Jund Axis Austin Lewis Good University of Pennsylvania, [email protected]

Unraveling the Beta Cell Translatome: Elucidation of an Erk/Hnrnpk/Jund Axis Austin Lewis Good University of Pennsylvania, Algood6@Gmail.Com

University of Pennsylvania ScholarlyCommons

Publicly Accessible Penn Dissertations

2019 Unraveling The Beta Cell Translatome: Elucidation Of An Erk/hnrnpk/jund Axis Austin Lewis Good University of Pennsylvania, [email protected]

Follow this and additional works at: https://repository.upenn.edu/edissertations Part of the Cell Biology Commons, Genetics Commons, and the Molecular Biology Commons

Recommended Citation Good, Austin Lewis, "Unraveling The Beta Cell Translatome: Elucidation Of An Erk/hnrnpk/jund Axis" (2019). Publicly Accessible Penn Dissertations. 3327. https://repository.upenn.edu/edissertations/3327

This paper is posted at ScholarlyCommons. https://repository.upenn.edu/edissertations/3327 For more information, please contact [email protected]. Unraveling The Beta Cell Translatome: Elucidation Of An Erk/hnrnpk/ jund Axis

Abstract In type 2 diabetes, oxidative stress contributes to the dysfunction and loss of pancreatic β cells. A highly conserved feature of the cellular response to stress is the regulation of mRNA translation, however, the mechanisms underlying this process in β cells are not fully understood. Here we use TRAP-seq to examine changes in the ribosome occupancy of mRNAs during conditions associated with β cell dysfunction, leading us to identify a cohort of translationally regulated genes with 3’UTR enrichment of a cytosine-rich motif. Of particular interest was the gene encoding JUND, a transcription factor with anti-oxidant functions in other cell types but whose role in cells is unknown. Interestingly, JUND is translationally upregulated in islets exposed to high glucose and free fatty acid levels, and depletion of JUND in β cells reduces oxidative stress and apoptosis caused by these conditions. Transcriptome assessment demonstrates that JUND regulates a cohort of genes that are commonly dysregulated during β cell dysfunction, including pro-oxidant and pro- inflammatory genes, consistent with this factor enhancing, rather than reducing, oxidative stress levels in cells. Further, hnRNPK, an RNA binding protein with specificity for cytosine-rich stretches, binds to the mRNA encoding JUND and is required for its post-transcriptional upregulation during metabolic stress. Although the absolute levels of hnRNPK do not change, there is a significant increase in hnRNPK phosphorylation during glucolipotoxicity. Importantly, this hnRNPK/JUND axis is activated in islets from diabetic db/db mice and in human islets exposed to metabolic stress. Finally, a series of mechanistic studies indicate that hnRNPK post- transcriptionally regulates JUND in a MEK/ERK- and DDX3X-dependent manner. Thus, a translation- centric approach uncovered hnRNPK and JUND as stress-responsive factors in β cells that contribute to redox imbalance and apoptosis during pathophysiologically relevant stress.

Degree Type Dissertation

Degree Name Doctor of Philosophy (PhD)

Graduate Group Cell & Molecular Biology

First Advisor Doris A. Stoffers

Keywords Beta cells, Diabetes, Oxidative stress, RNA binding proteins, Transcriptional regulation, Translation

Subject Categories Cell Biology | Genetics | Molecular Biology

This dissertation is available at ScholarlyCommons: https://repository.upenn.edu/edissertations/3327 UNRAVELING THE β CELL TRANSLATOME: ELUCIDATION OF AN ERK/HNRNPK/JUND AXIS

Austin L. Good

A DISSERTATION

Cell and Molecular Biology

Presented to the Faculties of the University of Pennsylvania

Partial Fulfillment of the Requirements for the

Degree of Doctor of Philosophy

2019

Supervisor of Dissertation

______

Doris A. Stoffers, M.D., Ph.D. Sylvan H. Eisman Professor of Medicine

Graduate Group Chairperson

______

Daniel S. Kessler, Ph.D. Associate Professor of Cell and Developmental Biology

Dissertation Committee

Klaus H. Kaestner, Ph.D. Thomas and Evelyn Suor Butterworth Professor in Genetics

Patrick Seale, Ph.D. Associate Professor of Cell and Developmental Biology

Stephen A. Liebhaber, M.D. Professor of Genetics

Zissimos Mourelatos, M.D. Professor of Pathology and Laboratory Medicine

ACKNOWLEDGMENTS

I would like to thank the many people that have contributed to my scientific development and research endeavors during my graduate work. I am very grateful to my thesis advisor, Dr. Doris

Stoffers, for her mentorship and support throughout this process. Her curiosity and excitement to more deeply understand scientific questions provide a wonderful model for her trainees.

Throughout my training, Doris encouraged me to develop my own ideas and hypotheses while also serving as the guard rails to keep me on track when necessary. Her mentorship allowed me to grow as an investigator and gave me confidence to tackle complex problems.

I am also very grateful for the opportunity to work alongside and interact with many bright and enthusiastic members of the Stoffers lab during my graduate work who have all contributed to my training. I especially want to thank Dr. Corey Cannon who served as a wonderful mentor for me when I joined the lab and who laid the foundation for all of the work presented here. I also want to thank Dave Groff and Dr. Andrea Rozo for putting up with all of my urgent order requests over the years. I also thank Drs. Christine Juliana, Diana Stanescu, Andrea Rozo, Jeff Raum, Ben Ediger, and Juxiang Yang for their many suggestions and advice on my project. I want to thank Matt

Haemmerle and Alexis Oguh for their grad student camaraderie, and I wish them the best of luck in their thesis work and future endeavors.

I would also like to thank Dr. Skip Brass for sharing his wisdom throughout the many stages of the MD-PhD program, and Maggie Krall for being a listening ear and giving perceptive advice.

I am very lucky to have a family that provides me with unwavering support and encouragement.

Amanda, Matt, Julian, and Camille, spending time together always provides a necessary escape from the world of science and gives me perspective to keep moving forward with a clear head.

Mom and Dad, I can’t thank you enough for the love and support you have always given me. I can’t imagine better role models. And finally, I thank my wife Charly Ann for always making me smile at the end of a long day. You are my rock and I can’t wait for the many adventures ahead.

ABSTRACT

UNRAVELING THE β CELL TRANSLATOME: ELUCIDATION OF AN ERK/HNRNPK/JUND AXIS

AUSTIN L. GOOD

DORIS A. STOFFERS, M.D., Ph.D.

In type 2 diabetes, oxidative stress contributes to the dysfunction and loss of pancreatic β cells. A highly conserved feature of the cellular response to stress is the regulation of mRNA translation, however, the mechanisms underlying this process in β cells are not fully understood. Here we use TRAP-seq to examine changes in the ribosome occupancy of mRNAs during conditions associated with β cell dysfunction, leading us to identify a cohort of translationally regulated genes with 3’UTR enrichment of a cytosine-rich motif. Of particular interest was the gene encoding JUND, a transcription factor with anti-oxidant functions in other cell types but whose role in β cells is unknown. Interestingly, JUND is translationally upregulated in islets exposed to high glucose and free fatty acid levels, and depletion of JUND in β cells reduces oxidative stress and apoptosis caused by these conditions. Transcriptome assessment demonstrates that JUND regulates a cohort of genes that are commonly dysregulated during β cell dysfunction, including pro-oxidant and pro-inflammatory genes, consistent with this factor enhancing, rather than reducing, oxidative stress levels in β cells. Further, hnRNPK, an RNA binding protein with specificity for cytosine-rich stretches, binds to the mRNA encoding JUND and is required for its post-transcriptional upregulation during metabolic stress. Although the absolute levels of hnRNPK do not change, there is a significant increase in hnRNPK phosphorylation during glucolipotoxicity.

Importantly, this hnRNPK/JUND axis is activated in islets from diabetic db/db mice and in human islets exposed to metabolic stress. Finally, a series of mechanistic studies indicate that hnRNPK post-transcriptionally regulates JUND in a MEK/ERK- and DDX3X-dependent manner. Thus, a translation-centric approach uncovered hnRNPK and JUND as stress-responsive factors in β cells that contribute to redox imbalance and apoptosis during pathophysiologically relevant stress.

iii

TABLE OF CONTENTS

ACKNOWLEDGMENTS ...... II

ABSTRACT ...... III

LIST OF TABLES ...... VII

LIST OF ILLUSTRATIONS ...... VIII

CHAPTER 1: INTRODUCTION ...... 1

1.1 Paradigms of β cell demise in type 2 diabetes ...... 1 1.1.1 Impaired insulin secretion ...... 2 1.1.2 β cell apoptosis ...... 4 1.1.3 Loss of β cell identify ...... 5

1.2 β cell stress in T2D ...... 6 1.2.1 Oxidative stress ...... 7 1.2.2 ER stress ...... 13

1.3 Translational regulation and cellular stress ...... 17 1.3.1 Overview of translation initiation ...... 18 1.3.2 Regulation of global translation rates ...... 23 1.3.3 Selective regulation of translation ...... 24 1.3.4 Translational regulation mediated by the PCBPs ...... 27

CHAPTER 2: TRAP-SEQ UNCOVERS POST-TRANSCRIPTIONAL REGULATION MEDIATED BY THE PCBPS IN β CELLS ...... 31

2.1 Summary ...... 31

2.2 Introduction ...... 32

2.3 Materials and Methods ...... 38

2.4 Results ...... 46 2.4.1 Establishment of TRAP in Min6 cells ...... 46 2.4.2 TRAP-seq during PDX1 deficiency ...... 48 2.4.3 Post-transcriptional regulation by the poly(C)-binding proteins ...... 50 2.4.4 Impact of the PCBPs on insulin secretion in Min6 cells ...... 58

2.5 Discussion ...... 61

CHAPTER 3: JUND PROMOTES β CELL APOPTOSIS AND OXIDATIVE STRESS DURING METABOLIC STRESS ...... 67

3.1 Summary ...... 67

3.2 Introduction ...... 68

3.3 Materials and Methods ...... 72

3.4 Results ...... 80 3.4.1 Translational upregulation of JUND in β cells during metabolic stress ...... 80 3.4.2 β cell-specific depletion of JUND in primary islets ...... 86 3.4.3 Reduced oxidative stress in β cells with JUND depletion during metabolic stress ...... 88 3.4.4 Reduced apoptosis in β cells with JUND depletion during metabolic stress ...... 90 3.4.5 JUND regulates pro-oxidant and pro-inflammatory genes in β cells ...... 92

3.5 Discussion ...... 96

CHAPTER 4: AN ERK/HNRNPK/DDX3X AXIS POST-TRANSCRIPTIONALLY REGULATES JUND DURING METABOLIC STRESS ...... 101

4.1 Summary ...... 101

4.2 Introduction ...... 102

4.3 Materials and Methods ...... 104

4.4 Results ...... 108 4.4.1 hnRNPK is required for the translational induction of JUND during metabolic stress ... 108 4.4.2 hnRNPK is phosphorylated during metabolic stress in a MEK-dependent manner ...... 110 4.4.3 No change in hnRNPK localization, RNA binding affinity, or nuclear retention of RNA during metabolic stress ...... 119 4.4.4 DDX3X interacts with hnRNPK and is required for efficient translation of JUND mRNA ...... 126

4.5 Discussion ...... 131

CHAPTER 5: CONCLUDING REMARKS AND FUTURE DIRECTIONS ...... 140

5.1 Discussion of key findings ...... 140

5.2 Additional screens for translational regulation in β cells ...... 144

5.3 Assessment of RBP functions in β cells ...... 146

5.4 Expanding the analysis of JUND function and regulation ...... 148

5.5 Translational regulation mediated by an hnRNPK-DDX3X complex ...... 152

BIBLIOGRAPHY ...... 156

LIST OF TABLES

Chapter 1

- Chapter 2

Table 1. Primer sequences for RT-qPCR Chapter 3

Table 2. Human islet donor characteristics Chapter 4

- Chapter 5

vii

LIST OF ILLUSTRATIONS

Chapter 1

Figure 1.1. β cell dysfunction, apoptosis, and loss of cell identity contribute to β failure in T2D.

Figure 1.2. ROS production from various subcellular locations contributes to the onset of oxidative stress.

Figure 1.3. The unfolded protein response (UPR) shapes cellular adaptation to ER stress.

Figure 1.4. Translation initiation is a complex, multi-step process. Chapter 2

Figure 2.1. TRAP in GFP-L10a Min6 cells pulls down ribosome-associated mRNA and detects shifts in the ribosome occupancy of ATF4 during ER stress.

Figure 2.2. TRAP in GFP-L10a Min6 detects shifts in ribosome occupancy after PDX1 depletion.

Figure 2.3. Enrichment of a cytosine-rich motif in genes from TRAP-seq identified by de novo motif analysis.

Figure 2.4. PCPB1, PCBP2, and hnRNPK bind to the mRNAs encoding NXK2-2 and JUND.

Figure 2.5. Detection of mRNAs bound by hnRNPK via RIP-seq.

Figure 2.6. Loss of PCBP1/2, but not hnRNPK, reduces NKX2-2 and JUND protein levels under baseline conditions.

Figure 2.7. Loss of PCBP1/2 or hnRNPK impairs insulin secretion from Min6 cells. Chapter 3

Figure 3.1. JUND is translationally upregulated in β cells during metabolic stress.

Figure 3.2. Post-transcriptional induction of JUND in islets from db/db mice and in human islets during metabolic stress.

Figure 3.3. β cell-specific depletion of JUND in mouse islets via lentiviral delivery of shRNA.

Figure 3.4. β cell-specific depletion of JUND in mouse islets reduces oxidative stress during glucolipotoxicity.

Figure 3.5. β cell-specific depletion of JUND in mouse islets reduces apoptosis during glucolipotoxicity.

Figure 3.6. JUND regulates pro-oxidant and pro-inflammatory genes associated with β cell stress and dysfunction. Chapter 4

viii

Figure 4.1. CRISPR-mediated depletion of hnRNPK blocks JUND induction during glucolipotoxicity.

Figure 4.2. Increased phosphorylation of hnRNPK during metabolic stress.

Figure 4.3. No change in phosphorylation of PCBP1/2 during metabolic stress as detected by Phos-tag analysis.

Figure 4.4. MEK inhibition blocks hnRNPK phosphorylation and JUND induction during metabolic stress.

Figure 4.5. Overexpression of constitutively active MEK1 in Min6 cells is sufficient to activate the hnRNPK/JUND axis.

Figure 4.6. MEK inhibition reduces islet cell apoptosis during glucolipotoxicity.

Figure 4.7. No change in hnRNPK subcellular distribution in Min6 cells during metabolic stress.

Figure 4.8. No change in hnRNPK RNA binding affinity or nuclear retention of mRNA during metabolic stress.

Figure 4.9. DDX3X is required for the efficient translation of JUND mRNA.

Figure 4.10. DDX3X interacts with the 18S ribosomal RNA in an hnRNPK-dependent manner.

Figure 4.11. Model diagram depicting an ERK/hnRNPK/JUND axis that promotes oxidative stress and apoptosis in β cells.

Figure 4.12. Model diagram depicting recruitment of the PIC by an hnRNPK-DDX3X complex during metabolic stress. Chapter 5

CHAPTER 1: INTRODUCTION

1.1 Paradigms of β cell demise in type 2 diabetes

By 2030, it is predicted that 439 million people across the world, or 7.7% of the total adult population, will have diabetes (Shaw et al., 2010). The peril posed by this global escalation in type 2 diabetes (T2D) prevalence is manifold. First, this disease can have debilitating consequences for patients including blindness, amputations, and kidney disease. Additionally, the treatment of patients with diabetes accounted for at least 1 in 8 US health care dollars in 2017 (American

Diabetes Association, 2018). Therefore, the identification of new treatment strategies to combat the rising prevalence of T2D is a critical task for the biomedical research community.

A key biological determinant of T2D pathogenesis is the functioning of highly specialized cells that reside within the pancreas known as pancreatic β cells.

These are the only cells in the body capable of producing and secreting the hormone insulin to regulate blood glucose levels. β cells, along with other endocrine cell types, are found within clusters of cells known as islets of

Langerhans, or islets for short. During the progression of T2D, insulin resistance of peripheral tissues, such as the liver, muscle, and adipose, places a growing burden on pancreatic β cells to produce and secrete more insulin. Eventually, the demand for insulin surpasses the functional capacity of β cells, leading to hyperglycemia (Leahy, 2005). Furthermore, this discrepancy worsens as the

1 disease progresses due to a decline in the number and in the functioning of β cells. This outcome, known as β cell failure, leads to severe hyperglycemia and diabetic complications (Prentki and Nolan, 2006). There are several proposed mechanisms to explain the development of β cell failure in T2D, including impaired insulin secretion, β cell apoptosis, and loss of β cell identity, termed

“dedifferentiation” (Figure 1.1) (Kitamura, 2013). However, the relative contribution of these processes towards T2D pathogenesis is unclear and may be dependent on the severity of the disease and the genetic makeup of the individual (Prentki and Nolan, 2006).

Figure 1.1. β cell dysfunction, apoptosis, and loss of cell identity contribute to β failure in T2D.

1.1.1 Impaired insulin secretion

In response to insulin resistance in obese individuals, β cells can augment insulin secretion as much as five times that seen in healthy controls despite only a 50%

2 increase in β cell mass (Kahn et al., 2006). This indicates that the insulin secretory capacity of a single β cell is dynamic and can be expanded to accommodate increased demand. Thus, the ability of β cells to compensate for insulin resistance is dependent not just on the number of β cells present in the pancreas, but also on the functional capacity of those cells. Indeed, it has been suggested that the initial cause of hyperglycemia in T2D is primarily a consequence of defects in insulin secretion, not a reduction in the number of β cells (Prentki and Nolan, 2006). For example, there is only a very small reduction in β cell mass in patients with a 5 year history of T2D, which is unlikely to explain the presence of hyperglycemia in the absence of defective insulin secretion

(Rahier et al., 2008). Similarly, C57BL/6 mice fed a high fat diet for 8 weeks develop hyperglycemia, and the severity of the rise in blood glucose levels correlates with impairments in insulin secretion, but not β cell mass (Peyot et al.,

2010).

This notion of β cell dysfunction caused by conditions associated with T2D is also supported by experiments performed on cultured islets obtained from mice or deceased human organ donors. In this ex vivo paradigm, the concentrations of glucose and free fatty acid levels in the culture media can be increased to mimic changes that occur in T2D, a model referred to as glucolipotoxicity. Exposure of either mouse or human islets to these conditions causes defects in glucose- stimulated insulin release (Doliba et al., 2017). Together, these findings indicate that impaired insulin secretion from β cells contributes to the inadequate insulin 3 levels in patients with T2D, and is especially central during early stages of the disease.

1.1.2 β cell apoptosis

Although reductions in β cell mass may be unable to fully explain the rises in blood glucose levels that occur during early stages of T2D, it is clear that long- standing cases of T2D and severe hyperglycemia are associated with a significant decline in the number of β cells (Rahier et al., 2008). This distinction is especially important when considering the likelihood of success for particular therapeutic interventions. For example, sulfonylureas, which increase insulin release from β cells, may be inadequate to restore normoglycemia if there is a significant reduction in β cell mass (Page and Reisman, 2013).

A main cause of reduced β cell mass in T2D is an increase in β cell apoptosis

(Prentki and Nolan, 2006). Examination of pancreatic tissue from autopsies indicated that individuals with T2D had reduced β cell mass that was associated with increased rates of β cell apoptosis (Butler et al., 2003). Many studies have also shown that elevations in glucose and/or free fatty acid levels lead to increased apoptosis in β cells during ex vivo culturing (Kharroubi et al., 2004;

Robertson et al., 2003; Shimabukuro et al., 1998).

1.1.3 Loss of β cell identify

Although significant evidence indicates that β cell apoptosis contributes to the reduction in β cell mass in T2D, it has recently come into question whether cell death can fully explain this decrease in β cell number. An alternative explanation has been proposed in which β cells lose their cell identity during the development of T2D, which entails the acquisition of features normally restricted to progenitor cells or other endocrine cell types (Accili et al., 2016; Dor and Glaser, 2013). This model has been considered plausible partly because there is known to be a high degree of cellular plasticity amongst endocrine cells of the pancreas (Puri et al.,

2015). For example, after near complete destruction of β cells in mice, α cells can convert into β cells to regenerate this population (Thorel et al., 2010).

This concept of cellular plasticity was first applied to β cell failure based on the observation that β cells lose markers of mature β cells and acquire markers of progenitor cells in mice with genetic deletion of the transcription factor FOXO1 or in murine models of diabetes (Talchai et al., 2012). The relevance of these findings to human β cells in T2D, however, is unresolved. A survey of human diabetic pancreata at autopsy did not find re-expression of NGN3, a marker of endocrine progenitor cells shown to be increased in mouse models with β cell dedifferentiation. However, there were some abnormal findings in this analysis that could support a loss of β cell identity, including cells with co-expression of

NKX6.1 and glucagon/somatostatin and the misexpression of ALDH1a3, another

5 progenitor cell marker (Cinti et al., 2015). Therefore, it will be critical to reach a consensus on the features that define a “dedifferentiated” β cell to properly assess its relevance for T2D. It will also be important to further clarify the fate of these cells: have they truly adopted a new cell identity or do they simply represent dysfunctional β cells? Regardless, the idea that reduced β cell mass in

T2D is not solely caused by apoptosis raises the exciting possibility that this loss may be reversible.

