The Critical Role of Cysteine Import and Metabolism in Pancreatic Cancer

The critical role of cysteine import and metabolism in pancreatic cancer

Michael Alexander Badgley

Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy under the Executive Committee of the Graduate School of Arts and Sciences

COLUMBIA UNIVERSITY

2018

Michael Alexander Badgley

ABSTRACT

The critical role of cysteine import and metabolism in pancreatic cancer

Michael Alexander Badgley

As of this writing, pancreatic cancer is the third leading cause of cancer-related death in the United States: approximately 53,670 new cases will be diagnosed in 2017. Of those patients living with this disease, 43,090 are estimated to die in 2017, accounting for an astounding 7.2 % of all cancer deaths. In fact, pancreatic cancer is estimated to rise to the second leading cause of cancer-related death in the United States by 2020, since it is one of the few cancers for which incidence is currently increasing over time. While precious gains have been made over the past

40 years to improve the survival rates of patients with this disease, still only 8.2% of patients will survive 5 years beyond their initial diagnosis, highlighting the need for researchers and clinicians to develop novel therapeutic strategies for what has heretofore been a highly intractable malignancy.

Recent efforts in the field of pancreatic cancer research have focused on identifying new biological dependencies in this cancer that may be leveraged to improve treatments. Many of these dependencies revolve around the unique metabolic demands of the cancer cell, highlighted in work first popularized in 1956 by Otto Warburg. New research has demonstrated that one of these unique metabolic demands may in fact be the tight regulation of redox homeostasis, spurred by the need of metabolically active pancreatic cancer cells to detoxify naturally occurring reactive oxygen species (ROS) in order to maintain a healthful state.

Experiments have shown that activating mutations in KRAS, which are present in 95% of pancreatic cancers, actually created a highly reducing intracellular environment to support the increased proliferative capacity of these cells.

Many cellular processes have evolved to cope with the stress of ROS production. Most

of these functions revolve around the unique chemistry of thiol functional groups, which are highly reactive and exquisitely sensitive to oxidation. Only two of the twenty canonical amino acids can provide the sulfur necessary for this chemistry: cysteine and methionine. Cysteine is the more reactive of these two amino acids and as such provides the basis for much of the redox chemistry of the cell. This is accomplished through its role as the rate-limiting precursor for the synthesis of glutathione which is the preeminent antioxidant molecule of the cell. To explore the possibility that the perturbation of redox homeostasis may present a unique therapeutic avenue in the context of pancreatic cancer cells, we have endeavored to illuminate the role of cysteine and its metabolic derivatives in maintaining the redox state of pancreatic cancer cells.

Here, we show that cysteine is critical for the growth of multiple pancreatic cancer cell lines, preventing the rapid and catastrophic accumulation of ROS, particularly lipid ROS. In the absence of exogenously provided cysteine, the quick accumulation of lipid ROS causes a cell death phenotype previously described as “ferroptosis,” an iron-dependent, oxidative process.

We also confirm that these effects can be mimicked pharmacologically by treating these cells

- with agents that block system xc , the only known transport mechanism of oxidized cysteine

(cystine). In addition, we identify cells that exhibit a priori resistance to these effects, but which can be sensitized to ferroptotic cell death by blocking macropinocytosis, an endocytic process that facilitates the bulk uptake of large extracellular substrates into the cell. Finally, we show that

GSH depletion is required, but not sufficient to mediate these effects, suggesting that additional cysteine-derived metabolites provide secondary functions critical to protecting against ferroptosis.

To explore whether these effects also hold true in vivo, we first assessed the mRNA

- expression levels of SLC7A11, a gene that encodes a protein subunit specific to system xc . We demonstrate that SLC7A11 is specifically upregulated in malignant pancreatic cancer tissue

when compared to a normal counterpart. Furthermore, this gene is upregulated as the cancer progresses from pre-neoplastic lesions to frank carcinoma. Also, we highlight that this gene is upregulated in other cancers as well, suggesting broadly applicable therapeutic potential.

Finally, to test if this system can be targeted in vivo, we designed a genetically engineered mouse model of pancreatic cancer in which we can temporally control the deletion of SLC7A11.

Using this tool, we demonstrate that deletion of SLC7A11 in the context of established pancreatic tumors can retard and even regress tumor growth, resulting in a notable increase in overall median survival. Taken together, our results indicate that cysteine import and metabolism are a critical dependency in pancreatic cancer, providing evidence that targeting these mechanisms may be useful in a clinical setting.

TABLE OF CONTENTS

LIST OF FIGURES...... iv

LIST OF TABLES...... vii

LIST OF ABBREVIATIONS AND ACRONYMS...... viii

ACKNOWLEDGEMENTS...... ix

DEDICATION...... xiv

Chapter 1 – Introduction...... 1

Pancreatic Cancer: History, epidemiology, biology, and models...... 2

Cysteine: Biological function and import...... 14

Ferroptosis: Mechanism and Clinical Relevance...... 18

Chapter 2 – Cysteine is critical for pancreatic cancer cell growth...... 21

Background and Significance...... 22

Results...... 23

PDA cell lines are sensitive to cysteine withdrawal...... 23

- Pharmacological inhibition of system xc phenocopies the effects of cysteine withdrawal. . . . . 25

Cysteine deprivation causes ferroptosis in pancreatic cancer cell lines...... 27

Discussion...... 31

Chapter 3 - Macropinocytosis, but not transsulfuration, mediates insensitivity to the

- effects of system xc inhibition in pancreatic cancer cells...... 34

Background and Significance...... 35

Results...... 35

Cysteine biosynthesis through transsulfuration provides a minor component of total cysteine levels in pancreatic cancer cells...... 35

Macropinocytosis is upregulated in MIA PaCA -2 cells, but not PANC-1 cells, in response to

- system xc inhibition...... 39

- Macropinocytosis mediates sensitivity to system xc inhibition in MIA PaCa-2 cells...... 41

Discussion...... 43

Chapter 4 – Cysteine, GSH, and their roles in mediating ferroptosis in pancreatic cancer cells...... 46

Background and significance...... 47

Results...... 48

GSH depletion is necessary, but not sufficient to cause ferroptotic cell death in pancreatic cancer cells...... 48

Exogenous cysteine is incorporated into acetyl CoA pools and CoA byproducts can rescue cell death caused by cysteine deprivation...... 50

Discussion...... 55

Chapter 5 – SLC7A11 expression in human tumors and the effect of its deletion on established tumors in a genetically engineered mouse model of PDA...... 60

Background and Significance...... 61

Results...... 62

SLC7A11 is overexpressed in human tumor tissue and correlates with poorer outcomes. . . . .62

SLC7A11 expression correlates with KRAS mutation...... 66

Deletion of SLC7A11 in established murine tumors slows growth and extends overall survival 67

Discussion...... 74

Chapter 6 – Conclusions and future directions...... 77

Cysteine import as a unique metabolic dependency in pancreatic cancer...... 78

Determining the role in cysteine import in pancreatic cancer development...... 79

- Overcoming resistance to system xc deletion...... 81

Targeting GSH synthesis and CoA function to cause ferroptosis...... 83

In vivo mechanisms of ferroptosis and its relevance in cancer...... 84

MATERIALS AND METHODS...... 87

Chapter 2...... 87

Chapter 3...... 89

Chapter 4...... 93

Chapter 5...... 93

REFERENCES...... 100

iii

LIST OF FIGURES

Chapter 1

Figure 1-1 The anatomy of the pancreas...... 3

Figure 1-2 PanIN progression model of pancreatic cancer...... 8

Figure 1-3 Metabolic changes associated with activating RAS mutations...... 11

Figure 1-4 The structure of cysteine and cystine...... 15

Figure 1-5 Structure of SLC7A11...... 17

Figure 1-6 Model of ferroptosis in KRAS mutant cells...... 19

Chapter 2

Figure 2-1 Pancreatic cancer cells are sensitive to cysteine withdrawal...... 24

- Figure 2-2 System xc inhibition phenocopies the effects of cysteine withdrawal...... 26

Figure 2-3 Cysteine deprivation causes ferroptosis in pancreatic cancer cells...... 28

Figure 2-4 Cysteine deprivation causes a catastrophic accumulation of lipid ROS characteristic of ferroptosis...... 30

Chapter 3

Figure 3-1 The transsulfuration pathway...... 37

- Figure 3-2 Neither system xc inhibition nor GSH synthesis impairment cause increase flux

iv through the transsulfuration pathway...... 38

Figure 3-3 Transsulfuration does not mediate sensitivity of pancreatic cancer cells to cysteine deprivation...... 40

Figure 3-4 MIA PaCa-2 cells, but not PANC-1 cells, upregulate macropinosome formation in

- response to system xc inhibition...... 42

Figure 3-5 Inhibiting macropinocytosis sensitizes previous unaffected pancreatic cancer cells to

- system xc inhibition...... 43

Chapter 4

Figure 4-1 GSH depletion is required for ferroptotic cell death...... 49

Figure 4-2 GSH depletion is not sufficient to case ferroptotic cell death...... 51

Figure 4-3 Cysteine and its biomolecular products...... 52

Figure 4-4 Cysteine does not enter the cysteine sulfinate production branch of cysteine metabolism in pancreatic cancer cells...... 53

Figure 4-5 Cysteine as incorporated into CoA pools and CoA products can rescue ferroptosis 54

Figure 4-6Select differentially modified metabolite abundances in PANC-1 cells treated with IKE or BSO...... 56

Figure 4-7 Select differentially modified metabolite abundances in MIA PaCa-2 cells treated with

IKE or BSO...... 57

Chapter 5

Figure 5-1 SLC7A11 transcript is overexpressed in human pancreatic cancer...... 64

Figure 5-2 SLC7A11 is overexpressed in many human cancers and can correlate with poor survival...... 65

Figure 5-3 SLC7A11 overexpression correlates with KRAS mutations...... 66

Figure 5-4 PdxFlpO allele validation and KPSFR breeding scheme...... 68

Figure 5-5 Histological comparison of PDA and PDA development in KPC and KPFSR mice. .69

Figure 5-6 Assessment of Slc7a11 recombination in tamoxifen treated mice...... 70

Figure 5-7 Slc7a11 deletion in established tumors slows growth and extends survival...... 72

Figure 5-8 PCR analysis of recombination in endpoint tumors...... 73

Figure 5-9 Histological analysis of proliferation markers and morphological effects in tamoxifen- treated tumors...... 73

Chapter 6

Figure 6-1 Effects of SLC7A11 deletion on acinar-ductal metaplasia (ADM) in KPC and KPFSR mice...... 81

Figure 6-2 High magnification images of the effects of Slc7a11 deletion on pancreatic tumors.86

LIST OF TABLES

Chapter 1

Table1-1 List of successful stage III clinical trials in pancreatic cancer...... 5

Table 1-2 Genetically engineered mouse models of pancreatic cancer...... 12

vii

LIST OF ABBREVIATIONS AND ACRONYMS a-KB alpha keto-butyrate BA1 bafilomycin A1 B-ME beta mercaptoethanol BSO buthionine sulfoxamine CBS cystathionine beta-synthase CGL cystathioine gamma-lyase CoA Coenzyme A DEG differential expressed gene DFO deferoxamine EIPA 5-(N-Ethyl-N-isopropyl)amiloride Fer-1 ferrostatin-1 GCL glutamate-cysteine ligase GPX4 glutathione peroxidase 4 GSH reduced glutathione GSH-EE glutathione ethyl ester GSS glutathione synthetase GSSG oxidized glutathione IKE imidazole ketone erastin NAC N-acetyl cysteine Nec1s Necrostatin-1s PanIN Pancreatic intraepithelial neoplasm PDA pancreatic ductal adenocarcinoma PE piperazine erastin PPG propargyglycine ROS reactive oxygen species RSL3 Ras-selective-lethal drug 3 SAH S-adenosyl homocysteine SAM S-adenosyl methionine Tro Trolox ZVAD Z-VAD-FMK

viii

ACKNOWLEDGEMENTS

The decision to complete a doctoral thesis is not one to be undertaken lightly, though wisdom such as this is sometimes wasted on the well-practiced. As such, I have only learned this fact after the process has been all but completed; after seven long years of training, epiphany, verification, and growth. The naiveté of the inaugural graduate student is counterbalanced by their seemingly indefatigable enthusiasm, but this too I have found eventually gives way to the hard reality that passion is little without loving support, readily and willingly given by those more qualified and those more patient than oneself. For these reasons, it is customary in a thesis dissertation to write acknowledgements, literally meaning as a verb

“to accept or admit the existence or truth of,” which is somewhat fitting given that science can be considered the ongoing attempt to accept the truth of things. In my case, however, I believe these not to be so much customary as they are of utmost necessity. To admit the truth, as it were, the completion of this work and the writing of this text would not have been possible without the thoughtful advice of caring mentors, the ongoing support of my fellow lab members, and the love and care of my family and friends. In the interest of offering these parties their due thanks, I gratefully take this opportunity to electronically scribble these few words of appreciation, paltry though they may be in comparison to the considerable support they are meant to redress.

To Ken—Before I begin, it should be noted that I have the distinct privilege of being Ken’s first graduate student (both taken into the lab and to successfully defend their work, thankfully). The resultant relationship is unique I believe, and I would not have traded any of the early hiccups in my graduate schooling (mostly self-inflicted, I suspect) for the deep level of trust and mutual respect I would like to think has developed between us. More to the point, however, I must thank Ken for being an irreplaceable mentor and friend. I first saw Ken when he gave a faculty lecture to first-year graduate students in the fall of 2011 and I remember thinking two things:

First, cool earring; and second, I need to rotate in this person’s lab. My rotation involved molecular cloning and a small bench fire. Despite the literally inflammatory start, Ken graciously accepted me into the lab, and I was eager to join. In the seven years we have known each other, I have come to learn that his passion for his work is as irrepressible as it is infectious; his knowledge of his field, encyclopedic. He even took the time to personally instruct me in the arcane skills of histopathological analysis (also known as the art of taking pretty pictures of terrifying things). Whether discussing scientific musings over a smoky glass of scotch, or tickling the ivories in the hopes of tickling our more creative neurons, Ken has proven to be an astute, thoughtful, and caring mentor. Not only this, but Ken also bravely entertains his student’s more hair-brained schemes (agreeing to pay for the breeding a six-allele mouse is never easy), fostering a fertile environment to test one’s own ideas and grow as a nascent scientist. Any modicum of success I have had thus far, I owe to him. I thank him for teaching a foolish student how to be a less (did I use that word right, Ken?) foolish scientist.

To my committee: Lori Sussel, Serge Cremers, Brent Stockwell, and Cory Abate-Shen—

Guideposts are indispensable for any scientist traveling through turbulent PhD waters, and I have been blessed with some of the best any student could have hoped to find. I thank Lori for being a thoughtful and kind member of the committee at the start of my PhD, and I thank Serge for stepping in so ably to provide a much needed boost to the metabolism bona fides of this project. To Brent, I offer thanks for continued support, from allowing me to attend Stockwell Lab meetings to get the proper background needed to understand the field of ferroptosis, to graciously being a co-mentor on grant applications. Brent has been a valuable collaborator on multiple projects and the short commentary here does little to circumscribe the full extent of his influence on this work and my PhD. Finally, I must give special thanks to Cory, the head of the committee. It is no small responsibility to serve as committee head, not only because the student may look to you for the most relevant and incisive advice, but because of the ongoing

x responsibility to the student and their continued career success. I often looked to her for the most blunt and necessary commentary, which inevitably improved and focused my research. I do believe that choosing her as the head of this committee may have been the best experiment

I ever undertook in my pursuit of this degree. I thank her for teaching me, in the end, that the shortest distance between two points, both in life and in science, is always a straight line.

To my collaborators—Since no one person can become an expert in all things, the multidisciplinary nature of this work was facilitated by fruitful and stimulating collaborations, both at Columbia and farther afield. To Costas and the team at the University of Michigan, I offer thanks for their time, their attentiveness, and their keen analyses of our metabolomics data. To the core facilities at Columbia University, especially the histology and microscopy cores, I owe a great deal of thanks for processing samples and helping with various analyses. Finally, I thank

Scott Seeley for providing us with vital mice to make our so-called KPFSR mice, which played such a critical role in this study.

To the Olive Lab, version 1.0—As a student, I found that while you must of course be inspired by the research you undertake, it is arguably more important for you to be inspired by the people who surround you daily. A caring, fun-loving, and friendly lab makes the most fertile ground for productive research, and I am privileged to say that over the course of seven years I have come to appreciate my fellow labmates as family. To my fellow graduate students,

Roshan and Jaime: I thank you so much for the commiseration, laughter, advice, late-night company, drunken revelry, and matching plaid outfits. I thank the CS club (Dafydd, Steve, and

Carmine) for being great friends that offered even greater advice when I needed it most. I will never forget our singing duets, late-night Dota rounds, the magic slippers, or the multiple fantasy football championships. To my Poke Bros (Chris and Wilson), I never did catch that

Squirtle. May justice always rain from above, and Chris, I am sorry for never having played

Ingress. To Barbara and Sam, any budding scientist would be so lucky to have such loving,

xi honest, and supportive friends. Thank you both for continuing to keep up with my progress and for making work seem more like fun. Ciao Ciao, Baby Cheese. To Carlo and Irina (our newest postdoctoral arrivals), I owe special thanks since not only have their sacrificed their time to me in lab but also out of it as I finally moved apartments towards the end of my PhD. Carlo I thank for his consistently engaging intellectual manner and numerous soda runs to the front, while

Irina I thank for making an exemplary baymate during this last year, especially as it pertains to head-ware. To my two undergraduate students, Peter and Christina (co-founder of Acrymalide

[sic] Industries), I must offer unique thanks since their youthful excitement, humor, hardwork, selflessness, snapchats, and consistent efforts have improved both this work and myself.

Though, I never did get over Peter throwing out all my cell culture plates. To Marina, I owe thanks for always making sure reagents were attained in timely fashion. To our Dutch contingent (Jennifer and Tessa), I share gratitude for the hard work both on collaborative projects and for general intellectual input. To our clinically minded fellows (Paul, Gulam, and

Dominic), sure…I can dose for you this weekend. And finally to our adopted labmates of the

Punterson lab: Long live the Puntoliversons. If I have left anyone out, please know that I thank the entire lab both past and present for making such good friends and making the laboratory feel much more like home.

To my family and friends: Mom, Dad, Jess, Michelle, Pat, Kevin, Conor, Marlena, Matt and

Brendan — None of this would have been possible without the loving and lifelong support of family and friends. I thank my wonderful parents for always letting a curious boy get into just a little bit of trouble and for always supporting my decision to pursue this PhD. To my sisters, thank you for making the long trek to New York City to visit your occasionally isolated brother.

As a transplant from Los Angeles and as someone who has sometimes found it difficult to feel at home in strange places, I owe a great deal of thanks to those people who were patient and thoughtful enough to take me into their home and always offer help when it was needed. For

xii this I thank my family-in-law, who have helped make New York a second hometown. Finally, to

Matt and Brendan: thank you for your lifelong friendship and every game of Smash Bros.

To my wife, Caroline—I first met Caroline when I was an aimless undergraduate at Harvard in

2007. While singing deliriously along to the tune of Mulan’s “I’ll Make a Man out of You,” I noticed a shy red-headed girl hunkered closely to her friends on the other side of the room. I had little knowledge at the time just how much this person would shape my life, but in the end it is likely to her I owe the most thanks of all those mentioned here. She is the person who actually convinced me to apply to graduate school and is the impetus for this entire endeavor. She has been my best friend, fiercest critic, most engaging interlocutor, and most reliable gaming partner. She even helped me design my only award-winning PowerPoint slide. She is an inspiring scientist in her own right, and has offered numerous comments on this dissertation, my defense presentation, and my submitted manuscript. I thank you for continually improving me and my research and am indebted to you always for agreeing to go through this life together with me.

Honorable mentions—I would also like to thank the Coca Cola Company for producing the most sublime beverage conceived by human ingenuity…Vanilla Coke.

I end with a quote from Kevin Garnett, a member of the championship Boston Celtics from

2007-2013, here discussing what it takes to achieve team chemistry:

“Timing is everything. Chemistry is something that you don't just throw in the frying pan and mix it up with another something, then throw it on top of something, then fry it up and put it in a tortilla and put in a microwave, heat it up and give it to you and expect it to taste good. You know? For those of you

who can cook, y'all know what I'm talking about. If y'all can't cook, this doesn't concern you."

-Kevin Garnett

I certainly hope that at this PhD’s end, I have at least learned to cook well with others.

xiii

DEDICATION

For Caroline, Frangipane, and all the mice that made this work possible.

xiv

Chapter 1 – Introduction

Pancreatic Cancer: History, epidemiology, biology, and models

History

In antiquity, anatomy was already a highly refined science, albeit a grotesque one. Some ancient anatomists have even been accused by historians of performing live vivisections on local criminals1. One such anatomist, the so-called “Father of Scientific Anatomy,” Herophilus of

Chalcedon, is generally credited with the first description of the pancreas in 300 BC2,3. Galen, the prominent Greek physician and surgeon, would later give an ambiguous assessment of the organ, and there is no evidence that the ancient Greeks had even a rudimentary understanding of the organ’s function4. Even with its formal anatomical description (Figure 1-1), the pancreas remained an enigmatic organ for some time: it was not until many centuries later that it would even be appreciated both for its critical roles in normal physiology and as a seat of diseases5.

While the pancreas is perhaps most famous for its endocrine function and the role it plays in the onset of type I diabetes, as early as the 16th century, anatomical scientists did note that cancerous malignancies were associated with the organ. The first known description of one such tumor was written in 1761 by the Italian anatomist Giovanni Battista Morgagni in his seminal work, De Sedibus Et Causis Morborum Per Anatomen Indagatis Libri Quinque (On the

Seat and Causes of Disease, Investigated Through Anatomy, in Five Books). In this tome,

Morgagni set forth the first description of a mass in the pancreas available in Western medicine.

It would take nearly 100 years before this finding would be confirmed in a series of cases described by Jacob Mendez Da Costa, an American clinician in Philadelphia. In a series of thirty-seven cases, Da Costa would describe in some detail this cancer and its associated effects, including its tendency to occlude the bile duct, its dense and fibrotic nature, and its associated co-morbidities. All these findings were set forth in his published work, On the Morbid

Anatomy and Symptoms of Cancer of the Pancreas, which would begin to dictate the

Figure 1-1: The anatomy of the pancreas

The pancreas is a retroperitoneal organ which serves both an exocrine and endocrine function. More than 90% of the pancreas is composed of acinar tissue tasked with the production of digestive enzymes to facilitate nutrient absorption in the small intestine. The endocrine function of the pancreas is carried out by cells in the islets of Langerhans, which produce a bevy of hormones to control blood glucose levels.

approaches physicians and surgeons of the day would take to treat this still relatively mysterious illness.

Early treatments for the disease revolved around surgical approaches. In the 19th century, an Italian surgeon named Alessandro Codivilla performed the first attempted pancreaticoduodenectomy (resection of the head of the pancreas and the neighboring duodenum) to remove a mass on the head of the pancreas, only to lose his patient 18 days

3 following the operation6. The first successful pancreaticoduodenectomy would be performed by

Georg Hirschel on a patient presenting with jaundice and suffering from an ampullary carcinoma. Finally in 1935, here at Columbia, the first Whipple procedure would be completed

(so-named for the American surgeon, Allen Oldfather Whipple, who popularized the technique and was a professor of surgery at Columbia University from 1921 to 1946). The Whipple

Procedure involved the removal of the C-loop of the duodenum and the pylorus junction of the stomach; resection of the head of the pancreas; and the ligation of the remaining pancreas, common bile duct, and stomach to facilitate functional digestion. The initial attempt at this surgery was described in three patients. While the first patient did not survive the ordeal, the second patient would live for nine months, eventually dying from an inflamed gall bladder. The third patient would live for two years, eventually succumbing to liver metastases arising from the original ampullary cancer7. The Whipple Procedure would go on to become a standard operation for patients with pancreatic cancer in the head of the pancreas, being thoroughly updated throughout the years to treat tail of the pancreas tumors and to improve outcomes for patients throughout the world.

