 2010).

Beyond transcriptional profiling, mechanistic studies give insight into cancer cells’ responses to LA. Multiple studies showed an up-regulation of the matrix metalloproteinase (MMP) proteins in response to LA (Rofstad et al., 2006; Martinez-

Zaguilan et al., 1996). Other effects of LA include triggering pro-angiogenic gene expression programs (Fukumura et al., 2001; Xu and Fidler, 2000; Shi et al., 2001), stabilizing HIF-1α (Mekhail et al., 2004) and inducing calcium signaling (Huang et al.,

2008) or autophagy (Wojtkowiak et al., 2012). Lactosis alone induced CD44 expression and altered the balance of NADH/NAD+, similar to hypoxia (Walenta and Mueller-

Klieser, 2004; Formby and Stern, 2003; Lu et al., 2002). Acidosis alone stimulated angiogenesis, cell migration and tissue remodeling, while also causing a shift in cellular metabolism mediated by p53 to preserve NADPH for redox homeostasis (Fukumura et al., 2001; Xu and Fidler, 2000; LaMonte et al., 2014). LA or acidosis can also select for cancer cells with stem-like characteristics (Hjelmeland et al., 2011; Fang et al., 2008).

Overall, the changes seen under LA reflect altered cellular metabolism and gene expression to maintain cellular energetics and redox homeostasis under this stress.

Generally, tumor LA reduces therapy efficacy and is associated with poor patient outcome. Several studies have shown increased invasion and metastasis correlating with or being caused by LA or acid (Martinez-Zaguilan et al, 1996; Rofstad et al., 2006; Robey et al., 2009; Silva et al., 2009;). Acidosis has also been associated with drug resistance due to increased activity of the multi-drug resistance protein 1 (MDR1) (Sauvant et al., 2008),

GRP78 (Visioli et al., 2014), or the P-glycoprotein (Thews et al., 2006) and altered protonation status of chemotherapies within the TME (Adams and Morgan, 2011).

Acidosis, similar to hypoxia, reduces the effects of radiation therapy by inhibiting apoptosis or prolonging the time for DNA repairs to be made in G2 (Rottinger and

Mendonca, 1982; Haveman, 1980; Lee et al., 1997; Park et al., 2003). All of these data support findings that lactosis, acidosis and lactic acidosis are negative prognostic features in clinical data (Walenta et al., 2000; Hirschhaeuser et al., 2011; Webb et al.,

2011; Calorini et al, 2012; Peppicelli et al., 2014). In conflict to the predictions of many studies, transcriptional gene expression signatures of HMEC and MCF7 breast cancer cells under LA or acidosis in culture correlated positively with breast cancer patients who had better outcomes (Chen et al., 2008 and 2010). While these transcriptional profiles suggest a multifaceted response to LA with different potential outcomes

depending on context, the prevalence of data that show the negative consequences of lactic acidosis are convincing enough to make attempts to target this feature of tumors in vivo.

Multiple therapeutic strategies alter the extracellular pH of the TME to reverse the negative effects of this TME stress (Neri and Supuran, 2011). Both sodium bicarbonate and 2-imidazole-1-yl-3-ethoxycarbonylpropionic acid (IEPA) have been used as systemic buffers to mitigate acidosis and decrease tumor metastases in murine models of breast and prostate cancers (Robey et al., 2009; Silva et al., 2009; Ibrahim-

Hashim et al., 2011a, 2011b and 2012). Although targeting H+ATPases has basal toxicities, recently developed proton pump inhibitors display low toxicity in in vitro and in vivo xenograft models and effectively prevented extracellular acidification and enhanced chemotherapeutic delivery (Wong et al., 2002; Luciani et al., 2004; Perez-

Sayans et al., 2012). Further investigation into the mechanisms of cancer cell adaptation to lactic acidosis will identify new drug targets that may have the potential to improve our ability to specifically target cancer cells under this stress.

1.2.5 Oxygen limitation (hypoxia)

As cancer cells proliferate and expand away from blood vessels, their distance from local blood vessels will exceed the oxygen diffusion limit of ~150um (Dang and

Semenza, 1999). Once this occurs, parts of the tumor experience low oxygen levels, or

hypoxia. In healthy tissue, there is typically 2-9% oxygen available to cells. Hypoxia is often defined as 2%-0.02% O2 and anoxia is below 0.02% O2 (Bertout et al., 2008). Cancer cells emit angiogenic signals in an attempt to trigger angiogenesis and overcome hypoxic stress. However, as mentioned, structurally unsound blood vessels and increased interstitial pressure prevent efficient delivery of nutrients, such as oxygen, from blood vessels to cells (Carmeliet and Jain, 2000). Thus, hypoxia is a dynamic and common stress in the solid tumor microenvironment.

All metazoan cells have mechanisms to sense oxygen levels in their environment and these oxygen-sensing and adaptive mechanisms are maintained in cancer cells. The major family of enzymes responsible for sensing oxygen levels is the 2-oxoglutarate- dependent dioxygenases (2-OGDDs) (Kaelin, 2011). These enzymes use molecular oxygen and 2-oxoglutarate (α-ketoglutarate) as substrates to modify their protein targets

(Markolovic et al., 2015). The primary oxygen sensing 2-OGDDs are the prolyl hydroxylases or PHD proteins, which control cellular responses to hypoxia by regulating protein levels of the hypoxia-inducible transcription factors (HIFs). Under normal oxygen conditions, the PHD proteins use their substrates to hydroxylate the

HIFs. These hydroxylations serve as a signal for the poly-ubiquitination and subsequent proteasome degradation of these proteins by the von Hippel-Lindau (VHL) ubiquitin ligase complex (Majmundar et al., 2010; Kaelin, 2008). Thus, under normoxia, the HIF-

1/2α proteins are maintained at low levels due to continuous proline hydroxylation and

VHL-mediated proteasomal degradation. As hypoxia develops and reduces the availability of oxygen, the PHD proteins become less active and so the HIFs become stabilized and enact their complex adaptive transcriptional response (see below)

(Majmundar et al., 2010). These 3 family members, PHD1-3, vary in their expression and preference for each HIF (Keith et al., 2012). The importance of these oxygen-sensing mechanisms to cancer biology is seen in the almost complete loss of the VHL complex in clear cell renal cell carcinomas (Kaelin, 2007) and the overexpression of HIF proteins in cancers (Bertout et al., 2008).

After sensing low oxygen, a cell enacts a number of survival and coping mechanisms. Arguably, the most important of these is the HIF-mediated transcriptional response. HIFs are bHLH transcription factors comprised of two subunits, an oxygen- responsive alpha subunit and a constitutive beta subunit (Majmundar et al., 2010). HIF-

1α and HIF-2α are the more prominent family members and they have shared and unique target genes (Keith et al., 2012). Both HIFs induce glycolytic metabolism (PDK1), stress response genes (ATF4), angiogenic pathways (VEGF, COX2, PDGF), invasion and motility genes (MET, LOX) and pH regulation (CA9) (Bertout et al., 2008). They are also implicated in cancer stem-ness (Keith and Simon, 2007) and inflammation (Palazon et al., 2014; Imtiyaz and Simon, 2010). HIF-1α is induced more acutely or under more

severe oxygen deprivation; HIF-2α is stabilized in response to more moderate or longer- term periods of hypoxia (Keith et al., 2012). While overexpression of HIF-1α is typically correlated with worse outcome, HIF-2α has been shown to be both a tumor suppressor and an oncogenic driver (Keith et al., 2012). Multiple co-factors can modulate their responses, such as p300, CBP, MYC, SP-1 and p53 (Bertout et al., 2008; Keith et al., 2012).

Interestingly, HIF-1α and HIF-2α differentially interact with some of these partners for opposing effects on transcriptional output (To et al., 2006; Koshiji et al., 2005; Keith et al.,

2012). Complex and multi-layered control mechanisms fine-tune the HIF-mediated transcriptional programs.

Hypoxia can also activate other transcription factors. NF-kB has a pro- inflammatory response under hypoxia (Koong, et al., 1994; Taylor and Cummins, 2009).

Hypoxia can induce transcription factors of the unfolded protein response, such as

PERK and ATF4, both through HIF-dependent and -independent ways (Bertout et al.,

2008; Ye and Koumenis, 2009). MYC transcriptionally influences the hypoxic metabolism of cancer cells, promoting glycolysis and glutaminolysis (Gordan et al., 2007). Other hypoxia-inducible transcription factors likely remain to be discovered. Overall, the transcriptional changes in response to low oxygen levels are complex, overlapping and context-dependent.

Besides through transcription factors like MYC and HIF, the metabolism of hypoxic cancer cells is also influenced by signaling pathways. The mTOR nutrient- sensing complexes are inhibited with decreased ATP/ADP levels under hypoxia, causing an inhibition of translation, induction of HIF-1α and induction of autophagy

(Shackelford and Shaw, 2009). AMPK is also activated in response to decreased ATP levels under hypoxia. AMPK has many targets and functional phenotypes, including inhibiting de novo lipogenesis by phosphorylating and inactivating the rate-limiting enzyme of lipogenesis, ACACA (ACC1) (Ruderman and Prentki, 2004). Hypoxia also promotes the process of reductive carboxylation of glutamine to glutamate to fuel lipid biosynthesis and decrease forward flux through the TCA (Metallo et al., 2012). Overall, metabolic shifts under hypoxia reduce the activity of ATP-consuming processes and reactions that consume molecular oxygen.

As with other TME stresses, the consequences of hypoxia on cancer cell phenotypes are negative for patient outcome. Hypoxia selects for radiation-resistant and chemotherapy-resistant cells, which are also more likely to metastasize (Wilson and

Hay, 2011; Bertout et al., 2008). Similarly, HIF-1α overexpression predicts multiple negative outcomes in clinical datasets for metastasis and survival in a wide range of cancers (Bertout et al., 2008). Therapeutically targeting hypoxic cells holds much potential value for physicians and patients.

A key aspect of treating hypoxic tumors is the ability to detect the presence, abundance and location of hypoxia within a tumor. Measuring oxygen tension in tumors is challenging, but methods exist to measure it in vivo or in biopsies. Studies with invasive probes revealed that hypoxia is common in tumors, especially in higher-grade tumors, and that the levels and importance of hypoxia vary with tumor type and patient

(Dhani et al., 2015). With the phase-out of probes, indirect approaches have been developed to measure oxygen levels in tumors, each with varying advantages and disadvantages. Bioreductive drugs can reveal specific patterns of hypoxia in tumor samples and have reached clinical trial (Guise et al., 2014), but they require taking biopsies, are invasive and their metabolism can vary with enzyme expression and flow rates of individual tumors (Dhani et al., 2015). Protein or RNA level biomarkers of hypoxia-induced changes can also give detailed mapping of hypoxic regions, but again rely on biopsies (Dhani et al., 2015). Lastly, PET or fMRI techniques are non-invasive imaging techniques that permit visualizing hypoxia over time in vivo, but often struggle against high background values, poor resolution and false positive readings. So far, these challenges have prevented them from widespread use in the clinic (Mees et al.,

2014). Additionally, the degree or cause of the tumor hypoxia will likely influence the success of its measurement and treatment (Koch and Evans, 2015). The spatial and temporal variability of hypoxia may challenge the efficacy of hypoxia-targeted

therapeutics in vivo, but an abundance of data suggests benefits for targeting cells under this stress.

Yet, current therapies targeting cells under hypoxia have significant limitations.

For example, angiogenesis is a well-established, valuable therapeutic target with many agents that have been developed to block it at various stages of tumor development.

However, many anti-angiogenic therapies fail over time, through acquired or inherited resistance that may involve the presence of tumor hypoxia (Abdollahi and Folkman,

2010). Therefore, a number of other strategies are being developed to directly target hypoxic cells, such as agents that block lactate transporters (Neri and Supuran, 2011;

Sonveaux et al., 2008), or pro-drugs that are activated only in the presence of low oxygen

(Wilson and Hay, 2011). These combination therapies targeting hypoxia, cancer cell metabolism and/or anti-angiogenesis mechanisms have promising pre-clinical results; the application and efficacy in the clinic remains to be tested (McIntyre and Harris,

2015).

Due to the negative clinical features associated with TME stresses, it will be important to develop therapies targeting cells under stress. The adaptive mechanisms to stresses create a significant therapeutic window in order to generate more selective therapies. Therefore, to better guide the development of future therapeutics, we need to

better understand the genes that mediate cellular survival under TME stresses. Many studies have investigated adaptive mechanisms in a hypothesis-driven fashion, learning more about known players in response to stresses. While these studies can inform therapeutic development through potentially identifying new ways to target known involved proteins, they are limited in what new biology they can discover as they build off of known pathways and proteins. Completely novel adaptive pathways, and therefore potential drug targets, are much more slowly discovered in this type of research. A way to combat this limitation is by performing an unbiased experiment to interrogate the effects of gene depletion genome-wide. Unbiased experimentation permits faster steps in new directions and can identify completely undocumented adaptive mechanisms since it does not rely on previous knowledge of stress pathways.

In order to employ such an unbiased approach, we applied functional RNAi screens to understand cancer cell survival under stress. Using this strategy, not only can we evaluate the contribution of nearly all genes in the human genome to stress survival, we can also isolate and, through additional detailed experimentation understand, novel adaptive mechanisms that could offer new targets for therapeutics.

1.3 Functional genomics through RNA interference screens

1.3.1 Introduction

The 2006 Nobel Prize in Chemistry was awarded to Andrew Fire and Craig

Mello for the discovery of the RNA interference (RNAi) pathway. Since its discovery, the RNAi pathway has become one of the most widely used tools for molecular and cellular biologists to interrogate the function of genes through loss-of-function, forward- genetic studies (Westbrook et al., 2005; Boutros and Ahringer, 2008; Mohr et al., 2010).

CRISPR-Cas9 technologies are rapidly becoming as widespread as RNAi tools and offer different advantages and challenges (Taylor and Woodcock, 2015; Wade, 2015). One key application of RNAi technology is the development of screens to interrogate, in an unbiased and a priori fashion, the function of either one gene across many conditions or, more frequently, many genes across one or more conditions (Westbrook et al., 2005;

Mohr et al., 2010). RNAi screens have led to discoveries across multiple model systems and biological processes. This section will describe designs for mammalian RNAi screening experiments, their advantages and disadvantages and what they have taught us about cancer cells under TME stresses.

1.3.2 Designing an RNAi screen experiment

As with any experiment, a careful and thoughtful consideration of experimental design is important to the successful execution of an RNAi screen. Controls should be

considered before each common “step” of an RNAi screen experiment: primary screen, secondary screen, quality control analysis, data normalization, hit identification analysis, hit validation, and detailed mechanistic investigation (Boutros and Ahringer, 2008; Mohr et al., 2010). The primary RNAi screen is designed to address a specific biological question and is often the largest and “dirtiest” step. Prior to beginning the primary screen, experiments should be conducted to determine the ideal treatment conditions with positive controls (Boutros and Ahringer, 2008). After the primary screen identifies an initial list of potential candidate genes involved in the process of interest, a secondary screen can be conducted on these selected candidates. A secondary screen applies more stringent filters or repetitions to the primary screen candidates (Boutros and Ahringer,

2008). It can use additional and independent RNAi reagents per gene, evaluate multiple

(lower-throughput) relevant phenotypes, or be conducted in a different genetic or sensitized background. The candidate hits that pass cutoff criteria in the secondary screen are higher confidence “hits”. After each primary and secondary screen, quality control evaluation of the data considers the possibility of technical errors such as researcher mistake, well- or plate-based biases, or differential response across replicates

(Boutros and Ahringer, 2008; Mohr et al., 2010 and 2014). Data normalization corrects for differences across pools of a library or different plates and allows the entire dataset to be compared. The process of hit identification analysis is still under active

development within the field, although certain criteria are common depending on the different format of the screen. Once any number of criteria is used to identify or prioritize the top candidates, these hits must be further validated for their phenotype in the biological question of interest (Mohr et al., 2010 and 2014; Sigoillot and King, 2011).

Then, if desired, a mechanistic study can be initiated to give a detailed explanation of how a gene or pathway contributes to the phenotype. This dissertation is an example of multiple types of RNAi screens carried out from primary screening through a mechanistic explanation of a top candidate gene.

The two most common formats for RNAi screens in mammalian cells are an arrayed format or a pooled format (Boutros and Ahringer, 2008; Mohr et al., 2010). In the arrayed format, RNAi reagents are separated in to different wells of a 96- or 384-well plate, to which the cells and treatment are added and the phenotype assayed. This type of screen often requires specialized automated equipment for liquid handling that allows the processing of many plates in order to reach statistical significance without human bias (Mohr et al., 2010). As a consequence, these screens are often more expensive. Arrayed screens can be applied to any phenotype measurable in a plate assay, including high content microscopy imaging (Krausz, 2007). As the effect of each gene in these screens is separated by wells, no deconvolution is necessary and the effect of individual wells is easily distinguishable without the potential complication of non-

representation or spurious enrichment or depletion in a pooled library (Mohr et al.,

2010).

The development of pooled RNAi libraries allowed single laboratories with no specialized equipment to perform RNAi screen experiments (Mohr et al., 2010; Sims et al., 2011). In the pooled format, the RNAi library is introduced en mass to a population of cells that remain pooled for the duration of the experiment; the researcher does not know during the experiment which cells have which gene targeted. Most commonly, these experiments are performed with shRNA libraries that generate a “stable” depletion of protein expression after integration into the cells’ genome through retro- or lentiviral transduction. Next, drug treatment selects for successful transduction before the experimental treatments are performed. After the treatment, cells are collected and

PCR amplifies the genomically integrated shRNA sequence, or a barcode that corresponds to individual shRNAs, so that the abundance of every shRNA in the experimental conditions can be measured (Sims et al., 2011). The abundance of shRNAs can then be resolved by next-generation sequencing or custom microarrays (Boutros and

Ahringer, 2008; Sims et al., 2011). While this type of experiment relies on the ability of the researcher to de-convolute the effect of each shRNA after the experiment through sequencing or arrays, it allows for longer time periods of treatment or in vivo xenograft growth to be performed (Mohr et al., 2010 and 2014). Due to the lack of specialized

equipment needed for pooled screens, the power of these screens can be applied to many biological questions in individual labs.

1.3.3 Advantages and Disadvantages of RNAi screens

RNAi screens, like any other type of large-scale profiling experiment, have both advantages and disadvantages. Although it is possible for individual labs to perform these experiments in house, they are costly in terms of reagents, time and labor.

Generating and infecting a pooled shRNA library involves considerable tissue culture costs for maintaining enough cells for sufficient shRNA representation in every treatment. Transfecting a library of transient siRNAs requires substantial transfection reagent and optimization of reagent delivery to achieve effective depletion of gene expression per well. If the arrayed format is used, then additional cost is associated with deployment and maintenance of specialized equipment. For pooled screens, there are money and time costs associated with de-convolution through next-generation sequencing.

The biggest challenges and disadvantages of performing these high throughput experiments are the possibilities of false-positive and false-negative hits due either to incomplete silencing and/or off-target effects due to the nature of RNAi (Westbrook et al., 2005; Mohr et al., 2010 and 2014). Off-target effects have been blamed for the lack of overlap between experiments conducted by different labs (Ma et al., 2006; Sigoillot et al.,

2012; Bhinder and Djaballah, 2013) or for the low validation rates often seen in RNAi screens (Sigoillot et al., 2012; Schultz et al., 2011; Sigoillot and King, 2011). Scientists have developed improved techniques for RNAi tool design to decrease the off-target effects and improve the efficiency of on-target knockdown (Fennell et al., 2014). This began with the development of short, chemically synthesized siRNAs to prevent the activation of the interferon response with the activation of the RNAi pathway (Elbashir et al.,

2001). The introduction of the miR-30 context promoted endogenous processing and thus improved efficiency of knockdown with shRNAs (Zeng et al., 2002; Cullen, 2004).

Additionally, the discovery that a Pol II-driven CMV promoter increased efficiency of transcription and downstream effects has improved RNAi application (Cai et al., 2004;

Steigmeier et al., 2005). These discoveries have been applied to a myriad of shRNA libraries for human and mouse genomes (Silva et al., 2005), inducible systems

(Steigmeier et al., 2005; GE Dharmacon), and use of fluorescent reporters (GE

Dharmacon); many libraries are commercially available.

In addition to RNAi design improvements, many analysis methods try “weed out” the off-target “hits” from RNAi screens. A simple and effective way to combat false-positive hits due to off-target effects is to use multiple RNAi reagents per gene and only focus on the genes that have multiple of these reagents passing established analysis criteria with the same phenotypic effect. In this work, this concept is referred to as

“multiple hairpin analysis”. Other methods include using RNAi tools of different construction (different vectors, siRNAs and shRNAs); rescuing the effect of loss of gene function with an RNAi-resistance cDNA (Mohr et al., 2010 and 2014); or complementing the sometimes hypomorphic response of RNAi with CRISPR reagents (Taylor and

Woodcock 2015; Mohr et al., 2014) (although explanations exist for incongruences between these two methods (Mohr et al., 2014)). When analyzing RNAi screens, the researcher must balance the false discovery rates due to inherent off-target effects of

RNAi and biases of different libraries with the stringency of the analysis criteria used to establish a “candidate hit”. In the end, Mohr and colleagues (2010) phrased it best:

“There is no substitute for verification of results”.

Considering these disadvantages and challenges that apply to performing, analyzing and validating an RNAi high-throughput screen, why go to all of the trouble?

In the end, these screens are a very powerful way to employ an unbiased experiment to discover something completely new and unexpected (Westbrook et al., 2005; Mohr et al.,

2010). The number and scope of important discoveries that have resulted from RNAi screens is vast and beyond proper citation, but arguably has been the most significant in signal transduction, host-pathogen interactions and cancer biology (Mohr et al., 2014).

The next section will detail some examples of RNAi screens in cancer biology, especially in the context of TME stresses.

1.3.4 Previous applications to cancer and the TME stresses

The field of cancer biology quickly adopted the use of RNAi screens to investigate the functional importance of genes to tumor biology (Westbrook et al., 2005).

This has scaled up immensely over a very short time and now Project Achilles™ is under way at the Dana-Farber Cancer Institute and the Broad Institute. The goal of this project is to comprehensively apply functional RNAi tools to identify essential genes for a large panel of cancer cells. This immense endeavor promises to improve our understanding of cancer vulnerabilities. In parallel, RNAi screens continue to be used to study more specific cancer biology questions, such as applying the concept of synthetic lethality (Westbrook et al., 2005; Nijman, 2010), which is very relevant under TME stresses. A synthetic lethal interaction is one in which the depletion of either of two genes individually does not affect cell survival (or whichever phenotype is being assayed), but the combinatorial loss of these two genes is lethal to the cell (Westbrook et al., 2005; Thompson et al., 2015). These “contextual phenotypes” can also refer to

“synthetic sick” combinations when the cells may not die, but have a reduced fitness advantage, or to “synthetic survival” combinations when the loss of two genes promotes survival under otherwise toxic conditions (Nijman, 2010). The concept of synthetic lethality originated from yeast and Drosophila studies (Nijman, 2010), but now applies to many fields of research, including cancer biology (Westbrook et al., 2005; Canaani,

2014; Thompson et al., 2015). While characteristics such as similar localizations, patterns of expression or physical interactions between proteins can suggest a synthetic lethal interaction, biological validation is the best way to identify these interactions and RNAi screens can be used to broadly search for them in a high throughput manner (Nijman,

2010; Westbrook et al., 2005).

Synthetic lethality has been applied to cancer biology by performing RNAi screens in isogenic cell lines harboring particular oncogene activations (Luo et al., 2009;

Canaani, 2014). Similarly, the concept has been applied by combining drug screening with particular driving genetic mutations (Mizuarai et al., 2008; Mohr et al., 2014), which can be especially useful when known oncogenes are difficult to target with small molecules (Nijam, 2010). For example, a famous synthetic lethal interaction in cancer biology is the combination of BRCA1/2 mutations and PARP inhibitors (Farmer et al.,

2005; Bryant et al., 2005). Treating BRCA1/2 mutated cancers, commonly found in breast and ovarian cancer patients, with PARP inhibitors prevented these cells from repairing

DNA lesions and induced apoptosis. In Luo et al. (2009), isogenic colon cancer cells with an activating KRAS mutation were screened in parallel for sensitizing secondary mutations. This work identified a pathway of regulation through the kinase PLK1, the anaphase promoting complex and proteasome inhibition that specifically killed RAS mutated cells. This screen exemplifies the challenge of validating RNAi screens,

reaching only a 26% validation rate in the secondary screen. Yet, it also demonstrates the power of RNAi screens as the identified RAS synthetic lethal genes have known small molecule inhibitors (Luo et al. 2009; Yu et al., 2015). In cells with oncogenic RAS mutations, these synthetic lethal genes offer novel targeting strategies for an oncogene that has been notoriously difficult to treat clinically. Another RNAi screen discovered that the MYC oncogene upregulates the Death Receptor 5 (DR5) and induces auto- cleavage and activation of caspase 8 to sensitize cancer cells to extrinsic apoptosis (Wang et al., 2004). Humanized antibodies that act as DR5 agonists have been tested in the clinic, although not specifically against MYC driven cancers, but showed promise in metastatic pancreatic cancer (Kindler et al., 2012). Similar work could be applied to investigate tumor suppressors without targeting potential. Thus, RNAi screens can identify new and unexpected drug targets.

Although many screens have been used to identify synthetic lethal drug combinations or drug and genetic alteration combinations, very few studies have examined contextual interactions the “natural occurring” TME stresses. In one elegant study, David Sabitini’s group developed a “Nutrostat” chemostat system to culture cancer cells under low glucose conditions for an extended period of time. They performed an RNAi screen of metabolic genes and transporters on cells grown in the

Nutrostat (Birsoy et al., 2014). This screen discovered that mitochondrial oxidative

phosphorylation was critical for Jurkat cells’ ability to survive low glucose conditions and that biguanides could be used to target cancer cells under low glucose conditions

(Birsoy et al., 2014). Additionally, two RNAi screens have been performed under low oxygen conditions. The first study employed a genome-wide library, but analyzed only

73 genes that showed expression in their cell line of interest (Yoshino et al., 2012). Of their 5 genes found to be overrepresented under hypoxia (predicting a “synthetic survival” phenotype), two genes were validated: GPR68 and RNF126. In addition, they validated 21.5% of their hypoxia-underrepresented genes (“synthetic sick/lethal”), identifying 11 genes that affected cell growth under hypoxia (Yoshino et al., 2012). The second study focused only on the kinome under extreme acute hypoxic conditions (0.1-

0.3% O2). This screen identified two kinases, GAK and IRAK4, which, when depleted, prevented sphere-formation under these conditions (Pan et al., 2013). These studies set excellent precedence for applying functional genomics to study cancer cells under stress.

Application of identified contextual lethalities to cancer drug development may continue to identify novel therapeutic strategies to treat cancer patients whose tumors do not have readily targetable mutations or targets.

1.4 Overview of Chapters

One way to understand such a complex environment as the TME is to break it into its parts and first understand each part in more detail. The work I describe herein employed this approach of understanding individual components to learn about the system as a whole. To this end, I focused my research on understanding the survival adaptations of cancer cells to the TME stresses of hypoxia and lactic acidosis. This

Chapter, Chapter 1, introduced the relevant topics and context for the rest of the thesis.

Chapters 2 and 3 describe each of the two functional RNAi screens that I performed in conjunction with a number of collaborators. Chapter 2 details the first screen, which was a positive selection screen that exposed cancer cells to an extended lactic acidosis treatment. Chapter 3 describes the second screen method-- a genome-wide investigation of both positive and negative regulators of cell survival under shorter term hypoxia and lactic acidosis treatments. Potential hypotheses to investigate the roles of validated genes in stress survival are proposed and discussed. Chapter 4 describes a detailed explanation for how a top candidate from the genome-wide screen, ACC1, affected cellular survival under hypoxia. Another candidate hit from the genome-wide screen,

ETV4, was also involved in ACC1’s regulation of hypoxia-induced apoptosis and so is also included in Chapter 4. Chapter 5 has two major parts. The first is a discussion of the current results (Sections 5.1-5.5). Second, in Section 5.6 I describe a future project that

stems from the work described in Chapter 4 on ACC1 and ETV4. Overall, my dissertation research has contributed to the scientific community both an overview of genes that potentially affect cell survival under hypoxia or lactic acidosis, as well as a detailed explanation of how decreased lipogenesis alters tumor cell metabolism and transcription for survival under hypoxia.

2. Positive Selection Screen under Lactic Acidosis

2.1 Introduction

In order to discover genes essential for continued survival under microenvironment stresses, I performed a positive selection RNAi screen on the cell line

H1975. H1975 is a non-small cell lung cancer cell line, and so, represents the largest sub- category of lung cancers, which is the most deadly form of cancer in the United States

(Siegal et al., 2014). This cell line is unique as it was derived from a non-smoker and has the T790M EGFR mutation. The T790M mutation is a commonly acquired mutation after

EFGR inhibitor treatment in the clinic, which renders cells resistant to further EGFR inhibitor treatment (Matikas et al., 2015). Although third-generation inhibitors show cautiously optimistic results in the clinic, their long term efficacy remains unknown

(Matikas et al., 2015). Positive selection RNAi screens are designed to create an experimental situation in which the majority of shRNAs or cells are not recovered due to extensive death or arrest, but the biologically interesting ones survive and can be analyzed (Paddison et al., 2004). These screens have identified new components of the p53 pathway (Berns et al., 2004), genes important for chemotherapy and apoptosis resistance (MacKeigan et al., 2005; Berns et al., 2007) and mechanisms of avoiding oncogene-induced senescence (Wajapeyee et al., 2008 and 2013). With positive screens there is no need to retain a control population to compare against, since all of the control

population should die or arrest during treatment. Consequently, for my application, stronger stress conditions can be applied, such as extended stress exposure or exposure to multiple stresses simultaneously. Few cells are anticipated to survive these extended and harsh treatments, so the background or noise should be low and the genes identified of higher confidence to confer robust stress survival phenotypes. In this section, I will describe the positive selection RNAi screen we performed under an extended lactic acidosis treatment, the results from this screen and validation attempts. For the genes that were validated, I outline potential hypotheses that were or could be investigated in future experiments.

2.2 Methods

2.2.1 Positive Selection Screen

A positive selection screen was conducted on a subset of cells transduced with a genome-wide shRNA library under a lactic acidosis (LA) stress. H1975 cells were spin- infected with the pMSCV-based retroviral genome-wide library (Schlabach et al., 2008) at an MOI of 0.3 and divided into six sub-pools to achieve a final library representation of 1000 cells per shRNA (Schlabach et al., 2008). The cells from Pool 4 of the library were plated and cultured under LA conditions (25mM LA, pH 6.7) for 3 weeks with continuous passaging to maintain sub-confluency and remove dead cells. In order to sustain a constant level of LA and pH, the media was changed every second day. In

parallel, the same library pool was maintained in a control condition with pH 7.4 media.

Both treatments included 25mM Hepes buffer in the treatment media in order to decrease fluctuations in pH during the course of the experiment. At the end of 3 weeks,

>90% of the cells in the LA condition had died. At this time, the cells were trypsinized to generate single cell suspensions, re-plated and allowed to recover in full media on 15cm dishes. Once single-cell colonies grew up to be distinguishable under a light microscope, sterile filter papers soaked in trypsin were used to collect individual colonies from the

15cm dish and transfer them to separate wells of a 12-well plate. These colonies were allowed to expand under regular culture conditions until the well was confluent, at which time they were split in half to 2 wells of a 6 well dish. Once the cells filled two wells of a 6-well plate, one well was collected and frozen in 10% DMSO for a cell stock and the other was pelleted and frozen at -20°C to determine the surviving shRNA.

Genomic DNA with the incorporated shRNA was isolated (QIAGEN’s DNeasy Kit) and

PCR amplified. The PCR products were run on a 1% TAE gel, the shRNA fragment was extracted and purified (QIAGEN Gel Extraction Kit). In order to generate pure PCR products in case the colonies grew from more than one different cell (with different shRNAs incorporated), PCR products were TOPO-TA cloned into the pGEM-TEasy vector according to the manufacturer’s instructions. At least 12 individual bacterial colonies per PCR product were prepared and sent for Sanger sequencing to determine the shRNA(s) present in that surviving cell population.

2.2.2 Cell culture, TME stress treatments and generation of stable shRNA cell lines

H1975 cells were cultured in RPMI media (GIBCO cat. no 11875) supplemented with 10% Fetal bovine serum (heat-inactivated), 1% glucose, 10mM HEPES, 1mM sodium pyruvate, and 1x antibiotics (penicillin, 10,000 UI/ml; streptomycin, 10,000

UI/ml), as directed by the Duke Cell Culture Facility. Cell lines, obtained from and initially validated by the Duke Cell Culture Facility (Durham, NC, USA), were maintained for fewer than 6 months and validated by microscopy every 1 to 2 days.

Lactic acidosis was generated via addition of lactic acid (Sigma-Aldrich, St.

Louis, MO, USA, cat. no L6402) and media pH adjustment to pH 6.7 by HCl immediately before use. For all stress experiments, cells were serum starved (0.5% FBS) for 24 hours before treated with stress, also under 0.5% FBS.

Stable cell lines were created with the pLKO.1 or pGIPZ shRNA constructs purchased from the Duke RNAi Core Facility or Sigma. Virus was generated by transfecting HEK-293T cells with a 1:1:1 ratio of pMDG2: pVSVG: pGIPZ or 1:0.1:1 ratio of pMDG2: pVSVG: pLKO.1 with Lipofectamine 2000 in the evening. Media was changed the following morning and virus collected 48 hours after transfection. Stable cell lines were generated by adding 500ul (pGIPZ vector) or 200ul (pLKO.1 vector) virus to a 60mm dish of parental cells with polybrene (final concentration 8ug/ml). Complete death in blank infection dishes was used to determine success of infection and puromycin selection. Efficiency of silencing or overexpression was determined by qPCR

or western blots. Concentration of puromycin needed for selection on H1975 cells=

1ug/ml.

2.2.3 Crystal violet staining

Cells were fixed either in 4% paraformaldehyde (PFA) overnight at 4°C or at room temperature for 30 min. PFA was removed and crystal violet staining solution

(0.2% crystal violet, 25% methanol, 75% water) gently shaken on cells for 30+ minutes at room temperature. Staining solution was removed and plates rinsed with tap water 2-3 times. For quantitation, completely dried stain was dissolved by adding 10% acetic acid and shaking gently at room temperature for 30+ min before reading absorbance at 570 nm.

2.2.4 Determination of cell number

Cell number was evaluated by direct cell counting with trypan blue exclusion of dead cells. After treatment, media was removed, cells were not rinsed for fear of losing loosely-attached cells, trypsinized, diluted 1:1 with trypan blue and immediate counted on a hemocytometer. All four quadrants of the hemocytometer were counted and an average of those 4 values was calculated as n=1.

2.2.5 Flow cytometry

For cell cycle analysis, after 4 days of stress treatment, media was collected, cells were trypsinized and pooled with the media. Cells were centrifuged (5 min, 1000rpm,

4°C) then fixed by resuspension in ice cold 70% ethanol while gently vortexing. Fixed

cells were placed at -20°C until prepared for FACS analysis. Immediately before FACS analysis, cells were centrifuged for 5 min at room temperature, washed twice in PBS

(spins of 5 min, 1000rpm, RT) then resuspended in freshly made 25ug/ml Propidium iodide (Sigma cat. no P4864) and 10ug/ml RNAse A in PBS. Cells were stained for 30+ min in the dark then 8000 events were captured on a Canto II Flow cytometer.

2.2.6 Protein lysate collection and Western blots

Cell lysis: Cells were washed once with ice cold PBS, lysed by RIPA buffer with protease and/or phosphatase inhibitors added fresh, scraped into a microcentrifuge tube, allowed to swell on ice for 15-20 min, vortexed briefly, then spun down at top speed for 15 min at 4°C. Supernatant was transferred to pre-cooled new tube and protein concentration assayed with the Pierce BCA kit (ThermoScientific, cat. no. 23225).

Western blots: Between 15-30ug of lysate was loaded on SDS-PAGE gels, wet- transferred to PDVF membrane, blocked with 5% milk in 1xTBST (0.1% Tween-20), then primary antibodies were incubated overnight at 4°C. Details on antibody usage are as standard in the Chi lab (refer to Google sheet of “Chi lab Antibodies”).

2.2.7 Quantitative real-time PCR

RNA was extracted using the RNeasy Kit (QIAGEN). 1 µg of total RNA was reverse transcribed by SuperScript II (Invitrogen) for real-time PCR with Power

SYBRGreen Mix (Applied Biosystems/Life Technologies (Grand Island, NY, USA)).

Primers were designed across exons whenever possible and were verified for specificity

by regular PCR prior to use in real-time PCR. Please refer to my “Designing real time primers” spread sheet for the sequences of primers used.

2.3 Initial Results

Of the 23 total colonies isolated during the positive selection screen under LA,

Table 1 shows the 13 colonies that had at least 40% of Sanger sequences match 1 shRNA sequence with a known target gene. Most of the isolated “colonies” were actually comprised of cells with more than one, if not several, shRNAs incorporated. While these data suggest that the current method of isolating “individual colonies” could be improved, some genes were successfully validated.

Table 1. Candidate hits from positive selection RNAi screen under LA in H1975 cells

Clone Gene Gene Name Gene Function # Symbol Vacuolar protein 1 VPS11 Sorts proteins into vacuoles sorting protein 11 H+/Cl- exchange 4 CLCN5 Chloride ion/proton channel transporter 5 Transcription Transcription termination factor, 5 TTF1 termination factor 1 role in ribosomal gene transcription Sel-1 suppressor of Adaptor protein involved in ERAD 6 SEL1L link-12-like-1 pathway E3 ubiquitin ligase, especially for 7 RNF123 Ring finger protein 123 p27 Cellular repressor of Negatively regulates transcription 8 CREG1 E1A-stimulated genes and transformation by E1A 1 10 ZNF26 Zinc finger protein 26 May play a role in transcription Ribosomal protein L7a 11 RPL7AL3 Pseudogene pseudogene 13 Neuropeptide Y GPCR whose activity is mediated by 12 NPYR5 receptor type 5 Gi/o-proteins (inhibit AC function) LIM domain Transcriptional activator, tumor 13 LIMD1 containing protein 1 suppressor in lung cancer Eukaryotic translation Involved in cap recognition and 15 EIF4A2 initiation factor 4A2 mRNA binding to ribosome Solute carrier family Carries solutes across mitochondrial 19R SLC25A40 25 member 40 membrane Transcription factor, promotes 22 MEIS1 Meis homeobox 1 tumorigenesis through inhibition of TGF-beta

2.4 Validation of Results

Our validation strategy was to first create stable cell lines with depletion of the target gene through lentiviral delivery of at least 2 independent shRNAs that were distinct from the shRNAs used during the screen. Loss of gene expression was tested at the mRNA level in the initial validation. Next, in parallel to a scramble control shRNA line, the gene knockdown cells were treated with LA for 2-4 weeks and the surviving number of cells counted by trypan blue exclusion. The experimental status of each of the

13 candidates (from Table 1) is outlined in Table 2. From validation experiments, three genes were considered validated: SEL1L, RNF123 and LIMD1. In the rest of this section,

I will describe these genes, their importance to cancer biology and outline potential hypotheses to test in future experiments.

Table 2. Experimental validation status and results of the 13 candidates from the Positive Selection Screen under LA

Gene Validation Validation result Current Experimental Status Symbol considered? Requires repetition in other SEL1L Yes Successfully validated cells and mechanistic investigation. Requires repetition in other RNF123 Yes Successfully validated cells and mechanistic investigation. Could not repeat phenotype Initially successfully LIMD1 Yes upon further attempts. Not validated pursued further. NPYR5 Yes Not attempted Validation could be pursued SLC25A40 Yes Not attempted Validation could be pursued CLCN5 Yes Not attempted Validation could be pursued EIF4A2 No Not attempted Validation could be pursued RPL7AL3 No Not attempted Validation could be pursued TTF1 No Not attempted Validation could be pursued VPS11 No Not attempted Validation could be pursued ZNF26 No Not attempted Validation could be pursued CREG1 Yes Failed validation Not being pursued further MEIS1 Yes Failed validation Not being pursued further

2.4.1 SEL1L

SEL1L was identified from the positive selection screen and so its loss was predicted to provide a survival benefit under lactic acidosis. Cell lines with shRNAs targeting SEL1L have been created and the efficiency of knockdown has been validated by qRT-PCR (Figure 1a). The shSEL1L cells were then tested for their survival under LA compared to a vector control (Figure 1b). By three different shRNAs, depletion of SEL1L protected cell number by approximately 50% after 2.5 or 3.5 weeks of LA treatment

(Figure 1b). Interestingly, although not identified as a top candidate, SEL1L also provided protection under LA in the independently performed genome-wide screen

(Figure 1c) (Chapter 3). Further validation of SEL1L’s ability to provide a survival advantage under LA should include extending this phenotype to other cell lines and to in vivo models.