1.2 β cell stress in T2D

In order to develop new therapeutic strategies to prevent or reverse β cell failure in T2D, the molecular mechanisms underlying these deleterious processes need to be elucidated. For example, what signaling pathways in β cells promote the adoption of an apoptotic fate? What factors control β cell identity and how are they dysregulated in disease conditions? A common theme that begins to address these questions is that β cells endure various forms of cellular stress in the pathogenesis of T2D. In this context, the term “stress” refers to a significant deviation from a homeostatic set point that causes cellular dysfunction and/or damage. In particular, an imbalance in redox homeostasis, or oxidative stress, and an imbalance in protein folding homeostasis, or ER stress, have been implicated in β cell demise in diabetes.

1.2.1 Oxidative stress

Reactive oxygen species (ROS) are normal byproducts of cellular respiration, and the major endogenously produced forms of ROS include hydrogen peroxide, superoxide, and hydroxyl radical (Lipinski, 2001). When generated at low levels,

ROS can function as critical signaling molecules. An important mechanism for

ROS-mediated signaling involves cysteine residues, which are particularly susceptible to oxidation by ROS due to unique properties of their thiol group

(Poole, 2015). Cysteine oxidation can cause rearrangements in protein conformation, leading to changes in function or activation of signaling pathways

(Ray et al., 2012). For example, oxidation of Cys-2991 of the kinase ATM by hydrogen peroxide increases its activity to promote DNA break repair processes

(Guo et al., 2010). When present at high levels, however, ROS can be very detrimental to the cell by damaging lipids, proteins, and DNA, and the onset of

ROS-mediated injury is referred to as oxidative stress (Schieber and Chandel,

2014). Lipids within cell membranes are particularly vulnerable to oxidation, which can severely disrupt the structure and function of lipid membranes throughout the cell leading to cell death (Rochette et al., 2014).

Oxidative stress is proposed to be a central contributor to β cell failure in T2D.

Islets from patients with T2D show increased levels of 8-hydroxy-2' – deoxyguanosine, a marker of oxidative damage to DNA (Del Guerra et al., 2005).

Animal models of T2D, such as chronic high fat diet feeding in mice, also display signs of oxidative damage in β cells (Hatanaka et al., 2017). Furthermore,

7 antioxidant treatment (Han et al., 2015) or overexpression of antioxidant genes

(Yagishita et al., 2014) can improve β cell function in models of T2D, suggesting that oxidative stress is an important determinant of β demise in T2D.

The development and resolution of oxidative stress depends on the balance between the generation of ROS and the levels of antioxidant enzymes that detoxify these molecules. Antioxidant genes can be subdivided into three groups based on their targeted ROS species and mechanisms of action: superoxide dismutases, peroxidases, and thiol-redox proteins (Gelain et al., 2009).

Antioxidant genes are broadly expressed across tissues due to their essential role in preventing oxidative stress, but their relative expression levels can vary to meet the needs of a particular cell type. For example, β cells are thought to be particularly sensitive to oxidative stress due to low expression levels of antioxidant genes (Lenzen et al., 1996). This is especially true for the hydrogen peroxide-inactivating enzymes catalase and glutathione peroxidase, which are expressed in islets at levels less than 5% of that in the liver (Tiedge et al., 1997).

The effect of ROS production on β cell homeostasis is complex because it is dependent on both the type and subcellular location of the oxidant (Figure 1.2).

For example, β cells seem particularly vulnerable to cell death caused by superoxide and hydrogen peroxide but are mostly resistant to the oxidant peroxynitrite, which is generated by the reaction of superoxide with nitric oxide

(Broniowska et al., 2015). This effect cannot be explained by the potency of

8 oxidation as peroxynitrite is a very strong oxidant (Pacher et al., 2007). Thus, the effect of ROS on β cell viability is partly determined by the types of oxidants being produced, which is further complicated by the reactivity of these molecules with each other. Along these lines, treatment of islets with low levels of hydrogen peroxide, but not other oxidants, has been found to increase insulin secretion (Pi et al., 2007).

Figure 1.2. ROS production from various subcellular locations contributes to the onset of oxidative stress. Adapted from (Kreuz and Fischle, 2016).

Besides the type of oxidant, the site of ROS production in β cells will also dictate its impact on apoptosis and can vary depending on the conditions invoking oxidative stress (Figure 1.2). Under conditions of excess glucose, for example, increased electron flux through the respiratory chain leads to increased production of superoxide in mitochondria (Robertson et al., 2003). Mitochondria contain a distinct set of antioxidant enzymes that detoxify ROS produced in this setting, such as mitochondrial superoxide dismutase that resides exclusively in the mitochondrial matrix (Ježek et al., 2012). Thus, the severity of oxidative stress in mitochondria will largely be determined by the functioning of mitochondrial, but not cytosolic, antioxidant enzymes. On the other hand, the presence of excess lipid can cause fatty acid catabolism to shift largely into peroxisomes, which leads to hydrogen peroxide, rather than superoxide, generation (Gehrmann et al., 2010). Interestingly, ROS production from peroxisomes, rather than mitochondria, has been suggested to promote apoptosis in β cells during glucolipotoxicity (Elsner et al., 2010).

The other major proposed sources of ROS production in β cells are located in the cytoplasm. Under conditions of excess glucose, for example, glucose metabolites can be shunted towards atypical pathways including the hexosamine pathway, which has been suggested to increase hydrogen peroxide levels in β cells

(Kaneto et al., 2001). Additionally, members of the NADPH oxidase (NOX) family can generate cytoplasmic superoxide by directly transferring electrons from

NADPH to oxygen (Maghzal et al., 2012). These enzymes are best known for

11 their role in innate immunity in which they produce large amounts of superoxide in neutrophils for release at sites of infection (Nguyen et al., 2017). NOX-1, -2, and -4 are expressed at low levels in β cells, and their activity may be increased during exposure to glucotoxicity or pro-inflammatory cytokines (Morgan et al.,

2006; Weaver et al., 2014). Similarly, there are many additional enzymes that catalyze redox reactions in common biological processes. For example, the metalloreductase STEAP4 facilitates the movement of electrons from NAPDH to extracellular iron or copper (Scarl et al., 2017). Under certain conditions, however, this enzymatic function of STEAP4 may promote oxidative stress in the cell (Zhou et al., 2013a). Thus, ROS production can also be derived from the enzymatic activity of pro-oxidant genes. Given the wide range of sources, subcellular locations, and types of oxidants produced in the cell, it is not surprising that the pertinent causes of oxidative stress in β cells during T2D are still uncertain.

Additionally, the upregulation of antioxidant genes to combat oxidative stress may be a coordinated process in the cell. For example, the transcription factor

NRF2 is a key regulator of antioxidant genes across different cell types (Itoh et al., 1997). In the absence of oxidative stress, NRF2 is targeted for ubiquitination by the adaptor protein KEAP1 leading to its proteasomal degradation. Increased

ROS levels, however, inhibit KEAP1 function via cysteine oxidation, and NRF2 levels accumulate in the nucleus to upregulate antioxidant genes (Uruno and

Motohashi, 2011). Interestingly, overexpression of NRF2 in β cells improves

12 hyperglycemia in several models of diabetes (Uruno et al., 2013; Yagishita et al.,

2014). Similarly, the transcription factor JUND has also been proposed to reduce oxidative stress by regulating antioxidant gene expression (Gerald et al., 2004;

Paneni et al., 2013). However, its role in β cell redox homeostasis is unknown.

In all, there is strong evidence linking oxidative stress to β cell dysfunction and apoptosis. However, there are many potential mechanisms that may contribute to

ROS accumulation during T2D pathogenesis and the relative contribution of these processes to β cell demise is unclear. Thus, advances in our understanding of how β cells adapt to environmental conditions associated with oxidative stress may provide new therapeutic opportunities for T2D.

1.2.2 ER stress

Another instance of homeostasis that is critical for cell viability is the balance between the abundance of nascent peptides and the protein folding capacity of the endoplasmic reticulum (ER). If protein synthesis exceeds the folding capacity of the ER, there will be an accumulation of unfolded proteins, termed ER stress.

Prolonged ER stress can lead to protein aggregation, cellular dysfunction, and apoptosis (Hetz, 2012). Since a major function of β cells is to synthesize and process proinsulin polypeptides, the balance between protein synthesis and folding is particularly important in these cells. In T2D, ER homeostasis is disrupted when insulin resistance causes the demand for insulin synthesis to overwhelm the folding capacity of β cells (Evans Molina et al., 2013).

The response to ER stress consists of a highly conserved set of processes known as the unfolded protein response (UPR). The presence of unfolded proteins in the ER activate three distinct branches of the UPR: IRE1α/XBP1,

PERK, and ATF6. Together, these factors activate a cellular program that initially aims to restore protein folding homeostasis by increasing expression of ER chaperones and degrading unfolded proteins (Figure 1.3). Chronic activation of the UPR, however, promotes cell death via pro-apoptotic factors, such as CHOP

(Hetz et al., 2013).

Figure 1.3. The unfolded protein response (UPR) shapes cellular adaptation to ER stress. Adapted from (Navid and Colbert, 2017).

Pancreata from T2D patients show evidence of ER stress in β cells, including induction of CHOP and dysregulation of the UPR mediators ATF6 and XBP1

(Engin et al., 2014; Huang et al., 2007). Mutations in the gene EIF2AK3, which encodes PERK, leads to permanent neonatal diabetes due to a reduction in the number of β cells (Delépine et al., 2000). Furthermore, culturing islets under glucolipotoxic conditions leads to ER stress, UPR activation, and apoptosis

(Laybutt et al., 2007). On the other hand, genetic deletion of CHOP rescues β cell survival and function in several mouse models of T2D (Song et al., 2008).

Thus, chronic activation of the UPR occurs in β cells during metabolic stress, leading to cellular dysfunction and apoptosis.

Beyond the canonical components of the UPR, β cell-specific factors may also shape the ER stress response to fit the particular needs of this cell type. For example, the transcription factor PDX1 is required for pancreas organogenesis and maintenance of cell identity in mature β cells (Gao et al., 2014; Jonsson et al., 1994; Stoffers et al., 1997a), and specific heterozygous mutations in PDX1 cause monogenic diabetes (Stoffers et al., 1997b). Furthermore, mice with heterozygous loss of PDX1 have defects in β cell compensation during high fat feeding, which is associated with increased ER stress and apoptosis in β cells.

PDX1 promotes adaptation to a high fat diet by regulating the expression of genes involved in ER function and the UPR (Sachdeva et al., 2009). This indicates that PDX1 enhances β cell resiliency during ER stress by shaping an adaptive gene regulatory network. 16

This example also serves to illustrate a general principle that deserves further elaboration. The identification of PDX1 as a key regulator of β cell adaptation to stress implies that a useful approach to better understand the mechanisms underlying the stress response in β cells is to closely scrutinize the network of genes that become dysregulated upon depletion of PDX1. This approach holds promise to advance our understanding of how β cells adapt to stress, how this adaptation may fail in disease, and what targets may be candidates for therapeutic intervention.

1.3 Translational regulation and cellular stress

A critical and prominent part of stress responses across cell types is the regulation of mRNA translation, which allows for rapid changes in gene expression to restore homeostasis. For example, one component of the UPR is increased translation of the mRNA encoding ATF4, a transcription factor that upregulates genes involved in protein folding (Harding et al., 2000; Lu et al.,

2004). Further, genome-wide assessment of ribosome occupancy, a proxy for translational efficiency, shows a striking reprogramming of translation for hundreds of genes during ER stress and heat shock (Sidrauski et al., 2015;

Ventoso et al., 2012; Zhou et al., 2015). Since translational regulation may not be accompanied by changes in mRNA abundance, common approaches such as

RNA-seq will not detect this layer of gene regulation. As such, translational regulation represents an essential, yet understudied, mechanism of cellular adaptation during stress.

β cells are highly secretory cells that require tight control of protein synthesis for proper functioning (Evans Molina et al., 2013; Scheuner et al., 2005). For example, the acute adaptation to high levels of glucose includes increased translation of the mRNA encoding proinsulin (Itoh and Okamoto, 1980).

Furthermore, excess translation rates during glucolipotoxicity have been linked to the onset of ER stress (Hatanaka et al., 2014). However, the mechanisms underlying translational regulation in β cells during stress conditions are poorly understood.

1.3.1 Overview of translation initiation

Translation initiation consists of a highly regulated, multi-step process that is generally considered the rate-limiting step of translation (Parsyan et al., 2011), although regulation can occur at the level of translation elongation in some circumstances (Hussey et al., 2011). The process of translation initiation requires recruitment of the 43S pre-initiation complex (PIC) to the 5’ end of the mRNA, scanning of the PIC to a start codon, and formation of a competent ribosomal complex for elongation (Figure 1.2). For most mRNAs, this process is dependent on the presence of two particular RNA modifications that are added post- transcriptionally: a 7-methylguanylate (m7G) cap at the 5’ end and a stretch of adenosine monophosphates at the 3’ end (poly(A) tail). Upon export to the cytoplasm, mRNA is bound at the 5’ cap by the eukaryotic translation initiation factor 4E (eIF4E), which is part of a heterotrimeric complex called eIF4F that also includes a scaffold protein eIF4G and an RNA helicase eIF4A. Separately, the

18 small ribosomal subunit (40S) is decorated by many additional translation initiation factors, including eIF2, the initiator tRNA, and the multi-subunit factor eIF3, and this large complex is collectively called the 43S PIC. Interaction between eIF4G and eIF3 facilitates the recruitment of the 43S PIC to the mRNA

5’ cap, after which the 43S PIC scans along the mRNA until a start codon is encountered. This is followed by the assembly of an 80S ribosomal complex, which requires GTP hydrolysis by eIF2 and recruitment of the 60S ribosomal subunit (Jackson et al., 2010).

Figure 1.4. Translation initiation is a complex, multi-step process. Adapted from (Parsyan et al., 2011).

In addition to the assembly of translation machinery at the 5’ end of mRNAs, efficient translation also depends upon factors binding to the 3’ end, most importantly the poly(A)-binding protein (PABP). This factor binds to the poly(A) tail and eIF4G to facilitate RNA circularization. These interactions are thought to increase the efficiency of translation by enhancing binding of eIF4F to the 5’ cap and facilitating the re-initiation of ribosomal subunits at start codons after termination, termed ribosome recycling (Dever and Green, 2012).

While these steps encompass translation initiation for most mRNAs, certain properties can alter the requirements for initiation to occur. For example, some mRNAs contain a highly structured 5’ UTR that inhibits scanning of the PIC. In this case, efficient translation initiation requires supplementary factors, such as additional RNA helicases, to unwind the RNA secondary structures and promote scanning (Marintchev, 2013). Besides providing structural blockades for translation, RNA secondary structure in 5’ UTRs can also influence the function of initiation factors. For example, eIF3 generally promotes translation initiation via recruitment of the PIC, however, certain mRNAs contain 5’ UTR structural elements that cause eIF3 to act as a translational repressor (Lee et al., 2015).

A useful paradigm for studying atypical modes of translation initiation is the translation of viral mRNAs. Viruses are completely dependent on the host translation machinery for their replication and propagation. Counterintuitively, however, many viruses incapacitate the translation machinery of infected cells to shut-off translation of host mRNAs. This is achieved by a broad range of 21 mechanisms, including expression of a viral protease that cleaves eIF4G

(poliovirus), reducing eIF4E expression levels (enterovirus 71), sequestration of eIF3 via viral proteins (measles virus), and cleavage of PABP by viral proteases

(retroviruses) (Walsh et al., 2013). While this impairment of translation cripples the host’s defense mechanisms, it also presents the conundrum of how to translate viral mRNAs. For this reason, viruses have evolved various mechanisms to initiate translation using host machinery in a cap-independent manner. One prominent example of this phenomenon is the presence of an internal ribosomal entry site (IRES) near the 5’ end of viral mRNAs. IRESs are highly structured RNA elements that recruit the 40S ribosomal subunit to mRNA in a cap-independent manner. There are several different classes of IRESs, each of which require a distinct subset of trans-acting factors, including translation initiation factors, RNA binding proteins, and viral proteins, for efficient translation

(Sweeney et al., 2013). For example, the IRES in poliovirus mRNA requires eIF4F and eIF3 for recruitment of the 40S ribosomal subunit, whereas hepatitis C virus mRNA requires eIF3 but not eIF4F and cricket paralysis virus mRNA does not require either of these factors (Komar et al., 2012). Importantly, many of these mechanisms for alternative initiation are thought to be pirated from their hosts, meaning that certain host mRNAs also employ these mechanisms under certain conditions. For example, the mRNA encoding c-MYC contains an IRES element, which confers cap-independent translation that is stimulated by the

RNA binding protein hnRNPK (Evans et al., 2003). The broad range of mechanisms employed by viruses for translation initiation raises the intriguing 22 possibility that we have only just begun to understand the usage of these processes for translational regulation of endogenous mRNAs.

Indeed, the multi-faceted process of translation initiation is regulated at various steps during stress conditions, allowing for the tight control of both global translation rates and selective translation of key mRNAs. Below, we outline several examples of translational regulation during stress that influence cellular homeostasis.

1.3.2 Regulation of global translation rates

Protein synthesis consumes a significant amount of cellular energy, thus dampening global translation rates during stress conserves cellular resources and promotes cell survival (Evans Molina et al., 2013). For example, the UPR influences global translation rates by several mechanisms. First, activated PERK phosphorylates eIF2α, which reduces its function and suppresses translation initiation (Hetz, 2012). This regulation is critical for β cell adaptation to stress in vivo as mutation of the phosphorylated site of eIF2α in mice led to the accumulation of unfolded proteins and defective insulin secretion during high fat feeding (Scheuner et al., 2005). Second, 4E-BP1, which reduces global translation rates by sequestering eIF4E, is upregulated by ATF4 and ATF5 during ER stress (Juliana et al., 2017; Pause et al., 1994; Spriggs et al., 2010).

Loss of 4E-BP1 in β cells leads to increased apoptosis during conditions of ER stress due to unchecked translation rates (Yamaguchi et al., 2008). Thus,

23 dampening translation is critical for the adaptation of β cells to ER stress because it reduces the protein folding burden and conserves cellular resources.

While we have highlighted the importance of regulating general protein synthesis during stress, it should be noted that control of this process is also essential for cell growth. For example, the mTOR signaling pathway integrates growth signals with downstream processes required for proliferation, including cap-dependent translation. Specifically, mTOR increases global translation rates by activation of the ribosomal protein S6K and inactivation of 4E-BPs, thereby increasing the availability of eIF4E for translation initiation (Mamane et al., 2006). This increase in translation allows for the synthesis of proteins to accommodate the expansion of cell size and the formation of new cells.

1.3.3 Selective regulation of translation

While the regulation of global translation rates promotes recovery from stress by conserving cellular resources and reducing the protein folding burden, the selective translation of mRNAs impacts cellular homeostasis indirectly by altering the gene expression program of the cell. This mechanism for altering gene expression may be advantageous over transcriptional regulation because it occurs downstream of transcription and thus can very quickly alter protein levels

(Sidrauski et al., 2015).

In order for selective translation to occur, this directive must be encoded in the mRNA, which can be achieved by three general mechanisms. First, selective

24 translation can be governed by the presence of regulatory sequences in the mRNA that are bound by sequence-specific factors that regulate translation, such as RNA binding proteins (RBPs) or microRNAs. Second, mRNAs can be translationally regulated based on the presence of RNA secondary structures rather than primary sequence elements. Lastly, upstream open reading frames

(uORFs) in 5’UTRs can affect the re-assembly of ribosome complexes on downstream canonical initiation codons (Jackson et al., 2010). The translational induction of genes containing uORFs, such as ATF4, ATF5, and CHOP, has been well characterized to be a consequence of eIF2α phosphorylation by PERK during ER stress (Somers et al., 2013). However, the extent to which primary sequence or secondary structural elements shape translational regulation during stress is less known.

Importantly, these regulatory elements may be present in a cohort of genes that are functionally related, allowing for the coordinated re-shaping of gene expression towards a specific purpose. This notion forms the basis of the “RNA- operon” theory, which postulates that trans-acting factors, such as RBPs, may integrate environmental signals with the post-transcriptional regulation of specific genes that share a common function (Keene, 2007). This model is especially intriguing in the context of β cell adaptation to pathophysiologically-relevant conditions; however, the role of RBPs in shaping the β cell gene expression program in response to external cues is largely unknown.

RBPs regulate all aspects of post-transcriptional RNA processing, including splicing, stability, localization, and translation. In fact, mRNAs are generally bound by a broad array of RBPs throughout their life cycle, positioning these factors to rapidly integrate cellular signals with changes in gene expression. The target specificity of an RBP is directed by its RNA-binding domain, which interacts with RNA in a sequence- and structure-dependent manner (Lukong et al., 2008). It is estimated that there are anywhere between 500 and 1500 RBPs in humans containing more than 40 different types of RNA-binding domains, underscoring the broad importance of these proteins in biological processes

(Gerstberger et al., 2014). For example, the most common RNA-binding domain, the RNA recognition motif (RRM), is found in more than half of all RBPs and is a highly plastic domain that can bind RNA with both high-affinity and high- specificity (Maris et al., 2005).

RBP-mediated post-transcriptional regulation is exemplified in β cells by the role of the polypyrimidine tract-binding protein (PTB) in adaptation to hyperglycemia.