Few patients, however, actually qualify for surgical interventions in pancreatic cancer

(estimates range from 10-20% of diagnosed patients), prompting later treatments to leverage the discovery and use of poisonous chemotherapeutics. While initially radiotherapy was commonly used in the treatment of locally invasive, unresectable pancreatic cancer, pioneering research in 1969 showed that combinations of radiation and 5-fluorouracil (5-FU), a nucleoside analog, could extend survival in pancreatic cancer patients8. This work was confirmed in another landmark study, showing that patients treated with radiation and 5-FU had a prolonged median survival of 10 months compared to 5.5 months with radiation alone9. New treatments of pancreatic cancer with chemotherapy would stagnate for 16 years until a trial showed improvements to quality of life and a modest improvement in overall survival when patients were

treated with gemcitabine, a cytidine analog10. Gemcitabine would remain the standard of care

treatment for advanced pancreatic cancer for nearly 15 years until the FDA approval of the new,

aggressive chemotherapeutic cocktail, FOLFIRINOX (a composite regimen containing folinic

acid, 5-FU, irinotecan, and oxaliplatin) which had been shown to increase overall survival in

metastatic pancreatic cancer from 6.8 to 11.1 months when compared to gemcitabine alone11.

Two years later, another treatment regimen featuring gemcitabine and nab-paclitaxel, an

albumin-bound form of taxol, would prove efficacious in metastatic pancreatic cancer patients,

shifting median overall survival from 6.7 to 8.5 months when compared to gemcitabine alone12.

Research continues to move apace, as researchers race to find new chemotherapeutic

regimens to extend survival in these patients (Table 1-1). However, few patients still qualify for

potentially curative surgery and 5-year survival rates continue to be stubbornly low, even as new

treatments are approved, highlighting the need for new and paradigm-shifting research to

challenge the devastation of this malignancy.

median overall Regimen number of patients Response rate % 1-year survival p-value Author (year) survival (months) 5-FU 63 0.00% 4.41 2% 0.0025 Burris HA III et al. (1997) Gemcitabine 63 5.40% 5.65 18% Gemcitabine 284 6.90% 5.9 17% 0.038 Moore MJ et al. (2005) Gemcitabine/erlotinib 285 8.20% 3.2 23% Gemcitabine 171 6.80% 6.8 21% <.001 Conroy T et al. (2011) FOLFIRINOX 171 11.10% 11.1 48% Gemcitabine 430 7.00% 6.7 22% <.001 Hoff DD et al. (2013) Gemcitabine/nab-paclitaxel 431 23.00% 8.5 35% 5-FU and folinic acid 149 1.00% 4.2 not reported Onivyde (irinotecan) 151 6.00% 4.9 not reported 0.012 Wang-Gillam et al. (2015) 5-FU and folinic acid/Onivyde 117 19% 6.1 not reported

Table 1-1: List of successful stage III clinical trials in pancreatic cancer

In the past two decades, only five major clinical trials in metastatic pancreatic cancer have resulted in new, FDA-approved treatment regimens in pancreatic cancer. They are listed above with relevant clinical details. Even with these advances, 5-year survival rates remain low.

Epidemiology

While the rates of death caused by cancer as a broad entity continue to decrease slowly over time, pancreatic cancer has the ominous distinction of being one of the few cancers that controvert that trend. While the current number of new cases of this cancer is 12.5 per 100,000, ranking it only the twelfth most common cancer in the United States, it is estimated that over

43,000 people will die of this disease in 2017 in the US, making it the third leading cause of cancer related death in this country. In fact, whereas the number of deaths per 100,000 persons per year in the cases of lung, breast, and colon cancer (the three other leading causes of cancer-related death in the US) has continued to drop since 1992, pancreas cancer mortality has risen over the same time frame. In 2017 it is estimated that while only 3.2% of all new cancer cases will be pancreatic cancer, it will account for 7.2% of all cancer deaths. By 2020, pancreatic cancer is predicted to become the second leading cause of cancer-related deaths in the US, trailing only lung cancers13,14.

These numbers alone paint a harrowing picture of the current state of the disease, but when analyzed alongside survival times post-diagnosis, the portrait becomes even more tragic.

Currently, only 8.2% of patients diagnosed with pancreatic cancer will survive beyond 5 years, and the median survival is estimated at 3-6 months15. Of these patients, 10-20% are eligible for surgery, the only ostensibly curative therapy available. While resected patients see an improved

20-25% five-year survival rate and a median survival time between 12-20 months, research has now shown that the majority of these patients will recur within a year16-19.

Many of these disappointing outcomes can be traced to the staging of pancreatic cancer at diagnosis. Only 10% of tumors are confined to the primary site, while nearly 30% have already spread to local lymph nodes. Over half of the time, the cancer has already metastasized to distant sites. As one might expect, 5- year relative survival correlates well with stage at diagnosis: 31.5% for patients with localized disease; 11.5% for patients with disease having

6 spread to regional lymph nodes; and a disastrous 2.7% for patients with distant metastases.

There is scant research to pinpoint a precise cause of pancreatic cancer, but we can say that age is the most obvious risk factor. The disease primarily afflicts persons between ages 65-

74, with this age group accounting for the plurality of new cases (27.8%). Most cases occur in patients over the age of 65, with 70 being the median age of diagnosis. That is not to say that the cancer cannot effect younger populations, though: 10% of cases are diagnosed in patients under the age of 54. Other more operable risk factors have been identified, including but not limited to cigarette smoking, heavy metal exposure, alcohol use, obesity, chronic pancreatitis, and family history20-23. Aspirin is known to decrease risk. These data paint a complicated epidemiological picture of a devastating disease, highlighting the need for improved screening techniques and treatment options for this cancer.

Biology

While pancreatic cancer continues to be a distressing diagnosis across the globe, illuminating research has eliminated some of the mystique this disease once held. While the molecular and biologically diverse functions of the pancreas can actually give rise to multiple pancreatic cancer types, we will restrict our discussion to by far the most common form of the disease: pancreatic ductal adenocarcinoma (PDA). PDA, so named due to its sharing histological features with ductal structures of the pancreas, accounts for more than 85% of pancreatic cancers24,25. PDA is characterized by glandular structures bearing strong resemblance to the epithelium of the ducts of the pancreas and a dense and desmoplastic stroma surrounding these structures (Figure 1-2)26. Tumors exhibit a wide range of differentiation states, from very glandular to highly mesenchymal. Often, these tumors invade local nerve bundles and lymph nests, spreading through the hematological system to invade distant sites like the liver and lung27,28.

Figure 1-2: PanIN progression model of pancreatic cancer

Pictured above is the current working model for the development of pancreatic cancer from PanIN precursor lesions. Mutations in the oncogene KRAS, highlighted in red, occur at the earliest stage of preneoplastic development and are thought to be critical for the development and maintenance of this disease. Histological images below each stage indicator typify the morphological features of these lesions and of PDA.

Adapted from Adapted from Hruban et al. 2003, Hingorani et al. 2003, Vincent et al. 2011, and Olive et al. 200629-32

Pancreatic cancer is now known to arise from multiple forms of precursor lesions, the most common of these being pancreatic intraepithelial neoplasms, or PanINs33. While these lesions are common in older adults, only a subset of these preneoplasms will transform into fully invasive carcinoma. PanIN lesions progress through three stages: PanIN-1 lesions are characterized by columnar, mucinous accumulations in ductal structures containing relatively normal nuclei; PanIN-2 lesions exhibit epithelial folding structures and nuclear atypia; PanIN-3 lesions contain increasingly irregular nuclei and display epithelial budding into the luminal space. PanIN-3 neoplasms are considered carcinoma in situ, or cancer which has not yet managed to breach the basement membrane of the epithelium34.

From a molecular genetic standpoint, pancreatic cancer has been very well studied.

Early events include activating mutations in KRAS and functional loss of the CDKN2A locus.

Subsequently, some precancerous cells will acquire loss-of-function mutations in p53, SMAD4, and BRCA2 (all tumor suppressor genes). Most commonly, however, the cancer is associated with its earliest genetic lesion, activating mutations in KRAS, which are present in 95-100% of all PDA cases35,36. The most common KRAS mutations in PDA occur at codon 12, however other prevalent mutations include those at codons 13 and 6037. These mutations limit the ability of KRAS to carry out its GTPase function or to bind GTPase activating proteins, leaving the protein bound to GTP and thereby in its active signaling state38. The resulting signaling cascade is thought to precipitate many of the oncogenic events that give rise to pancreatic cancer. Given the early timing of this mutation in PDA, it has been the subject of intensive study over many decades. Results have shown that activating KRAS mutations play a critical role in both the initiation and maintenance of PDA39,40, making it an attractive pharmacological target in this disease. To date, however, attempts to drug mutant KRAS directly have proven difficult due it its nature as a small GTPase and its lack of structurally appropriate binding pockets41. Recent efforts using in silico approaches have proven, however, that mutant RAS is targetable42.

Since drugging KRAS directly has not borne clinical fruit as of yet, many researchers have focused studies on understanding downstream biological implications of KRAS mutation in order to leverage possible dependencies precipitated by these mutations. One such burgeoning field of study has revolved around understanding the critical metabolic shifts initiated by KRAS mutation in cancer. Consequently, altered metabolic function is newly emerging as a well- appreciated hallmark of cancer43, although it may be better to say that this concept is actually re-emerging. As many as 60 years ago, Otto Warburg famously contended that cellular metabolic shifts from primarily mitochondrial oxidative phosphorylation to aerobic glycolysis were actually the cause of cancer. And, while most researchers now contend that these

9 phenotypes, the so-called “Warburg Effect,” are the result of changes in cancer cells rather than their cause, Warburg’s hypothesis has continued to be the basis for a renewed and expansive interest in metabolic biology in the context of cancer.

Given that the vast preponderance of PDAs harbor activating KRAS mutations, many studies have focused on elucidating the myriad metabolic shifts that occur in RAS-mutant cancer cells (Figure 1-3). Current work suggests that many of the metabolic shifts in these cells occur to support a high rate of proliferation by facilitating the production of adequate levels of cellular biomass substrates, including proteins, lipids, and DNA44. Interestingly, PDA is generally considered nutrient-poor, thanks in part to its deposition of a stifling desmoplastic stroma45. This results in PDA utilizing novel methods to accumulate biomolecular building blocks, including by using autophagy and macropinocytosis46-48. To acquire essential amino acids, some research suggests that PDA in patients can cause body-wide metabolic reprogramming, sending branched-chain amino acids into circulation via whole-body muscle protein breakdown and that this process even precedes disease diagnosis by as many as 5 years49. To acquire the molecular precursors necessary for the production of both lipids and DNA, PDA cells grow addicted to glutamine consumption. Glutamine, an incredibly metabolically flexible molecule, provides carbon for the TCA cycle, nitrogen for the production of purine and pyrimidine rings of

DNA bases, and contributes to the synthesis of CoA to support lipid production50-54.

In addition to increasing biomass needs, high rates of cellular proliferation can cause an accumulation of reactive oxygen species (ROS) caused by mitochondrial respiration in both precancerous lesions and PDA55-58. ROS, while natural byproducts of mitochondrial respiration, are highly reactive molecules than can damage proteins, lipids, and DNA. To combat this, mutant KRAS pancreatic cancer cells create a uniquely reducing intracellular environment by engaging a Nrf2-mediated antioxidant program59. The hypothesized reliance of

Figure 1-3 Metabolic changes in cells associated with activating RAS mutations

Multiple metabolic shifts occur in RAS-mutant cancer cells. Classically, RAS mutations cause an increase an aerobic glycolysis, the “Warburg effect,” resulting in the production of pyruvate and lactate from glucose (left). RAS-mutant cells are also addicted to glutamine, using this flexible biomolecule to create substrates for the TCA cycle and also maintain a healthy redox state (center). RAS-mutant cells can upregulate macropinocytosis, an endocytic cellular process, allowing them to take up large extracellular components, which are then transported to the lysosome for the breakdown and release of their constituent amino acids (right, bottom). Mutant RAS pancreatic cancer cells also activate autophagy to catabolize damaged organelles and proteins for reuse in essential biomolecular processes (right, top). Metabolic enzymes regulated by mutant RAS are highlighted in blue, with arrows indicating up or down regulation. Enzymes in yellow are not known to be affected by mutant RAS.

Abbreviations: GLUT1, glucose transporter-1; HK1 and HK2; hexokinase 1 and 2; PFK1, phosphofructokinase 1; ENO1, enolase1; PKM, pyruvate kinase; LDHA, lactate dehydrogenase; GLUD1, glutamate dehydrogenase 1; GLS1, glutaminase 1; GOT1 and 2; aspartate transaminase 1 and 2; MDH1, malate dehydrogenase 1; ME1, malic enzyme 1

Figure adapted from Cox et al., 201460

these cells on a highly reducing environment to promote tumor formation and maintenance in

spite of the increased production of ROS, may represent a critical opportunity for highly specific

and potent targeting of this disease, which we will expound on in later chapters of this work.

Models

While clinical research has continued apace on pancreatic cancer patients since the

1960’s, models that would allow researchers to delicately interrogate specific aspects of PDA

biology in vivo would not arise until the early 2000’s (Table 1-2). Although early models of the

1980’s used harsh carcinogens to facilitate the production of pancreatic tumors in mice61,62, it

was the creation of the Cre-inducible KRASLSL-G12D allele in 2001 that would revolutionize the

Genotype PanINs? Time to cancer onset Cancer type Metastasis? Median Survival Comments Early model of PDA, Pdx1-Cre;KrasG12D Yes Over a year PDA Yes Over a year though only a subset of mice develop cancer

Very similar to Pdx1- p48+/Cre;KrasG12D Yes Over a year PDA Yes Over a year Cre;KrasG12D mouse; p48 expression is less leaky

Pdx1- Short latency model with Yes 1-2 months PDA No 3 months Cre;KrasG12D;p53lox/lox multi-focal tumors

Recapitulates the Pdx1- Yes 2-3 months PDA Yes 5 months phenotypic constellation Cre;KrasG12D;p53R172H/+ of the human disease

Ptf1a+/Cre;KrasG12D;p53R Model of familial Yes 2 months PDA, ACC Yes 2.5 months 270H/+;Brca2Tr/D11 pancreatic cancer

Table 1-2 Genetically engineered mouse models of pancreatic cancer

Listed above are some of the most commonly employed genetically engineered mouse models (GEMMs) in the field of pancreatic cancer.

Abbreviations: PanIN, pancreatic intraepithelial neoplasia; PDA, pancreatic ductal adenocarcinoma; ACC, adenoid cystic carcinoma

creation of genetically engineered mouse models (GEMMs) in this disease63. The unique scientific achievement of this allele was the utilization of the “Lox-stop-lox” cassette (the “LSL” of the abbreviation above), which prevented the expression of the constitutively active mutant

KRAS allele until this DNA was brought into the proximity of Cre-recombinase. Tissue specific drivers of Cre expression were already available by this time, allowing researchers spatial control of the activation of this oncogenic allele. In 2003, work was published characterizing what is now commonly referred to as the KC model, in which either PDX-Cre or p48-Cre (both pancreas-specific transcription factors) mice are bred with KRASLSL-G12D mice to create KC mice30. Remarkably, these mice are born with relatively normal pancreata but later develop PDA deriving from precursor PanIN lesions, albeit at low penetrance.

The KC model, while highly informative and continuingly useful, develops malignant tumors at too low a frequency and too slow a rate for preclinical research utility. To surmount this shortcoming, additional genetic modifications have been layered onto this mouse. The first description of what would become for many the current standard in GEMMs of PDA came only two years after the description of the KC mouse. These mice, KRASLSL-G12D; p53LSL-R172H;

PdxCre, aptly termed KPC mice, develop precursor PanIN lesions more quickly than KC mice, and develop focal pancreatic tumors with 100% penetrance64. Importantly, KPC mice recapitulate many aspects of the human disease with high fidelity, including the presence of distant metastases to the liver, lungs, and peritoneum; and widespread chromosomal instability in the genomes of tumor cells. Histopathologically, they express markers found in human PDA and are characterized by the same densely fibrotic stroma seen in human patients. As such, they serve as the primary reagent in translational studies hoping to elucidate the relevance of interrogated pathways to clinical application and form the basis for the mouse model discussed in Chapter 5.

Cysteine: Functions and Transport

Function

Discovered first in its dimeric form (cystine) in 1810 by the chemist William Hyde

Hollaston65, and later in its monomeric form (cysteine) in 188466, cysteine has long held a unique place in amino acid biology (Figure 1-4). Considered a semi-essential amino acid, it is the preponderant source of free sulfur in mammalian cells. As a result, cysteine plays a critical role in many, indispensable biological processes. The highlight of its structure is its highly reactive thiol group, which is readily oxidized and can act as a potent nucleophile in many enzymatic processes. The only other amino acid to contain sulfur is methionine, which can act as a sulfur-donor in the biosynthesis of cysteine through the transsulfuration pathway, but which itself is not particularly a reactive molecule. Thanks is no small part to the reactivity of its sulfhydryl group, cysteine can be considered the primary amino acid responsible for stabilizing secondary and tertiary protein structure, facilitated by the oxidation of this thiol group and the formation of covalent disulfide bridges. Cysteine residues also serve a critical functional role in many proteins, acting as unique sensors of oxidative stress67 and dynamically regulating enzyme function in response to changes in intracellular redox state.

Clearly, cysteine is vital for the role it plays in protein synthesis and structure, but it is perhaps equally appreciated for its more general antioxidant properties. The active thiol group can provide the reducing equivalents necessary for the neutralization of ROS, classically doing so in the form of glutathione (GSH), the preeminent non-protein antioxidant of the cell. GSH is synthesized in two enzymatic steps: first, the protein glutamate-cysteine ligase (GCL) catalyzes the condensation of glutamate and cysteine to form gamma-glutamylcysteine; second, glutathione synthetase (GSS) catalyzes the creation of glutathione by adding glycine to the product of the previous step. Cysteine is considered the rate-limiting precursor to the synthesis

Figure 1-4 The structure of cysteine and cystine

The semi-essential amino acid cysteine can readily be oxidized to from its dimer, cystine. The dimer is formed by the creation of a disulfide bond between two free cysteine molecules. The sulhdryl group, highlighted here in gold, is the bioreactive portion of the molecule, lending it the capacity for many of its vital functions.

of GSH primarily due to its relative intracellular scarcity when compared to glutamate and glycine. From a functional perspective, it is the sulfur provided by cysteine that acts as the bioactive portion of GSH, donating electrons to stabilize ROS and subsequently binding another oxidized GSH to form GSSG. GSSG can then be recycled by glutathione reductase at the expense of NADPH to give rise to GSH once more. Through GSH, cysteine can be considered one of the most potent molecules responsible for redox homeostasis in mammalian cells.

Cysteine also serves as a component of the biosynthesis of Coenzyme A (CoA), which is critical for the creation and oxidation of lipids. CoA plays an important role in the synthesis of

HMG-CoA from acetyl-CoA and acetoacetyl-CoA. HMG-CoA is a precursor to the production of

15 mevalonate, the critical substrate from which cholesterol and other isoprenoids are synthesized, contributing to lipid synthesis and overall cell health. Furthermore, cysteine serves important bioenergetic functions by indirectly giving rise to increased pyruvate production through its degradation. Cysteine can be degraded, beginning with its oxidation by cysteine dioxygenase to yield 3-sulfinoalanine. This molecule can then be transaminated to yield 3-sulfinopyruvate, which will quickly degrade to pyruvate and sulfite. Pyruvate is then free to enter the TCA cycle in metabolically active cells. Taken together, cysteine plays a unique role in protein function, redox homeostasis, and even cellular bioenergetics—all functions that underscore the fascination that chemists and biologists have had with cysteine for more than 200 years.

Transport

As highlighted previously, cysteine exists in two forms depending on oxidation state: as a monomer (cysteine) or as a dimer joined by a disulfide bond (cystine). Reduced cysteine can be imported into the cell primarily through the sodium-dependent ASC system, which also favors the transport of alanine, serine, and threonine68,69. System ASC is widely expressed in various tissues and cell types, modulating its affinity for certain transport candidates across species and even based on cellular context70-72. It should be noted, however, that most cysteine in the extracellular space is present in its oxidized form, cystine, and thus System ASC accounts for a low percentage of total cysteine import in cells73.

- Cystine import into cells is facilitated by the heterodimeric transporter, system xc .

Composed of two subunits, the heavy chain subunit SLC3A2 and the light chain subunit

- SLC7A11, system xc pairs the import of cystine with the export of glutamate at a one-to-one molar ratio. Of its two component subunits, SLC3A2 is common to many amino acid transporters, whereas SLC7A11 is specific to this transporter74,75. Given its specificity, SLC7A11 has been of particular interest functionally since perturbation of its expression impedes the

16 function of cystine import precisely. Located on chromosome 4 in humans and chromosome 3 in mice, the SLC7A11 gene contains 12 transmembrane domains and is expressed in a wide array of tissues, including the nervous system, liver, lung, and pancreas. The expression of this transporter is thought to be tightly regulated by levels of extracellular glutamate, amino acid starvation, and oxidative stress. Naturally occurring mutations of SLC7A11 are found in the subtle gray (sut) mouse, so-named due to its characteristic gray coat. Impressively, that absence of functional transporter in the sut mouse produces relatively mild phenotypes, including increased bleeding time, higher sensitivity to chemical carcinogens, and discolored skin and coat pigmentation. Otherwise, these mice seem to fare quite well. SLC7A11 knockout mice have also been generated and exhibit mild neurological phenotypes, but otherwise normal

Figure 1-5 Structure of SLC7A11

- SLC7A11 is the gene encoding the specific subunit of the heterodimeric system xc antiporter. Located on chromosome 3 in mice, the gene contains 12 exons (E), spanning 134,151 base pairs. The naturally occurring mutation in sut mice deletes the entirety of exon12 and more neighboring DNA.

- lifespans. While it is known that system xc can be targeted pharmacologically in culture, no conditional models of SLC7A11 deletion had yet been created or studied prior to this work, previously limiting our ability to study the transport of cystine in vivo.

Ferroptosis: Mechanism and Clinical Relevance

Mechanism

Understanding how and why cells die has been a field of active research dating back to the 19th century, but the notion that cell death can be tightly biologically regulated was an epiphany reserved for the mid-20th century. The term “apoptosis,” deriving from the Greek term to describe a “falling off”, as a petal from a flower or a leaf from a tree, was first coined in 1972.

While apoptosis had actually been identified in principle over one hundred years before, it would only become the thriving field of research it is today once it was discovered that this process was precisely regulated molecularly and it was dysregulated in some human diseases, notably in cancer.

Apoptosis, by far the most famous and well-studied cell death mechanism, has nourished a wide field of research now extending into intriguing, new mechanisms of cell death: among these, autophagic cell death, necrosis, necroptosis (programmed necrosis), and ferroptosis. Ferroptosis will be the focus of this discussion because of its close ties to oxidative stress, cystine import, and GSH: all major subjects of this work. Ferroptosis is an iron- dependent, oxidative form of cell death characterized morphologically by membrane

“ballooning” coupled with the presence of relatively intact nuclei. First formally described in

2013, ferroptosis is genetically and morphologically distinct from apoptosis, autophagic cell death, and necroptosis mediated by RIPK1.The first compounds found to cause ferroptosis were initially identified in large chemical library screens targeting RAS-mutant cell lines. These

- compounds, termed erastin and RSL3, targeted system xc and glutathione peroxidase 4

(GPX4), respectively, providing evidence that ferroptosis could be achieved through mechanistically distinct means. It is currently thought, however, that the primary effector of ferroptotic cell death is the compromised function of GPX4, whether directly (e.g. RSL3), or

18 indirectly by GSH depletion (e.g. erastin and BSO). In either situation, the cell’s inability to detoxify lipid ROS results in catastrophic damage to cellular membranes, causing the resultant cell death phenotype (Figure 1-6).

Functionally, ferroptosis is identified by the accumulation of damaging lipid ROS and its dependence on large pools of available iron. Lipophilic antioxidants including Trolox (a water- soluble vitamin E analog), α-tocopherol, and β-carotene have all been shown to potently rescue ferroptotic cell death, implicating lipid ROS as the primary mediators of this process.