The protein encoded by SEL1L plays a major role in the Endoplasmic Reticulum-

Associated Degradation (ERAD) pathway (Mueller et al., 2006; Iida et al., 2011; Lilley and Ploegh, 2005). As an adaptor protein, it forms a complex that facilitates cytosolic translocation of both glycosylated and non-glycosylated proteins for degradation

(Mueller et al., 2008; Ushioda et al., 2013) in vitro and in vivo (Sun et al., 2014). In The

Cancer Genome Atlas (TCGA) data, SEL1L was missense mutated in 3-9% of uterine carcinomas, missense mutated or truncated in 5-7% of squamous cell lung carcinomas and differentially altered in 5% of bladder cancers (cBIOPortal). Additionally, increased

SEL1L expression was associated with early stages of tumorigenesis in many cancer types and its expression was maintained throughout progression in some cancers

(Biunno et al., 2006). In breast cancer however, its decreased expression correlated with worse prognosis (Orlandi et al., 2002). Many outstanding questions remain unknown about the role of the ERAD pathway, ER stress and SEL1L’s function in tumorigenesis

(Kim et al., 2015). Persistent ER stress can cause cell death. However, tumor cells can hijack this system for their benefit (Kim et al., 2015, Croft et al., 2014). I will now outline one hypothesis I have to potentially explain the survival advantage under LA with the loss of SEL1L.

Hypothesis: Decreased SEL1L expression causes elevated ER stress, priming the ER-stress response pathway for an enhanced response to allow cells to better cope with the ER stress induced by LA.

Justification: Depletion of SEL1L leads to increased ER stress, expression of unfolded protein response (UPR) genes and decreased levels of translation (Francisco et al., 2010; Sun et al, 2014). Lactic acidosis induces ER stress and UPR genes as well

(Prabhu et al., 2012; Tang et al., 2012). If basal levels of the ER-stress/UPR pathways are increased prior to the actual exposure to LA, cancer cells may be better able to cope with higher levels of stress without causing cell death. In other words, with pre-emptive activation of this stress pathway, tumor cells may more quickly respond to stress and thus survive stress more effectively.

Approach: First, I suggest examining the extent of ER stress induction upon

SEL1L knockdown, both under basal conditions and LA treatment. If levels of ER stress- response genes, such as CHOP, ATF3 and spliced XBP-1, are up-regulated on a basal level in the SEL1L depleted cells, this may suggest a “primed” ER-stress response capable of coping with higher levels of stress. This “primed” ER-stress response pathway could permit faster and/or higher induction of the stress-response pathway.

Since the survival advantage of this knockdown is not observable until after weeks of treatment, it will be important to test the activation of ER stress and the UPR at multiple time points during the first two weeks of treatment. If the proposed experiments suggest that ER-stress induction has been altered with SEL1L knockdown, then blocking the ER- stress pathway in SEL1L knockdown cells should abolish their resistance and restore their sensitivity to LA. Extending these experiments to xenograft or inducible genetic cancer models will determine if SEL1L exhibits anti-tumor effects in vivo. Systemic treatments targeting acidosis responses, such as bicarbonate administration, will help determine if SEL1L’s role in tumorigenesis is dependent on its loss providing a protective effect under LA.

Results from testing these hypotheses will show if SEL1L knockdown alters the cells’ ER-stress response system to eventually allow for improved cell survival under

LA. These experiments will help determine the role of SEL1L in tumorigenicity.

Figure 1. Validation of SEL1L depletion as protective under LA.

(A) qPCR data showing relative abundance of SEL1L in shRNA cell lines. (B) Viable cell number as determined by trypan blue exclusion and cell counting in indicated shRNA cell lines after extended LA treatment (n=3 per bar). (C) Heat map showing result of SEL1L in genome-wide shRNA pooled screen (Chapter 3). Each column is a different shRNA targeting SEL1L. Each row is a different screen replicate under the designated condition. Green box highlights the screen where SEL1L loss is predicted to be protective. All data are from the H1975 cell line.

2.4.2 RNF123

As a hit in the positive selection screen, RNF123 depletion was predicted to protect cells from an extended LA stress. While the RNF123 gene provided a benefit under 2.5 weeks of LA stress, its protective phenotype decreased over time so it might not be as strong of a candidate to pursue (Figure 2b). CRISPR-Cas9-mediated loss of protein may be necessary to sustain the RNF123 depletion to create a more significant effect. Testing this phenotype in other cell lines or in vivo will help to determine whether loss of RNF123 offers protection in multiple cancer cells to support its role as an interesting candidate.

The protein encoded by RNF123 (KPC1) is the E3 ubiquitin ligase of the KPC complex that degrades CDKN1B (p27) during G1 (Kamura et al., 2004). TCGA data shows that RNF123 is mutated in 5-13% of stomach carcinomas, deleted in 4-12% of ccRCC and mutated or deleted to a modest extent in a number of other cancers

(cBIOPortal). These data on the loss and mutation of RNF123 are consistent with our finding that loss of this gene’s function may offer a survival benefit under LA. However, there is limited literature about this protein. While RNF123 is best recognized as a negative regulator of the cell cycle regulator p27Kip1, a recent study showed it also targeted the p105 isoform of NF-kB for degradation (Kravtsova-Ivantsiv et al., 2015). The hypothesis that I outline next suggests that NF-kB is the critical target of the KPC complex to promote cell survival under LA.

Figure 2. Validation of RNF123 depletion as protective under lactic acidosis.

(A) qPCR data showing relative abundance of RNF123 in shRNA cell lines. (B) Viable cell number as determined by trypan blue exclusion and cell counting in indicated shRNA cell lines after extended LA treatment (n=3 per bar). (C) Heat map showing result of RNF123 in genome-wide shRNA pooled screen (Chapter 3). Each column is a different shRNA targeting RNF123. Each row is a different replicate of genome-wide screen under the designated condition. Green box highlights the screen where RNF123 loss is predicted to be protective. All data are from the H1975 cell line.

Hypothesis: Loss of RNF123 provides a survival advantage under LA by relieving the negative regulation on NF-kB under LA, thus allowing NF-kB to activate anti-apoptotic and pro-cell cycle progression gene expression programs under LA.

Justification: NF-kB has been implicated in the response to acidosis in cancer cells

(Gupta et al., 2014; Peppicelli et al, 2013; Karashima, et al., 2003) and as an activator of anti-apoptotic mechanisms (Karin and Lin, 2002), proliferation and cell cycle genes

(Dolcet et al., 2005). RNF123 was recently described as a negative regulator of NF-kB

(Kravtsova-Ivantsiv et al., 2015). Importantly, NF-kB activation was only seen in cancer cells in response to acidosis, not non-malignant cells (Gupta et al., 2014). If the loss of

RNF123 can enhance the baseline and/or acidosis-induced expression of NF-kB, then the enhanced NF-kB activities can induce a proliferative and anti-apoptotic response to promote cell survival under LA.

Approach: To test this hypothesis, we will determine the level of NF-kB pathway activity in control vs. RNF123-depleted cells under control, acidosis or LA conditions.

We can determine the NF-kB activity by examining induction of target genes and activation of a luciferase reporter. In parallel, we will determine if NF-kB depletion under LA phenocopies the loss of RNF123 and protects cell viability after an extended

LA treatment. If the loss of NF-kB is protective under LA, then NF-kB overall protein levels and nuclear localization should be tested by western blot in response to LA with and without RNF123 depletion. Additionally, since NF-kB can regulate invasive

properties of cancer cells under acidosis (Gupta et al., 2014); NF-kB’s role in the invasion and mobility of RNF123 depleted cells under LA should be measured. Testing this hypothesis will be an important first step to determining if RNF123 mediates cancer cell survival under LA through NF-kB.

2.4.3 LIMD1

LIMD1 was identified from the positive screen and its phenotype of protecting cells under LA was preliminarily validated in H1975 lung cancer cells (Figure 3b). After

4 days of LA treatment, LIMD1 depletion increased the surviving number of cells by approximately 2-fold when compared with control cells (Figure 3b). A significant caveat of this data was my inability to reproduce this phenotype when I began to change the media at day 2 of treatment to maintain acidosis, rather than no media change for 4 days. Without a media change in 4 days, the pH increases to a more neutral pH and so the stress condition is not maintained. Thus, the slight increased basal growth rate of the shLIMD1 cells could explain the ~2-fold increase in cell number. This suggests that the phenotype I documented could be observable under only specific experimental conditions, so careful re-examination of the phenotype would be required if this gene was pursued further.

However, the overexpression of LIMD1 did show sensitivity to LA, thus showing a “rescued” phenotype and lending more credence to the original data. Due to these

Figure 3. Validation of LIMD1 loss as protecitve under LA.

(A) Western blot showing decreased LIMD1 expression at the protein level with 2 shRNAs. (B) Cell number as determined by trypan blue exclusion and counting in indicated shRNA cell line after 4 days of LA (n=9). (C) Western blot of LIMD1 structure- function mutants: sh=shRNA cell line, rr=RNAi resistant overexpression, delPHD2/VHL= loss of both PHD2 and VHL binding domains. (D) Cell number as determined by cell counting with trypan blue exclusion after 4 days of treatment in indicated cell line (n=9). All data are from the H1975 cell line.

complications, I believe further investigation of LIMD1 should be conducted only after further successful validation. Below, I outline the hypotheses that I have tested about

LIMD1’s role under LA.

The gene LIMD1 encodes an adaptor protein that interacts with a number of major signaling and stress response pathways in lung cancer cells (Kadrmas and

Beckerle, 2004; Sharp et al., 2004 and 2008; James et al., 2010; Foxler et al., 2012). LIMD1 expression is altered in breast cancers (Huggins et al., 2007; Huggins and Andrulis, 2008;

Spendlove et al., 2008), decreased in head and neck cancers (Ghosh et al., 2008 and 2010), and it is a bona-fide tumor suppressor in non-small cell lung cancer (Sharp et al., 2008).

The domains necessary for many of LIMD1’s protein-protein interactions have been determined. Dr. Tyson Sharp generously provided several structure-function mutant constructs of LIMD1 to facilitate this investigation. Among different partners, LIMD1 simultaneously binds PHD2 and VHL to allow for more efficient degradation of HIF-1α

(Foxler et al., 2012). Therefore, decreased LIMD1 expression may stabilize HIF-1α and activate its target genes to allow for tumor cells’ survival under LA.

Hypothesis: LIMD1 depletion leads to HIF-1/2α stabilization and activation of target genes to protect from a LA stress.

Justification: Hypoxia-induced glycolysis increases lactate production and contributes to the LA and acidosis in the tumor (Chiche et al., 2010). Although, HIF-1/2α protein levels are repressed under a combined hypoxic + acidic condition (Tang et al.,

2012; Parks et al., 2013), their hypoxic responses can protect against acidosis (Parks et al.,

2013; Chiche et al, 2009). LIMD1’s interaction with PHD2 and VHL promotes efficient degradation of HIF-1/2α (Foxler et al., 2012). Thus, the loss of LIMD1 could promote survival under LA through stabilization of HIF-1/2, which are normally repressed by acidosis.

Approach: We will determine how the overexpression of different mutant

LIMD1 constructs with deletions of specific protein-interaction domains affects cell survival under LA. The relative survival of different mutant LIMD1 constructs under LA will suggest the most important LIMD1 domains, their relevant binding partners and downstream effectors for the survival phenotype. Our initial transient transfection experiments of the LIMD1 mutant without PHD2 and VHL binding domains failed to reverse the protective phenotype of LIMD1 loss under LA. Therefore, we generated cell lines stabling expressing these constructs of LIMD1 (Figure 3c). Stable cell lines with a mutated VHL/PHD2 interaction domain in LIMD1 showed subtle differential growth, but not sensitivity to LA, even with strong overexpression (Figure 3d).

Since LIMD1’s interaction with PHD2 and VHL did not affect cell survival under

LA, other LIMD1-regulated pathways, such as pRB and the Hippo Pathway, may be responsible for this phenotype. Similar structure-function phenotype experiments could be done with a mutant LIMD1 missing its pRB-binding domain (Sharp et al, 2004).

LIMD1 regulates the Hippo pathway causing transcriptional activation by inhibiting the

phosphorylation of YAP/TAZ (Das Thakur et al., 2010). To test if the Hippo Pathway is involved, the phospho- and total-Yap levels with or without LIMD1 knockdown in control or LA conditions should be measured. Since cell density and media sphingosine levels affect the Hippo Pathway, it will be important to measure Yap levels at a variety of cell densities and with a constant media volume.

These structure-function analyses would determine which interaction domain and putative downstream pathway of LIMD1 confers a survival advantage under LA. If a survival phenotype under LA were confirmed and narrowed down to a particular interaction domain, this would lend mechanistic insight into how LIMD1 acts as a tumor suppressor in lung cancer and would suggest that LA-targeting therapies may be beneficial in tumors with defective LIMD1.

2.4.4 Other candidates of interest

CLCN5 is a chloride ion-proton antiporter. This gene is interesting for its role in lactic acidosis survival because of the importance of pH balance under this stress and the obvious impact the role of this protein could have on that process. Its function has not been studied in cancer biology.

SLC25A40 is a mitochondrial ion transporter, located in the inner mitochondrial membrane. Considering the connections between lactic acidosis, metabolism and mitochondria, this gene is also particularly interesting under LA. Its function has not been studied in cancer biology.

2.5 Future considerations

Overall, this experiment showed that a positive screen under TME stress conditions was possible. However, the high level of background (surviving cells in the control condition) likely resulted in a number of false-positives in the potential candidates. To reduce the background and improve the efficiency of the screen, the screen setup should be altered to include multiple rounds of stress or combinations of stress. Additionally, the mixed populations of shRNAs in each “colony” suggest that a more thorough cell suspension technique than trypsinization should be used.

Integration of these candidates with a genome-wide screen would also give confidence to genes that show a similar phenotype in both experiments.

3. Genome-wide Functional Genomic Screens under Hypoxia and Lactic Acidosis

The validation of two genes presented in this Chapter (ACC1 and ETV4) is included in the published, peer-reviewed manuscript: “ACLY and ACC1 Regulate

Hypoxia-Induced Apoptosis by Modulating ETV4 via α-ketoglutarate.” Keenan MM,

Liu B, Tang X, Wu J, Cyr D, Stevens RD, Ilkayeva O, Huang Z, Tollini LA, Murphy SK,

Lucas J, Muoio DM, Kim SY, Chi JT. PLoS Genet. 2015 Oct 9;11(10):e1005599. doi:

10.1371/journal.pgen.1005599. eCollection 2015 Oct.

3.1 Introduction

The goal of this portion of my thesis was to assess how individual gene knockdown, on a genome-wide scale, affected cell survival under TME stresses.

Genome-wide RNAi contextual screens were used to identify genes that specifically affected cell survival under hypoxia or LA in an unbiased manner. These two stresses were focused on to maintain focus and feasibility. The design of these genome-wide screens was such that both the negative and positive regulators of stress survival could be detected through microarray technology. Thus, genes that cause “synthetic sick” phenotypes were detected even though they decreased cell fitness under stress. While mimicking the effects of synthetic lethal genes is the easiest logical step toward developing new therapeutics, reversing the effects of “synthetic survival” genes is an equally useful strategy to prevent cell survival under these TME stresses. Thus, through 58

the detection of both positive and negative regulators of stress survival and by using a genome-wide library, we were able to broaden our scope of understanding TME-stress adaptation mechanisms.

When this study began, it was the only contextual screen done under TME stress conditions. Now, 5 years later, two RNAi screens under hypoxia are described in the literature (Yoshino et al., 2012; Pan et al., 2013), but so far, no other studies have published genome-wide screens under hypoxia or LA. This Chapter will detail the initial screens, multiple types of analyses I conducted to identify top candidates and describe five genes that were successfully validated. Most analyses and genes highlighted are of interest for future studies, but two of the validated genes (ACC1 and ETV4) were researched in detail and this is described in Chapter 4. Further interrogation of top candidates would continue to identify novel mechanisms of stress survival by cancer cells and could lead to identification of additional therapeutic targets in the future.

3.2 Methods

3.2.1 Genome-wide pooled shRNA screen

To identify genes that modulate cell survival under lactic acidosis and hypoxia, we conducted genome-wide, shRNA-based, contextual pooled screens in the lung cancer cell line H1975 under hypoxia or lactic acidosis (Figure 5a). To preferentially discover

genes important for survival rather than proliferation, the screen was done in low proliferative conditions due to reduced serum in the culture media.

H1975 cells were spin-infected with the pMSCV-based retroviral genome-wide library (Schlabach et al., 2008) at an MOI of 0.3 and divided into six sub-pools to achieve a final library representation of 1000 cells per shRNA (Schlabach et al., 2008). After three days of selection under 1ug/ml puromycin, cells were split into control and stress conditions, maintaining 1000-fold representation of each shRNA per triplicate. While the puromycin selection could introduce a survival bias, it was a necessary step for selection for integrated and expressing shRNAs. Cells were serum starved to 0.1% FBS 24 hours after plating. 24 hours after serum starvation, media was changed to treatment media

(0.1% FBS, 25mM Hepes) with control (21% O2) and hypoxia (2% O2) media at a pH=7.4 and lactic acidosis (21% O2) media (25mM lactic acid, Sigma cat. no L6402) adjusted to pH=6.7. All medias were pH’d the day before use and filter sterilized. These conditions mimicking the TME stresses are relevant to conditions measured in vivo and were as performed previously (Chen et al., 2008; Chen et al., 2010a). At these stress treatments, there was a ~50% reduction in cell number, which allowed us to uncover both genes whose suppression either reduced or improved survival under stress. After 4 days of treatment, cells were harvested, centrifuged, and frozen at -80°C.

Genomic DNA was extracted with the QIAamp DNA Blood Maxi kit (QIAGEN, cat.no 51194) then the shRNA sequences were PCR amplified. The amplified products

from the control and each stress were labeled (Cy3 and Cy5, respectively) and then interrogated by a custom Agilent microarray, which contained probes against the library’s shRNA sequences (Schlabach et al., 2008). The custom array was modified slightly from similar arrays used in other shRNA screens (Luo et al., 2009; Schlabach et al., 2008). We validated the sensitivity and specificity of the array to different ratios of labeled PCR product in our hands (Figure 4a). Therefore, the relative hybridization of the Cy5/Cy3 labeled shRNA populations determined the abundance of each shRNA under control, hypoxia or LA. The Cy3 and Cy5 signals across the three biological replicates were highly reproducible (Figure 4b). Probes with signal intensities of less than 2-fold above background were discarded.

The abundance of each shRNA sequence in the stress versus control condition reflected the effect of its target gene on cell survival under stress: if the shRNA was depleted in the stress treatment, the gene it targeted had a “synthetic sick/lethal” phenotype; if the shRNA was enriched in the stress treatment, the gene it targeted had a

“synthetic survival/protective” phenotype under stress. In order to analyze the effect of each shRNA in stress, we calculated an “R/G” ratio of Cy5/Cy3 values for each shRNA.

R/G ratios for probes above background were calculated, log2 transformed and quantile normalized across pools. R/G ratios were distributed on a scale of +/- 4.0 that was highly consistent between replicates and stresses, although many fell in the “unchanged” range of +/-0.5 (Figure 4c). The R/G ratios were then analyzed by a number of methods to

determine the top candidate hit genes that affected cell survival under hypoxia or LA.

The details of these analyses are presented in Section 3.3.

3.2.2 Cell culture, TME stress treatments and generation of stable shRNA cell lines

Lactic acidosis was generated via addition of lactic acid (Sigma-Aldrich, St.

polybrene (final concentration 8ug/ml). Complete death in blank infection dishes was used to determine success of infection and puromycin selection. Efficiency of silencing or overexpression was determined by qPCR or western blots. Concentration of puromycin needed for selection on H1975 cells= 1ug/ml.

3.2.3 Crystal violet staining

Cells were fixed either in 4% paraformaldehyde (PFA) overnight at 4°C or at room temperature for 30 min. PFA was removed and crystal violet staining solution

3.2.4 Determination of cell number

3.2.5 Flow cytometry

For cell cycle analysis, after 4 days of stress treatment, media was collected, cells were trypsinized and pooled with the media. Cells were centrifuged (5 min, 1000rpm,

4°C) then fixed by resuspension in ice cold 70% ethanol while gently vortexing. Fixed cells were placed at -20°C until prepared for FACS analysis. Immediately before FACS analysis, cells were centrifuged for 5 min at room temperature, washed twice in PBS

3.2.6 Protein lysate collection and Western blots

3.2.7 Quantitative real-time PCR

RNA was extracted using the RNeasy Kit (QIAGEN). 1 µg of total RNA was reverse transcribed by SuperScript II (Invitrogen) for real-time PCR with Power

SYBRGreen Mix (Applied Biosystems/Life Technologies (Grand Island, NY, USA)).

Primers were designed across exons whenever possible and were verified for specificity by regular PCR prior to use in real-time PCR. Please refer to my “Designing real time primers” spread sheet for the sequences of primers used.

3.3 Analysis of genome-wide shRNA pooled screens

3.3.1 Multiple Hairpin Analysis

To minimize false positives due to potential off-target effects of individual shRNAs, we first performed a multiple hairpin analysis to specifically identify the genes that had at least two distinct shRNA sequences that were enriched or depleted above defined cutoffs. The numbers of genes that resulted from this analysis are outlined in

Table 3. Appendix A includes a list of all of these multiple hairpin hits (Table 12). For this “Multiple Hairpin Analysis,” genes were considered a hit Multiple Hairpin Hit

(MHH) when they had

(1) at least 2 different shRNAs with an (absolute value R/G) > 0.7 in at least 2 of the three biological replicates, and

(2) the (standard deviation/average) of the biological replicates was <0.5.

Importantly, this “Multiple Hairpin Analysis” identified EPAS1 (hypoxia- inducible factor 2α, HIF-2α) as a synthetic lethal gene under hypoxia. We further functionally validated this by showing that silencing EPAS1 by shRNA significantly reduced cell survival under hypoxia (Figure 5b, c). This result was consistent with the critical role of EPAS1 in cellular adaptation to hypoxia (Majmundar et al., 2010).

Therefore, the “re-discovery” of EPAS1 as a synthetic lethal hit provided confidence in our screen and analysis methods.

Although an attempt was made to perform Gene Ontology analyses with multiple online programs (DAVID, GOrilla, Panther), with either the library or the human genome as the “total gene list”, no pathways or gene ontology groups reached statistical significance for different categories of “multiple hairpin hits” (MHHs).

Additionally, there was little overlap between the enriched or depleted MHHs under LA and hypoxia (Figure 5e). This was consistent with past reports of distinct responses and adaptations to hypoxia and lactic acidosis (Chen et al., 2010a; Tang et al., 2012).

Interestingly, this MHH analysis identified the same number of hits in the two LA categories (enriched or depleted shRNAs). More genes were hits in hypoxia overall than

LA. This could originate from these lung cancers cells being exposed to different oxygen levels throughout their lifetime and the critical necessity for these cells to function under hypoxia. The category with the most number of MHH candidate genes was the hypoxia synthetic survival category. While no pathways or larger groups of genes were

identified in this set of hits, a careful analysis of these genes does reveal a number of related genes and pathways within these MHHs.

Figure 4. Quality control analysis of shRNA microarray and screens.

(A) Correlation plots showing distinguishable ratios of differentially mixed PCR products by the custom microarray. (B) 3-D scatterplots showing reproducibility between biological triplicates’ of hypoxia and lactic acidosis treated samples for both Cy3 and Cy5 signals. (C) Distribution of R/G ratios by number of shRNAs, separated by treatment (LA=lactic acidosis, H=hypoxia) and replicate (n=3).

Table 3. Overview of results from Multiple Hairpin Analysis

Category of Number of Lactic Acidosis Hypoxia shRNA Hairpins shRNAs enriched in stress 2 83 158 3 3 13 4 0 2 total 86 173 shRNAs depleted in stress 2 83 114 3 3 7 4 0 1 total 86 122

First, the gene ontology group of nuclear hormone receptor pathway was modestly enriched when all of the MHHs were considered as one group against a background of all the genes in the library. These genes included: NR2F2 (Hypoxia-

Lethal); NR4A3 (LA-Lethal); AR, ESRRB, NR1H4, NR2C1, NR3C2 and NR5A1

(Hypoxia-Survival); ESR2, NF0B2, NR1I2, NR2E3, NR5A1, NR6A1 and NRIP1 (LA-

Survival). So, although these hits do not fall in to one category, this result suggested the involvement of various nuclear hormone receptor signaling pathways in cancer cell adaptation to stress. ESRRB and ESR2 are involved in estrogen receptor signaling and are critically important in cancer risk, prognosis and treatment (Hynes and Lane, 2005).

AR encodes the androgen receptor, which is vital for responding to androgens. In prostate cancer, androgen-insensitivity after initial treatment of androgen withdrawal is a major challenge that limits the clinical efficacy of this treatment (Feldman and

Feldman, 2001). While some of these nuclear hormone receptors are well understood, others remain uninvestigated and so their relevance to cancer biology should include consideration of the effects of TME stresses.

Second, two isoforms of the same enzyme, acetyl-CoA carboxylase, were hits in multiple categories. ACACA (ACC1), which encodes the mainly cytosolic isoform, was predicted as a high confidence MHH in the hypoxia survival category. ACACB (ACC2), which encodes the isoform mainly located at the outer mitochondrial membrane, was predicted as a MHH in the LA-survival and hypoxia-lethal categories. These results

suggested an important role for this enzyme, which performs the rate-limiting step in de novo lipogenesis, in regulating stress survival. It also suggested that these different isoforms, while sharing enzymatic activities, have different functions under stress. The appearance of these two isoforms in multiple stresses further supported the functional and mechanistic investigation into ACACA (validated in Section 3.4.1 and investigated mechanistically in Chapter 4).

There were nineteen genes that appeared as hits in more than one category

(survival/lethal/hypoxia/LA) (Table 4). If the assumption is made that appearance in more than one category more strongly implicates that gene in stress response, then these genes may deserve additional attention. However, for three genes (CPLX2, NR1I2 and

SKP1) the screen predicted both survival and lethal phenotypes in the same stress treatment. This could be due to a number of technical artifacts (off-target of RNAi, greater representation in the library) and should be removed from further consideration.

Table 4. Genes that were Multiple Hairpin Hits (MHHs) in more than one category

Gene Symbol shRNA Hit Category 1 shRNA Hit Category 2 Stress Phenotype Stress Phenotype ACACB Hypoxia SL LA Sur AHCYL1 LA SL Hypoxia Sur CPLX2 Hypoxia SL Hypoxia Sur CXorf34 Hypoxia SL LA SL DEFB107A Hypoxia Sur LA Sur EXOSC8 Hypoxia SL LA Sur FANCA Hypoxia SL LA SL LOC730273 Hypoxia Sur LA SL NR1I2 LA SL LA Sur NR5A1 Hypoxia Sur LA Sur PRDX3 Hypoxia Sur LA Sur RNF14 Hypoxia SL LA SL SKP1 LA SL LA Sur SYT6 Hypoxia Sur LA Sur TRBV24-1 LA SL Hypoxia Sur UNQ6125 Hypoxia Sur LA Sur UTP11L Hypoxia SL LA SL USH1C Hypoxia Sur LA Sur ZNF571 LA SL Hypoxia Sur

Upon literature investigation, there were seven MHHs related to the E3 ubiquitin ligase, COP1 (RFWD2). COP1 was identified as a negative regulator of light responsive growth in plants, but recently has been linked to cancer biology (Marine, 2012). Often considered a tumor suppressor, its role in tumor progression may vary as its targets include both tumor suppressors, such as p53 (Dornan et al., 2004), and oncogenes, such as the PEA3 family (Vitari et al., 2011; Baert et al., 2010), JUN (Wertz et al., 2004;

Migliorini et al., 2011) and MTA1 (Li et al., 2009). One MHH, ETV4, is a direct target of

COP1 (Vitari et al., 2011; Baert et al., 2010). The hypoxic synthetic survival phenotype for

ETV4 was validated (Section 3.4.3) and investigated in detail (Chapter 4). Four other genes facilitate the degradation of proteins by the COP1 complex. COPS6, a MHH predicted as hypoxia lethal, is a component of the COP9 signalosome and it stabilizes

COP1 protein levels (Choi et al., 2001). Cullin 2 (CUL2) was a hypoxia survival hit that promotes the ubiquitination of substrates by COP1 (Olma et al., 2009). DDA1, also predicted as a hypoxia survival hit, interacts with DET1, an important binding partner to stimulate ubiquitination by COP1 (Pick et al., 2007; Lau and Deng, 2012). YWHAZ is a monooxygenase whose knockdown was also predicted to be protective under hypoxia; it promotes ubiquitination by altering COP1 localization (Su et al., 2010). Additionally,

MLF1, a synthetic lethal gene under LA, inhibits the activity of COP1 to stabilize p53 through CSN3, another MHH (Yoneda-Kato et al., 2005). Careful examination of the library revealed that it does not contain any shRNAs targeting COP1; therefore this gene

was not evaluated in this experiment. No current studies have investigated this potential tumor suppressor under hypoxia. These analyses make the investigation the importance of COP1 under hypoxia a high priority for future studies.

The multiple hairpin analysis was a stringent filter, with only 467 genes out of approximately 19,000 evaluated passing as candidate hit genes across all categories of shRNA phenotype. The stringency of this analysis likely contributed to the inability to identify GO enrichments. This stringency also means that the MHH genes were high confidence candidates. Yet, additional analysis methods can have different advantages and overlapping genes between multiple modes of analysis are the genes of highest confidence to be successfully validated.

3.3.2 RIGER Analysis

In order to analyze the genome-wide shRNA screen dataset in another way, we performed a RIGER analysis with the GENE-E software using the log-fold change and the second best shRNA for each gene criteria for each category of shRNA phenotype

(Gould, www.broadinstitute.org). The second best shRNA method ignores the top performing shRNA and ranks genes based on their second-best performing shRNA.

Thus, it relies on a robust and consistent result in additional hairpins. The RIGER analysis is always a comparison of how different the phenotype for one shRNA is between two “Classes” (or screens). In these analyses we always used the H1975 proliferation screen, which was conducted under non-stressed growth conditions,

performed at Duke by So Young Kim and Beiyu Liu as the “other” Class. RIGER produces a ranked list of all genes in the library, rather than establishing a cutoff to consider a select number of genes, which allowed for more successful gene ontology analyses. Tables 5-8 list the top 20 genes in each of the four “categories” with their 1/NES

(normalized enrichment score) scores (which reflect the strength of effect). For both stresses, the 1/NES scores were higher (the Classes were more different) for the top survival genes than the lethal genes. This may reflect a greater similarity between lethal genes under stresses vs. proliferation and a greater difference between survival genes under stresses vs. proliferation. Nevertheless, by the 20th hit in each category, the 1/NES values were similar, meaning that past the top performing few hairpins, the strength of phenotype predicted for these different categories was similar. A discussion of how these results compare to the multiple hairpin analysis is in Section 3.3.4. I will now highlight the results of these analyses from each category of shRNA phenotype.

In the hypoxia survival group, there was an enrichment of genes affecting mRNA regulation and binding, as well as membrane dynamics and nuclear localization

(using GOrilla, Eden et al., 2009). Interestingly, there was also an enrichment of genes associated with the kinetochore and lysosomes. The top gene of this analysis was HSPE1

(HSP10), Heat Shock Protein 10kDa Protein 1, a chaperonin that forms heterodimers with other heat shock proteins (HSP60) to promote protein folding (David et al., 2013).

This protein is essential for mitochondrial biogenesis (David et al., 2013), which

suggested an important function for mitochondria under hypoxia. Although HSPE1 was up-regulated under hypoxia in the Indian Catfish (Mohindra et al., 2015) and has been implicated in cancer biology (David et al., 2013), no studies have examined its importance in tumor hypoxia.

For the hypoxia synthetic lethal genes, the only statistically significant GO enrichment was a modest enrichment of ubiquitin processing and binding when the top

500 genes in the RIGER analysis were compared to the library background (Eden et al.,

2009). The two top ranking genes both had a 1/NES=206: ZDHHC20 and OPRMI.

ZDHHC20 (Zinc Finger, DHHC-Type Containing 20) has only one paper published on it, which shows that it acts as a palmitoyl acyltransferase with transformative abilities when overexpressed and that it is up-regulated in a number of cancers (Draper and

Smith, 2010). OPRMI (Opioid Receptor, Mu 1) is a GPCR known for its role in pain sensation, but has also been proposed as a target in cancer therapy as studies have found a correlation between its expression or stimulation and angiogenesis and tumor progression (Singleton et al., 2015; Koodie et al., 2010).

The LA survival genes were enriched in genes that encode RNA binding proteins, ribonucleoprotein complex members and components of the spliceosome.

Although not statistically significant (p=0.0667), another identified ontology group was the Hrd1p ubiquitin ligase complex that included the gene SEL1L, which was identified in our positive selection screen to promote survival under LA (Figure 1). SEL1L ranked

#118 by the RIGER analysis, further supporting the validated phenotype for this gene under LA during the positive screen in Chapter 2. This category also contained the strongest single performing hairpin in any of the RIGER analyses, with PLK1 being predicted as synthetic lethal under LA with a 1/NES of 1697 (a value nearly 3 times greater than the next highest value across any of the analyses). PLK1 has been identified in other genome-wide screens as essential to RAS-driven cancers’ proliferation (Luo et al., 2009) and a target in osteosarcoma (Duan et al., 2010), neuroblastoma (Grinshtein et al., 2011), pediatric rhabdomyosarcomas (Hu et al., 2009), malignant pleural mesothelioma (Linton et al., 2014) and hormone-independent ER-positive breast cancer

(Bhola et al., 2015). While this kinase has an important role in sustaining viability of many cancer cells, this predicted phenotype suggests caution when targeting it, since its depletion may protect cells under LA. Second to PLK1, the completely unannotated and unstudied LOC642250 was predicted as a synthetic survival gene under LA. The third hit was a mitochondrial pyrimidine nucleotide transporter (SLC25A33) that maintains mitochondrial respiration (Di Noia et al., 2014; Favre et al., 2010). These genes have not been implicated in LA biology so, if validated, these results would reveal novel roles for these genes in cancer biology.

Finally, for the LA synthetic lethal RIGER analysis, enriched pathways were inositol phosphate dephosphorylation and ubiquitin conjugating enzyme binding (Eden et al., 2009). Thus, ubiquitin processing appeared in both of the synthetic lethal stress

categories. The top gene of this list was RNU12P, which encodes the U12 small nuclear

RNA that has not been functionally investigated. The top synthetic lethal gene under hypoxia, ZDHHC20, ranked 15th in this LA-synthetic lethal list, suggesting this gene may function under multiple stress conditions.

Since the GENE-E software permits analyses with multiple different criteria, additional analyses could be done on these genome-wide screen datasets in the future, especially if improved methods are developed and integrated into the GENE-E platform.

The top 20 gene lists highlight the very top performing genes by this analysis. Highest confidence candidates are considered top candidates by multiple analysis methods. In the next Section, I briefly describe the overlap between the Multiple Hairpin Analysis

(3.3.1) and these RIGER gene ranking analyses (3.3.2).

Table 5. Top 20 genes from RIGER analysis of the hypoxia survival condition

RANK Gene Symbol Gene Name 1/NES 1 HSPE1 Heat Shock 10kDa Protein 1 449.8426 Nudix (Nucleoside Diphosphate Linked 2 NUDT21 138.6001 Moiety X)-Type Motif 21 Ubiquitin-Like Modifier Activating 3 UBE1 108.1666 Enzyme 1 LanC Lantibiotic Synthetase 4 LANCL3 102.3856 Component C-Like 3 (Bacterial) Chromosome 11 Open Reading Frame 5 C11orf71 101.6363 71 6 NCKAP1L NCK-Associated Protein 1-Like 67.11409 DnaJ (Hsp40) Homolog, Subfamily C, 7 DNAJC6 56.56109 Member 6 8 PLK1 Polo-Like Kinase 1 56.33803 Ubiquitin A-52 Residue Ribosomal 9 UBA52 55.67929 Protein Fusion Product 1 STT3B, Subunit Of The 10 STT3B Oligosaccharyltransferase Complex 50.65856 (Catalytic) 11 RPL23 Ribosomal Protein L23 49.48046 12 NRCAM Neuronal Cell Adhesion Molecule 45.89261 13 ACACA Acetyl-CoA Carboxylase Alpha 41.98153 14 ARCN1 Archain 1 40.63389 Adaptor-Related Protein Complex 1, 15 AP1S1 38.58025 Sigma 1 Subunit 16 XRN1 5'-3' Exoribonuclease 1 38.05175 17 C15orf2 Nuclear Pore Associated Protein 1 37.67898 18 NUP210 Nucleoporin 210kDa 37.59398 19 NUP43 Nucleoporin 43kDa 35.89375 20 NOTCH3 Notch 3 35.77818

Table 6. Top 20 genes from RIGER analysis of the hypoxia synthetic lethal condition

RANK Gene Symbol Gene Name 1/NES Zinc Finger, DHHC-Type Containing 1 ZDHHC20 206.5688907 20 2 OPRM1 Opioid Receptor, Mu 1 206.4409579 3 TROVE2 TROVE Domain Family, Member 2 180.7664497 Amyloid Beta (A4) Precursor Protein- 4 APBA3 97.27626459 Binding, Family A, Member 3 5 CLCC1 Chloride Channel CLIC-Like 1 86.13264427 RNA Pseudouridylate Synthase 6 RPUSD2 71.68458781 Domain Containing 2 7 CHIC1 Cysteine-Rich Hydrophobic Domain 1 67.84260516 8 C8orf4 Chromosome 8 Open Reading Frame 4 66.57789614 Kelch Repeat And BTB (POZ) Domain 9 KBTBD7 62.81407035 Containing 7 10 RP11-139H14.4 Not annotated 61.462815 11 ASGR1 Asialoglycoprotein Receptor 1 58.96226415 12 CCBL2 Cysteine Conjugate-Beta Lyase 2 54.73453749 13 ELOVL6 ELOVL Fatty Acid Elongase 6 51.07252298 UDP-Glucose Glycoprotein 14 UGCGL2 49.11591356 Glucosyltransferase 2 Geranylgeranyl Diphosphate Synthase 15 GGPS1 45.47521601 1 16 TSC2 Tuberous Sclerosis 2 44.70272687 17 ZNRD1 Zinc Ribbon Domain Containing 1 40.8496732 18 LOC339535 Not annotated 40.12841091 ArfGAP With Coiled-Coil, Ankyrin 19 CENTB1 38.59513701 Repeat And PH Domains 1 20 TMEM169 Transmembrane Protein 169 37.97949107

Table 7. Top 20 genes from RIGER analysis of the LA survival condition

RANK Gene Symbol Gene Name 1/NES 1 PLK1 Polo-Like Kinase 1 1696.928559 2 LOC642250 Not annotated 602.4096386 Solute Carrier Family 25 (Pyrimidine 3 SLC25A33 74.8502994 Nucleotide Carrier), Member 33 Olfactory Receptor, Family 8, Subfamily 4 OR8D4 67.88866259 D, Member 4 Synovial Apoptosis Inhibitor 1, 5 SYVN1 61.65228113 Synoviolin Chromosome 1 Open Reading Frame 6 C1orf174 61.42506143 174 7 EXT2 Exostosin Glycosyltransferase 2 59.31198102 8 VMD2L3 Bestrophin 3 57.50431282 9 LOC729324 HCG1986447 56.02240896 Nudix (Nucleoside Diphosphate Linked 10 NUDT21 52.41090147 Moiety X)-Type Motif 21 Ash2 (Absent, Small, Or Homeotic)-Like 11 ASH2L 49.87531172 (Drosophila) 12 KIF7 Kinesin Family Member 7 48.75670405 13 NRCAM Neuronal Cell Adhesion Molecule 47.82400765 DEAD (Asp-Glu-Ala-Asp) Box 14 DDX49 43.42162397 Polypeptide 49 Amyotrophic Lateral Sclerosis 2 15 ALS2CR12 (Juvenile) Chromosome Region, 41.10152076 Candidate 12 16 CBLN4 Cerebellin 4 Precursor 40.93327876 17 H2AFZ H2A Histone Family, Member Z 39.3081761 18 LOC389970 Not annotated 39.26187672 Toll-Interleukin 1 Receptor (TIR) 19 TIRAP 38.52080123 Domain Containing Adaptor Protein Minichromosome Maintenance 20 MCM6 37.99392097 Complex Component 6

Table 8. Top 20 genes from RIGER analysis of the LA synthetic lethal condition

RANK Gene Symbol Gene Name 1/NES 1 RNU12P RNA, U12 Small Nuclear 2, Pseudogene 308.4515731 2 FLJ20035 Not annotated 283.0455703 Amyloid Beta (A4) Precursor Protein- 3 APBA3 266.8089648 Binding, Family A, Member 3 4 LOC730273 Not annotated 243.4274586 5 CHIC1 Cysteine-Rich Hydrophobic Domain 1 175.9633996 6 H2AFJ H2A Histone Family, Member J 132.5205407 7 DDA1 DET1 And DDB1 Associated 1 128.1558375 Zinc Finger And BTB Domain 8 ZNF297B 124.4555072 Containing 43 9 LOC154822 Not annotated 100.7861318 10 SNAI2 Snail Family Zinc Finger 2 92.76437848 11 LOC730926 Not annotated 79.17656374 12 THRB Thyroid Hormone Receptor, Beta 78.98894155 13 GGPS1 Geranylgeranyl Diphosphate Synthase 1 78.67820614 14 RNU5F RNA, U5F Small Nuclear 1 72.35890014 15 ZDHHC20 Zinc Finger, DHHC-Type Containing 20 72.20216606 16 SRPK3 SRSF Protein Kinase 3 71.94244604 17 SLC22A10 Solute Carrier Family 22, Member 10 68.77579092 Translocase Of Inner Mitochondrial 18 TIMM23 67.88866259 Membrane 23 Homolog (Yeast) 19 MAGEH1 Melanoma Antigen Family H1 67.38544474 20 FXN Frataxin 60.06006006

3.3.3 Integration of multiple hairpin changes with RIGER analysis

Higher confidence hits are determined by identifying the overlapping top candidates from both the multiple hairpin analysis and the RIGER ranking analyses.