Part of the β cell response to high glucose levels is the upregulation of genes involved in insulin processing and secretory granule biogenesis, and many of these genes were found to be upregulated at the level of mRNA stability, not transcription (Knoch et al., 2004; Martin et al., 1994). Interestingly, loss of PTB blocked the post-transcriptional induction of these genes and significantly reduced insulin granule biogenesis (Knoch et al., 2004). Thus, by coordinating the post-transcriptional regulation of genes with a common function, constituting

26 an RNA operon, PTB augments β cell insulin secretory capacity during high glucose conditions. It is enticing to speculate that this example is just the tip of the iceberg for RBPs mediating adaptive mechanisms in β cell. Therefore, a systematic evaluation of post-transcriptional changes in β cells, and the RBPs governing these changes, warrants close examination.

1.3.4 Translational regulation mediated by the PCBPs

The poly(C)-binding protein (PCBP) family consists of five members: hnRNPK and PCBP1-4. These factors all contain three K homology (KH) domains, which provide the specific targeting of cytosine-rich sequences (Makeyev and

Liebhaber, 2002). The PCBP members are highly multifunctional RBPs involved in many aspects of gene regulation, including transcription, splicing, alternative polyadenylation, mRNA stability, and translation (Bomsztyk et al., 2004; Ji et al.,

2013; 2016; Makeyev and Liebhaber, 2002; Ostareck-Lederer et al., 2002;

Waggoner et al., 2009). Their impact on gene expression is likely determined by a host of factors, including their subcellular localization, post-translational modifications, and interacting co-factors (Bomsztyk et al., 2004). The members hnRNPK and PCBP1-2 are ubiquitously expressed and play critical roles in a range of cellular processes, some of which are cell type-specific (Makeyev and

Liebhaber, 2002). For example, PCBP1-2 promote the differentiation of functional erythrocytes by binding to the 3’UTR of α-globin mRNA and significantly stabilizing this transcript (Kiledjian et al., 1995). These factors are also regulated extensively by post-translational modifications, especially phosphorylation. This 27 allows for the function of these proteins, including their impact on mRNA translation, to be modulated downstream of various signaling pathways

(Chaudhury et al., 2010a).

The first example of translational regulation mediated by the PCBPs was the translational silencing of the mRNA encoding erythroid 15-lipoxygenase. This enzyme is responsible for the breakdown of phospholipids during late stages of erythrocyte maturation and must be silenced at earlier stages. Interestingly, 15- lipoxygenase mRNA is abundantly expressed throughout the stages of differentiation. Rather, its temporally restricted expression pattern is entirely dependent on translational silencing via a large cytosine-rich stretch in the 3’

UTR. hnRNPK and PCBP1 were identified as the factors binding to this regulatory element and inhibiting translation by preventing 60S ribosomal subunit joining (Ostareck et al., 2001; 1997). Curiously, translation of the mRNA encoding c-Src kinase is also translationally silenced in erythroid progenitor cells by hnRNPK, but not PCBP1 (Naarmann et al., 2008). Thus, three different mRNAs expressed in erythroid cells, 15-lipoxygenase, c-Src, and α-globin, are post-transcriptionally regulated by members of the PCBP family via binding to C- rich sequences in the 3’UTR. Despite these similarities, the regulation of these mRNAs are distinct based on the mode of regulation (RNA stability or translation) and on the PCBP family members imparting this regulation. It is unclear how these RBPs provide distinct functions depending on their target mRNA but likely depends on the presence of additional cofactors.

Along these lines, the translational regulation provided by PCBP family members can be either translational enhancement or repression. Like 15-lipoxygenase and c-Src, hnRNPK represses translation of the mRNAs encoding transforming growth factor-β-activated kinase 1 (TAK1) and the androgen receptor (Liepelt et al., 2014; Mukhopadhyay et al., 2009). On the other hand, hnRNPK has been implicated in the translational enhancement of the mRNAs encoding Bruton’s tyrosine kinase (p65BTK), myelin basic protein (MBP), neurofilament medium

(NF-M), vascular endothelial growth factor (VEGF), and c-MYC (Evans et al.,

2003; Grassilli et al., 2016; Hutchins and Szaro, 2013; Laursen et al., 2011;

Sataranatarajan et al., 2008). PCBP1 and PCBP2 have also been found to either stimulate or repress translation depending on the target mRNA (Chaudhury et al.,

2010b; Eiring et al., 2010; Evans et al., 2003; Laursen et al., 2011; Wang et al.,

2010).

Furthermore, translational regulation mediated by the PCBPs can be modified by phosphorylation of these RBPs in response to external cues. For example, the mRNAs encoding DAB2 and ILEI, which are critical for epithelial-mesenchymal transdifferentiation (EMT), are translationally repressed by PCBP1 via 3’UTR binding. Treatment of cells with TGF-β, however, leads to PCBP1 phosphorylation, reduced PCBP1 binding to the 3’UTRs, and enhanced translation of DAB2 and ILEI to promote EMT (Chaudhury et al., 2010b).

Similarly, phosphorylation of hnRNPK can lead to changes in its subcellular localization and RNA binding affinity, both of which can impact target mRNA

29 translation (Habelhah et al., 2001; Ostareck-Lederer et al., 2002; Ostrowski et al.,

2001). hnRNPK likely impacts translation by interacting with cofactors that either repress or promote translation initiation (Bomsztyk et al., 2004), and these interactions may also be modulated by phosphorylation. For example, phosphorylation of hnRNPK by JNK leads to enhanced translation of cytoskeletal

RNAs required for axon outgrowth in developing neurons of Xenopus laevis. This effect on hnRNPK was not attributable to a change in its subcellular localization nor a change in its RNA binding affinity. Instead, it was suggested that JNK signaling altered the interaction of hnRNPK with translation machinery, however, the precise mechanisms by which this occurred were not elucidated (Hutchins and Szaro, 2013).

In the work described herein, we investigate translational regulation and the mechanisms controlling these processes in the context of β cell dysfunction. We explore the role of the PCBPs in β cells, including their post-transcriptional control of gene expression during stress conditions. Together, these studies aim to uncover novel mechanisms governing β cell adaptation to stress that may pertain to the pathogenesis of T2D.

CHAPTER 2: TRAP-SEQ UNCOVERS POST-TRANSCRIPTIONAL

REGULATION MEDIATED BY THE PCBPS IN β CELLS

2.1 Summary

Translational regulation is a highly conserved and critical component of stress responses, however, the mechanisms underlying this process in β cells are not fully understood. Here we used TRAP-seq to uncover a subset of genes with changes in ribosome occupancy in Min6 cells with PDX1 deficiency, including genes encoding the transcription factors NKX2-2 and JUND. De novo motif analysis identified enrichment of a poly(C) motif in the 3’UTRs of genes with increased ribosome occupancy, suggesting a functional role for the poly(C)- binding protein (PCBP) family of RNA binding proteins in β cells. The PCBP members PCBP1, PCBP2, and hnRNPK all bound to the mRNAs encoding

NKX2-2 and JUND, and loss of PCBP1/2 led to a post-transcriptional reduction in these genes. Additionally, glucose-stimulated insulin secretion was impaired in

Min6 cells with depletion of PCBP1/2 or hnRNPK, suggesting that these factors are functionally important in β cells.

2.2 Introduction

The specialization of cells is bestowed largely by the distinct catalog of proteins present in a given cell type, yet all cells of an organism contain the same genetic information. Thus, mechanisms to selectively decode genetic information are crucial for the functional specificity of cells. Advances in our understanding of the cellular controls that decode genetic information, and how they become dysfunctional in disease states, will be a central step towards unlocking the next generation of medical therapeutics. For example, we are on the cusp of treating an array of diseases with cell-based therapies. The advancement and expansion of these approaches, however, will rely on a better understanding of the genetic controls shaping cell identity and how environmental cues impact these processes.

The decoding of genetic information for the synthesis of proteins is dynamic and requires a full cycle of events to take place: transcription of DNA into RNA, processing RNA to mRNA, translation of mRNA to protein, and degradation of mRNA/protein. Importantly, all of these steps represent potential regulatory points to shape the gene expression profile of a cell. Due to rapid advances in the depth and accuracy of DNA sequencing technologies, however, it is commonplace to use mRNA abundance as a proxy for the functional output of genetic information. This approach is certainly useful when focusing on transcriptional regulators, but it will miss regulation at the level of mRNA translation or protein degradation. This has led to a bias in our understanding of

32 gene regulation, including in β cells: the factors shaping transcriptional regulation are much better understood than those regulating mRNA translation.

Translational regulation is particularly important in the context of dynamic shifts in gene expression required for adaptation to environmental conditions, external stimuli, or stress conditions. Since translation is downstream of transcription, a cell can respond to a stimulus more quickly by increasing translation of an mRNA compared to increasing its transcription. The regulation of translation can be grouped into two broad categories: regulation of global translation rates and selective translational regulation of specific mRNAs. Tight control of global translation rates is critical for cells because protein synthesis is associated with high cellular energy consumption. A common mechanism for this type of translational regulation is post-translational modification of a component of the translation machinery. For example, mTOR signaling integrates growth signals with translation rates by phosphorylating eukaryotic translation initiation factor

4E-binding protein 1 (4E-BP1) and ribosomal protein S6K1/2 (Mamane et al.,

2006).

Selective regulation of translation, on the other hand, allows for rapid changes in gene regulatory networks that shape the cellular response to various conditions.

This mechanism for translational regulation seems to be particularly important in the context of cellular stress responses. For example, during ER stress there is increased translation of select mRNAs with upstream open reading frames

(uORFs), such as ATF4. This response promotes restoration of ER homeostasis 33 by increasing the protein folding capacity of the ER (Harding et al., 2000).

Similarly, inhibition of cap-dependent translation occurs under various stress conditions, such as during infection of cells with picornaviruses due to cleavage of eIF4G by a viral protease or during ER stress due to upregulation of 4E-BP1.

These alterations of translational machinery promote the cap-independent translation of specific mRNAs containing internal ribosomal entry sites (IRESs) while global translation rates are suppressed (Spriggs et al., 2008). For example, the transcript encoding the chaperone BiP was the first eukaryotic mRNA found to contain an IRES, which allows for the translation of this factor during ER stress to enhance cellular protein folding capacity (Macejak and Sarnow, 1991).

Translation rates of mRNAs can be directly measured by pulse labeling followed by mass spectrometry, however, this approach is technically challenging and requires large amounts of input material. Accordingly, alternatives to this method have been developed that determine the relative density of ribosomes bound to mRNAs, which can be assessed on a genome-wide scale with RNA-seq (Kapeli and Yeo, 2012). The use of these approaches to assess translational regulation, however, relies on several assumptions because they are indirect measurements of translation. Nevertheless, an understanding of these assumptions and careful scrutiny of the obtained results will allow for the successful implementation of these methodologies. For example, an increase in ribosome binding to an mRNA cannot distinguish between actively translating or stalled ribosomes. As such, an observed change in mRNA ribosome occupancy should be confirmed to cause

34 an appropriate change in protein level by western blot analysis. Additionally, ribosome run-off assays can be used to rule out the presence of stalled ribosomes (Pelechano et al., 2015). It should also be noted that these methodologies are useful for detecting changes in translation initiation, which is thought to be the major regulatory step of translation, but not for changes in translation elongation or termination. Instead, the study of translational regulation occurring at the elongation step requires specific methodologies to assess the kinetics of ribosome run-off (Richter and Coller, 2015).

The first methodology that used changes in ribosome occupancy to assess translation was polysome profiling. This method separates mRNAs into different fractions based on the relative abundance of bound ribosomes using sucrose gradient sedimentation. For example, during ER stress, the mRNA encoding

ATF4 shifts towards greater representation in the polysome fraction compared to the monosome fraction, which indicates an increase in ribosome binding during stress (Hatanaka et al., 2014). However, this methodology has several disadvantages including the requirement for significant starting material and the inability to distinguish between cell types in a heterogeneous tissue.

Furthermore, this method uses density gradient centrifugation to separate mRNAs into polysome and monosome populations. This assumes that the main factor contributing to the density of an mRNA is the number of bound ribosomes.

It is unclear if this assumption is a valid one or if large RNP complexes other than ribosomes may contribute to mRNA density (Kuersten et al., 2013).

To circumvent some of these disadvantages, methods that rely on the immunoprecipitation of tagged ribosomal subunits have also been developed, such as Translating Ribosome Affinity Purification (TRAP, GFP-tagged RPL10a) and RiboTag (HA-tagged RPL22) (Heiman et al., 2008; Sanz et al., 2009). In these methods, comparison of transcript abundance in the immunoprecipitated

RNA fraction to that in total RNA can be used to determine the ribosome occupancy of mRNAs, an approximation of translational efficiency. By expressing these transgenes from a cell type-specific promoter, ribosome-bound mRNA can be purified for a specific subpopulation of cells within a heterogeneous tissue.

Therefore, these approaches are particularly useful when attempting to assess translational regulation in vivo because most tissues are composed of a complex mixture of cell types. Also, these methods use a highly specific immunoprecipitation step to purify ribosomes and their associated mRNAs and do not rely on the assumption that ribosomes are the main determinant of mRNA density.

One disadvantage to all of the methodologies mentioned above is that they involve sequencing of full-length mRNA. For this reason, the positioning of ribosomes along mRNAs cannot be discerned by these approaches. In contrast, ribosome profiling, which is the most recent advance in studying translation, utilizes an RNase digestion step to generate ribosome-protected RNA fragments.

High-throughput sequencing of these footprints can then be used to determine not only the density of ribosomes binding to an mRNA, but also the location of

36 ribosomes along the transcript, including the relative usage of start codons

(Ingolia et al., 2009; 2011). This allows for the study of alternative translation initiation, which is particularly interesting given that initiation from non-AUG codons may be a common mechanism to increase protein diversity and can be regulated under stress conditions (Touriol et al., 2012).

2.3 Materials and Methods

Cell line culture

Min6 mouse insulinoma cells passage 20-30 were cultured in high glucose

DMEM as described (Claiborn et al., 2010), unless otherwise noted. For siRNA- mediated depletion of Pdx1, cells were nucleofected by AMAXA with siRNA for

Pdx1 (Dharmacon L-040402-01) or non-targeting control (Dharmacon D-001810-

10) and collected 72hrs post-transfection. For lentiviral infections, Min6 cells were transduced for 6 hours with virus and polybrene (Sigma) at 8ug/mL. Cells were allowed to recover for 4-5 days before collection or stress treatments.

HEK293T cells were cultured in DMEM containing 25mM glucose.

GFP-RPL10A Min6 stable cell line

The GFP-RPL10A transgene was generated by cloning PCR amplified fragments for GFP-RPL10A or GFP into the pBABE-puro retroviral vector (Morgenstern and

Land, 1990) digested with SalI. Retrovirus was produced in HEK293T cells and added to Min6 cells, followed by two rounds of puromycin selection (5 days,

2ug/mL).

Glucose-stimulated insulin secretion in Min6 cells

Min6 cells were seeded in a 12 well plate 3 days prior to GSIS experiment

(~400,000 cells/well). On day of experiment, fresh KRBH buffer (15mM HEPES,

120mM NaCl, 4.7mM KCl, 1.2mM MgSO4, 1.2mM KH2PO4, 20mM NaHCO3,

2mM CaCl2, 0.01% BSA) was prepared and the pH was adjusted to 7.4. Glucose 38 was dissolved in KRBH to obtain solutions with glucose concentrations of 2.5mM

(low glucose) and 25mM (high glucose).

Wells were washed with PBS and KRBH (no glucose). 500ul of glucose-free

KRBH was added to each well and plates were incubated for 1 hour at 37C.

Wells were washed once with glucose-free KRBH and 500ul of 2.5mM glucose

KRBH was added to each well, then incubated at 37C for 30 minutes. 100ul of supernatant was transferred to tubes and placed on ice. The remaining KRBH was removed from each well and one wash with glucose-free KRBH was performed. 500ul of 25mM glucose KRBH was added to each well, then incubated at 37C for 30 minutes. 100ul of supernatant was transferred to tubes and placed on ice. The remaining KRBH was removed, wells were washed twice with cold PBS, and cells were harvested with 100ul lysis buffer containing protease inhibitor. LG and HG supernatant samples were centrifuged at 1500g for 5 min at 4C. 50ul of supernatant was taken and froze at -20C along with cell lysates. Insulin levels in LG, HG, and cell lysate samples were measured by

ELISA (ALPCO, 80-INSMSU-E01), as per manufacturer instructions.

RNA isolation and RT-qPCR

For Min6, cells were washed 2X with cold PBS before addition of TRIZOL

(Invitrogen) and RNA was extracted according to manufacturer’s instructions.

RNA was reverse transcribed with random hexamers using High Capacity cDNA

Reverse Transcription Kit (Applied Biosystems). Quantitative PCR (BioRad

CFX384) was used to measure transcript abundance and normalized to HPRT.

See Table 1 for a list of primer sequences used for these analyses.

Western blot

Proteins were separated by SDS-PAGE and immunoblotted with the following antibodies: mouse anti-NKX2-2 (Hybridoma Bank, 74.5A5), rabbit anti-JUND

(Santa Cruz, sc-74), rabbit anti-PCBP1 (Liebhaber lab), rabbit anti-PCBP2

(Liebhaber lab), rabbit anti-hnRNPK (Bethyl Laboratories, A300-674A), mouse anti-Tubulin (Sigma, T9026), mouse anti-Ran (B.D. 610340), goat anti-GFP

(Abcam, 6673), mouse anti-RPL10A (Novus, 3G2), rabbit anti-RPL7 (Novus,

NB100-2269), and rabbit anti-RPS6 (Abcam, ab40820).

Co-immunoprecipitation

Min6 cells were lysed in buffer containing 20mM Tris pH 8.0, 137 mM NaCl, 1mM

MgCl2, 1mM CaCl2, 1% NP-40, 10% glycerol, protease inhibitor cocktail

(Millipore), and Benzonase (Sigma) at 12.5 U/mL. Lysates were rotated at 4°C for 1 hour. Protein concentration was determined using a Micro BCA Protein

Assay Kit (Thermo). 1ug of primary antibody was added to lysate encompassing

500ug of protein and incubated overnight at 4°C. Protein A Dynabeads were washed 3 times in lysis buffer then resuspended in lysate/antibody mixture and incubated for 3 hours at 4°C. The immunoprecipitations were washed 4 times with lysis buffer then eluted in NuPAGE LDS Sample Buffer (Invitrogen) by

40 heating at 70°C for 10 minutes. Eluted proteins and input samples were analyzed by western blot.

TRAP

TRAP was performed as described (Heiman et al., 2008), with minor modifications. Briefly, after stress treatments of GFP-RPL10A Min6 cells, cycloheximide (Sigma) was added to the culture media at 100ug/mL for 10 minutes prior to washing 2X with cold PBS. Cells were lysed and protein concentration was measured by BCA (Thermo). Total RNA was isolated (1-5% input) with the RNeasy Mini Kit (Qiagen). For IP RNA, cell lysates encompassing

200ug of protein were added to Protein G Dynabeads bound to GFP antibodies

(19C8 and 19F7, Memorial Sloan-Kettering Monoclonal Antibody Facility) and incubated overnight at 4°C. The next day, the beads were washed 4X with high salt buffer, as described8. IP RNA was eluted from beads in RLT buffer and extracted using the RNeasy Mini Kit. Ribosome occupancy was determined by dividing transcript abundance for each gene in the IP RNA fraction by its level in the Total RNA fraction. For TRAP followed by RT-qPCR, RNA was reverse transcribed with random hexamers and Superscript III (Invitrogen) and transcript abundance was first normalized to HPRT for each fraction before determining ribosome occupancy.

RNA immunoprecipitation

RNA immunoprecipitation was performed using rabbit anti-hnRNPK (Bethyl

Laboratories, A300-674A) or rabbit anti-PCBP1/2 antibodies (Liebhaber lab).

Antibodies or IgG were bound to Protein A Dynabeads in buffer containing 20mM

HEPES pH 7.4, 5mM MgCl2, 150mM KCl, 2mM DTT, 1% NP-40. Min6 cells were lysed in buffer containing 50mM Tris pH 8.0, 150mM NaCl, and 1% NP-40 with

RNase, protease, and phosphatase inhibitors. Total RNA was extracted from lysate (10% input) using RNeasy Mini Kit (Qiagen). Lysate encompassing 75ug of protein was added to beads bound to antibody for hnRNPK or PCBP1/2 and an IgG control, then incubated overnight at 4°C. The next day, IPs were washed with buffer containing 20mM HEPES pH 7.4, 5mM MgCl2, 350mM KCl, 2mM

DTT, 1% NP-40, and RNase inhibitors. RNA was eluted from beads in RLT buffer

(Qiagen) and RNA was isolated using RNeasy Mini Kit (Qiagen). For RIP followed by RT-qPCR, enrichment was calculated by first normalizing transcript abundance in IP RNA to that in total RNA and then to the IgG control.

CRISPR design and cloning

CRISPR gRNAs were designed using http://crispr.mit.edu/ to minimize off-target binding (ROSA26: AAGATGGGCGGGAGTCTTCT, hnRNPK:

GTTTAATACTTACGTCTGTA, PCBP1: TCGGCTGCTGATGCACGGAA,

PCBP2: GACACCGGTGTGATTGAAGG). gRNAs were cloned into lentiCRISPRv2, as described (Sanjana et al., 2014).