Figure 1-6 Model of ferroptosis in KRAS mutant cells

Furthermore, deferoxamine (DFO), a powerful iron chelator, is able to protect cells otherwise sensitive to ferroptosis from death. Iron is a critical cofactor for the production and propagation of free radicals via the Fenton reaction. In fact, it is this dependence on iron from which ferroptosis derives its name (“ferrum,” latin for iron). To date, ferroptotic cell death is confirmed through either rescue by lipophilic antioxidant or iron chelation.

Clinical relevance

Much of the current data characterizing ferroptosis details its mechanistic underpinnings, but a burgeoning corner of the field is beginning to address critical questions determining the relevance that this cellular process might play in human diseases. Recent work has used ferostatin-1, the current canonical inhibitor of ferroptosis, to rescue the deleterious effects of pathogenic huntingtin protein in a brain slice model, the effects of cystine deprivation on cultured oligodendrocytes in a model of periventricular luekomalacia, and the cell death seen in models of rhabdomyolysis-induced kidney injury. Of critical importance for the work we lay out here, ferroptosis has been shown to occur in some cancer types, including cell culture models of diffuse large B cell lymphoma. However, better models are still needed to address whether ferroptosis truly can occur in vivo in the context of disease—a major aspect of the field pushed forward by this work and addressed in more detail in Chapter 5.

Chapter 2 – Cysteine is critical for pancreatic cancer cell growth

Background and Significance

Current treatment for metastatic PDA is based primarily on cytotoxic chemotherapeutic combinations. While increasingly aggressive regimens have produced modest increases in overall survival, only a small minority of tumors harbor “actionable” genetic alterations targeted by current precision medicine paradigms76. There is a great need for novel, translatable approaches for this disease. In recent years, a number of groups have revitalized the study of cancer metabolism with a focus on cancer–specific dependencies that might be amenable to therapeutic intervention. Notable examples include preferential reliance on serine to provide intermediates for the 1-carbon cycle 77, aspartate aminotransferase (GOT1) for the production of

NADPH in pancreatic cancer 54, IDH1 mutations for the production of 2-hydroxyglutarate in a subset of gliomas 78, glutathione (GSH) synthesis in breast cancer79, and glutamine addiction50.

Metabolic dependencies often vary by cancer type, but generally reflect the dominantly anabolic needs of tumor cells for generating the raw materials required for cell proliferation.

Recent work has shown that mutant KRAS, the oncogene most typically associated with pancreatic cancer development and maintenance, results in a cascade of metabolic changes that may offer therapeutic opportunities80. Interestingly, mutant KRAS was shown to create a highly reducing intracellular environment in pancreatic cancer cells by promoting the overexpression of Nrf2, a transcription factor controlling the activity of antioxidant genes 59.

While redox biology seems to be a critical component of the development and maintenance of

PDA and other cancers, efforts to translate agents such as buthionine sulfoxamine (BSO), an inhibitor of GSH synthesis, have not been successful. A greater understanding of the specific metabolic pathways that contribute to redox hemostasis in PDA is warranted.

In this chapter we explore the hypothesis that redox homeostasis represents a critical dependency in pancreatic cancer cells. We find that cysteine, being one of only two biological

22 sources of the fixed sulfur required for most antioxidant processes, is critical to PDA cell proliferation and survival. When deprived of cysteine by exogenous removal, pancreatic cancer cells die very quickly, exhibiting morphology characterized by a membrane “ballooning:” effect while maintaining relatively intact nuclei. We also show that these effects can be phenocopied

- by inhibition of system xc . Characterizing this cell death phenotype, we conclude that cysteine deprivation causes ferroptosis in pancreatic cancer cells, characterized by early accumulation of lipid ROS and rescuable by lipophilic antioxidants and the iron chelator DFO. These experiments show, for the first time, that pancreatic cancer cells require cysteine for viability.

Results

PDA cell lines are sensitive to cysteine withdrawal

Cysteine is a semi-essential amino acid that plays a critical role in intracellular redox homeostasis owing to its uniquely reactive sulfhydryl group, which provides reducing equivalents necessary for the function of many antioxidant molecules 81. Cysteine is the rate- limiting amino acid in the biosynthesis of glutathione (GSH), a non-protein, tripeptide thiol that mediates detoxification of reactive oxygen species (ROS) and redox maintenance82-86. Given the high rate of intracellular GSH turnover and the highly reducing intracellular environment specifically in KRAS mutant pancreatic cancer cells59, we hypothesized that exogenous cysteine sources may play a key role in pancreatic cancer cell biology.

To test this, we cultured human PDA cell lines in media containing varying concentrations of cysteine. In three of four lines (PANC-1, BxPC-3, and AsPC-1), complete cysteine withdrawal was uniformly lethal within 24 hours, with PANC-1 cells exhibiting the highest sensitivity to de-escalating concentrations of cystine. Early morphological indications of these effects were apparent visually by 16 hours (Figure 2-1A, B). Using time-lapse video microscopy, we found that cysteine-deprived cells undergo a catastrophic destabilization of their

Figure 2-1 Pancreatic cancer cells are sensitive to cysteine withdrawal

A. Visualization of PANC-1 cells over time by transmitted light. Cells were incubated in media containing 200 µM cysteine (+cys), lacking cysteine (-cys), and without cysteine but supplemented with 100 µM Trolox, a water soluble analog of vitamin E, a powerful antioxidant (-cys + Tro).

B. Cysteine dose de-escalation curves in a small panel of pancreatic cancer cell lines (PANC-1, BxPC-3, AsPC-1, and MIA PaCa-2). Cells were cultured in increasing amounts of cysteine, ranging from no cysteine to 200uM cysteine (curves shown in red), and then this experiment was performed in the presence of Trolox at 50uM (curves shown in blue). N=3 biological replicates. Error = +/- standard deviation.

plasma membranes, characterized by sudden membrane “ballooning” events, but lacking visual evidence of nuclear fragmentation. We hypothesized that due to the key role cysteine plays in redox biology, that the cell death in these cells was oxidative in nature. To test this, we attempted to rescue the effects of cysteine withdrawal with the antioxidant Trolox. In all sensitive lines, Trolox provided a near complete rescue, even in the total absence of cysteine (Figure 2-

1A, B). In contrast, MIA PaCa-2 cells proved resilient to these cell cultures conditions, a phenomenon which we will discuss in detail in Chapter 3. The partial loss of relative cell viability

24 observed at the lowest cysteine concentrations was not rescued by Trolox, likely indicating a different form of cell death or a cell proliferation defect occurring in these cells.

- Pharmacological inhibition of system xc phenocopies the effects of cysteine withdrawal

Exogenous cystine enters the cell primarily through the glutamate/cystine antiporter,

- 87 system xc . To test whether the cell death phenotype induced by cysteine withdrawal could be reproduced pharmacologically, we performed dose–escalation viability experiments on the

- same PDA lines using piperazine erastin (PE), a potent analog of the system xc inhibitor erastin88. PDA cells that exhibited sensitivity to cysteine withdrawal were also highly sensitive to

- system xc inhibition (Figure 2-2A, B). Microscopic examination found that sensitive cells died beginning at 8 hours after treatment with PE and almost completely by 24 hours. This cell death could largely be rescued by co-treatment with Trolox, just as in the case of the cysteine withdrawal discussed above. Promoting cystine uptake by the neutral amino acid transporter system L through treatment with β-ME89 also rescued these effects. To test that cysteine itself

- could rescue the effects of system xc inhibition, ruling out off target effects of our inhibitor, we treated cells with both PE and N-acetyl cysteine (NAC), a cell permeable cysteine precursor.

- Results showed that NAC could ably rescue the effects of system xc inhibition, confirming cysteine itself is mediating the effects of this drug treatment (Figure 2-2C, D). Once again, MIA

PaCa-2 cells were largely insensitive to these perturbations. Interestingly, the cellular phenotype

- in PANC-1 cells treated with a system xc inhibitor was morphological distinct from the apoptosis caused by staurosporine (Figure 2-2E). We also noted that among these four lines, there was

- parity between the sensitivity to cysteine depletion and system xc inhibition. Taken together, these data indicate that pancreatic cancer cells are sensitive to cysteine deprivation, whether directly or pharmacologically.

A B

C D

- Figure 2-2 System xc inhibition phenocopies the effects of cysteine withdrawal

A. A. Visualization of PANC-1 cells over time in media containing vehicle (.1% DMSO, veh), 10 µM piperazine erastin (PE), and 10uM PE with 100uM Trolox (PE + Tro).

B. Standard dose response curves in a small battery of pancreatic cancer cells lines ranging from 0 µM PE to 100 µM PE (curves in red). Cells were also treated with increasing amounts of PE in combination with 50 µM Trolox (curves in blue). Viability was assessed at 24 hours following treatment. N=3 biological replicates. Error = +/- SD.

C. Cell viability of PANC-1 cells cultured with 10 µM PE in combination with 100 µM Trolox, 50 µM β-ME, or 1mM NAC. Error is +/- SD, * = p < .05 by student’s t-test. N = 3 technical replicates.

D. Dose response curve testing the effects of PE alone or PE in combination with 1 mM NAC on PANC-1 cells. Error is +/- SD, n = 3 biological replicates.

E. 100x image of PANC-1 cells cultured in the 5uM PE or .2uM staurosporine after 16 hours.

Cysteine deprivation causes ferroptosis in pancreatic cancer cell lines

While it was clear from our observations that some PDA cell lines were exquisitely

- sensitive to cysteine withdrawal or system xc inhibition, we had not fully interrogated the means by which these cells were dying. Rescue with the antioxidant Trolox hinted at a rapid, oxidative cell death. This is reminiscent of a form of an iron-dependent, non-apoptotic cell death called ferroptosis associated with the accumulation of lipid ROS species 90. To assess whether cysteine deprivation induces ferroptosis in sensitive PDA lines, we cultured PANC-1 cells in media containing 10 µM PE and attempted to rescue these effects with two canonical inhibitors of ferroptosis, Ferrostatin-1 (Fer-1, a lipophilic antioxidant) and deferoxamine (DFO, an iron chelator). Microscopically, rescue with fer-1 or DFO restored normal morphology to these cells even at 24 hours (Figure 2-3A).

To quantify these results and rule out other possible forms of cell death, we treated our panel of PDA cell lines with media lacking cysteine or containing 10 µM PE and attempted to rescue these effects with a variety of cell death inhibitors. Fer-1 and DFO provided the most consistent and highest level of rescue in all sensitive lines. By contrast, inhibitors of apoptosis

(Z-VAD-FMK), necroptosis (Necrostatin-1s), or autophagy (Bafilomycin A1) failed to rescue the death of these cells as effectively following cysteine depletion. In the insensitive MIA PaCa-2 line, no tested agent effectively rescued the effects of cystine withdrawal, consistent with a cell proliferation defect rather than a cell death phenotype (Figure 2-3B). Similarly, cells treated with

- a system xc inhibitor were rescued most effectively by either fer-1 or DFO. It should be noted that when some cells were treated with PE, other agents could provide substantial rescue, however. AsPC1 cells, for instance could be rescued with a variety of agents, suggesting that the specificity of these effects may vary by cell line.

Figure 2-3 Cysteine deprivation causes ferroptosis in pancreatic cancer cells

A. Images of PANC-1 cells treated for 24 hours with vehicle (.2% DMSO), 10 µM PE, 10 µM PE with 500 nM ferrostatin -1 (PE + Fer-1), and 10uM PE with 100uM DFO (PE + DFO). Scale bars indicate 50 µM.

B. Cell viability of pancreatic cancer cells cultured in vehicle (.1% DMSO, veh, shown in gray), the absence of cysteine (no cys, shown in red), and in the absence of cysteine but in the presence of 500 nM ferrostatin-1 (Fer-1, shown in blue), 10uM Necrostatin 1s (Nec1s, shown in green), 50uM ZVAD-FMK (ZVADFMK, shown in purple), 1nM bafilomycin A1 (BA1, shown in orange), and 100uM deferoxamine (DFO, shown in pink). Viability was assessed after 24 hours of treatment . Error equals +/- standard deviation. N = 3 biological replicates. * = p<.05 by an unpaired, non-parametric student’s t test. x = no significant difference.

C. Cell viability of PANC-1 and MiaPACA-2 cells cultured in vehicle (.2% DMSO, veh, shown in gray), 5 µM PE (shown in red), and the aforementioned compounds and the concentrations indicated in figures G and I. Viability was assessed at 24 hours after treatment. Error equals +/- standard deviation. N = 3 biological replicates. * = p<.05 by an unpaired, non-parametric student’s t test. x = no significant difference

Another characteristic feature of ferroptosis is the rapid accumulation of lipid ROS in the plasma membrane. Using the lipid ROS sensor C-11 Bodipy, we observed a sharp increase in lipid oxidation in PANC-1 cells during PE treatment that temporally preceded the onset of cell death. This effect was fully rescued by co-treatment with Trolox as quantified by flow cytometric analysis (Figure 2-4A). In contrast, plasma membrane ROS accumulation was not induced by buthionine sulfoxamine (BSO), a well characterized inhibitor of GSH synthesis 91, a finding we will explore in more detail in Chapter 4. These effects were absent from MIA PaCa-2 cells under any condition (Figure 2-4A), save treatment with cumene hydroperoxide, a well-known inducer of lipid peroxidation92. These results are consistent with the lack of overt cell death in MIA

PaCa-2 cells that we observed in previous experiments.

To explore whether the morphological effects we see in sensitive cells are correlated directly with lipid ROS accumulation, we visualized both sensitive and insensitive cells under the effects of cysteine deprivation (Figure 2-4B). Interestingly, lipid oxidation is present at relatively high levels in PANC-1 cells, but not MIA PaCa-2 cells, by 6 hours following either cysteine

- withdrawal or treatment with the system xx inhibitor IKE. To quantify these observations, we used a flow cytometric approach. PANC-1 cells treated with IKE exhibited a sharp increase in C-

11 Bodipy positivity by 6 hours after treatment (Figure 2-4C). In these samples, up to 72.3% of detected cells were positive for the oxidized form of C-11Bodipy. This lipid oxidation accumulation could not be alleviated by ZVAD-FMK, Nec1s, or BA1. However, in Fer-1 treated samples positivity was reduced to only 14.1%. When we examined the effects of these treatments on MIA PaCa-2 cells, we noted no profound increase in C-11 Bodipy positivity. Thus, we found that the cell death phenotype induced by cysteine deprivation in PANC1 cells was iron-dependent, non-apoptotic, oxidative, and tightly associated with the accumulation of lipid

ROS. From these collective observations, we conclude that cysteine deprivation in PANC-1 cells induces ferroptosis.

Figure 2-4 Cysteine deprivation causes a catastrophic accumulation of lipid ROS characteristic of ferroptosis

A. Flow cytometric analysis of PANC-1 and MIA PaCa 2 cells stained with C-11 Bodipy, a marker of lipid oxidative stress, after 6 hours of treatment with indicated compounds. Cells were treated with vehicle (.2 % DMSO, gray); 10 µM PE (red); 10 µM PE and 100 µM Trolox (blue); 600 µM of the GSH-synthesis inhibitor buthionine sulfoxamine (BSO, in orange); and 100 µM cumene hydroperoxide, a lipid free-radical initiator, (cumene, in green). Representative data from one of three experiments are shown.

B. Fluorescent visualization of PANC-1 and MIA PaCa 2 cells stained with C-11 Bodipy. Reduced lipids are shown in red, oxidized lipids are shown in green, overlayed with a transmitted light image. Cells were treated with vehicle (.2% DMSO, veh), incubated in media lacking cysteine (-cys); or treated with 5uM IKE , a more potent analog of the system xc- inhibitor erastin (IKE). All images were captured after 6-8 hours of treatment.

C. Flow cytometric quantification of PANC-1 and MIA PaCa-2 cells stained with C-11 Bodipy after six hours of treatment with 5 µM IKE Positive cells fall into quadrant 3 (Q3). Representative data from one of three experiments are shown. 30

Discussion

While research has shown that mutant RAS pancreatic cancer cells create a uniquely reducing intracellular environment via Nrf2 upregulation59, research detailing if this biological change can be targeted therapeutically is still scant. Recent work has indicated that targeting this system in conjunction with the Akt pathway can inhibit pancreatic cancer cell growth93, however, it has not yet been shown that targeting a metabolic dependency based on redox biology alone can inhibit tumor cell growth. Here we show for the first time that depriving pancreatic cancer cells of cysteine, whether directly by exogenous removal from culture media

- or indirectly by pharmacological blockade of system xc , can result in abrupt cell death. Other cell lines are resistant to cysteine-depleted culture conditions94, highlighting that cysteine withdrawal does not represent a universally lethal stress. Interestingly, while previous research has shown that various forms of nutrient deprivation can cause cell death by apoptosis or regulated necrosis in other cellular contexts95-98, we show that pancreatic cancer cells do not die from apoptosis when deprived of cysteine. However, pancreatic cancer cells are not uniformly sensitive to this metabolic stress: MIA PaCa-2 cells were not particularly sensitive to the effects of cysteine deprivation. This can be interpreted in multiple ways: perhaps these cells create cysteine from other sources of sulfur (transsulfuration, catabolism of extracellular proteins); upregulate other antioxidant pathways to accommodate cysteine depletion; or these cells simply do not produce ROS at a high enough rate to create the accumulation of lipid damage necessary to execute the phenotypes we see. This phenomenon will be the subject of more extensive inquiry in the next chapter.

Cysteine is typically present in the extracellular space as cystine, which is imported

- specifically by system xc . Here, we confirm the cystine is the major source of intracellular

- cysteine in standard culture conditions. Importantly, pharmacological blockade of system xc phenocopies the effects of cysteine withdrawal in sensitive pancreatic cancer cells. This result

31 highlights the inherent druggability of this dependency and bears considerable importance on

- the clinical value of this finding. Previous work has highlighted that system xc inhibition can cause an iron-dependent, oxidative cell death called ferroptosis99. Here we show that, indeed, cysteine deprivation in pancreatic cancer also causes ferroptosis, as evidenced by rescue with both ferrostatin-1 and DFO. PANC-1 cells appear to be the most sensitive to the effects of cysteine deprivation and also appear to die almost exclusively from ferroptosis in our experiments, making them the ideal model cell line to study these effects moving forward. It is clear from our work however, that ferroptosis may not account for the only death we see in all cell lines. AsPC-1 cells, for example, can be rescued from these stresses quite well when using an inhibitor of necroptosis or an inhibitor of autophagy, suggesting interplay between multiple avenues of cell death. ZVAD-FMK consistently provides the worse rescue effects across all cell lines, underscoring though that apoptosis does not contribute to any appreciable cell death in these contexts. Fer-1 and DFO both provide the most reliable rescue in all cell lines, suggesting that ferroptosis plays a critical role in mediating the effects in these cells, nonetheless.

For the first time we show live-cell imaging implicating the accumulation of lipid ROS in the potentiation of all of these effects. Consistent with an overall resistance to cysteine deprivation, MIA PaCa-2 cells do not accumulate lipid peroxides under these conditions. It is important to note, however, that cumene hydroperoxide can create lipid peroxidation in these cells, suggesting that they are not simply incapable for the production and propagation of this type of damage. Imaging data and time-lapse microscopy showed that the accumulation of lipid peroxides and the morphological changes associated with ferroptosis occur in sequence, suggesting that lipid peroxides may tightly regulate the effects of cysteine deprivation, consistent with previous reports100-103. In these experiments, we also tested a known inhibitor of

GSH synthesis, buthionine sulfoxamine (BSO), since we hypothesized that GSH depletion mediated the accumulation of lipid peroxides, based on previous work88. Surprisingly, there was

32 no evidence that BSO could cause the same accumulation of lipid damage over the equivalent time span. These results prompts questions regarding the precise role that cysteine, GSH, and other cysteine derivatives actually play in ferroptosis in these cells, forming the basis for inquiry that we describe in more detail in Chapter 4.

Chapter 3 - Macropinocytosis, but not transsulfuration, mediates insensitivity to the

- effects of system xc inhibition in pancreatic cancer cells

Background and Significance

In a clinical setting, cancers do not respond equally well to all treatment regimens. In fact, one of the major goals of modern precision medicine is to determine, whether by genomic profiling or master-regulator analyses, what drugs will work most effectively in a given patient.

As such, modelling variable drug response in cancer may prove important in foreseeing possible modes of resistance to a proposed means of treatment. In our previous experiments, we found that three out of four pancreatic cancer cell lines tested were sensitive to cysteine depletion via ferroptotic cell death. However, Mia PACA-2 cells exhibited a priori insensitivity to this nutrient deprivation. We thought it valuable to study the biology of cells which were insensitive to cysteine deprivation as a means to model the biology of tumor that may not respond to system

- xc inhibition.

We hypothesized that Mia PACA-2 cells were not in fact resistant to cysteine deprivation per se, but instead able to acquire cysteine by alternative means. In addition to importing cystine, cells can synthesize cysteine by transferring the sulfhydryl group of methionine to the carbon backbone of serine using the transsulfuration pathway104-108. Previous work has highlighted that induction of the transsulfuration pathway can inhibit ferroptosis induced by cysteine deprivation109, making this an attractive pathway to interrogate in MIA PaCa-2 cells.

Cysteine could also be acquired through the uptake of exogenous materials via vesicle- mediated processes such as macropinocytosis. Protein uptake via macropinocytosis has been shown to provide amino acids in RAS–driven cancers, and inhibition of this process compromises the growth of pancreatic tumor xenografts 48. In this chapter, we explore the hypotheses that these pathways confer resistance to ferroptosis in MIA PaCa-2 cells.

Results

Cysteine biosynthesis through transsulfuration provides a minor component of total

35 cysteine levels in pancreatic cancer cells

To interrogate if transsulfuration plays a critical role in providing free cysteine to pancreatic cancer cells, we undertook a series of labeling experiments to track the movement of carbons donated by methionine to other metabolites (Figure 3-1). Transsulfuration begins with methionine, which is enzymatically converted to S-adenosylmethioine (SAM) by methionine adenosyl transferase. Various methyl transferases may act on SAM to yield S-

Adenosylhomocysteine (SAH). SAH is converted into homocysteine by a specific hydrolase.

Homocysteine then goes on to one of two fates: it can be converted back into methionine by the enzyme methionine synthase, or it can be converted to cystathionine by the addition of a serine as catalyzed by cystathionine β-synthase (CBS). To complete the production of cysteine, cystathionine γ-lyase (CGL) decomposes cystathionine into cysteine, producing an ammonium ion and α-ketobutyrate. In our experiment, we cultured PANC-1 and MIA PaCa-2 cells in medium containing methionine synthesized with isotopically-labeled carbon (13C). In this experiment, cells were treated concomitantly with labeled substrate and various compounds.

Metabolites were extracted and then subject to mass spectrometry-based profiling to analyze

13C-label incorporation into components of the transsulfuration pathway. In both cell lines we found that nearly 50% of the cystathionine pool was labeled with 13C after 6 hours, and this reached 90% by 24 hours. At no point did we detect labeled α-ketobutyrate. Of note, pathway labeling was not altered by treatment with either BSO or PE, suggesting that neither cell line modulates activity of this pathway in response to cysteine deprivation or GSH depletion.

Furthermore, metabolite abundance of transsulfuration intermediates did not change in either sensitive or insensitive cells upon treatment with BSO or PE (Figure 3-2A, B).

To test if long-term incorporation of labeled methionine could be perturbed in the context

- of system xc inhibition, we performed a second labeling experiment in which cells were incubated with 13C labeled methionine for 16 hours prior to drug administration, allowing the

Figure 3-1 The transsulfuration pathway

Detailed above are key steps in the transsulfuration pathway, which in mammals catalyzes the biosynthesis of cysteine from methionine. In the described labeling experiment, 13C labeled carbon is used to synthesize methionine, which can then be tracked by mass spectrometry. Metabolites that derive some subset of their carbon skeleton from carbons in methionine are indicated in red typeface. Enzymes that catalyze these reactions are boxed in blue.

Abbreviations: MAT, methyl adenosyl transferase; MT, methy transferase; SAHH, S-adenosyl homocysteine hydrolase; CBS, cystathionine β-synthase; CGL, cystathionine γ-lyase; MS, methionine synthetase.