Tables 9-10 outline the genes that were MHH hits and fell within the top 100 ranked genes of the RIGER lists for the same category of shRNA phenotype.

This analysis showed 47 genes were identified as hits by both methods; that number decreased to 26 when only the top 50 RIGER genes are considered. For both LA and hypoxia, the overlap between analysis methods was higher for the contextual survival category than the synthetic lethal categories. In addition to multiple methods of analyzing the screening data, integration of other types of data can help identify top candidate genes. The next section discusses how we integrated transcriptomics data with the genome-wide shRNA screen data.

Table 9. Candidate hits under hypoxia in both the multiple hairpin analysis and in the top 100 genes of the RIGER analysis, separated by gene knockdown phenotype

shRNA RANK in Gene Gene Name phenotype RIGER Symbol LanC Lantibiotic Synthetase Component C- Survival 4 LANCL3 Like 3 (Bacterial) Survival 6 NCKAP1L NCK-Associated Protein 1-Like Survival 13 ACACA Acetyl-CoA Carboxylase Alpha Adaptor-Related Protein Complex 1, Sigma 1 Survival 15 AP1S1 Subunit Solute Carrier Family 25 (Mitochondrial Survival 23 SLC25A14 Carrier, Brain), Member 14 Survival 27 HEPH Hephaestin Survival 29 LOC388946 Not annotated Dehydrogenase E1 And Transketolase Survival 31 DHTKD1 Domain Containing 1 Survival 36 HIST4H4 Histone Cluster 4, H4 Survival 42 CUL2 Cullin 2 Survival 49 IL3RA Interleukin 3 Receptor, Alpha (Low Affinity) Survival 51 AHNAK AHNAK Nucleoprotein Survival 64 CPOX Coproporphyrinogen Oxidase Olfactory Receptor, Family 1, Subfamily B, Survival 67 OR1B1 Member 1 (Gene/Pseudogene) Survival 75 FBXL13 F-Box And Leucine-Rich Repeat Protein 13 Survival 77 SOX5 SRY (Sex Determining Region Y)-Box 5 Protein Tyrosine Phosphatase, Receptor Survival 78 PTPRF Type, F Survival 81 CD6 CD6 Molecule Survival 82 ZFP42 ZFP42 Zinc Finger Protein Survival 85 DEFB107A Defensin, Beta 107A Eukaryotic Translation Initiation Factor 3, Survival 98 EIF3A Subunit A Lethal 45 ASB2 Ankyrin Repeat And SOCS Box Containing 2 Lethal 54 ZFP14 ZFP14 Zinc Finger Protein Lethal 56 R3HDM1 R3H Domain Containing 1 Lethal 60 CCDC92 Coiled-Coil Domain Containing 92 Taurine Up-Regulated 1 (Non-Protein Lethal 92 TUG1 Coding)

Table 10. Candidate hits in LA in both the multiple hairpin analysis and in the top 100 genes in the RIGER analysis, separated by gene knockdown phenotype

shRNA RANK in Gene Gene Name phenotype RIGER Symbol Survival 8 VMD2L3 Bestrophin 3 Survival 9 LOC729324 HCG1986447 Survival 12 KIF7 Kinesin Family Member 7 Survival 16 CBLN4 Cerebellin 4 Precursor Toll-Interleukin 1 Receptor (TIR) Domain Survival 19 TIRAP Containing Adaptor Protein Survival 23 LOC652818 Not annotated Survival 33 ZFP57 ZFP57 Zinc Finger Protein Ribosomal RNA Processing 15 Homolog Survival 45 RRP15 (S. Cerevisiae) Nuclear Receptor Subfamily 0, Group B, Survival 46 NR0B2 Member 2 Sarcoglycan, Beta (43kDa Dystrophin- Survival 48 SGCB Associated Glycoprotein) Survival 49 C10orf79 Cilia And Flagella Associated Protein 43 Elongator Acetyltransferase Complex Survival 50 ELP4 Subunit 4 Survival 61 LOC402420 Not annotated Survival 74 MGC16275 Uncharacterized Protein MGC16275 Survival 78 DEFB107A Defensin, Beta 107A Survival 80 ECH1 Enoyl CoA Hydratase 1, Peroxisomal Survival 85 PRDX3 Peroxiredoxin 3 Lethal 4 LOC730273 Not annotated Lethal 28 BEST4 Bestrophin 4 Phosphatidylinositol Glycan Anchor Lethal 72 PIGP Biosynthesis, Class P TRNA Methyltransferase 2 Homolog B (S. Lethal 100 CXorf34 Cerevisiae)

Figure 5. Overview of genome-wide shRNA screen strategy and results.

(A) The overall design of genome-wide pooled shRNA screen protocol. Each treatment was performed in n=3. (B) Western blot showing decreased protein level expression of EPAS1 (HIF-2α) by shRNA. (C) Viable cell number counts by trypan blue exclusion after indicated cells in 4 days of hypoxia (n=9). (D) Venn diagrams showing the overlap of genes in the multiple hairpin analysis (see Methods) across both TME stress treatments, separated by shRNA enrichment (top) or depletion (bottom) in the stress condition. (E) Analysis of screen data by RIGER with GENE-E program with the probability of a gene

being a hit indicated by 1/NES (normalized enrichment score). With the second best hairpin method, ACC1 is 13th best performing gene overall (indicated by red circle).

3.3.4 Integration of gene expression data with multiple hairpin analysis

One strategy to improve the validation rate from pooled shRNA screens is to integrate the RNAi data with other independently obtained –omics datasets. In this vein, we completed microarrays to measure relative RNA abundances and changes in the same cell line in response to hypoxia or LA. I then integrated the MHHs with the microarray data. The steps I completed for this analysis are as follows:

1) Perform Multiple Hairpin Analysis (data described here used the list from the document named “One gene hit list version 2”) 2) Perform RMA normalization and zero-transformation of microarrays 3) Create a probe list of all probes for the genes in (1) 4) File merge (3) with (2) 5) Open with Cluster 3.0, cluster by genes only, complete linkage analysis 6) Observe in Java Treeview

This analysis makes two assumptions: First, that the microarray dataset shows which genes are expressed at reasonable levels in these cells, to help rule-out hits that are not expressed. Second, that genes regulated at the mRNA level under stress are of greater importance to stress adaptation in this cell line. The mRNA changes could reflect either of two effects: they could be part of the cellular response of stress survival or their changes could contribute to cell death or cell cycle arrest. Depending on these functionalities, the expected direction of mRNA change (up/down-regulation) would be opposite for genes with synthetic sick or survival predicted phenotypes. Therefore, a more general consideration is that any mRNA change increases the confidence of a candidate “hit” gene. However, two genes that were validated for a hypoxia survival

phenotype when depleted, ETV4 and ACACA, had down-regulated mRNAs under hypoxia. Considering this, I believe that the highest confidence genes for further validation efforts are those that show one of the following patterns:

(a) A gene with shRNAs enriched in the screen and down-regulated mRNA, or

(b) A gene with shRNAs depleted in the screen and up-regulated mRNA.

For each “category” of MHH (survival/lethal/LA/hypoxia), I highlight the genes that follow this pattern and then list the additional MHHs with opposite mRNA expression changes from those outlined in (a) or (b). Overall, the genes highlighted in this section are of higher possible validation success as they were multiple hairpin hits and had altered mRNA level expression under the relevant stresses.

To reiterate, for hypoxia survival genes, the MHHs of particular interest had down-regulated mRNA under hypoxia. Both ACACA and ETV4 followed this pattern and the mechanistic explanation of their phenotype under hypoxia is the subject of

Chapter 4. STOM had 3 probes in the array data; two of these were down-regulated under hypoxia. AP1S1 (3 probes) and DHTKD1 (2 probes) had multiple microarray probes down-regulated under hypoxia; these genes are addressed again at the end of this Section. An additional 4 genes had up-regulated mRNA: PTPRF, TTF1, SLC40A1

(FPN), and COL4A3BP.

For the hypoxia synthetic lethal genes, there was increased confidence when the

MHHs had up-regulated mRNA during hypoxia. DSC2 had 3 probes that clustered

together that were hypoxia up-regulated and LA down-regulated. DSC2 was not predicted as a MHH under LA. MYBL1 had one probe up-regulated, but another that hardly responded to hypoxia. RASA1 had 3 up-regulated probes under hypoxia and LA.

NR2F2, one of the nuclear hormone receptors previously mentioned, had 4 probes up- regulated under hypoxia. CASP6 had 2 probes mildly up-regulated under hypoxia. Two genes had “opposite” from the expected mRNA regulation: AZGP1 has strongest down- regulation under hypoxia of any gene and 3 probes for SIP1 were mildly down- regulated under hypoxia.

For the synthetic survival shRNAs in LA, 4 MHH genes had down-regulated mRNAs by multiple probe sets. PIN4 had 3 probes mildly down-regulated under LA;

CDH1 had 2 probes down-regulated under LA; SGCB had 3 probes with strong down- regulation in LA; and ACACB had 6 probes changed in the arrays, but the magnitude of change was quite subtle on 5 of them. MHHs with mRNA probes up-regulated or inconsistently changed were NRIP1, ASPH, IL1RAP, PARVB, and LOC253842/GABPB2.

The last category is the LA synthetic lethal predicted genes. MATR3 had 10 probes, all of which trended toward up-regulation, but 5 were up-regulated to a greater degree. AHCYL1, a MHH in both the LA synthetic lethal and the hypoxia survival categories, had 3 probe sets that clustered tightly together and were up-regulated in LA.

MLF1 and SLC12A6 each had 2 probe sets modestly up-regulated; the predicted phenotype for MLF1 was validated (Section 3.4.4). FANCA, a lethal MHH in both

stresses, had 4 probe sets in the array dataset, but only one showed mild (less than 2- fold) up-regulation under LA. WHSC1L1 had 6 probe sets that changed with 3 strongly up-regulated. All 8 probe sets for UBE2H trended toward up-regulation, but the change was subtle (less than 2-fold). 2 probes for COL27A1 were up-regulated under both LA and hypoxia. The single probe set for CA9 changed only under hypoxia, consistent with it being a canonical HIF target gene (Wykoff et al., 2000; Bertout et al., 2008). MHH genes in this category with down-regulated mRNAs were ACTR2, SESN3, KIAA1377,

HECTD1, WWOX, SLC12A9 and C18orf8.

Additionally, all three types of analysis (MHH, RIGER, gene expression) predicted that the loss of three genes (ACACA, AP1S1, and DHTKD1) could promote hypoxic cell survival. AP1S1 and DHTKD1 only had two hairpins in the multiple hairpin analysis, while ACACA had 4 passing the cutoff criteria. This lends additional weight to focusing on ACACA in Chapter 4. One additional gene, PTPRF, was up- regulated under hypoxia and was predicted to be a hypoxic survival gene when depleted, strongly implicating the regulation of PTPRF as important under hypoxia.

Only the hypoxia survival category had genes that appeared in all three types of analysis. Considering all three types of analyses, I propose that the top candidates for future validation attempts are AP1S1, DHTKD1 and COP1-interacting genes.

3.4 Validation of Results

Our validation strategy for the genome-wide screen was similar to the positive selection screen: we used lentiviral delivery of at least 2 shRNAs, distinct from those used in the screen by vector and sequence, to create stable cell lines with loss of gene expression of the candidate gene. Then I treated these cells, in parallel to a scramble control shRNA line, with the relevant stress treatment and measured cell survival after 4 days under stress. Previous to this stable-knockdown strategy, I tried a transient siRNA transfection strategy, but the results were not robust, and so the genes that failed in those experiments could be re-examined with stably-depleted cell lines.

Table 11 details validation attempts from the genome-wide screens. As LA seemed to cause cell cycle arrest rather than active cell death, a reproducible, but not dramatic, pattern of fewer surviving cells was common for the LA hits. Partly for this reason, the genes followed up on in greater detail were from the hypoxic stress that triggered active apoptosis (Chapter 4). It will be important to test the LA hits in other assays for stronger phenotypes. In this section, I briefly describe the top five successfully validated genes and their known importance in cancer biology and then outline potential hypotheses to test in future experiments.

Table 11. Summary of validation work from genome-wide pooled screen

Category of Hit Gene Symbol Current Status Phenotype ACACA Hypoxia-Survival Validated, mechanistic studies in Ch.3 ETV4 Hypoxia-Survival Validated, mechanistic studies in Ch.3 EPAS1 Hypoxia-Lethal Validated, positive control STK39 LA-Lethal Validated by counting MLF1 LA-Lethal Validated by counting SART1 LA-Lethal Validated by counting SENP1 Hypoxia-Lethal Have KD cell lines, very subtle phenotype ANTXR1 LA-lethal Have “KD” cell line, PCR primers not good CRMP1 LA-Survival Have virus to make KD cell lines Hypoxia, LA- EVI5 Have virus to make KD cell lines Lethal DYRK3 LA-Lethal In process of cloning into pLKO.1 Hypoxia, LA- FANCA In process of cloning into pLKO.1 Lethal TMEM158/RIS1 Hypoxia-Survival In process of cloning into pLKO.1 TFG LA-Survival In process of cloning into pLKO.1 PTK2 Hypoxia-Lethal Failed to generate KD cell lines COPS6 Hypoxia-Lethal Failed validation with transient transfection Hypoxia, LA- ZDHHC20 Failed validation with transient transfection Lethal MAP3K7 Hypoxia-Lethal Failed validation with transient transfection GRPEL2 Hypoxia-Lethal Failed validation with transient transfection SLC9A3 Hypoxia-Survival Failed validation with transient transfection RAPGEF1 Hypoxia-Lethal Failed validation with transient transfection NSD1 Hypoxia-Lethal Failed validation with transient transfection NMI Hypoxia-Lethal Failed validation with transient transfection HSP90AB1 LA-Lethal Failed validation UTP11L LA-Lethal Failed validation

3.4.1 ACACA

We defined high confidence hits based on their selection in both the multiple hairpin and RIGER analyses and showing mRNA changes under the relevant stress.

From these considerations, ACC1 (acetyl-CoA carboxylase alpha or ACACA) was a top candidate. ACC1 encodes the cytosolic isoform of acetyl-CoA carboxylase, which converts acetyl-CoA to malonyl-CoA in the rate-limiting step of de novo fatty acid synthesis. There was an enrichment of shRNAs targeting ACC1 in the hypoxic versus the control condition, suggesting that ACC1 knockdown allowed for improved survival under hypoxia. ACC1 had 4 hairpins enriched under hypoxia (Figure 6a), scored as the

13th best gene in the RIGER analysis using the second best shRNA metric (Figure 5d) and had down-regulated mRNA in the microarrays under hypoxia (Section 3.3.4).

Additionally, the down-regulation of ACC1 was previously shown to protect cancer cells from glucose deprivation and matrix detachment stresses (Jeon et al., 2012).

Together, these data prompted us to validate and investigate the role of ACC1 under hypoxia.

To validate the shRNA screen result, we silenced ACC1 expression through lentiviral infection of multiple shRNAs that targeted sequences distinct from those shRNAs used in the screen. We confirmed the successful reduction of ACC1 protein by these shRNAs (Figure 6b). In the control cells transduced with a scramble shRNA, hypoxia significantly decreased cell viability and induced apoptosis (Figure 6).

However, silencing ACC1 by multiple shRNAs inhibited the hypoxia-induced apoptosis as shown by crystal violet staining (Figure 6c), cell counting (Figure 6d), propidium- iodide staining (flow cytometry) (Figure 6e) and PARP cleavage (Figure 6f). This hypoxic protection associated with ACC1 silencing was also reproduced in additional hypoxia-sensitive cell lines, including MDA-MB-231 (breast cancer; Figure 12a, b), and

PANC-1 (pancreatic cancer; Figure 12e, g). Furthermore, chemical inhibition of ACC1 through the AMPK agonist metformin also protected H1975 cells from hypoxia-induced apoptosis (Figure 12h, i). Collectively, these data successfully validated the screen results and showed that the depletion of ACC1 enhanced cell survival under hypoxia in multiple cancer cells from different tissues of origin. Additional introduction to ACC1 and details of how ACC1 protects from hypoxia-induced apoptosis are presented in

Chapter 4.

Figure 6. Depletion of ACC1 protects cells from hypoxia-induced apoptosis.

(A) GENE-E heat map results of ACC1 hairpins in hypoxia screen (top 3 rows) and lactic acidosis screen (bottom 3 rows). Each row is one biological replicate (n=3 per treatment). Each column is a different shRNA targeting ACC1. Green box highlights the hypoxia result where the gene is predicted as a “protective” hit. (B) Western blot showing efficiency of 4 independent shRNAs targeting ACC1 in H1975 cells. (C) Crystal violet staining of scramble and ACC1 shRNA cells after 4 days of hypoxia. (D) Viable cell numbers of indicated cells after 4 days of hypoxia as determined by counting nuclei (n=9). (E) Percent of cells with less than 2N DNA content (sub-G1 in cell cycle analysis) as determined by PI staining and FACS analysis of indicated cell line after 4 days of hypoxia (n=9). (F) Western blot showing cleaved PARP levels with scramble and shACC1 cells after hypoxia for 48 hours. All data is from the H1975 cell line.

Figure 7. Depletion of ETV4 protects cells from hypoxia-induced apoptosis.

(A) Western blot showing decreased ETV4 protein levels in shRNA cell lines targeting ETV4. (B) Crystal violet assay of shETV4 cells after 4 days of indicated treatment. (C) Percent of cells from indicated cell line with less than 2N DNA content (sub-G1 in cell cycle analysis) after 4 days of normoxia or hypoxia treatment. (D) Western blot of the canonical apoptosis marker PARP in indicated cell lines after 48 hours of hypoxia. All data is from the H1975 cell line.

3.4.2 ETV4

In the multiple hairpin analysis of the hypoxia genome-wide screen, there was an enrichment of shRNAs targeting a PEA3 transcription factor family member, ETV4. This suggested that silencing ETV4 might be protective under hypoxia. Additionally, hypoxia down-regulated ETV4 in the microarray analysis. The other PEA3 family members,

ETV1 and ETV5, were not identified as “multiple hairpin hits” in the shRNA screen, nor were their mRNAs changed in the microarrays. We validated the hypoxia-protective phenotype of ETV4 loss with two different shRNAs targeting ETV4 (Figure 7a, b). The depletion of ETV4 maintained cell viability in a crystal violet assay (Figure 7b), reduced the number of apoptotic cells as measured by the percentage of cells in the sub-2N DNA content (termed “sub-G1” in figures) (Figure 7c) and decreased PARP cleavage (Figure

7d) under hypoxia. These data validated the phenotype for ETV4 silencing as predicted by the genome-wide screen and specifically showed that shETV4 decreased hypoxia- induced apoptosis. Chapter 4 includes an introduction to ETV4, its importance in cancer biology, and detailed experiments to explain how the loss of ETV4 protects from hypoxia-induced cell death.

3.4.3 STK39

STK39 (SPAK) was predicted as a two hairpin MHH in the LA synthetic lethal category of the genome-wide LA screen. This phenotype was validated with multiple shRNAs that reduced the expression of STK39 in both low and high serum conditions

(Figure 8a, b, c). The high growth condition may be more relevant for STK39 experiments as this is where increased cell death was observed, rather than an arrest in

G1 (Figure 8d, e). STK39 also showed a synthetic lethal phenotype under LA in MDA-

MB-231 breast cancer cells in a siRNA screen that was performed by Jianli Wu in the Chi

Lab. STK39 is a serine/threonine kinase critical for maintaining osmotic pressure in cells, especially in the context of kidney physiology. It has a highly redundant functional protein, OSR1 (Gagnon and Delpire, 2012), which was not a hit in the pooled genome- wide screen. Both of these kinases phosphorylate to activate three Na+/Cl-/K+-

Transporters (NKKCs) to promote their function and maintain osmotic pressure in response to upstream stresses. The NKKC transporters are targets of the widely prescribed blood pressure medications, loop and thiazide diuretics (Gagnon and

Delpire, 2012). Therefore, if this phenotype were validated in vivo, this would suggest the potential use of thiazide and loop diuretics as cancer therapeutics.

STK39 is not strongly connected to cancer biology. It is not altered in many cancer patient samples by TCGA data. In the literature, its expression varies depending on cell types: it was robustly expressed in prostate normal and cancer tissues (Qi et al,

2001) and gliomas (Haas et al, 2011), but was down-regulated in B cell lymphomas

(Balatoni et al., 2009). A SNP in STK39 was associated with overall survival in NSCLC

(Huang et al., 2009) and its reduced mRNA expression correlated with increased incidence of metastasis in prostate cancer (Hendriksen et al., 2006). In cervical cancer,

100

STK39 mediated the aggressive phenotype of KCC3 overexpressing cells, so that its loss decreased these cells’ invasive ability and reduced xenograft tumor growth (Chiu et al.,

2014). Therefore, much remains to understand about the importance of this gene under stress in cancer biology.

I did not complete a detailed investigation into the role of STK39 under LA.

Instead, here I propose a potential hypothesis to explain the validated phenotype, a justification for this hypothesis and suggest some initial experiments to begin to better understand the functional role of STK39 in cancer cells under LA.

Hypothesis: STK39 is essential for cancer cell viability under LA due to its importance in maintaining ionic balance across the plasma membrane.

Justification: STK39 plays a critical role in osmotic pressure maintenance and activates the NKKC transporters in response to multiple stress signals (Gagnon and

Delpire, 2012). A build-up of extracellular lactic acidosis is intricately involved in cellular water, ion and pH balance; maintenance of these balances is critical to cell survival. It is logical for STK39 to play an important role in ion homeostasis in cancer cells under LA and for the maintenance of this balance to be critical for cell survival.

Approach: If STK39 regulates ionic and osmotic balance to maintain cancer cell survival under LA, then the loss of the NKKC transporters should phenocopy the synthetic lethal effect of STK39. Depletion of multiple NKKC transporters may be necessary to see a phenotype. Using qPCR or western blots to determine the relative

101

expression of the different NKKC transporters can help suggest if any particular transporter is more important. Especially interesting will be to determine if pharmacological inhibition of the relevant NKKCs can affect cancer cell survival under

LA. Loop diuretics have different NKKC target specificity and so the expression of the transporters may help select the most relevant drugs to test for affecting survival of cancer cells under LA. If the NKKCs are important for cell survival under LA, then experiments should measure intracellular and extracellular pH and ion gradients with or without STK39 or NKCC transporter function. Intracellular pH can be detected in multiple ways including pH-sensitive probes or fluorescent proteins, NMR or microelectrodes (Loiselle and Casey, 2010). A recent study measured pH differences to

0.038 in single cancer cells with nanoprobes (Yang et al., 2015). Multiple kits are commercially available to measure intracellular pH or intracellular sodium and potassium through fluorescent probes (ThermoScientific, Sigma, AAT Bioquest). Since

OSR1 and STK39 can be functionally redundant (Gagnon and Delpire, 2012), it may be necessary to deplete both simultaneously or inhibit multiple of their downstream targets to see a robust phenotype. Completing these experiments will determine if a dysregulated balance of cellular pH or ions results with loss of STK39 under LA.

The most high impact part of STK39 will be to determine if the loop or thiazide diuretics can induce a synthetic lethal phenotype under LA in cultured cells, and if this can be extended to decreased lung cancer tumor growth in xenograft or genetic mouse

102

models in vivo, as has been seen in GBM models (Ru et al., 2014). The dose and potential combination of drugs will need to be carefully chosen, as these drugs affect normal cells and had toxic effects at high doses in a neuronal model of oxygen-glucose deprivation stress (Pond et al., 2004). Furthermore, retrospective epidemiological studies of the incidence of cancers, metastases or disease progression in patients on diuretics could be investigated. The role of STK39 in cancer biology has significant translational potential if diuretics can decrease the negative features associated with cancer cells under LA.

103

Figure 8. Validation of STK39 depletion as synthetic sick under LA.

(A) qPCR analysis of relative STK39 mRNA abundance in 3 shRNA cell lines. (B) Crystal violet assay of shSTK39 cells after 4 days of indicated treatment (0.5% FBS). (C) Crystal violet assay of shSTK39 cells after 4 days of indicated treatment (10% FBS). (D) Percent of cells from indicated shRNA cell line in the G1 phase of the cell cycle after 4 days of normoxia or hypoxia treatment (0.5% FBS). (E) Percent of cells from indicated shRNA cell line with less than 2N DNA content (sub-G1 in cell cycle analysis) after 4 days of normoxia or hypoxia treatment (10% FBS).

104

3.4.4 MLF1

MLF1 was predicted as a synthetic lethal gene under LA by the multiple hairpin and microarray analysis. Knockdown of MLF1 decreased the number of viable cells under LA by ~30% compared to the control, however the level of LA sensitivity did not correspond with percent of mRNA depletion (Figure 9a, b). MLF1 is particularly interesting since it is highly amplified in cancer patient samples based on TCGA data, with the highest levels of amplification found in squamous cell lung carcinomas, reaching 22-32% of patient samples (Figure 9c). MLF1 is located at 3q25. Amplifications of the 3q(25) arm are common in lung, head and neck squamous, prostate and ovarian cancers (Pei et al., 2001; Redon et al., 2002; Sticht et al., 2005; Jung et al., 2006). This amplification has been proposed as a defining and potentially targetable feature of lung squamous carcinoma (Brunelli et al, 2012). Amplifications in this region have also been proposed to drive expression of TLOC (Jung et al., 2006) or Cyclin L (Redon et al., 2002;

Sticht et al., 2005). Since loss of MLF1 was detrimental to cell survival under LA, amplification of this region may also provide a selective advantage for cancer cells under LA through increased expression of MLF1.

MLF1 is a transcription factor known to promote myeloid cell differentiation through the erythroid lineage. Fusions with MLF1 and NPM are common occurrences and drivers in adult acute myeloid leukemia (Falini et al., 2006 and 2007). MLF1 can

105

inhibit cell cycle progression by preventing the accumulation of p27KIP1 (Winteringham et al., 2004) and through p53 (Yoneda-Kato et al., 2005). It also alters transcriptional programs in development and cancer by preventing the degradation of RUNX family transcription factors (Bras et al., 2012). Thus, while the importance of MLF1 to cancer biology is recognized, no studies have investigated its role under LA.

As with STK39, I did not complete experiments on MLF1 beyond the initial validation of the screen-predicted phenotype (Figure 9). Further investigation into the hypothesis I propose below will uncover additional details about the role of MLF1 under lactic acidosis.

Hypothesis: MLF1 loss is detrimental to cell survival by relieving the negative regulation of COP1 and permitting a COP1-mediated decrease in cell survival under

LA.

Justification: Through its direct interaction with the COP9 signalosome component CSN3, MLF1 prevents the E3 ubiquitin ligase COP1, from degrading its target p53 (Yoneda-Kato et al., 2005). Therefore, loss of MLF1 should allow COP1 to target p53, and other targets, for degradation. Our analysis that identified multiple

COP1-interacting proteins as MHHs implicated COP1 in stress survival. Since decreased p53 is most often associated with apoptosis-inhibition (Vousden and Lu, 2002), and loss of ETV4 prevented apoptosis under hypoxia (Chapter 4, Keenan et al., 2015), these are not likely the targets of interest here. However, COP1 can have both tumor suppressive

106

and oncogenic roles (Marine, 2012), so its role under LA may be context-dependent and is so far unknown. More global approaches may help to investigate the mediators

MLF1’s phenotype under LA.

Approach: To begin investigating this hypothesis, COP1 protein levels should be investigated with and without MLF1 depletion under basal and LA conditions. If COP1 activity acts downstream of MLF1 here, double silencing of MLF1 and COP1 should rescue the LA-lethal phenotype. If COP1 loss rescues the effect of MLF1 loss, then a more comprehensive profiling approach could be used to determine the potential target of COP1 that mediates this effect. I suggest to two parallel experiments. The first is a microarray or RNA-seq analysis with MLF1 loss under LA to determine the gene expression changes that could be mediating the phenotype. Second, I suggest immunoprecipitation of COP1 follow by mass spectrometry of its interacting partners under LA. The overlapping proteins identified in these two experiments would likely be important for the effect of MLF1 on cancer cell survival under LA.

If COP1 loss does not rescue the MLF1 phenotype, then the microarray or RNA- seq experiment could still identify potential downstream mediators of MLF1 under LA.

An understanding of how MLF1 affects cell survival under LA would help explain this common genetic aberration in cancer and would warrant further investigation of this mechanism in other cancer types where MLF1 alteration is a driving genetic feature.

107

Figure 9. Validation of MLF1 depletion as synthetic sick under LA.

(A) qPCR analysis of relative MLF1 mRNA abundance in 3 shRNA cell lines. (B) Viable cell number as determined by trypan blue exclusion and cell counting of indicated shRNA cell lines after 4 days of control or LA treatment (n=9). (C) Alteration frequency of MLF1 in TCGA data as represented in cBIOPortal. Legend indicates type of alteration (red=amplification, blue=deletion, green=mutation, gray=multiple alterations). Each bar is a different study. Colored circles under each bar indicate type of cancer studied.

108

3.4.5 SART1

SART1 was predicted as a synthetic lethal MHH by 3 hairpins under lactic acidosis. I validated this phenotype with three stable shRNA cell lines targeting SART1

(Figure 10a, b). This phenotype was reproducible and statistically significant, but modest. SART1, or Hypoxia-Associated Factor, is essential for spliceosome assembly

(Makarova et al., 2001) and regulates a shift from a HIF-1α to a HIF-2α transcriptional program (Koh et al., 2008 and 2011). In the cBioPortal data, SART1 is amplified in 5-10% in HNSCC and pancreatic cancers, deleted in ~6% of malignant peripheral nerve sheath tumor samples and then altered in less than 4% of other cancer types. The connection of

SART1 to the HIFs may be most relevant for its effect under acidosis. Multiple studies have shown the inhibition of HIF-1α under acidosis (Tang et al., 2012; Parks et al., 2013), but none have looked at the relative contributions of HIF-1α or HIF-2α under LA.

Hypothesis: Silencing SART1 prevents a shift from HIF-1α to HIF-2α, which promotes an unfavorable metabolic program through a HIF-1α-mediated transcriptional response.

Justification: SART1 promotes the degradation of HIF-1α and the transition to a

HIF-2α transcriptional program (Koh et al., 2008 and 2011). Thus, its depletion should prevent this shift. While both HIFs contribute to hypoxic cell survival, they have unique targets and roles in cancer progression (Keith et al., 2012). In particular, HIF-1α has more metabolic targets, promoting glycolysis by multiple mechanisms (Keith et al., 2012).

109

Figure 10. Validation of SART1 depletion as synthetic sick under LA.

(A) qPCR data showing relative mRNA abundance of SART1 shRNA cell lines in either of two plasmids (pLKO.1 or GIPZ). (B) Viable cell number as determined by trypan blue exclusion and cell counting of indicated shRNA cell line after 4 days of control or lactic acidosis treatment (n=9).

110

On the other hand, LA inhibits glycolytic gene expression (Chen et al., 2008 and

2010). Possibly, the promotion of glycolysis with maintained HIF-1α expression under

LA is detrimental to these cells’ survival. No one has examined the importance of the balance between HIF-1α and HIF-2α expression under a low pH or lactic acidosis condition.

Approach: First, examine the protein levels of HIF-1α and HIF-2α under LA with or without SART1 depletion to determine if SART1 regulates the HIFs in our experimental setup. If a connection to the HIFs is established, experiments that deplete and overexpress the HIFs should determine if cell survival under LA is affected by the balance between these two proteins. Additionally, reporter assays and ChIP-Seq could be used to test the activity and promoter occupancy, respectively, of the HIFs with or without LA and SART1 expression. To test the specific downstream characteristics of the

HIFs, I suggest first investigating metabolic characteristics of cells with SART1 depletion, HIF-1α overexpression (using a non-hydroxylatable version) and HIF-2α loss, as these are predicted to be similar in this model. The Seahorse Machine (Seahorse

Bioscience) would be useful to measure glucose uptake, oxygen consumption, and further acidification of the media under LA conditions. According to this hypothesis,

HIF-2α-only expressing cells under LA would have “opposite” phenotypes in these assays. If these experiments show the predicted trends, HIF-2α overexpression should be able to rescue the survival of SART1 depleted cells under LA. It might be interesting

111

to test this hypothesis in ccRCC where there is often constitutive HIF expression and

HIF-2α is on oncogene (Kondo et al., 2003). Does the loss of SART1 influence HIF-2α’s ability to drive tumor progression in this context? Understanding the importance of

SART1 under LA could offer an upstream way to target the HIF-1/2 balance in tumor progression.

3.5 Future considerations

This Section described the analysis and validation attempts of the genome-wide pooled shRNA screens under LA and hypoxia in H1975 lung cancer cells. Clearly, this work established many potential directions of future investigation. My analysis highlights a few top candidates that have been phenotypically validated or are top predicted candidates across multiple lines of analysis: AP1S1, DHTKD1, COP1- interacting proteins, MLF1, STK39 and SART1. Phenotypes validated in H1975 cells should be extended to other cell lines and tested in vivo. Genes with many different functions were identified as hits across these contextual stress screens. This demonstrates the utility of performing a genome-wide screen and emphasizes the multiple modes of adaptation that a cancer cell can employ to survive TME stresses. It is my hope that the analysis and detailed future experiments outlined here encourage future research from this screen project. The next Chapter will described the detailed

112

investigation I conducted about how the loss of ACC1 or ETV4 protected cells from hypoxia-induced apoptosis.

113

4. The Role of ACC1 in Hypoxic Cancer Cell Survival

This Chapter contains data and text published in the peer-reviewed manuscript:

“ACLY and ACC1 Regulate Hypoxia-Induced Apoptosis by Modulating ETV4 via α- ketoglutarate.” Keenan MM, Liu B, Tang X, Wu J, Cyr D, Stevens RD, Ilkayeva O, Huang

Z, Tollini LA, Murphy SK, Lucas J, Muoio DM, Kim SY, Chi JT. PLoS Genet. 2015 Oct

9;11(10):e1005599. doi: 10.1371/journal.pgen.1005599. eCollection 2015 Oct.”

4.1 Introduction

Due to the nature of RNAi screens, the particular hit that would be the focus of detailed investigation was not known until validation experiments were completed.

Therefore, I first include the background specific to this Chapter. The initial validation data for ACC1 and ETV4 from the genome-wide screen are presented in Chapter 3

(Figure 6 and Figure 7). After the introduction, the methods are described in detail and then I discuss our continued investigation in to how the loss of ACC1 protects cancer cells from hypoxia-induced apoptosis.

4.1.1 Lipogenic enzymes in cancer cell metabolism

In general, cancer cells up-regulate their de novo lipogenic programs to supply the necessary lipids for membrane expansion and signaling molecules (Currie et al.,

2013). This cytosolic pathway is composed of multiple enzymes that synthesize long chain fatty acids from short carbon metabolites (acetyl-CoA and other acyl-CoAs), while

114

consuming significant amounts of NADPH and ATP. The regulation of this pathway is complex and finely tunable to various oncogenic alterations and TME conditions and has been considered a target for cancer therapeutics (Currie et al., 2013; Tong and

Harwood, 2006; Zu et al., 2013).

The rate-limiting enzyme of de novo lipogenesis is acetyl-CoA carboxylase, which has two isoforms, ACC1 (ACACA) and ACC2 (ACACB). The ACC’s convert acetyl-CoA to malonyl-CoA in multi-step enzymatic reactions, performing first the biotin carboxylase reaction, followed by the carboxyltransferase reaction (Tong and

Harwood, 2006). This enzyme commits these metabolites to fatty acid synthesis (Barber et al., 2005). ACC1 and ACC2 differ somewhat in their size (265 kDa and 280 kDa), tissue expression and subcellular localization. ACC1 is expressed ubiquitously and more in lipogenic tissues such as liver and adipose, while ACC2 is expressed more in oxidative tissues such as the heart and skeletal muscle (Barber et al., 2005). The enzymes’ differential localizations may be related to the distinct roles of their metabolite pools:

ACC1 is a cytosolic enzyme, mostly involved in de novo lipogenesis, while ACC2 is localized to the outer mitochondrial membrane and therefore potentially more involved with the regulation of β-oxidation in the mitochondria (Abu-Elheiga et al., 2000 and

2003). Transcriptionally, ACC is induced by lipogenic transcription factors such as the

SREBPs (Horton et al., 2002). The activity of both ACCs is sensitive to post-translational modifications and allosteric inhibition (Barber et al., 2005). The most recognized

115

posttranslational regulation of ACC is its inactivating phosphorylation by AMPK or

PKA (Sim and Hardie, 1988; Haystead et al., 1990). Citrate, isocitrate and, to a lesser extent, glutamate can all activate the ACCs in vitro (Barber et al., 2005). ACC is competitively inhibited by malonyl-CoA and long chain acyl-CoAs to form negative feedback loops to reduce lipogenesis (Barber et al., 2005). ACC1 is overexpressed in many cancers, supporting the importance of increased de novo lipogenesis for proliferation (Currie et al., 2013). Small molecule inhibitors targeting ACC1/2 have been developed to treat metabolic diseases, but their therapeutic potential for cancer remains to be determined (Tong and Harwood, 2006).

ATP-citrate lyase (ACLY) converts citrate and ATP to acetyl-CoA, and therefore acts immediately upstream of ACC. Although this enzymatic reaction is critical for lipogenesis, this step does not commit metabolites to lipogenesis. Instead, the acetyl-

CoA generated by ACLY is more widely used in other aspects of cell metabolism and protein modifications (Zaidi et al., 2012). ACLY is found in both the nucleus and the cytosol (Wellen et al., 2009; Chypre et al., 2012), although the nuclear ACLY maybe more important in affecting histone acetylation patterns through its product acetyl-CoA

(Wellen et al., 2009). ACLY is post-translationally activated through phosphorylation by

Akt (Migita et al., 2008) and stabilized through acetylation (Lin et al., 2013). It is over expressed in many cancer types (Zaidi et al., 2012), its activity or expression correlated with poor prognosis in lung cancers (Migita et al., 2008), increased stages of gastric

116

carcinomas (Qian et al., 2015) and local tumor stage in NSCLC (Csanadi et al., 2015).

Furthermore, SNPs in ACLY were associated with survival in colon cancer (Xie et al.,

2015) and NSCLC (Jin et al., 2014). Since ACLY can control both gene expression and lipogenesis, targeting it therapeutically could have multiple implications for cancer biology. While naturally occurring compounds or citrate mimics have been tested as inhibitors of ACLY, none of these have produced clinically relevant products yet (Zaidi et al., 2012).

Fatty acid synthase, FASN or FAS, is the enzyme that condenses malonyl-CoA and acetyl-CoA, with NADPH as a reducing equivalent, to form palmitate. FASN has been extensively targeted in cancer biology since it is the most commonly overexpressed enzyme in de novo lipogenesis in cancer cells, its overexpression correlates with increased death and poor prognosis, and its inhibition is toxic in multiple cancer cell types (Menendez and Lupu, 2007). FASN is overexpressed in cancers due to increased transcription by SREBP1c responding to growth factor signaling through the ERK and

Akt pathways and due to increased protein stability (Menendez and Lupu, 2007). Many drugs have been developed to target FASN, but their toxicities to normal cells make their application challenging (Menendez and Lupu, 2007).

SCD, stearoyl-CoA desaturase, desaturates lipids in an oxygen-dependent reaction that produces monounsaturated lipids, including the most abundant of these, oleate (Ntambi et al., 2004). While less studied in cancer biology, this enzyme is also up-

117

regulated in response to a number of growth factors and in colon, esophageal, and hepatocellular carcinomas (Igal, 2010). In renal cell carcinoma, SCD and HIF-2α promote each other’s expression (Zhang et al., 2013a) and thus SCD represents a particularly interesting clinical target in this disease (von Roemeling et al, 2013; Leung and Kim,

2013; Fenner, 2013). Inhibitors are being developed and have anti-proliferative effects in vitro, but still remain outside of clinical effectiveness (Uto, 2015).