RNA-seq and analysis

RNA was isolated using the RNeasy Mini Kit (Qiagen). Libraries were prepared using NEB Next Ultra RNA Library Prep Kit according to manufacturer’s instructions with polyA enrichment followed by paired-end sequencing of 150bp using HiSeq (Illumina). Reads were mapped to mm10 using TopHat2 (Kim et al.,

2013) and read counts per gene were determined using featureCounts (Liao et al., 2014). For TRAP-seq analysis, ribosome occupancy was calculated by dividing normalized gene counts in the IP RNA samples by that in the total RNA samples. Differential expression analysis was performed using edgeR (Robinson et al., 2010) with significant genes called using fold-change cutoffs of greater than 1.5 or less than -1.5 and FDR less than 0.05. Gene ontology analysis was performed using DAVID (Huang et al., 2009). For RIP-seq analysis, edgeR was used to determine enrichment in the hnRNPK immunoprecipitated RNA compared to total RNA with significant genes called using a fold-change cutoff of greater than 1.5 and FDR less than 1x10-4. An IgG control was performed but not sequenced because no RNA was detected in the pull-down by Qubit RNA High

Sensitivity quantitation. The overlap between RNA-seq data sets was determined using hypergeometric tests.

Statistics

Data are presented as mean ± standard error of the mean unless otherwise noted in figure legends. Statistical analyses were performed using GraphPad

PRISM 7 software. Statistical tests used are noted in figure legends and include

43 unpaired two-tailed Student’s t-test, one-way analysis of variance (ANOVA), two- way ANOVA, and hypergeometric test.

Motif analysis

De novo motif discovery was performed using the MEME software suite (Bailey et al., 2009) with the following parameters: strand-specific search, motif width 4-

10. For motif frequency comparison, control sets of genes were generated by randomly selecting 100 genes 10 times from a list of genes with detectable expression in Min6 cells but not meeting the criteria for differential ribosome occupancy with PDX1 deficiency. The FIMO tool of the MEME suite (Bailey et al.,

2009) was used to determine motif frequencies. The Tomtom tool of the MEME suite (Bailey et al., 2009) was used to compare discovered motifs to a database of motifs recognized by RBPs (Ray et al., 2014).

Gene Name Forward primer Reverse primer HPRT TGCTCGAGATGTCATGAAGGA CCAGCAGGTCAGCAAAGAACT Beta_actin gctacagcttcaccaccaca tctccagggaggaagaggat ATF4 GCAGTGTTGCTGTAACGGACA CGCTGTTCAGGAAGCTCATCT JunD TCAAGACCCTCAAAAGCCAGA CGTGGCTGAGGACTTTCTGTT hnRNPK GAAGAAACCTTCCCCAACACC CGCAATTCAACCATCTCATCA NOS2 CTGAACTTGAGCGAGGAGCAG TTGCCCCATAGGAAAAGACTG Ccl2 CAGGTGTCCCAAAGAAGCTGT ATTTGGTTCCGATCCAGGTTT Ptgs2 CCTCCTGGAACATGGACTCAC GCTTGTACAGCAATTGGCACA Steap4 TGCCATCAGTAAGCAACATGG TACACCAAAGTGTGGGCTGTG Cxcl1 TCCAGAGCTTGAAGGTGTTGC TTCTGAACCAAGGGAGCTTCA Cxcl2 TGCCAAGGGTTGACTTCAAGA AACTTTTTGACCGCCCTTGAG Lrrtm2 ACTGAATGCAGCCTCCAATGT AGGGCAGGCAGCATTTTTAAT Lamc2 GGGCAATGCCACTTTTTATGA TCTCTTCATGGCCTCTTCAGC Crispld2 TTGGGCTCCTGTGTATGGAAC ACATCTGCATAGCCACCAACC Inhba GATCATCACCTTTGCCGAGTC TTCTGCACGCTCCACTACTGA GAPDH aggccggtgctgagtatgtc tgcctgcttcaccaccttct SOD2 GCCTACGTGAACAATCTCAACG TTGAACTTCAGTGCAGGCTGA SOD3 GAGTCCAGCTTCGACCTAGCA CTCCATCCAGATCTCCAGCAC ALDH2 GATTGGCGGATCTCATTGAAC TTCAGGACCATGTCCAAATCC GPX1 GACTGGTGGTGCTCGGTTTC ACCAGGTCGGACGTACTTGAG MGST1 CGCATTCCAGAGGATAACCAA GTTCCACCTTCTCGTCAGTGC RDH10 TGTAGACACGGGCATGTTCAG CAGTGAGGATGGCCCTCATAG CDO1 TGGGCTTTGTATGCCAAATTC ACCCCAGCACAGAATCATCAG SCD1 GCTCAGTCCCTGTTTGTTTGC CTTTGGAGGGTGGACAGACAC G6PDX AACCCCAATGGAGAAGGAGAA GAGGACAGCTGCTGCAAAAGT DHX9 CCACACAAGTTCCACAGTACATT TTTCCAGGCTCTTCTCCTCTC DDX1 AGTGGCAGATGAACCCATATGA CCTCATAGTAGTGCTTCCCTTTC DDX3X GAAAATGGAAGATATGGCCGTCG TTCAGCACCACCATAAACCACG 18S AACCCGTTGAACCCCATT CCATCCAATCGGTAGTAGCG Neat1 tcttggggccacattaatcac gaggggcaagagtagggaaga PDX1_PT TGAAATCCACCAAAGCTCACG tgaaggcagtagcagccaagt

Table 1. Primer sequences for RT-qPCR

2.4 Results

2.4.1 Establishment of TRAP in Min6 cells

In order to detect changes in mRNA translation on a genome-wide scale, we have employed the Translating Ribosome Affinity Purificaition (TRAP) methodology in which expression of a tagged ribosomal subunit (GFP-RPL10A) is used to immunoprecipitate ribosomes and their associated mRNA (Doyle et al.,

2008). This approach can be used to detect changes in ribosome occupancy, or the density of ribosomes binding to mRNA, which serves as a proxy for the efficiency of translation (Zhou et al., 2013b). To use TRAP in Min6 cells, we established a stable cell line using retroviral transduction to deliver the GFP-

RPL10A transgene. This allowed for the isolation of intact ribosomes and their associated mRNAs by GFP immunoprecipitation (Figure 2.1a-c). To confirm that this approach can detect dynamic shifts in ribosome occupancy, TRAP was performed in GFP-RPL10A Min6 cells treated with thapsigargin to induce ER stress, and ribosome occupancy was determined by normalizing transcript abundance in the ribosome pull-down fraction to that in total RNA. This gave a near doubling of ribosome occupancy for the stress-responsive factor ATF4 during thapsigargin treatment (Figure 2.1d), consistent with previous findings using polysome profiling (Guan et al., 2014; Hatanaka et al., 2014).

Figure 2.1. TRAP in GFP-L10a Min6 cells pulls down ribosome-associated mRNA and detects shifts in the ribosome occupancy of ATF4 during ER stress. a, Immunoprecipitation of GFP in Min6 cells with stable overexpression of either GFP only or the GFP-RPL10A fusion protein, as per TRAP protocol. Compared to GFP only, immunoprecipitation of GFP-RPL10A enriches for the ribosomal proteins RPL7 and RPS6. b, Comparison of input (Inp) and unbound (UB, supernatant of IP reaction) fractions shows that immunoprecipitation of GFP led to depletion of ribosomal proteins RPL10A, RPL7, and RPS6 from the UB fractions for GFP-RPL10A Min6 cells, but not for cells only expressing GFP. c, Representative bioanalyzer traces on IP RNA showing that TRAP in Min6 cells allows for the isolation of high quality, ribosome-associated RNA with overexpression of GFP-RPL10A transgene, but not GFP only. d, TRAP in GFP-RPL10a Min6 cells was used to determine ribosome occupancy after 3hr treatment with thapsigargin (Tg, 1uM) or vehicle (DMSO), followed by RT-qPCR of IP RNA and Total RNA (n=3). P values were calculated by unpaired two-tailed Student’s t-test. * = p < 0.05.

2.4.2 TRAP-seq during PDX1 deficiency

To screen for translationally regulated genes in β cells, we used deficiency of the transcription factor PDX1 as a means to disrupt normal β cell homeostasis. PDX1 is a human diabetes gene (Stoffers et al., 1997b), and reduced PDX1 levels cause β cell dysfunction and an impaired stress response (Brissova et al., 2002;

Sachdeva et al., 2009). Given these connections to human disease and β cell stress, we reasoned that PDX1 deficiency was a promising model to uncover functionally important translational controls in β cells. Indeed, TRAP-seq on GFP-

RPL10A Min6 cells with siRNA-mediated depletion of PDX1 identified 53 genes with an increase in ribosome occupancy after PDX1 depletion while 57 genes had a reduction (Figure 2.2a,b). A subset of these genes demonstrated no change in total mRNA abundance but a significant change in ribosome association (Figure 2.2c), indicating that the dominant regulatory mechanism for these genes is post-transcriptional. This list included several genes with known importance in β cells such as the transcription factor Nkx2-2, a critical regulator of β cell identity and function (Doyle and Sussel, 2007; Gutiérrez et al., 2017).

We also identified several genes that have been implicated in various stress responses in other cell types. Of particular interest was the transcription factor

JUND, which is important for redox homeostasis in other cell types (Gerald et al.,

2004; Paneni et al., 2013) but whose role in β cells is unknown.

Figure 2.2. TRAP in GFP-L10a Min6 detects shifts in ribosome occupancy after PDX1 depletion. a,b, Volcano plot depicting changes in ribosome occupancy with Pdx1 depletion (a) and comparison of the change in transcript abundance with Pdx1 depletion in the IP RNA fraction to that in the Total RNA fraction (b) as determined by TRAP-seq in GFP-RPL10a Min6 cells transfected with siRNA targeting Pdx1 or non-targeting (NT) control (n=3). Significant changes in ribosome occupancy shown in red (upregulated) or blue (downregulated). c, Heatmap showing genes identified by TRAP-seq as having significant changes in ribosome occupancy but no significant change in total RNA levels after Pdx1 depletion.

2.4.3 Post-transcriptional regulation by the poly(C)-binding proteins

To investigate the mechanism underlying the translational regulation of these genes, a de novo motif analysis was used to search for a common regulatory element within the untranslated regions (UTRs) of genes from our TRAP screen

(Figure 2.2a). This approach identified 4 motifs as occurring more frequently in the UTRs of genes with increased ribosome occupancy than expected by chance

(Figure 2.3a). However, when compared to the motif frequency (number of motifs per sequence length) in UTRs of randomly sampled control genes, two of these motifs showed no significant enrichment over controls, indicating they are generally enriched in UTRs (Figure 2.3b). In contrast, the motif frequency for the cytosine-rich motif 3 was significantly higher in the upregulated genes compared to that in both the downregulated genes and control genes, and it was specifically enriched in the 3’UTR of the upregulated genes (Figure 2.3b). 31 of the 53 genes with increased ribosome occupancy after PDX1 depletion contained at least one occurrence of motif 3 in their 3’ UTR. Further, the 3’UTRs of Jund and Nkx2-2 contain sequence elements matching this cytosine-rich motif.

Figure 2.3. Enrichment of a cytosine-rich motif in genes from TRAP-seq identified by de novo motif analysis. a, Motifs found by de novo motif analysis on UTRs of genes with increased ribosome occupancy after PDX1 depletion. b, Assessment of motif frequencies in UTRs of genes with changes in ribosome occupancy after PDX1 depletion and sets of randomly sampled control genes. Box, 25– 75th percentile; bar, median; whiskers, range. c, Comparison of Motif 3 to binding motifs for hnRNPK, PCBP1, and PCBP2, as depicted.

We hypothesized that the cytosine-rich motif 3 serves as the binding site for an

RBP that post-transcriptionally regulates these target mRNAs. Comparison of motif 3 to a database of motifs for RBPs (Ray et al., 2014) found a high similarity to the motif recognized by the poly(C)-binding protein family of RBPs, including the members PCBP1, PCBP2, and hnRNPK (Figure 2.3c), which are known regulators of mRNA translation (Bomsztyk et al., 2004; Makeyev and Liebhaber,

2002). Indeed, RNA immunoprecipitation (RIP) for PCBP1, PCBP2, or hnRNPK demonstrated that the mRNAs encoding JUND and NKX2-2 are significantly bound by these RBPs in Min6 cells compared to housekeeping genes (Figure

2.4a-c). The interaction between hnRNPK and the JUND mRNA was also confirmed in eCLIP data for HepG2 and K562 cells and was specifically localized in the 3’UTR of this transcript (The ENCODE Project Consortium et al., 2012)

(Figure 2.4d).

Figure 2.4. PCPB1, PCBP2, and hnRNPK bind to the mRNAs encoding NXK2-2 and JUND. a-c, Interaction of PCBP1 (a), PCBP2 (b), or hnRNPK (c) with the mRNAs encoding NKX2-2 and JUND compared to housekeeping controls, as determined by RNA immunoprecipitation (n=3-4). P values were calculated by a two-way ANOVA. * = p < 0.05. d, eCLIP binding peaks for hnRNPK show binding in the 3’UTR of JUND in both HepG2 and K562 cells.

To broadly examine the connection between hnRNPK and genes from our

TRAP-seq experiment, RIP for hnRNPK followed by RNA-seq (RIP-seq) was performed in Min6 cells treated with glucolipotoxicity (Figure 2.5a). As a control,

RNA-seq on input RNA was also performed. To determine enrichment for hnRNPK binding, the immunoprecipitated RNA was compared to total RNA. An

IgG control was also performed but not sequenced because no RNA was detected in this group. Comparison of RIP-seq and TRAP-seq results showed that 64% of genes with increased ribosome occupancy and at least one occurrence of motif 3, including JUND and NKX2-2, were enriched for hnRNPK binding, constituting a statistically significant overlap (Figure 2.5b, p = 1.0x10-4).

Figure 2.5. Detection of mRNAs bound by hnRNPK via RIP-seq. a, Comparison of transcript expression level to its enrichment in the hnRNPK IP over total RNA as determined by RIP-seq for hnRNPK in Min6 cells treated with glucolipotoxicity. Genes with a statistically significant enrichment in hnRNPK binding are shown in red (n=3). b, Comparison of the percentage of genes enriched for hnRNPK binding by RIP-seq. Control group is 100 sets of 100 randomly sampled genes from all expressed genes. Box, 25–75th percentile; bar, median; whiskers, range. Increased ribosome occupancy (RO) and decreased RO groups are genes determined to have significant changes in RO by TRAP-seq. Motif 3 group is the set of genes that had increased RO and contained at least one occurrence of motif 3 in the 3’ UTR.

Given the interaction between the PCBPs and the mRNAs encoding JUND and

NKX2-2, we next investigated whether loss of these RBPs would impact JUND or

NKX2-2 expression. PCBP1 and PCBP2 have high amino acid sequence homology and serve redundant roles for some biological processes. For example, co-depletion of PCBP1/2, but not depletion of either individually, leads to increased mRNA stability of CDKN1A in K562 cells (Waggoner et al., 2009).

Therefore, we targeted both PCBP1/2 in Min6 with CRISPR-Cas9, which allowed for efficient co-depletion of these factors. Interestingly, loss of PCBP1/2 led to a significant reduction in protein levels of NKX2-2 and JUND (Figure 2.6a). In contrast, the transcript level of JUND was slightly increased after loss of

PCBP1/2, indicating that the reduction in protein is not caused by reduced transcription, but rather is post-transcriptional (Figure 2.6b). For NKX2-2, there was a slight reduction in mRNA levels but a significant increase in primary transcript, which was measured using primers spanning an intron-exon junction

(Figure 2.6b). This is consistent with loss of PCBP1/2 leading to reduced stability of the NKX2-2 mRNA and/or reduced mRNA translation. Together, these findings implicate PCBP1/2 as post-transcriptional regulators of NKX2-2 and JUND, likely through a combination of reduced translation and mRNA stability. In contrast,

CRISPR-mediated depletion of hnRNPK did not impact protein levels of NKX2-2 or JUND under baseline conditions (Figure 2.6c).

Figure 2.6. Loss of PCBP1/2, but not hnRNPK, reduces NKX2-2 and JUND protein levels under baseline conditions. a, CRISPR-mediated depletion of PCBP1 and PCBP2 in Min6 cells causes reduction in NKX2-2 and JUND protein levels 8 days post-transduction, as assessed by Western blot (n=3). b, RT- qPCR analysis of NKX2-2, NKX2-2 primary transcript (PT), and JUND in Min6 cells with CRISPR- mediated depletion of PCBP1/2, 8 days post-transduction (n=3). c, Western blot depicting no change in NKX2-2 or JUND protein levels after CRISPR-mediated knockout of hnRNPK in Min6 cells, 8 days post-transduction (n=3). P values were calculated by unpaired two-tailed Student’s t- test. * = p < 0.05.

2.4.4 Impact of the PCBPs on insulin secretion in Min6 cells

NKX2-2 is a critical regulator of β cell identity, and genetic deletion of NKX2-2 in

β cells leads to impaired insulin secretion in mice (Gutiérrez et al., 2017). Given their role in the post-transcriptional regulation of NKX2-2, we hypothesized that loss of PCBP1/2 would also negatively impact insulin secretion. To test this, we performed glucose-stimulated insulin secretion assays in Min6 cells after

CRISPR-mediated depletion of PCBP1/2. As expected, Min6 cells in the

ROSA26 control group secreted approximately 10 times as much insulin when exposed to high glucose levels (25mM) compared to low glucose (2.5mM)

(Figure 2.7a,b). Interestingly, loss of PCBP1/2 led to a significant reduction in insulin secretion at 25mM compared to the ROSA26 group, and it caused a reduction in the fold-change from low to high glucose (Figure 2.7a,b). In contrast, there was no change in the total insulin content of the cells after loss of PCBP1/2

(Figure 2.7c), which indicates that there is not a defect in the synthesis of insulin but rather in its secretion.

To assess whether loss of hnRNPK also impacts insulin secretion, we measured glucose-stimulated insulin secretion in Min6 cells after CRISPR-mediated depletion of hnRNPK. Similar to the PCBP1/2 depletion, loss of hnRNPK led to a significant impairment in insulin secretion at high glucose levels compared to the

ROSA26 group and a reduction in the fold-change from low to high glucose concentrations (Figure 2.7d,e). In contrast to PCBP1/2, however, loss of hnRNPK led to a significant increase in insulin content, which was independent

58 of changes in insulin mRNA (Figure 2.7f,g). Together these data indicate that hnRNPK is required for normal insulin secretion in Min6 cells. This defect cannot be explained by reduced insulin levels as there was actually an increase in the total insulin content of Min6 cells with depletion of hnRNPK. Instead, these findings suggest a secretory defect in these cells. The cause of elevated insulin levels is unclear but appears to be independent of insulin mRNA transcription.

Although loss of hnRNPK did not affect NKX2-2 or JUND levels under baseline conditions (Figure 2.6), it did bind to a significant number of genes from our

TRAP screen containing the polyC motif by RIP-seq. Interestingly, our RIP-seq data indicated that hnRNPK binds not only to NKX2-2, but also other transcription factors critical for maintenance of β cell identity, including PDX1,

NKX6-1, and FOXO1. To assess whether loss of hnRNPK may impact β cell identity, we depleted hnRNPK in Min6 cells using CRISPR-Cas9 and assessed levels of NGN3, a marker of pancreatic endocrine progenitor cells that has been used to identify β cell “dedifferentiation” (Talchai et al., 2012). Interestingly, loss of hnRNPK led to a significant increase in NGN3 levels in Min6 cells (Figure

2.7h), suggesting this factor may be important for maintenance of β cell identity.

Figure 2.7. Loss of PCBP1/2 or hnRNPK impairs insulin secretion from Min6 cells. a-f, Assessment of glucose-stimulated insulin secretion (GSIS) in Min6 cells with CRISPR- mediated depletion of PCBP1/2 (a-c) or hnRNPK (d-f). Data presented as the amount of insulin secreted into media during a 30-minute incubation in low glucose (LG, 2.5mM) or high glucose (HG, 25mM) conditions with normalization to the total insulin content of the cells (a,d). GSIS results shown as the fold change in insulin secretion from LG to HG conditions (b,e). Total insulin content determined for each group with normalization to total protein levels (c,f). For GSIS, each group run in quadruplicate. g,h, RT-qPCR analysis of INS1 (n=4) (g) and NGN3 (n=3) (h) in Min6 cells with CRISPR-mediated depletion of hnRNPK, 8 days post-transduction. P values were calculated by unpaired two-tailed Student’s t-test, except in (a,d) in which a 2-way ANOVA was used. * = p < 0.05.

2.5 Discussion

To study translational regulation in pancreatic β cells, we have adapted the

TRAP methodology for use in a Min6 stable cell line expressing the GFP-

RPL10A transgene. One advantage of TRAP over other methodologies, such as ribosome profiling, is that expression of the transgene can be controlled in a cell type-specific manner. This is particularly attractive for assessing changes in ribosome occupancy in vivo as β cells reside alongside other endocrine cell types in pancreatic islets. Thus, an exciting and promising direction is to assess translational regulation in β cells under in vivo stress conditions, such as a high fat diet, using TRAP. This will present some challenges given the limiting material of primary islets, however, the continuing improvements in sequencing technologies for low input RNA indicate that this approach is likely to be successful.

Another enticing application of this methodology is studying translational regulation in human islets by lentiviral delivery of the GFP-L10a transgene expressed from a β cell-specific promoter. Since β cells represent a smaller fraction of the islet cell population in human compared to mouse islets, the cell type specificity of TRAP is particularly important in this context. One large hurdle to overcome, however, will be the time required for ribosomal turnover to occur.

For example, the half-lives of ribosomal proteins were determined to be approximately 9 days in rat brain tissue (Retz and Steele, 1980). This suggests that it may require prolonged culturing of human islets for a sufficient number of

61 ribosomes to incorporate the GFP-RPL10a protein. The time required for adequate incorporation will need to be determined empirically, but limitations on culturing human islets for extended periods may cause this approach to be problematic.