- Figure 3-2 Neither system xc inhibition nor GSH synthesis impairment cause increase flux through the transsulfuration pathway

Measurements of incorporation of 13C-labeled methionine into transsulfuration intermediates. Cells were pretreated for 16 hours with labeled substrate and then treated with vehicle (.05% DMSO, shown gray) or 5 µM PE (shown in red), for six hours. N = 3 biological replicates. Error = +/- SD. Of note, alpha-ketobutyrate is unlabeled in all treatments, showing that the final step of transsulfuration is not active in these cells under any of the treatment conditions.

38 entire pool of transsulfuration intermediates to be labeled prior to metabolic stress. Structured this way, the experiment would allow us to detect steady state levels of transsulfuration products. Examining the percent of total metabolite pools labeled, we detect nearly complete labeling of all transsulfuration intermediates in both vehicle and PE treated samples in both cell lines. In contrast, α-KB labeling was not detected in either PANC-1 or MIA PaCa-2 cells under either treatment condition, which may reflect low flux through the final step of the transsulfuration pathway (Figure 3-3A, see Figure 3-1). To complement these biochemical data, we assessed whether transsulfuration pathway activity functionally affects the sensitivity of

PDA cells to cysteine depletion. MIA PaCa-2 cells were treated with propargylglycine (PPG), which inhibits cystathionase (CGL), the enzyme which catalyzes the final step of transsulfuration110. PPG alone did not alter cell viability, even at high concentrations (Figure 3-

3B). Moreover, co-treatment of PE at varying concentrations with 1mM PPG did not alter the sensitivity of MiaPACA-2 cells to PE–induced cell death (Figure 3-3C). These data are consistent with the results of the 13C-methionine labeling experiment above that indicate transsulfuration pathway activity is not altered by PE treatment within 6 hours. From these collective experiments, we conclude that the transsulfuration pathway does not contribute appreciably to cysteine pools in these PDA cell lines and is not a major determinant of sensitivity to cysteine depletion in this context.

Macropinocytosis is upregulated in MIA PaCA -2 cells, but not PANC-1 cells, in response

- to system xc inhibition

Cysteine could also be acquired through the uptake of exogenous materials via vesicle- mediated processes such as macropinocytosis. Protein uptake via macropinocytosis has been shown to provide amino acids in RAS–driven cancers, and inhibition of this process compromises the growth of pancreatic tumor xenografts 48. To test whether macropinocytosis alters the sensitivity of PDA cells to cysteine depletion, we first profiled its activity in PANC-1

Figure 3-3 Transsulfuration does not mediate sensitivity of pancreatic cancer cells to cysteine deprivation

A. PANC-1 and MIA PaCa-2 cells were cultured with 13C labeled methionine for 16 hours and then treated with either vehicle (.1% DMSO, in gray), or with 10 µM PE (in red) for 6 hours. Metabolites were then extracted and subject to analysis by mass spectrometry. No significant changes in labeling were identified in the metabolites shown here, and notably, no labeling was detected in α-KB. N = 3 biological replicates. Error = +/- SD.

B. Dose response of propargyglycine (PPG), a transsulfuration inhibitor, in MIA PaCa-2 cells. Cells were treated for 24 hours. N = 3 technical replicates. Error = +/- SD.

C. Dose response of PE alone, or in combination with 1 mM PPG for 24 hours in MIA PaCa-2 cells. Cells were treated for 24 hours. N = 3 biological replicates. Error = +/- SD.

and MIA PaCa-2 cells by measuring uptake of a high molecular weight fluorescently labeled dextran, as described previously111. Using high magnification fluorescent microscopy, we noted that healthy PANC-1 cells in their logarithmic growth phase did not form macropinosomes to any appreciable extent (Figure 3-4A, B). Notably, MIA PaCa-2 cells, which were insensitive to cysteine deprivation, showed high macropinocytotic activity in vehicle treated samples. In the

MIA PaCa-2 cells, dextran uptake was further increased by co-treatment with PE, and was fully blocked by the macropinocytosis inhibitor ethylisopropylamiloride (EIPA)112 (Figure 3-4C, D).

These results suggest that macropinocytosis is upregulated in direct response to metabolic stress, a means to acquire the biomolecular substrates required for survival in these conditions.

- Macropinocytosis mediates sensitivity to system xc inhibition in MIA PaCa-2 cells

Next we examined the impact of macropinocytosis inhibition on the cell viability of MIA

- PaCa-2 cells following system xc inhibition. Whereas treatment of these cells with PE at increasing concentrations had no effect on viability, the addition of EIPA led to near-complete

- sensitization to system xc inhibition (Figure 3-5A). The loss of viability was almost completely rescued by NAC and Trolox, similar to results seen in PANC-1 cells treated with PE alone, as previously discussed (See Figure 2-2C). To confirm that EIPA alone was not dictating these effects, we carried out a cell viability assay, treating MIA PaCa-2 cells with an array of compounds to attempt to rescue these effects, which we hypothesized were mediated by ferroptosis, as in previous examples. While neither PE nor EIPA alone killed MIA PaCa-2 cells, the combination of these drugs managed to achieve nearly complete cell killing. These effects were rescued at least partially by co-treatment with β-mercaptoethanol (β-ME), a potent reducing agent; DFO; Fer-1; and NAC, providing strong evidence that these otherwise–resistant

- cells are dying at least partially of ferroptosis following a combination system xc and macropinocytosis inhibition (Figure 3-5B). We therefore conclude that macropinocytosis contributes to the sensitivity of pancreatic tumor cells to cysteine deprivation.

Figure 3-4 MIA PaCa-2 cells, but not PANC-1 cells, upregulate macropinosome formation in - response to system xc inhibition A, B. Fluorescent microscopic visualization of PANC-1 and MIA PaCa-2 cells treated for six hours with vehicle (.5 % DMSO); 10 µM PE; 50 µM EIPA, an inhibitor of micropinocytosis; and with PE and EIPA in combination.

C, D. Quantification of fluorescence microscopy data presented above. The average number of macropinosomes per cell was counted for 10 randomly selected fields per condition to calculate average macropinoctytotic index. Error bars indicate mean +/- S.D, with N=3 biological replicates. * = p<.05, Kruskal-Wallis ANOVA Error = +/- SD.

Figure 3-5 Inhibiting macropinocytosis sensitizes previous unaffected pancreatic cancer cells to - system xc inhibition A. Cell viability assessment of MIA PaCa-2 cells treated with vehicle (.5% DMSO), varying concentrations of PE alone; PE in combination with 50 µM EIPA; PE, EIPA and 1 mM N-acetyl cysteine (NAC), a cell membrane permeable analog of cysteine; and PE, EIPA, and 100 µM Trolox, a lipohilic antioxidant. Each condition was normalized to its respective vehicle. Error bars indicate +/- SD, N=3 biological replicates. Statistical significance was determined via two-way ANOVA then Tukey’s multiple comparisons test; *p<0.0001. All measurements were taken at 24 hours post initiation of treatment.

B. Cell viability assessment of MIA PaCa-2 cells treated for 24 hours with vehicle (0.5% DMSO), PE (15 μM), EIPA (50 μM), Trolox (100 μM), BME (100 μM), DFO (100 μM), Fer-1 (500 nM), and NAC (500 μM), or as indicated. Error bars indicate +/- SD, N=3 technical replicates. Statistical significance was determined via ANOVA; *p<0.0001.

Discussion

In Chapter 2, we determined that a subset of pancreatic cancer cell lines were

exquisitely sensitive to cysteine deprivation. Notably, one cell line, MIA PaCa-2, exhibited

almost no ill effects from these perturbations within 24 hours. In efforts to model potential

43 mediators of sensitivity to cysteine deprivation and thereby ferroptotic cell death in a cancer context, we utilized the marked differences in PANC1 and MIA PaCa-2 cells to execute the experiments discussed herein. Previous research had shown that transsulfuration was a potential mediator of resistance to ferroptosis in the context of cysteine deprivation109, providing an attractive candidate to test in our experimental paradigm. Using a methionine-based labeling approach, we assessed flux through the transsulfuration pathway in our two model cell lines.

Interestingly, transsulfuration intermediates were in fact produced in these cells in vehicle samples, proving that at least portions of the pathway are active under normal conditions.

Fascinatingly, we failed to detect labeled αKB in either cell line under any conditions, which is consistent with a model in which pancreatic cancer cells actively create cystathionine but fail to catalyze the final reaction in the transsulfuration pathway. It is possible that we did not detect

αKB because this metabolite is produced but is quickly and regularly transported out of the cell,

- but results in which we use PPG (a transsulfuration inhibitor) and a system xc inhibitor in combination fail to produce any synergistic effects, consistent with limited pathway activity. This leaves us with the much more likely model that cysteine is simply not biosynthesized at the level required to confer resistance to either PANC-1 cells or MIA PaCa-2 cells, suggesting that any

- resistances to system xc inhibition in pancreatic cancer are not likely to be overcome by transsulfuration inhibition.

Macropinocytosis is a well-studied endocytic process that has been linked to the expression of oncogenes, including mutant RAS113. In pancreatic cancer specifically, it has been identified as an alternate route by which amino acids can be acquired through bulk uptake of extracellular contents48. In that study, the model cell lines chosen to explore these effects were in fact MIA PaCa-2 cells and BxPC3 cells, wherein MIA PaCa2-cells exhibited high levels of macropinocytosis. Here, we recapitulate that finding, and show for the first time that PANC-1 cells do not produce macropinosomes at appreciable numbers. We also show for the first time

44 that otherwise resilient cells can be sensitized to ferroptosis by way of cysteine deprivation by inhibiting macropinocytosis. These data support the possibility of using macropinocytotic index as a biomarker of potential sensitivity to both cysteine deprivation and ferroptosis, but more research is warranted in this vein to confirm this prospect. From a clinical perspective, this

- research is important because it models the possible outcomes of system xc inhibition in a patient setting. One of the major goals of precision medicine is to define subset of patients populations that will respond best to a given treatment regimen. Put more simply: to find the right treatment for the right patients. Here, we model a possible mechanism by which cells can

- avoid the effects of system xc inhibition and provide a solution to overcome this resistance.

There are no doubt other mechanisms by which cells can avoid the ill effects of nutrient deprivation, and more work to highlight these potential mechanisms will serve us well as we make efforts to translate these results into clinically actionable pancreatic cancer treatments.

Chapter 4 – Cysteine, GSH, and their roles in mediating ferroptosis

in pancreatic cancer cells

Background and Significance

We have shown that pancreatic cancer cells are sensitive to cysteine deprivation and even those that are insensitive can be induced to undergo a similar mechanism of cell death under those conditions (ferroptosis), provided we inhibit other modes of cysteine acquisition

(macropinocytosis). The prevailing mechanism for how cysteine depletion may lead to ferroptosis is based on its role as the rate-limiting precursor for glutathione (GSH) 88,114. GSH itself can directly react with ROS, or it may be used as a cofactor by enzymes that catalyze the detoxification of particular ROS. One example of such an enzyme is glutathione peroxidase 4

(GPX4), which is the only known lipid peroxidase in mammals115,116. To date, ferroptosis–

- inducing compounds have been found either to inhibit system xc or GPX4, which has led to a

- model in which depletion of glutathione pools by system xc inhibition leads to inactivation of

GPX4, causing the catastrophic accumulation of lipid ROS that are responsible for killing these cells.

Because GSH levels correlate with response to cytotoxic treatment117,118, limiting its production has been the subject of clinical inquiry for over 50 years. BSO has been found to profoundly deplete intracellular stores of GSH in patient tumor samples, but has not been tested as a monotherapy119. Early clinical trial work focused on the combination of BSO with the alkylating agent L-PAM (melphalan), but adverse toxicity profiles and limited in vivo efficacy may limit the potential of this dual-combinatorial approach to sensitizing tumors to chemotherapy120. In previous experiments, we noted that the GSH-synthesis inhibitor BSO failed to recapitulate some of the phenotypic responses of pancreatic cancer cells to cysteine deprivation (Figure 2-4A).However, we had not formally tested the effects of BSO treatment over longer periods nor had we definitively shown that GSH levels in these cells were completely ablated by BSO treatment. In efforts to more thoroughly characterize the roles of cysteine, GSH, and other possible cysteine derivatives in causing ferroptosis in PDA cells, we

47 describe in this chapter a series of experiments that show for the first time that GSH depletion is necessary, but not sufficient to cause the deleterious effects of cysteine deprivation in PDA cell lines. Furthermore, using a cysteine-labeling approach analogous to the system described for methionine-tracing in Chapter 3, we attempt to delineate key differences between cells that have been treated with a GSH-depleting agent versus cells that have been wholly deprived of cysteine. We implicate other cysteine derivatives required for the de novo synthesis of undamaged lipids or for the production of lipophilic antioxidants as a necessary component of engaging ferroptotic cell death in these cells, refining the current model of ferroptotic cell death.

Results

GSH depletion is necessary, but not sufficient to cause ferroptotic cell death in PANC-1 cells

To test if GSH depletion is necessary for the cell death effects we see in PDA cells, we attempted to rescue cells deprived of cysteine by the addition of a cell permeable form of GSH,

GSH-ethyl ester (GSH-EE)121. We found that high millimolar concentrations of GSH-EE were able to provide complete rescue from culture conditions absent of cysteine in PANC-1 cells

(Figure 4-1A), consistent with research showing that intracellular concentrations of GSH can

122 - range from 1-10 mM . PANC-1 cells treated with a system xc inhibitor were similarly able to be rescued with GSH-EE (Figure 4-1B). Therefore, providing the cell with exogenous GSH-EE is

- sufficient to rescue PDA cells from ferroptosis. To confirm that system xc inhibition did in fact deplete GSH levels in these cells, we used a mass spectrometric platform to assess the levels of intracellular GSH directly. Consistent with a model in which GSH depletion is critical for ferroptosis, we see that both GSH and GSSG (oxidized GSH) levels drop precipitously in

PANC-1 cells (Figure 4-1C). To get a clearer picture of the redox state of these cells, we next calculated the ratio of GSH:GSSG, a marker of oxidative stress and overall cell health123-126. As

expected, GSH:GSSG ratio dropped very low in PANC-1 cells, confirming the compromised

capacity of these cells to maintain proper redox balance under the constraints of cysteine

deprivation. Interestingly, we were not able to measure any significant drop in GSH, GSSG, or

- GSH:GSSG levels in MIA PaCa-2 cells, consistent with their overall insensitivity to system xc

inhibition (Figure 4-1D). Taken together, these data suggest that GSH depletion is necessary to

achieve ferroptotic cell death in PDA cells.

Figure 4-1 GSH depletion is required for ferroptotic cell death

A, B. Rescue of the effects of growing PANC-1 cells in the absence of cysteine or when treated - with the system xc inhibitor PE by supplementing with GSH-ethyl esther (GSH-EE). N = 3 biological replicates. Error bars +/- SD. * = p < .05 by t test. Measurements are taken 24 hours after treatment initiation.

C, D. The effects of system xc- inhibition on the depletion of intracellular GSH, oxidized GSH (GSSG), and the ratio of these two molecules in both PANC-1 and Mia-PACA2 cells, as measured by mass spectrometry. Analytes were extracted and processed 6 hours after treatment initiation. N = 3 biological replicates, error bars +/- SD. * = p < .05 by t test

Another prediction of the model above is that glutathione depletion alone may be sufficient to induce ferroptosis. We therefore evaluated the effects of GSH depletion on PDA cell viability using buthionine sulfoxamine (BSO), an inhibitor of gamma-glutamylcysteine synthetase, the penultimate step of GSH synthesis127. Using mass spectrometry, we affirmed that BSO effectively depleted both reduced and oxidized glutathione (GSH and GSSG, respectively) from PANC-1 cells, leading to near-complete loss of GSH pools over 24 hours.

(Figure 4-2A).Contrastingly, MIA PaCa-2 cells exhibited slightly differently GSH depletion rates.

While initial abundance of GSH was comparable across the two lines, BSO also managed to deplete intracellular GSH, as expected, albeit at a slightly slower rate when compared to PANC-

1 cells. (Figure 4-2B). However, in stark contrast to cysteine deprivation, GSH depletion had no effect on cell viability at 24 hours (Figure 4-2C). We conclude from these results that GSH depletion is necessary, but not sufficient, for the induction of ferroptosis following cysteine depletion.

Exogenous cysteine is incorporated into acetyl CoA pools and CoA byproducts can rescue cell death caused by cysteine deprivation

We infer from the findings above that, in concert with GSH depletion, the loss of one or more additional cysteine-derived functions or metabolites contributes to the induction of ferroptosis. In mammals, cysteine is utilized in a limited number of reactions (Figure 4-3).

Outside of GSH, additional products of these reactions include proteins (via translation), taurine and pyruvate production (via cysteine sulfinate production), and Coenzyme A (CoA, via the pantothenate pathway). We reasoned that the timescale over which ferroptosis induction occurs (~8 hours) following cysteine deprivation is not consistent with an effect on protein translation, which would take a much longer time to affect protein pools and cell viability128. In a manner akin to the 13C- methionine labeling experiment described in Chapter 3, we analyzed metabolite fractions by LC-MS based metabolomics from 13C-labeled cystine samples. Here,

Figure 4-2 GSH depletion is not sufficient to case ferroptotic cell death

A, B. Mass spectrometric measurements of intracellular pools of GSH in PANC-1 and MIA PaCa-2 cells treated with vehicle (.2% DMSO) or 600 µM BSO Cells were processed at 6 hours or 24 hours post treatment initiation. N=3 biological replicates, errors bars +/- SD.

C. The effects of a dose escalation of BSO on PANC-1 and MIA PaCa-2 cells’ viability. N = 3 biological replicates. Error bars +/- SD.

we found that labeling of taurine, lactate, glutamate, and citrate were negligible (i.e. below level

of detection) after 24 hours (Figure 4-4). Pyruvate is not readily detectable by LC-MS. As such,

metabolites derived from pyruvate in the cytosol (i.e. lactate) or mitochondria (i.e. citrate and

glutamate) were analyzed. These results suggest that neither taurine biosynthesis nor

cysteine’s contribution to pyruvate synthesis and the production of TCA cycle intermediates is

minor.

Figure 4-3 Cysteine and its biomolecular products

Pictured above is a working model of cysteine and its various biomolecular roles in the cell. It is critical for protein synthesis, some portion of pyruvate production, taurine biosynthesis, the production of Coenzyme A, and the synthesis of GSH. Isotopically labeled carbons are highlighted in red, while enzymes are boxed in blue.

Abbreviations: HDH, hypotaurine dehydrogenase; CSAD, sulfinoalanine decarboxylase; CDO, cysteine dioxygenase; GCL, glutamate-cysteine ligase; GSS; glutathione synthetase; STA, sulfino- transaminase.

Figure 4-4 Cysteine does not enter the cysteine sulfinate production branch of cysteine metabolism in pancreatic cancer cells

The measurement of 13C-labeled cysteine products in a variety of treatment settings. PANC-1 and MIA PaCa-2 cells were treated with vehicle (.05% DMSO, in gray and dark gray), 5 µM IKE (in red), or 600 µM BSO (in gold and dark gold) for indicated time points. No labeled metabolites were detected in this experiment under any condition.

Given these results, we focused our investigation on Coeznyme A (CoA) and its derivatives. Indeed, under basal conditions in both PANC-1 and MIA PaCa-2 cells, fully 50% of cellular CoA becomes labeled over the course of 24 hours in baseline conditions, indicating a steady contribution of exogenous cysteine to CoA pools (Figure 4-5A).CoA is an essential substrate for the de novo synthesis of fatty acids, sterols, and the production of lipophilic antioxidants, all of which could play a critical role in alleviating the toxic effects of lipid damage that we see in pancreatic cancer cells deprived of cysteine. To explore if known fatty acid

- antioxidants downstream of CoA synthesis could abrogate the effects of system xc - treated samples, we attempted rescue experiments using idebenone, a cell permeable form of coenzyme Q10, a product of the mevalonate pathway which utilizes CoA as a primary carbon

- source. The results indicated the system xc inhibition could be completely rescued by

53 idebenone across all tested concentrations (Figure 4-5B). The rescue was even more pronounced than that achieved by Trolox in previous experiments.

Figure 4-5 Cysteine as incorporated into CoA pools and CoA products can rescue ferroptosis

A. The incorporation of C13 labeled cysteine into intracellular CoA pools as measure by mass spectrometry. Cells were treated with vehicle (.5% DMSO) for indicated times prior to sample collection.

B. Dose response curves of PANC-1 cells treated with IKE (red curve), IKE and 100 µM Trolox (blue curve), and IKE and 10 µM idebenone (light blue curve) for 24 hours.

Lipid synthesis is another possible means to overcome lipid ROS damage. Consistent with the idea that de novo lipid synthesis itself is critical to mediate ferroptotic cell death, we see in PANC-1 cells that palmitate is among the top metabolites depleted in IKE when compared to

BSO treated samples (Figure 4-6). Interestingly, this trend is reversed in MIA PaCa-2 cells, consistent with the evidence that these cells acquire cysteine by other means and use it to create essential cysteine-based metabolites to avoid ferroptosis (Figure 4-7). Taken together,

- our labeling data suggests the cysteine deprivation caused by system xc inhibition causes unique metabolic dysfunctions when compared to the blockade of GSH synthesis by BSO.

Interestingly, some of the most clear metabolite differences exist with regards to lipid production and CoA biosynthesis, consistent with a model that stipulates that both GSH and other cysteine- derived materials are critical for averting ferroptosis in pancreatic cancer cells.

Discussion

Current models of ferroptosis center on the depletion of GSH and its subsequent effect on GPX4 function. GPX4 uses GSH as a cofactor to specifically detoxify lipid ROS, believed to be the harbingers of ferroptotic cell death. Here, we show for the first time that GSH depletion, while necessary for pancreatic cancer cells to undergo ferroptosis, is not sufficient to cause this phenotype in these cells. Clearly, sensitive PANC-1 cells exhibit a sharp decline in intracellular

- GSH stores when treated with a system xc inhibitor, whereas insensitive MIA PaCa-2 cells do not. We initially hypothesized that direct GSH depletion would circumvent any differences in sensitivity since it targeted antioxidant systems much more downstream than cysteine deprivation. To our surprise, however, BSO treatment accomplished its pharmacological goal without causing any ill effects. Interestingly, other research has shown that GSH is actually required for cancer initiation, but once tumors are established, they upregulate other antioxidant production to dispense with their reliance on GSH synthesis79. In this work, the authors highlighted that inhibition of GSH synthesis and the thioredoxin recycling system by treatment

treatedcells (barsin red,palmitate boxed in red). BSO In treatedcells, palmitate isnearly un metaboliteabundances afterdrug treatment.Note palmitatethat isone of the most downregulatedmetabolites in IKE efforts In to discern differencesmajor between GSH depletion cystei and 4 Figure

6 Select 6 differentially modified metaboliteabundances in PANC

nedeprivation, we analyzed differences in

1 cells 1 treated IKEorwith BSO

changed(barin blue).

control. treatedIKE cells (redbars)when compared BSO to (bluebars), in which levels palmitate of are actually lower than in with weHere, present graphsshowing differential metaboliteabundances in MIA PaCa 4 Figure

BSO.Again, we draw attention presenceto the of palmitate (boxedin red) as oneof the most enriched metabolitesin

7 Select 7 differentially modified metaboliteabundances in MIA PaCa

2 cells 2 treated with IKEor BSO

2 cells 2 deprived cysteine of or treated

57 with BSO and auranofin, respectively, could reduce tumor growth in a xenograft model of breast cancer. In a similar vein, our work suggests that BSO alone is not enough to cause deleterious phenotypes in cancer cells.

These findings prompted us to interrogate further what the precise roles of cysteine and

GSH are in these cells. Using a 13C-labeling approach, we tracked the movement of cysteine carbons throughout various metabolic systems to get hints as to what functions of cysteine, in addition to its role in GSH synthesis, may be critical to maintaining low levels of lipid ROS.

Cysteine’s role in the production of bioenergetic TCA intermediates does not seem to be critical in these cells, since we see little to no labeling pyruvate products in either PANC-1 or MIA

PaCa-2 cells. We rule out taurine biosynthesis as the prime cysteine product responsible for these differences due to the lack of labeled taurine production under any of our experimental conditions. We were left to explore either protein production or the critical role of cysteine in

CoA synthesis. Cyclohexamide can be used to inhibit protein synthesis129, but we reasoned that given its cytotoxicity profile, testing this compound in combination with BSO would not yield a potential therapeutic regimen. Furthermore, the median half-life of eukaryotic proteins is 46 hours130, making it unlikely that the effects we see are caused by some deficiency in protein synthesis caused by cysteine deprivation.