118

Figure 11. Model of enzymes and metabolites important for de novo lipogenesis.

Simplified schematic of the enzymes and metabolites important for de novo lipogenesis. Many levels of interaction and regulation are omitted for clarity. Purple rectangles indicate enzymes. Green boxes indicate metabolites.

119

4.1.2 Lipogenesis under stress

Since lipogenesis is an energy-consuming pathway, cells have multiple mechanisms to regulate its activity. For example, under stresses and nutrient deprivation, lipogenesis is often down-regulated to conserve energy and reduce anabolism in favor of stress-survival. Specifically, both glucose deprivation and hypoxia reduce levels of ATP and activate AMPK to phosphorylate and inactivate ACC

(Ruderman and Prentki, 2004), to inhibit lipogenesis. However, recently, the reductive carboxylation of glutamine to generate citrate and fuel lipogenesis was shown to occur under hypoxia (Metallo et al., 2012; Wise et al., 2011). Increased lipogenesis protected cells under oxidative stress by promoting saturation of lipids (Rysman et al., 2010), but its inhibition was protective under matrix detachment or short-term glucose deprivation to preserve levels of NADPH for redox homeostasis (Jeon et al., 2012). Interestingly, lipogenesis also affects cellular responses to drug treatments, as removal of anti- angiogenesis therapy enhanced lipogenesis and promoted tumorigenesis, but the subsequent inhibition of lipogenesis re-instated anti-angiogenic drug effects (Sounni et al., 2014). In response to an altered metabolism that inhibits lipogenesis under hypoxia, cancer cells driven by RAS transformation take up lipids from the extracellular environment (Kamphorst et al., 2013). Although this process of lipid scavenging is independent from de novo lipogenesis, it emphasizes the importance of lipid metabolism to cancer cells under stress. Overall, depending on the particular stresses

120

and oncogenic genetic events, the activation of lipogenesis under stress can be beneficial or harmful to stress survival.

4.1.3 ETV4: A PEA3 family member

4.1.3.1 An overview of the PEA3 Family

The ETS family of transcription factors includes 28 different members that are highly conserved in mammals, but are missing from plants, fungi and bacteria. The family is divided into 12 subgroups and 9 subfamilies with 2-3 members each. This family is defined by the presence of the ~85 amino acid long ETS DNA binding motif

(Oh et al., 2012; Cooper et al., 2015). Based on structural information, DNA binding assays in vitro and chromatin immunoprecipitation experiments on both human and mouse ETS proteins, the greater ETS family is divided in to four different DNA-binding

“Classes”, each with slightly different binding specificities outside of 4 base-pair core residues (Wei et al., 2010).

Within the larger ETS family, “Class I” includes the PEA3 subfamily that has three members: ETV1 (ER81), ETV4 (PEA3/E1AF), and ETV5 (ERM) (Oh et al, 2012; Wei et al., 2010). The general consensus site for the PEA3 members has been determined (5’-

ACCGGAAGT-3’) (Wei et al., 2010). These family members have both overlapping and unique functions in physiological and pathological settings. Generally, there is greater overlap in functional redundancy between ETV4 and ETV5 than ETV1, and this is mirrored in the phenotypes of the knockout mice for each of these proteins (Oh et al.,

121

2012; Arber et al., 2000; Laing et al., 2000). Since the DNA binding motifs and domain structures are quite similar, only with further understanding of expression differences, interacting co-factors and upstream regulation will we understand how the specificity and selectivity of this family is determined. This introduction is centered on ETV4, as it is the focus of data in this Chapter, but includes information about ETV1 and ETV5 as well.

4.1.3.2 Regulation of ETV4 activity

Multiple signal transduction pathways regulate the expression and activity of the

PEA3 family members in response to extracellular stimuli. The best recognized upstream signals are the ERK and JNK pathways, two MAPK pathways (O’Hagan et al.,

1996). HER2 also activates the ERK pathway and has been shown to stimulate PEA3 expression and nuclear localization (Menendez et al., 2004). Various growth factors

(FGF, GDNF, HGF, EGF) can stimulate PEA3 transcription as well (Hanzawa et al., 2000;

Hiroumi et al., 2001; Shindoh et al., 2004; Hakuma et al., 2005; Kherrouche et al., 2015).

In addition to signaling pathways, posttranslational modifications alter the function and activity of the PEA3 family. The phosphorylation of ETV1 or ETV5 can inhibit their DNA binding. Interestingly, these inhibitory phosphorylated residues on

ETV1/5 are not found in ETV4 (Baert et al., 2002). As mentioned, phosphorylation downstream of MAPK pathways stimulates its transcriptional activity (O’Hagan et al.,

1996; Menendez et al., 2004). p300-dependent lysine acetylation enhanced transcriptional

122

activation of ETV4, likely by blocking ubiquitination at these same lysine residues and subsequent degradation (Guo et al., 2011a). Studies from one group showed that both acetylation and sumoylation events stimulate ETV4 activity (Guo et al., 2009 and 2011a).

However, a different study showed that sumoylation of ETV4 inhibited its transcriptional activation (Bojovic and Hassell, 2008). COP1 (RFWD2) targets the PEA3 family for ubiquitination and degradation (Baert et al., 2010; Vitari et al., 2011). The details of other specific proteins necessary for the writing or removing of these PTMs remains unknown. Thus, a complex set of interacting posttranslational modifications play a significant role in regulating ETV4’s levels and transcriptional capabilities.

Multiple co-factors interact with the PEA3 family members and contribute to regulating and defining their specificity. Both ETV5 and ETV4 interact with LPP

(lipoma-preferred partner) to stimulate their transcriptional responses (Colas et al., 2012;

Guo et al., 2006). USF-1 binds and relieves the auto-inhibitory function of an internal

ETV4 domain to promote DNA binding activity (Greenall et al., 2001). ETV4 also directly interacts with CBP/p300 as a co-activator to influence HIF-1α transcription under hypoxia (Liu et al., 2004; Wollenick et al., 2012). Direct physical interaction of Sp1 and ETV4 in glioma cells, independent of DNA binding, led to increased reporter gene activity for the Sp1 target gene GalT V (Jiang et al., 2007). It is likely that additional undetermined co-factors contribute to the specificity and modulate the transcriptional function of each PEA3 family member.

123

4.1.3.3 Role of ETV4 in normal physiology and cancer biology

The PEA3 family members are expressed throughout development and function in a variety of normal physiological processes. All three members influence sonic hedgehog (Shh) expression to control patterning in limb bud development and outgrowth (Mao et al., 2009; Zhang et al., 2009; Zhang et al., 2010; Lettice et al., 2012). All three PEA3 family members are also expressed during branching morphogenesis for the development of mammary epithelial cells (Oh, et al., 2012), the kidneys (Lu et al., 2009;

Kuure at al., 2010) or the lungs (Liu et al., 2008 and 2003). While ETV4 knockout mice display some features of motor and muscle disability, other than being infertile, they lack obvious phenotypes (Laing et al., 2000; Livet et al., 2002; Ladle and Frank, 2002).

Collectively, these studies show that the PEA3 transcription factors play important roles in various tissues throughout mammalian development.

Generally, the PEA3 family members are overexpressed in multiple cancers and are considered to promote oncogenic phenotypes (Oh et al., 2012). Hyper-activated upstream signaling pathways and translocations to form active fusion proteins are two major mechanisms of oncogenic up-regulation for these proteins (Tomlins et al., 2006;

Helgeson et al., 2008). Interestingly, these fusions proteins are more resistant to proteasomal degradation (Vitari et al., 2008) and have altered DNA binding patterns

(Wei et al., 2010) that likely contribute to their pro-tumorigenic phenotype.

124

The major biological consequences of PEA3 overexpression are the promotion of invasive and migratory phenotypes and the development of metastases, which have been documented in prostate cancer (Aytes et al., 2013; Mesquita et al., 2015), lung cancer (Hiroumi et al., 2001; Hakuma et al., 2005; Kherrouche et al., 2015) breast cancer

(Ladam et al., 2013), oral squamous cancer (Hanzawa et al., 2000), and gliomas (Jiang et al., 2007). Two drugs, phenamil and YK-4-279, can modulate the activity of this family, but neither has been used in the clinic (Park et al., 2010; Erkizan et al., 2009; Rahim et al.,

2011 and 2014; Lamhamedi-Cherradi et al., 2015). A better understanding of the regulatory mechanisms up and downstream to PEA3 overexpression will improve our likelihood of developing ways to target the over-activation of this family in cancer patients.

4.2 Methods

In this section, I detail the methods used throughout this Chapter to obtain the presented data. Some of these methods were also described in Chapters 2 and 3. The details of the genome-wide screens are provided in Chapter 3.

4.2.1 Cell culture, TME stress treatments and generation of stable shRNA cell lines

Cell culture was the same as described on page 62. Additionally, MDA-MB-231 and PANC-1 cells were cultured in DMEM (GIBCO cat.no. 11995) supplemented with

10% Fetal bovine serum (heat-inactivated) and 1x antibiotics (penicillin, 10,000 UI/ml; 125

streptomycin, 10,000 UI/ml). Cell lines, obtained from and initially validated by the

Duke Cell Culture Facility (Durham, NC, USA), were maintained for fewer than 6 months and validated by microscopy every 1 to 2 days.

Lactic acidosis was generated via addition of lactic acid (Sigma-Aldrich, St.

Louis, MO, USA, cat. no L6402) and media pH adjustment to pH 6.7 by HCl immediately before use. Hypoxia was generated with a cell culture incubator with 93-

94% N2, 5% CO and 1-2% O2. For the α-KG rescue experiments, media was supplemented with 0.875-4mM dimethyl α-KG as indicated in figure legends (Sigma, cat. no. 349631). For all stress experiments, cells were serum starved (0.5% FBS) for 24 hours before treated with stress under 0.5% FBS. All survival/viability measurements were made after 4 days of stress treatment.

Stable cell lines were created with the pLKO.1 shRNA constructs purchased from the Duke RNAi Core Facility. Virus was generated by transfecting HEK-293T cells with a 1: 0.1: 1 ratio of pMDG2: pVSVG: pLKO.1 with Lipofectamine 2000 in the evening.

Media was changed the following morning and virus collected 48 hours after transfection. Stable cell lines were generated by adding 200ul virus to a 60mm dish of parental cells with polybrene (final concentration 8ug/ml). Complete death in blank infection dishes was used to determine success of infection and puromycin selection.

Efficiency of silencing or overexpression was determined by western blots.

Concentrations of puromycin needed for selection: H1975 cells= 1ug/ml, MDA-MB-231

126

cells = 1ug/ml, PANC-1 cells= 2ug/ml. For stable overexpression of ETV4, concentration of blasticidin used was 2.5ug/ml in H1975 cells.

4.2.2 Crystal violet staining

Cells were fixed either in 4% paraformaldehyde (PFA) overnight at 4°C or at room temperature for 30 min. PFA was removed and crystal violet staining solution

4.2.3 Determination of cell number

Cell number was evaluated by either direct cell counting (with trypan blue exclusion of dead cells) or high-throughput microscopic counting (HTC) of fixed and stained nuclei. For direct cell counting, at designated time after treatment, media was removed, cells were not rinsed for fear of losing loosely-attached cells, trypsinized, diluted 1:1 with trypan blue and immediate counted on a hemocytometer. For HTC experiments, after designated time period, cells were fixed in 4% PFA either overnight at

4°C or for 30 min at RT. Cells were washed 2x, permeabilized with 0.1% Triton-X in PBS, wash 2x, stained with 50ug/ml Hoescht 33342 (Sigma cat. no B2261) for 30 min, at RT in the dark, then washed 2x and PBST added to each well and scanned by the Cellomics

127

high-throughput microscope at the Duke RNAi Core Facility. Both lower and upper limits on the size of nuclei counted were established to avoid counting pyknotic, fragmented or enlarged nuclei, as is standard procedure suggested by the Cellomics software.

4.2.4 Flow cytometry

For cell cycle analysis, after 4 days of stress treatment, media was collected, cells were trypsinized and pooled with the media. Cells were centrifuged (5 min, 1000rpm,

(spins of 5 min, 1000rpm, RT) then resuspended in 25ug/ml Propidium iodide (Sigma cat. no P4864) and 10ug/ml RNAse A in PBS. Cells were stained for 30+ min in the dark then 8000 events were captured on a Canto II Flow cytometer.

4.2.5 Protein lysate collection and Western blots

Cell lysis: Cells were washed once with ice cold PBS, lysed by RIPA buffer with protease and phosphatase inhibitors added fresh, scraped into a microcentrifuge tube, allowed to swell on ice for 15-20 min, vortexed briefly, then spun down at top speed for

15 min at 4°C. Supernatant was transferred to pre-cooled new tube and protein concentration assayed with the Pierce BCA kit (ThermoScientific, cat. no. 23225). Effort was made to immediately rinse and lyse cells coming from a hypoxia condition as cells

128

very quickly re-equilibrate to normoxic conditions. Western blots: Between 15-30ug of lysate was loaded on SDS-PAGE gels, wet-transferred to PDVF membrane, blocked with

5% milk in 1xTBST (0.1% Tween-20), then primary antibodies were incubated overnight at 4°C. For analysis of histones, protein was extracted with the EpiQuick Total Histone

Extraction Kit (Epigentek, cat.no. OP-0006) and 2ug of protein were resolved on 15%

SDS-PAGE gels or nuclear fractions were collected by the REAP fractionation method

(Suzuki et al., 2010) and 7.5-30 ul of lysate were run in each lane. Details on antibody usage are as standard in the Chi lab (refer to Chi lab antibody Google sheet).

4.2.6 Quantitative real-time PCR

RNA was extracted using the RNeasy Kit (QIAGEN). 1 µg of total RNA was reverse transcribed by SuperScript II (Invitrogen) for real-time PCR with Power

SYBRGreen Mix (Applied Biosystems/Life Technologies (Grand Island, NY, USA)).

Primers were designed across exons whenever possible and were verified for specificity by regular PCR prior to use in real-time PCR. Please refer to “Designing real time primers” spread sheet for the sequences of primers used.

4.2.7 Microarrays and analysis

Samples were collected on ice and RNA was isolated with QIAGEN’s RNeasy

Mini Kit (cat. no 74104) according to manufacturer’s instructions. After quality control assessment with the Agilent BioAnalyzer, cDNA was amplified from 200ng RNA with the Ambion® MessageAmp™ Premier RNA Amplification (Life Technologies, Grand

129

Island NY, USA). The gene expression pattern of the RNA samples were interrogated with Affymetrix U133A gene chips and normalized by the RMA (Robust Multi-Array) algorithm. cDNA synthesis and microarray interrogation was performed by the Duke

Microarray Core. The influence of the silencing of ACC1 or ACLY on gene expression was derived by a zero transformation process, in which we compared the transcript level for each gene in cells with stably integrated shRNAs targeting ACC1 or ACLY to the average transcript levels in control scramble shRNA cell line samples. Data was filtered as described with Cluster 3.0 software and heat maps were generated with (Java)

TreeView. To generate gene signatures of knockdown of ACLY, ACC1 or ETV4, the

CreateSignature module in GenePattern

(https://genepattern.uth.tmc.edu/gp/pages/login.jsf) was used with scramble cells expression pattern as the train0 set, and the knockdown cells’ gene expression pattern as train1 set and the Gray dataset (Chin et al., 2006) used as the test set. Default parameters were used for the analysis similar to (Gatza et al., 2011), including using 100 genes in each signature. The resulting probabilities of gene signature expression in each patient for each knockdown signature were analyzed by simple linear regression in the JMP Pro

11 software.

4.2.8 Statistical Analysis

Data, unless otherwise noted, represent the mean +/- the standard error of the mean and n indicates number of replicates used to generate the SEM. P-values were

130

determined either by a two-tailed Student’s t-test in Excel or by a two-way ANOVA with StatView.

4.2.9 Metabolomics profiling and analysis

After the collection of lysates, the metabolomics profiling and running of mass spectrometer was conducted by the mass spectrometry group within the Stedman center at the Duke Molecular Physiology Institute. The measurement of amino acids and acyl carnitines was performed using stable isotope dilution techniques and flow injection tandem mass spectrometry and sample preparation methods described previously (An et al., 2004; Ferrara et al., 2008; LaMonte et al., 2013). Derivatized organic acids were analyzed by capillary gas chromatography/mass spectrometry (GC/MS) using a TRACE

DSQ instrument (Thermo Electron Corporation; Austin, TX) (An et al., 2004; Ferrara et al., 2008; LaMonte et al., 2013). All MS analyses employed stable-isotope-dilution. The standards serve both to help identify each of the analyte peaks and provide the reference for quantifying their levels. Quantification was facilitated by addition of mixtures of known quantities of stable-isotope internal standards from Isotec (St. Louis, MO),

Cambridge Isotope Laboratories (Andover, MA), and CDN Isotopes (Pointe-Claire,

Quebec, CN) to samples. Sample Preparation: Biological triplicates of 15cm plates under each treatment condition were placed on ice and washed twice with ice-cold PBS before as much PBS as possible was removed. Cells were lysed in 620ul ml of 0.78% Formic

Acid in water and scraped to collect. 30ul were removed for protein quantification. 1x

131

volume of the collected pellet (~800ul) of acetonitrile was added and sample was vortexed vigorously (10-15 seconds on level 8-9). Aliquots were separated for mass spectrometry measurements (300 ul for organic acids, 100 ul for amino acids/acyl- carnitines) and were immediately frozen on dry ice and transferred to -80°C. Protein concentration per replicate was determined by the Pierce BCA Kit (ThermoScientific,

Waltham, MA, USA) and used to normalize all metabolite levels.

4.2.10 NAP+/NADPH measurements

A ratio of NADP+/NADPH was calculated after measuring each molecule separately with the Amplite Fluorimetric NADP/NADPH Ratio Assay Kit from AAT

Bioquest, Inc (Sunnyvale, CA). The protocol was conducted as the manufacturer suggested and all values were normalized to protein content, as measured by the Pierce

BCA kit, on plated and treated samples done in parallel.

4.2.11 Chromatin Immunoprecipitation

3 million H1975 cells were plated in 15 cm dishes. After 24 hours the cells were serum starved to 0.5% FBS and 24 hours later they were treated with either a 4-hour treatment of α-KG or an 8 hour treatment of hypoxia before collection for a native ChIP.

The protocol was carried out as the manufacturer suggested with the SimpleChIP Plus

Enzymatic Chromatin IP Kit (Agarose Beads) (Cell Signaling Tech., cat. no. 9004).

Sonication and digestion were performed to obtain chromatin 1-4 nucleosomes in size, which was verified by gel electrophoresis. An extra seeded and treated plate was

132

counted immediately before collection to adjust collection for each IP to 4-5 million cells.

The immunoprecipitations were performed overnight.

4.2.12 DNA methylation

After genome DNA isolation, the remaining portion of this protocol was conducted in the laboratory of Susan K. Murphy. Genomic DNA was extracted with the

DNeasy Blood & Tissue Kit according to the protocol provided by the manufacturer

(QIAGEN). The genomic DNAs (800 ng) were modified by treatment with sodium bisulfite using the Zymo EZ DNA Methylation kit (Zymo Research, Irvine, CA). Bisulfite treatment of denatured DNA converts all unmethylated cytosines to uracils, leaving methylated cytosines unchanged, allowing for quantitative measurement of cytosine methylation status. Pyrosequencing was performed using a Pyromark Q96 MD pyrosequencer (QIAGEN). The bisulfite pyrosequencing assays were used to quantitatively measure the level of methylation at CpG sites contained. Assays were designed to query CpG islands using the Pyromark Assay Design Software (QIAGEN).

Pyrosequencing was performed using the sequencing primer. PCR conditions were optimized to produce a single, robust amplification product. Defined mixtures of fully methylated and unmethylated control DNAs were used to show a linear increase in detection of methylation values as the level of input DNA methylation increased

(Pearson r > 0.98 for all regions). Once optimal conditions were defined, each assay was analyzed using the same amount of input DNA from each specimen (40 ng, assuming

133

complete recovery after bisulfite modification). Percent methylation for each CpG cytosine was determined using Pyro Q-CpG Software (QIAGEN).

4.3 Investigating the specificity of the protective phenotype of ACC1 depletion

To begin studying the phenotype of ACC1 depletion under hypoxia in more detail, we investigated the specificity of this hypoxia protection by ACC1 depletion.

First, we tested the phenotype of the other acetyl-CoA isoform, ACC2, under hypoxia.

Its silencing by shRNA did not offer a similar hypoxia protection as seen with ACC1 depletion (Figure 12j). Next, we determined whether depletion of ACC1 protected against other TME stresses (lactic acidosis, glutamine deprivation and glucose deprivation). We found that the protective effect of shACC1 was seen only under hypoxia (Figure 12k, l). Since the HIFs are the major transcriptional responders to hypoxia, we examined how loss of ACC1 affected HIF-1α levels. Interestingly, we found

ACC1 depletion decreased levels of HIF-1α under hypoxia across multiple cell lines

(Figure 13a-c). These data suggested that the protective phenotype of ACC1 depletion was not due to up-regulation of the HIF response. Overall, these results indicated that only the cytosolic isoform of acetyl-CoA carboxylase (encoded by ACC1) was essential for apoptosis under a hypoxic stress.

When we examined the effect of blocking enzymes up- or downstream of ACC1, we found that silencing ATP citrate lyase (ACLY) also enhanced survival under hypoxia 134

(Figure 14a, b). ACLY encodes the enzyme immediately upstream of ACC1 in lipogenesis, catalyzing the formation of acetyl-CoA and oxaloacetate from citrate.

Similar to ACC1, this protection results from the inhibition of hypoxia-induced apoptosis (Figure 14c). The protective effect of ACLY depletion was also reproduced in

MDA-MB-231 cells (Figure 12c, d) and PANC-1 cells (Figure 12f, g). H1975 and MDA-

MB-231 shACLY cells also exhibited decreased HIF-1α expression under hypoxia as was seen in the shACC1 cells (Figure 13d, e). These data showed that blocking lipogenesis at the points of either ACLY or ACC1 inhibited apoptosis and permitted the survival of cancer cells from multiple tissue types under hypoxia.

135

Figure 12. The depletion of ACC1, but not ACC2, protects cells specifically under a hypoxic stress.

136

(A) Western blot of ACC1 protein knockdown by 2 shRNAs in MDA-MB-231 cells. (B) Quantified crystal violet of shACC1 MDA-MB-231 cells after 4 days of hypoxia (n=3). (C) Western blot of ACLY protein knockdown by 2 shRNAs in MDA-MB-231 cells. (D) Quantified crystal violet of shACLY MDA-MB-231 cells after 4 days of hypoxia (n=3). (E) Western blot of ACC1 protein knockdown in PANC-1 cells. (F) Western blot of ACLY protein knockdown in PANC-1 cells. (G) Quantified crystal violet staining of indicated shRNA PANC-1 cells after 6 days of hypoxia (n=3). (H) Crystal violet staining of H1975 cells with simultaneous metformin treatment and hypoxia for 4 days. (I) Western blot of PARP in H1975 cells with hypoxia and metformin treatment. (J) Counts of viable cell number by trypan blue exclusion of shACC2 cells under normoxia or hypoxia for 4 days (n=9). (K) and (L) Crystal violet of ACC1 and scramble (Scr) cells (boxed in blue rectangles) under indicated stresses (K-hypoxia, L- LA (lactic acidosis), no glutamine or no glucose (Glu)) for 4 days. Data are from indicated cell line.

137

4.4 Loss of ACC1 or ACLY did not protect cells from hypoxia- induced apoptosis through NADPH conservation to relieve oxidative stress

Lipogenesis is a highly anabolic process that requires significant amounts of

NADPH and ATP. As mentioned, silencing ACC1 protected cells from death caused by glucose deprivation and matrix detachment by preserving NADPH and ATP to counteract the ensuing oxidative stresses (Jeon et al., 2012; Schafer et al., 2009). We tested the relevance of these factors in our system. In H1975 cells, silencing ACC1 trended toward increasing the NADP+/NADPH ratio, suggesting a decrease in available

NADPH (Figure 15a). We reasoned that if the NADPH were being used to combat elevated reactive oxygen species under hypoxia, then supplementation with antioxidants should protect control cells from hypoxia-induced death similar to Jeon et al., 2012. However, neither the addition of N-acetyl cysteine nor glutathione antioxidants rescued hypoxia-induced death in control cells (Figure 15b, c). While ACC1 silencing increased ATP levels, the change in ATP levels from normoxia to hypoxia was consistent in control and knockdown cells and thus could not readily explain the hypoxia protection (Figure 15d). Therefore, in these cells with ACC1 or ACLY depleted, changes in NADPH and ATP levels may not be the primary mechanism for cell survival under hypoxia. Therefore, we sought to identify another mechanistic explanation for this hypoxia protection phenotype.

138

Figure 13. Effect of depleted ACC1 or ACLY on levels of HIF-1α.

(A) Western blot of HIF-1α protein levels with ACC1 knockdown by 2 shRNAs in H1975 cells. (B) Western blot of HIF-1α protein levels with ACC1 knockdown by 2 shRNAs in MDA-MB-231 cells. (C) Western blot of HIF-1α protein levels with ACC1 knockdown by 2 shRNAs in PANC-1 cells. (D) Western blot of HIF-1α protein levels with ACLY knockdown by 2 shRNAs in H1975 cells. (E) Western blot of HIF-1α protein levels with ACLY knockdown by 2 shRNAs in MDA-MB-231 cells. 139

Figure 14. Depletion of ACLY protects from hypoxia-induced apoptosis.

(A) Crystal violet staining of scramble and ACLY shRNA cells after 4 days of hypoxia. (B) Western blot showing efficiency of shRNAs targeting ACLY and inhibition of PARP cleavage after 48 hours of hypoxia. (C) Viable cell numbers of indicated cells after 4 days of hypoxia as determined by counting nuclei (n=9). All data is from the H1975 cell line.

140

Figure 15. Loss of ACC1 does not protect cells from hypoxia-induced apoptosis through preservation of NADPH to relieve oxidative stress.

(A) NADP+/NADPH ratio under normoxia and hypoxia in shScr or shACC1 H1975 cells (n=6). (B) Crystal violet staining of shScramble H1975 cells treated with N-acetyl cysteine (2mM) under normoxia or hypoxia for 4 days. (C) Quantified crystal violet staining of shScramble H1975 cells after addition of glutathione under normoxia or hypoxia (n=3). (D) Protein-normalize ATP levels in indicated shRNA cell line under normoxia or hypoxia (n=9). (E) qPCR results of ETV1 and ETV5 mRNA levels in shACC1 cells under hypoxia or normoxia (n=6). (F, G) GSEA analysis showing high overlap of genes changed with ETV4 and ACC1 (left panels) or ACLY (right panels) depletion. (F) Enrichment of ETV4-up-regulated genes in shACC1 (left panel) or shACLY (right panel) cells. (G) Depletion of ETV4-down-regulated genes in shACC1 (left panel) or shACLY (right panel) cells. All data are from the H1975 cell line.

141

4.5 Depletion of the transcription factor ETV4 led to similar hypoxia-protection and gene expression phenotypes as with the loss of ACC1 or ACLY

As described in Chapter 3, our “multiple hairpin analysis” of the genome-wide screen identified the loss of a PEA3 transcription factor family member, ETV4, as protective under hypoxia (Figure 7). A link between lipogenesis and ETV4 was previously established when levels of malonyl-CoA were associated with ETV4 activity

(Menendez et al., 2004). This prompted us to validate the phenotype for ETV4, as described in Chapter 3 (Figure 7). Once this phenotype was validated, we investigated a potential regulatory relationship between ACLY, ACC1 and ETV4. Real-time PCR analysis showed that hypoxia led to a reduction of ETV4 mRNA in the ACC1-depleted, but not control cells (Figure 16a). Additionally, we noted correspondingly reduced ETV4 protein in the shACC1 cells as compared to the scramble cells (Figure 16b). Reduced

ETV4 mRNA and protein levels were also noted in the ACLY-depleted cells (Figure 16c, d). While ETV4 protein levels were somewhat decreased under normoxia, the down- regulation was stronger under hypoxia. We are investigating the normoxic regulation between ETV4, ACLY and ACC1 currently, and these efforts are discussed as a future direction in Section 5.7. The regulation by ACC1 and ACLY was mostly specific to ETV4; the other PEA3 subfamily members, ETV1 and ETV5, were not consistently altered by

ACC1 depletion (Figure 15e). Additionally, neither ETV1 nor ETV5 were identified as

“multiple hairpin hits” in the shRNA screen, therefore we focused our studies on ETV4.

142

Since ETV4 is a transcription factor, we investigated the contribution of reduced

ETV4 activity to the transcriptional response of ACLY or ACC1 depletion. To do this, we used microarrays to analyze the global transcriptional response to the silencing of each

ACC1, ACLY or ETV4 by two independent shRNAs under hypoxia (each shRNA was done in triplicate). The transcriptional responses were determined by zero- transformation against the shScramble cells (Tang et al., 2015). Next, the data were filtered with a 1.7-fold change in at least six arrays and the selected 641 probe sets were grouped by hierarchical clustering (Figure 17a). This analysis revealed a remarkable similarity between the transcriptional responses to the depletion of ETV4, ACC1 or

ACLY with the induction and repression of common sets of genes (Figure 17a). Using the GATHER algorithm (Chang and Nevins, 2006), we noticed an “anti-apoptotic expression program” that included both the induction of negative regulators of apoptosis signaling such as NQO1, CYP1B1 and SERPINE1 (Zhang et al., 2013b; Zhu et al., 2014) and the repression of the apoptosis-promoting genes BIK, TNFRSF9, TNFAIP3,

GLIPR1, DDIT and TRIB3 (Figure 17a). The induction and repression of multiple genes in the shACC1, shACLY and shETV4 cells were confirmed by real-time qPCR (Figure

17b, c). Using GSEA, these gene expression changes were highly overlapping in all pairwise comparisons for both up- and down-regulated genes (Figure 15f, g). Since

ETV4 is most often considered a transcriptional activator, it is likely that these changed mRNAs represent both direct and indirect targets of ETV4.

143

Figure 16. Depletion of ACC1 or ACLY led to loss of ETV4 protein levels under hypoxia.

(A) qPCR results of ETV1 and ETV5 mRNA levels in shACC1 cells under hypoxia or normoxia (n=6). (B) Western blot of ETV4 protein levels with ACC1 knockdown under normoxia or hypoxia. (C) qPCR results of ETV1 and ETV5 mRNA levels in shACC1 cells under hypoxia or normoxia (n=6). (D) Western blot of ETV4 protein levels with ACLY knockdown under normoxia or hypoxia.

144

We used two different approaches to further evaluate whether the ACC1- affected genes were direct transcriptional targets of ETV4. First, we compared publicly available ETV4 ChIP-seq data from PC3 cells (Cistrome Finder (Sun et al., 2013)) with the genes that were changed in our microarray analysis of H1975 cells with loss of ETV4,

ACC1 or ACLY (Figure 17a). While performed in a different cell line, these analyses still identified at least two potential direct ETV4 target genes, PLEC (Figure 18a) and DUSP6

(Figure 18b). For both genes, there were ChIP peaks indicating direct ETV4 binding that overlapped with histone H3 lysine 27 acetylation (a mark of actively transcribed gene bodies) and DNase hypersensitive regions of open chromatin (Figure 18iv,v). DUSP6 has been previously described as an ETS transcription factor family target (Znosko et al.,

2010). While PLEC was reported to interact with an ETV4 direct target (Chen et al.,

1996), this analysis suggested that PLEC itself might represent a novel ETV4 target.

In the second approach, we used qPCR to determine if the ACC1-affected genes could be “rescued” by ETV4 over-expression. We found that CTSS, COL13A1, DUSP6 and SERPINE1 could be reversed by ETV4 re-expression (Figure 19a, b). In contrast, the

ACC1-altered expression of other genes was either partially or not restored upon ETV4 overexpression (Figure 19c). Interestingly, many of the non-rescuable genes were immune-related genes, many of which were documented as NF-kB targets (Chang and

Nevins, 2006). This analysis suggested that some of the gene expression changes discovered by our microarray analysis may represent direct effects of changed ETV4

145

transcription, while others likely represent more indirect changes with ETV4 loss.

Together, these data indicated that the repression of ETV4 played an important role in a substantial subset of the transcriptional response to ACC1 or ACLY depletion.

In order to better understand if these changes reflected an in vivo biological regulation between these genes, we developed “gene signatures” associated with the silencing of ETV4, ACC1 or ACLY using the CreateSignature algorithm (Chang et al.,

2011). These gene expression signatures represent “quantitative phenotypes” that reflect the loss of these genes. Comparing their similarity in different expression datasets allowed us to recognize similar quantitative changes in these genes in both in vitro experimental perturbation and human tumors. Similar “gene signature” approaches have been used to define the influences of oncogenic signaling and TME stresses in multiple cancer types (Chen et al., 2010b; Gatza et al., 2011; Lucas et al., 2010). Gene expression patterns from human tumor samples (Chin et al., 2006) were separated by their similarity to our developed gene signatures associated with loss of ETV4 (shETV4),

ACC1 (shACC1) or ACLY (shACLY). Binary regression from this analysis in human tumors showed highly statistically significant correlations between the shACC1 or shACLY and shETV4 signatures (Figure 17d). In other words, patient tumors with gene expression patterns more similar to the ACC1-depletion (shACC1) signature had expression patterns that were also more similar to the ETV4-depletion (shETV4) signature; likewise, patient tumors with gene expression patterns similar to the ACLY-

146

depletion (shACLY) signature also had similar gene expression patterns with the ETV4- depletion (shETV4) signature. Importantly, this showed that the regulation between

ACC1/ACLY and ETV4 was relevant in tumor expression datasets. Overall, multiple analyses demonstrated the similarity of the transcriptional responses to the depletion of

ACC1, ACLY or ETV4 and suggested that ETV4 mediated a portion of the transcriptional effect downstream of ACLY or ACC1 both in vitro and in vivo.

147

Figure 17. Global transcriptional response to depletion of ACC1, ACLY or ETV4 is highly similar.

148

(A) Heat map of transcriptional responses to the depletion of indicated genes (ETV4, ACC1 or ACLY) by two shRNAs each (n=3 per shRNA) under hypoxia. Data are log2 values that have been zero transformed to the scramble shRNA cells. Filtering criteria of at least six occurrences with values greater than 1.7 resulted in 641 probe sets. (B, C) qPCR validation of mRNA changes of similarly (B) down-regulated and (C) up- regulated genes with depletion of ETV4, ACLY or ACC1 (n=6). (D) Binary regressions of shACC1 (left panel) or shACLY (right panel) and shETV4 gene signatures when compared across gene expression patterns of 130 breast cancer tumor samples (see text and Methods).

149

Figure 18. PLEC and DUSP6 may be direct transcriptional targets of ETV4.

Modified UCSC Genome Browser and CistromeFinder interfaces showing ETV4 binding in the regulatory regions of (A) PLEC and (B) DUSP6. For both (A) and (B): (i) shows location of gene in genome; (ii) shows peaks of binding from ChIP-Seq data with ETV4 in PC3 cells, highlighted by red box; (iii) shows the annotated gene structures for each gene; (iv) shows abundance of acetylated-Histone H3 lysine 27 (H3K27Ac) at these locations; (v) dark bars to represent DNase hypersensitivity clusters at these genomic locations.

150

Figure 19. ACC1-altered genes likely represent both ETV4-dependent and - independent transcriptional targets.

151

(A) Western blot showing overexpression of ETV4 in ACC1 depleted cells by 2 shRNAs. (B) qPCR analysis of a set of indicated genes whose ACC1-affected changes can be reversed with ETV4 expression, consistent with a pattern consistent of being downstream targets of ETV4 (n=6). (C) qPCR analysis of a set of indicated genes whose changes could not be reversed with ETV4 expression, consistent with a pattern of not being downstream targets of ETV4 (n=6). Data are represented as mean values +/- SEM. All data are from the H1975 cell line.

152

4.6 Global metabolomics revealed that hypoxia-induced elevated α-ketoglutarate levels protected cells from apoptosis

Considering that ACC1 and ACLY are critical lipogenic enzymes, we performed a metabolomics experiment to interrogate the metabolic effects of ACC1 or ACLY depletion under normoxia or hypoxia. Five cell lines were evaluated in triplicate: 1 control “hypoxia-sensitive” cell line (shScramble line) and four “hypoxia-survival” cell lines (2 shACC1 lines, 2 shACLY lines). After 36 hours of treatment, cells were lysed on ice and collected to measure the intracellular levels of 15 amino acids and 45 acyl- carnitines by tandem mass spectrometry (MS/MS) and levels of 7 organic acids by gas chromatography and mass spectrometry (GC/MS). All measurements were normalized by total protein content per sample.

We observed several expected metabolic changes to validate our approach.

Silencing ACC1 depleted basal and hypoxia-induced palmitate levels, reflecting reduced de novo lipogenesis in these cells (Figure 20a). Consistent with hypoxia-induced inhibition of pyruvate utilization in the TCA cycle in favor of glycolysis, hypoxia modestly increased the levels of pyruvate and lactate in control cells (Figure 20b, c). In addition, as previously noticed (Tennant et al., 2009), hypoxia generally reduced the levels of TCA metabolites succinate, fumarate, malate, and citrate in most cells (Figure

21a). These results indicated that our metabolomics assay accurately detected the expected metabolic changes associated with inhibited lipogenesis and hypoxia exposure.

153

The pattern of α-ketoglutarate (α-KG) levels in this experiment suggested that it might be an interesting candidate for offering protection under hypoxia. In the control cells, hypoxia reduced the levels of α-KG. However, in the hypoxia-resistant cells with depleted ACC1 and ACLY, hypoxia increased the α-KG levels (Figure 21b).

We tested the possibility that levels of α-KG contributed to survival by adding cell-permeable dimethyl-α-KG to H1975 cells. In order to ensure that our supplementation treatment would be relevant to the levels of α-KG seen with ACC1 or

ACLY depletion, we determined the level of α-KG achieved intra-cellularly after extracellular supplementation with different levels of dimethyl-α-KG. Mass spectrometry analysis showed increasing amounts of intracellular α-KG after supplementation in a dose-dependent manner (Figure 21c). Dimethyl-α-KG supplementation at 1mM achieved levels of α-KG comparable to the hypoxia-induced increase found in the ACC1 and ACLY depleted cells under hypoxia (Figure 21b, c).

Supplementation with relevant levels α-KG also inhibited PARP cleavage under hypoxia

(Figure 21d). These data indicated that the increased α-KG under hypoxia in the ACC1 and ACLY depleted cells recapitulated the hypoxia-protective phenotype of these cells and thus is a downstream mediator of ACC1 and ACLY depletion under hypoxia.

154

Figure 20. Metabolomics assay accurately reflects expected changes and the gene expression effects of different doses of α-KG.

(A) Protein-normalized levels of palmitate measured in the indicated shRNA cell lines under normoxia (n=3). (B, C) Protein-normalized levels of pyruvate (B) and lactate (C) in shScramble cells under normoxia and hypoxia (n=3). (D) qPCR analysis of ETV4 mRNA levels after supplementation of α-KG at 3.5mM for 24 hours (n=6). (E) Western blot of ETV4 protein levels after supplementation of α-KG at 3.5mM for 24 hours. (F, G) qPCR analysis of the changes of gene expression in the indicated genes after supplementation of different levels of α-KG for (F) up-regulated and (G) down-regulated genes (n=6). (H) 155

Western blot of HIF-1α protein levels after supplementation of α-KG at indicated doses.***** = p<0.0001. All data are from the H1975 cell line.

156

2 1

Figure 21. Loss of ACC1 or ACLY results in elevated levels of α-KG under hypoxia that protects from hypoxia-induced apoptosis.

(A) Heat map showing percent change under hypoxia of various organic acids in indicated cell lines. Yellow=increase in hypoxia; Blue=decrease in hypoxia. Each shRNA cell line (2 per gene) was done in triplicate (n=3). (B) Protein-normalized fold change of intracellular α-KG levels in hypoxia in indicated cell lines. Dashed line indicates no change in hypoxia from normoxia (n=3). (C) Normalized intracellular levels of α-KG after supplementation with indicated concentration dimethyl-α-KG for 24 hours (n=3). (D) Western blot of intact and cleaved PARP after α-KG supplementation in control cells under normoxia or hypoxia for 48 hours. Data are represented as mean values +/- SEM, unless otherwise noted. Data are from the H1975 cell line.

157

4.7 Hypoxia-induced increased α-ketoglutarate levels regulated ETV4, possibly through 2-oxoglutarate/Fe(II)-dependent dioxygenases

With both α-KG and ETV4 acting downstream of ACC1 or ACLY, we next wanted to determine if α-KG was mediating the effects of ACC1 or ACLY on ETV4 expression. α-KG supplementation reduced mRNA and protein levels of ETV4 at both lower (Figure 22a, b) and higher concentrations (Figure 20d, e). Additionally, α-KG supplementation in control cells caused similar changes in repressed and induced genes as was caused by ACC1, ACLY or ETV4 silencing (Figure 22c, d), and some of these mRNA effects were dose-dependent with α-KG supplementation (Figure 20f, g).