As a preliminary TRAP screen, we have used PDX1 deficiency to uncover novel translational regulation in β cells. This method, however, could be applied to a variety of stresses or stimuli that are relevant for β cell biology. For example, ER stress and oxidative stress are thought to contribute to β cell demise in T2D.

Therefore, a thorough characterization of translational changes that occur in response to these stresses may uncover novel mechanisms for adaptation to stress. Similarly, high glucose levels have been reported to acutely increase translation of insulin mRNA (Itoh and Okamoto, 1980), but the extent to which this occurs for other mRNAs under these conditions is unknown. TRAP could also be applied to determine the subset of mRNAs translationally regulated by particular RNA binding proteins in loss-of-function studies. Taken together with genome-wide assessments of RBP binding, such as eCLIP (Van Nostrand et al.,

2016), this could provide a comprehensive picture of the translational targets for a particular RBP.

TRAP can detect shifts in the ribosome occupancy of mRNAs on a genome-wide scale, however, it cannot detect changes in the usage of translation initiation sites due to the sequencing of full-length mRNA. This is particularly notable because translation of an alternative open reading frame of insulin generates an 62 immunogenic peptide that is potentially contributing to type 1 diabetes pathogenesis (Kracht et al., 2017). Therefore mapping the usage of alternative translation initiation sites on a genome-wide scale, and assessing conditions that alter the relative usage of these sites, may advance our understanding of β cell autoimmunity. To address this question, TRAP could be adapted to include an

RNase digestion step prior to immunoprecipitation, which would generate ribosome-protected RNA fragments similar to ribosome profiling. This would allow for the assessment of ribosome localization along mRNAs while retaining the advantage of cell type-specificity.

A de novo motif analysis on genes from our TRAP screen identified enrichment of a poly(C) motif in the 3’ UTR of genes with increased ribosome occupancy, which is similar to the preferred binding sequence for the poly(C)-binding protein family of RNA binding proteins. Two of the genes with enrichment of this motif

(NKX2-2 and JUND) are also bound by the PCBPs by RIP and require PCBP1/2 for normal expression under baseline conditions. These data suggest that these

RBPs directly regulate the expression of NKX2-2 and JUND. However, we cannot rule out the possibility that loss of PCBP1/2 indirectly affects the expression of these genes. One approach to expand on the level of regulation is to use a reporter gene flanked by the 5’ and 3’UTR of the gene of interest.

Deletion constructs can then be generated to assess the importance of the poly(C) regions for maintaining normal expression levels or for the control of stress-dependent changes in expression. Although this approach is commonly

63 used for this purpose, we have been unsuccessful in our attempts to model the regulation of NKX2-2 or JUND using reporter assays. There are many reasons that an artificial reporter gene may not fully recapitulate the complex regulation of endogenous genes. For example, RNA secondary structure may be altered due to presence of remaining plasmid sequence, or the assembly of RNP complexes may require factors recruited by the local chromatin environment. Nevertheless, we cannot currently rule out the possibility that the PCBPs regulate NKX2-2 or

JUND independent of binding to 3’UTR poly(C) motifs. In the future, it would be interesting to generate deletions of the poly(C) motifs by genome editing to assess their impact on gene expression and stress adaptation.

Loss of PCBP1/2 or hnRNPK leads to impaired glucose-stimulated insulin secretion in Min6 cells. As an insulinoma cell line, Min6 cells are an imperfect model of primary β cells. However, our methodology provides a robust increase in insulin secretion from Min6 cells during high glucose conditions compared to low glucose, suggesting these cells are a useful model for preliminary experiments. To follow-up on these results, our lab is actively characterizing the impacts of β cell-specific ablation of PCBP1 and PCBP2 in vivo on glucose homeostasis and insulin secretion.

The exact mechanisms by which loss of PCBP1/2 or hnRNPK reduces insulin secretion are currently unclear. In both cases, total cellular insulin levels cannot explain this phenotype, but rather there is likely a secretory impairment in these cells. One useful approach to further localize this defect along the pathway of 64 glucose-stimulated insulin secretion is calcium imaging. For example, if loss of these RBPs was associated with normal glucose-stimulated calcium influx, it would suggest there is not a problem with glucose sensing but rather in a more distal step such as insulin granule exocytosis. An in-depth investigation of the mRNA targets regulated by the PCBPs in β cells will also shed light on the precise cause of the impaired insulin secretion. This objective is complicated by the fact that the PCBPs can regulate gene expression at many levels including transcription, splicing, mRNA stability, and translation. Nevertheless, a systematic examination of gene expression changes downstream of PCBP depletion holds promise to elucidate novel regulatory networks governing β cell functionality.

Loss of β cell identity has received growing attention as a cause of impaired insulin secretion in T2D. Therefore, it will be important to identify the full inventory of factors shaping β cell identity and to understand the dynamics and adaptability of these gene regulatory networks during environmental perturbations. Our finding that loss of hnRNPK in Min6 cells leads to an upregulation of NGN3, a marker of endocrine progenitor cells, suggests that hnRNPK may be a novel regulator of β cell identity. On the other hand, hnRNPK depletion leads to increased insulin levels in Min6 cells, which seems to contradict the adoption of an endocrine progenitor fate. One explanation for these findings is that loss of hnRNPK leads to the formation of distinct subpopulations of cells, including a

65 subset of NGN3+ cells. An interesting approach to assess this possibility is performing single cell RNA-seq on β cells after depletion of hnRNPK.

The role of hnRNPK in regulating β cell identity is particularly interesting because it can integrate various signaling pathways with multiple levels of gene expression, allowing for the swift reshaping of regulatory networks. hnRNPK binds to the mRNAs of several genes with critical roles in β cell identity, including

NKX2-2, PDX1, NKX6-1, and FOXO1. However, we have been unable to identify changes in the expression level of NKX2-2 or PDX1 with loss of hnRNPK. One possible explanation for these findings is that hnRNPK regulates these genes under stress conditions or as cells recover from stress, but not under baseline conditions. PCBP1/2 may also be important for maintenance of β cell identity as loss of these factors leads to reduced NKX2-2 levels. An active area of investigation for our lab is to better understand whether PCBP1/2 depletion causes loss of β cell identity and reprogramming to a new cell type.

CHAPTER 3: JUND PROMOTES β CELL APOPTOSIS AND OXIDATIVE STRESS DURING METABOLIC STRESS 3.1 Summary

During the development of T2D, prolonged oxidative stress conditions contribute to β cell dysfunction and apoptosis; however, the factors shaping the oxidative stress response in β cells is poorly understood. By focusing on translational regulation, we have identified the transcription factor JUND as a potential stress- responsive factor in β cells. Indeed, exposure of mouse islets or Min6 cells to metabolic stress caused by high glucose and free fatty acid levels leads to the translational induction of JUND. This post-transcriptional regulation of JUND also occurs during in vivo stress conditions and is conserved in human islets.

Surprisingly, depletion of JUND in β cells reduces oxidative stress and apoptosis caused by metabolic stress. Consistent with this, JUND regulates the transcription of pro-oxidant and pro-inflammatory genes in β cells. Furthermore, there is a significant overlap between JUND target genes and genes increased in islets from diabetic db/db mice. Thus, we have uncovered the novel translational regulation of JUND during metabolic stress and implicate this transcription factor in activating a maladaptive response in β cells during pathophysiologically relevant conditions.

3.2 Introduction

In T2D, β cell dysfunction occurs due to prolonged stress conditions that impact the function and viability of these cells. Of particular importance is chronic oxidative stress, or the accumulation of reactive oxygen/nitrogen species

(ROS/RNS) in the cell. Chronic oxidative stress has been suggested to impact various aspects of β cell function including reduced expression of key β cell genes such as those encoding insulin, PDX1, and MAFA (Jonas et al., 1999;

Robertson et al., 2007). In fact, β cells are though to be particularly vulnerable to oxidative stress due to low expression of antioxidant genes (Lenzen et al., 1996).

Furthermore in animal models of diabetes, antioxidant treatment can reverse the onset of hyperglycemia by improving β cell function (Han et al., 2015), however, the use of antioxidants for treatment of humans with T2D has as of yet not proven beneficial (Rochette et al., 2014).

The cause of oxidative stress in β cells during T2D is complex and may be manifold. Part of this complexity arises due to the generation of ROS in various cellular compartments. For example, ROS are a normal byproduct of oxidative phosphorylation in mitochondria, and increased flux through the electron transport chain can cause increased mitochondrial oxidative stress (Sakai et al.,

2003). On the other hand, high levels of free fatty acids cause lipid oxidation to occur predominantly in peroxisomes, leading to hydrogen peroxide generation in this organelle (Gehrmann et al., 2010). Additionally, many cytoplasmic enzymes, such as NADPH oxidases, generate ROS and may be induced in β cells under 68 stress conditions, such as cytokine treatment (Weaver et al., 2014). It is unclear to what extent these different sources of ROS contribute to β cell dysfunction in

T2D. This is further complicated by the fact that different models of β cell stress

(ie: glucotoxicity vs. lipotoxicity) implicate different ROS sources as the critical mediators of dysfunction (Elsner et al., 2010; Sakai et al., 2003).

Despite extensive evidence linking oxidative stress to β cell demise in T2D, the factors shaping the oxidative stress response in β cells, and how these factors are regulated during disease progression, are unclear. For this reason, we were particularly interested in further exploring our finding that the transcription factor

JUND may be translationally regulated in β cells (Figure 2.2), because this factor has been linked to redox homeostasis in other cell types (Gerald et al., 2004;

Paneni et al., 2013).

JUND is a member of the Jun family of transcription factors, which also includes

C-JUN and JUNB. These factors form either homo- or hetero-dimers among themselves or with factors from other transcription factor families, including Fos,

Atf, and Maf. Importantly, the composition of these dimers will dictate the DNA binding preference and transcriptional effect of JUND (Mechta-Grigoriou et al.,

2001). Consequently, the transcriptional and phenotypic effects of JUND may vary across cell types based on the relative abundance of its binding partners.

The effect of JUND on oxidative stress has mostly been studied in the context of a global JUND knockout mouse model. Fibroblasts and endothelial cells isolated

69 from these mice display elevated ROS levels compared to those from wild type mice, which was linked to reduced expression of antioxidant genes (Gerald et al.,

2004; Paneni et al., 2013). JUND knockout mice have a shortened lifespan and hyperinsulinemic hypoglycemia, and pancreatic islets from these mice are hypervascularized (Laurent et al., 2008). However, whether loss of JUND also causes cell-autonomous changes in β cells has not been explored.

Although the role of JUND in β cells has not been studied, it is worth noting that one factor known to interact with JUND and regulate its function is MENIN, a transcriptional cofactor with repressive function. Mutations in the gene encoding this cofactor, Men1, cause multiple endocrine neoplasia, type 1 (MEN1). This disorder is characterized by tumors of the pancreatic islets, parathyriods, anterior pituitary, and duodenum (Feng et al., 2017). MENIN represses the transcriptional activity of JUND (Agarwal et al., 1999), raising the intriguing possibility that JUND may be important for regulating β cell proliferation. However, whether this interaction between JUND and MENIN is important for the role of either factor in

β cells is unknown.

One approach to address the role of JUND in β cells would be to use a mouse model in which the gene encoding JUND is flanked by loxP sites to allow for its conditional ablation using a β cell-specific Cre driver. Unfortunately, a JUND conditional knockout mouse is currently unavailable. To circumvent this problem, we have used lentiviral delivery of an shRNA targeting JUND from the rat insulin

70 promoter to achieve β cell-specific depletion of JUND in primary mouse islets, leading to our novel findings that depletion of JUND in β cells reduces apoptosis and oxidative stress during glucolipotoxicity.

3.3 Materials and Methods

Animals

Animal studies were approved by the University of Pennsylvania Institutional

Animal Care and Use Committee. Wild type CD1 males were purchased from

Jackson Laboratory. Male db/db mice (C57BLKS/J Leprdb/db) or db/+ mice

(C57BLKS/J Leprdb/+) were purchased from Jackson Laboratory. Blood glucose levels were determined by handheld glucometers (One Touch). Serum NEFA levels were determined by fluorometric assay (Abcam). Mice were housed in a

12hr light/dark cycle and had ad libitum access to food.

Lentivirus production

293T cells were transfected for 8 hrs in OptiMEM using Lipofectamine 2000

(Invitrogen), after which the media was changed to standard high glucose

DMEM. psPAX2 and pMD2.G were used for packaging and envelope vectors.

These plasmids were a gift from Didier Trono (Addgene plasmid #12260 and #

12259). Media containing virus was collected 2 and 3 days post-transfection.

Ultracentrifugation of collected media (19,000rpm for 1.5hrs at 4°C) was used to concentrate virus. Lentivirus was titered by RT-PCR (Sastry et al., 2002).

Islet isolation and culture

Mouse islets were isolated from 6-12 week old CD1 male mice unless otherwise noted. Briefly, ductal inflation of the pancreas was performed followed by collagenase digestion (Roche 11213873001). Islets were enriched by density 72 gradient centrifugation with Ficoll-Paque (GE 45-001-751). After handpicking 3-4 times, islets were collected for RNA/protein isolation or cultured overnight for recovery from isolation and stress treatments were started the next day.

Human islets were obtained through the NIH-supported Human Pancreas

Analysis Program via the University of Pennsylvania Islet Core facility. The islets were harvested from non-diabetic deceased donors without any identifying information at NIH-approved centers with informed consent and IRB approval at the islet isolation centers. Human islet donor characteristics are provided below in Table 2.

The culture media used for mouse and human islets was RPMI 1640 (11mM glucose) supplemented with 10% FBS, 2mM glutamine, 1mM sodium pyruvate,

10mM HEPES, 1% antibiotic antimycotic (Thermo 15240096), and pH was adjusted to 7.3-7.4.

Islet transductions

Lentiviral infection of mouse islets was performed as described (Jimenez-Moreno et al., 2015). 100-200 islets were cultured overnight in serum-free islet media containing lentivirus at an MOI of 20.

Palmitate preparation and glucolipotoxicity conditions

Palmitate (Sigma P9767) was dissolved in 50% ethanol at 65°C and diluted in

10% BSA to a concentration of 7mM. The mixture was incubated at 37°C for 1

73 hour to allow for conjugation before diluting in culturing media to a final concentration of 0.5mM. Control media was made by performing the same procedure with 50% ethanol and no palmitate.

For islets, control media (described above) contained 11mM glucose with no added palmitate. Media for glucolipotoxic conditions contained 25mM glucose with 500uM palmitate.

For Min6, control media (DMEM) contained 5.6mM glucose with no added palmitate while glucolipotoxic conditions had 25mM glucose and 500uM palmitate.

RNA isolation and RT-qPCR

For islets, handpicked islets were washed 2X with cold PBS and RNA was extracted using RNeasy Mini Kit (Qiagen). RNA was reverse transcribed using oligo(dT) and Superscript III (Invitrogen). Quantitative PCR (BioRad CFX384) was used to measure transcript abundance and normalized to HPRT. See Table

1 for a list of primer sequences used for these analyses.

Western blot

Proteins were separated by SDS-PAGE and immunoblotted with the following antibodies: rabbit anti-JUND (Santa Cruz, sc-74), mouse anti-ATF4 (Santa Cruz, sc-390063), rabbit anti-cleaved caspase-3 (Cell Signaling, 9664S), mouse anti-

Tubulin (Sigma, T9026), and mouse anti-Ran (B.D. 610340).

Cycloheximide chase assay

Min6 cells were cultured in control or glucolipotoxic conditions for 30hrs prior to treatment with cycloheximide at a concentration of 200ug/mL. Cells here harvested at 0hr, 3hr, 6hr, and 9hr time points post-cycloheximide treatment and analyzed by western blot. shRNA design and cloning

The lentiviral backbone was generated by cloning the rat insulin II promoter (-405 to +7 relative to the TSS), GFP, and the UltramiR mir-30 scaffold (Knott et al.,

2014) into pLenti CMV puro (Campeau et al., 2009) by PCR and Gibson

Assembly (replaced CMV promoter with rat insulin promoter). shRNA sequences were designed using the shERWOOD algorithm (Knott et al., 2014) (shJunD:

AGCAGCTCAAACAGAAAGTCC, shNT: GCGCGATAGCGCTAATAATTT).

CRISPR design and cloning

CRISPR gRNAs were designed using http://crispr.mit.edu/ to minimize off-target binding (ROSA26: AAGATGGGCGGGAGTCTTCT, JUND:

CAGCTTGCGCTTGCGGCATT). gRNAs were cloned into lentiCRISPR v2, as described (Sanjana et al., 2014).

Immunofluorescence staining of isolated islets

Transduced islets were collected, dispersed to single cell suspension, washed in

PBS, and fixed for 10 minutes at room temperature in 4% PFA. Fixed cells were

75 attached to slides using cytospin. Following permeabilization (0.1% Triton x-100 for 10 min at room temperature), immunofluorescence staining was performed using the following antibodies: guinea pig anti-insulin (Dako, A0564) and goat anti-GFP (Abcam, 6673). Images were taken on a Keyence BZ-X700 microscope and images were analyzed using the BZ-X Advanced Analysis Software.

Oxidative stress measurement

For mouse islets, following stress treatment, 50-100 intact islets were transferred to a polysterene round bottom tube and CellROX Deep Red Reagent (Invitrogen) was added at 1:500 dilution to culture media followed by incubation at 37°C for

45 min. Islets were washed 2X with PBS, dispersed to single cell suspension, attached to slides using cytospin, and imaged using fluorescence microscopy

(Keyence BZ-X700 microscope). GFP signal was used to identify transduced cells and the CellROX signal from each cell was quantified and normalized by cell area using BZ-X Advanced Analysis Software. Signal from at least 60 GFP positive cells was assessed for each condition per experiment.

For Min6 cells, following stress treatment, CellROX Deep Red Reagent

(Invitrogen) was added at 1:500 dilution directly to culture media followed by incubation at 37°C for 30 min. Cells were washed 2X with PBS and imaged using fluorescence microscopy (Keyence BZ-X700 microscope). Bright field images were used to determine cell areas and CellROX signal was quantified from at least 6 imaging fields for each group using BZ-X Advanced Analysis Software.

Caspase-3 and-7 activation assay

Following stress treatment, 50-100 intact islets were transferred to a polysterene round bottom tube and fluorescent inhibitor of caspases (FLICA) reagent (Image- iT LIVE Red Caspase-3 and -7 Detect kit, Invitrogen) was added at 1:150 dilution followed by incubation at 37°C for 1 hour. Islets were washed 2X with wash buffer (provided by manufacturer), dispersed to single cell suspension, attached to slides using cytospin, and imaged using fluorescence microscopy (Keyence

BZ-X700 microscope). Double positive cells were determined using BZ-X

Advanced Analysis Software, and at least 200 GFP positive cells were counted for each condition per experiment.

RNA-seq and analysis

2013) and read counts per gene were determined using featureCounts (Liao et al., 2014). Differential expression analysis was performed using edgeR

(Robinson et al., 2010) with significant genes called using fold-change cutoffs of greater than 1.5 or less than -1.5 and FDR less than 0.05. Gene ontology analysis was performed using DAVID (Huang et al., 2009). The overlap between

RNA-seq data sets was determined using hypergeometric tests.

Statistics

Data are presented as mean ± standard error of the mean unless otherwise noted in figure legends. Statistical analyses were performed using GraphPad

PRISM 7 software. Statistical tests used are noted in figure legends and include unpaired two-tailed Student’s t-test, one-way analysis of variance (ANOVA), two- way ANOVA, and hypergeometric test.

Cold Medical Cause of Gender Age Race BMI HbA1c ischemia History death time (hr) Non- Anoxia, drug Donor 1 F 39 Caucasian 34.8 4.7 8:33 diabetic intoxication

Non- Anoxia, drug Donor 2 M 24 Caucasian 20.8 4.9 16:27 diabetic intoxication

Non- Anoxia, Donor 3 F 31 Caucasian 32.7 4.4 9:59 diabetic cardiovascular

Non- Anoxia, drug Donor 4 M 23 unknown - 5.3 - diabetic intoxication

Table 2. Human islet donor characteristics

3.4 Results

3.4.1 Translational upregulation of JUND in β cells during metabolic stress

Our TRAP-seq results indicated that a subset of genes had increased ribosome occupancy in Min6 cells with PDX1 deficiency despite no change in mRNA abundance (Figure 2.2c). Of particular interest was the transcription factor JUND, which is important for redox homeostasis in various cell types (Gerald et al.,

2004; Paneni et al., 2013). Given the particular vulnerability of β cells to oxidative stress (Lenzen et al., 1996; Robertson et al., 2003), we hypothesized that JUND plays an important, but as of yet unappreciated, role during the β cell stress response.

We first investigated whether JUND upregulation extends to other stress conditions with relevance for T2D pathophysiology. To this end, we exposed mouse islets to a panel of stressors, including hydrogen peroxide (oxidative stress), high levels of glucose and the free fatty acid palmitate, termed glucolipotoxicity (metabolic stress), and thapsigargin (ER stress). While ATF4 was upregulated in all of these stress models, JUND was only induced by metabolic stress (Figure 3.1a). Despite the increase in JUND protein levels during glucolipotoxicity, there was no change in the abundance of JUND mRNA

(Figure 3.1b). This discordance could be caused by either an increase in the translation of JUND mRNA or an increase in the stability of JUND protein. A

TRAP assay performed in GFP-RPL10A Min6 cells during metabolic stress

80 showed a significant shift towards greater ribosome occupancy of the JUND mRNA (Figure 3.1c), consistent with increased translation of JUND mRNA during glucolipotoxicity. We next performed a cycloheximide chase assay in Min6 cells and found no increase in JUND protein stability during glucolipotoxicity (Figure

3.1d,e). Furthermore, there was no difference in TUBULIN levels throughout the time course between the two treatment groups, providing an additional control for the experiment (Figure 3.1f). We also confirmed that there was a significant induction of JUND protein levels at the 0hr time point (Figure 3.1g). Together, these data indicate that JUND is translationally upregulated in β cells during metabolic stress.