Instead, we focused our efforts on CoA, a notable coenzyme critical for the production of fatty acids and fatty acid oxidation. We show for the first time that cysteine is relatively quickly incorporated into CoA pools, showing hints of incorporation by 6 hours and over 50% incorporation by 24 hours, at baseline conditions, in either PANC1- or MIA PaCa-2 cells. CoA is a critical substrate for the mevalonate pathway, which is responsible for the production of isoprenoids, like cholesterol and coenzyme Q. Coenzyme Q is a lipophilic antioxidant and we show that a derivative of this molecule, idebenone, can rescue ferroptosis, consistent with previous reports131. These data suggest that other antioxidants of the cell can surmount the

58 effects of GSH depletion but only in the presence of adequate cysteine, though more research remains to be done to highlight the precise role that the mevalonate pathway and its derivatives may play in protecting cells from ferroptosis.

Finally, we used a non-biased metabolite screening approach to characterize major differences between cells treated with a cysteine depleting agent versus a GSH depleting agent.

Here, we note that a key component of this signature associated with IKE treatment in PANC-1 cells is the depletion of DNA base precursors (cytidine diphosphate, guanosine diphosphate, adenosine, adenosine monophosphate). We suspect this is a byproduct of the unhealthful state of IKE-treated PANC-1 cells, but have not formally tested this. Instead, we highlight that palmitate, the first fatty acid produced in fatty acid synthesis, is actually one of the most differentially downregulated metabolites in PANC-1 cells treated with BSO versus IKE by 6 hours. We hypothesize that this is due to compromised fatty acid synthesis as a result of cysteine depletion, which limits production of CoA , acetyl-CoA and thereby production of palmitate, though this remains to be fully explored in future work (see Chapter 6 for discussion).

Chapter 5 – SLC7A11 expression in human tumors and the effect of its deletion on

established tumors in a genetically engineered mouse model of PDA

Background and Significance

While modeling human cancer biology in cell culture systems is an important tool for studying the mechanistic underpinnings of a given effect in a well-controlled setting, most agree that a thorough preclinical evaluation of the potential value of targeting a specific biological system requires relevant human data and in vivo modeling. Heretofore, our work has focused on carefully characterizing the effects of cysteine deprivation, both directly and pharmacologically, on pancreatic cancer cells in culture. In efforts to determine if cystine transport might be relevant in more clinically translatable settings, we explored the expression of

- SLC7A11, the specific subunit of the system xc transporter, in human tumors, and generate an entirely new, genetically engineered mouse model to study the effects of SLC7A11 deletion on established pancreatic tumors.

While there is some work to suggest that SLC7A11 is overexpressed in human cancers132,133, a thorough analysis of publicly available, large-scale data sets containing comprehensive RNAseq data has not been carried out. In this chapter, we detail our work characterizing mRNA expression levels of SLC7A11 on an in-house RNA-seq data set derived from laser-captured, microdissected human PDA samples. We also show for the first time that

SLC7A11 is overexpressed in pancreatic cancers compared to bulk normal pancreatic tissue in multiple, independent data sets. We also show for the first time that this gene is overexpressed in a multitude of other human cancers, suggesting that some of the findings detailed in this work may have broader applicability. Furthermore, we report that mutant KRAS expression correlates positively with the expression of this transcript in human cancers, corroborating previous studies suggesting that mutant KRAS can directly control expression of SLC7A11. Finally, we detail novel data correlating SLC7A11 with poor prognosis in multiple cancer types. Together, these data imply that SLC7A11 may in fact play a vital role in human cancer biology and support our argument that it may be a valuable new pharmacological target in pancreatic cancer.

While the human data present in the chapter present a compelling correlational case for the importance of SLC7A11 in tumor development, no research has yet been done to show that targeting this transporter in a mouse model of pancreatic cancer can have beneficial effects.

Genetically engineered mouse models of cancer, in which careful, genetic control of endogenous alleles is leveraged to promote tissue-specific tumor formation, provide the most validated model of cancer that we currently have available134,135. In efforts to more fully understand the importance of cystine import in vivo, we undertook a three-year long breeding protocol to produce a genetically engineered mouse model (GEMM) of PDA in which we can also control global deletion of other genes via a dual recombinase system, building on previous modelling strategies136. This is the first report of a dual-recombinase GEMM based on the biology of the KPC model64. In the latter half of this chapter, we report that global deletion of exon three of Slc7a11 (the mouse homolog of SLC7A11) in the context of established pancreatic tumors can slow their growth and extend overall survival. These benefits seem to be conferred due to the increased oxidative burden on these tumors, as suggested by the abrogation of this effect in mice co-treated with NAC. Taken together, these results make a compelling case that cystine import is critical for pancreatic tumor growth in vivo and presents a novel therapeutic avenue in pancreatic cancer.

Results

SLC7A11 is overexpressed in human tumor tissue and overexpression correlates with poorer outcomes

- System xc is comprised of two protein subunits, SLC3A2 and SLC7A11. While SLC3A2

- is a member of many amino acid transporters, SLC7A11 is specific to system xc . To study the potential clinical relevance of our previously reported findings, we examined mRNA expression of SLC7A11 in human cancer. One confounding factor in the analysis of gene expression from

62 bulk PDA samples is the presence of large amounts of stromal or infiltrating–normal cells within the tumor mass. Our group recently performed laser capture microdissection and RNA– sequencing on a large cohort of human pancreatic tumors, deriving a stromal and an epithelial profile for each. To analyze SLC7A11 expression, we first determined the distribution of this gene’s expression in all de-identified samples. This distribution yielded a normalized expression plot centered on 0 and with a standard deviation of 1. We then replotted these data, separating the distributions for epithelial samples versus stromal samples plotted on the same scale. These results revealed that SLC7A11 was overexpressed in malignant epithelial tissue as compared to stromal samples (Figure 5-1A). We extended our analysis of malignant tissue to test if there were any expression differences in SLC7A11 between preneoplastic lesions versus established primary tumors. Here, we found that increasing expression was associated with primary tumors when compared to PanINs (Figure 5-1B), suggesting that SLC7A11 expression increases may be a late occurrence in tumor development. All of these data, however, represent only one cohort of patients from one institution. To make our analyses more thorough, we then examined six published reports of PDA gene expression data that included matched normal tissues.

Notably, SLC7A11 expression was significantly increased over normal in four of the six datasets. Effects sizes varied from 1.33 to 2.38 fold change, while a meta-analysis of all data sets corroborated these results (Figure 5-1C). Taken together, these data show for the first time that SLC7A11 is overexpressed in human PDA compared to normal pancreatic tissue.

Next, we extended our research question to a broader cohort of cancer types to test the widespread applicability of this finding. When we examined SLC7A11 expression in the TCGA data set for a variety of cancers, we found multiple cancer types showed significant upregulation of this transcript (Figure 5-2A), suggesting the importance of this transporter in pancreatic cancer may be extrapolated to many cancer types. We sought to explore the correlation of

SLC7A11 expression level with survival in cancer patients. For a subset of those tumors in the

Figure 5-1 SLC7A11 transcript is overexpressed in human pancreatic cancer

A, B. LCM RNA-Seq from matched human PDA epithelium and stroma shows epithelial expression of SLC7A11. When we examine early lesions (PanINs) and late PDA, we note that expression of SLC7A11 is enriched in fully developed cancer.

C. A meta-analysis showing that SLC7A11 is overexpressed in primary PDA tumors as compared to normal control tissue in of multiple pancreatic cancer data sets (right panel). Left panel shows effect size estimates with their 95% confidence interval per study as well as meta-analytic summaries using random (RE) and fixed effect (FE) models. Q-test and I2 indicate measures of inter-study heterogeneity.

TCGA data set for which there was a significant difference in SLC7A11 expression, we find that high SLC7A11 expression is consistently correlated with a poorer survival outcome (Figure 5-

2B).

Figure 5-2 SLC7A11 is overexpressed in many human cancers and can correlate with poor survival

A. Analysis of SLC7a11 mRNA expression in multiple cancer types from the TCGA data set repeatedly show overexpression of SLC7a11 as compared to normal control tissue. Abbreviations: BRCA, breast cancer; KIRC, kidney renal clear cell; LUAD, lung adenocarcinoma; LIHC, liver hepatocellular carcinoma; PRAD, prostate adenocarcinoma; LUSC, lung squamous cell carcinoma; KICH, kidney chromophobe; KIRP; kidney renal papillary cell carcinoma; HNSC, head and neck squamous cell; CHOL, cholangiocarcinoma; STAD, stomach adenocarcinoma; THCA, thyroid cancer; ESCA, esophageal carcinoma; USEC, uterine corpus endometrial carcinoma; BLCA, urothelial bladder carcinoma

B. In TCGA tumors for which there is a difference in outcome between patients expressing varying levels of SLC7a11, high levels are consistently associated with a worse prognosis (logrank test). See above for abbreviations. 65

SLC7A11 expression correlates with KRAS mutation

We previously stipulated that KRAS mutation may be a driving factor in the high production of ROS in cancer cells and their hypothesized sensitivity to alterations of redox homeostasis. Consistent with this model, we find that KRAS mutational status correlates positively with SLC7A11 expression (Figure 5-3). In summary, we find that SLC7A11 is overexpressed in malignant tissue, commonly elevated in many human tumors, is frequently associated with poor outcomes in multiple independent datasets, and particularly associated with activating KRAS mutations.

Figure 5-3 SLC7A11 overexpression correlates with KRAS mutations

KRAS mutant tumors in the TCGA (n = 424) data set show higher SLC7A11 expression levels than wildtype tumors (n = 2188)

Deletion of Slc7a11 in established murine tumors slows growth and extends overall survival

- The following features make the system xc transporter an attractive drug target. First, genetic deletion or mutation of Slc7a11 in the mouse germline is well tolerated, resulting in generally healthy adult animals; however, when challenged with ROS stress, the animals are prone to ROS related pathologies such as kidney disease and neuronal toxicity at advanced ages137-141. Second, as a cell surface channel, the protein is easily accessible to therapeutic

- agents. Finally, a number of small molecules have been identified that inhibit system xc activity with varying potency90,99,142, illustrating that it is druggable. PE, IKE, and other erastin analogs potently inhibit SLC7A11 in vitro, but lack the appropriate pharmacological properties for effective systemic activity in mammals. Conversely, two FDA-approved drugs, sulfasalazine and

- 99,143 sorafenib, exhibit off-target inhibition of system xc , but at substantially lower potencies .

- In the absence of a clinic-ready system xc inhibitor, we employed a dual recombinase

- genetic engineering strategy to evaluate the potential clinical utility of targeting system xc in established pancreatic tumors. Our strategy was guided by the successful “KPC” mouse model of PDA, which relies on conditional point-mutation of the endogenous KRAS and p53 genes with expression activated in pancreatic progenitors by Pdx1-Cre 64. In this strategy, we cross a novel

Pdx-FlpO driver allele (Figure 5-4A) to mice bearing a homozygosed, floxed Slc7a11 allele in addition to a Rosa26CreERT2/CreERT2 allele, giving us tissue specific control over the recombination of FRT-based recombination and temporal control on CRE-based recombination. These mice are then crossed to KRASFSF-G12D/+; p53R172H/+; Slc7a11fl/fl mice, to produce our KRASFSF-G12D/+; p53R172H/+; PdxFlpOtg/+; Slc7a11fl/fl; Rosa26CreERT2/+ mice, hereafter referred to as KPFSR mice

(Figure 5-4B). For a more thorough description of this breeding strategy, please refer to the materials and methods for this chapter.

Figure 5-4 PdxFlpO allele validation and KPSFR breeding scheme

A. The PdxFlpo mouse was a gift of Scott Seeley. These mice were validated by crossing the strain with a FRT-STOP-FRT alkaline phosphatase reporter in which reporter expression was silenced until recombination via FRT would occur. Images show founder lines exhibiting prominent alkaline phosphatase activity in the pancreas (frozen sections).

B. Breeding strategy describing the production of KPFSR mice.

The resulting KPFSR mice were born at expected Mendelian ratios and appeared grossly

healthy throughout development. To accelerate the development of PDA, KPFSR mice were

injected with cerulean at ~8 weeks of age utilizing a dosing regimen known to cause chronic

pancreatitis 144. These animals all developed PDA within several months, as detected by high

resolution ultrasound and later confirmed by histopathology. Tumors from KPFSR mice

appeared histopathologically identical to KPC mice, typically developing a single, focal tumor

with moderate differentiation and evidence of a robust stromal desmoplasia (Figure 5-5A). The

surrounding pancreatic tissues showed widespread evidence of pancreatitis, acinar-to-ductal

metaplasia, and varying grades of PanINs (Figure 5-5B).

Figure 5-5 Histological comparison of PDA and PDA development in KPC and KPFSR mice

A. PDA in KPC and KPFSR mice bost present with a robust desmoplastic stroma and moderately differentiated, glandular structures. Two representative mice shown.

B. Both KPC and KPFSR mice proceed to PDA formation through PanIN progression.

- To directly assess the impact of blocking system xc function in pancreatic tumors, we

intended to use tamoxifen to induce Cre-mediated deletion of Slc7a11 in mice with established

tumors. To determine an optimal dose, first we carried out a pilot study to assess recombination

efficiency in tumor tissues. After six days of daily tamoxifen administration at 200 mg/kg, near-

complete recombination was observed in the tumors assessed (Figure 5-6A), whereas there

was incomplete recombination in neighboring tissues (Figure 5-6B).

Figure 5-6 Assessment of Slc7a11 recombination in tamoxifen treated mice

A. Analysis of DNA recombination in tumors of mice treated with 200 mg/kg of tamoxifen for 6 consecutive days. Samples were collected at 14 days post study initiation. Vehicle (corn oil) treated tumors exhibited essentially no recombination, producing one, clear amplicon at ~1300 base pairs (bp). Tumors treated with tamoxifen exhibited nearly complete recombination, producing one, clear ~500 bp amplicon.

B. Vehicle mice showed no off target recombination in neighboring tissues. Tamoxifen treated mice exhibited some degree of recombination in these tissues, producing two amplicons, one at ~1300 bp and another at ~500 bp. DNA from liver, spleen, and kidney was analyzed for this experiment.

Using this dosing regimen, we then carried out an intervention survival study. Following

cerulean induction, KPFSR mice were imaged periodically using high resolution 3D ultrasound

70 to identify nascent tumors. Upon detection of a 4–7mm diameter tumor, mice were randomized to either tamoxifen or vehicle (corn oil) treatment arms, and then tumor growth was monitored twice weekly by 3D ultrasound. Post-hoc analysis shows that the distribution of tumor volume at enrollment was equivalent among treatment groups (Figure 5-7A). Following tamoxifen treatment, tumors typically grew more slowly or exhibited a period of stable disease; a subset showed transient regressions and one tumor completely regressed to the point of being undectable by ultrasound (Figure 5-7B, C). This mouse, however, eventually succumbed to a new, independent tumor. No comparable responses were observed in vehicle-treated animals.

Tumor growth rates in mice treated with tamoxifen were significantly lower than those in the control arm (Figure 5-7D)145. Overall, tamoxifen-treated mice had a 2-fold increase in overall survival. Importantly, mice fed NAC in drinking water that had also been dosed with tamoxifen survived times comparable with our control arm, suggesting that oxidative damage is critical for

- mediating the benefits of system xc deletion, in vivo (Figure 5-7E). We also examined the degree of recombination in tumors at endpoint after they had escaped the effects of Slc7a11 deletion, using PCR. Notably, recombination was largely intact in the majority of tumors, highlighting the abilities to tumors to eventually overcome this genetic perturbation (Figure 5-8).

To more fully assess the pathological effects of Slc7a11 deletion on tumors, we performed a standard histological examination of these samples. Our results showed no clear differences in cleaved caspase-3 staining (a marker of apoptosis), Ki67 staining ( a cell proliferation marker), or PHH3 ( a marker of cells in mitosis) staining in endpoint samples (Figure 5-9A). However, histological analysis showed unique morphological features in tamoxifen-treated tumors when compared to controls (Figure 5-9B). These features were characterized by pockets of lipid inclusions surround intact nuclei, lending the region a “swiss cheese” appearance. Lipid oxidation damage was sometimes associated with these features as evidenced by malondialdehyde staining, a marker of lipid peroxidation (Figure 5-9B). Taken

Figure 5-7 Slc7a11 deletion in established tumors slows growth and extends survival

A. Analysis of tumor volumes at day of enrollment across groups of survival study. No significant changes were identified. Error = +/- SD.

B. Tumor volumes are plotted as a function of days on study. N = 11, vehicle. N = 9, tamoxifen. N = 4, tamoxifen/NAC. Error = +/- SD. * = p < .05 by unpaired t test. Vehicle is shown in gray, tamoxifen treated animals in red, and tamoxifen/NAC shown in blue.

C. Tumor volumes for mouse in which Slc7a11 caused complete regression on initial tumor. A second tumor presented at day 18, however, and grew with normal kinetics.

D. Tumor growth rates compared between treatment arms. Error = +/-SD. * = p<.05

E. Kaplan-Meier survival curve showing significant extension of survival in tamoxifen-treated mice. * = p =.0295. N=11, vehicle; N=9, tamoxifen; N = 5, tamoxifen + NAC.

Figure 5-8 PCR analysis of recombination in endpoint tumors

Agarose gel showing that in vehicle tumors, no recombination was readily apparent. Conversely, full recombination was observed in all endpoint tumors.

Figure 5-9 Histological analysis of proliferation markers and morphological effects in tamoxifen-treated tumors

A. Cleaved-caspase 3 (CC3), Ki67, and phosphohistone H3 (PHH3) staining qauntifiction.

B. Histological comparison of endpoint tumors on survival study. Black arrows indicate areas of cellular damage. Red errors indicate areas of positive malondialdehyde staining.

73 together, these data suggest that therapeutic targeting of SLC7A11 may prove clinical beneficial in pancreatic cancer.

Discussion

Multiple lines of research have begun to show the value of targeting unique metabolic shifts in cancer 146. In redox biology, cysteine has long been appreciated for its critical role in the production of GSH and in providing the basis for the disulfide chemistry that supports secondary and tertiary protein structure, but the precise relevance of these functions and the ability to target them for clinical benefit in cancer patients is still the subject of ongoing research. Here we

- show for the first time that the membrane amino acid transporter system xc can be specifically targeted to leverage a previously unidentified metabolic dependency in pancreatic cancer.

Using human expression data we confirm that SLC7A11 is overexpressed in human pancreatic cancer when compared to normal pancreatic tissue. Interestingly, when comparing

PanINs to PDA samples, we still see an upregulation of this gene, implying that SLC7A11 overexpression may be a relatively late event in pancreatic cancer development. This is consistent with the current model of the relationship between oxidative stress and cancer onset:

DNA damage caused by ROS is mutagenic and pro-oncogenic but cancer cells must eventually upregulate antioxidant function late in tumor development to promote survival in the face of mitochondrial oxidative stress56. Our results imply that SLC7A11 may play an important role in the progression of preneoplastic lesions to frank carcinoma in pancreatic cancer. As such,

SLC7A11 activity may be an interesting biomarker for future study, a topic we will discuss in more detail in chapter 6.

While the primary focus of this work has been on pancreatic cancer, we also use this chapter as an opportunity to explore the broad applicability of these results to other human cancers, using expression data as a surrogate for functional relevance in human cancer. We

74 show using a broad number of data sets that SLC7A11 is overexpressed in many human cancers, and as such may hold critical function in general cancer maintenance. We also highlight that in some cancers, SLC7A11 expression is correlated positively with poor prognosis.

It may be possible that high expression of SLC7A11 can confer a survival advantage to tumor cells, especially those exposed to cytotoxic therapies. Another interpretation of these data is that SLC7A11 may promote the formation of more aggressive tumors in some cancer types, though it should be noted that these correlations do not apply to every tumor type. Finally, we show that KRAS mutational status positively correlates with SLC7A11 expression. This is consistent with previous reports that KRAS mutation can activate a cellular antioxidant program mediated by Nrf2 transcription of antioxidant genes59.

While there is some work to suggest that targeting redox dependencies in pancreatic cancer can retard tumor growth54,93, we provide the first evidence that targeting cystine transport directly in a GEMM can cause slowdown or even regression in the growth of established pancreatic tumors and extend overall survival in mice. In fact, we report the first complete regression of a pancreatic tumor in a model of this type. It is important to note that in our model, we cause global deletion of Slc7a11, making it impossible to known if these effects cell autonomous or non-autonomous. However, we confirm that treatment of mice with an antioxidant (NAC) can “rescue” the phenotype in these tumors, restoring their initial growth rates, suggesting that these effects are primarily oxidative in nature. While the survival benefit conferred by this genetic therapy is statistically significant, it is interesting to note that the survival rates in all mice that die before 8 days on study are essentially equivalent. This may highlight one of two points: first, that Slc7a11 gene function is not rendered completely ineffective until up to 8 days following enrolment on study; or second, that only a subset of mice responds to this gene therapy. It is also critical not to lose sight of the fact that eventually all mice do succumb to their tumors, despite the loss of Slc7a11 function. This underscores the

75 stubborn ability of pancreatic tumors to overcome stress, but may also offer an opportunity to study modes of resistance to cystine import blockade in efforts to make this treatment approach more efficacious. In closing, we offer the first report of SLC7A11 expression in a variety of human cancers and the creation of a novel mouse model of pancreatic cancer in which we delete Slc7a11 to assess its functional relevance in PDA. We show that SLC7A11 is overexpressed in multiple cancer data sets when compared to normal tissue and that deletion of the murine homolog of this gene in established pancreatic tumors confers survival benefit.

Chapter 6 – Conclusions and Future Directions

Cysteine import as a unique metabolic dependency in pancreatic cancer

Many lines of current inquiry in pancreatic cancer have focused on illuminating the effects of KRAS mutation on vital metabolic systems in the cancer cell. While many breakthroughs have been made with regards to glucose metabolism, amino acid metabolism, and nutrient import and scavenging40,46,48,49,54; redox biology has recently become an emerging theme in this field59,93. Cysteine, given its critical antioxidant functions, presented an intriguing metabolite to explore in the context of pancreatic cancer, a prototypical RAS-driven malignancy.

As such, we endeavored to elucidate what role this amino acid may play in maintaining the redox state of pancreatic cancers and if targeting its import would yield any potentially useful therapeutic effects.

A primary concern when testing new therapeutic approaches is that targeting vital cellular systems will limit their clinical utility given the chances that it may cause undue toxicity in normal tissues. Such is the classical problem associated with chemotherapeutic cancer interventions, which are associated with myriad side effects and whose administration must often be de-escalated or even halted in light of declining patient health. Like pancreatic cancer cells, it is possible that normal cells also require cysteine for survival. However, perhaps somewhat surprisingly, germline knockout of Slc7a11 produces viable litters with no obvious

147 - phenotypes . This finding is actually consistent with the expression pattern of system xc , which has limited expression in human tissues. SLC7A11 mRNA is most prominently expressed in the central nervous system, spinal cord, and immune organs (thymus, spleen, lymphatic tissue), with some expression in the pancreas148, suggesting it may play a rather limited role in normal tissue function. In keeping with these phenotypic motifs, the sut mouse, which bears a

- large deletion on chromosome 3 that compromises system xc function, exhibits only two

78 reported phenotypes: a deficiency in pigment formation yielding its characteristic gray fur, and some brain atrophy140, neither of which appear to affect overall lifespan or neurological function.

As such, we suggest that cysteine may represent a selective and potent dependency in pancreatic cancer. Most importantly, this dependency can be targeted pharmacologically by way of its primary transporter, setting the stage for translational work to confirm the validity of these results in human patients.

To formally test differences in sensitivity between normal and malignant tissue, we propose using a recently pioneered organoid model which allows for the facile three- dimensional culture of normal pancreas, early pancreatic cancer lesions, and frank carcinoma149. In these experiments, organoids derived from both normal and neoplastic human tissue would be cultured and could be tested in manners analogous to methods laid out herein.