Unexpectedly, dimethyl-α-KG supplementation increased HIF-1α protein levels (Figure

20h). Since the regulation of HIF-1α was different with depletion of ACC1/ACLY or α-

KG supplementation, changes in HIF-1α protein likely did not explain the improved hypoxic cell survival of the ACC1/ACLY depleted cells. Combined, these data indicated that, in the ACLY or ACC1 depleted cells, the increase in α-KG was a hypoxic trigger that reduced ETV4 levels and activity to mediate an anti-apoptotic gene expression response.

158

Figure 22. α–ketoglutarate affects ETV4 levels and activity possibly through histone demethylase 2-OGDDs.

(A) qPCR showing ETV4 mRNA levels after α-KG supplementation (1mM) over time (n=6). (B) Western blot showing ETV4 protein levels after α-KG supplementation (1mM) over time. (C,D) qPCR of gene expression changes after α-KG supplementation (3.5mM, 24h) for both (C) down-regulated genes (similar to Fig. 4b) and (D) up-regulated genes (similar to Fig. 4b,c) (n=6). (E) Ratio of α-KG/succinate in normoxia and hypoxia in the indicated cell lines from the metabolomics experiment (n=3). (F) Western blot showing 159

effect of α-KG supplementation (1mM) on histone methylation marks after indicated time of treatment. Numbers indicate densitometry analysis by Image J relative to that sample’s loading of total H3. Data are from the H1975 cell line.

160

In addition to being a TCA cycle intermediate, α-KG is a substrate for the 2- oxoglutarate/Fe(II)-dependent dioxygenases (2-OGDDs), of which there are at least 60 in humans. 2-OGDDs use α-KG (aka 2-oxoglutarate) and molecular oxygen as substrates to perform a number of different protein modification reactions (Loenarz and Schofield,

2008). These enzymes include families of histone demethylases that recognize and remove methylation marks from histones, as well as the TET family of proteins that facilitate DNA demethylation (Loenarz and Schofield, 2008). Thus, α-KG levels can affect gene expression through the activities of these 2-OGDDs (Carey et al., 2015).

An elevated ratio of α-KG/succinate (substrate/product ratio) has been proposed as a potential indicator of increased 2-OGDD activity (Carey et al., 2015; Kaelin, 2011).

We found that the α-KG/succinate ratio was significantly elevated in all of the shACC1 and shACLY cells under hypoxia (Figure 22e). To better understand if the level of α-KG or the α-KG/succinate ratio determined our hypoxia-survival phenotypes, we supplemented ACLY or ACC1 depleted cells with cell-permeable dimethyl-succinate to theoretically drive the α-KG/succinate ratio in the opposite direction from when α-KG was added. Succinate supplementation did not affect the survival of either shACC1 or shACLY cells under hypoxia (Figure 23a, b) and also did not affect the regulation of

ETV4 by ACC1 or ACLY (Figure 23c, d) Therefore, in our experimental system, we concluded that the levels of α-KG, rather than the α-KG/succinate ratio, drove the hypoxia survival phenotype.

161

Figure 23. Effects of succinate or α–KG supplementation on survival, ETV4 regulation and epigenetic changes.

(A) Quantified crystal violet stain of shACC1 and shScr cells supplemented with indicated dose of dimethyl-succinate under normoxia or hypoxia for 4 days (n=3). (B) Quantified crystal violet stain of shACLY and shScr cells supplemented with indicated dose of dimethyl-succinate under normoxia or hypoxia for 4 days (n=3). (C) qPCR analysis of the relative change in ETV4 mRNA levels in two shACC1 cells with the addition of DMSO or succinate (4mM). Ratio of 1 (dashed line) indicates no change with treatment (n=6). (D) qPCR analysis of the relative change in ETV4 mRNA levels in two shACLY cells with the addition of DMSO or succinate (4mM). Ratio of 1 (dashed line) indicates no change with treatment (n=6). (E) Percent of methylated CpG sites across the

162

two shore regions and center of the ETV4 promoter CpG island as determined by bisulfite sequencing (n=5). (F) Western blot analysis of indicated histone modification when supplemented with either DMSO or α-KG (1mM) for indicated length of time. (G) Western blot of indicated histone modifications with α-KG supplementation (3.5mM) for indicated length of time. All data are from the H1975 cell line.

163

While succinate supplementation did not affect our phenotype, the elevated α-

KG in the ACC1 or ACLY depleted cells under hypoxia still suggested that 2-OGDDs may be relevant in our system. Therefore, we hypothesized that α-KG affected ETV4 mRNA abundance by altering the activity of 2-OGDDs and subsequent histone methylations. As a control, we tested the DNA methylation status of the two shore regions (at the edges) and the center region of the ETV4 promoter CpG islands by bisulfite pyro-sequencing and saw no significant change upon α-KG supplementation

(Figure 23e). However, α-KG supplementation caused a global reduction in two

(H3K4me2 and H3K4me3) “active” histone methylation marks and also reduced the

“repressive” marks, H3K27me3 and, to a lesser extent, H3K9me3 (Figure 22f). When we compared the epigenetic changes associated with α-KG supplementation with Carey et al. (2015), we found that H3K27me3 was consistently reduced with the addition of α-KG in both studies. However, there were also differences in the histone methylation changes caused by α-KG between the two studies: 1) H4K20me3 was reduced with α-KG previously and we saw an increase in this mark relative to the control (Figure 23f); 2) we saw a decrease in the levels of H3K4me3 in the current experiment, although no changes were seen previously. These differences could be due to different cellular contexts

(embryonic stem cells vs. cancer cells) or the presence of glutamine deprivation during the previous examination of α-KG effects (Carey et al., 2015). Similar changes were observed using either water or DMSO as control (Figure 21g). Collectively, the reduced

164

levels of H3K4me2/3, H3K27me3 and H3K9me3 marks were consistent with our hypothesis that predicted more active histone demethylases as a result of increased levels of α-KG.

To extend evidence in support of our model, we also examined various histone methylation modifications with the depletion of ACC1 or ACLY. Overall, we saw similar global histone methylation changes in both the shACC1 (Figure 24a) and shACLY (Figure 24b) cells as compared to α-KG supplementation. Among all the tested epigenetic markers, the H3K4me3 mark was the most strikingly decreased across both gene depletions and the α-KG treatment. The H3K4me2 mark was modestly decreased in all three conditions (shACC1, shACLY, α-KG supplementation). Similarly, H3K9me3 was decreased somewhat with the 8 hour α-KG treatment and in the shACC1 cells, while it was more strongly decreased in the shACLY cells. H3K27me3 was lowered by both α-KG and ACC1 depletion. Levels of H4K20me3 were unchanged in the shACC1 and shACLY cells while they were increased with α-KG treatment, and so this suggested that the changes in this methyl mark were likely not due to the changes we explain in our model.

Besides global epigenetic changes, we also sought to determine if histone methylation specifically at the ETV4 locus was changed by chromatin immunoprecipitation of the “active” histone H3 lysine 4 tri-methylation (H3K4me3) mark. This mark was chosen because it showed the most consistent and strongest

165

changes across both genetic depletions and α-KG supplementation. In addition, a loss of this “active” mark would be consistent with decreased ETV4 expression in these conditions. ChIP experiments showed that α-KG treatment decreased the abundance of the H3K4me3 modification at the ETV4 locus by ~30% (Figure 24c). Additionally, there was decreased abundance of H3K4me3 in both ACC1- and ACLY-depleted cells under hypoxia as compared to control cells (Figure 24d, e). Together, these data provide compelling evidence that elevated levels of α-KG affected the histone, but not DNA, methylation status of the ETV4 locus and this pattern was consistent with the histone methylation changes under hypoxia with either α-KG supplementation or ACC1 or

ACLY depletion.

Collectively, our data are consistent with a model in which, under hypoxia, the inhibition of ACC1 or ACLY increased levels of α-ketoglutarate to block hypoxia- induced apoptosis by reducing the levels and activity of ETV4, possibly through altered histone methylation patterns (Figure 25). These data reveal a novel molecular connection showing that the transcriptional program induced by altered lipogenic metabolism can modulate survival under hypoxia.

166

Figure 24. Global and ETV4-specific histone methylation changes in shACC1 and shACLY cells.

167

(A) Western blot showing levels of the indicated histone methylation marks in shACC1 cells under normoxia or hypoxia (24h). (B) Western blot showing levels of the indicated histone methylation marks in shACLY cells under normoxia or hypoxia (8h). (C) ChIP- qPCR analysis of the abundance of the H3K4me3 mark at two locations in the promoter region of ETV4 after α-KG treatment (n=6). (D, E) ChIP-qPCR analysis of the abundance of the H3K4me3 mark at two locations in the promoter region of ETV4 after ACC1 or ACLY depletion under normoxia and hypoxia (n=9). n.s.= not significant; * = p<0.05, ** = p<0.01, *** = p<0.005, ***** = p<0.0001. Data are from the H1975 cell line.

168

Figure 25. Model of how the ACLY-ACC1-KG-ETV4 axis protects from hypoxia-induced apoptosis.

169

5. Discussion

In this dissertation, I have described two parallel RNAi screen strategies that interrogated the genetic determinants of cancer cell survival under hypoxia or lactic acidosis. The positive selection screen design could be improved to reduce background and false-positive hits, but it still identified SEL1L and other candidate regulators of cell survival under LA. The potential hypotheses and suggested future experiments for candidate genes from the positive selection screen were discussed in Chapter 2.

The genome-wide pooled screen successfully identified multiple genes that influence cancer cell line survival under hypoxia or lactic acidosis (Chapter 3). In addition to validating the phenotypes of five genes, I have developed a detailed model for how one top candidate gene (ACC1) protected cells from hypoxia-induced apoptosis.

Our data revealed that blocking de novo lipogenesis through the genetic depletion of

ACLY or ACC1 protected cancer cells from hypoxia-induced apoptosis by increasing hypoxic levels of α-ketoglutarate, which inhibited ETV4 and its pro-apoptotic transcriptional activities. Therefore, inhibition of ACLY or ACC1 affected both metabolism and transcription to inhibit hypoxia-induced apoptosis. Thus, one mechanism of hypoxia-induced apoptosis was through the metabolite α-KG, whose depletion under hypoxia normally maintained the levels and activity of ETV4 through histone modifications to trigger apoptosis. This molecular mechanism provides an explanation for the observed hypoxia-protection phenotype, but leaves a number of

170

points for discussion and suggests specific questions for future experiments. I discuss these results and their implications in Sections 5.1-5.5. In Section 5.6 I discuss one future/on-going research project that is an extension of the work I described here. I end this Chapter with a final Conclusion in Section 5.7.

5.1 Model of tumorigenic potential versus stress survival

Multiple previous reports show that increased activity of ACLY (Bhalla et al.,

2011; Migita et al., 2008; Zaidi et al., 2012), ACC1 (Brusselmans et al., 2005; Chajes et al.,

2006; Zhan et al., 2008) or ETV4 (Aytes et al., 2013; Keld et al., 2011; Pellecchia et al.,

2012; Yuen et al., 2011) is associated with increased tumorigenicity and/or poor patient outcome, or that inhibiting these genes’ activities reduces tumorigenicity and improves patient outcome. The presence of tumor hypoxia also is generally a negative prognostic feature. We show that inhibition of ACLY, ACC1 or ETV4 paradoxically allows tumor cells to survive better under hypoxia. To address this apparent paradox, we propose a conceptual model in which there are two cellular states: one of activating oncogenesis and proliferation versus another of stress survival. This model includes a trade-off between the two states, such that the promotion of one comes at the expense of the other. Stated another way, the activation of various proliferative programs can accelerate many oncogenic features, but may also render cells susceptible to apoptosis, especially

171

under stresses. Likewise, reduced oncogenic programs could slow cellular proliferation to a “dormant” state and allow for better stress survival.

A similar model has been proposed for several oncogenes. Oncogenesis driven by MYC rendered non-transformed cells vulnerable to hypoxia-induced apoptosis

(Brunelle et al., 2004; Schmaltz et al., 1998); the degradation or cleavage of c-MYC under hypoxia allowed tumor cells to evade hypoxia-induced apoptosis (Conacci-Sorrell et al.,

2014; Wong et al., 2013). Different E2Fs can promote proliferation (oncogenesis) or apoptosis in different cellular contexts, such as with differing PI3K activity (Hallstrom et al., 2008). A recent paper also indicated that oncogenic HIF-1α repressed the ATF4- mediated stress response pathway to allow for the expansion of fetal cardiomyocytes

(Guimarães-Camboa et al.), reflecting an occurrence of increased proliferation and decreased stress survival pathways. Thus, there is precedence for oncogenic programs to be in a mutually exclusive balance with stress survival mechanisms.

The concept of reduced proliferation permitting stress survival is wide-reaching across multiple organisms and phylogenies. While gene expression changes and signaling pathways are classically considered the major responders to stress (Lopez-

Maury et al., 2008), altering metabolic programs (generally to decrease anabolic growth) to cope with stress survival is also seen across phylogenies. In bears, massive depressions in metabolism are necessary to sustain the animal through hibernation periods (Carey et al., 2003). In multiple organisms, reduced nutrient availability and

172

growth leads to increased stress tolerance, which can also lead to increased lifespan

(Bishop and Guarente, 2007). In yeast, a lack of nutrient availability can induce quiescence, filamentous growth or sporulation, all of which represent hardy, stress- resistant cellular forms (Lopez-Maury et al., 2008). Also in yeast, stress resistance was negatively correlated with growth rate (Elliott and Futcher, 1993). In response to stresses, Arabidopsis thaliana decreases expression of DNA synthesis and fatty acid metabolism, showing reduced cellular proliferative capacity under stress (Ma and

Bohnert, 2007). Caenorhabditis elegans, while in their quiescence lifecycle stage of dauer, are also robustly resistant to stress (Fielenbach and Antebi, 2008). Our finding exemplifies this wide reaching concept and shows similarities between organismal stress survival and cellular physiology in cancer biology.

The presence of slowly proliferating or quiescent cancer cells is most often discussed through the concept of “tumor dormancy”, which could also reflect the presence of tumor stem cells. There are multiple models of cancer cell dormancy that differ in expectations of cell cycle arrest and the influences of angiogenesis and cytotoxic immunity (Aguirre-Ghiso, 2007). Importantly for relevance to our data, these quiescent or slowly proliferating cells have an acute ability to survive stress (Wells et al, 2013;

Skvortsova et al., 2015), and can influence tumor evolution through their ability to survive stress (Yeh and Ramaswamy, 2015). I propose that our experimental conditions have unintentionally established a simplified in vitro model for investigating the

173

properties and survival mechanisms of dormant-like cancer cells. In other words, our cell culture model of low growth combined with a hypoxic stress can induce a growth- constrained phase, similar to tumor dormancy. This could be advantageous as one challenge to understanding dormant cells is the lack of good models (Yeh and

Ramaswamy, 2015). Studies of other genes identified as hits in our screen may reveal additional mechanisms of “dormant” cell survival. Of course, our conditions are not a perfect model, especially considering the different time scales in question, but still provide a way to study mechanisms of survival for slow-growing cancer cells.

In addition to being consistent with our proposed model of proliferation and stress survival in a state of balance, tumor dormancy specifically fits with our data on loss of ACC1 or ACLY protecting cells from hypoxia-induced apoptosis. Overall, these cells are in a state of reduced proliferation, yet show remarkable ability to survive a particularly stressful event. A major assumption of the angiogenic model of tumor dormancy is that low oxygen levels induce this state (Wells et al., 2013); our experimental results rely on low oxygen levels. Our use of low serum cell culture conditions reduced cell proliferation so that total cell number over the course of our experiments remained essentially the same; therefore, like “tumor mass dormancy”, there was no overall growth (Aguirre-Ghiso, 2007). Before our study, ACC1, ACLY and

ETV4 were not implicated in tumor dormancy; however, some histone demethylases have been implicated in dormancy-like characteristics (See section 5.3).

174

Applying this proposed model to our data, cancer cells treated with ACLY or

ACC1 inhibitors may not proliferate due to blocked lipogenesis, but also may survive in hypoxic regions. As these cells resist death under stresses, they may become the

“dormant” cells that have the potential to recur after treatment regimens end. However, these hypoxia-tolerant and persisting cells may also become more targetable if we block these cells’ stress survival adaptive mechanisms. While data to prove such a model remains necessary, this study suggests that the ACLY-ACC1-ETV4 axis might mediate the balance between oncogenesis and stress survival under hypoxia. As discussed, similar mechanisms of balance between proliferation and stress survival exist in a wide variety of biological contexts, but have not been deeply explored in cancer biology.

5.2 Implications of elevated α-ketoglutarate in hypoxia with inhibition of ACC1 or ACLY

5.2.1 Source of elevated α-ketoglutarate in hypoxia with inhibition of ACC1 or ACLY

The observed elevated α-KG levels in the shACLY and shACC1 cells under hypoxia could be due to either an increased generation or a decreased consumption of this metabolite. A portion of α-KG is consumed to fuel lipogenesis, which is further promoted under hypoxia by reductive carboxylation of glutamine (Metallo et al., 2012;

Wise et al., 2011). Therefore, the simplest explanation is that blocking lipogenesis under hypoxia may lead to a “build-up” of upstream metabolites, including α-KG. However,

175

since lipogenesis can also be decreased under hypoxia due to the inactivation of ACC1 by AMPK, other sources of the elevated α-KG could be considered.

As a critical metabolite in the tricarboxylic acid (TCA) cycle, α-KG is both generated and consumed as the TCA cycle flux occurs. Under hypoxia, reduced flux into the TCA cycle occurs as cytosolic glycolysis is favored, often resulting in a decrease of

TCA intermediates, just as we saw in our metabolomics data (Figure 21). The influx of metabolites to the TCA cycle is inhibited largely through the HIFs, and my data showed reduced HIF levels with loss of ACC1 or ACLY. Therefore, influx to the TCA cycle could be maintained, but the reduced oxygen levels would still prevent the efficient generation of ATP through the electron transport chain. Therefore, without the oxygen acceptor to allow the ETC to proceed, a “build-up” of TCA cycle-generated α-KG could also explain the increased α-KG levels. However, we did not see an elevation of all TCA intermediates with ACC1/ACLY depletion under hypoxia, so this is a less likely explanation of the α-KG elevation.

In addition to differential cell metabolism, α-KG levels could be altered under hypoxia through differential consumption by the 2-oxoglutarate dependent dioxygenases (2-OGDDs). This class of enzymes will be discussed in detail in Section 5.3.

Generally, these enzymes utilize α-KG (2-oxoglutarate), iron and molecular oxygen to modify many proteins, which results in a number of functional consequences. Therefore, elevated α-KG levels could be due to overall reduced activity of the 2-OGDDs.

176

Considering that many of these enzymes are induced by hypoxia (Melvin and Rocha,

2012), and our data showed patterns consistent with their increased function (see Section

4.7), this also seems like an unlikely explanation for elevated α-KG under hypoxia with

ACC1 or ACLY inhibited.

Since our metabolomics data is only a snapshot of the stable levels of cellular metabolites rather than their flux through metabolic pathways, it is impossible to tell from our data if the elevated α-KG is due to decreased consumption or increased generation. Regardless, since the supplementation of α-KG recapitulated the phenotypes of ACC1 and ACLY silencing, elevated levels of α-KG were important to maintaining cell survival under hypoxia.

5.2.2 α-ketoglutarate implications for stress response vs. oncogenesis

Here, our data indicated that α-KG can affect cellular survival under hypoxia by regulating ETV4 expression, possibly through 2-OGDDs and epigenetic mechanisms.

Other studies have also implicated increased α-KG levels as protective to cancer cells under stress. In a model of glucose deprivation, sustained or extreme stress triggered an

ERK2-mediated signaling cascade that induced apoptosis; α-KG was elevated when this pathway was inhibited pharmacologically (Shin et al., 2015). Sufficient α-KG levels are essential for reductive carboxylation to generate citrate and lipids to maintain cellular metabolism under hypoxia (Wise et al., 2011; Metallo et al., 2012), although the relative flux through reductive versus oxidative metabolism is still debated (Fan et al., 2013). 177

Our model of proliferation vs. stress survival predicts that increased levels of α-

KG are anti-tumorigenic and likewise, decreased α-KG is pro-tumorigenic. Elevated α-

KG inhibited angiogenesis and reduced Lewis lung cancer tumor growth in vitro and in vivo (Matsumoto et al., 2009). Where elevated levels of α-KG induced apoptosis (in conflict with our data), increased α-KG levels were still anti-tumorigenic either through

PHD3 (Tennant and Gottlieb, 2010) or through activation of HIF-1α as a result of decreased O-linked glycosylation (Ferrer et al., 2014). Further metabolomics and metabolic flux experiments of cancer cells under stress will continue to inform our understanding of the importance of this metabolite to stress survival and tumorigenicity.

5.2.3 Implications of α-ketoglutarate results with the “oncometabolite” 2-hydroxyglutarate (2-HG)

The interactions between α-KG and 2-hydroxyglutarate (2-HG) and their distinct effects on oncogenesis are also consistent with our model of balanced stress survival and proliferation. 2-HG is generated when isocitrate dehydrogenase (IDH) is somatically mutated to gain a novel enzymatic function, producing 2-HG instead of α-KG (Dang et al., 2009). Therefore, the production of these two metabolites is mutually exclusive depending on the cells’ IDH mutation status. Additionally, 2-HG is a competitive inhibitor of α-KG so its accumulation prevents dioxygenase enzymes from using α-KG

(Xu et al., 2011; Chowdhury et al., 2011). Thus, α-KG and 2-HG are in competition with each other for functional outcomes. Consistent with our prediction that decreased α-KG is worse for tumors, increased levels of 2-HG (reduced α-KG levels) promote tumor 178

growth and are found in more advanced cancer patients (Dang et al, 2009; Wang et al.,

2013; Terunuma et al., 2014). However, this is a cancer-specific effect since the opposite is true in gliomas: the presence of IDH mutations and 2-HG is better for prognosis (Yan et al., 2009; Guo et al., 2011b). The applicability of α-KG being better for patients is likely limited to those cancers where 2-HG acts as an onco-metabolite. Under proliferative conditions, hypoxia can promote the production of L-2-HG (Intlekofer et al., 2015).

Therefore, phenotypes of the “oncometabolite” 2-HG also support our model and hypothesis that elevated α-KG is associated with decreased oncogenic potential.

5.3 Potential role of 2-OGDDs in hypoxia survival and implications for cancer therapies

In addition to the importance of α-KG in the TCA cycle, our experiments showed that elevated α-KG could alter histone methylation patterns, likely via α-ketoglutarate- dependent dioxygenases, to regulate ETV4. While causal relationships predicted by this hypothesis were not tested by rigorous genetic manipulations, our data was consistent with such a model. Previous reports suggested that α-KG/succinate ratios determined the direction and activities of dioxygenases (Carey et al., 2015; Kaelin, 2011), yet our data indicated that it was increased α-KG, not succinate or α-KG/succinate, that drove the epigenetic changes, ETV4 repression and hypoxia survival phenotypes. The implications of the source and potential role of α-KG in oncogenesis was the topic of Section 4.1.2, so

179

here I focus on the implications of its potential role in modulating the activity of 2- oxoglutarate dependent dioxygenases (2-OGDDs) to promote hypoxia survival.

As their substrate, α-KG affects the enzymatic activities of the 2-OGDDs, which are involved in many different cellular functions (Loenarz and Schofield, 2008). In addition to the histone demethylases that likely regulated ETV4 in our experiments, 2-

OGDDs are involved in hypoxic signaling (Loenarz and Schofield, 2008). The HIFs are negatively regulated by the 2-OGDD PHD proteins and FIH (factor inhibiting HIF) through hydroxylation (Majmundar et al., 2010; Kaelin, 2011). Additionally, the hydroxylation of EGF by EGFH affects signaling through this pathway under hypoxia

(Dinchuk et al., 2002). Higher levels of α-KG under hypoxia with ACC1 or ACLY could lead to more active 2-OGDDs. Increased α-KG available for the PHDs and FIH would lead to the decreased HIF expression we saw with loss of ACC1 or ACLY (Figure 13).

Interestingly, the hydroxylation of ASB4 by FIH allows for protein degradation in an oxygen dependent manner (Loenarz and Schofield, 2008) and the closely related protein,

ASB2 was a hypoxia lethal multiple hairpin hit in our genome-wide screen. Thus, our data is consistent with increased activity of the hypoxia-regulating 2-OGDDs with ACC1 or ACLY silencing under hypoxia.

With respect to the regulation of ETV4 under hypoxia, we focused on the 2-

OGDD histone demethylases because changes in their function could explain the altered

RNA abundance of ETV4 (Figure 16). Since multiple histone methylation marks

180

decreased with depletion of ACC1, ACLY or α-KG treatment (Figure 22 and Figure 24), and histone demethylases act upon different lysine residues (Loenarz and Schofield,

2008), under these conditions, many demethylases could have elevated activity.

Our data showed that H3K4me3 abundance was the most dramatically and consistently affected by depletion of ACC1, ACLY or α-KG treatment; all three conditions reduced the abundance of H3K4me3 globally and at the ETV4 promoter. The

JARID1 (KDM5) family of 2-OGDD histone demethylases specifically demethylates

H3K4me2/3 (Blair et al., 2011; Johansson et al., 2014). Therefore, we speculate that KDM5 family members are the most likely candidates to affect the abundance of this mark at the ETV4 locus and subsequently affect ETV4 activity. These enzymes range in their expression and activity by cell type and are differentially influenced by oxygen levels

(Mimura et al., 2011; Pollard et al., 2008; Shmakova et al., 2014; Yang et al., 2009).

KDM5B has been reported as a direct HIF-1α target that is upregulated in response to hypoxia (Xia et al., 2009). KDM5A and KDM5B are more strongly implicated in cancer biology, promoting the proliferation of breast cancer cells (Hou et al., 2012; Catchpole et al., 2011), regulating oxidative stress-responsive genes (Liu et al., 2014), drug tolerance

(Hou et al., 2012; Sharma et al., 2010; Roesch et al., 2013; Kuo et al., 2015; Wang et al.,

2015) and stem-like characteristics (Kuo et al., 2015). Overexpression of KDM5C is a negative prognostic feature in prostate cancer (Stein et al., 2014) and it promotes tumor formation through interactions with VHL, thus connecting it to hypoxic biology (Niu et

181

al., 2012). While different splice isoforms of KDM5D may have cancer relevance

(Jangravi et al., 2015), it has not been functionally studied in cancer models. This literature shows that these enzymes can influence both proliferation and characteristics of dormant tumor cells. Based on these observations, ACC1/ACLY could influence tumor dormancy indirectly through α-KG, histone demethylases and their effects on gene expression. However, their oncogenic functions have made the KDM5 family of

H3K4 demethylases targets of cancer drugs (Blair et al., 2011; Mannironi et al., 2014;

Sayegh et al., 2013; Rasmussen and Staller, 2014; Johansson et al., 2014). Our data was consistent with increased histone demethylase activity being anti-tumorigenic, but also with the possibility that their activity could promote a dormant-like stress survival phenotype. Context likely determines the oncogenic or stress resistance effects of these enzymes’ activity. Our data suggests that the effects of hypoxia on these drugs should be considered. A full investigation into the mRNA, protein, activity and targets of these histone regulators with ACC1/ACLY loss and under hypoxia will be necessary to determine the extent to which each plays a role in regulating ETV4. If none of the KDM5 family members affect ETV4 function, then a more unbiased approach through a targeted RNAi screen of a broader set of histone demethylases or 2-OGDDs could help to determine the specific regulator of ETV4 in this context.

While there are no histone methyltransferases currently known to use α-KG as a substrate, several histone methyltransferases are regulated by hypoxia (Melvin and

182

Rocha, 2012), so we do not exclude the potential for changes in histone methylation patterns to be due to changes in both methyltransferase and demethylase enzyme activities. Additionally, the changes in global histone methylation patterns very likely affect a number of other genes’ expression patterns, in addition to the changes we see at the ETV4 locus. Other changes elicited by 2-OGDDs also potentially contribute to increased cell survival under hypoxia with loss of ACC1 or ACLY.

5.4 Implications for the role of ETV4 in cancer biology and therapeutics

Transcriptional adaptation to hypoxia is most often orchestrated by the HIFs.

However, here we show that cancer cells’ hypoxic survival can be mediated by a different transcription factor, ETV4. Our data reveal that ETV4 is critical for hypoxia- induced apoptosis. Although this regulation was mostly specific to ETV4, we do not fully exclude the possibility that other ETS transcription factor family members also contribute to an apoptotic, hypoxic transcriptional program. ETV4 was previously proposed as an essential co-activator of HIF-1α that mediates a hypoxic transcriptional program (Wollenick et al., 2012). Another ETS family transcription factor, ETS1, is a hypoxia-responsive gene that interacts with HIF-1α, but the other PEA3 transcription factors have not been previously implicated in oxygen responses. The levels of ACC1 or

ACLY influencing α-KG and ETV4 under hypoxia provide a link between lipogenesis, a

TCA cycle intermediate and transcription. Therefore, ETV4 mediates the transcriptional 183

response to varying degrees of active lipogenesis caused by changing ACC1 and ACLY levels. Our gene signature analyses suggested that this regulation was preserved in human tumors in vivo.

While the global and local ETV4 epigenetic changes we describe herein were consistent with decreased activity of ETV4, our evidence provides correlative support for reduced promoter activity and does not provide a causal explanation for ETV4 repression. Silencing and overexpression of different 2-OGDD enzymes can determine which of the candidate enzymes are responsible for our observed epigenetic and gene expression changes (discussed in Section 5.3). Recently, the CRISPR system was adapted for gene activation through manipulation of acetylation of histone H3 lysine 27 at any desired locus in human cells (Hilton et al., 2015). Modifying this technique to alter histone methylation events at the ETV4 locus would allow us to determine the causal relationship between histone modifications at the ETV4 promoter and a decrease in

ETV4 expression under hypoxia with loss of ACC1 or ACLY. Overall, our data establish

ETV4 as one critical factor that influences hypoxic cell survival and transcriptional responses downstream from ACC1 and ACLY, and reveal the metabolite α- ketoglutarate as a molecular link between metabolic and transcriptional adaptation to hypoxia.

Since our data shows that the loss of ETV4 prevents hypoxia-induced apoptosis, its presence must be necessary for initiating or eliciting the apoptotic response. Hypoxia

184

can induce multiple mechanisms of apoptosis (Greijer and van der Wall, 2004;

Hammond and Giaccia, 2005). While this is not interrogated in detail here, additional data suggested that ETV4 was responsible for enacting a caspase 8-dependent mode of extrinsic apoptosis (Fulda and Debatin, 2006), possibly by modulating NF-kB activity.

This hypothesis requires much additional experimental evidence to be conclusive.

Overall, our results suggest that lipogenic inhibitors that block ACC1 or ACLY may be particularly effective for tumors driven by ETV4. Since transcription factors are notoriously difficult to target with small molecules, these enzymes potentially offer a different route to prevent the activation of ETV4. Likewise, this reduction of the oncogenic driver ETV4 may account for portions of the therapeutic potential of lipogenic inhibitors. Somewhat uniquely, a small molecule (YK-4-279) has been developed to target ETS fusion genes (Erkizan et al., 2009) and it prevents ETV1-driven tumor growth and metastasis in prostate cancers (Rahim et al., 2011 and 2014). I encourage continued investigation into the consequences of lipogenic or ETV4-targeted molecules on cell invasion, mobility, tumor growth and further analysis of their effects under hypoxia.

5.5 Implications for inhibitors of ACC1 and ACLY in cancer therapies

Inhibitors that target ACLY and ACC1 have been proposed as cancer therapeutics (Currie et al., 2013; Migita et al., 2008). ACC1 has been targeted based on its role encoding the rate-limiting enzyme in lipogenesis, a process critical for cancer cells’ 185

rapid proliferation (Currie et al., 2013). For ACLY, not only is it a critical enzyme for active lipogenesis, but it also regulates epigenetic states by generating acetyl-CoA and integrates glucose and lipid metabolism. These essential roles offer multiple mechanisms to therapeutically target tumor biology (Wellen et al., 2009; Zaidi et al., 2012). Our data provides several novel insights on the biological effects of ACC1 and ACLY in hypoxic cancer cells that should be considered when targeting these enzymes.

First, while we expect ACLY and ACC1 inhibition to have opposite effects on levels of acetyl-CoA, their effects on gene expression and hypoxia survival were highly similar. This suggests that levels of acetyl-CoA (and subsequent histone acetylation or epigenetic changes) may not readily explain the majority of gene expression responses to the inhibition of ACC1 or ACLY under hypoxia. Instead, the ACC1/ACLY-induced reduction in ETV4 levels and activity seemed to account for a significant portion of the hypoxic transcriptional changes. We show that α-KG can mediate multiple downstream events of these two enzymes under hypoxia, especially the regulation of ETV4. We do not rule out that effects we did not investigate also contribute to survival with ACC1 or

ACLY loss. Possibly, targeting of either of these enzymes may be sufficient to achieve relevant clinical results. These data support an expectation of similar responses to therapeutically targeting ACC1 or ACLY and emphasize the importance of targeting the early steps in de novo lipogenesis for modulating cancer cell hypoxic survival.

186

While ACC1 and ACLY had similar effects on phenotypes and gene expression under hypoxia, loss of ACC1 or ACC2 had different effects on cell survival under hypoxia. The metabolite pools generated from each of these isoforms can have different functions: the malonyl-CoA generated by ACC1 feeds cytosolic lipogenesis and the malonyl-CoA generated by ACC2 at the outer mitochondrial membrane inhibits CPT1 and β-oxidation (Wakil and Abu-Elheiga, 2009). Our data imply a possible difference in the importance of fatty acid synthesis versus oxidation under hypoxia, with cytosolic synthesis (through ACC1) having a greater role in mediating hypoxic cell survival.

These data also support the concept that the pools of metabolites from each of these enzymes are functionally distinct (Wakil and Abu-Elheiga, 2009). The different phenotypes of ACC1 and ACC2 under hypoxia suggest that inhibitors with isozyme specificity will be important to develop and consider in regards to cancer treatments

(Tong and Harwood, 2006).

Targeting fatty acid synthesis in cancer is not a new concept. ACC, ACLY and

FASN, as well as their upstream transcriptional activators the SREBPs have all been considered targets for cancer cells and studied extensively in vitro and in mouse models

(Currie et al., 2013). Inhibitors of ACLY, as mentioned, could have pleiotropic effects on cancer cell metabolism, and so should be considered with caution without a complete understanding of their effects in vivo. Citrate analogs inhibit ACLY, but the small molecule SB-204990 probably has better clinical potential; the effects of this drug under

187

hypoxia are not known. ACLY inhibition must contend with multiple avenues of acetyl-

CoA generation in the cell (Zaidi et al. 2012); our data also suggests α-KG generation may be relevant to targeting ACLY. The importance of developing drugs to selectively inhibit ACC1 or ACC2, depending on the circumstance, complicates targeting the ACCs.

While their affinities for acetyl-CoA and citrate are different enough to suggest that selective inhibition is possible, it will still be challenging considering their structural and sequence similarity (Tong and Harwood, 2006). Both naturally occurring and chemically synthesized inhibitors of ACC have been developed (Zu et al., 2013), but these direct inhibitors have not been studied under hypoxia or stress. In the substantial literature suggesting the therapeutic potential of targeting this pathway in cancer, I cannot find evidence of clinical trials specifically targeting ACC1, ACC2 or ACLY in cancer. When specific, efficient drugs can target these enzymes, it will be important to understand how hypoxia influences their in vivo behavior and efficacy.

In addition to hypoxia, our model suggests that the effects of lipogenic inhibitors on dormant cells should also be considered. Since re-activation of dormant cells in order to kill them is risky, targeting dormant cells’ stress survival adaptations is a preferred strategy for killing dormant cells (Sosa et al., 2014). If lipogenic inhibitors affect tumor dormancy, their use could impact cancer recurrence and metastasis, the leading cause of cancer-related deaths (Wang and Lin, 2013; Sosa et al., 2014). Therefore, targeting lipogenesis in cancer may need to be combined with other therapeutic approaches that

188

target hypoxic regions (such as hypoxia-activated pro-drugs), to eliminate the cancer cells that may be dormant and protected in the hypoxia-niche. Consistent with past literature and clinical attempts to target these genes, our model and data advise that treatment regimens be carefully considered to achieve an optimal balance between blocking proliferation and potentiating tumor cell stress survival.

5.6 Normoxic post-translational regulation of ETV4 by ACC1 or ACLY

This section details a potential future line of investigation that stems from the data gathered during my investigation of ACC1, ACLY and ETV4 under hypoxia. While my dissertation work answered a number of questions about cancer cell adaptation under hypoxia, it has also opened up additional questions. Here, I outline one of these outstanding questions and a hypothesis and potential experiments to address it.

While we noticed a hypoxic regulation of ETV4 through α-KG, as detailed in

Chapter 4, we also noticed a normoxic regulation of ETV4 by ACC1 and ACLY. The loss of ACC1 or ACLY led to a decrease in the protein, but no consistent change in the RNA, of ETV4 under normoxia (Figure 16b, d). This result suggested additional post- transcriptional relationships between these proteins and begged the question, what is the nature of normoxic regulation between ACC1, ACLY and ETV4? Below, I outline a hypothesis regarding this regulation, the literature background and my data that 189

supports this hypothesis and what questions remain to be answered through experimentation. I began the work on these experiments and they are ongoing by Po-

Han Chen in the laboratory.

Hypothesis: The loss of ACC1 promotes COP1-mediated ubiquitination and degradation of ETV4.

Justification: By western blot, protein levels of ETV4 were reduced with genetic depletion of ACC1 or ACLY (Figure15b, d). In addition to the hypoxic data already presented, we conducted microarrays of the shETV4, shACC1 and shACLY cells under normoxia. Analysis of these arrays showed a similar finding as under hypoxia: there was a highly similar global transcriptional response to the depletion of these three genes.

This extensively similar transcriptional profile suggests that these proteins are functionally related under normoxia as well.

The E3 ligase COP1 targets both ACC1 and ETV4 for degradation (Qi et al., 2006;

Vitari et al, 2011; Baert et al., 2010), thus regulating the abundance of these two proteins.

For ACC1, this regulation occurred under fasting conditions, which may be relevant to our low serum growth conditions (Qi et al., 2006). These published studies were done in other cell types, but I have confirmed that in H1975 cells, the siRNA-mediated depletion of COP1 leads to increased protein levels of ETV4. Additionally, MG132 treatment of

H1975 leads to robust increase in ETV4 protein abundance. These two pieces of data suggest that ETV4 may also be a target of COP1 for proteasomal degradation in H1975

190

cells. Therefore, we hypothesize that the loss of ACC1 promotes COP1-mediated ubiquitination and proteasomal degradation of ETV4. To test this hypothesis, I outline four questions to address experimentally:

Question 1: Is the degree or pattern of ETV4 ubiquitination affected by the loss of

ACC1 or ACLY?

If loss of ACC1 or ACLY alter ETV4 protein levels due to COP1-mediated degradation, ETV4 is expected to have increased levels of ubiquitination in the ACC1 or

ACLY knockdown cells. To test this, I first suggest an immunoprecipitation of ETV4 from the control and ACC1/ACLY knockdown cells, followed by blotting with a pan- ubiquitin or lysine 48-linked ubiquitin antibody. Lysine 63 linked ubiquitin could also be investigated, but would suggest different fates for ETV4 than degradation (Weissman,

2001). Mass spectrometry of ETV4 after its immunoprecipitation could also test this possibility that ACC1/ACLY silencing alters ETV4 protein modifications. I have immunoprecipitated ETV4, but I have not been able to detect ubiquitin modifications from the pulled down ETV4, even though the input lysates show increased ubiquitin in

MG132-treated cells. Therefore, more technical optimization and inclusion of positive controls will be needed to address this question. Alternatively, there may not be differences in ETV4 ubiquitination with depletion of ACC1, in which case a decrease in the stability of ETV4 protein should be investigated since this could also explain decreased abundance of the protein.

191

Question 2: What is the pattern of posttranslational modifications on ETV4 with the loss of ACC1 or ACLY?

There is considerable precedence for the posttranslational modifications of ETV4 and its PEA3 family members to affect their stability and activities, resulting in altered transcriptional output (see Section 4.1.3). These posttranslational modifications could be altered in ACC1 or ACLY depleted cells to affect ETV4 abundance and activity.