Figure 3.1. JUND is translationally upregulated in β cells during metabolic stress. a, Western blot depicting ATF4 and JUND levels in mouse islets treated with hydrogen peroxide (H2O2, 200uM) for 1 hr, glucolipotoxicity (GLT, 25mM glucose, 500uM palmitate) for 2 days, or thapsigargin (Tg, 1uM) for 3 hrs (n=3). b, Increased protein levels of JUND as determined by Western blot, but not transcript levels by RT-qPCR, in mouse islets treated with glucolipotoxicity 82 for 2 days (n=5). c, Increased ribosome occupancy of JUND in GFP-RPL10a Min6 cells after 30 hours of glucolipotoxic conditions, as determined by TRAP followed by RT-qPCR (n=3). d, Representative western blot depicting reduction in JUND protein levels at indicated times after addition of cycloheximide (CHX) to Min6 cells. CHX added after culturing for 30hrs in control or glucolipotoxic (GLT) culturing conditions (n=3). e,f, Quantification of western blot signal for JUND (e) or TUBULIN (f) normalized to RAN. Each group normalized to value at 0hr time point to depict reduction in protein over time. g, Quantification of JUND western blot at 0hr time points with normalization to the control group to depict increased JUND levels prior to CHX addition. P values were calculated by unpaired two-tailed Student’s t-test, except in (e,f) in which a two-way ANOVA was used. For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non-specific band. Otherwise, * = p < 0.05. Cont denotes control culturing conditions and GLT denotes glucolipotoxic culturing conditions.

To determine if this induction of JUND also occurs in β cells under metabolic stress in vivo, we used 12-week-old db/db mice, which are obese with elevated serum glucose and free fatty acid levels (Figure 3.2a-c). Indeed, compared to non-diabetic db/+ mice, islets from db/db mice had a significant increase in JUND protein, but not mRNA (Figure 3.2d). Thus, we have uncovered the post- transcriptional upregulation of JUND as a novel component of the β cell response to metabolic stress.

To extend these findings to a human model, we assessed the induction of JUND during metabolic stress by culturing human islets with high levels of glucose and palmitate. Consistent with our findings in mouse islets, this excess of metabolic fuel caused an increase in JUND protein levels in human islets (Figure 3.2e), indicating that the induction of JUND during metabolic stress is conserved in humans.

Figure 3.2. Post-transcriptional induction of JUND in islets from db/db mice and in human islets during metabolic stress. a-c, Compared to age- and gender-matched db/+ mice, db/db male mice at 12-weeks of age had increased body weight (a), blood glucose levels (b), and serum non-esterified fatty acid levels (c). n = 3-4 per group. d, Western blot showing increased JUND levels in islets isolated from db/db mice compared to db/+ at 12 wks of age. No change in JUND transcript as determined by RT- qPCR (n=3). e, Representative Western blot showing increased JUND levels in human islets treated with glucolipotoxic (GLT) conditions for 2 days. Quantification of results from four human donors are shown. * = p < 0.05. P values were calculated by unpaired two-tailed Student’s t-test. For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non- specific band. Otherwise, * = p < 0.05.

3.4.2 β cell-specific depletion of JUND in primary islets

To study the function of JUND during metabolic stress, we developed a system for β cell-specific depletion of JUND in isolated islets. To this end, we adopted a protocol to allow for efficient transduction of primary islets with lentiviral constructs (Jimenez-Moreno et al., 2015). Initial attempts to deplete genes using lentiviral delivery of CRISPR-Cas9 were unsuccessful in primary mouse islets, thus we decided to use an shRNA system. A lentiviral vector was constructed to co-express GFP and an shRNA from the rat insulin promoter, and its functionality was confirmed in Min6 cells (Figure 3.3a,b). Lentiviral transduction of intact mouse islets was used to deliver shRNA constructs, followed by a recovery period in culture media. To assess transduction efficiency, islets were dispersed and assessed for GFP positivity by immunofluorescence staining. Nearly half of all islet cells were GFP positive, and all GFP positive cells stained positive for insulin, indicating efficient and β cell-specific transgene delivery (Figure 3.3c).

Further, the shRNA targeting JUND provided a 50% reduction in its transcript level (Figure 3.3c), similar to the rate of transduction and suggesting a robust depletion of JUND in those cells expressing the transgene.

Figure 3.3. β cell-specific depletion of JUND in mouse islets via lentiviral delivery of shRNA. a, Schematic depicting shRNA system for β cell-specific depletion of JunD. b, Depletion of JunD transcript in Min6 cells transduced with shRNA targeting JunD or non-targeting control (n = 5). c, Immunofluorescence images showing β cell-specific expression of GFP in half of mouse islet cells after lentiviral transduction. Arrows denote β cells staining positive for GFP. Delivery of shRNA targeting JUND provides a significant reduction in JUND transcript compared to a non- targeting (NT) control (n=3-4). * = p < 0.05. Significance determined by unpaired two-tailed Student’s t-test.

3.4.3 Reduced oxidative stress in β cells with JUND depletion during metabolic stress

Chronically elevated glucose and free fatty acid levels cause β cell dysfunction and apoptosis at least partly due to increased oxidative stress (Piro et al., 2002;

Poitout and Robertson, 2008), but whether JUND contributes to β cell redox homeostasis during metabolic stress is unknown. As expected, β cells exposed to glucolipotoxicic conditions showed elevated oxidative stress as seen by increased fluorescence from an oxidation-sensitive dye (Figure 3.4a,b). If JUND were playing an antioxidant role in β cells, we predicted that there would be an exacerbation of oxidative stress caused by glucolipotoxicity. Surprisingly, however, depletion of JUND in β cells blocked the increase in oxidative stress caused by high levels of glucose and palmitate (Figure 3.4a,b). This is in contrast to findings in other cell types where loss of JUND leads to elevated reactive oxygen species (ROS) levels (Gerald et al., 2004; Paneni et al., 2013), highlighting the cell-type specificity of this factor. This prevention of redox imbalance during metabolic stress was also observed in Min6 cells with CRISPR- mediated depletion of JUND compared to a control group with a gRNA targeting the ROSA26 locus (Figure 3.4c,d).

Figure 3.4. β cell-specific depletion of JUND in mouse islets reduces oxidative stress during glucolipotoxicity. a,b, Assessment of oxidative stress in β cells transduced with shRNA vectors targeting JUND or NT control by fluorescence imaging of CellROX Deep Red reagent. Whole mouse islets cultured for 2 days in control or glucolipotoxic conditions prior to oxidative stress assessment. Representative fluorescence image for GFP and CellROX in islet cells after cytospin and frequency distribution of CellROX signal (a) and quantification of CellROX signal in GFP positive cells (b). The frequency distribution of CellROX intensity is a representative finding and the averages are for 3 independent experiments. c, Western blot showing depletion of JUND in Min6 cells using lentiviral delivery of CRISPR-Cas9. A gRNA targeting the ROSA26 locus is used as a negative control. d, Quantification of CellROX signal in Min6 cells with CRISPR-mediated depletion of JUND after culturing in glucolipotoxic conditions for 30hrs (n=4). P values were calculated by two-way ANOVA. For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non-specific band. Otherwise, * = p < 0.05. Cont denotes control culturing conditions and GLT denotes glucolipotoxic culturing conditions.

3.4.4 Reduced apoptosis in β cells with JUND depletion during metabolic stress

β cells are thought to be particularly sensitive to oxidative stress due to low expression of antioxidant genes (Lenzen et al., 1996). Thus, we investigated whether the impact of JUND on redox homeostasis would also affect β cell apoptosis during metabolic stress. Using a fluorescence readout of caspase-3/7 activation, there was a 3-fold increase in the number of apoptotic β cells after culturing islets with high levels of glucose and palmitate for 3 days (Figure 3.5a).

Consistent with its impact on redox imbalance, depletion of JUND significantly reduced this induction of apoptosis during metabolic stress (Figure 3.5a). This improvement in cell survival was confirmed in Min6 cells as CRISPR-mediated depletion of JUND completely abrogated the increase in cleaved caspase-3 levels caused by metabolic stress (Figure 3.5b).

Figure 3.5. β cell-specific depletion of JUND in mouse islets reduces apoptosis during glucolipotoxicity. a, Assessment of apoptosis in transduced β cells by fluorescence imaging of FLICA reagent, which marks apoptotic cells as red, after 3 days of culturing mouse islets in control or glucolipotoxic conditions. The percentage of GFP positive cells that were apoptotic were averaged for 3 independent experiments. b, Western blot of cleaved caspase-3 used to assess apoptosis in Min6 cells with CRISPR-mediated depletion of JUND and cultured in glucolipotoxic conditions for 30hrs (n=3). P values were calculated by two-way ANOVA. For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non-specific band. Otherwise, * = p < 0.05. Cont denotes control culturing conditions and GLT denotes glucolipotoxic culturing conditions.

3.4.5 JUND regulates pro-oxidant and pro-inflammatory genes in β cells

In accordance with the distinct pro-oxidant role of JUND in β cells, none of the antioxidant genes reported to be targets of JUND in other cell types was dysregulated in β cells (Figure 3.6a). This indicates that both the transcriptional and phenotypic effects of JUND are cell-type specific. To investigate how JUND impacts redox homeostasis and cell survival in β cells, we assessed the transcriptome of Min6 cells cultured with high levels of glucose and palmitate after CRISPR-mediated depletion of JUND, leading to the identification of 27 downregulated genes and 10 upregulated genes (Figure 3.6b). Gene ontology analysis on the downregulated genes demonstrated a significant enrichment in processes including stress response, ROS metabolism, and inflammation (Figure

3.6c). β cell-specific depletion of JUND in mouse islets confirmed most of these genes to indeed be targets of JUND in primary β cells (Figure 3.6d). Interestingly, several of these genes, including Nos2, Ptgs2, and Steap4, encode proteins with enzymatic activity leading to ROS generation (Jin et al., 2015; Maciag, 2004;

Tabatabaie et al., 2003; Xia and Zweier, 1997; Zhou et al., 2013a). Thus, JUND induction during metabolic stress likely contributes to β cell demise by activating several deleterious genes, including pro-oxidants. Indeed, these JUND targets were increased in mouse islets exposed to high levels of glucose and palmitate

(Figure 3.6e). Strikingly, nearly half of the genes downregulated with depletion of

JUND were previously identified via RNA-seq to be upregulated in islets from diabetic db/db mice (Neelankal John et al., 2018) (Figure 3.6f), constituting a

92 statistically significant overlap (Figure 3.6g, p = 6.0x10-5). The JUND target genes also significantly overlapped with genes upregulated during PDX1 deficiency (Figure 2.2) or after treatment of human islets with palmitate (Cnop et al., 2014) (Figure 3.6g). Together, these findings indicate that JUND regulates a set of genes that are commonly increased in models of β cell dysfunction.

Figure 3.6. JUND regulates pro-oxidant and pro-inflammatory genes associated with β cell stress and dysfunction. a, RT-qPCR analysis of mouse islets with shRNA-mediated depletion of JUND for panel of genes reported to be downregulated with loss of JUND in denoted publications (n=3). b, Volcano plot depicting changes in transcript levels after JUND depletion in Min6 cells cultured in glucolipotoxic conditions for 30hrs, as determined by RNA-seq (n=3). Significant changes shown in red (upregulated) or blue (downregulated). c, Gene ontology analysis for downregulated genes. d, RT-qPCR panel of select JUND target genes in mouse islets with shRNA-mediated depletion of JUND (n=3). e, Increased transcript levels of JUND target genes in mouse islets cultured for 2 days in glucolipotoxic (GLT) conditions (n=3). f, Heatmap of the genes downregulated after JUND depletion with detectable expression in islets from db/db mice. The change in expression after JUND depletion (JUND KO) is compared to the change in expression in islets from db/db vs db/+ mice. Genes with an asterisk had a statistically significant change in both data sets. g, The significance of the overlap between gene sets is plotted. Genes downregulated after JUND depletion are compared to genes upregulated in islets from db/db mice, with PDX1 depletion in Min6 cells, or in human islets treated with palmitate (PA). P values were calculated by unpaired two-tailed Student’s t-test (a,d,e) or hypergeometric tests (g). * = p < 0.05, unless otherwise noted.

3.5 Discussion

Oxidative stress has long been recognized as a major contributor to β cell demise in T2D, thus new insights into the molecular mechanisms controlling redox homeostasis in β cells are central to our understanding of diabetes pathophysiology. Here, we have used a translation-centric approach to uncover

JUND as a stress-responsive gene and a novel regulator of β cell redox homeostasis. JUND was upregulated in isolated mouse and human islets exposed to metabolic stress, as well as islets from diabetic db/db mice, indicating that this response is conserved across species and relevant for in vivo stress conditions. In β cells, JUND depletion dampens oxidative stress caused by the presence of excess metabolic fuel, which is in contrast to reports in other cell types where loss of JUND enhances oxidative stress due to downregulation of antioxidant genes (Gerald et al., 2004; Paneni et al., 2013). Although this discrepancy was unexpected, the cell-type specific function of JUND is consistent with its ability to dimerize with a variety of binding partners, which likely shapes its tissue specific effects (Hai and Curran, 1991). Further, the amelioration of ROS accumulation with JUND depletion agrees with our finding that JUND positively regulates several genes with the capacity to increase ROS production in β cells, including Ptgs2, Steap4, and Nos2, while having no impact on the expression of antioxidant genes.

It is unclear whether one or several of the identified JUND targets contribute to its effect on oxidative stress and apoptosis in β cells. Interestingly, both Ptgs2 and

Steap4 are upregulated in islets from db/db mice and have some genetic association with T2D in humans (Konheim and Wolford, 2003; Sharma et al.,

2015). Ptgs2, which encodes cyclooxygenase-2 (COX-2), imparts deleterious effects in β cells, including impaired insulin secretion, reduced cell proliferation, and increased free radical levels (Oshima et al., 2006; Persaud et al., 2007;

Tabatabaie et al., 2003). Steap4 encodes a metalloreductase involved in iron and copper reduction (Scarl et al., 2017). While Steap4-/- mice develop hyperglycemia likely due to adipose inflammation and insulin resistance (Wellen et al., 2007), its function in β cells is unknown. In osteoclast differentiation and prostate carcinogenesis, however, increased STEAP4 levels cause an elevation in cellular

ROS (Jin et al., 2015; Zhou et al., 2013a). Further, Nos2, or inducible nitric oxide synthase (iNOS), has strong connections to cytokine-mediated β cell dysfunction and death (Zumsteg et al., 2000).

Several other genes downregulated by JUND depletion are implicated in islet inflammation, including Ccl2, Cxcl1, and Cxcl2, which have been linked to poor outcomes for islet transplantation (Citro et al., 2012; Piemonti et al., 2002). Thus, targeting JUND provides a new avenue for dampening oxidative stress and inflammation in islets, which warrants further investigation for clinical applications such as diabetes therapy and islet transplantation.

The connection between these JUND target genes and β cell demise strengthens our finding that JUND induction during glucolipotoxicity is maladaptive. Although we initially expected the opposite outcome, the data presented here consistently support this model. Furthermore, the significant overlap between JUND targets and genes upregulated in islets from db/db mice indicates that the regulation of these genes is relevant for in vivo stress conditions and is not an artifact of culturing conditions. To follow-up on these studies, it would be ideal to conditionally ablate JUND in β cell in vivo. However, this mouse model will need to be carefully designed to avoid disrupting regulatory elements with loxP sites since JUND is an intronless gene.

The discrepancy between JUND’s effect on oxidative stress in β cells compared to endothelial cells or fibroblasts suggests that its interacting factors may also be distinct between these cell types. The precise interacting factors that give rise to these opposite phenotypes are currently unclear. Given the broad array of hetero-dimers that can be formed with JUND, it may be difficult to pinpoint exactly which combinations give rise to pro-oxidant or anti-oxidant effects.

The reported interaction between JUND and MENIN also raises the interesting possibility that JUND could impact β cell proliferation (Agarwal et al., 1999). We did not find any evidence that loss of JUND impacted cell cycle progression based on our RNA-seq results. However, we did not directly assess proliferation in our experiments. It should also be noted that the RNA-seq was performed in

Min6 cells, which are constantly in a proliferative state. Finally, JUND may 98 regulate β cell proliferation only in the context of stress or particular environmental conditions. Therefore, a closer examination of the impact of JUND on β cell proliferation in primary islets, or ideally in vivo, is warranted.

Our results show for the first time that translational induction of JUND occurs in β cells during metabolic stress conditions. To assess the conservation of this finding in humans, we have shown a significant increase in JUND levels in human islets cultured with high glucose and free fatty acid levels. This finding, combined with the significant overlap between JunD targets and genes upregulated during treatment of human islets with palmitate, indicates that JUND induction is pertinent to the β cell response to metabolic stress in humans. To expand on these findings, it would be interesting to examine whether there is an increase in JUND levels in islets from individuals with T2D by immunofluorescence staining. Given that the induction of JUND during in vitro stress conditions is 2-3 fold and that there is typically high variability between human samples, however, it may be difficult to confidently discern an increase in

JUND levels using this approach.

Given its effects on cell survival and oxidative stress, we propose a maladaptive role for JUND in β cells under metabolic stress. These findings are based on culturing islets under high glucose and palmitate levels comparable to that which would be seen in late stages of T2D. Similarly, the in vivo assessment of JUND induction utilized the db/db model at a stage of severe hyperglycemia. This is

99 consistent with the notion that β cell apoptosis contributes to the worsening of hyperglycemia at later stages of T2D rather than at the onset of the disease

(Prentki and Nolan, 2006). It is currently unclear whether JUND also influences β cell function under milder stress conditions that would occur at earlier stages of

T2D.

Furthermore, one may contemplate why this maladaptive response occurs in β cells under metabolic stress conditions. It is possible that this response is adaptive under certain situations and becomes inappropriately activated under prolonged disease conditions. Such a notion is particularly relevant for the generation of ROS in cells. At low doses, ROS can act as critical signaling molecules influencing a wide range of cellular processes, including insulin secretion (Pi et al., 2007; Ray et al., 2012). When ROS accumulation becomes excessive, however, these molecules become damaging. This imbalance, which is the cause of oxidative stress, leads to impaired insulin secretion and increased apoptosis in β cells (Poitout and Robertson, 2008). As such, it is possible that

JUND induction in the acute setting leads to constructive ROS generation and activation of adaptive signaling pathways, whereas chronically it promotes β cell demise.

100

CHAPTER 4: AN ERK/HNRNPK/DDX3X AXIS POST-TRANSCRIPTIONALLY

REGULATES JUND DURING METABOLIC STRESS

4.1 Summary

Translational regulation is a critical component of stress responses, however, the mechanisms controlling this process in pancreatic β cells during pathophysiologically-relevant conditions are poorly understood. Here we show that the translational induction of JUND during metabolic stress requires the RNA binding protein hnRNPK. Furthermore, activation of the MEK/ERK signaling pathway is both necessary and sufficient for phosphorylation of hnRNPK and induction of JUND during glucolipotoxicity. Consistent with the effect of JUND on

β cell apoptosis, MEK inhibition significantly reduced cell death in mouse islets during metabolic stress. The translational upregulation of JUND could not be attributed to a change in hnRNPK subcellular localization, RNA binding affinity, or nuclear retention of mRNA. Rather, we found that efficient JUND translation is dependent on the RNA helicase DDX3X, an hnRNPK-interacting factor. During glucolipotoxicity, there is increased interaction between DDX3X and the translation pre-initiation complex (PIC), and this interaction is abrogated with loss of hnRNPK. Thus, hnRNPK regulates JUND translation in part by facilitating the recruitment of the PIC via DDX3X.

101

4.2 Introduction

In order to adapt to stress conditions imposed by changes in the extracellular environment, β cells must dynamically shift their gene expression profile. This allows for the synthesis of proteins that promote restoration of cellular homeostasis and recovery from stress. On the other hand, changes in gene expression can be maladaptive during disease states and exacerbate cellular dysfunction. Therefore, the identification of factors that shape the β cell response to stress conditions is a critical step towards better understanding the mechanisms of β cell failure in T2D. In the previous chapters, we used a translation-centric approach to uncover the transcription factor JUND as part of a

β cell maladaptive response to high levels of glucose and free fatty acids, conditions relevant for T2D pathophysiology. Thus, understanding the mechanisms controlling JUND translation during stress could provide novel therapeutic strategies to modulate redox balance in β cells.

Translational regulation of specific mRNAs is dependent on the presence of regulatory elements, typically in the 5’ or 3’ UTRs, which permit targeting by sequence-specific factors, including microRNAs and RNA binding proteins

(RBPs). The enrichment of a poly(C) motif in genes from our TRAP screen suggested this element may impart translational control mediated by the poly(C)- binding protein family of RBPs.