As an example, early experiments would involve culturing these cells in media containing diminishing concentrations of cysteine and then subjected to standard viability assays. A similar

- series of dose escalation experiments testing the efficacy of a system xc inhibitor in these systems would follow. Treating these cells concomitantly with various rescue agents could help interrogate what sorts of cell death effects, in any, were occurring in these various organoids.

Furthermore, it is possible to use KPFSR mice to derive novel organoids over which genetic control of Slc7a11 could be exerted. In this system, a viral Cre could be utilized to cause deletion of Slc7a11 with temporal control and could help us untangle if the effects of Slc7a11 deletion that we have discussed in our model are in fact cell autonomous or non-autonomous.

Furthermore, this system could be refined to allow us to control KRAS mutational activation to test if RAS mutations themselves bear any importance on conferring sensitivity to cysteine deprivation, as is implied by some our work.

Determining the role in cysteine import in pancreatic cancer development

- In this study, we chose to focus on the role that system xc plays in tumor maintenance, since patients present clinically with late-stage cancer and that approach is the most stringent for probing for translatable results. However, we left unexplored what role, if any, that this system plays in the early evolution of PDA. ROS and the role they play as signaling molecules is not to be understated: certainly their toxicity makes them a palatable bogeymen for many cancer researchers and, indeed, in the popular scientific consciousness, but a large body of record confirms that they are important signaling molecules with complicated regulatory functions150.

It may be possible that SLC7A11 plays an important role in limiting ROS accumulation in developing preneoplastic lesions and that in the absence of this control, initiating lesions fail to progress. Preliminary results from our study in mice that were euthanized at earlier time points during our survival study show evidence of a very pronounced cell death effect in nests of acinar-ductal metaplasia (ADM) in diseased pancreata (Figure 6-1). Here, we see the tissue is lent the same “swiss cheese” appearance seen in tamoxifen treated tumors. Interestingly, whereas ADM in KPC often show evidence of apoptosis as measured by cleaved-caspase 3 staining, the ADM we observe in KPFSR mice do not. These early results suggest that

SLC7A11 function may be critical for PanIN progression, but further work would be required to validate this claim. We would propose an early intervention study, in which mice that do not have signs of overt disease are treated with tamoxifen using the dosing regimen we have already established in this work. These mice could then be monitored for tumor formation as compared to control animals, and the time to tumor onset could be recorded. Mice would be sacrificed at established time points following study initiation, in efforts to capture effects of

Slc7a11 deletion on PanIN formation and tumor initiation. Once again, our model also allows us to also test the genetic contribution of KRAS or p53 mutations to this process. Using caerulein to induce ADM in the normal pancreas of mice with difference genetic background (KRAS

80 mutant, p53 mutant, or double mutant) and then deleting Slc7a11 in these mice using tamoxifen

Normal KPC-ADM KPFSR-ADM HE HE

CC3 CC3

Figure 6-1 Effects of Slc7a11 deletion on acinar-ductal metaplasia (ADM) in KPC and KPFSR mice

Histological examination of ADM lesions in KPC and KPFSR mice. Here, KPFSR mice have been treated with tamoxifen to induced recombination of Slc7a11. Arrows indicate CC3 positive cells.

could help us determine if these effects are merely associated with any pro-inflammatory event or if they require a specific mutational burden. Using these approaches, we could determine if cystine import may play a critical role in tumor development and if it may be leveraged in the context of a preventative therapy.

- Overcoming resistance to system xc deletion

- Our work highlights the first reported case of targeting system xc to yield appreciable survival benefit in a GEMM, but we are forced to reckon with the fact that, in spite of these efforts, all of these experimental mice eventually succumb to their disease. Certainly, this is the not the first case of a highly intractable cancer overcoming therapeutic intervention. Indeed, there is a large body of scientific literature attempting to solve the problem of overcoming drug resistance, both acquired and intrinsic, in many cancers151-156. Our results do raise a pertinent question in this regard: how exactly do pancreatic tumors in mice overcome the deletion of

Slc7a11?

We provide evidence that these tumors in most cases either fail to grow or even decrease in size when faced with this obstacle, but still they manage to surmount this difficulty and eventually grow to appreciable sizes. We have a hint as to how this may be happening from some of our in vitro work: in cell culture conditions, we show one method that pancreatic cancer

- cells use to overcome the effects of system xc blockade is by upregulating macropinocytosis.

To test if this is true, the tumors of KPSFR mice treated with tamoxifen or vehicle would be directly injected with a high molecular weight fluorescent dextran (analgous to our in vitro experiment) to help us quantify macropinosome formation. Should these results confirm that macropinocytotic activity was increased in tamoxifen treated tumors, we would undertake an intervention study featuring co-treatment of tamoxifen and the macropinocytosis inhibitor, EIPA, utilizing an administration protocol as previously describe48. Monitoring tumor growth by ultrasound, we would be able to determine response rates and discern the effect of EIPA co- treatment on the health of these tumors.

While macropinocytosis would provide one possible answer to the question of how

- tumors overcome system xc dysfunction, other possibilities cannot be ruled out. While we did

- not find that transsulfuration mediated sensitivity to system xc inhibition, it still offers a compelling explanation for how cells can overcome cysteine deprivation. A more exploratory

82 approach in the context of our GEMM may yield more inclusive answer regarding how tumors overcome this metabolic stress. We propose using an RNA-seq analysis to unveil potential transcriptional changes that cooperate to promote tumor survival in the absence of cystine import.

Many current treatment modalities in pancreatic cancer focus on using chemotherapeutic combinations to generate the keenest cytotoxic effects in tumors. In our study, we limit our research to exploring one therapeutic intervention, but would suggest that combination of this genetic therapy with other agents may prove even more efficacious. FDA- approved drugs which are known to cause oxidative stress would be an attractive class of agents to test in this model, but certainly other standard-of-care chemotherapeutics would also be appropriate for testing. As an example, we propose testing cisplatin, a platinum-based chemotherapy that has recently been reported to cause ROS accumulation in normal tissues157-

159, in combination with Slc7a11 deletion to determine efficacy. Furthermore, focused radiation treatment can also cause ROS accumulation and may synergize impressively with Slc7a11 deletion160. It may even be compelling to turn to a long debated anti-cancer therapy, vitamin C, in this context to exacerbate these phenotypes, since work from the Cantley laboratory has shown that vitamin C is effective in KRAS and BRAF mutant colorectal cancer by exacerbating oxidative stress161.

Targeting GSH synthesis and CoA function to cause ferroptosis

Our work refines the current model of the mechanism of ferroptosis by suggesting that, while GSH depletion is critical for mediating these effects, it alone is not sufficient to cause them in this context. These results are consistent with early phase I trials of BSO in combination with melphalan, which demonstrated excellent pharmacodynamic effects and little associated toxicity, but only produced measurable responses of 2 of 28 patients162. The combination has

83 moved into a phase II trial that was completed in 2016, but not results have yet been reported for this study. We contend that BSO alone is unlikely to cause responses as a monotherapy, but are encouraged by the possibility that, given its favorable toxicity profile, it can be paired with other agents to create robust effects. Our work suggests that targeting other cysteine-regulated pathways may enhance this effect, but this remains to be borne out experimentally. We hypothesize that CoA synthesis and its numerable functions are the secondary mediator of the effects of cysteine deprivation. As such, targeting CoA synthesis in combination with BSO treatment may prove beneficial in pancreatic cancer. One function of CoA is to serve as a precursor for the production of mevalonate, a key molecule required for the synthesis of cholesterols and other isoprenoids (e.g, coenzymeQ10). In this work, we show that the effects

- of system xc inhibition can be fully rescued treatment with idebanone, a cell-permeable form of coenzyme Q10. With these data in mind, we propose dosing tumor bearing mice with BSO in combination with statin class drugs to inhibit the mevalonate pathway. This is a particularly appealing combination because both compounds are FDA approved and have limited side effects. Interestingly, statin use has now been correlated with a 34% decrease in pancreatic cancer risk in males, suggesting that this approach may even have a preventative benefit163.

In vivo mechanisms of ferroptosis and its relevance in cancer

The primary goals of breeding the KPFSR mouse were two-fold: first, to test the effect of

Slc7a11 deficiency in intact tumors; second, to create a conditional model system in which we

- can determine if system xc inhibition in vivo can cause ferroptosis. To date, precious little work has been done to show if ferroptosis occurs in living tissues. One recent publication highlights that GPX4 knockout induced by tamoxifen treatment causes acute renal failure and kills mice164.

This work shows profound images of the damage incurred in renal tubules and the liver after deletion of GPX4, which the authors claim is in vivo ferroptosis. Here, the authors perform elegant work showing that these effects can be ameliorated by liproxstatin-1, which they identify

84 as an inhibitor of ferroptosis by way of large-library compound screen to identify chemicals that rescue Gpx4 knockout MEFs from ill effects. Using this compound, they rescue the effects of

GPX4 deletion on livers subjected to transient ischaemia/reperfusion. These may be some of the first reports of in vivo ferroptosis in native animal tissues, albeit not in a therapeutic setting.

Building on this type of research, our in vivo work shows that the survival benefit conferred by Slc7a11 deletion cancer be abrogated by co-treatment with NAC, a known rescuer

- of ferroptotic cell death in the context of system xc malfunction. These findings are also consistent with data presented in our in vitro work. However, it may still be too early to claim that we are seeing ferroptosis in tumors. Certainly, the lesions we see in endpoint tumors are unlike those seen in apoptosis and seem analogous to the morphological effects of ferroptosis on cells in culture (Figure 6-2). To date, characterizing ferroptosis has been accomplished by rescuing its effects using both a lipophilic antioxidant (e.g. Fer-1) and by iron chelation (e.g. DFO). To explore this question, we propose attempting to “rescue” to survival benefit in our model with liproxstatin-1, ferostatin-1, and DFO. Adding these arms to the study will help elucidate the precise mechanism by which Slc7a11 deletion exerts its effects. It will also be possible to plan to short-term study in which animals are sacrificed at early signs of tumor response to generate samples critical for determining the molecular mediators of the effects we see herein. In vivo metabolomics using a mass spectrometric readout can be used to determine metabolite levels in tumor samples, comparing responding tumors and vehicle tumors to generate data implicating particular molecules in the tumor growth inhibition we see in our experiments. It would also be fruitful to perform a full lipidomics screen on these same samples to determine what role, if any, changes in lipid biosynthesis or fatty acid oxidation state play in executing the effects in these tumors. In conclusion, we believe that the KPSFR model can play a vital role in untangling the precise mechanisms of ferroptosis in living tissues and can be easily modified to test other hypothesis regarding cystine import and other vital metabolic functions. The mice

85 generated in this study will hopefully go on to serve as an invaluable resource, not only in our laboratory, but also in the greater cancer biological field, to help test new therapeutic regimens and to answer some of the burning questions in cancer metabolism research.

Figure 6-2 High magnification images of the effects of Slc7a11 deletion on pancreatic tumors

100x images of normal pancreatic tumor (left panel) and a characteristic lesion in tumors in mice that were treated with tamoxifen to delete Slc7a11 (right panel). We note the presence of cells with intact nuclei, ballooned membranes, and lipid inclusions (marked with black arrows). Scale bars are 20μm.

MATERIALS AND METHODS

Chapter 2

Cell lines, cell culture, and viability assays

All cell lines were obtained from ATCC and tested negatively for mycoplasma infection. Cells were maintained under standard conditions at 37ºC and 5% CO2. Cells were grown in DMEM

(Life Technologies, 12430-054) supplemented with penicillin and streptomycin (Corning, 30-

003-CI), 10% FBS (Life Technologies, 10438-034), and MEM NEAA (Life Technologies, 11140-

050), unless otherwise indicated. For cysteine depletion experiments, cells were cultured in

DMEM lacking met/cys (Life Technologies, 21013-024), supplemented with 200 mM methionine

(Sigma, M9625), 4mM glutamine (Life Technologies, 25030-081), 10% FBS, penicillin and streptomycin and varying concentrations of cystine (Sigma, C8755), depending on the given experiment. For cell viability experiments cells were plated in a 96-well plate format. PANC-1 cells were plated at 4000 cells per well; MIA PaCA-2, BxPC3, and AsPC1 cells were plated at

8000 cells per well. Cells were allowed to seed overnight, and subsequently treated with compounds at indicated concentrations and for indicated lengths of time. The following is a list of all chemical compounds used in cell culture experiments: ZVAD-FMK (SelleckChem, S7023),

Bafilomycin A1 (Sigma, B1793), Necrostatin-1s (BioVision, 2263-1), Ferrostatin-1 (Sigma,

SML0583), N-acetyl-L-cysteine (Sigma, A9615), buthionine sulfoxamine (Sigma, B2515), deferoxamine (Sigma, D9533), EIPA (Sigma, A3085), PE and IKE (provided by the Stockwell laboratory), staurosporine (Sigma, S4400), N-acetyl-cysteine (Sigma, A9165), beta- mercaptoethanol (Sigma, M6250). Compounds were formulated according to the manufacturer’s instructions. All viability assays were carried out using the AlamarBlue reagent (Thermofisher,

88952) according to the manufacturer’s instructions. In brief, 10 μL of AlamarBlue reagent was added each well containing 100 μL of experimental media. Plates were gently agitated for 1

87 minute to promote adequate mixing. Once wells had reached uniform color, plates were incubated at standard culture conditions indicated above for 1-4 hours. Plates were subsequently assessed for fluorescent readout on a Promega plate reader using the appropriate filter.

Light and fluorescent microscopy

Still transmitted light images were captured on an Olympus CKX41. Time lapse images and fluorescent microscopy images were captured on a Nikon A1RMP. One image was captured every two minutes for up to 24 hours post treatment initiation. Fluorescent time lapse microscopic images were captured on an IncuCyte ZOOM system. One image was captured in every channel (transmitted light, red, and green) once every 15 minutes between 2 and 12 hours post treatment initiation. Cells were plated in a 6-well plate format, (PANC-1: 250,000 cells per well, MIA PaCa2: 400,000 cells per well) 16 hours prior to drug treatment and subsequent staining. Cells were stained with C-11 Bodipy (Invitrogen, C10445) at a concentration of 2uM for 30 minutes prior to time lapse imaging. Cells were washed with PBS three times and incubated in Live-Cell Imaging Solution (Life Technologies, A14291DJ) during the duration of live-cell imaging. To detect lipid ROS at static time points by fluorescence microscopy, cells were plated in 1u-Slide 8 well ibiTreat dishes (ibidi, 80827). PANC-1 cells were plated at 30,000 cells per chamber, MIA PaCa2 cells were plated at 60,000 cells per chamber. Cells were seeded overnight and subsequently subjected to indicated treatments for indicated times. Upon experiment completion, cells were visualized on a Nikon A1RMP using a

FITC filter.

Flow cytometric detection of ROS

PANC-1 and MIA PaCa-2 cells were plated in a 12 well format at 100,000 cells and 200,000 cells per well, respectively. Cells were allowed to seed overnight and were subjected to drug

88 treatments for indicated times. Cells were then incubated for 30 minutes in live-cell imaging solution containing the pertinent ROS dyes at the following concentrations: C-11 Bodipy, 2 µM;

H2DCFDA, 10 µM (Invitrogen, C6827). Cells were then washed with PBS, trypsinized with .25% trypsin (Life Technologies, 25200-056), and neutralized with FBS. Cells were strained through a

40uM strainer (BD Falcon), and analyzed on a MACSQuant Analyzer 10. A minimum of 8000 cells were analyzed per condition. Software analysis was carried out using FlowJo V10.

Chapter 3

13C labeling and metabolomics analysis (NOTE: all metabolomics materials and methods are detailed here and will be referred to in future chapters as necessary)

PANC-1 and MIA PaCa-2 cells were plated 16 hours prior to drug treatment at 1.5 million millions cells and 3 million cells per 10 cm dish, respectively. Cells were cultured in DMEM containing 10% dialyzed FBS (ThermoScientific, 26400036), but lacking either cysteine or methionine (see Chapter 2). All experiments were set up in biological triplicate. 13C carbon in metabolites was assessed by growing PANC-1 and MIA PaCa-2 in the media described above, which was then supplemented with 13C methionine or 12C cystine, in presence or absence of inhibitors.

At the end of indicated time points, the medium was aspirated saving 1 ml for polar metabolite extraction. Cells were lysed with dry-ice cold 80% methanol and extracts were incubated in -

80ºC for 10 min. Samples were then centrifuged at 4000 rpm for 10 min at 4 °C and the supernatant was stored at -80 °C till further analyses. Metabolite extraction utilizing medium samples was done by adding 1 ml of medium to 4 ml 100% methanol, followed by incubation at

-80 °C for 30 minutes and centrifugation at 14,000 rpm for 10 min at 4 °C. Protein concentrations were determined by processing a parallel 10 cm dish and used to normalize

89 metabolite fractions. Based on protein concentrations, aliquots of the supernatants were transferred to a fresh micro centrifuge tube and lyophilized using a speed vac. Dried metabolite pellets from cells or media were re-suspended in 35 μl 50:50 MeOH: H2O mixture for LC–MS analysis.

For steady state, 5 μl of samples was injected onto a Agilent 1290 UHPLC as above and separated using a Waters Acquity UPLC BEH amide column (2.1 x 100mm, 1.7µm) column. For negative ion acquisition, the mobile phase (A) consisted of 20mM ammonium acetate, pH 9.6 in water, and mobile phase (B) was ACN. Gradient program: mobile phase (B) was held at 85% for 1 min, increased to 65% in 12 min, then to 40% in 15 min and held for 5 min before going to initial condition and held for 10 min. The column was kept at 40 ̊C and 3 µl of sample was injected into the LC-MS with a flow rate of 0.2 ml/min. Calibration of TOF MS was achieved through Agilent ESI- Low Concentration Tuning Mix.

Optimization was performed on the 6490 QqQ in negative or positive mode individually for each of 220 standard compounds to get the best fragment ion and other MS parameters for each standard. Retention time for each standard of the 220 standards was measured from pure standard solution or a mix standard solution. The LC-MS/MS method was created with dynamic dMRMs with RTs, RT windows and MRMs of all the 220 standard compounds.

In both acquisition modes, key parameters of AJS ESI were: Gas temp 275 ̊C, Gas Flow 14 l/min, Nebulizer at 20 psi, Sheath Gas Heater 250 ̊C, Sheath Gas Flow 11 L/min, Capillary 3000

V. For negative mode MS: Delta EMV was 350 V, Cycle Time 500 ms and Cell accelerator voltage was 4 V, whereas for positive acquisition mode MS: Delta EMV was set at 200 V with no change in cycle time and cell accelerator voltage.

The QqQ data pre-processed with Agilent MassHunter Workstation Software Quantitative

Analysis were post-processed for further quality control in the programming language R. We

90 calculated coefficient of variation (CV) across replicate samples for each metabolite given a cut- off value of peak areas in both the positive and the negative modes. We then compared distributions of CVs for the whole dataset for a set of peak area cut-off values of 0, 1000, 5000,

10000, 15000, 20000, 25000 and 30000 in each mode. A noise cut-off value of peak areas in each mode was chosen by manual inspection of the CV distributions. Each sample is then normalized by the total intensity of all metabolites to reflect the same protein content as a normalization factor. We then retained only those metabolites with at least 2 replicate measurements. The remaining missing value in each condition for each metabolite was filled with the mean value of the other replicate measurements. Finally, each metabolite abundance level in each sample was divided by the median of all abundance levels across all samples for proper comparisons, statistical analyses, and visualizations among metabolites. The statistical significance test was done by a two-tailed t-test with a significance threshold level of 0.05. The p-values were not adjusted in favor of more flexible biological interpretation.

For the 13C cystine/methionine label incorporation studies, a Agilent 1260 UHPLC combined with a 6520 Accurate-Mass Q-TOF LC/MS was utilized. Agilent MassHunter Workstation

Software LC/MS Data Acquisition for 6200 series TOF/6500 series Q-QTOF (B.06.01) was used for calibration and data acquisition. A Waters Acquity UPLC BEH amide column (2.1 x 100mm,

1.7µm) column was used with mobile phase (A) consisting of 20 mM NH4OAc in water pH 9.6, and mobile phase (B) consisting of ACN. Gradient program: mobile phase (B) was held at 85% for 1 min, increased to 65% in 12 min, then to 40% in 15 min and held for 5 min before going to initial condition and held for 10 min. The column was at 40 ̊C and 3 µl of sample was injected into the LC-MS with a flow rate of 0.2 ml/min. Calibration of TOF MS was achieved through

Agilent ESI-Low Concentration Tuning Mix.

Key parameters for both acquisition modes were: mass range 100-1200 da, Gas temp 350 C ̊,

Fragmentor 150 V, Skimmer 65 v, Drying Gas 10 l/min, Nebulizer at 20 psi and Vcap 3500 V,

Ref Nebulizer at 20 psi. For negative mode the reference ions were at 119.0363 and 980.01637 m/z whereas for positive acquisition mode, reference ions at 121.050873 and 959.9657 m/z

For 13C-labeling data analysis, we used Agilent MassHunter Workstation Software Profinder

B.08.00 with Batch Targeted Feature Extraction and Batch Isotopologue Extraction and

Qualitative Analysis B.07.00. Various parameter combinations, e.g. mass and RT tolerance, were used to find best peaks and signals by manual inspection. Key parameters were: mass tolerance = 20 or 10 ppm and RT tolerance = 1 or 0.5 min. Isotopologue ion thresholds, the anchor ion height threshold was set to 250 counts and the threshold of the sum of ion heights to

500 counts. Coelution correlation threshold was set to 0.3.

All other bioinformatics analyses including graphs and plots were done using R/Bioconductor.

Macropinosome Visualization and Quantification

Glass coverslips were treated with poly-L-lysine (PLL) (Sigma, P5899), according to manufacturer’s instructions. Cells were seeded onto glass coverslips. 24 hours after seeding, cells were treated with PE, EIPA (Sigma, A3085), vehicle, or PE+EIPA co-treatment at the indicated doses.After preliminary experiments, 50 μM was determined to be the effective dose for EIPA that both inhibits macropinosome formation and limits off-target cytotoxic effects. This dose was used for all experiments thereafter. Macropinosomes were visualized by the uptake of high molecular weight (70kDa) TMR-dextran (ThermoFisher, D1818), which was added to the medium prior to the conclusion of the drug treatment time-point, at a final concentration of 500

μg/mL for 30 minutes at 37℃. After the 30-minute incubation the cells were gently rinsed three times with cold PBS then immediately fixed in formalin (4% formaldehyde in PBS for 30 minutes at room temperature. Cells were DAPI-treated to stain the nuclei, and the coverslips were mounted onto slides using Vectashield Mounting Medium (Vector Laboratories, H-1000). Slides were left to dry overnight in dark room at 4℃. Images were captured using an Axiovert 200

92 inverted fluorescent microscope (Zeiss, 100x oil immersion lens, 10 randomly selected fields per sample) and analyzed as previously described 48,165. Macropinosomes were defined as those particles with a circularity index between 0.5 and 1.0, and with an area of 0.20-20.0 microns2, in keeping with literature classification of macropinosomes. The number of macropinosomes per cell was divided by the number of intact cells present, as visualized via DAPI and brightfield staining. We termed this the “macropinocytotic index” (average number of macropinosomes per cell). All 10 fields were analyzed for each sample.

Cell Culture Viability Assays

Cell culture assays were carried out as described in the Chapter 2 section. New compounds used in this section: propargyglycine (Sigma, 81838), EIPA (Sigma, A3085), buthionine sulfoxamine (Sigma, B2515).

Chapter 4

13C labeling and metabolomics analysis

See Chapter 3

Cell Culture Viability Assays

Cell culture assays were carried out as described in Chapter 2. New compounds used in this section: GSH-ethyl ester (Sigma, G1404), idebenone (Sigma, I5659).

Chapter 5

Bioinformatic analyses

Microarray data. Gene expression data from studies comparing pancreatic ductal adenocarcinoma specimen with normal pancreas parenchyma were downloaded from the Gene

Expression Omnibus (GEO) using the GEOquery R166 package 167 and the following accession numbers: GSE32676 168, GSE19650 169, GSE16515 170, GSE62452 171 and GSE71729 172.

U133A-CEL files from 173,174 were obtained from ArrayExpress under accession number E-

MEXP-950 and normalized using GCRMA 175. The following samples were excluded because of outlier behavior during exploratory data analysis: NPD15, T55, TPK9, NPK13. For all studies, probes were collapsed at the gene level using their mean normalized expression.