Therefore, regardless of the ubiquitination pattern of ETV4, changes in the other posttranslational modifications of ETV4 could also explain decreased protein stability and functionality with loss of ACC1 or ACLY. Similar to the experiments outlined for

Question 1, I suggest two approaches to investigate the status of posttranslational modifications on ETV4. The first is to immunoprecipitate ETV4 and blot with antibodies that detect different modifications (serine/threonine phosphorylation, tyrosine phosphorylation, acetylation, malonylation etc.). The second approach is to use mass spectrometry to identify the altered modifications, as well as the particular modified residues. Mass spectrometry experiments could be better designed if the immunoprecipitation-blotting strategy identified one or multiple candidate modification(s), but could also be run sequentially to discover different modifications if the blotting technique does not work. If posttranslational modifications on ETV4 are affected by the loss of ACC1/ACLY, then site-directed mutagenesis to alter the modified

192

residues and abolish or mimic this modification will determine if it leads to stabilization or increased activity of the protein.

Question 3: Does the depletion of ACC1 or ACLY affect COP1’s regulation of

ETV4?

To address this question, it is necessary to examine the function of COP1 on

ETV4 with ACC1/ACLY loss. First, knockdown COP1 in the ACC1 or ACLY depleted cells and examine ETV4 protein levels. The expectation is that loss of COP1 will restore

ETV4 protein levels in the shACC1/shACLY cells. The RING domain of E3 ligases is important for localizing the substrate and catalyzing the ubiquitination reaction (Zheng et al., 2000). Examining ETV4 protein levels after overexpression of wildtype or mutated

RING domain versions of COP1 will determine if the ubiquitin ligase activity of COP1 is necessary for this regulation. Some preliminary biochemical fractionation experiments suggest that the subcellular localization of ACC1, ACLY, ETV4 and COP1 may influence their regulation. The subcellular localization of these proteins could be determined using a cleaner, but simpler, fractionation protocol (Suzuki et al., 2010) or with confocal microscopy. The levels of COP1 protein should be investigated with ACC1, ACLY or

ETV4 depletion, even though preliminary experiments have not seen changes in overall abundance of COP1 with these genetic manipulations.

Question 4: How does stable ETV4 overexpression sustain dramatic loss of

ACC1?

193

Although addressing this question does not investigate the role of COP1 in the

ACC1-ETV4 interaction, my data suggested an additional unexpected level of regulation from ETV4 on ACC1. In the course of past experiments, we discovered that stable overexpression of ETV4 led to a dramatic and sustained repression of both the mRNA and protein of ACC1, but not ACLY. This down-regulation was apparent even in the shACC1 cells—there was essentially undetectable ACC1 with stable ETV4 overexpression, even as the shACC1 cells lost their stable knockdown over time. ETV4 overexpression was not dominant negative as gene expression and survival phenotypes of ETV4 loss were rescued with this overexpression. Since ETV4 is most often considered a transcriptional activator, it is curious that it mediates such a strong repression of ACC1 at the RNA level. To investigate this regulation of ACC1 by ETV4 overexpression, I would first determine if ACC1 is a direct target of ETV4 through chromatin immunoprecipitation of ETV4 and qPCR of the ACC1 locus. The promoter region of ACC1 could also be searched for canonical PEA3 family binding sites. Since the PEA3 family interacts with many co-factors, this non-canonical role of ETV4 function could be mediated through differential co-factors. While the literature suggests a number of possible interactors, an unbiased approach would be mass spectrometry after immunoprecipitation of ETV4 with overexpressed ETV4 and non-overexpressed ETV4.

Investigating the experiments outlined in this section will elucidate the molecular interactions between ETV4, ACC1 and ACLY and their functional outputs.

194

These studies will help the scientific community better understand this point of interaction between cellular metabolism and transcription.

5.7 Conclusion

Overall, this dissertation work described two parallel RNAi screens that were conducted to investigate the genetic determinants of cancer cell survival under lactic acidosis or hypoxia. Results from these screens and their analysis and validation identified many genes with diverse functions that affect cancer cell survival under these stresses. While a number of these genes remained to be investigated to achieve a mechanistic understanding of their role under stress, I have defined in detail the mechanism underpinning how silencing ACC1 or ETV4 protects cells under hypoxia.

Loss of ACC1, the related enzyme ACLY, or of ETV4 prevented hypoxia-induced apoptosis and triggered highly similar global transcriptional responses under hypoxia.

The similarity of response induced by loss of these three genes suggested they acted in the same pathway. Further analysis revealed that the loss of ACC1 or ACLY elevated levels of the metabolite α-KG under hypoxia which mediated global and local histone methylation changes that could explain the loss of ETV4 protein and activity.

Comparative analysis of the global transcriptional responses with patient gene expression data sets suggested that this regulation is maintained in vivo. There was a conflict of logic between the literature supporting pro-tumorigenic effects of these genes

195

and hypoxia, and the survival under hypoxia with the loss of these genes. Therefore, we proposed a model in which oncogenic programs and stress survival programs are maintained by cancer cells in a mutually exclusive balance dictated by the genetic alterations and tumor microenvironment in which they are growing. This model has implications for tumor cell dormancy and the use of lipogenic inhibitors, such as metformin, in the clinic. It is my intention that the detailed future directions outlined from this work will facilitate elaboration on these findings of cancer cell survival mechanisms and hopefully contribute to improved design of cancer therapeutics.

196

Appendix A. Annotation of Multiple Hairpin Hits from genome-wide pooled shRNA screens.

This Appendix includes a detailed list of the Multiple Hairpin Hits from the

H1975 genome-wide pooled shRNA screen (Table 12). The detailed description of the analysis conducted to generate these top candidate genes is described in Section 3.3.1.

Table 12. List of Candidates Identified in Multiple Hairpin Analysis

Contextual Survival Hits Contextual Lethal Hits Hypoxia (173) LA (86) Hypoxia (123) LA (86) ACACA ACACB ACACB ACTR2 AHCYL1 AMBN ADORA3 AHCYL1 AHNAK ASPH AFF3 ARHGEF16 ANKRD17 C10orf28 APOL1 ARMETL1 AP1S1 C10orf79 ARIH1 ARNTL AR C20ORF112 ASB2 ATP6V0A4 BEX4 C20orf177 ASNA1 BCL2L1 BIRC4 C9orf32 ATF1 BEST4 BNC2 CACNA1C ATP13A5 C18orf8 BRD3 CBLN4 AUP1 CA9 BSN CCDC7 AZGP1 CD53 C13orf33 CDH1 BARD1 CEP55 C1orf101 CLASP2 BCYRN1 CLCA1 C1orf163 CLIC2 BRMS1L COL27A1 C1orf32 CNBP BTN2A2 CSTA C6orf12 CPO C12ORF27 CXorf34 C6orf128 CRMP1 C1orf51 DC2 C6orf167 DAND5 C2orf30 DDHD2 C7orf54 DDR1 C6orf184 DDI2 C9orf75 DEFB107A CAND1 DYRK3 CCDC74A DEFB118 CASP6 EMID2 CD6 ECH1 CCDC110 FAM3B CLEC4D ELP4 CCDC28A FANCA CLEC4G ESR2 CCDC92 GEMIN5 CNIH EXOSC6 CDS1 GLI1 CNOT10 EXOSC8 COPS6 HCG2P7

197

CNTROB FBN1 CPLX2 HECTD1 COL4A3BP FBXO33 CSF1 HSP90AB1 CPLX2 FLJ11235 CXorf34 ICHTHYIN CPOX FOXF1 DKFZP547G183 IL12B CSN3 FST DKFZP779B1634 ITGA8 CUL2 GABPB2 DNAJC16 KCNC2 CXCL12 HARS DSC2 KIAA1377 DDA1 HIP2 E2F5 KLHDC5 DDX56 HIRA EPAS1 LOC126826 DEFB107A HIVEP2 ERRFI1 LOC401898 DHTKD1 IL13RA1 EXOSC8 LOC646879 DIDO1 IL1RAP FAHD2A LOC730273 DKFZP566F0947 INTS9 FANCA LONRF3 DNM1L KIF7 FBXL6 LRRC41 DYNLRB1 KRTAP8-1 FGFR1OP2 MATR3 E2F6 LEF1 FLRT2 MLF1 EDG2 LOC285638 FMO5 MPO EIF3A LOC402420 FRAS1 NEDD4L EIF4G2 LOC441898 GRPEL2 NOD1 EML1 LOC442308 HIST3H2BA NR1I2 EPHA7 LOC644558 HOXB4 NR4A3 ERN1 LOC646048 HSBP1 OR2AG1 ESRRB LOC652818 HYDIN PDE1A ETV4 LOC729324 IKBKE PIGP F7 LOC93463 IRF2 PMCH FAM120C MGC16275 JARID1D PNOC FAM12CP NR0B2 KCNH4 PROZ FAM20B NR1I2 KIAA0528 PSIP1 FAT1P1 NR2E3 LEMD1 PTPRE FAUP1 NR5A1 LOC130355 RAB3IP FBXL13 NR6A1 LOC642609 RASGRP3 FBXL7 NRIP1 LOC642929 RNF14 FBXW12 OR4E2 LOC644006 RP11-49G10.8 FBXW7 OR9G4 LOC731751 RPL34P2 FLJ12056 P2RX2 LRRC52 SART1 FLJ20699 PARVB MAP3K7 SCGB1A1 FLJ21062 PDE11A MARVELD2 SESN3 FNDC4 PDXDC1 MAST1 SF3B14 FUSIP1 PIN4 MICAL1 SKP1 GABRB2 PKIG MOXD1 SLC12A6 GBF1 POLR1B MYBL1 SLC12A9 GTF2E2 PPM1K NFATC2 SLC26A4 HDAC9 PPP1R8P NMI SLC6A8 198

HEPH PRDX3 NR2F2 STK39 HIPK4 PRNT NSD1 TGDS HIST1H2BA RNF139 OBSCN THOC2 HIST2H2AB RRP15 OSTalpha TMTC1 HIST4H4 SGCB PAG1 TRBV24-1 HSFX1 SKP1 PCDH9 TRIM26 HTR1E SLC9A9 PEA15 TRIM6 HUS1 SMURF1 PPP2R5C UBE2H IGKDEL SNORA67 PRKAR1A UBE3C IGKV3OR2-5 SYT6 PRKCH USP48 IL3RA TIRAP PSMA1 UTP11L INMT TRPC4AP PTCH1 VAMP5 KBTBD3 UNQ6125 PTGFR WHSC1L1 KCNG3 USH1C PTPRJ WWOX KLHL15 VMD2L3 PTPRK ZFP37 LANCL3 WIT1 R3HDM1 ZNF571 LOC100101120 ZFP57 RAB23 ZNF615 LOC146053 RAPGEF1 LOC148145 RASA1 LOC158435 RBM20 LOC284701 RBM35A LOC285943 RNF14 LOC339240 RPL28 LOC387978 RPP30 LOC388588 SFRS1 LOC388946 SFRS2 LOC389892 SIGIRR LOC653414 SIL1 LOC730273 SIP1 LOXL2 SIX4 MAP4K4 SKIV2L2 MED26 SLC25A3 MTHFD2 SLC25A37 MYO9A SLC26A2 NCKAP1L TDRKH NMU TES NR1H4 TM2D3 NR2C1 TMEM117 NR3C2 TNS4 NR5A1 TRAPPC2 NUDT13 TRIM24 NUDT16 TRIM9 NXN TRMEP1 199

OR1A1 TSSC1 OR1B1 TUG1 OR51F2 UBXD5 OR5AC2 UTP11L PDPR WDR76 POFUT1 XKR6 POLD1 ZDHHC2 PRDX3 ZFP14 PROK1 ZNF165 PSG6 ZNF510 PTK7 ZNF71 PTPRF RBM33 RDH12 RFFL RIOK2 RNU1P8 RNU7P4 RORA RPS14 RSPH1 SAR1A SEC23A SHMT2 SLAIN1 SLC25A14 SLC30A8 SLC40A1 SLC5A5 SLC9A3 SLCO4C1 SOAT2 SOX5 SRF ST7L ST8SIA4 STAU1 STOM SYT6 TESC THAP11 TLOC1 TMCC3 200

TMEM136 TNFSF18 TOP2B TRBV24-1 TREX1 TRIP12 TSR2 TTF1 TXNL4B UNQ6125 USH1C XKR8 YWHAZ ZFP42 ZNF12 ZNF439 ZNF514 ZNF571

201

Appendix B. The potential gender bias of ETV4 mRNA up-regulation in TCGA datasets

The Cancer Genome Atlas project makes a huge amount of data on cancer patients’ tumors publicly available. During my investigations of ETV4, I combed TCGA data and found a statistically significant up-regulation of ETV4 mRNA in the gender for which the incidence of cancer occurrence was more common (Table 13, Table 14). These findings included altered ETV4 expression in bladder urothelial carcinoma, head and neck carcinoma, colorectal adenocarcinoma, lung squamous and adenocarcinoma, and thyroid cancers. Table 14 includes data from all of the studies I analyzed, both with and without the correlation. Table 13 includes the studies that showed gender-biased ETV4 up-regulation and the relation to gender-prevalence. The null hypothesis tested was that the probability of ETV4 mRNA up-regulation was equal in men and women (expected frequencies of each gender =0.5). I did not examine cancers from gender-specific organs, for obvious reasons of bias. Studies with less than 10 patients with ETV4 up-regulated were also not included. These data are not corrected for differential gender frequency in the populations studied. As of the date of analysis (3/27/2015), many studies in TCGA did not have mRNA regulation data, so it was not possible to analyze them for this trend. As the TCGA database continues to grow, these analyses could be extended to additional datasets.

202

As with many documented gender associations in cancer, the meaning and cause of this preliminary finding for ETV4 remains at large (Dorak and Karpuzoglu, 2012). Of course, gender can dictate the presence or absence of cancers of specific reproductive organs for each sex. Years of studies show that the overall risk of developing cancer is higher for men, and this is commonly thought to be due to differential environmental factors (diet, smoking, occupational preferences) (Dorak and Karpuzoglu, 2012).

However, since the same trend occurs for childhood cancers, there are likely underlying genetic factors affecting cancer susceptibility (Dorak and Karpuzoglu, 2012). ETV4 expression has only been tangentially linked to gender. ETV4 expression was associated with COX2 expression and COX2 expression was statistically correlated with gender in colorectal cancers (Nosho et al., 2005). The cancers with a gender-specific bias do not overlap well with the organs for which ETV4 plays a role in normal physiology (Oh et al., 2012). While this potential relationship requires more rigorous statistical testing in larger datasets to be substantiated, there is a potential differential expression pattern of

ETV4 associated with gender occurrence in patients with multiple cancer types.

203

Table 13. Summary of increased ETV4 mRNA expression in different cancer types, considering gender

Gender with Gender with TCGA Cancer and Study Chi squared skewed increased Name p-value expression prevalence Bladder Urothelial Carcinoma-TCGA 0.0325 Male Male Provisional Head & Neck Carcinoma 0.000157 Male Male TCGA Provisional CRC Adenocarcinoma- 0.0339 Male Male TCGA Nature Lung Squamous-TCGA 0.000239 Male Male Provisional Lung Adenocarcinoma- 0.0833 Maybe Female Female TCGA Nature Lung Adenocarcinoma- 0.0280 Female Female TCGA Provisional Thyroid Carcinoma-TCGA 0.000579 Female Female Provisional Thyroid Papillary 0.00389 Female Female

204

Table 14. Calculating significance of ETV4 mRNA upregulation in TCGA datasets by gender with Chi squared test

# females # males Total Chi TCGA Cancer and with Skewed with high patient squared Study Name high expression? ETV4 considered p-value ETV4 Bladder Urothelial Carcinoma-TCGA 11 3 14 0.0325 Male Provisional Head & Neck Carcinoma TCGA 24 4 28 0.000157 Male Provisional CRC Adenocarcinoma- 7 1 8 0.0339 Male TCGA Nature Lung Squamous- 21 3 24 0.000239 Male TCGA Provisional Lung Maybe Adenocarcinoma- 3 9 12 0.0833 Female TCGA Nature Lung Adenocarcinoma- 11 24 35 0.0280 Female TCGA Provisional Thyroid Carcinoma- 2 17 19 0.000579 Female TCGA Provisional Thyroid Papillary 1 11 12 0.00389 Female Stomach Adenocarcinoma- 15 10 25 0.317 No TCGA Nature CRC-TCGA 13 9 22 0.394 No Provisional CRC Adenocarcinoma- 13 9 22 0.394 No TCGA Provisional Brain, low grade 5 4 9 0.7388 No glioma Glioblastoma 3 3 6 1 No Liver Hepatocellular Carcinoma- TCGA 4 7 11 0.366 No Provisional Skin cutaneous melanoma- TCGA 6 4 10 0.527 No Provisional

205

References

Abdollahi, A., and Folkman, J. (2010). Evading tumor evasion: current concepts and perspectives of anti-angiogenic cancer therapy. Drug resistance updates: reviews and commentaries in antimicrobial and anticancer chemotherapy 13, 16-28.

Abu-Elheiga, L., Brinkley, W.R., Zhong, L., Chirala, S.S., Woldegiorgis, G., and Wakil, S.J. (2000). The subcellular localization of acetyl-CoA carboxylase 2. Proceedings of the National Academy of Sciences of the United States of America 97, 1444-1449.

Abu-Elheiga, L., Oh, W., Kordari, P., and Wakil, S.J. (2003). Acetyl-CoA carboxylase 2 mutant mice are protected against obesity and diabetes induced by high-fat/high-carbohydrate diets. Proceedings of the National Academy of Sciences of the United States of America 100, 10207- 10212.

Adams, D.J., and Morgan, L.R. (2011). Tumor physiology and charge dynamics of anticancer drugs: implications for camptothecin-based drug development. Current medicinal chemistry 18, 1367-1372.

Aguirre-Ghiso, J.A. (2007). Models, mechanisms and clinical evidence for cancer dormancy. Nature reviews Cancer 7, 834-846.

An, J., Muoio, D.M., Shiota, M., Fujimoto, Y., Cline, G.W., Shulman, G.I., Koves, T.R., Stevens, R., Millington, D., and Newgard, C.B. (2004). Hepatic expression of malonyl-CoA decarboxylase reverses muscle, liver and whole-animal insulin resistance. Nature medicine 10, 268-274.

Arber, S., Ladle, D.R., Lin, J.H., Frank, E., and Jessell, T.M. (2000). ETS gene Er81 controls the formation of functional connections between group Ia sensory afferents and motor neurons. Cell 101, 485-498.

Ariffin, A.B., Forde, P.F., Jahangeer, S., Soden, D.M., and Hinchion, J. (2014). Releasing pressure in tumors: what do we know so far and where do we go from here? A review. Cancer research 74, 2655-2662.

Aytes, A., Mitrofanova, A., Kinkade, C.W., Lefebvre, C., Lei, M., Phelan, V., LeKaye, H.C., Koutcher, J.A., Cardiff, R.D., Califano, A., et al. (2013). ETV4 promotes metastasis in response to activation of PI3-kinase and Ras signaling in a mouse model of advanced prostate cancer. Proceedings of the National Academy of Sciences of the United States of America 110, E3506- 3515.

Baert, J.L., Beaudoin, C., Coutte, L., and de Launoit, Y. (2002). ERM transactivation is up- regulated by the repression of DNA binding after the PKA phosphorylation of a consensus site at the edge of the ETS domain. The Journal of biological chemistry 277, 1002-1012.

206

Baert, J.L., Monte, D., Verreman, K., Degerny, C., Coutte, L., and de Launoit, Y. (2010). The E3 ubiquitin ligase complex component COP1 regulates PEA3 group member stability and transcriptional activity. Oncogene 29, 1810-1820.

Balatoni, C.E., Dawson, D.W., Suh, J., Sherman, M.H., Sanders, G., Hong, J.S., Frank, M.J., Malone, C.S., Said, J.W., and Teitell, M.A. (2009). Epigenetic silencing of Stk39 in B-cell lymphoma inhibits apoptosis from genotoxic stress. The American journal of pathology 175, 1653-1661.

Baluk, P., Hashizume, H., and McDonald, D.M. (2005). Cellular abnormalities of blood vessels as targets in cancer. Current opinion in genetics & development 15, 102-111.

Banerjee, H.N., Hyman, G., Evans, S., Manglik, V., Gwebu, E., Banerjee, A., Vaughan, D., Medley, J., Krauss, C., Wilkins, J., et al. (2014). Identification of the Transmembrane Glucose Regulated Protein 78 as a Biomarker for the Brain Cancer Glioblastoma Multiforme by Gene Expression and Proteomic Studies. Journal of membrane science & technology 4.

Barber, M.C., Price, N.T., and Travers, M.T. (2005). Structure and regulation of acetyl-CoA carboxylase genes of metazoa. Biochimica et biophysica acta 1733, 1-28.

Berns, K., Hijmans, E.M., Mullenders, J., Brummelkamp, T.R., Velds, A., Heimerikx, M., Kerkhoven, R.M., Madiredjo, M., Nijkamp, W., Weigelt, B., et al. (2004). A large-scale RNAi screen in human cells identifies new components of the p53 pathway. Nature 428, 431-437.

Bertout, J.A., Patel, S.A., and Simon, M.C. (2008). The impact of O2 availability on human cancer. Nature reviews Cancer 8, 967-975.

Bhalla, K., Hwang, B.J., Dewi, R.E., Ou, L., Twaddel, W., Fang, H.B., Vafai, S.B., Vazquez, F., Puigserver, P., Boros, L., et al. (2011). PGC1alpha promotes tumor growth by inducing gene expression programs supporting lipogenesis. Cancer research 71, 6888-6898.

Bhinder, B., and Djaballah, H. (2013). Systematic analysis of RNAi reports identifies dismal commonality at gene-level and reveals an unprecedented enrichment in pooled shRNA screens. Combinatorial chemistry & high throughput screening 16, 665-681.

Bhola, N.E., Jansen, V.M., Bafna, S., Giltnane, J.M., Balko, J.M., Estrada, M.V., Meszoely, I., Mayer, I., Abramson, V., Ye, F., et al. (2015). Kinome-wide functional screen identifies role of PLK1 in hormone-independent, ER-positive breast cancer. Cancer research 75, 405-414.

Birsoy, K., Possemato, R., Lorbeer, F.K., Bayraktar, E.C., Thiru, P., Yucel, B., Wang, T., Chen, W.W., Clish, C.B., and Sabatini, D.M. (2014). Metabolic determinants of cancer cell sensitivity to glucose limitation and biguanides. Nature 508, 108-112.

Bishop, N.A., and Guarente, L. (2007). Genetic links between diet and lifespan: shared

207

mechanisms from yeast to humans. Nature reviews Genetics 8, 835-844.

Biunno, I., Cattaneo, M., Orlandi, R., Canton, C., Biagiotti, L., Ferrero, S., Barberis, M., Pupa, S.M., Scarpa, A., and Menard, S. (2006). SEL1L a multifaceted protein playing a role in tumor progression. Journal of cellular physiology 208, 23-38.

Blair, L.P., Cao, J., Zou, M.R., Sayegh, J., and Yan, Q. (2011). Epigenetic Regulation by Lysine Demethylase 5 (KDM5) Enzymes in Cancer. Cancers 3, 1383-1404.

Bojovic, B.B., and Hassell, J.A. (2008). The transactivation function of the Pea3 subfamily Ets transcription factors is regulated by sumoylation. DNA and cell biology 27, 289-305.

Boutros, M., and Ahringer, J. (2008). The art and design of genetic screens: RNA interference. Nature reviews Genetics 9, 554-566.

Bras, S., Martin-Lanneree, S., Gobert, V., Auge, B., Breig, O., Sanial, M., Yamaguchi, M., Haenlin, M., Plessis, A., and Waltzer, L. (2012). Myeloid leukemia factor is a conserved regulator of RUNX transcription factor activity involved in hematopoiesis. Proceedings of the National Academy of Sciences of the United States of America 109, 4986-4991.

Brunelle, J.K., Santore, M.T., Budinger, G.R., Tang, Y., Barrett, T.A., Zong, W.X., Kandel, E., Keith, B., Simon, M.C., Thompson, C.B., et al. (2004). c-Myc sensitization to oxygen deprivation-induced cell death is dependent on Bax/Bak, but is independent of p53 and hypoxia-inducible factor-1. The Journal of biological chemistry 279, 4305-4312.

Brunelli, M., Bria, E., Nottegar, A., Cingarlini, S., Simionato, F., Calio, A., Eccher, A., Parolini, C., Iannucci, A., Gilioli, E., et al. (2012). True 3q chromosomal amplification in squamous cell lung carcinoma by FISH and aCGH molecular analysis: impact on targeted drugs. PloS one 7, e49689.

Brusselmans, K., De Schrijver, E., Verhoeven, G., and Swinnen, J.V. (2005). RNA interference- mediated silencing of the acetyl-CoA-carboxylase-alpha gene induces growth inhibition and apoptosis of prostate cancer cells. Cancer research 65, 6719-6725.

Bryant, H.E., Schultz, N., Thomas, H.D., Parker, K.M., Flower, D., Lopez, E., Kyle, S., Meuth, M., Curtin, N.J., and Helleday, T. (2005). Specific killing of BRCA2-deficient tumours with inhibitors of poly(ADP-ribose) polymerase. Nature 434, 913-917.

Butcher, D.T., Alliston, T., and Weaver, V.M. (2009). A tense situation: forcing tumour progression. Nature reviews Cancer 9, 108-122.

Cai, X., Hagedorn, C.H., and Cullen, B.R. (2004). Human microRNAs are processed from capped, polyadenylated transcripts that can also function as mRNAs. RNA (New York, NY) 10, 1957- 1966.

208

Calorini, L., Peppicelli, S., and Bianchini, F. (2012). Extracellular acidity as favouring factor of tumor progression and metastatic dissemination. Experimental oncology 34, 79-84.

Canaani, D. (2014). Application of the concept synthetic lethality toward anticancer therapy: a promise fulfilled? Cancer letters 352, 59-65.

Cardone, R.A., Casavola, V., and Reshkin, S.J. (2005). The role of disturbed pH dynamics and the Na+/H+ exchanger in metastasis. Nature reviews Cancer 5, 786-795.

Carey, B.W., Finley, L.W., Cross, J.R., Allis, C.D., and Thompson, C.B. (2015). Intracellular alpha- ketoglutarate maintains the pluripotency of embryonic stem cells. Nature 518, 413-416.

Carey, H.V., Andrews, M.T., and Martin, S.L. (2003). Mammalian hibernation: cellular and molecular responses to depressed metabolism and low temperature. Physiological reviews 83, 1153-1181.

Carmeliet, P., and Jain, R.K. (2000). Angiogenesis in cancer and other diseases. Nature 407, 249- 257.

Catchpole, S., Spencer-Dene, B., Hall, D., Santangelo, S., Rosewell, I., Guenatri, M., Beatson, R., Scibetta, A.G., Burchell, J.M., and Taylor-Papadimitriou, J. (2011). PLU-1/JARID1B/KDM5B is required for embryonic survival and contributes to cell proliferation in the mammary gland and in ER+ breast cancer cells. International journal of oncology 38, 1267-1277.

Chang, J.T., Gatza, M.L., Lucas, J.E., Barry, W.T., Vaughn, P., and Nevins, J.R. (2011). SIGNATURE: a workbench for gene expression signature analysis. BMC bioinformatics 12, 443.

Chang, J.T., and Nevins, J.R. (2006). GATHER: a systems approach to interpreting genomic signatures. Bioinformatics (Oxford, England) 22, 2926-2933.

Chen, J.H., Vercamer, C., Li, Z., Paulin, D., Vandenbunder, B., and Stehelin, D. (1996). PEA3 transactivates vimentin promoter in mammary epithelial and tumor cells. Oncogene 13, 1667- 1675.

Chen, J.L., Lucas, J.E., Schroeder, T., Mori, S., Wu, J., Nevins, J., Dewhirst, M., West, M., and Chi, J.T. (2008). The genomic analysis of lactic acidosis and acidosis response in human cancers. PLoS genetics 4, e1000293.

Chen, J.L., Merl, D., Peterson, C.W., Wu, J., Liu, P.Y., Yin, H., Muoio, D.M., Ayer, D.E., West, M., and Chi, J.T. (2010). Lactic acidosis triggers starvation response with paradoxical induction of TXNIP through MondoA. PLoS genetics 6, e1001093.

Chiche, J., Brahimi-Horn, M.C., and Pouyssegur, J. (2010). Tumour hypoxia induces a metabolic shift causing acidosis: a common feature in cancer. Journal of cellular and molecular medicine 14,

209

771-794.

Chiche, J., Ilc, K., Laferriere, J., Trottier, E., Dayan, F., Mazure, N.M., Brahimi-Horn, M.C., and Pouyssegur, J. (2009). Hypoxia-inducible carbonic anhydrase IX and XII promote tumor cell growth by counteracting acidosis through the regulation of the intracellular pH. Cancer research 69, 358-368.

Chin, K., DeVries, S., Fridlyand, J., Spellman, P.T., Roydasgupta, R., Kuo, W.L., Lapuk, A., Neve, R.M., Qian, Z., Ryder, T., et al. (2006). Genomic and transcriptional aberrations linked to breast cancer pathophysiologies. Cancer cell 10, 529-541.

Chiu, M.H., Liu, H.S., Wu, Y.H., Shen, M.R., and Chou, C.Y. (2014). SPAK mediates KCC3- enhanced cervical cancer tumorigenesis. The FEBS journal 281, 2353-2365.

Choi, H.H., Gully, C., Su, C.H., Velazquez-Torres, G., Chou, P.C., Tseng, C., Zhao, R., Phan, L., Shaiken, T., Chen, J., et al. (2011). COP9 signalosome subunit 6 stabilizes COP1, which functions as an E3 ubiquitin ligase for 14-3-3sigma. Oncogene 30, 4791-4801.

Chowdhury, R., Yeoh, K.K., Tian, Y.M., Hillringhaus, L., Bagg, E.A., Rose, N.R., Leung, I.K., Li, X.S., Woon, E.C., Yang, M., et al. (2011). The oncometabolite 2-hydroxyglutarate inhibits histone lysine demethylases. EMBO Rep 12, 463-469.

Chypre, M., Zaidi, N., and Smans, K. (2012). ATP-citrate lyase: a mini-review. Biochemical and biophysical research communications 422, 1-4.

Colas, E., Muinelo-Romay, L., Alonso-Alconada, L., Llaurado, M., Monge, M., Barbazan, J., Gonzalez, M., Schoumacher, M., Pedrola, N., Ertekin, T., et al. (2012). ETV5 cooperates with LPP as a sensor of extracellular signals and promotes EMT in endometrial carcinomas. Oncogene 31, 4778-4788.

Conacci-Sorrell, M., Ngouenet, C., Anderson, S., Brabletz, T., and Eisenman, R.N. (2014). Stress- induced cleavage of Myc promotes cancer cell survival. Genes & development 28, 689-707.

Cooper, C.D., Newman, J.A., Aitkenhead, H., Allerston, C.K., and Gileadi, O. (2015). Structures of the Ets Protein DNA-binding Domains of Transcription Factors Etv1, Etv4, Etv5, and Fev: DETERMINANTS OF DNA BINDING AND REDOX REGULATION BY DISULFIDE BOND FORMATION. The Journal of biological chemistry 290, 13692-13709.

Croft, A., Tay, K.H., Boyd, S.C., Guo, S.T., Jiang, C.C., Lai, F., Tseng, H.Y., Jin, L., Rizos, H., Hersey, P., et al. (2014). Oncogenic activation of MEK/ERK primes melanoma cells for adaptation to endoplasmic reticulum stress. The Journal of investigative dermatology 134, 488-497.

Csanadi, A., Kayser, C., Donauer, M., Gumpp, V., Aumann, K., Rawluk, J., Prasse, A., zur Hausen, A., Wiesemann, S., Werner, M., et al. (2015). Prognostic Value of Malic Enzyme and ATP-

210

Citrate Lyase in Non-Small Cell Lung Cancer of the Young and the Elderly. PloS one 10, e0126357.

Cullen, B.R. (2004). Transcription and processing of human microRNA precursors. Molecular cell 16, 861-865.

Currie, E., Schulze, A., Zechner, R., Walther, T.C., and Farese, R.V., Jr. (2013). Cellular fatty acid metabolism and cancer. Cell metabolism 18, 153-161.

Curthoys, N.P., Taylor, L., Hoffert, J.D., and Knepper, M.A. (2007). Proteomic analysis of the adaptive response of rat renal proximal tubules to metabolic acidosis. American journal of physiology Renal physiology 292, F140-147.

Dang, C.V. (2011). Therapeutic targeting of Myc-reprogrammed cancer cell metabolism. Cold Spring Harbor symposia on quantitative biology 76, 369-374.

Dang, C.V., and Semenza, G.L. (1999). Oncogenic alterations of metabolism. Trends in biochemical sciences 24, 68-72.

Dang, L., White, D.W., Gross, S., Bennett, B.D., Bittinger, M.A., Driggers, E.M., Fantin, V.R., Jang, H.G., Jin, S., Keenan, M.C., et al. (2009). Cancer-associated IDH1 mutations produce 2- hydroxyglutarate. Nature 462, 739-744.

Das Thakur, M., Feng, Y., Jagannathan, R., Seppa, M.J., Skeath, J.B., and Longmore, G.D. (2010). Ajuba LIM proteins are negative regulators of the Hippo signaling pathway. Current biology: CB 20, 657-662.

David, S., Bucchieri, F., Corrao, S., Czarnecka, A.M., Campanella, C., Farina, F., Peri, G., Tomasello, G., Sciume, C., Modica, G., et al. (2013). Hsp10: anatomic distribution, functions, and involvement in human disease. Frontiers in bioscience (Elite edition) 5, 768-778.

Delage, B., Fennell, D.A., Nicholson, L., McNeish, I., Lemoine, N.R., Crook, T., and Szlosarek, P.W. (2010). Arginine deprivation and argininosuccinate synthetase expression in the treatment of cancer. International journal of cancer Journal international du cancer 126, 2762-2772.

Dhani, N., Fyles, A., Hedley, D., and Milosevic, M. (2015). The clinical significance of hypoxia in human cancers. Seminars in nuclear medicine 45, 110-121.

Di Noia, M.A., Todisco, S., Cirigliano, A., Rinaldi, T., Agrimi, G., Iacobazzi, V., and Palmieri, F. (2014). The human SLC25A33 and SLC25A36 genes of solute carrier family 25 encode two mitochondrial pyrimidine nucleotide transporters. The Journal of biological chemistry 289, 33137- 33148.

Dinchuk, J.E., Focht, R.J., Kelley, J.A., Henderson, N.L., Zolotarjova, N.I., Wynn, R., Neff, N.T.,

211

Link, J., Huber, R.M., Burn, T.C., et al. (2002). Absence of post-translational aspartyl beta- hydroxylation of epidermal growth factor domains in mice leads to developmental defects and an increased incidence of intestinal neoplasia. The Journal of biological chemistry 277, 12970- 12977.

Dolcet, X., Llobet, D., Pallares, J., and Matias-Guiu, X. (2005). NF-kB in development and progression of human cancer. Virchows Archiv: an international journal of pathology 446, 475- 482.

Dong, L., Li, Z., Leffler, N.R., Asch, A.S., Chi, J.T., and Yang, L.V. (2013). Acidosis activation of the proton-sensing GPR4 receptor stimulates vascular endothelial cell inflammatory responses revealed by transcriptome analysis. PloS one 8, e61991.

Dorak, M.T., and Karpuzoglu, E. (2012). Gender differences in cancer susceptibility: an inadequately addressed issue. Frontiers in genetics 3, 268.

Dornan, D., Wertz, I., Shimizu, H., Arnott, D., Frantz, G.D., Dowd, P., O'Rourke, K., Koeppen, H., and Dixit, V.M. (2004). The ubiquitin ligase COP1 is a critical negative regulator of p53. Nature 429, 86-92.

Draper, J.M., and Smith, C.D. (2010). DHHC20: a human palmitoyl acyltransferase that causes cellular transformation. Molecular membrane biology 27, 123-136.

Duan, Z., Ji, D., Weinstein, E.J., Liu, X., Susa, M., Choy, E., Yang, C., Mankin, H., and Hornicek, F.J. (2010). Lentiviral shRNA screen of human kinases identifies PLK1 as a potential therapeutic target for osteosarcoma. Cancer letters 293, 220-229.

Eden, E., Navon, R., Steinfeld, I., Lipson, D., and Yakhini, Z. (2009). GOrilla: a tool for discovery and visualization of enriched GO terms in ranked gene lists. BMC bioinformatics 10, 48.

El Mjiyad, N., Caro-Maldonado, A., Ramirez-Peinado, S., and Munoz-Pinedo, C. (2011). Sugar- free approaches to cancer cell killing. Oncogene 30, 253-264.

Elbashir, S.M., Harborth, J., Lendeckel, W., Yalcin, A., Weber, K., and Tuschl, T. (2001). Duplexes of 21-nucleotide RNAs mediate RNA interference in cultured mammalian cells. Nature 411, 494- 498.

Elliott, B., and Futcher, B. (1993). Stress resistance of yeast cells is largely independent of cell cycle phase. Yeast (Chichester, England) 9, 33-42.

Erkizan, H.V., Kong, Y., Merchant, M., Schlottmann, S., Barber-Rotenberg, J.S., Yuan, L., Abaan, O.D., Chou, T.H., Dakshanamurthy, S., Brown, M.L., et al. (2009). A small molecule blocking oncogenic protein EWS-FLI1 interaction with RNA helicase A inhibits growth of Ewing's sarcoma. Nature medicine 15, 750-756.

212

Falini, B., Bigerna, B., Pucciarini, A., Tiacci, E., Mecucci, C., Morris, S.W., Bolli, N., Rosati, R., Hanissian, S., Ma, Z., et al. (2006). Aberrant subcellular expression of nucleophosmin and NPM- MLF1 fusion protein in acute myeloid leukaemia carrying t(3;5): a comparison with NPMc+ AML. Leukemia 20, 368-371.

Falini, B., Nicoletti, I., Bolli, N., Martelli, M.P., Liso, A., Gorello, P., Mandelli, F., Mecucci, C., and Martelli, M.F. (2007). Translocations and mutations involving the nucleophosmin (NPM1) gene in lymphomas and leukemias. Haematologica 92, 519-532.

Fan, J., Kamphorst, J.J., Mathew, R., Chung, M.K., White, E., Shlomi, T., and Rabinowitz, J.D. (2013). Glutamine-driven oxidative phosphorylation is a major ATP source in transformed mammalian cells in both normoxia and hypoxia. Molecular systems biology 9, 712.

Fang, J.S., Gillies, R.D., and Gatenby, R.A. (2008). Adaptation to hypoxia and acidosis in carcinogenesis and tumor progression. Seminars in cancer biology 18, 330-337.

Farmer, H., McCabe, N., Lord, C.J., Tutt, A.N., Johnson, D.A., Richardson, T.B., Santarosa, M., Dillon, K.J., Hickson, I., Knights, C., et al. (2005). Targeting the DNA repair defect in BRCA mutant cells as a therapeutic strategy. Nature 434, 917-921.

Farrow, B., Albo, D., and Berger, D.H. (2008). The Role of the Tumor Microenvironment in the Progression of Pancreatic Cancer. Journal of Surgical Research 149, 319-328.

Favre, C., Zhdanov, A., Leahy, M., Papkovsky, D., and O'Connor, R. (2010). Mitochondrial pyrimidine nucleotide carrier (PNC1) regulates mitochondrial biogenesis and the invasive phenotype of cancer cells. Oncogene 29, 3964-3976.

Feldman, B.J., and Feldman, D. (2001). The development of androgen-independent prostate cancer. Nature reviews Cancer 1, 34-45.

Fennell, M., Xiang, Q., Hwang, A., Chen, C., Huang, C.H., Chen, C.C., Pelossof, R., and Garippa, R.J. (2014). Impact of RNA-guided technologies for target identification and deconvolution. Journal of biomolecular screening 19, 1327-1337.

Fenner, A. (2013). Kidney cancer: Stearoyl-CoA desaturase: a novel therapeutic target for RCC. Nature reviews Urology 10, 370.

Ferrara, C.T., Wang, P., Neto, E.C., Stevens, R.D., Bain, J.R., Wenner, B.R., Ilkayeva, O.R., Keller, M.P., Blasiole, D.A., Kendziorski, C., et al. (2008). Genetic networks of liver metabolism revealed by integration of metabolic and transcriptional profiling. PLoS genetics 4, e1000034.

Ferrer, C.M., Lynch, T.P., Sodi, V.L., Falcone, J.N., Schwab, L.P., Peacock, D.L., Vocadlo, D.J., Seagroves, T.N., and Reginato, M.J. (2014). O-GlcNAcylation regulates cancer metabolism and

213

survival stress signaling via regulation of the HIF-1 pathway. Molecular cell 54, 820-831.

Fielenbach, N., and Antebi, A. (2008). C. elegans dauer formation and the molecular basis of plasticity. Genes & development 22, 2149-2165.

Formby, B., and Stern, R. (2003). Lactate-sensitive response elements in genes involved in hyaluronan catabolism. Biochemical and biophysical research communications 305, 203-208.

Foxler, D.E., Bridge, K.S., James, V., Webb, T.M., Mee, M., Wong, S.C., Feng, Y., Constantin- Teodosiu, D., Petursdottir, T.E., Bjornsson, J., et al. (2012). The LIMD1 protein bridges an association between the prolyl hydroxylases and VHL to repress HIF-1 activity. Nature cell biology 14, 201-208.