102

The PCBP family consists of five members (PCBP1-4 and hnRNPK), which play diverse roles in gene regulation, including regulation of transcription, splicing, alternative polyadenylation, mRNA stability, and translation (Bomsztyk et al.,

2004; Ji et al., 2013; 2016; Makeyev and Liebhaber, 2002; Ostareck-Lederer et al., 2002; Waggoner et al., 2009). This diversity of function allows the PCBPs to affect a broad array of cellular processes, including apoptosis, cell cycle progression, cellular differentiation, etc. (Makeyev and Liebhaber, 2002;

Ostareck-Lederer and Ostareck, 2012; Waggoner et al., 2009). While the members PCBP1-2 and hnRNPK are broadly expressed across cell types, they may have cell-type specific effects based on the expression patterns of cofactors and target mRNAs. Furthermore, a common mechanism for the functional regulation of this family of RBPs is post-translational modification via phosphorylation (Bomsztyk et al., 2004; Chaudhury et al., 2010b; Kimura et al.,

2010). This allows for the rapid integration of various signaling pathways with multiple steps of gene expression. Thus, the PCBPs are fascinating candidates for shaping the β cell gene expression profile in response to various stimuli and stressors. Given their diversity of action, this regulation may be complex but nevertheless deserves further scrutiny to uncover novel aspects of β cell biology.

In this chapter, we focus on the role of hnRNPK in regulating JUND translation in

β cells, leading to the identification of an ERK/hnRNPK/DDX3X pathway that is activated during metabolic stress.

103

4.3 Materials and Methods

Cell line culture

Min6 mouse insulinoma cells passage 20-30 were cultured in high glucose

DMEM as described (Claiborn et al., 2010), unless otherwise noted. For lentiviral infections, Min6 cells were transduced for 6 hours with virus and polybrene

(Sigma) at 8ug/mL. Cells were allowed to recover for 4-5 days before collection or stress treatments. HEK293T cells were cultured in DMEM containing 25mM glucose.

Western blot

Proteins were separated by SDS-PAGE and immunoblotted with the following antibodies: rabbit anti-JUND (Santa Cruz, sc-74), rabbit anti-hnRNPK (Bethyl

Laboratories, A300-674A), rabbit anti-DDX3X (Bethyl Laboratories, A300-474A), rabbit anti-phospho-ERK1/2 (Cell Signaling, 4377), rabbit anti-total-ERK1/2 (Cell

Signaling, 4695), mouse anti-Tubulin (Sigma, T9026), and mouse anti-Ran (B.D.

610340).

Co-immunoprecipitation

Min6 cells were lysed in buffer containing 20mM Tris pH 8.0, 137 mM NaCl, 1mM

MgCl2, 1mM CaCl2, 1% NP-40, 10% glycerol, protease inhibitor cocktail

(Millipore), and Benzonase (Sigma) at 12.5 U/mL. Lysates were rotated at 4°C for 1 hour. Protein concentration was determined using a Micro BCA Protein

Assay Kit (Thermo). 1ug of primary antibody was added to lysate encompassing 104

Phos-tag analysis

Phos-tag was performed as described (Kinoshita et al., 2006). Briefly, lysates were run on SDS-PAGE gels containing 25uM Phos-tag acrylamide (Wako),

50uM MnCl2, and 8% acrylamide. The gels were run for 4hrs at 90V to achieve optimal separation of bands for hnRNPK. Prior to transfer, the gels were soaked in buffer containing 10mM EDTA for 1 hour.

RNA immunoprecipitation

RNA immunoprecipitation was performed using rabbit anti-hnRNPK (Bethyl

Laboratories, A300-674A) or rabbit anti-DDX3X antibodies (Bethyl Laboratories,

A300-474A). hnRNPK or DDX3X antibodies or IgG were bound to Protein A

Dynabeads in buffer containing 20mM HEPES pH 7.4, 5mM MgCl2, 150mM KCl,

2mM DTT, 1% NP-40. Min6 cells were lysed in buffer containing 50mM Tris pH

8.0, 150mM NaCl, and 1% NP-40 with RNase, protease, and phosphatase inhibitors. Total RNA was extracted from lysate (10% input) using RNeasy Mini

Kit (Qiagen). Lysate encompassing 75ug or 200ug of protein was added to beads

105 bound to antibody for hnRNPK or DDX3X, respectively, and an IgG control, then incubated overnight at 4°C. The next day, IPs were washed with buffer containing 20mM HEPES pH 7.4, 5mM MgCl2, 350mM KCl, 2mM DTT, 1% NP-

40, and RNase inhibitors. RNA was eluted from beads in RLT buffer (Qiagen) and RNA was isolated using RNeasy Mini Kit (Qiagen). For RIP followed by RT- qPCR, enrichment was calculated by first normalizing transcript abundance in IP

RNA to that in total RNA and then to the IgG control.

CRISPR design and cloning

CRISPR gRNAs were designed using http://crispr.mit.edu/ to minimize off-target binding (ROSA26: AAGATGGGCGGGAGTCTTCT, hnRNPK:

GTTTAATACTTACGTCTGTA, DHX9: TTCAGTTGTGATTATCCGAG and

GAGCGAGTTGCTTATGAGAG, DDX1: CAGATGAACCCATATGATAG and

GGAACTAGAGGACTGCTGAA, DDX3X: AGATTGGATACTGTTTACGA and

GCACCACCATAAACCACGCA). gRNAs were cloned into lentiCRISPR v2, as described (Sanjana et al., 2014).

MEK1-CA Cloning

Constitutively active MEK1 fragments were generated by PCR with primers to introduce S->D mutations at Ser218 and Ser222. CA-MEK1 and the rat insulin promoter fragments were cloned into pLenti CMV blast (Campeau et al., 2009) using Gibson Assembly (replaced CMV promoter with RIP).

Immunofluorescence staining of Min6 cells 106

Min6 cells were cultured in 4-well chamber slides. After culturing under desired conditions, cells were washed in PBS, and fixed for 15 minutes at room temperature in 4% PFA. Following permeabilization (0.1% Triton x-100 for 10 min at room temperature), immunofluorescence staining was performed using the following antibodies: guinea pig anti-insulin (Dako, A0564) and goat anti-GFP

(Abcam, 6673). Images were taken on a Keyence BZ-X700 microscope and images were analyzed using the BZ-X Advanced Analysis Software.

107

4.4 Results

4.4.1 hnRNPK is required for the translational induction of JUND during metabolic stress

Despite binding to its mRNA, loss of hnRNPK had no impact on JUND levels under baseline conditions (Figures 2.4 and 2.6). Instead, we hypothesized that hnRNPK controls the induction of JUND during metabolic stress. To assess this possibility, we used CRISPR-Cas9 to deplete hnRNPK in Min6 cells followed by treatment with high levels of palmitate. While control cells receiving a ROSA26 gRNA demonstrated the expected upregulation of JUND, depletion of hnRNPK completely blocked this induction despite having no impact on JUND mRNA abundance (Figure 4.1a,b). To confirm the specificity of this effect, we performed a knockout/rescue experiment for hnRNPK in Min6 cells. The gRNA targeting hnRNPK spans an intron-exon junction and thus should not affect vector- mediated expression of hnRNPK due to the absence of introns. Indeed, expression of hnRNPK from a lentiviral construct restored its expression in the presence of the targeting gRNA (Figure 4.1c). While loss of hnRNPK prevented

JUND induction during glucolipotoxicity, re-expression of hnRNPK rescued this effect comparable to the ROSA26 control group (Figure 4.1c). This indicates that the impact of hnRNPK on JUND induction is not due to an off-target effect of

CRISPR editing. Thus, hnRNPK binds to the JUND mRNA and is required for its post-transcriptional induction during metabolic stress.

108

Figure 4.1. CRISPR-mediated depletion of hnRNPK blocks JUND induction during glucolipotoxicity. a, Western blot and quantification showing depletion of hnRNPK in Min6 cells using CRISPR- Cas9 and its impact on the induction of JUND after 2 days of palmitate treatment (n=6). b, No change in JUND transcript levels with depletion of hnRNPK in Min6 cells during palmitate treatment, as determined by RT-qpCR (n=6). c, Western blot and quantification showing depletion of hnRNPK in Min6 cells using CRISPR-Cas9 and rescue with a vector to overexpress hnRNPK. Cells were treated for 30hrs in control (Cont) or glucolipotoxic (GLT) conditions (n=2). P values were calculated by two-way ANOVA (a) or unpaired two-tailed Student’s t-test (b). For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non-specific band. Otherwise, * = p < 0.05.

109

4.4.2 hnRNPK is phosphorylated during metabolic stress in a MEK- dependent manner

A common mechanism for regulation of hnRNPK function is post-translational modification via phosphorylation (Bomsztyk et al., 2004). To study changes in hnRNPK phosphorylation, we used Phos-tag SDS-PAGE, which unveils phosphorylated forms of hnRNPK via electrophoresis (Kimura et al., 2010). This approach clearly demonstrated a shift towards increased phosphorylation of hnRNPK in mouse islets or Min6 cells exposed to high levels of glucose and palmitate while total hnRNPK levels did not change (Figure 4.2a,b). This increase in phosphorylation was specific for metabolic stress, as it did not occur during stress induced by hydrogen peroxide or thapsigargin, paralleling the pattern of

JUND induction (Figure 4.2c). Similarly, there was an increase in hnRNPK phosphorylation in isolated islets from diabetic db/db mice, indicating that this pathway is also activated during metabolic stress in vivo (Figure 4.2d). To assess hnRNPK phosphorylation in a human model, we cultured human islets in glucolipotoxic conditions and observed a significant increase in hnRNPK phosphorylation by Phos-tag and no change in total hnRNPK levels (Figure

4.2e).

110

Figure 4.2. Increased phosphorylation of hnRNPK during metabolic stress. a, Phos-tag SDS-PAGE analysis shows changes in phosphorylation of hnRNPK in mouse islets treated with glucolipotoxic (GLT) conditions for 2 days (n=5). b, Phos-tag SDS-PAGE and Western blot showing increased phosphorylation of hnRNPK in Min6 cells treated with increasing doses of palmitate for 2 days (n=3). c, Phos-tag for hnRNPK showing changes in phosphorylation 111 in mouse islets treated with a panel of stressors including hydrogen peroxide (H2O2, 200uM) for 1 hr, GLT for 2 days, or thapsigargin (Tg, 1uM) for 3 hrs (n=3). d, Phos-tag SDS-PAGE analysis shows changes in phosphorylation of hnRNPK in islets from 12-wk old db/db or db/+ male mice (n=3). e, Phos-tag SDS-PAGE analysis shows changes in phosphorylation of hnRNPK in human islets treated with glucolipotoxic (GLT) conditions for 2 days (n=3). np denotes band for non- phosphorylated hnRNPK and ph denotes bands for phosphorylated hnRNPK. P values were calculated by unpaired two-tailed Student’s t-test. For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non-specific band. Otherwise, * = p < 0.05.

112

To assess whether this increase in phosphorylation is unique to hnRNPK or common amongst PCBP family members, we performed Phos-tag analysis for

PCBP1 and PCBP2 in Min6 cells exposed to metabolic stress conditions.

Compared to analysis using standard SDS-PAGE, Phos-tag revealed additional bands for PCPB1 and PCBP2 correlating to phosphorylated forms of these proteins (Figure 4.3). However, unlike hnRNPK, there was no obvious change in the relative pattern nor intensity of these bands in Min6 cells exposed to high palmitate levels (Figure 4.3). This suggests that the increase in phosphorylation during metabolic stress is unique to hnRNPK and does not occur for PCBP1/2.

However, Phos-tag analyses may not resolve all phosphorylated sites of a particular protein (Kinoshita et al., 2008), thus we cannot rule out the possibility that there is an increase in phosphorylation of PCBP1/2 at a site that cannot be identified by Phos-tag.

113

Figure 4.3. No change in phosphorylation of PCBP1/2 during metabolic stress as detected by Phos-tag analysis. Phos-tag SDS-PAGE and Western blot showing no change in phosphorylation or total protein levels of PCBP1/2 in Min6 cells treated with indicated concentration of palmitate for 24 hours (n=3).

114

One of the kinase cascades that targets hnRNPK and modulates its function is the MEK/ERK signaling pathway (Habelhah et al., 2001), which was activated in mouse islets treated with high levels of glucose and palmitate (Figure 4.4a).

Treatment of mouse islets with the potent and specific MEK1/2 inhibitor trametinib during metabolic stress completely abolished p-ERK1/2 levels while also blocking the increase in phosphorylation of hnRNPK and the induction of

JUND (Figure 4.4a). Trametinib treatment also reduced expression levels of most

JUND target genes in mouse islets, including Ptgs2 and Steap4 (Figure 4.4b). To test the sufficiency of ERK to activate the hnRNPK/JUND axis, we overexpressed a constitutively active form of MEK1 (S218/222D) in Min6 cells.

As expected, this led to a robust increase in phosphorylation of ERK1/2 (Figure

4.5a). Consistent with the effect of MEK inhibition, expression of constitutively active MEK1 was sufficient to increase JUND levels, hnRNPK phosphorylation, and expression of many JUND targets, most notably Ptgs2 (Figure 4.5a,b). To test whether blocking the hnRNPK/JUND axis via MEK inhibition would also reduce apoptosis, we assessed caspase activation in mouse islets treated with trametinib and found a striking decrease in apoptosis during metabolic stress

(Figure 4.6).

115

Figure 4.4. MEK inhibition blocks hnRNPK phosphorylation and JUND induction during metabolic stress. a,b Western blot, Phos-tag SDS-PAGE (a), and RT-qPCR (b) analyses show that trametinib treatment (1uM) blocks ERK phosphorylation and activation of the hnRNPK/JUND axis in mouse islets cultured for 2 days in glucolipotoxic (GLT) or control conditions. Quantification is the average of 4 (a) or 3 (b) biological replicates. * = p < 0.05. P values were calculated by two-way ANOVA (a) or unpaired two-tailed Student’s t-test (b). For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non-specific band. Otherwise, * = p < 0.05.

116

Figure 4.5. Overexpression of constitutively active MEK1 in Min6 cells is sufficient to activate the hnRNPK/JUND axis. a,b Lentiviral overexpression of constitutively active (CA) MEK1 (S218/222D) in Min6 cells activates the ERK/hnRNPK/JUND pathway as determined by Western blot and Phos-tag SDS- PAGE (a) or RT-qPCR (b) (n=3). EV denotes lentiviral transduction with empty vector. * = p < 0.05. P values were calculated by unpaired two-tailed Student’s t-test. For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non-specific band. Otherwise, * = p < 0.05.

117

Figure 4.6. MEK inhibition reduces islet cell apoptosis during glucolipotoxicity. Trametinib treatment (1uM) reduces the percentage of apoptotic cells, as determined by fluorescence imaging of FLICA reagent, in mouse islets cultured for 3 days in GLT conditions (n=3). * = p < 0.05. P values were calculated by two-way ANOVA.

118

4.4.3 No change in hnRNPK localization, RNA binding affinity, or nuclear retention of RNA during metabolic stress

To further investigate how hnRNPK may impact JUND translation during metabolic stress, we considered several possible mechanisms based on modifiable properties of hnRNPK. For example, hnRNPK is a nucleo-cytoplasmic shuttling protein and its subcellular distribution can vary amongst cell types

(Bomsztyk et al., 2004). Furthermore, this property of hnRNPK is dynamic and can be modified by its phosphorylation state. Indeed, it has been reported that phosphorylation of hnRNPK by ERK leads to its cytoplasmic accumulation in

HeLa cells (Habelhah et al., 2001). To assess whether phosphorylation of hnRNPK during glucolipotoxicity also influenced the subcellular distribution of hnRNPK, we performed immunofluorescence staining of Min6 cells. First, we confirmed the specificity of the hnRNPK antibody in immunostaining analyses by showing that CRISPR-mediated depletion of hnRNPK in Min6 cells leads to the complete loss of the immunofluorescence signal in the targeted cells (Figure

4.7a). Next, we assessed the subcellular localization of hnRNPK in Min6 cells treated with glucolipotoxicity and observed no change in its distribution compared to control conditions (Figure 4.7b). Given the reported connection between hnRNPK localization and ERK signaling, we further investigated this possibility by using overexpression of constitutively active MEK1, which leads to robust phosphorylation of ERK1/2 (Figure 4.5). In contrast to previous reports, expression of constitutively active MEK1 had no impact on hnRNPK subcellular

119 distribution (Figure 4.7c). Together, these results indicate that the impact of hnRNPK on JUND translation during glucolipotoxicity cannot be explained by a change in its nucleo-cytoplasmic shuttling.

120

121

Figure 4.7. No change in hnRNPK subcellular distribution in Min6 cells during metabolic stress. a, Representative immunofluorescence images of fixed Min6 cells with CRISPR-mediated depletion of hnRNPK in a subset of cells to confirm the specificity of the hnRNPK antibody. A gRNA targeting the ROSA26 locus was used as a negative control. b,c, Representative immunofluorescence images of fixed Min6 cells that were treated with control or glucolipotoxic conditions for 30hrs (b) or transduced with lentivirus to overexpress constitutively active MEK1 (CA-MEK1) (c), which show no detectable change in the subcellular localization of hnRNPK under these conditions.

122

Another modifiable property of hnRNPK is its RNA binding affinity. For example, treatment of a rat hepatoma cell line with insulin led to phosphorylation of hnRNPK, and this increased the binding affinity of hnRNPK for RNA containing poly(C) stretches (Ostrowski et al., 2001). Therefore, we hypothesized that phosphorylation of hnRNPK during glucolipotoxicity affected JUND translation by increasing the interaction between hnRNPK and the mRNA encoding JUND. To assess this possibility, we performed RNA immunoprecipitation for hnRNPK in

Min6 cells treated with control or glucolipotoxic conditions. This showed that there was no change in the amount of JUND mRNA pulled down with hnRNPK in the RIP experiment, suggesting that a change in hnRNPK binding to JUND mRNA cannot explain its effect on translation (Figure 4.8a).

The export of mRNA from the nucleus to the cytoplasm can be regulated to provide an additional post-transcriptional control on gene expression. In fact, some mRNAs are preferentially retained in the nucleus, and this has recently been attributed at least in part to targeting by hnRNPK (Lubelsky and Ulitsky,

2018). Therefore, we next considered the possibility that hnRNPK regulates the nuclear retention of JUND mRNA in a stress-dependent manner. To address this, we first investigated whether there is a shift in the nucleo-cytoplasmic localization of the JUND mRNA during metabolic stress. We performed fractionation experiments in Min6 cells to enrich for nuclear and cytoplasmic compartments and isolated RNA from these pools. To validate our methodology of fractionation, we assessed the nuclear/cytoplasmic distribution of genes with known

123 localization patterns. For example, mRNAs encoding housekeeping factors, such as HPRT and BETA-ACTIN are highly translated and should be predominantly cytoplasmic. In agreement with this, approximately 80% of the mRNA for these genes was located in the cytoplasmic fraction (Figure 4.8b). On the other hand,

NEAT1 is a nuclear enriched long non-coding RNA, and accordingly 90% of its mRNA was present in the nuclear fraction. As an additional nuclear control, we used primers spanning an intron-exon junction for PDX1 to detect its primary transcript, which should be found in the nucleus. Indeed, 85% of this mRNA was located in the nuclear compartment. Assessment of JUND mRNA showed that is was predominantly located in the cytoplasm, and there was no change in its localization during glucolipotoxicity (Figure 4.8b). Thus, a shift in the nucleo- cytoplasmic distribution of the JUND mRNA cannot explain its induction during metabolic stress.

124

Figure 4.8. No change in hnRNPK RNA binding affinity or nuclear retention of mRNA during metabolic stress. a, No change in the interaction of hnRNPK with the mRNA encoding JUND in Min6 cells after 30hrs treatment with glucolipotoxicity (GLT), as determined by RNA immunoprecipitation (n=4). b, Fractionation of Min6 cells into nuclear and cytoplasmic pools followed by RT-qPCR analysis shows no change in the subcellular localization of the JUND mRNA after culturing for 30hrs in GLT conditions (n=2). Transcript level in the cytoplasmic fraction was normalized to the combined transcript level in the cytoplasmic and nuclear fractions to determine the fraction of mRNA in the cytoplasm. HPRT and BETA-ACTIN were used as controls for cytoplasmic enriched RNA while NEAT1 and PDX1 primary transcript (PT) were used as controls for nuclear enriched RNA to assess fractionation quality. P values were calculated by unpaired two-tailed Student’s t-test. Cont denotes control culturing conditions and GLT denotes glucolipotoxic culturing conditions.

125

4.4.4 DDX3X interacts with hnRNPK and is required for efficient translation of JUND mRNA

Since we were unable to detect a change in some of the known modifiable properties of hnRNPK during metabolic stress (Figures 4.7 and 4.8), we hypothesized that hnRNPK impacts JUND translation via a binding partner. The

JUND mRNA contains a highly structured 5’UTR (Short and Pfarr, 2002), and efficient translation initiation is dependent on RNA helicases, such as DHX9, to resolve these secondary structures (Hartman et al., 2006). Upon searching mass spectrometry data sets for hnRNPK interacting factors (Chen et al., 2002; Mikula et al., 2015), we noticed the presence of the DEAD-box helicases DHX9, DDX1, and DDX3X in the hnRNPK interactome. Given the connection of DEAD-box helicases to the regulation of translation (Hartman et al., 2006; Linder and

Jankowsky, 2011), we depleted these genes in Min6 cells via CRISPR-Cas9 to determine their impact on JUND induction during metabolic stress (Figure

4.9a,b). Interestingly, loss of DDX3X, but not DHX9 nor DDX1, led to reduced

JUND protein levels in Min6 cells during glucolipotoxicity (Figure 4.9c). On the other hand, loss of DDX3X also led to a significant increase in JUND mRNA levels (Figure 4.9d). This decrease in the amount of JUND protein despite an increase in mRNA abundance likely signifies a defect in translation initiation caused by loss of DDX3X, consistent with its ability to enhance translation of select mRNAs (Soto-Rifo et al., 2012). Indeed, depletion of DDX3X in GFP-

RPL10A Min6 cells led to a significant reduction in the density of ribosomes

126 binding to the JUND mRNA, as determined by TRAP (Figure 4.9e). These data indicate that DDX3X is required for the efficient translation of JUND mRNA.