TCGA data. RNA-Seq V2 expression, clinical and mutation data were downloaded for all available TCGA tumor types using the RTCGAToolbox R package 176. The run date was set to

“2016-01-28”. A detailed script on further pre-processing steps is provided separately.

Differential Gene Expression analysis

Genome-wide differential gene expression analysis normal tissue of origin and tumor tissue were calculated using the limma 177 R package using (gc)rma-normalized expression data for microarray studies and log2-transformed normalized count data for the TCGA cohorts. If normal and tumor samples were matched, i.e. from the same patient, a paired design was specified in the design formula.

Effect size meta-analysis.

In order to evaluate evidence on the differential expression of SLC7A11 from 6 studies where global expression in normal pancreatic tissue was compared to PDA, the effect size (i.e. log2 fold change) and its standard error were extracted from the respective genome-wide differential expression analysis, which was carried out using the limma R package. Meta-analysis was carried out using the metafor 178 R package. Both random and fixed effect models were fit using the rma function (method = “REML” and method = “FE”, respectively).

Survival analysis

The association of SLC7A11 expression status with disease outcome was evaluated using a log-rank test as implemented in the survdiff function from the survival 179 R package. Patients were grouped into tertiles according to normalized SLC7A11 expression and differences in outcome were assessed between the upper and lower tertile.

Animal Breeding and genotyping

All studies were carried out in accordance with the relevant institutional guidelines of Columbia

University.

Generation of Pdx1-FlpO mice

The proximal 6kb promoter of the Pdx1-Cre transgene (Gu et al., 2002; Hingorani et al., 2003) was fused to the start codon of mammalian codon-optimized, thermostabile Flp recombinase

(FlpO)(Raymond and Soriano, 2007), with subsequent fusion of the 5’ end of the FlpO open reading frame to the hGH polyadenylation signal sequence using the In-Fusion cloning system

(Clontech) according to the manufacturer’s instructions. The Pdx1-FlpO cassette was excised from a large-scale plasmid preparation, gel-purified, and microinjected into the pronuclei of fertilized FVB oocytes at the Gladstone Transgenic Mouse Core. Following implantation, birth, and weaning, transgene-bearing founder mice were identified via PCR and maintained on an

FVB background.

Pdx1-FlpO founders were crossed to homozygous FVB alkaline phosphatase Flp reporter mice

(gift, Dr. Susan Dymecki, Harvard University), Pdx1-FlpO-harboring progeny sacrificed at 2 months of age, and skin, brain, liver, pancreas, kidney, lung, stomach, duodenum, small intestine, colon collected and flash frozen in liquid nitrogen. Frozen sections of each were cut, alkaline phosphatase histochemistry performed (Vector Laboratories), and sections examined microscopically. Founder lines exhibiting prominent alkaline phosphatase activity in the

95 pancreas were backcrossed an additional five generations and this process was repeated to identify founder lines exhibiting stable transgene expression.

Pdx1-FlpO mice were crossed with Kras+/LSL-G12D;p53FRT/FRT, Pdx1-FplO; Kras+/LSL-G12D;p53FRT/+ progeny aged for 3 months or until they had reached humane endopoint, and pancreas, stomach, duodenum, small intestine, colon, skin, lung, and brain examined grossly and microscopically for the presence of neoplasia.

Generation of SLC7A11 conditional knockout mice

Initially, Slc7a11 conditional potential mice were imported from the IMPC repository (EMMA ID:

10001, strain designation Slc7a11tm1a(EUCOMM)Wtsi). These mice were crossed to homozygous

ACTB: FLPe mice (Jackson Laboratory, Stock no. 003800) to create SLC7A11 conditional mice in which loxP sites surround exon three of the gene. Using a combination of the Pdx1-Flpo,

KRASFSFG12D (Jackson Laboratory, Stock no. 008653), p53R172H (Jackson Laboratory, Stock no.

008652), Rosa26CreERT2 (Jackson Laboratory, Stock no. 008463), and our newly made

Slc7a11fl/fl mice, we were able to generate a new strain of genetically engineered mouse model akin to the KrasLSL-G12D/+;p53LSL-R172H;Pdx1-Cre (KPC) mouse 64, but in which we have temporal control of the deletion of Slc7a11 using tamoxifen. We term these mice KRASFSF-G12D/+; p53R172H; Pdx1-Flpotg/+; Slc7a11fl/fl; Rosa26CreERT2/+, KPFSR mice.

Genotyping

Genotyping was carried out using protocols provided by Jackson Laboratory for all strains obtained from this vendor.

Slc7a11fl/fl genotyping was carried out as follows:

Forward primer: 5’-tgggttggtctctggtgatc-3’

Reverse primer: 5’-cctgtgaagatccgcctact-3’

Cycling conditions:

1. 94ºC 3 minutes

2. 94ºC 1 minute

3. 60ºC 2 minutes

4. 72ºC 1 minute

Cycle to step 2, 29 times.

5. 72ºC 5 minutes

6. 4ºC forever

Expected amplicons:

Slc7a11 mice: 386 bp

WT mice: no amplicon

Slc7a11 recombinatorial genotyping was custom designed as follows:

Forward primer: 5’-tgg gtt ggt ctc tgg tga tc-3’

Reverse primer:5’-ctt aac ccc agc acc att cg-3’

Cycling conditions:

1. 95ºC 2 min

2. 95ºC 1 min

3. 56ºC 30 seconds

4. 72ºC 1 min

Cycle to step 2, 34 times

5. 72ºC 5 min

6. 4ºC forever

Expected amplicons:

Slc mouse unrecombined approx. 1300 bp

Slc mouse recombined approx. 500 bp

WT mouse approx 1240.

Survival study

KPSFR mice were treated at approximately 42 days of age with caerulein (Sigma, C9026) at

250 ug/kg by intraperitoneal injection for 5 consecutive days to induce chronic pancreatitis and accelerate tumor formation, as per previous reports 180-182. Tumor formation was monitored initially by once weekly palpation and upon positive palpation, but twice-weekly ultrasound.

Once the average diameter of tumors of the pancreas was 4-7 mm in size, mice were randomly enrolled into the survival study. Mice were treated with either corn oil (Sigma, C8267) or tamoxifen (Sigma, T5648) at 200 mg/kg by oral gavage for 6 days. Mice were euthanized once they reached endpoint criteria in concordance with IACUC regulations. Samples were either fixed in formalin overnight at 4ºC, fixed in PFA at 4ºC overnight followed by sucrose-mediated water displacement for 24 hours at 4ºC, frozen in OCT, or flash frozen in liquid nitrogen.

Pathology and immunohistochemistry

Samples that had been fixed in formalin were washed in 70% ethanol and subjected to standard dehydration processing to prepare them for mounting in paraffin wax blocks. Paraffin blocks were sectioned on a Leica RM 2235 at 5 μM thickness. The sections were mounted on charged slides and heated to 60ºC to melt wax and ensure tissue adherence to the slide. Slides were then subjected to standard rehydration and antigen retrieval was carried out for five minutes in boiling 10 mM sodium citrate buffer pH 6, .05% Tween-20 using a pressure cooker. Slides were allowed to cool to room temperature in an ice-cold water bath. Slides were then incubated in 3% hydrogen peroxide for 20 minutes at room temperature to block endogenous peroxidases.

Slides were blocked in 1.5% horse serum and 2% animal free blocker (Vector Laboratories, SP-

5030) for 1 hour at room temperature. Malondialdehyde-stained slides however were blocked in

5% BSA. Slides were then stained with the appropriate antibody: Cleaved caspase-3 (Cell

Signaling, 9664S), 1:100 dilution; Ki67 (Cell Signaling, 12202S), 1:100 dilution; phosphohistone-

H3 (Cell Signaling, 9716S), 1:100 dilution; malondialdehyde (Abcam, ab6463), 1:500. Primary antibody incubation was carried out overnight at 4ºC. Slides were then washed 3x with PBS-T and incubated with the appropriate secondary antibody for 30 minutes at room temperature.

Staining was developed using the DAB reagent (Vector Labs, VV-93951085).

Histological staining on paraffin sections (hematoxylin and eosin, Masson’s Trichrome, and

Movatt’s Pentachrome) were carried out using standard protocols. Following digital capture on

Olympus BX51, histology images were processed using Affinity Photo software, using three filters/adjustments applied evenly across the entire image: unsharp mask, white balance, and levels adjustment.

REFERENCES

1 Bay, N. S. & Bay, B. H. Greek anatomist herophilus: the father of anatomy. Anatomy & cell biology 43, 280-283, doi:10.5115/acb.2010.43.4.280 (2010). 2 Potter, P. Herophilus of Chalcedon: an assessment of his place in the history of anatomy. Bulletin of the history of medicine 50, 45-60 (1976). 3 von Staden, H. The discovery of the body: human dissection and its cultural contexts in ancient Greece. The Yale journal of biology and medicine 65, 223-241 (1992). 4 Tsuchiya, R., Kuroki, T. & Eguchi, S. The pancreas from Aristotle to Galen. Pancreatology : official journal of the International Association of Pancreatology 15, 2-7, doi:10.1016/j.pan.2014.10.011 (2015). 5 Busnardo, A. C., DiDio, L. J., Tidrick, R. T. & Thomford, N. R. History of the pancreas. American journal of surgery 146, 539-550 (1983). 6 Are, C., Dhir, M. & Ravipati, L. History of pancreaticoduodenectomy: early misconceptions, initial milestones and the pioneers. HPB : the official journal of the International Hepato Pancreato Biliary Association 13, 377-384, doi:10.1111/j.1477-2574.2011.00305.x (2011). 7 Whipple, A. O., Parsons, W. B. & Mullins, C. R. Treatment of Carcinoma of the Ampulla of Vater. Annals of surgery 102, 763-779 (1935). 8 Moertel, C. G., Childs, D. S., Jr., Reitemeier, R. J., Colby, M. Y., Jr. & Holbrook, M. A. Combined 5- fluorouracil and supervoltage radiation therapy of locally unresectable gastrointestinal cancer. Lancet 2, 865-867 (1969). 9 Moertel, C. G. et al. Therapy of locally unresectable pancreatic carcinoma: a randomized comparison of high dose (6000 rads) radiation alone, moderate dose radiation (4000 rads + 5- fluorouracil), and high dose radiation + 5-fluorouracil: The Gastrointestinal Tumor Study Group. Cancer 48, 1705-1710 (1981). 10 Burris, H. A., 3rd et al. Improvements in survival and clinical benefit with gemcitabine as first- line therapy for patients with advanced pancreas cancer: a randomized trial. Journal of clinical oncology : official journal of the American Society of Clinical Oncology 15, 2403-2413, doi:10.1200/JCO.1997.15.6.2403 (1997). 11 Conroy, T. et al. FOLFIRINOX versus gemcitabine for metastatic pancreatic cancer. The New England journal of medicine 364, 1817-1825, doi:10.1056/NEJMoa1011923 (2011). 12 Von Hoff, D. D. et al. Increased survival in pancreatic cancer with nab-paclitaxel plus gemcitabine. The New England journal of medicine 369, 1691-1703, doi:10.1056/NEJMoa1304369 (2013). 13 Muniraj, T., Jamidar, P. A. & Aslanian, H. R. Pancreatic cancer: a comprehensive review and update. Disease-a-month : DM 59, 368-402, doi:10.1016/j.disamonth.2013.08.001 (2013). 14 Siegel, R. L., Miller, K. D. & Jemal, A. Cancer statistics, 2016. CA: a cancer journal for clinicians 66, 7-30, doi:10.3322/caac.21332 (2016). 15 Gong, Z., Holly, E. A. & Bracci, P. M. Survival in population-based pancreatic cancer patients: San Francisco Bay area, 1995-1999. American journal of epidemiology 174, 1373-1381, doi:10.1093/aje/kwr267 (2011). 16 Klinkenbijl, J. H. et al. Adjuvant radiotherapy and 5-fluorouracil after curative resection of cancer of the pancreas and periampullary region: phase III trial of the EORTC gastrointestinal tract cancer cooperative group. Annals of surgery 230, 776-782; discussion 782-774 (1999). 17 Oettle, H. et al. Adjuvant chemotherapy with gemcitabine vs observation in patients undergoing curative-intent resection of pancreatic cancer: a randomized controlled trial. Jama 297, 267-277, doi:10.1001/jama.297.3.267 (2007).

100

18 Herman, J. M. et al. Analysis of fluorouracil-based adjuvant chemotherapy and radiation after pancreaticoduodenectomy for ductal adenocarcinoma of the pancreas: results of a large, prospectively collected database at the Johns Hopkins Hospital. Journal of clinical oncology : official journal of the American Society of Clinical Oncology 26, 3503-3510, doi:10.1200/JCO.2007.15.8469 (2008). 19 Neoptolemos, J. P. et al. Adjuvant 5-fluorouracil and folinic acid vs observation for pancreatic cancer: composite data from the ESPAC-1 and -3(v1) trials. British journal of cancer 100, 246- 250, doi:10.1038/sj.bjc.6604838 (2009). 20 Hidalgo, M. Pancreatic cancer. The New England journal of medicine 362, 1605-1617, doi:10.1056/NEJMra0901557 (2010). 21 Ezzati, M., Henley, S. J., Lopez, A. D. & Thun, M. J. Role of smoking in global and regional cancer epidemiology: current patterns and data needs. International journal of cancer 116, 963-971, doi:10.1002/ijc.21100 (2005). 22 Willett, W. C. Diet and cancer. The oncologist 5, 393-404 (2000). 23 Ekbom, A. et al. Pancreatitis and pancreatic cancer: a population-based study. Journal of the National Cancer Institute 86, 625-627 (1994). 24 Warshaw, A. L. & Fernandez-del Castillo, C. Pancreatic carcinoma. The New England journal of medicine 326, 455-465, doi:10.1056/NEJM199202133260706 (1992). 25 Li, D., Xie, K., Wolff, R. & Abbruzzese, J. L. Pancreatic cancer. Lancet 363, 1049-1057, doi:10.1016/S0140-6736(04)15841-8 (2004). 26 Hruban, R. H. & Fukushima, N. Pancreatic adenocarcinoma: update on the surgical pathology of carcinomas of ductal origin and PanINs. Modern pathology : an official journal of the United States and Canadian Academy of Pathology, Inc 20 Suppl 1, S61-70, doi:10.1038/modpathol.3800685 (2007). 27 Makino, I. et al. Nerve plexus invasion in pancreatic cancer: spread patterns on histopathologic and embryological analyses. Pancreas 37, 358-365 (2008). 28 Yachida, S. & Iacobuzio-Donahue, C. A. The pathology and genetics of metastatic pancreatic cancer. Archives of pathology & laboratory medicine 133, 413-422, doi:10.1043/1543-2165- 133.3.413 (2009). 29 Iacobuzio-Donahue, C. A. & Hruban, R. H. Gene expression in neoplasms of the pancreas: applications to diagnostic pathology. Advances in anatomic pathology 10, 125-134 (2003). 30 Hingorani, S. R. et al. Preinvasive and invasive ductal pancreatic cancer and its early detection in the mouse. Cancer cell 4, 437-450 (2003). 31 Vincent, A., Herman, J., Schulick, R., Hruban, R. H. & Goggins, M. Pancreatic cancer. Lancet 378, 607-620, doi:10.1016/S0140-6736(10)62307-0 (2011). 32 Olive, K. P. & Tuveson, D. A. The use of targeted mouse models for preclinical testing of novel cancer therapeutics. Clinical cancer research : an official journal of the American Association for Cancer Research 12, 5277-5287, doi:10.1158/1078-0432.CCR-06-0436 (2006). 33 Andea, A., Sarkar, F. & Adsay, V. N. Clinicopathological correlates of pancreatic intraepithelial neoplasia: a comparative analysis of 82 cases with and 152 cases without pancreatic ductal adenocarcinoma. Modern pathology : an official journal of the United States and Canadian Academy of Pathology, Inc 16, 996-1006, doi:10.1097/01.MP.0000087422.24733.62 (2003). 34 Makohon-Moore, A. & Iacobuzio-Donahue, C. A. Pancreatic cancer biology and genetics from an evolutionary perspective. Nature reviews. Cancer 16, 553-565, doi:10.1038/nrc.2016.66 (2016). 35 Klimstra, D. S. & Longnecker, D. S. K-ras mutations in pancreatic ductal proliferative lesions. The American journal of pathology 145, 1547-1550 (1994). 36 Rozenblum, E. et al. Tumor-suppressive pathways in pancreatic carcinoma. Cancer research 57, 1731-1734 (1997).

101

37 Smit, V. T. et al. KRAS codon 12 mutations occur very frequently in pancreatic adenocarcinomas. Nucleic acids research 16, 7773-7782 (1988). 38 Trahey, M. & McCormick, F. A cytoplasmic protein stimulates normal N-ras p21 GTPase, but does not affect oncogenic mutants. Science 238, 542-545 (1987). 39 Collins, M. A. et al. Oncogenic Kras is required for both the initiation and maintenance of pancreatic cancer in mice. The Journal of clinical investigation 122, 639-653, doi:10.1172/JCI59227 (2012). 40 Ying, H. et al. Oncogenic Kras maintains pancreatic tumors through regulation of anabolic glucose metabolism. Cell 149, 656-670, doi:10.1016/j.cell.2012.01.058 (2012). 41 Walker, K. & Olson, M. F. Targeting Ras and Rho GTPases as opportunities for cancer therapeutics. Current opinion in genetics & development 15, 62-68, doi:10.1016/j.gde.2004.11.001 (2005). 42 Welsch, M. E. et al. Multivalent Small-Molecule Pan-RAS Inhibitors. Cell 168, 878-889 e829, doi:10.1016/j.cell.2017.02.006 (2017). 43 Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646-674, doi:10.1016/j.cell.2011.02.013 (2011). 44 Keibler, M. A. et al. Metabolic requirements for cancer cell proliferation. Cancer & metabolism 4, 16, doi:10.1186/s40170-016-0156-6 (2016). 45 Kamphorst, J. J. et al. Human pancreatic cancer tumors are nutrient poor and tumor cells actively scavenge extracellular protein. Cancer research 75, 544-553, doi:10.1158/0008- 5472.CAN-14-2211 (2015). 46 Yang, A. & Kimmelman, A. C. Inhibition of autophagy attenuates pancreatic cancer growth independent of TP53/TRP53 status. Autophagy 10, 1683-1684, doi:10.4161/auto.29961 (2014). 47 Yang, S. & Kimmelman, A. C. A critical role for autophagy in pancreatic cancer. Autophagy 7, 912-913 (2011). 48 Commisso, C. et al. Macropinocytosis of protein is an amino acid supply route in Ras- transformed cells. Nature 497, 633-637, doi:10.1038/nature12138 (2013). 49 Mayers, J. R. et al. Elevation of circulating branched-chain amino acids is an early event in human pancreatic adenocarcinoma development. Nature medicine 20, 1193-1198, doi:10.1038/nm.3686 (2014). 50 Wise, D. R. & Thompson, C. B. Glutamine addiction: a new therapeutic target in cancer. Trends in biochemical sciences 35, 427-433, doi:10.1016/j.tibs.2010.05.003 (2010). 51 Eagle, H. Nutrition needs of mammalian cells in tissue culture. Science 122, 501-514 (1955). 52 DeBerardinis, R. J. et al. Beyond aerobic glycolysis: transformed cells can engage in glutamine metabolism that exceeds the requirement for protein and nucleotide synthesis. Proceedings of the National Academy of Sciences of the United States of America 104, 19345-19350, doi:10.1073/pnas.0709747104 (2007). 53 DeBerardinis, R. J. & Cheng, T. Q's next: the diverse functions of glutamine in metabolism, cell biology and cancer. Oncogene 29, 313-324, doi:10.1038/onc.2009.358 (2010). 54 Son, J. et al. Glutamine supports pancreatic cancer growth through a KRAS-regulated metabolic pathway. Nature 496, 101-105, doi:10.1038/nature12040 (2013). 55 Liou, G. Y. et al. Mutant KRas-Induced Mitochondrial Oxidative Stress in Acinar Cells Upregulates EGFR Signaling to Drive Formation of Pancreatic Precancerous Lesions. Cell reports 14, 2325- 2336, doi:10.1016/j.celrep.2016.02.029 (2016). 56 Liou, G. Y. & Storz, P. Reactive oxygen species in cancer. Free radical research 44, 479-496, doi:10.3109/10715761003667554 (2010). 57 Siegel, D., Shieh, B., Yan, C., Kepa, J. K. & Ross, D. Role for NAD(P)H:quinone oxidoreductase 1 and manganese-dependent superoxide dismutase in 17-(allylamino)-17-

102

demethoxygeldanamycin-induced heat shock protein 90 inhibition in pancreatic cancer cells. The Journal of pharmacology and experimental therapeutics 336, 874-880, doi:10.1124/jpet.110.176438 (2011). 58 Lewis, A. et al. Metastatic progression of pancreatic cancer: changes in antioxidant enzymes and cell growth. Clinical & experimental metastasis 22, 523-532, doi:10.1007/s10585-005-4919-7 (2005). 59 DeNicola, G. M. et al. Oncogene-induced Nrf2 transcription promotes ROS detoxification and tumorigenesis. Nature 475, 106-109, doi:10.1038/nature10189 (2011). 60 Cox, A. D., Fesik, S. W., Kimmelman, A. C., Luo, J. & Der, C. J. Drugging the undruggable RAS: Mission possible? Nature reviews. Drug discovery 13, 828-851, doi:10.1038/nrd4389 (2014). 61 Corbett, T. H. et al. Induction and chemotherapeutic response of two transplantable ductal adenocarcinomas of the pancreas in C57BL/6 mice. Cancer research 44, 717-726 (1984). 62 Longnecker, D. S. Lesions induced in rodent pancreas by azaserine and other pancreatic carcinogens. Environmental health perspectives 56, 245-251 (1984). 63 Jackson, E. L. et al. Analysis of lung tumor initiation and progression using conditional expression of oncogenic K-ras. Genes & development 15, 3243-3248, doi:10.1101/gad.943001 (2001). 64 Hingorani, S. R. et al. Trp53R172H and KrasG12D cooperate to promote chromosomal instability and widely metastatic pancreatic ductal adenocarcinoma in mice. Cancer cell 7, 469-483 (2005). 65 Wollaston, W. H. On Cystic Oxide, a New Species of Urinary Calculus. Philosophical Transactions of the Royal Society of London 100, 223-230 (1810). 66 Baumann, E. Ueber Cystin und Cystein. Zeit. Physiol. Chem. 8, 299-305 (1884). 67 Klomsiri, C., Karplus, P. A. & Poole, L. B. Cysteine-Based Redox Switches in Enzymes. Antioxidants & redox signaling 14, 1065-1077, doi:10.1089/ars.2010.3376 (2011). 68 Christen.Hn, Liang, M. & Archer, E. G. A Distinct Na+-Requiring Transport System for Alanine Serine Cysteine and Similar Amino Acids. Journal of Biological Chemistry 242, 5237-& (1967). 69 Alsaleh, E. A. & Wheeler, K. P. Transport of Neutral Amino-Acids by Human-Erythrocytes. Biochimica et biophysica acta 684, 157-171, doi:Doi 10.1016/0005-2736(82)90001-3 (1982). 70 Guidotti, G. G., Gazzola, G. C. & Borghetti, A. F. Regulation of Amino-Acid Transport in Animal- Cells. Biochimica et biophysica acta 515, 329-366, doi:Doi 10.1016/0304-4157(78)90009-6 (1978). 71 Eavenson, E. & Christen.Hn. Transport Systems for Neutral Amino Acids in Pigeon Erythrocyte. Journal of Biological Chemistry 242, 5386-& (1967). 72 Young, J. D., Jones, S. E. M. & Ellory, J. C. Amino-Acid-Transport in Human and in Sheep Erythrocytes. Proc R Soc Ser B-Bio 209, 355-375, doi:DOI 10.1098/rspb.1980.0100 (1980). 73 Bannai, S. & Tateishi, N. Role of Membrane-Transport in Metabolism and Function of Glutathione in Mammals. J Membrane Biol 89, 1-8, doi:Doi 10.1007/Bf01870891 (1986). 74 Sato, H., Tamba, M., Ishii, T. & Bannai, S. Cloning and expression of a plasma membrane cystine/glutamate exchange transporter composed of two distinct proteins. Journal of Biological Chemistry 274, 11455-11458, doi:DOI 10.1074/jbc.274.17.11455 (1999). 75 Verrey, F., Meier, C., Rossier, G. & Kuhn, L. C. Glycoprotein-associated amino acid exchangers: broadening the range of transport specificity. Pflug Arch Eur J Phy 440, 503-512, doi:DOI 10.1007/s004240000274 (2000). 76 Waddell, N. et al. Whole genomes redefine the mutational landscape of pancreatic cancer. Nature 518, 495-501, doi:10.1038/nature14169 (2015). 77 Mullarky, E. et al. Identification of a small molecule inhibitor of 3-phosphoglycerate dehydrogenase to target serine biosynthesis in cancers. Proceedings of the National Academy of Sciences of the United States of America 113, 1778-1783, doi:10.1073/pnas.1521548113 (2016). 78 Gross, S. et al. Cancer-associated metabolite 2-hydroxyglutarate accumulates in acute