Francisco, A.B., Singh, R., Li, S., Vani, A.K., Yang, L., Munroe, R.J., Diaferia, G., Cardano, M., Biunno, I., Qi, L., et al. (2010). Deficiency of suppressor enhancer Lin12 1 like (SEL1L) in mice leads to systemic endoplasmic reticulum stress and embryonic lethality. The Journal of biological chemistry 285, 13694-13703.

Fukumura, D., Xu, L., Chen, Y., Gohongi, T., Seed, B., and Jain, R.K. (2001). Hypoxia and acidosis independently up-regulate vascular endothelial growth factor transcription in brain tumors in vivo. Cancer research 61, 6020-6024.

Fulda, S., and Debatin, K.M. (2006). Extrinsic versus intrinsic apoptosis pathways in anticancer chemotherapy. Oncogene 25, 4798-4811.

Gagnon, K.B., and Delpire, E. (2012). Molecular physiology of SPAK and OSR1: two Ste20-related protein kinases regulating ion transport. Physiological reviews 92, 1577-1617.

Gatenby, R.A., and Gillies, R.J. (2008). A microenvironmental model of carcinogenesis. Nature reviews Cancer 8, 56-61.

Gatza, M.L., Kung, H.N., Blackwell, K.L., Dewhirst, M.W., Marks, J.R., and Chi, J.T. (2011). Analysis of tumor environmental response and oncogenic pathway activation identifies distinct basal and luminal features in HER2-related breast tumor subtypes. Breast cancer research: BCR 13, R62.

Ghosh, S., Ghosh, A., Maiti, G.P., Alam, N., Roy, A., Roy, B., Roychoudhury, S., and Panda, C.K. (2008). Alterations of 3p21.31 tumor suppressor genes in head and neck squamous cell carcinoma: Correlation with progression and prognosis. International journal of cancer Journal international du cancer 123, 2594-2604.

Ghosh, S., Ghosh, A., Maiti, G.P., Mukherjee, N., Dutta, S., Roy, A., Roychoudhury, S., and Panda, C.K. (2010). LIMD1 is more frequently altered than RB1 in head and neck squamous cell carcinoma: clinical and prognostic implications. Molecular cancer 9, 58.

214

Gordan, J.D., Thompson, C.B., and Simon, M.C. (2007). HIF and c-Myc: sibling rivals for control of cancer cell metabolism and proliferation. Cancer cell 12, 108-113.

Greenall, A., Willingham, N., Cheung, E., Boam, D.S., and Sharrocks, A.D. (2001). DNA binding by the ETS-domain transcription factor PEA3 is regulated by intramolecular and intermolecular protein.protein interactions. The Journal of biological chemistry 276, 16207-16215.

Greijer, A.E., and van der Wall, E. (2004). The role of hypoxia inducible factor 1 (HIF-1) in hypoxia induced apoptosis. Journal of clinical pathology 57, 1009-1014.

Griffith, L.G., and Swartz, M.A. (2006). Capturing complex 3D tissue physiology in vitro. Nature reviews Molecular cell biology 7, 211-224.

Grinshtein, N., Datti, A., Fujitani, M., Uehling, D., Prakesch, M., Isaac, M., Irwin, M.S., Wrana, J.L., Al-Awar, R., and Kaplan, D.R. (2011). Small molecule kinase inhibitor screen identifies polo- like kinase 1 as a target for neuroblastoma tumor-initiating cells. Cancer research 71, 1385-1395.

Guimaraes-Camboa, N., Stowe, J., Aneas, I., Sakabe, N., Cattaneo, P., Henderson, L., Kilberg, M.S., Johnson, R.S., Chen, J., McCulloch, A.D., et al. (2015). HIF1alpha Represses Cell Stress Pathways to Allow Proliferation of Hypoxic Fetal Cardiomyocytes. Developmental cell 33, 507- 521.

Guise, C.P., Mowday, A.M., Ashoorzadeh, A., Yuan, R., Lin, W.H., Wu, D.H., Smaill, J.B., Patterson, A.V., and Ding, K. (2014). Bioreductive prodrugs as cancer therapeutics: targeting tumor hypoxia. Chinese journal of cancer 33, 80-86.

Gulledge, C.J., and Dewhirst, M.W. (1996). Tumor oxygenation: a matter of supply and demand. Anticancer research 16, 741-749.

Guo, B., Panagiotaki, N., Warwood, S., and Sharrocks, A.D. (2011a). Dynamic modification of the ETS transcription factor PEA3 by sumoylation and p300-mediated acetylation. Nucleic acids research 39, 6403-6413.

Guo, C., Pirozzi, C.J., Lopez, G.Y., and Yan, H. (2011b). Isocitrate dehydrogenase mutations in gliomas: mechanisms, biomarkers and therapeutic target. Current opinion in neurology 24, 648- 652.

Guo, B., Sallis, R.E., Greenall, A., Petit, M.M., Jansen, E., Young, L., Van de Ven, W.J., and Sharrocks, A.D. (2006). The LIM domain protein LPP is a coactivator for the ETS domain transcription factor PEA3. Molecular and cellular biology 26, 4529-4538.

Guo, B., and Sharrocks, A.D. (2009). Extracellular signal-regulated kinase mitogen-activated protein kinase signaling initiates a dynamic interplay between sumoylation and ubiquitination to

215

regulate the activity of the transcriptional activator PEA3. Molecular and cellular biology 29, 3204-3218.

Gupta, S.C., Singh, R., Pochampally, R., Watabe, K., and Mo, Y.Y. (2014). Acidosis promotes invasiveness of breast cancer cells through ROS-AKT-NF-kappaB pathway. Oncotarget 5, 12070- 12082.

Haas, B.R., Cuddapah, V.A., Watkins, S., Rohn, K.J., Dy, T.E., and Sontheimer, H. (2011). With- No-Lysine Kinase 3 (WNK3) stimulates glioma invasion by regulating cell volume. American journal of physiology Cell physiology 301, C1150-1160.

Hakuma, N., Kinoshita, I., Shimizu, Y., Yamazaki, K., Yoshida, K., Nishimura, M., and Dosaka- Akita, H. (2005). E1AF/PEA3 activates the Rho/Rho-associated kinase pathway to increase the malignancy potential of non-small-cell lung cancer cells. Cancer research 65, 10776-10782.

Hallstrom, T.C., Mori, S., and Nevins, J.R. (2008). An E2F1-dependent gene expression program that determines the balance between proliferation and cell death. Cancer cell 13, 11-22.

Hammond, E.M., and Giaccia, A.J. (2005). The role of p53 in hypoxia-induced apoptosis. Biochemical and biophysical research communications 331, 718-725.

Hanahan, D., and Weinberg, R.A. (2000). The hallmarks of cancer. Cell 100, 57-70.

Hanahan, D., and Weinberg, R.A. (2011). Hallmarks of cancer: the next generation. Cell 144, 646- 674.

Hanzawa, M., Shindoh, M., Higashino, F., Yasuda, M., Inoue, N., Hida, K., Ono, M., Kohgo, T., Nakamura, M., Notani, K., et al. (2000). Hepatocyte growth factor upregulates E1AF that induces oral squamous cell carcinoma cell invasion by activating matrix metalloproteinase genes. Carcinogenesis 21, 1079-1085.

Haveman, J. (1980). The influence of pH on the survival after X-irradiation of cultured malignant cells. Effects of carbonylcyanide-3-chlorophenylhydrazone. International journal of radiation biology and related studies in physics, chemistry, and medicine 37, 201-205.

Haystead, T.A., Moore, F., Cohen, P., and Hardie, D.G. (1990). Roles of the AMP-activated and cyclic-AMP-dependent protein kinases in the adrenaline-induced inactivation of acetyl-CoA carboxylase in rat adipocytes. European journal of biochemistry / FEBS 187, 199-205.

Helgeson, B.E., Tomlins, S.A., Shah, N., Laxman, B., Cao, Q., Prensner, J.R., Cao, X., Singla, N., Montie, J.E., Varambally, S., et al. (2008). Characterization of TMPRSS2:ETV5 and SLC45A3:ETV5 gene fusions in prostate cancer. Cancer research 68, 73-80.

Helmlinger, G., Yuan, F., Dellian, M., and Jain, R.K. (1997). Interstitial pH and pO2 gradients in

216

solid tumors in vivo: high-resolution measurements reveal a lack of correlation. Nature medicine 3, 177-182.

Hendriksen, P.J., Dits, N.F., Kokame, K., Veldhoven, A., van Weerden, W.M., Bangma, C.H., Trapman, J., and Jenster, G. (2006). Evolution of the androgen receptor pathway during progression of prostate cancer. Cancer research 66, 5012-5020.

Hilton, I.B., D'Ippolito, A.M., Vockley, C.M., Thakore, P.I., Crawford, G.E., Reddy, T.E., and Gersbach, C.A. (2015). Epigenome editing by a CRISPR-Cas9-based acetyltransferase activates genes from promoters and enhancers. Nature biotechnology 33, 510-517.

Hiroumi, H., Dosaka-Akita, H., Yoshida, K., Shindoh, M., Ohbuchi, T., Fujinaga, K., and Nishimura, M. (2001). Expression of E1AF/PEA3, an Ets-related transcription factor in human non-small-cell lung cancers: its relevance in cell motility and invasion. International journal of cancer Journal international du cancer 93, 786-791.

Hirschhaeuser, F., Sattler, U.G., and Mueller-Klieser, W. (2011). Lactate: a metabolic key player in cancer. Cancer research 71, 6921-6925.

Hjelmeland, A.B., Wu, Q., Heddleston, J.M., Choudhary, G.S., MacSwords, J., Lathia, J.D., McLendon, R., Lindner, D., Sloan, A., and Rich, J.N. (2011). Acidic stress promotes a glioma stem cell phenotype. Cell death and differentiation 18, 829-840.

Horton, J.D., Goldstein, J.L., and Brown, M.S. (2002). SREBPs: activators of the complete program of cholesterol and fatty acid synthesis in the liver. The Journal of clinical investigation 109, 1125- 1131.

Hou, J., Wu, J., Dombkowski, A., Zhang, K., Holowatyj, A., Boerner, J.L., and Yang, Z.Q. (2012). Genomic amplification and a role in drug-resistance for the KDM5A histone demethylase in breast cancer. American journal of translational research 4, 247-256.

Hu, K., Lee, C., Qiu, D., Fotovati, A., Davies, A., Abu-Ali, S., Wai, D., Lawlor, E.R., Triche, T.J., Pallen, C.J., et al. (2009). Small interfering RNA library screen of human kinases and phosphatases identifies polo-like kinase 1 as a promising new target for the treatment of pediatric rhabdomyosarcomas. Molecular cancer therapeutics 8, 3024-3035.

Huang, W.C., Swietach, P., Vaughan-Jones, R.D., Ansorge, O., and Glitsch, M.D. (2008). Extracellular acidification elicits spatially and temporally distinct Ca2+ signals. Current biology: CB 18, 781-785.

Huang, Y.T., Heist, R.S., Chirieac, L.R., Lin, X., Skaug, V., Zienolddiny, S., Haugen, A., Wu, M.C., Wang, Z., Su, L., et al. (2009). Genome-wide analysis of survival in early-stage non-small-cell lung cancer. Journal of clinical oncology: official journal of the American Society of Clinical Oncology 27, 2660-2667.

217

Huggins, C.J., and Andrulis, I.L. (2008). Cell cycle regulated phosphorylation of LIMD1 in cell lines and expression in human breast cancers. Cancer letters 267, 55-66.

Huggins, C.J., Gill, M., and Andrulis, I.L. (2007). Identification of rare variants in the hLIMD1 gene in breast cancer. Cancer genetics and cytogenetics 178, 36-41.

Hynes, N.E., and Lane, H.A. (2005). ERBB receptors and cancer: the complexity of targeted inhibitors. Nature reviews Cancer 5, 341-354.

Ibrahim-Hashim, A., Cornnell, H.H., Coelho Ribeiro Mde, L., Abrahams, D., Cunningham, J., Lloyd, M., Martinez, G.V., Gatenby, R.A., and Gillies, R.J. (2011). Reduction of metastasis using a non-volatile buffer. Clinical & experimental metastasis 28, 841-849.

Ibrahim-Hashim, A., Wojtkowiak, J.W., de Lourdes Coelho Ribeiro, M., Estrella, V., Bailey, K.M., Cornnell, H.H., Gatenby, R.A., and Gillies, R.J. (2011). Free Base Lysine Increases Survival and Reduces Metastasis in Prostate Cancer Model. Journal of cancer science & therapy Suppl 1.

Ibrahim-Hashim, A., Cornnell, H.H., Abrahams, D., Lloyd, M., Bui, M., Gillies, R.J., and Gatenby, R.A. (2012). Systemic buffers inhibit carcinogenesis in TRAMP mice. The Journal of urology 188, 624-631.

Igal, R.A. (2010). Stearoyl-CoA desaturase-1: a novel key player in the mechanisms of cell proliferation, programmed cell death and transformation to cancer. Carcinogenesis 31, 1509-1515.

Iida, Y., Fujimori, T., Okawa, K., Nagata, K., Wada, I., and Hosokawa, N. (2011). SEL1L protein critically determines the stability of the HRD1-SEL1L endoplasmic reticulum-associated degradation (ERAD) complex to optimize the degradation kinetics of ERAD substrates. The Journal of biological chemistry 286, 16929-16939.

Imtiyaz, H.Z., and Simon, M.C. (2010). Hypoxia-inducible factors as essential regulators of inflammation. Current topics in microbiology and immunology 345, 105-120.

Intlekofer, A.M., Dematteo, R.G., Venneti, S., Finley, L.W., Lu, C., Judkins, A.R., Rustenburg, A.S., Grinaway, P.B., Chodera, J.D., Cross, J.R., et al. (2015). Hypoxia Induces Production of L-2- Hydroxyglutarate. Cell metabolism 22, 304-311.

Jain, R.K., Martin, J.D., and Stylianopoulos, T. (2014). The role of mechanical forces in tumor growth and therapy. Annual review of biomedical engineering 16, 321-346.

James, V., Zhang, Y., Foxler, D.E., de Moor, C.H., Kong, Y.W., Webb, T.M., Self, T.J., Feng, Y., Lagos, D., Chu, C.Y., et al. (2010). LIM-domain proteins, LIMD1, Ajuba, and WTIP are required for microRNA-mediated gene silencing. Proceedings of the National Academy of Sciences of the United States of America 107, 12499-12504.

218

Jangravi, Z., Tabar, M.S., Mirzaei, M., Parsamatin, P., Vakilian, H., Alikhani, M., Shabani, M., Haynes, P.A., Goodchild, A.K., Gourabi, H., et al. (2015). Two Splice Variants of Y Chromosome- Located Lysine-Specific Demethylase 5D Have Distinct Function in Prostate Cancer Cell Line (DU-145). Journal of proteome research 14, 3492-3502.

Jeon, S.M., Chandel, N.S., and Hay, N. (2012). AMPK regulates NADPH homeostasis to promote tumour cell survival during energy stress. Nature 485, 661-665.

Jiang, J., Wei, Y., Shen, J., Liu, D., Chen, X., Zhou, J., Zong, H., Yun, X., Kong, X., Zhang, S., et al. (2007). Functional interaction of E1AF and Sp1 in glioma invasion. Molecular and cellular biology 27, 8770-8782.

Jin, X., Zhang, K.J., Guo, X., Myers, R., Ye, Z., Zhang, Z.P., Li, X.F., Yang, H.S., and Xing, J.L. (2014). Fatty acid synthesis pathway genetic variants and clinical outcome of non-small cell lung cancer patients after surgery. Asian Pacific journal of cancer prevention: APJCP 15, 7097-7103.

Johansson, C., Tumber, A., Che, K., Cain, P., Nowak, R., Gileadi, C., and Oppermann, U. (2014). The roles of Jumonji-type oxygenases in human disease. Epigenomics 6, 89-120.

Jones, R.G., Plas, D.R., Kubek, S., Buzzai, M., Mu, J., Xu, Y., Birnbaum, M.J., and Thompson, C.B. (2005). AMP-activated protein kinase induces a p53-dependent metabolic checkpoint. Molecular cell 18, 283-293.

Jung, V., Kindich, R., Kamradt, J., Jung, M., Muller, M., Schulz, W.A., Engers, R., Unteregger, G., Stockle, M., Zimmermann, R., et al. (2006). Genomic and expression analysis of the 3q25-q26 amplification unit reveals TLOC1/SEC62 as a probable target gene in prostate cancer. Molecular cancer research: MCR 4, 169-176.

Kadrmas, J.L., and Beckerle, M.C. (2004). The LIM domain: from the cytoskeleton to the nucleus. Nature reviews Molecular cell biology 5, 920-931.

Kaelin, W.G., Jr. (2007). The von Hippel-Lindau tumor suppressor protein and clear cell renal carcinoma. Clinical cancer research: an official journal of the American Association for Cancer Research 13, 680s-684s.

Kaelin, W.G., Jr. (2008). The von Hippel-Lindau tumour suppressor protein: O2 sensing and cancer. Nature reviews Cancer 8, 865-873.

Kaelin, W.G., Jr. (2011). Cancer and altered metabolism: potential importance of hypoxia- inducible factor and 2-oxoglutarate-dependent dioxygenases. Cold Spring Harbor symposia on quantitative biology 76, 335-345.

Kamphorst, J.J., Cross, J.R., Fan, J., de Stanchina, E., Mathew, R., White, E.P., Thompson, C.B., and

219

Rabinowitz, J.D. (2013). Hypoxic and Ras-transformed cells support growth by scavenging unsaturated fatty acids from lysophospholipids. Proceedings of the National Academy of Sciences of the United States of America 110, 8882-8887.

Karashima, T., Sweeney, P., Kamat, A., Huang, S., Kim, S.J., Bar-Eli, M., McConkey, D.J., and Dinney, C.P. (2003). Nuclear factor-kappaB mediates angiogenesis and metastasis of human bladder cancer through the regulation of interleukin-8. Clinical cancer research: an official journal of the American Association for Cancer Research 9, 2786-2797.

Karin, M., and Lin, A. (2002). NF-kappaB at the crossroads of life and death. Nature immunology 3, 221-227.

Keenan, M.M., and Chi, J.T. (2015). Alternative fuels for cancer cells. Cancer journal (Sudbury, Mass) 21, 49-55.

Keenan, M.M., Liu, B., Tang, X., Wu, J., Cyr, D., Stevens, R.D., Ilkayeva, O., Huang, Z., Tollini, L.A., Murphy, S.K., et al. (2015). ACLY and ACC1 Regulate Hypoxia-Induced Apoptosis by Modulating ETV4 via alpha-ketoglutarate. PLoS genetics 11, e1005599.

Keith, B., Johnson, R.S., and Simon, M.C. (2012). HIF1alpha and HIF2alpha: sibling rivalry in hypoxic tumour growth and progression. Nature reviews Cancer 12, 9-22.

Keith, B., and Simon, M.C. (2007). Hypoxia-inducible factors, stem cells, and cancer. Cell 129, 465- 472.

Keld, R., Guo, B., Downey, P., Cummins, R., Gulmann, C., Ang, Y.S., and Sharrocks, A.D. (2011). PEA3/ETV4-related transcription factors coupled with active ERK signalling are associated with poor prognosis in gastric adenocarcinoma. British journal of cancer 105, 124-130.

Kelloff, G.J., Hoffman, J.M., Johnson, B., Scher, H.I., Siegel, B.A., Cheng, E.Y., Cheson, B.D., O'Shaughnessy, J., Guyton, K.Z., Mankoff, D.A., et al. (2005). Progress and promise of FDG-PET imaging for cancer patient management and oncologic drug development. Clinical cancer research: an official journal of the American Association for Cancer Research 11, 2785-2808.

Kharaishvili, G., Simkova, D., Bouchalova, K., Gachechiladze, M., Narsia, N., and Bouchal, J. (2014). The role of cancer-associated fibroblasts, solid stress and other microenvironmental factors in tumor progression and therapy resistance. Cancer Cell International 14, 41.

Kherrouche, Z., Monte, D., Werkmeister, E., Stoven, L., De Launoit, Y., Cortot, A.B., Tulasne, D., and Chotteau-Lelievre, A. (2015). PEA3 transcription factors are downstream effectors of Met signaling involved in migration and invasiveness of Met-addicted tumor cells. Molecular oncology 9, 1852-1867.

Kim, H., Bhattacharya, A., and Qi, L. (2015). Endoplasmic reticulum quality control in cancer:

220

Friend or foe. Seminars in cancer biology 33, 25-33.

Kindler, H.L., Richards, D.A., Garbo, L.E., Garon, E.B., Stephenson, J.J., Jr., Rocha-Lima, C.M., Safran, H., Chan, D., Kocs, D.M., Galimi, F., et al. (2012). A randomized, placebo-controlled phase 2 study of ganitumab (AMG 479) or conatumumab (AMG 655) in combination with gemcitabine in patients with metastatic pancreatic cancer. Annals of oncology: official journal of the European Society for Medical Oncology / ESMO 23, 2834-2842.

Koch, C.J., and Evans, S.M. (2015). Optimizing hypoxia detection and treatment strategies. Seminars in nuclear medicine 45, 163-176.

Koh, M.Y., Darnay, B.G., and Powis, G. (2008). Hypoxia-associated factor, a novel E3-ubiquitin ligase, binds and ubiquitinates hypoxia-inducible factor 1alpha, leading to its oxygen- independent degradation. Molecular and cellular biology 28, 7081-7095.

Koh, M.Y., Lemos, R., Jr., Liu, X., and Powis, G. (2011). The hypoxia-associated factor switches cells from HIF-1alpha- to HIF-2alpha-dependent signaling promoting stem cell characteristics, aggressive tumor growth and invasion. Cancer research 71, 4015-4027.

Kondo, K., Kim, W.Y., Lechpammer, M., and Kaelin, W.G., Jr. (2003). Inhibition of HIF2alpha is sufficient to suppress pVHL-defective tumor growth. PLoS biology 1, E83.

Koodie, L., Ramakrishnan, S., and Roy, S. (2010). Morphine suppresses tumor angiogenesis through a HIF-1alpha/p38MAPK pathway. The American journal of pathology 177, 984-997.

Koong, A.C., Chen, E.Y., and Giaccia, A.J. (1994). Hypoxia causes the activation of nuclear factor kappa B through the phosphorylation of I kappa B alpha on tyrosine residues. Cancer research 54, 1425-1430.

Koshiji, M., To, K.K., Hammer, S., Kumamoto, K., Harris, A.L., Modrich, P., and Huang, L.E. (2005). HIF-1alpha induces genetic instability by transcriptionally downregulating MutSalpha expression. Molecular cell 17, 793-803.

Krausz, E. (2007). High-content siRNA screening. Molecular bioSystems 3, 232-240.

Kravtsova-Ivantsiv, Y., Shomer, I., Cohen-Kaplan, V., Snijder, B., Superti-Furga, G., Gonen, H., Sommer, T., Ziv, T., Admon, A., Naroditsky, I., et al. (2015). KPC1-mediated ubiquitination and proteasomal processing of NF-kappaB1 p105 to p50 restricts tumor growth. Cell 161, 333-347.

Kuo, Y.T., Liu, Y.L., Adebayo, B.O., Shih, P.H., Lee, W.H., Wang, L.S., Liao, Y.F., Hsu, W.M., Yeh, C.T., and Lin, C.M. (2015). JARID1B Expression Plays a Critical Role in Chemoresistance and Stem Cell-Like Phenotype of Neuroblastoma Cells. PloS one 10, e0125343.

Kuure, S., Chi, X., Lu, B., and Costantini, F. (2010). The transcription factors Etv4 and Etv5

221

mediate formation of the ureteric bud tip domain during kidney development. Development (Cambridge, England) 137, 1975-1979.

Ladam, F., Damour, I., Dumont, P., Kherrouche, Z., de Launoit, Y., Tulasne, D., and Chotteau- Lelievre, A. (2013). Loss of a negative feedback loop involving pea3 and cyclin d2 is required for pea3-induced migration in transformed mammary epithelial cells. Molecular cancer research: MCR 11, 1412-1424.

Ladle, D.R., and Frank, E. (2002). The role of the ETS gene PEA3 in the development of motor and sensory neurons. Physiology & behavior 77, 571-576.

Laing, M.A., Coonrod, S., Hinton, B.T., Downie, J.W., Tozer, R., Rudnicki, M.A., and Hassell, J.A. (2000). Male sexual dysfunction in mice bearing targeted mutant alleles of the PEA3 ets gene. Molecular and cellular biology 20, 9337-9345.

Lamhamedi-Cherradi, S.E., Menegaz, B.A., Ramamoorthy, V., Aiyer, R.A., Maywald, R.L., Buford, A.S., Doolittle, D.K., Culotta, K.S., O'Dorisio, J.E., and Ludwig, J.A. (2015). An Oral Formulation of YK-4-279: Preclinical Efficacy and Acquired Resistance Patterns in Ewing Sarcoma. Molecular cancer therapeutics 14, 1591-1604.

Lamonte, G., Tang, X., Chen, J.L., Wu, J., Ding, C.K., Keenan, M.M., Sangokoya, C., Kung, H.N., Ilkayeva, O., Boros, L.G., et al. (2013). Acidosis induces reprogramming of cellular metabolism to mitigate oxidative stress. Cancer & metabolism 1, 23.

Lau, O.S., and Deng, X.W. (2012). The photomorphogenic repressors COP1 and DET1: 20 years later. Trends in plant science 17, 584-593.

Ledoux, S., Yang, R., Friedlander, G., and Laouari, D. (2003). Glucose depletion enhances P- glycoprotein expression in hepatoma cells: role of endoplasmic reticulum stress response. Cancer research 63, 7284-7290.

Lee, H.S., Park, H.J., Lyons, J.C., Griffin, R.J., Auger, E.A., and Song, C.W. (1997). Radiation- induced apoptosis in different pH environments in vitro. International journal of radiation oncology, biology, physics 38, 1079-1087.

Lettice, L.A., Williamson, I., Wiltshire, J.H., Peluso, S., Devenney, P.S., Hill, A.E., Essafi, A., Hagman, J., Mort, R., Grimes, G., et al. (2012). Opposing functions of the ETS factor family define Shh spatial expression in limb buds and underlie polydactyly. Developmental cell 22, 459-467.

Leung, J.Y., and Kim, W.Y. (2013). Stearoyl co-A desaturase 1 as a ccRCC therapeutic target: death by stress. Clinical cancer research: an official journal of the American Association for Cancer Research 19, 3111-3113.

Li, B., and Simon, M.C. (2013). Molecular Pathways: Targeting MYC-induced metabolic

222

reprogramming and oncogenic stress in cancer. Clinical cancer research: an official journal of the American Association for Cancer Research 19, 5835-5841.

Li, D.Q., Ohshiro, K., Reddy, S.D., Pakala, S.B., Lee, M.H., Zhang, Y., Rayala, S.K., and Kumar, R. (2009). E3 ubiquitin ligase COP1 regulates the stability and functions of MTA1. Proceedings of the National Academy of Sciences of the United States of America 106, 17493-17498.

Lilley, B.N., and Ploegh, H.L. (2005). Multiprotein complexes that link dislocation, ubiquitination, and extraction of misfolded proteins from the endoplasmic reticulum membrane. Proceedings of the National Academy of Sciences of the United States of America 102, 14296-14301.

Lin, R., Tao, R., Gao, X., Li, T., Zhou, X., Guan, K.L., Xiong, Y., and Lei, Q.Y. (2013). Acetylation stabilizes ATP-citrate lyase to promote lipid biosynthesis and tumor growth. Molecular cell 51, 506-518.

Linton, A., Cheng, Y.Y., Griggs, K., Kirschner, M.B., Gattani, S., Srikaran, S., Chuan-Hao Kao, S., McCaughan, B.C., Klebe, S., van Zandwijk, N., et al. (2014). An RNAi-based screen reveals PLK1, CDK1 and NDC80 as potential therapeutic targets in malignant pleural mesothelioma. British journal of cancer 110, 510-519.

Liu, X., Greer, C., and Secombe, J. (2014). KDM5 interacts with Foxo to modulate cellular levels of oxidative stress. PLoS genetics 10, e1004676.

Liu, Y., Borchert, G.L., and Phang, J.M. (2004). Polyoma enhancer activator 3, an ets transcription factor, mediates the induction of cyclooxygenase-2 by nitric oxide in colorectal cancer cells. The Journal of biological chemistry 279, 18694-18700.

Liu, Y., Jiang, H., Crawford, H.C., and Hogan, B.L. (2003). Role for ETS domain transcription factors Pea3/Erm in mouse lung development. Developmental biology 261, 10-24.

Liu, Y., Martinez, L., Ebine, K., and Abe, M.K. (2008). Role for mitogen-activated protein kinase p38 alpha in lung epithelial branching morphogenesis. Developmental biology 314, 224-235.

Livet, J., Sigrist, M., Stroebel, S., De Paola, V., Price, S.R., Henderson, C.E., Jessell, T.M., and Arber, S. (2002). ETS gene Pea3 controls the central position and terminal arborization of specific motor neuron pools. Neuron 35, 877-892.

Loenarz, C., and Schofield, C.J. (2008). Expanding chemical biology of 2-oxoglutarate oxygenases. Nature chemical biology 4, 152-156.

Loiselle, F.B., and Casey, J.R. (2010). Measurement of Intracellular pH. Methods in molecular biology (Clifton, NJ) 637, 311-331.

Lopez-Maury, L., Marguerat, S., and Bahler, J. (2008). Tuning gene expression to changing

223

environments: from rapid responses to evolutionary adaptation. Nature reviews Genetics 9, 583- 593.

Lu, B.C., Cebrian, C., Chi, X., Kuure, S., Kuo, R., Bates, C.M., Arber, S., Hassell, J., MacNeil, L., Hoshi, M., et al. (2009). Etv4 and Etv5 are required downstream of GDNF and Ret for kidney branching morphogenesis. Nature genetics 41, 1295-1302.

Lu, H., Forbes, R.A., and Verma, A. (2002). Hypoxia-inducible factor 1 activation by aerobic glycolysis implicates the Warburg effect in carcinogenesis. The Journal of biological chemistry 277, 23111-23115.

Lucas, J.E., Kung, H.N., and Chi, J.T. (2010). Latent factor analysis to discover pathway-associated putative segmental aneuploidies in human cancers. PLoS computational biology 6, e1000920.

Luciani, F., Spada, M., De Milito, A., Molinari, A., Rivoltini, L., Montinaro, A., Marra, M., Lugini, L., Logozzi, M., Lozupone, F., et al. (2004). Effect of proton pump inhibitor pretreatment on resistance of solid tumors to cytotoxic drugs. Journal of the National Cancer Institute 96, 1702- 1713.

Luo, J., Emanuele, M.J., Li, D., Creighton, C.J., Schlabach, M.R., Westbrook, T.F., Wong, K.K., and Elledge, S.J. (2009). A genome-wide RNAi screen identifies multiple synthetic lethal interactions with the Ras oncogene. Cell 137, 835-848.

Ma, S., and Bohnert, H.J. (2007). Integration of Arabidopsis thaliana stress-related transcript profiles, promoter structures, and cell-specific expression. Genome biology 8, R49.

Ma, Y., Creanga, A., Lum, L., and Beachy, P.A. (2006). Prevalence of off-target effects in Drosophila RNA interference screens. Nature 443, 359-363.

MacKeigan, J.P., Murphy, L.O., and Blenis, J. (2005). Sensitized RNAi screen of human kinases and phosphatases identifies new regulators of apoptosis and chemoresistance. Nature cell biology 7, 591-600.

Majmundar, A.J., Wong, W.J., and Simon, M.C. (2010). Hypoxia-inducible factors and the response to hypoxic stress. Molecular cell 40, 294-309.

Makarova, O.V., Makarov, E.M., and Luhrmann, R. (2001). The 65 and 110 kDa SR-related proteins of the U4/U6.U5 tri-snRNP are essential for the assembly of mature spliceosomes. The EMBO journal 20, 2553-2563.

Mannironi, C., Proietto, M., Bufalieri, F., Cundari, E., Alagia, A., Danovska, S., Rinaldi, T., Famiglini, V., Coluccia, A., La Regina, G., et al. (2014). An high-throughput in vivo screening system to select H3K4-specific histone demethylase inhibitors. PloS one 9, e86002.

224

Mao, J., McGlinn, E., Huang, P., Tabin, C.J., and McMahon, A.P. (2009). Fgf-dependent Etv4/5 activity is required for posterior restriction of Sonic Hedgehog and promoting outgrowth of the vertebrate limb. Developmental cell 16, 600-606.

Marine, J.C. (2012). Spotlight on the role of COP1 in tumorigenesis. Nature reviews Cancer 12, 455-464.

Marjon, P.L., Bobrovnikova-Marjon, E.V., and Abcouwer, S.F. (2004). Expression of the pro- angiogenic factors vascular endothelial growth factor and interleukin-8/CXCL8 by human breast carcinomas is responsive to nutrient deprivation and endoplasmic reticulum stress. Molecular cancer 3, 4.

Markolovic, S., Wilkins, S.E., and Schofield, C.J. (2015). Protein Hydroxylation Catalyzed by 2- Oxoglutarate-dependent Oxygenases. The Journal of biological chemistry 290, 20712-20722.

Martinez-Zaguilan, R., Seftor, E.A., Seftor, R.E., Chu, Y.W., Gillies, R.J., and Hendrix, M.J. (1996). Acidic pH enhances the invasive behavior of human melanoma cells. Clinical & experimental metastasis 14, 176-186.

Marusyk, A., Almendro, V., and Polyak, K. (2012). Intra-tumour heterogeneity: a looking glass for cancer? Nature reviews Cancer 12, 323-334.

Matikas, A., Mistriotis, D., Georgoulias, V., and Kotsakis, A. (2015). Current and Future Approaches in the Management of Non-Small-Cell Lung Cancer Patients with Resistance to EGFR TKIs. Clinical lung cancer 16, 252-261.

Matsumoto, K., Obara, N., Ema, M., Horie, M., Naka, A., Takahashi, S., and Imagawa, S. (2009). Antitumor effects of 2-oxoglutarate through inhibition of angiogenesis in a murine tumor model. Cancer science 100, 1639-1647.

McIntyre, A., and Harris, A.L. (2015). Metabolic and hypoxic adaptation to anti-angiogenic therapy: a target for induced essentiality. EMBO molecular medicine 7, 368-379.

Meacham, C.E., and Morrison, S.J. (2013). Tumour heterogeneity and cancer cell plasticity. Nature 501, 328-337.

Mees, G., Sathekge, M., Maes, A., and Van de Wiele, C. (2014). Radiolabelled probes targeting tumor hypoxia for personalized medicine. Current pharmaceutical design 20, 2308-2318.

Mekhail, K., Gunaratnam, L., Bonicalzi, M.E., and Lee, S. (2004). HIF activation by pH-dependent nucleolar sequestration of VHL. Nature cell biology 6, 642-647.

Melvin, A., and Rocha, S. (2012). Chromatin as an oxygen sensor and active player in the hypoxia response. Cellular signalling 24, 35-43.

225

Menendez, J.A., and Lupu, R. (2007). Fatty acid synthase and the lipogenic phenotype in cancer pathogenesis. Nature reviews Cancer 7, 763-777.

Menendez, J.A., Oliveras-Ferraros, C., Cufi, S., Corominas-Faja, B., Joven, J., Martin-Castillo, B., and Vazquez-Martin, A. (2012). Metformin is synthetically lethal with glucose withdrawal in cancer cells. Cell cycle (Georgetown, Tex) 11, 2782-2792.

Menendez, J.A., Vellon, L., Mehmi, I., Oza, B.P., Ropero, S., Colomer, R., and Lupu, R. (2004). Inhibition of fatty acid synthase (FAS) suppresses HER2/neu (erbB-2) oncogene overexpression in cancer cells. Proceedings of the National Academy of Sciences of the United States of America 101, 10715-10720.

Mesquita, D., Barros-Silva, J.D., Santos, J., Skotheim, R.I., Lothe, R.A., Paulo, P., and Teixeira, M.R. (2015). Specific and redundant activities of ETV1 and ETV4 in prostate cancer aggressiveness revealed by co-overexpression cellular contexts. Oncotarget 6, 5217-5236.

Metallo, C.M., Gameiro, P.A., Bell, E.L., Mattaini, K.R., Yang, J., Hiller, K., Jewell, C.M., Johnson, Z.R., Irvine, D.J., Guarente, L., et al. (2012). Reductive glutamine metabolism by IDH1 mediates lipogenesis under hypoxia. Nature 481, 380-384.

Migita, T., Narita, T., Nomura, K., Miyagi, E., Inazuka, F., Matsuura, M., Ushijima, M., Mashima, T., Seimiya, H., Satoh, Y., et al. (2008). ATP citrate lyase: activation and therapeutic implications in non-small cell lung cancer. Cancer research 68, 8547-8554.

Migliorini, D., Bogaerts, S., Defever, D., Vyas, R., Denecker, G., Radaelli, E., Zwolinska, A., Depaepe, V., Hochepied, T., Skarnes, W.C., et al. (2011). Cop1 constitutively regulates c-Jun protein stability and functions as a tumor suppressor in mice. The Journal of clinical investigation 121, 1329-1343.

Mimura, I., Tanaka, T., Wada, Y., Kodama, T., and Nangaku, M. (2011). Pathophysiological response to hypoxia - from the molecular mechanisms of malady to drug discovery: epigenetic regulation of the hypoxic response via hypoxia-inducible factor and histone modifying enzymes. Journal of pharmacological sciences 115, 453-458.

Mizuarai, S., Irie, H., Schmatz, D.M., and Kotani, H. (2008). Integrated genomic and pharmacological approaches to identify synthetic lethal genes as cancer therapeutic targets. Current molecular medicine 8, 774-783.

Mohindra, V., Tripathi, R.K., Yadav, P., Singh, R.K., and Lal, K.K. (2015). Hypoxia induced altered expression of heat shock protein genes (Hsc71, Hsp90alpha and Hsp10) in Indian Catfish, Clarias batrachus (Linnaeus, 1758) under oxidative stress. Molecular biology reports 42, 1197- 1209.

226

Mohr, S., Bakal, C., and Perrimon, N. (2010). Genomic screening with RNAi: results and challenges. Annual review of biochemistry 79, 37-64.

Mohr, S.E., Smith, J.A., Shamu, C.E., Neumuller, R.A., and Perrimon, N. (2014). RNAi screening comes of age: improved techniques and complementary approaches. Nature reviews Molecular cell biology 15, 591-600.

Mueller, B., Klemm, E.J., Spooner, E., Claessen, J.H., and Ploegh, H.L. (2008). SEL1L nucleates a protein complex required for dislocation of misfolded glycoproteins. Proceedings of the National Academy of Sciences of the United States of America 105, 12325-12330.

Mueller, B., Lilley, B.N., and Ploegh, H.L. (2006). SEL1L, the homologue of yeast Hrd3p, is involved in protein dislocation from the mammalian ER. The Journal of cell biology 175, 261-270.

Muller, H.J., and Boos, J. (1998). Use of L-asparaginase in childhood ALL. Critical reviews in oncology/hematology 28, 97-113.

Neri, D., and Supuran, C.T. (2011). Interfering with pH regulation in tumours as a therapeutic strategy. Nature reviews Drug discovery 10, 767-777.

Nijman, S.M. (2011). Synthetic lethality: general principles, utility and detection using genetic screens in human cells. FEBS letters 585, 1-6.

Niu, X., Zhang, T., Liao, L., Zhou, L., Lindner, D.J., Zhou, M., Rini, B., Yan, Q., and Yang, H. (2012). The von Hippel-Lindau tumor suppressor protein regulates gene expression and tumor growth through histone demethylase JARID1C. Oncogene 31, 776-786.

Nosho, K., Yoshida, M., Yamamoto, H., Taniguchi, H., Adachi, Y., Mikami, M., Hinoda, Y., and Imai, K. (2005). Association of Ets-related transcriptional factor E1AF expression with overexpression of matrix metalloproteinases, COX-2 and iNOS in the early stage of colorectal carcinogenesis. Carcinogenesis 26, 892-899.

Nowik, M., Lecca, M.R., Velic, A., Rehrauer, H., Brandli, A.W., and Wagner, C.A. (2008). Genome-wide gene expression profiling reveals renal genes regulated during metabolic acidosis. Physiological genomics 32, 322-334.

Ntambi, J.M., Miyazaki, M., and Dobrzyn, A. (2004). Regulation of stearoyl-CoA desaturase expression. Lipids 39, 1061-1065.

Oh, S., Shin, S., and Janknecht, R. (2012). ETV1, 4 and 5: an oncogenic subfamily of ETS transcription factors. Biochimica et biophysica acta 1826, 1-12.

O'Hagan, R.C., Tozer, R.G., Symons, M., McCormick, F., and Hassell, J.A. (1996). The activity of the Ets transcription factor PEA3 is regulated by two distinct MAPK cascades. Oncogene 13,

227

1323-1333.

Olma, M.H., Roy, M., Le Bihan, T., Sumara, I., Maerki, S., Larsen, B., Quadroni, M., Peter, M., Tyers, M., and Pintard, L. (2009). An interaction network of the mammalian COP9 signalosome identifies Dda1 as a core subunit of multiple Cul4-based E3 ligases. Journal of cell science 122, 1035-1044.

Orlandi, R., Cattaneo, M., Troglio, F., Casalini, P., Ronchini, C., Menard, S., and Biunno, I. (2002). SEL1L expression decreases breast tumor cell aggressiveness in vivo and in vitro. Cancer research 62, 567-574.

Paddison, P.J., Silva, J.M., Conklin, D.S., Schlabach, M., Li, M., Aruleba, S., Balija, V., O'Shaughnessy, A., Gnoj, L., Scobie, K., et al. (2004). A resource for large-scale RNA-interference- based screens in mammals. Nature 428, 427-431.

Palazon, A., Goldrath, A.W., Nizet, V., and Johnson, R.S. (2014). HIF transcription factors, inflammation, and immunity. Immunity 41, 518-528.

Pan, J., Zhang, J., Hill, A., Lapan, P., Berasi, S., Bates, B., Miller, C., and Haney, S. (2013). A kinome-wide siRNA screen identifies multiple roles for protein kinases in hypoxic stress adaptation, including roles for IRAK4 and GAK in protection against apoptosis in VHL-/- renal carcinoma cells, despite activation of the NF-kappaB pathway. Journal of biomolecular screening 18, 782-796.

Park, H.J., Lee, S.H., Chung, H., Rhee, Y.H., Lim, B.U., Ha, S.W., Griffin, R.J., Lee, H.S., Song, C.W., and Choi, E.K. (2003). Influence of environmental pH on G2-phase arrest caused by ionizing radiation. Radiation research 159, 86-93.

Park, K.W., Waki, H., Choi, S.P., Park, K.M., and Tontonoz, P. (2010). The small molecule phenamil is a modulator of adipocyte differentiation and PPARgamma expression. Journal of lipid research 51, 2775-2784.

Parks, S.K., Mazure, N.M., Counillon, L., and Pouyssegur, J. (2013). Hypoxia promotes tumor cell survival in acidic conditions by preserving ATP levels. Journal of cellular physiology 228, 1854- 1862.

Pei, J., Balsara, B.R., Li, W., Litwin, S., Gabrielson, E., Feder, M., Jen, J., and Testa, J.R. (2001). Genomic imbalances in human lung adenocarcinomas and squamous cell carcinomas. Genes, chromosomes & cancer 31, 282-287.

Pellecchia, A., Pescucci, C., De Lorenzo, E., Luceri, C., Passaro, N., Sica, M., Notaro, R., and De Angioletti, M. (2012). Overexpression of ETV4 is oncogenic in prostate cells through promotion of both cell proliferation and epithelial to mesenchymal transition. Oncogenesis 1, e20.

228

Peppicelli, S., Bianchini, F., and Calorini, L. (2014). Extracellular acidity, a "reappreciated" trait of tumor environment driving malignancy: perspectives in diagnosis and therapy. Cancer metastasis reviews 33, 823-832.

Peppicelli, S., Bianchini, F., Contena, C., Tombaccini, D., and Calorini, L. (2013). Acidic pH via NF-kappaB favours VEGF-C expression in human melanoma cells. Clinical & experimental metastasis 30, 957-967.

Perez-Sayans, M., Garcia-Garcia, A., Scozzafava, A., and Supuran, C.T. (2012). Inhibition of V- ATPase and carbonic anhydrases as interference strategy with tumor acidification processes. Current pharmaceutical design 18, 1407-1413.

Pick, E., Lau, O.S., Tsuge, T., Menon, S., Tong, Y., Dohmae, N., Plafker, S.M., Deng, X.W., and Wei, N. (2007). Mammalian DET1 regulates Cul4A activity and forms stable complexes with E2 ubiquitin-conjugating enzymes. Molecular and cellular biology 27, 4708-4719.

Pickup, M.W., Mouw, J.K., and Weaver, V.M. (2014). The extracellular matrix modulates the hallmarks of cancer. EMBO reports 15, 1243-1253.

Pollard, P.J., Loenarz, C., Mole, D.R., McDonough, M.A., Gleadle, J.M., Schofield, C.J., and Ratcliffe, P.J. (2008). Regulation of Jumonji-domain-containing histone demethylases by hypoxia- inducible factor (HIF)-1alpha. The Biochemical journal 416, 387-394.

Pond, B.B., Galeffi, F., Ahrens, R., and Schwartz-Bloom, R.D. (2004). Chloride transport inhibitors influence recovery from oxygen-glucose deprivation-induced cellular injury in adult hippocampus. Neuropharmacology 47, 253-262.

Prabhu, A., Sarcar, B., Kahali, S., Shan, Y., and Chinnaiyan, P. (2012). Targeting the unfolded protein response in glioblastoma cells with the fusion protein EGF-SubA. PloS one 7, e52265.

Qi, H., Labrie, Y., Grenier, J., Fournier, A., Fillion, C., and Labrie, C. (2001). Androgens induce expression of SPAK, a STE20/SPS1-related kinase, in LNCaP human prostate cancer cells. Molecular and cellular endocrinology 182, 181-192.

Qi, L., Heredia, J.E., Altarejos, J.Y., Screaton, R., Goebel, N., Niessen, S., Macleod, I.X., Liew, C.W., Kulkarni, R.N., Bain, J., et al. (2006). TRB3 links the E3 ubiquitin ligase COP1 to lipid metabolism. Science (New York, NY) 312, 1763-1766.

Qian, X., Hu, J., Zhao, J., and Chen, H. (2015). ATP citrate lyase expression is associated with advanced stage and prognosis in gastric adenocarcinoma. International journal of clinical and experimental medicine 8, 7855-7860.

Rahim, S., Beauchamp, E.M., Kong, Y., Brown, M.L., Toretsky, J.A., and Uren, A. (2011). YK-4-279 inhibits ERG and ETV1 mediated prostate cancer cell invasion. PloS one 6, e19343.

229

Rahim, S., Minas, T., Hong, S.H., Justvig, S., Celik, H., Kont, Y.S., Han, J., Kallarakal, A.T., Kong, Y., Rudek, M.A., et al. (2014). A small molecule inhibitor of ETV1, YK-4-279, prevents prostate cancer growth and metastasis in a mouse xenograft model. PloS one 9, e114260.

Rasmussen, P.B., and Staller, P. (2014). The KDM5 family of histone demethylases as targets in oncology drug discovery. Epigenomics 6, 277-286.

Redon, R., Hussenet, T., Bour, G., Caulee, K., Jost, B., Muller, D., Abecassis, J., and du Manoir, S. (2002). Amplicon mapping and transcriptional analysis pinpoint cyclin L as a candidate oncogene in head and neck cancer. Cancer research 62, 6211-6217.

Robey, I.F., Baggett, B.K., Kirkpatrick, N.D., Roe, D.J., Dosescu, J., Sloane, B.F., Hashim, A.I., Morse, D.L., Raghunand, N., Gatenby, R.A., et al. (2009). Bicarbonate increases tumor pH and inhibits spontaneous metastases. Cancer research 69, 2260-2268.

Roesch, A., Vultur, A., Bogeski, I., Wang, H., Zimmermann, K.M., Speicher, D., Korbel, C., Laschke, M.W., Gimotty, P.A., Philipp, S.E., et al. (2013). Overcoming intrinsic multidrug resistance in melanoma by blocking the mitochondrial respiratory chain of slow-cycling JARID1B(high) cells. Cancer cell 23, 811-825.

Rofstad, E.K., Mathiesen, B., Kindem, K., and Galappathi, K. (2006). Acidic extracellular pH promotes experimental metastasis of human melanoma cells in athymic nude mice. Cancer research 66, 6699-6707.

Rottinger, E.M., and Mendonca, M. (1982). Radioresistance secondary to low pH in human glial cells and Chinese hamster ovary cells. International journal of radiation oncology, biology, physics 8, 1309-1314.

Ru, Q., Tian, X., Wu, Y.X., Wu, R.H., Pi, M.S., and Li, C.Y. (2014). Voltage-gated and ATP- sensitive K+ channels are associated with cell proliferation and tumorigenesis of human glioma. Oncology reports 31, 842-848.

Rucki, A.A., and Zheng, L. (2014). Pancreatic cancer stroma: understanding biology leads to new therapeutic strategies. World journal of gastroenterology 20, 2237-2246.

Ruderman, N., and Prentki, M. (2004). AMP kinase and malonyl-CoA: targets for therapy of the metabolic syndrome. Nature reviews Drug discovery 3, 340-351.

Rysman, E., Brusselmans, K., Scheys, K., Timmermans, L., Derua, R., Munck, S., Van Veldhoven, P.P., Waltregny, D., Daniels, V.W., Machiels, J., et al. (2010). De novo lipogenesis protects cancer cells from free radicals and chemotherapeutics by promoting membrane lipid saturation. Cancer research 70, 8117-8126.

230

Sauvant, C., Nowak, M., Wirth, C., Schneider, B., Riemann, A., Gekle, M., and Thews, O. (2008). Acidosis induces multi-drug resistance in rat prostate cancer cells (AT1) in vitro and in vivo by increasing the activity of the p-glycoprotein via activation of p38. International journal of cancer Journal international du cancer 123, 2532-2542.

Sayegh, J., Cao, J., Zou, M.R., Morales, A., Blair, L.P., Norcia, M., Hoyer, D., Tackett, A.J., Merkel, J.S., and Yan, Q. (2013). Identification of small molecule inhibitors of Jumonji AT-rich interactive domain 1B (JARID1B) histone demethylase by a sensitive high throughput screen. The Journal of biological chemistry 288, 9408-9417.

Schafer, Z.T., Grassian, A.R., Song, L., Jiang, Z., Gerhart-Hines, Z., Irie, H.Y., Gao, S., Puigserver, P., and Brugge, J.S. (2009). Antioxidant and oncogene rescue of metabolic defects caused by loss of matrix attachment. Nature 461, 109-113.

Schlabach, M.R., Luo, J., Solimini, N.L., Hu, G., Xu, Q., Li, M.Z., Zhao, Z., Smogorzewska, A., Sowa, M.E., Ang, X.L., et al. (2008). Cancer proliferation gene discovery through functional genomics. Science (New York, NY) 319, 620-624.

Schmaltz, C., Hardenbergh, P.H., Wells, A., and Fisher, D.E. (1998). Regulation of proliferation- survival decisions during tumor cell hypoxia. Molecular and cellular biology 18, 2845-2854.

Schornack, P.A., and Gillies, R.J. (2003). Contributions of cell metabolism and H+ diffusion to the acidic pH of tumors. Neoplasia (New York, NY) 5, 135-145.

Schultz, N., Marenstein, D.R., De Angelis, D.A., Wang, W.Q., Nelander, S., Jacobsen, A., Marks, D.S., Massague, J., and Sander, C. (2011). Off-target effects dominate a large-scale RNAi screen for modulators of the TGF-beta pathway and reveal microRNA regulation of TGFBR2. Silence 2, 3.

Scott, L., Lamb, J., Smith, S., and Wheatley, D.N. (2000). Single amino acid (arginine) deprivation: rapid and selective death of cultured transformed and malignant cells. British journal of cancer 83, 800-810.

Semenza, G.L. (2013). HIF-1 mediates metabolic responses to intratumoral hypoxia and oncogenic mutations. The Journal of clinical investigation 123, 3664-3671.

Shackelford, D.B., and Shaw, R.J. (2009). The LKB1-AMPK pathway: metabolism and growth control in tumour suppression. Nature reviews Cancer 9, 563-575.

Sharma, S.V., Lee, D.Y., Li, B., Quinlan, M.P., Takahashi, F., Maheswaran, S., McDermott, U., Azizian, N., Zou, L., Fischbach, M.A., et al. (2010). A chromatin-mediated reversible drug-tolerant state in cancer cell subpopulations. Cell 141, 69-80.

Sharp, T.V., Al-Attar, A., Foxler, D.E., Ding, L., de, A.V.T.Q., Zhang, Y., Nijmeh, H.S., Webb,

231

T.M., Nicholson, A.G., Zhang, Q., et al. (2008). The chromosome 3p21.3-encoded gene, LIMD1, is a critical tumor suppressor involved in human lung cancer development. Proceedings of the National Academy of Sciences of the United States of America 105, 19932-19937.

Sharp, T.V., Munoz, F., Bourboulia, D., Presneau, N., Darai, E., Wang, H.W., Cannon, M., Butcher, D.N., Nicholson, A.G., Klein, G., et al. (2004). LIM domains-containing protein 1 (LIMD1), a tumor suppressor encoded at chromosome 3p21.3, binds pRB and represses E2F- driven transcription. Proceedings of the National Academy of Sciences of the United States of America 101, 16531-16536.

Shaw, R.J., and Cantley, L.C. (2012). Decoding key nodes in the metabolism of cancer cells: sugar & spice and all things nice. F1000 biology reports 4, 2.

Shi, Q., Le, X., Wang, B., Abbruzzese, J.L., Xiong, Q., He, Y., and Xie, K. (2001). Regulation of vascular endothelial growth factor expression by acidosis in human cancer cells. Oncogene 20, 3751-3756.

Shim, H., Chun, Y.S., Lewis, B.C., and Dang, C.V. (1998). A unique glucose-dependent apoptotic pathway induced by c-Myc. Proceedings of the National Academy of Sciences of the United States of America 95, 1511-1516.

Shin, S., Buel, G.R., Wolgamott, L., Plas, D.R., Asara, J.M., Blenis, J., and Yoon, S.O. (2015). ERK2 Mediates Metabolic Stress Response to Regulate Cell Fate. Molecular cell 59, 382-398.

Shindoh, M., Higashino, F., and Kohgo, T. (2004). E1AF, an ets-oncogene family transcription factor. Cancer letters 216, 1-8.

Shmakova, A., Batie, M., Druker, J., and Rocha, S. (2014). Chromatin and oxygen sensing in the context of JmjC histone demethylases. The Biochemical journal 462, 385-395.

Siegel, R., Ma, J., Zou, Z., and Jemal, A. (2014). Cancer statistics, 2014. CA: a cancer journal for clinicians 64, 9-29.

Sigoillot, F.D., and King, R.W. (2011). Vigilance and validation: Keys to success in RNAi screening. ACS chemical biology 6, 47-60.

Sigoillot, F.D., Lyman, S., Huckins, J.F., Adamson, B., Chung, E., Quattrochi, B., and King, R.W. (2012). A bioinformatics method identifies prominent off-targeted transcripts in RNAi screens. Nature methods 9, 363-366.

Silva, A.S., Yunes, J.A., Gillies, R.J., and Gatenby, R.A. (2009). The potential role of systemic buffers in reducing intratumoral extracellular pH and acid-mediated invasion. Cancer research 69, 2677-2684.

232

Silva, J.M., Li, M.Z., Chang, K., Ge, W., Golding, M.C., Rickles, R.J., Siolas, D., Hu, G., Paddison, P.J., Schlabach, M.R., et al. (2005). Second-generation shRNA libraries covering the mouse and human genomes. Nature genetics 37, 1281-1288.

Sim, A.T., and Hardie, D.G. (1988). The low activity of acetyl-CoA carboxylase in basal and glucagon-stimulated hepatocytes is due to phosphorylation by the AMP-activated protein kinase and not cyclic AMP-dependent protein kinase. FEBS letters 233, 294-298.

Sims, D., Mendes-Pereira, A.M., Frankum, J., Burgess, D., Cerone, M.A., Lombardelli, C., Mitsopoulos, C., Hakas, J., Murugaesu, N., Isacke, C.M., et al. (2011). High-throughput RNA interference screening using pooled shRNA libraries and next generation sequencing. Genome biology 12, R104.

Singleton, P.A., Moss, J., Karp, D.D., Atkins, J.T., and Janku, F. (2015). The mu opioid receptor: A new target for cancer therapy? Cancer 121, 2681-2688.

Skvortsova, I., Debbage, P., Kumar, V., and Skvortsov, S. (2015). Radiation resistance: Cancer stem cells (CSCs) and their enigmatic pro-survival signaling. Seminars in cancer biology.

Sonveaux, P., Vegran, F., Schroeder, T., Wergin, M.C., Verrax, J., Rabbani, Z.N., De Saedeleer, C.J., Kennedy, K.M., Diepart, C., Jordan, B.F., et al. (2008). Targeting lactate-fueled respiration selectively kills hypoxic tumor cells in mice. The Journal of clinical investigation 118, 3930-3942.

Sosa, M.S., Bragado, P., and Aguirre-Ghiso, J.A. (2014). Mechanisms of disseminated cancer cell dormancy: an awakening field. Nature reviews Cancer 14, 611-622.

Sounni, N.E., Cimino, J., Blacher, S., Primac, I., Truong, A., Mazzucchelli, G., Paye, A., Calligaris, D., Debois, D., De Tullio, P., et al. (2014). Blocking lipid synthesis overcomes tumor regrowth and metastasis after antiangiogenic therapy withdrawal. Cell metabolism 20, 280-294.

Spendlove, I., Al-Attar, A., Watherstone, O., Webb, T.M., Ellis, I.O., Longmore, G.D., and Sharp, T.V. (2008). Differential subcellular localisation of the tumour suppressor protein LIMD1 in breast cancer correlates with patient survival. International journal of cancer Journal international du cancer 123, 2247-2253.

Spitz, D.R., Sim, J.E., Ridnour, L.A., Galoforo, S.S., and Lee, Y.J. (2000). Glucose deprivation- induced oxidative stress in human tumor cells. A fundamental defect in metabolism? Annals of the New York Academy of Sciences 899, 349-362.

Stegmeier, F., Hu, G., Rickles, R.J., Hannon, G.J., and Elledge, S.J. (2005). A lentiviral microRNA- based system for single-copy polymerase II-regulated RNA interference in mammalian cells. Proceedings of the National Academy of Sciences of the United States of America 102, 13212- 13217.

233

Stein, J., Majores, M., Rohde, M., Lim, S., Schneider, S., Krappe, E., Ellinger, J., Dietel, M., Stephan, C., Jung, K., et al. (2014). KDM5C is overexpressed in prostate cancer and is a prognostic marker for prostate-specific antigen-relapse following radical prostatectomy. The American journal of pathology 184, 2430-2437.

Sticht, C., Hofele, C., Flechtenmacher, C., Bosch, F.X., Freier, K., Lichter, P., and Joos, S. (2005). Amplification of Cyclin L1 is associated with lymph node metastases in head and neck squamous cell carcinoma (HNSCC). British journal of cancer 92, 770-774.

Su, C.H., Zhao, R., Velazquez-Torres, G., Chen, J., Gully, C., Yeung, S.C., and Lee, M.H. (2010). Nuclear export regulation of COP1 by 14-3-3sigma in response to DNA damage. Molecular cancer 9, 243.

Sun, H., Qin, B., Liu, T., Wang, Q., Liu, J., Wang, J., Lin, X., Yang, Y., Taing, L., Rao, P.K., et al. (2013). CistromeFinder for ChIP-seq and DNase-seq data reuse. Bioinformatics (Oxford, England) 29, 1352-1354.

Sun, S., Shi, G., Han, X., Francisco, A.B., Ji, Y., Mendonca, N., Liu, X., Locasale, J.W., Simpson, K.W., Duhamel, G.E., et al. (2014). Sel1L is indispensable for mammalian endoplasmic reticulum- associated degradation, endoplasmic reticulum homeostasis, and survival. Proceedings of the National Academy of Sciences of the United States of America 111, E582-591.

Suzuki, K., Bose, P., Leong-Quong, R.Y., Fujita, D.J., and Riabowol, K. (2010). REAP: A two minute cell fractionation method. BMC research notes 3, 294.

Tang, X., Keenan, M.M., Wu, J., Lin, C.A., Dubois, L., Thompson, J.W., Freedland, S.J., Murphy, S.K., and Chi, J.T. (2015). Comprehensive profiling of amino acid response uncovers unique methionine-deprived response dependent on intact creatine biosynthesis. PLoS genetics 11, e1005158.

Tang, X., Lin, C.C., Spasojevic, I., Iversen, E.S., Chi, J.T., and Marks, J.R. (2014). A joint analysis of metabolomics and genetics of breast cancer. Breast cancer research: BCR 16, 415.

Tang, X., Lucas, J.E., Chen, J.L., LaMonte, G., Wu, J., Wang, M.C., Koumenis, C., and Chi, J.T. (2012). Functional interaction between responses to lactic acidosis and hypoxia regulates genomic transcriptional outputs. Cancer research 72, 491-502.

Taylor, C.T., and Cummins, E.P. (2009). The role of NF-kappaB in hypoxia-induced gene expression. Annals of the New York Academy of Sciences 1177, 178-184.

Taylor, J., and Woodcock, S. (2015). A Perspective on the Future of High-Throughput RNAi Screening: Will CRISPR Cut Out the Competition or Can RNAi Help Guide the Way? Journal of biomolecular screening 20, 1040-1051.

234

Tennant, D.A., Frezza, C., MacKenzie, E.D., Nguyen, Q.D., Zheng, L., Selak, M.A., Roberts, D.L., Dive, C., Watson, D.G., Aboagye, E.O., et al. (2009). Reactivating HIF prolyl hydroxylases under hypoxia results in metabolic catastrophe and cell death. Oncogene 28, 4009-4021.

Tennant, D.A., and Gottlieb, E. (2010). HIF prolyl hydroxylase-3 mediates alpha-ketoglutarate- induced apoptosis and tumor suppression. Journal of molecular medicine (Berlin, Germany) 88, 839-849.

Terunuma, A., Putluri, N., Mishra, P., Mathe, E.A., Dorsey, T.H., Yi, M., Wallace, T.A., Issaq, H.J., Zhou, M., Killian, J.K., et al. (2014). MYC-driven accumulation of 2-hydroxyglutarate is associated with breast cancer prognosis. The Journal of clinical investigation 124, 398-412.

Thews, O., Gassner, B., Kelleher, D.K., Schwerdt, G., and Gekle, M. (2006). Impact of extracellular acidity on the activity of P-glycoprotein and the cytotoxicity of chemotherapeutic drugs. Neoplasia (New York, NY) 8, 143-152.

Thompson, J.M., Nguyen, Q.H., Singh, M., and Razorenova, O.V. (2015). Approaches to identifying synthetic lethal interactions in cancer. The Yale journal of biology and medicine 88, 145-155.

To, K.K., Sedelnikova, O.A., Samons, M., Bonner, W.M., and Huang, L.E. (2006). The phosphorylation status of PAS-B distinguishes HIF-1alpha from HIF-2alpha in NBS1 repression. The EMBO journal 25, 4784-4794.

Tomlins, S.A., Mehra, R., Rhodes, D.R., Smith, L.R., Roulston, D., Helgeson, B.E., Cao, X., Wei, J.T., Rubin, M.A., Shah, R.B., et al. (2006). TMPRSS2:ETV4 gene fusions define a third molecular subtype of prostate cancer. Cancer research 66, 3396-3400.

Tong, L., and Harwood, H.J., Jr. (2006). Acetyl-coenzyme A carboxylases: versatile targets for drug discovery. Journal of cellular biochemistry 99, 1476-1488.

Ushioda, R., Hoseki, J., and Nagata, K. (2013). Glycosylation-independent ERAD pathway serves as a backup system under ER stress. Molecular biology of the cell 24, 3155-3163.

Uto, Y. (2015). Recent progress in the discovery and development of stearoyl CoA desaturase inhibitors. Chemistry and physics of lipids.

Vander Heiden, M.G. (2011). Targeting cancer metabolism: a therapeutic window opens. Nature reviews Drug discovery 10, 671-684.

Vander Heiden, M.G., Cantley, L.C., and Thompson, C.B. (2009). Understanding the Warburg effect: the metabolic requirements of cell proliferation. Science (New York, NY) 324, 1029-1033.

Visioli, F., Wang, Y., Alam, G.N., Ning, Y., Rados, P.V., Nor, J.E., and Polverini, P.J. (2014).

235

Glucose-regulated protein 78 (Grp78) confers chemoresistance to tumor endothelial cells under acidic stress. PloS one 9, e101053.

Vitari, A.C., Leong, K.G., Newton, K., Yee, C., O'Rourke, K., Liu, J., Phu, L., Vij, R., Ferrando, R., Couto, S.S., et al. (2011). COP1 is a tumour suppressor that causes degradation of ETS transcription factors. Nature 474, 403-406. von Roemeling, C.A., Marlow, L.A., Wei, J.J., Cooper, S.J., Caulfield, T.R., Wu, K., Tan, W.W., Tun, H.W., and Copland, J.A. (2013). Stearoyl-CoA desaturase 1 is a novel molecular therapeutic target for clear cell renal cell carcinoma. Clinical cancer research: an official journal of the American Association for Cancer Research 19, 2368-2380.

Vousden, K.H., and Lu, X. (2002). Live or let die: the cell's response to p53. Nature reviews Cancer 2, 594-604.

Wade, M. (2015). High-Throughput Silencing Using the CRISPR-Cas9 System: A Review of the Benefits and Challenges. Journal of biomolecular screening 20, 1027-1039.

Wajapeyee, N., Deibler, S.K., and Green, M.R. (2013). Genome-wide RNAi screening to identify regulators of oncogene-induced cellular senescence. Methods in molecular biology (Clifton, NJ) 965, 373-382.

Wajapeyee, N., Serra, R.W., Zhu, X., Mahalingam, M., and Green, M.R. (2008). Oncogenic BRAF induces senescence and apoptosis through pathways mediated by the secreted protein IGFBP7. Cell 132, 363-374.

Wakil, S.J., and Abu-Elheiga, L.A. (2009). Fatty acid metabolism: target for metabolic syndrome. Journal of lipid research 50 Suppl, S138-143.

Walenta, S., and Mueller-Klieser, W.F. (2004). Lactate: mirror and motor of tumor malignancy. Seminars in radiation oncology 14, 267-274.

Walenta, S., Wetterling, M., Lehrke, M., Schwickert, G., Sundfor, K., Rofstad, E.K., and Mueller- Klieser, W. (2000). High lactate levels predict likelihood of metastases, tumor recurrence, and restricted patient survival in human cervical cancers. Cancer research 60, 916-921.

Wang, J.H., Chen, W.L., Li, J.M., Wu, S.F., Chen, T.L., Zhu, Y.M., Zhang, W.N., Li, Y., Qiu, Y.P., Zhao, A.H., et al. (2013). Prognostic significance of 2-hydroxyglutarate levels in acute myeloid leukemia in China. Proceedings of the National Academy of Sciences of the United States of America 110, 17017-17022.

Wang, L., Mao, Y., Du, G., He, C., and Han, S. (2015). Overexpression of JARID1B is associated with poor prognosis and chemotherapy resistance in epithelial ovarian cancer. Tumour biology: the journal of the International Society for Oncodevelopmental Biology and Medicine 36, 2465-

236

2472.

Wang, S.H., and Lin, S.Y. (2013). Tumor dormancy: potential therapeutic target in tumor recurrence and metastasis prevention. Experimental hematology & oncology 2, 29.

Wang, Y., Engels, I.H., Knee, D.A., Nasoff, M., Deveraux, Q.L., and Quon, K.C. (2004). Synthetic lethal targeting of MYC by activation of the DR5 death receptor pathway. Cancer cell 5, 501-512.

Warburg, O. (1956). On respiratory impairment in cancer cells. Science (New York, NY) 124, 269- 270.

Webb, B.A., Chimenti, M., Jacobson, M.P., and Barber, D.L. (2011). Dysregulated pH: a perfect storm for cancer progression. Nature reviews Cancer 11, 671-677.

Wei, G.H., Badis, G., Berger, M.F., Kivioja, T., Palin, K., Enge, M., Bonke, M., Jolma, A., Varjosalo, M., Gehrke, A.R., et al. (2010). Genome-wide analysis of ETS-family DNA-binding in vitro and in vivo. The EMBO journal 29, 2147-2160.

Weissman, A.M. (2001). Themes and variations on ubiquitylation. Nature reviews Molecular cell biology 2, 169-178.

Wellen, K.E., Hatzivassiliou, G., Sachdeva, U.M., Bui, T.V., Cross, J.R., and Thompson, C.B. (2009). ATP-citrate lyase links cellular metabolism to histone acetylation. Science (New York, NY) 324, 1076-1080.

Wells, A., Griffith, L., Wells, J.Z., and Taylor, D.P. (2013). The dormancy dilemma: quiescence versus balanced proliferation. Cancer research 73, 3811-3816.

Wertz, I.E., O'Rourke, K.M., Zhang, Z., Dornan, D., Arnott, D., Deshaies, R.J., and Dixit, V.M. (2004). Human De-etiolated-1 regulates c-Jun by assembling a CUL4A ubiquitin ligase. Science (New York, NY) 303, 1371-1374.

Westbrook, T.F., Stegmeier, F., and Elledge, S.J. (2005). Dissecting cancer pathways and vulnerabilities with RNAi. Cold Spring Harbor symposia on quantitative biology 70, 435-444.

White, E. (2013). Exploiting the bad eating habits of Ras-driven cancers. Genes & development 27, 2065-2071.

Wilson, W.R., and Hay, M.P. (2011). Targeting hypoxia in cancer therapy. Nature reviews Cancer 11, 393-410.

Winteringham, L.N., Kobelke, S., Williams, J.H., Ingley, E., and Klinken, S.P. (2004). Myeloid Leukemia Factor 1 inhibits erythropoietin-induced differentiation, cell cycle exit and p27Kip1 accumulation. Oncogene 23, 5105-5109.

237

Wise, D.R., Ward, P.S., Shay, J.E., Cross, J.R., Gruber, J.J., Sachdeva, U.M., Platt, J.M., DeMatteo, R.G., Simon, M.C., and Thompson, C.B. (2011). Hypoxia promotes isocitrate dehydrogenase- dependent carboxylation of alpha-ketoglutarate to citrate to support cell growth and viability. Proceedings of the National Academy of Sciences of the United States of America 108, 19611- 19616.

Wojtkowiak, J.W., Rothberg, J.M., Kumar, V., Schramm, K.J., Haller, E., Proemsey, J.B., Lloyd, M.C., Sloane, B.F., and Gillies, R.J. (2012). Chronic autophagy is a cellular adaptation to tumor acidic pH microenvironments. Cancer research 72, 3938-3947.

Wollenick, K., Hu, J., Kristiansen, G., Schraml, P., Rehrauer, H., Berchner-Pfannschmidt, U., Fandrey, J., Wenger, R.H., and Stiehl, D.P. (2012). Synthetic transactivation screening reveals ETV4 as broad coactivator of hypoxia-inducible factor signaling. Nucleic acids research 40, 1928- 1943.

Wong, P., Kleemann, H.W., and Tannock, I.F. (2002). Cytostatic potential of novel agents that inhibit the regulation of intracellular pH. British journal of cancer 87, 238-245.

Wong, W.J., Qiu, B., Nakazawa, M.S., Qing, G., and Simon, M.C. (2013). MYC degradation under low O2 tension promotes survival by evading hypoxia-induced cell death. Molecular and cellular biology 33, 3494-3504.

Xia, X., Lemieux, M.E., Li, W., Carroll, J.S., Brown, M., Liu, X.S., and Kung, A.L. (2009). Integrative analysis of HIF binding and transactivation reveals its role in maintaining histone methylation homeostasis. Proceedings of the National Academy of Sciences of the United States of America 106, 4260-4265.

Xie, S., Zhou, F., Wang, J., Cao, H., Chen, Y., Liu, X., Zhang, Z., Dai, J., and He, X. (2015). Functional polymorphisms of ATP citrate lyase gene predicts clinical outcome of patients with advanced colorectal cancer. World journal of surgical oncology 13, 42.

Xu, L., and Fidler, I.J. (2000). Acidic pH-induced elevation in interleukin 8 expression by human ovarian carcinoma cells. Cancer research 60, 4610-4616.

Xu, W., Yang, H., Liu, Y., Yang, Y., Wang, P., Kim, S.H., Ito, S., Yang, C., Wang, P., Xiao, M.T., et al. (2011). Oncometabolite 2-hydroxyglutarate is a competitive inhibitor of alpha-ketoglutarate- dependent dioxygenases. Cancer cell 19, 17-30.

Yan, H., Parsons, D.W., Jin, G., McLendon, R., Rasheed, B.A., Yuan, W., Kos, I., Batinic-Haberle, I., Jones, S., Riggins, G.J., et al. (2009). IDH1 and IDH2 mutations in gliomas. The New England journal of medicine 360, 765-773.

Yang, J., Ledaki, I., Turley, H., Gatter, K.C., Montero, J.C., Li, J.L., and Harris, A.L. (2009). Role of

238

hypoxia-inducible factors in epigenetic regulation via histone demethylases. Annals of the New York Academy of Sciences 1177, 185-197.

Yang, Q., Wang, H., Chen, S., Lan, X., Xiao, H., Shi, H., and Ma, Y. (2015). Fiber-Optic-Based Micro-Probe Using Hexagonal 1-in-6 Fiber Configuration for Intracellular Single-Cell pH Measurement. Analytical chemistry 87, 7171-7179.

Ye, J., and Koumenis, C. (2009). ATF4, an ER stress and hypoxia-inducible transcription factor and its potential role in hypoxia tolerance and tumorigenesis. Current molecular medicine 9, 411- 416.

Yeh, A.C., and Ramaswamy, S. (2015). Mechanisms of Cancer Cell Dormancy-Another Hallmark of Cancer? Cancer research.

Yoshino, S., Hara, T., Weng, J.S., Takahashi, Y., Seiki, M., and Sakamoto, T. (2012). Genetic screening of new genes responsible for cellular adaptation to hypoxia using a genome-wide shRNA library. PloS one 7, e35590.

Yu, B., Swatkoski, S., Holly, A., Lee, L.C., Giroux, V., Lee, C.S., Hsu, D., Smith, J.L., Yuen, G., Yue, J., et al. (2015). Oncogenesis driven by the Ras/Raf pathway requires the SUMO E2 ligase Ubc9. Proceedings of the National Academy of Sciences of the United States of America 112, E1724- 1733.

Yuen, H.F., Chan, Y.K., Grills, C., McCrudden, C.M., Gunasekharan, V., Shi, Z., Wong, A.S., Lappin, T.R., Chan, K.W., Fennell, D.A., et al. (2011). Polyomavirus enhancer activator 3 protein promotes breast cancer metastatic progression through Snail-induced epithelial-mesenchymal transition. The Journal of pathology 224, 78-89.

Yun, J., Rago, C., Cheong, I., Pagliarini, R., Angenendt, P., Rajagopalan, H., Schmidt, K., Willson, J.K., Markowitz, S., Zhou, S., et al. (2009). Glucose deprivation contributes to the development of KRAS pathway mutations in tumor cells. Science (New York, NY) 325, 1555-1559.

Zaidi, N., Swinnen, J.V., and Smans, K. (2012). ATP-citrate lyase: a key player in cancer metabolism. Cancer research 72, 3709-3714.

Zeng, Y., Wagner, E.J., and Cullen, B.R. (2002). Both natural and designed micro RNAs can inhibit the expression of cognate mRNAs when expressed in human cells. Molecular cell 9, 1327-1333.

Zhan, Y., Ginanni, N., Tota, M.R., Wu, M., Bays, N.W., Richon, V.M., Kohl, N.E., Bachman, E.S., Strack, P.R., and Krauss, S. (2008). Control of cell growth and survival by enzymes of the fatty acid synthesis pathway in HCT-116 colon cancer cells. Clinical cancer research: an official journal of the American Association for Cancer Research 14, 5735-5742.

Zhang, Y., Wang, H., Zhang, J., Lv, J., and Huang, Y. (2013a). Positive feedback loop and

239

synergistic effects between hypoxia-inducible factor-2alpha and stearoyl-CoA desaturase-1 promote tumorigenesis in clear cell renal cell carcinoma. Cancer science 104, 416-422.

Zhang, Y.P., Wang, W.L., Liu, J., Li, W.B., Bai, L.L., Yuan, Y.D., and Song, S.X. (2013b). Plasminogen activator inhibitor-1 promotes the proliferation and inhibits the apoptosis of pulmonary fibroblasts by Ca(2+) signaling. Thrombosis research 131, 64-71.

Zhang, Z., Sui, P., Dong, A., Hassell, J., Cserjesi, P., Chen, Y.T., Behringer, R.R., and Sun, X. (2010). Preaxial polydactyly: interactions among ETV, TWIST1 and HAND2 control anterior- posterior patterning of the limb. Development (Cambridge, England) 137, 3417-3426.

Zhang, Z., Verheyden, J.M., Hassell, J.A., and Sun, X. (2009). FGF-regulated Etv genes are essential for repressing Shh expression in mouse limb buds. Developmental cell 16, 607-613.

Zheng, N., Wang, P., Jeffrey, P.D., and Pavletich, N.P. (2000). Structure of a c-Cbl-UbcH7 complex: RING domain function in ubiquitin-protein ligases. Cell 102, 533-539.

Zhu, E.D., Li, N., Li, B.S., Li, W., Zhang, W.J., Mao, X.H., Guo, G., Zou, Q.M., and Xiao, B. (2014). miR-30b, down-regulated in gastric cancer, promotes apoptosis and suppresses tumor growth by targeting plasminogen activator inhibitor-1. PloS one 9, e106049.

Znosko, W.A., Yu, S., Thomas, K., Molina, G.A., Li, C., Tsang, W., Dawid, I.B., Moon, A.M., and Tsang, M. (2010). Overlapping functions of Pea3 ETS transcription factors in FGF signaling during zebrafish development. Developmental biology 342, 11-25.

Zu, X., Zhong, J., Luo, D., Tan, J., Zhang, Q., Wu, Y., Liu, J., Cao, R., Wen, G., and Cao, D. (2013). Chemical genetics of acetyl-CoA carboxylases. Molecules (Basel, Switzerland) 18, 1704-1719.

240

Biography

Melissa Marie Keenan was born Melissa Marie Kirkpatrick on April 26, 1986 to parents Janet Dawn Keremitsis and Bruce Charles Kirkpatrick in Walnut Creek,

California. She grew up in Davis, California. Melissa attended the University of

California at San Diego for her undergraduate education. Upon graduating with a

Bachelor’s of Science in Molecular Biology, Cum laude, in June 2008, she worked for

Celgene Corporation until enrolling in the Cell and Molecular Biology PhD Program at

Duke University in 2009. She joined the University Program in Genetics and Genomics for her degree in 2010.

During her time at Duke, Melissa authored and co-authored a number of research manuscripts and a book chapter. Abbreviated citations of these publications:

Melissa M. Keenan, et al., “ACLY and ACC1 Regulate Hypoxia-Induced Apoptosis by Modulating ETV4 via α-ketoglutarate”. PLoS Genetics. 2015 Oct 9;11(10):e1005599. X Tang, Melissa M. Keenan, et al. “Comprehensive profiling of amino acid response uncovers unique methionine-deprived response dependent on intact creatine biosynthesis”. PLoS Genetics. 2015 Apr 7;11(4):e1005158. Melissa M. Keenan and JT Chi. “Alternative Fuels for Cancer Cells”. Cancer Journal. 2015 Mar-Apr;21(2):49-55. Melissa M. Keenan, CK Ding and JT Chi. “An unexpected alliance between stress responses to drive oncogenesis”. Breast Cancer Research. 2014 Nov 6;16:471. Melissa M. Keenan, CC Lin and JT Chi. “A Genomic Analysis of Cellular Responses and Adaptations to Extracellular Acidosis”. Molecular Genetics of Dysregulated pH Homeostasis. Springer, 2014. CC Lin, Melissa M. Keenan and JT Chi. “Introduction: Molecular Genetics of Acid Sensing and Response”. Molecular Genetics of Dysregulated pH Homeostasis. Springer, 2014.

241

G LaMonte, X Tang, JLY Chen, J Wu, CK Ding, Melissa M. Keenan, et al., “Acidosis induces reprogramming of cellular metabolism to mitigate oxidative stress”. Cancer Metabolism. 2013 Dec 23;1(1):23. JC Lloyd, EM Masko, C Wu, Melissa M. Keenan, et al., “Fish Oil Slows Prostate Cancer Xenograft Growth Relative to Other Dietary Fats and is Associated with Decreased Mitochondrial and Insulin Pathway Gene Expression”. Prostate Cancer and Prostatic Diseases. 2013 Dec;16(4):285-91.

While in graduate school she scored in the top 2% of all proposals to receive a

Ruth L. Kirschstein National Research Service Award from the National Cancer

Institute. In 2014, Melissa received a Scholar-in-Training Award from Aflac, Inc. and the

American Association for Cancer Research to attend and present her work at the AACR

Special Topics Conference, “Cellular Heterogeneity in the Tumor Microenvironment”.

She completed her PhD studies in 2015 and continued on to a Postdoctoral

Research Fellow Position in San Diego. Melissa hopes to continue to contribute to the success of the biomedical research field throughout her career.

242