103

myelogenous leukemia with isocitrate dehydrogenase 1 and 2 mutations. The Journal of experimental medicine 207, 339-344, doi:10.1084/jem.20092506 (2010). 79 Harris, I. S. et al. Glutathione and thioredoxin antioxidant pathways synergize to drive cancer initiation and progression. Cancer cell 27, 211-222, doi:10.1016/j.ccell.2014.11.019 (2015). 80 Kimmelman, A. C. Metabolic Dependencies in RAS-Driven Cancers. Clin Cancer Res 21, 1828- 1834, doi:10.1158/1078-0432.CCR-14-2425 (2015). 81 Paulsen, C. E. & Carroll, K. S. Cysteine-mediated redox signaling: chemistry, biology, and tools for discovery. Chemical reviews 113, 4633-4679, doi:10.1021/cr300163e (2013). 82 Lu, S. C. Regulation of glutathione synthesis. Molecular aspects of medicine 30, 42-59, doi:10.1016/j.mam.2008.05.005 (2009). 83 Meister, A. & Anderson, M. E. Glutathione. Annual review of biochemistry 52, 711-760, doi:10.1146/annurev.bi.52.070183.003431 (1983). 84 Richman, P. G. & Meister, A. Regulation of gamma-glutamyl-cysteine synthetase by nonallosteric feedback inhibition by glutathione. The Journal of biological chemistry 250, 1422-1426 (1975). 85 Bannai, S. & Tateishi, N. Role of membrane transport in metabolism and function of glutathione in mammals. The Journal of membrane biology 89, 1-8 (1986). 86 Lu, S. C., Ge, J. L., Kuhlenkamp, J. & Kaplowitz, N. Insulin and glucocorticoid dependence of hepatic gamma-glutamylcysteine synthetase and glutathione synthesis in the rat. Studies in cultured hepatocytes and in vivo. The Journal of clinical investigation 90, 524-532, doi:10.1172/JCI115890 (1992). 87 Bannai, S. Exchange of cystine and glutamate across plasma membrane of human fibroblasts. The Journal of biological chemistry 261, 2256-2263 (1986). 88 Yang, W. S. et al. Regulation of ferroptotic cancer cell death by GPX4. Cell 156, 317-331, doi:10.1016/j.cell.2013.12.010 (2014). 89 Ishii, T., Bannai, S. & Sugita, Y. Mechanism of growth stimulation of L1210 cells by 2- mercaptoethanol in vitro. Role of the mixed disulfide of 2-mercaptoethanol and cysteine. The Journal of biological chemistry 256, 12387-12392 (1981). 90 Dixon, S. J. et al. Ferroptosis: an iron-dependent form of nonapoptotic cell death. Cell 149, 1060- 1072, doi:10.1016/j.cell.2012.03.042 (2012). 91 Griffith, O. W. & Meister, A. Potent and specific inhibition of glutathione synthesis by buthionine sulfoximine (S-n-butyl homocysteine sulfoximine). The Journal of biological chemistry 254, 7558- 7560 (1979). 92 Weiss, R. H. & Estabrook, R. W. The mechanism of cumene hydroperoxide-dependent lipid peroxidation: the function of cytochrome P-450. Archives of biochemistry and biophysics 251, 348-360 (1986). 93 Chio, II et al. NRF2 Promotes Tumor Maintenance by Modulating mRNA Translation in Pancreatic Cancer. Cell 166, 963-976, doi:10.1016/j.cell.2016.06.056 (2016). 94 Lee, J. I. et al. HepG2/C3A cells respond to cysteine deprivation by induction of the amino acid deprivation/integrated stress response pathway. Physiological genomics 33, 218-229, doi:10.1152/physiolgenomics.00263.2007 (2008). 95 Caro-Maldonado, A. et al. Glucose deprivation induces an atypical form of apoptosis mediated by caspase-8 in Bax-, Bak-deficient cells. Cell death and differentiation 17, 1335-1344, doi:10.1038/cdd.2010.21 (2010). 96 Akakura, N. et al. Constitutive expression of hypoxia-inducible factor-1alpha renders pancreatic cancer cells resistant to apoptosis induced by hypoxia and nutrient deprivation. Cancer research 61, 6548-6554 (2001). 97 Braun, F., Bertin-Ciftci, J., Gallouet, A. S., Millour, J. & Juin, P. Serum-nutrient starvation induces cell death mediated by Bax and Puma that is counteracted by p21 and unmasked by Bcl-x(L)

104

inhibition. PloS one 6, e23577, doi:10.1371/journal.pone.0023577 (2011). 98 Tang, X. et al. Cystine Deprivation Triggers Programmed Necrosis in VHL-Deficient Renal Cell Carcinomas. Cancer research 76, 1892-1903, doi:10.1158/0008-5472.CAN-15-2328 (2016). 99 Dixon, S. J. et al. Pharmacological inhibition of cystine-glutamate exchange induces endoplasmic reticulum stress and ferroptosis. eLife 3, e02523, doi:10.7554/eLife.02523 (2014). 100 Yang, W. S. & Stockwell, B. R. Ferroptosis: Death by Lipid Peroxidation. Trends in cell biology 26, 165-176, doi:10.1016/j.tcb.2015.10.014 (2016). 101 Matsushita, M. et al. T cell lipid peroxidation induces ferroptosis and prevents immunity to infection. The Journal of experimental medicine 212, 555-568, doi:10.1084/jem.20140857 (2015). 102 Yang, W. S. et al. Peroxidation of polyunsaturated fatty acids by lipoxygenases drives ferroptosis. Proceedings of the National Academy of Sciences of the United States of America 113, E4966- 4975, doi:10.1073/pnas.1603244113 (2016). 103 Doll, S. et al. ACSL4 dictates ferroptosis sensitivity by shaping cellular lipid composition. Nature chemical biology 13, 91-98, doi:10.1038/nchembio.2239 (2017). 104 Lu, S. C., Martinez-Chantar, M. L. & Mato, J. M. Methionine adenosyltransferase and S- adenosylmethionine in alcoholic liver disease. Journal of gastroenterology and hepatology 21 Suppl 3, S61-64, doi:10.1111/j.1440-1746.2006.04575.x (2006). 105 Beatty, P. W. & Reed, D. J. Involvement of the cystathionine pathway in the biosynthesis of glutathione by isolated rat hepatocytes. Archives of biochemistry and biophysics 204, 80-87 (1980). 106 Banerjee, R., Evande, R., Kabil, O., Ojha, S. & Taoka, S. Reaction mechanism and regulation of cystathionine beta-synthase. Biochimica et biophysica acta 1647, 30-35 (2003). 107 Yamanishi, T. & Tuboi, S. The mechanism of the L-cystine cleavage reaction catalyzed by rat liver gamma-cystathionase. Journal of biochemistry 89, 1913-1921 (1981). 108 Jhee, K. H. & Kruger, W. D. The role of cystathionine beta-synthase in homocysteine metabolism. Antioxidants & redox signaling 7, 813-822, doi:10.1089/ars.2005.7.813 (2005). 109 Hayano, M., Yang, W. S., Corn, C. K., Pagano, N. C. & Stockwell, B. R. Loss of cysteinyl-tRNA synthetase (CARS) induces the transsulfuration pathway and inhibits ferroptosis induced by cystine deprivation. Cell death and differentiation 23, 270-278, doi:10.1038/cdd.2015.93 (2016). 110 Washtien, W. & Abeles, R. H. Mechanism of inactivation of gamma-cystathionase by the acetylenic substrate analogue propargylglycine. Biochemistry 16, 2485-2491 (1977). 111 Commisso, C., Flinn, R. J. & Bar-Sagi, D. Determining the macropinocytic index of cells through a quantitative image-based assay. Nature protocols 9, 182-192, doi:10.1038/nprot.2014.004 (2014). 112 Koivusalo, M. et al. Amiloride inhibits macropinocytosis by lowering submembranous pH and preventing Rac1 and Cdc42 signaling. The Journal of cell biology 188, 547-563, doi:10.1083/jcb.200908086 (2010). 113 Bar-Sagi, D. & Feramisco, J. R. Induction of membrane ruffling and fluid-phase pinocytosis in quiescent fibroblasts by ras proteins. Science 233, 1061-1068 (1986). 114 Cao, J. Y. & Dixon, S. J. Mechanisms of ferroptosis. Cellular and molecular life sciences : CMLS 73, 2195-2209, doi:10.1007/s00018-016-2194-1 (2016). 115 Ursini, F. et al. Diversity of glutathione peroxidases. Methods in enzymology 252, 38-53 (1995). 116 Arthur, J. R. The glutathione peroxidases. Cellular and molecular life sciences : CMLS 57, 1825- 1835 (2000). 117 Ball, C. R., Connors, T. A., Double, J. A., Ujhazy, V. & Whisson, M. E. Comparison of nitrogen- mustard sensitive and -resistant Yoshida sarcomas. International journal of cancer 1, 319-327 (1966).

105

118 Suzukake, K., Petro, B. J. & Vistica, D. T. Reduction in glutathione content of L-PAM resistant L1210 Cells confers drug sensitivity. Biochemical pharmacology 31, 121-124 (1982). 119 Bailey, H. H. L-S,R-buthionine sulfoximine: historical development and clinical issues. Chemico- biological interactions 111-112, 239-254 (1998). 120 Bailey, H. H. et al. Phase I study of continuous-infusion L-S,R-buthionine sulfoximine with intravenous melphalan. Journal of the National Cancer Institute 89, 1789-1796 (1997). 121 Wellner, V. P., Anderson, M. E., Puri, R. N., Jensen, G. L. & Meister, A. Radioprotection by glutathione ester: transport of glutathione ester into human lymphoid cells and fibroblasts. Proceedings of the National Academy of Sciences of the United States of America 81, 4732-4735 (1984). 122 Meister, A. Glutathione metabolism and its selective modification. The Journal of biological chemistry 263, 17205-17208 (1988). 123 Carelli, S. et al. Cysteine and glutathione secretion in response to protein disulfide bond formation in the ER. Science 277, 1681-1684 (1997). 124 Locigno, R. & Castronovo, V. Reduced glutathione system: role in cancer development, prevention and treatment (review). International journal of oncology 19, 221-236 (2001). 125 Noctor, G. & Foyer, C. H. ASCORBATE AND GLUTATHIONE: Keeping Active Oxygen Under Control. Annual review of plant physiology and plant molecular biology 49, 249-279, doi:10.1146/annurev.arplant.49.1.249 (1998). 126 Townsend, D. M., Tew, K. D. & Tapiero, H. The importance of glutathione in human disease. Biomedicine & pharmacotherapy = Biomedecine & pharmacotherapie 57, 145-155 (2003). 127 Griffith, O. W. Mechanism of action, metabolism, and toxicity of buthionine sulfoximine and its higher homologs, potent inhibitors of glutathione synthesis. The Journal of biological chemistry 257, 13704-13712 (1982). 128 Demetriades, C., Doumpas, N. & Teleman, A. A. Regulation of TORC1 in response to amino acid starvation via lysosomal recruitment of TSC2. Cell 156, 786-799, doi:10.1016/j.cell.2014.01.024 (2014). 129 Obrig, T. G., Culp, W. J., McKeehan, W. L. & Hardesty, B. The mechanism by which cycloheximide and related glutarimide antibiotics inhibit peptide synthesis on reticulocyte ribosomes. The Journal of biological chemistry 246, 174-181 (1971). 130 Schwanhausser, B. et al. Global quantification of mammalian gene expression control. Nature 473, 337-342, doi:10.1038/nature10098 (2011). 131 Shimada, K. et al. Global survey of cell death mechanisms reveals metabolic regulation of ferroptosis. Nature chemical biology 12, 497-503, doi:10.1038/nchembio.2079 (2016). 132 Lo, M., Ling, V., Wang, Y. Z. & Gout, P. W. The xc- cystine/glutamate antiporter: a mediator of pancreatic cancer growth with a role in drug resistance. British journal of cancer 99, 464-472, doi:10.1038/sj.bjc.6604485 (2008). 133 Jiang, L. et al. Ferroptosis as a p53-mediated activity during tumour suppression. Nature 520, 57- 62, doi:10.1038/nature14344 (2015). 134 Day, C. P., Merlino, G. & Van Dyke, T. Preclinical mouse cancer models: a maze of opportunities and challenges. Cell 163, 39-53, doi:10.1016/j.cell.2015.08.068 (2015). 135 Van Dyke, T. & Jacks, T. Cancer modeling in the modern era: progress and challenges. Cell 108, 135-144 (2002). 136 Schonhuber, N. et al. A next-generation dual-recombinase system for time- and host-specific targeting of pancreatic cancer. Nature medicine 20, 1340-1347, doi:10.1038/nm.3646 (2014). 137 Swank, R. T., Reddington, M. & Novak, E. K. Inherited prolonged bleeding time and platelet storage pool deficiency in the subtle gray (sut) mouse. Laboratory animal science 46, 56-60 (1996).

106

138 Huizing, M., Anikster, Y., White, J. G. & Gahl, W. A. Characterization of the murine gene corresponding to human Hermansky-Pudlak syndrome type 3: exclusion of the Subtle gray (sut) locus. Molecular genetics and metabolism 74, 217-225, doi:10.1006/mgme.2001.3233 (2001). 139 Chintala, S. et al. Slc7a11 gene controls production of pheomelanin pigment and proliferation of cultured cells. Proceedings of the National Academy of Sciences of the United States of America 102, 10964-10969, doi:10.1073/pnas.0502856102 (2005). 140 Shih, A. Y. et al. Cystine/glutamate exchange modulates glutathione supply for neuroprotection from oxidative stress and cell proliferation. The Journal of neuroscience : the official journal of the Society for Neuroscience 26, 10514-10523, doi:10.1523/JNEUROSCI.3178-06.2006 (2006). 141 Liu, R. R., Brown, C. E. & Murphy, T. H. Differential regulation of cell proliferation in neurogenic zones in mice lacking cystine transport by xCT. Biochemical and biophysical research communications 364, 528-533, doi:10.1016/j.bbrc.2007.10.036 (2007). 142 Sleire, L. et al. Drug repurposing: sulfasalazine sensitizes gliomas to gamma knife radiosurgery by blocking cystine uptake through system Xc-, leading to glutathione depletion. Oncogene 34, 5951-5959, doi:10.1038/onc.2015.60 (2015). 143 Gout, P. W., Buckley, A. R., Simms, C. R. & Bruchovsky, N. Sulfasalazine, a potent suppressor of lymphoma growth by inhibition of the x(c)- cystine transporter: a new action for an old drug. Leukemia 15, 1633-1640 (2001). 144 Guerra, C. et al. Chronic pancreatitis is essential for induction of pancreatic ductal adenocarcinoma by K-Ras oncogenes in adult mice. Cancer cell 11, 291-302, doi:10.1016/j.ccr.2007.01.012 (2007). 145 Wilkerson, J. et al. Estimation of tumour regression and growth rates during treatment in patients with advanced prostate cancer: a retrospective analysis. The Lancet. Oncology 18, 143- 154, doi:10.1016/S1470-2045(16)30633-7 (2017). 146 Zhao, Y., Butler, E. B. & Tan, M. Targeting cellular metabolism to improve cancer therapeutics. Cell death & disease 4, e532, doi:10.1038/cddis.2013.60 (2013). 147 Sato, H. et al. Redox imbalance in cystine/glutamate transporter-deficient mice. The Journal of biological chemistry 280, 37423-37429, doi:10.1074/jbc.M506439200 (2005). 148 Kim, J. Y. et al. Human cystine/glutamate transporter: cDNA cloning and upregulation by oxidative stress in glioma cells. Biochimica et biophysica acta 1512, 335-344 (2001). 149 Boj, S. F. et al. Organoid models of human and mouse ductal pancreatic cancer. Cell 160, 324- 338, doi:10.1016/j.cell.2014.12.021 (2015). 150 D'Autreaux, B. & Toledano, M. B. ROS as signalling molecules: mechanisms that generate specificity in ROS homeostasis. Nature reviews. Molecular cell biology 8, 813-824, doi:10.1038/nrm2256 (2007). 151 Hu, C. M. & Zhang, L. Nanoparticle-based combination therapy toward overcoming drug resistance in cancer. Biochemical pharmacology 83, 1104-1111, doi:10.1016/j.bcp.2012.01.008 (2012). 152 Xu, R. H. et al. Inhibition of glycolysis in cancer cells: a novel strategy to overcome drug resistance associated with mitochondrial respiratory defect and hypoxia. Cancer research 65, 613-621 (2005). 153 Holohan, C., Van Schaeybroeck, S., Longley, D. B. & Johnston, P. G. Cancer drug resistance: an evolving paradigm. Nature reviews. Cancer 13, 714-726, doi:10.1038/nrc3599 (2013). 154 Dean, M., Fojo, T. & Bates, S. Tumour stem cells and drug resistance. Nature reviews. Cancer 5, 275-284, doi:10.1038/nrc1590 (2005). 155 Pao, W. et al. Acquired resistance of lung adenocarcinomas to gefitinib or erlotinib is associated with a second mutation in the EGFR kinase domain. PLoS medicine 2, e73, doi:10.1371/journal.pmed.0020073 (2005).

107

156 Engelman, J. A. et al. MET amplification leads to gefitinib resistance in lung cancer by activating ERBB3 signaling. Science 316, 1039-1043, doi:10.1126/science.1141478 (2007). 157 Santos, N. A. et al. Cisplatin-induced nephrotoxicity is associated with oxidative stress, redox state unbalance, impairment of energetic metabolism and apoptosis in rat kidney mitochondria. Archives of toxicology 81, 495-504, doi:10.1007/s00204-006-0173-2 (2007). 158 Dehne, N., Lautermann, J., Petrat, F., Rauen, U. & de Groot, H. Cisplatin ototoxicity: involvement of iron and enhanced formation of superoxide anion radicals. Toxicology and applied pharmacology 174, 27-34, doi:10.1006/taap.2001.9171 (2001). 159 Jiang, Y., Guo, C., Vasko, M. R. & Kelley, M. R. Implications of apurinic/apyrimidinic endonuclease in reactive oxygen signaling response after cisplatin treatment of dorsal root ganglion neurons. Cancer research 68, 6425-6434, doi:10.1158/0008-5472.CAN-08-1173 (2008). 160 Azzam, E. I., Jay-Gerin, J. P. & Pain, D. Ionizing radiation-induced metabolic oxidative stress and prolonged cell injury. Cancer letters 327, 48-60, doi:10.1016/j.canlet.2011.12.012 (2012). 161 Yun, J. et al. Vitamin C selectively kills KRAS and BRAF mutant colorectal cancer cells by targeting GAPDH. Science 350, 1391-1396, doi:10.1126/science.aaa5004 (2015). 162 Bailey, H. H. et al. Phase I clinical trial of intravenous L-buthionine sulfoximine and melphalan: an attempt at modulation of glutathione. Journal of clinical oncology : official journal of the American Society of Clinical Oncology 12, 194-205, doi:10.1200/JCO.1994.12.1.194 (1994). 163 Walker, E. J., Ko, A. H., Holly, E. A. & Bracci, P. M. Statin use and risk of pancreatic cancer: results from a large, clinic-based case-control study. Cancer 121, 1287-1294, doi:10.1002/cncr.29256 (2015). 164 Friedmann Angeli, J. P. et al. Inactivation of the ferroptosis regulator Gpx4 triggers acute renal failure in mice. Nature cell biology 16, 1180-1191, doi:10.1038/ncb3064 (2014). 165 Wang, J. T., Teasdale, R. D. & Liebl, D. Macropinosome quantitation assay. MethodsX 1, 36-41, doi:10.1016/j.mex.2014.05.002 (2014). 166 Team, R. C. R: A Language and Environment for Statistical Computing. (2016). 167 Davis, S. & Meltzer, P. S. GEOquery: a bridge between the Gene Expression Omnibus (GEO) and BioConductor. Bioinformatics 23, 1846-1847, doi:10.1093/bioinformatics/btm254 (2007). 168 Donahue, T. R. et al. Integrative Survival-Based Molecular Profiling of Human Pancreatic Cancer. Clinical Cancer Research 18, 1352 (2012). 169 Hiraoka, N. et al. CXCL17 and ICAM2 Are Associated With a Potential Anti-Tumor Immune Response in Early Intraepithelial Stages of Human Pancreatic Carcinogenesis. Gastroenterology 140, 310-321.e314, doi:10.1053/j.gastro.2010.10.009. 170 Pei, H. et al. FKBP51 Affects Cancer Cell Response to Chemotherapy by Negatively Regulating Akt. Cancer cell 16, 259-266, doi:10.1016/j.ccr.2009.07.016. 171 Yang, S. et al. A Novel MIF Signaling Pathway Drives the Malignant Character of Pancreatic Cancer by Targeting NR3C2. Cancer Research 76, 3838 (2016). 172 Moffitt, R. A. et al. Virtual microdissection identifies distinct tumor-and stroma-specific subtypes of pancreatic ductal adenocarcinoma. Nat. Genet. 47, 1168-1178 (2015). 173 Ruckert, F. et al. Co-expression of KLK6 and KLK10 as prognostic factors for survival in pancreatic ductal adenocarcinoma. British journal of cancer 99, 1484-1492, doi:10.1038/sj.bjc.6604717 (2008). 174 Grützmann, R. et al. Gene expression profiling of microdissected pancreatic ductal carcinomas using high-density DNA microarrays. Neoplasia 6, 611-622, doi:10.1593/neo.04295 (2004). 175 Wu, Z., Irizarry, R. A., Gentleman, R., Martinez-Murillo, F. & Spencer, F. A Model-Based Background Adjustment for Oligonucleotide Expression Arrays. Journal of the American Statistical Association 99, 909-917, doi:10.1198/016214504000000683 (2004). 176 Samur, M. K. RTCGAToolbox: A New Tool for Exporting TCGA Firehose Data. PLOS ONE 9,

108

e106397, doi:10.1371/journal.pone.0106397 (2014). 177 Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 43, e47, doi:10.1093/nar/gkv007 (2015). 178 Viechtbauer, W. Conducting Meta-Analyses in R with the metafor Package. Journal of Statistical Software, Articles 36, 1--48, doi:10.18637/jss.v036.i03 (2010). 179 Therneau, T. M. & Grambsch, P. M. Modeling Survival Data: Extending the Cox Model. (Springer New York, 2013). 180 Rhim, A. D. et al. EMT and dissemination precede pancreatic tumor formation. Cell 148, 349- 361, doi:10.1016/j.cell.2011.11.025 (2012). 181 Carriere, C., Young, A. L., Gunn, J. R., Longnecker, D. S. & Korc, M. Acute pancreatitis accelerates initiation and progression to pancreatic cancer in mice expressing oncogenic Kras in the nestin cell lineage. PLoS One 6, e27725, doi:10.1371/journal.pone.0027725 (2011). 182 Carriere, C., Young, A. L., Gunn, J. R., Longnecker, D. S. & Korc, M. Acute pancreatitis markedly accelerates pancreatic cancer progression in mice expressing oncogenic Kras. Biochemical and biophysical research communications 382, 561-565, doi:10.1016/j.bbrc.2009.03.068 (2009).

109