127

Figure 4.9. DDX3X is required for the efficient translation of JUND mRNA. a,b RT-qPCR for DHX9, DDX1, and DDX3X (a) and Western blot for DDX3X (b) show a significant depletion of these RNA helicases using CRISPR-Cas9 in Min6 cells (n=3). c, Representative western blot depicting JUND levels in Min6 cells transduced with lentivirus to deplete the indicated RNA helicase using CRISPR-Cas9 and cultured in glucolipotoxic conditions for 30hrs (n=3). d, Assessment of JUND mRNA abundance in Min6 cells by RT-qPCR after CRISPR-mediated depletion of DDX3X and culturing in glucolipotoxic conditions for 30hrs (n=3). e, Decreased ribosome occupancy of JUND in GFP-RPL10a Min6 cells after depletion of DDX3X by CRISPR-Cas9 and 30 hours of glucolipotoxic conditions, as determined by TRAP followed by RT-qPCR (n=3). * = p < 0.05 and ns = not significant. P values were calculated by two-way ANOVA except in (b) in which an unpaired two-tailed Student’s t-tests was used. In (d) * denotes significance compared to ROSA26 group of the same treatment. For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non-specific band. Otherwise, * = p < 0.05. 128

To assess the interaction between hnRNPK and DDX3X, we performed co- immunoprecipitation experiments in Min6 cells in the presence of a nuclease to degrade both RNA and DNA. This showed a clear nucleic acid-independent interaction between these two factors that was maintained during glucolipotoxicity (Figure 4-10a). DDX3X and its yeast homolog Ded1 can impact translation through interaction with components of the translation pre-initiation complex (PIC), especially the 18S ribosomal RNA (Guenther et al., 2018; Oh et al., 2016). Indeed, RIP for DDX3X showed a robust binding to 18S in Min6 cells, and this interaction was increased during metabolic stress (Figure 4-10b).

Intriguingly, CRISPR-mediated depletion of hnRNPK in Min6 cells significantly reduced the association between DDX3X and 18S (Figure 4-10c). DDX3X has been reported to bind to all expressed mRNAs, likely via its interaction with the

PIC (Guenther et al., 2018; Oh et al., 2016). Consistent with this, DDX3X bound to all analyzed mRNAs by RIP, and these interactions were significantly reduced with loss of hnRNPK (Figure 4-10d). These data support a model whereby hnRNPK impacts translation in part by promoting the interaction between DDX3X and the PIC. Thus, we have identified DDX3X as a novel regulator of JUND translation during metabolic stress that interacts with the 18S ribosomal RNA in an hnRNPK-dependent manner.

129

Figure 4.10. DDX3X interacts with the 18S ribosomal RNA in an hnRNPK-dependent manner. a, Western blots from pull-down of hnRNPK (left) or DDX3X (right) in Min6 cells treated with glucolipotoxic conditions for 30 hrs. b,c RNA immunoprecipitation for DDX3X followed by RT- qPCR for 18S ribosomal RNA in Min6 cells treated with glucolipotoxic conditions for 30hrs (n=3) (b) and in Min6 cells with CRISPR-mediated depletion of hnRNPK and cultured in glucolipotoxic conditions for 30hrs (n=3) (c). d, Decreased interaction between DDX3X and indicated mRNAs after CRISPR-mediated depletion of hnRNPK in Min6 cells treated with glucolipotoxic conditions for 30hrs as determined by RNA immunoprecipitation for DDX3X (n=3). * = p < 0.05 and ns = not significant. P values were calculated by two-way ANOVA except in (b) in which unpaired two- tailed Student’s t-tests were used. In (c,d), * denotes significance compared to ROSA26 group of the same treatment unless otherwise noted. For Western blot images of JUND, arrows denote two bands for JUND and * denotes a non-specific band. Otherwise, * = p < 0.05.

130

4.5 Discussion

Although the transcriptional regulators that shape the proper β cell gene expression program have received much attention, the factors important for post- transcriptional regulation in β cells, such as RBPs, are less defined but may be critical for adaptation to environmental stressors. Here, an unbiased search for putative regulatory motifs led to the identification of hnRNPK as a post- transcriptional regulator of JUND expression. While we implicate hnRNPK as a regulator of translation in β cells, its highly multi-functional nature suggests it likely serves other roles as well, such as regulating transcription, splicing, and mRNA stability (Bomsztyk et al., 2004). This versatility is intriguing in that hnRNPK can integrate signaling pathways with multiple aspects of RNA processing, but it also complicates the design and interpretation of loss-of- function studies. Nevertheless, given its activation during metabolic stress in vivo, a systematic analysis of hnRNPK function holds promise to elucidate novel aspects of β cell biology. By focusing on hnRNPK interacting factors, we identified the RNA helicase DDX3X as a novel hnRNPK partner that participates in the regulation of JUND translation. While the function of DDX3X in β cells is unknown, its links to cell cycle progression, apoptosis, and innate immunity in other cell types (Ariumi, 2014) warrant a broader examination of DDX3X function in β cells.

131 hnRNPK can be phosphorylated downstream of multiple signaling pathways, including Protein kinase C, JNK, and ERK (Habelhah et al., 2001; Schullery et al., 1999). In β cells, we have found that activation of the MEK/ERK signaling pathway is both necessary and sufficient to increase hnRNPK phosphorylation and induce JUND levels (Figure 4.11). Furthermore, in the context of metabolic stress, inhibition of this ERK/hnRNPK/JUND axis with trametinib reduced apoptosis in mouse islets. This is consistent with previous findings that MEK inhibition reduced apoptosis in human islets treated with high glucose levels or

IL-1β (Maedler et al., 2004). In contrast, activation of the MEK/ERK signaling pathway has been shown to promote β cell survival in the context of incretin signaling (Campbell et al., 2016; Quoyer et al., 2010). Thus, the impact of MEK inhibition on β cell viability is context-dependent. Interestingly, treatment of ob/ob mice with trametinib improves glucose homeostasis in these animals, however, this effect appears to be largely attributable to an improvement in insulin sensitivity (Banks et al., 2015). Thus, the use of MEK inhibition in metabolic syndrome warrants further investigation for improving both insulin sensitivity and

β cell viability.

132

Figure 4.11. Model diagram depicting an ERK/hnRNPK/JUND axis that promotes oxidative stress and apoptosis in β cells.

133

To advance our understanding of translational regulation mediated by hnRNPK, we systematically evaluated modifiable properties of hnRNPK during metabolic stress including its subcellular localization, RNA binding affinity, and nuclear retention of mRNA. None of these features, however, could explain the induction of JUND during glucolipotoxicity. This led us to consider the possibility that hnRNPK regulates translation via a cofactor. To further explore this prospect, we utilized published mass spectrometry data sets for hnRNPK interacting factors

(Chen et al., 2002; Mikula et al., 2015), however, this included nearly 200 proteins. Accordingly, we focused our search on RNA helicases because JUND has been reported to contain a highly structured 5’UTR (Short and Pfarr, 2002), which narrowed our search to three candidates: DHX9, DDX1, and DDX3X.

DHX9 has previously been shown to regulate JUND translation in COS cells

(Hartman et al., 2006) and it was highly enriched for hnRNPK binding in one mass spectrometry data set (Mikula et al., 2015); therefore this target seemed particularly promising. Surprisingly, however, depletion of DHX9 had no impact on the induction of JUND during glucolipotoxicity or on its expression under baseline conditions in Min6 cells. One possibility is that other DEAD-box helicases expressed in β cells provide redundancy for DHX9. Indeed, the DEAD- box protein family contains 37 members in humans, many of which regulate translation (Linder and Jankowsky, 2011).

In contrast to DHX9, loss of DDX3X clearly reduced JUND translation in Min6 cells based on reductions in both protein levels and ribosome occupancy. In

134 other cell types, DDX3X has been implicated in translational regulation of specific mRNAs, such as cyclin E1 (Lai et al., 2010), and several viral genes (ie: HIV,

Japanese encephalitis virus, etc.) (Soto-Rifo et al., 2012). In Min6 cells, DDX3X depletion significantly reduced the ribosome occupancy of JUND, but not BETA-

ACTIN, mRNA, indicating that loss of DDX3X impacts the translation of specific genes. In the future, it would be interesting to define the full catalog of DDX3X targets in β cells by TRAP-seq. Additionally, we cannot rule out the possibility that loss of DDX3X also has an effect on global translation rates, which would not be picked up by TRAP analyses.

We also investigated the mechanism by which DDX3X and hnRNPK may cooperate to impact translation. Under metabolic stress, increased phosphorylation of hnRNPK does not change its binding to the JUND mRNA.

Instead, we considered the possibility that phosphorylation of hnRNPK may enhance its interaction with DDX3X, thus recruiting the helicase to JUND mRNA in a stress-dependent manner. However, co-immunoprecipitation experiments showed no change in the interaction between these factors under control versus glucolipotoxic conditions (Figure 4.10).

The mechanisms by which RNA helicases impact translation of specific mRNAs include remodeling secondary structures in the 5’UTR and recruitment of the translation pre-initiation complex (PIC) (Marintchev, 2013). DDX3X has been shown to directly bind to the 18S ribosomal RNA, a component of the PIC (Oh et al., 2016). This led us to hypothesize that hnRNPK may impact the interaction 135 between DDX3X and the PIC. Interestingly, during metabolic stress, there was increased interaction between DDX3X and the 18S ribosomal RNA, and this interaction was significantly reduced in Min6 cells with loss of hnRNPK. This suggests that hnRNPK may influence translation in part by facilitating recruitment of the PIC via DDX3X. It is unclear how hnRNPK impacts the DDX3X-PIC interaction. One possibility is that hnRNPK promotes formation of the complex by directly recruiting the PIC in a stress-dependent manner. On the other hand, hnRNPK could indirectly facilitate this interaction via assembly of a larger complex that includes DDX3X and the PIC, and the composition of this complex may be altered during metabolic stress in an hnRNPK- and ERK-dependent manner (Figure 4.12). Additionally, this complex may contain cofactors that modulate DDX3X activity, as has been reported for EZRIN and GLE1 (Bolger and Wente, 2011; Çelik et al., 2015). Thus, it would be interesting to more comprehensively define the hnRNPK and DDX3X interactome in β cells under control and glucolipotoxic conditions to identify other components of this hnRNPK-DDX3X complex, and to determine whether they are recruited in a stress-dependent manner.

136

Figure 4.12. Model diagram depicting recruitment of the PIC by an hnRNPK-DDX3X complex during metabolic stress.

137

Another mechanistic detail that deserves further elaboration is the localization of the DDX3X-hnRNPK complex on mRNA. Our de novo motif analysis and the eCLIP data for hnRNPK indicate that it binds to the 3’UTR of the JUND mRNA.

DDX3X, on the other hand, has been reported to predominantly bind to the

5’UTR of genes (Oh et al., 2016). Therefore, one possibility is that the interaction occurs while DDX3X and hnRNPK are bound to the 5’ and 3’ UTR, respectively, thus facilitating circularization of the JUND mRNA (Figure 4.11) similar to that mediated by PABP and eIF4G to promote efficient translation initiation

(Kahvejian et al., 2005). Interestingly, another binding partner for DDX3X is

PABP (Soto-Rifo et al., 2012), however, it is unclear if this factor is part of the hnRNPK-DDX3X complex. Circularization of mRNA has been proposed to stimulate ribosome recycling (Dever and Green, 2012), thus it is enticing to speculate that hnRNPK could promote the re-assembly of initiation complexes after termination. Another possibility is that the hnRNPK-DDX3X interaction occurs independent of their respective binding to RNA. In this situation, hnRNPK may prime a translation initiation complex by facilitating the interaction between

DDX3X and the PIC prior to loading onto mRNA.

Translation initiation can occur independent of binding to the 5’ cap, and this mode of translation initiation is particularly important during certain stress conditions due to sequestration of eIF4E (Spriggs et al., 2010). It is unclear whether translation initiation regulated by the DDX3X-hnRNPK complex can occur independent of binding to the 5’ cap. Interestingly, hnRNPK mediates

138 translation initiation at some cellular IRESs, such as in the 5’UTR of c-myc

(Evans et al., 2003). Similarly, DDX3X is critical for translation initiation at several viral IRESs, including that of hepatitis C virus (Geissler et al., 2012). Thus, it will be interesting to examine whether this complex functions during cap- independent conditions to enhance JUND translation.

It has recently been shown that loss of Ded1, the yeast homolog of DDX3X, promotes alternative start codon usage in the 5’ UTR of certain genes (Guenther et al., 2018). Intriguingly, close examination of western blots for JUND indicates the appearance of an additional faint band after depletion of DDX3X (Figure

4.9c). One possibility is that DDX3X prevents the inappropriate translation initiation at alternative start codons in the 5’UTR of JUND. The mapping of translation initiation sites can be achieved using ribosome profiling after treatment of cells with the drug harringtonine to freeze ribosomes at sites of initiation (Ingolia et al., 2011). This approach could be used to determine if loss of

DDX3X dysregulated start codon usage in β cells on a genome-wide scale. This would be particularly interesting to assess in β cells because it has been suggested that translation from an alternative initiation site in the insulin mRNA produces an immunogenic peptide that may contribute to the development of type 1 diabetes (Kracht et al., 2017).

139

CHAPTER 5: CONCLUDING REMARKS AND FUTURE DIRECTIONS

5.1 Discussion of key findings

The dysfunction and apoptosis of pancreatic β cells are central to the progression of T2D, thus identifying new strategies to ameliorate these processes holds promise to reduce the burden of this chronic disease. In this work, I have approached this goal by first studying the mechanisms underlying β cell adaptation to stress conditions that are pertinent to diabetes. By focusing on translational regulation, I uncovered a novel ERK/hnRNPK/JUND axis that is activated during metabolic stress. There were several advantages to the translation-centric approach used in these studies. First, the translational induction of JUND was not accompanied by an increase in mRNA abundance, therefore examining the transcriptome would not have identified JUND as a stress-responsive factor. Second, assessing translational regulation on a genome-wide scale with TRAP-seq allowed me to probe for upstream regulators of translation by searching for common features of genes with changes in ribosome occupancy. This led to the identification of a poly(C) motif that was enriched in 3’UTRs, which implicated the PCBP family members in post- transcriptional regulation in β cells.

The initial TRAP screen led me to investigate the role of JUND in β cells. Given its previous link to regulating antioxidant genes, I hypothesized that JUND induction during metabolic stress was a protective mechanism, and that loss of

140

JUND would exacerbate ROS accumulation. Surprisingly, however, I found the opposite effect: depletion of JUND blocked the increase in oxidative stress during glucolipotoxicity.

As a transcription factor, the role of JUND in a particular cell type is completely dependent on the cohort of genes it regulates. Consistent with its effect on oxidative stress, loss of JUND in β cells led to the downregulation of pro-oxidant, rather than anti-oxidant genes. The mechanism underlying this cell type-specific role for JUND is likely related to its array of interacting factors. JUND binds to

DNA as a homo- or hetero-dimer with other members of the Jun transcription factor family or with members of the Fos, Atf, and Maf transcription factor families, and the composition of these dimers will dictate the DNA binding preference and transcriptional effect of JUND (Mechta-Grigoriou et al., 2001).

Thus, the discrepancy in transcriptional targets between cell types may be attributable to the relative expression of JUND binding partners.

Furthermore, I used RNA-seq to perform an unbiased search for JUND targets, which led to the identification of a relatively small cohort of genes, many of which have been implicated in ROS production and inflammation. Strikingly, this set of genes significantly overlapped with genes upregulated in various models of β cell dysfunction, including islets from db/db mice, human islets treated with palmitate, and Min6 cells with PDX1 deficiency. This supports a model in which JUND regulates genes implicated in β cell dysfunction and suggests my findings may be generalizable and not specific to the model of glucolipotoxicity. 141

Lastly, if JUND were playing an antioxidant role in β cells, as has been suggested in other cell types, one would predict an exacerbation of stress- induced cell death with loss of JUND because β cells are particularly sensitive to oxidative stress (Lenzen et al., 1996). Instead, I observed that depletion of JUND in β cells significantly blunted apoptosis during metabolic stress, consistent with the proposed pro-oxidant role for JUND. Collectively, these results significantly strengthen the novel finding that JUND has a maladaptive function in β cells by promoting oxidative stress.

I also identified the MEK/ERK signaling pathway as being necessary and sufficient to activate the hnRNPK/JUND pathway in β cells. Given the pro- apoptotic role ascribed to JUND during metabolic stress, I investigated whether blocking JUND induction via MEK inhibition would also improve cell viability and found a striking reduction in apoptosis during glucolipotoxicity. While this result is consistent with the effect of JUND depletion on apoptosis, it is unclear to what extent the benefit of MEK inhibition can be attributed to reduced JUND levels.

Inhibition of the MEK/ERK signaling pathway will have broad effects on the cell, therefore it is very likely that other downstream targets also contribute to this phenotype. For example, ERK-mediated upregulation of p53 has been found to increase levels of pro-apoptotic factors, such as BAX (Cagnol and Chambard,

2009).

142

Given these pro-oxidant and pro-apoptotic roles for JUND in β cells, I reasoned that advancing our mechanistic understanding of JUND induction could provide new targets for diabetes therapy. To this end, I utilized publicly available data sets of hnRNPK interacting factors to identify candidate cofactors, leading to the discovery that the RNA helicase DDX3X is required for efficient JUND translation. This factor can impact translation via its ATP-dependent helicase activity or by recruiting components of the translation pre-initiation complex (PIC)

(Marintchev, 2013). Interestingly, there was increased interaction between

DDX3X and the 18S ribosomal RNA, a component of the PIC, during metabolic stress. This indicates that the function of DDX3X is regulated during glucolipotoxicity. Further, loss of hnRNPK significantly reduced the interaction between DDX3X and 18S. These findings suggest a model in which an hnRNPK-

DDX3X complex increases JUND translation in part via recruitment of the PIC.

To study translational regulation in the setting of β cell dysfunction, I employed the models of PDX1 deficiency and glucolipotoxicity. While my data implicate the

PCBP family members as translational regulators in both of these models, the mechanisms by which the PCBPs impact translation are likely distinct in these two scenarios. For example, NKX2-2 exhibited increased translation during

PDX1 deficiency, but I did not detect any induction of NKX2-2 during glucolipotoxicity. Furthermore, I did not observe any change in phosphorylation of hnRNPK during PDX1 deficiency, unlike glucolipotoxicity.

143

It is currently unclear how depletion of PDX1 affects the ribosome occupancy of the poly(C)-containing mRNAs. One possibility is that PDX1 controls expression of a component of the translation machinery that is not required for genes with

PCBP binding in their 3’UTR. For example, PDX1 deficiency causes a modest reduction (~50%) in eIF1AX and eIF1AD, which have been implicated in PIC recruitment and start codon selection (Passmore et al., 2007), similar to DDX3X

(Guenther et al., 2018). Therefore, one interesting possibility is that genes regulated by an hnRNPK-DDX3X complex can “escape” translational repression caused by reductions in eIF1AX/D. However, PDX1 regulates a very broad set of genes in β cells, so there are many possible mechanisms that could explain altered translation of the poly(C)-containing genes. Indeed, I have tested a number of potential mechanisms based on genes regulated by PDX1 but have been unable to establish a connection to the post-transcriptional regulation of

NKX2-2 or JUND. These mechanisms include: sequestration by a decoy RNA

(insulin or PDX1 mRNAs), competition with other RBPs (PTB or YBX1), loss of eIF4H, and altered recruitment of BRG1 or the long-noncoding RNA NEAT1.

5.2 Additional screens for translational regulation in β cells

In the current work, I have used TRAP-seq to study translational regulation in

Min6 cells with PDX1 deficiency. This approach was successful in that it led to the identification of JUND as a translationally regulated gene and suggested a role for the PCBP family of RBPs in mediating this post-transcriptional regulation.

However, there are a number of additional screens one could imagine to further

144 investigate translational regulation in β cells. Given the induction of JUND during glucolipotoxicity, it would be interesting to determine how widespread translatome remodeling is in this model. Alternatively, TRAP could be performed in Min6 cells treated with conditions to invoke particular forms of stress, such as thapsigargin/tunicamycin (ER stress) or hydrogen peroxide/nitric oxide

(oxidative/nitrosative stress). These treatment strategies may be favorable over glucolipotoxicity as they would be focused on a particular stress response and therefore be potentially less complex. For example, glucolipotoxicity may simultaneously invoke a number of pathways in β cells, such as responses to ER stress, oxidative stress, and mitochondrial dysfunction.

Beyond additional TRAP-seq studies, a very exciting extension of this work would be to use ribosome profiling (or an adaptation of this method) to globally assess start codon usage in β cells. Alternative translation initiation appears to be widespread and can increase the diversity of proteins expressed in a cell

(Ingolia et al., 2011). Furthermore, start codon usage may be altered under stress or disease conditions (Kearse and Wilusz, 2017; Sendoel et al., 2017).

However, mapping of start codon usage to identify alternative initiation sites under stress conditions has not been performed in β cells. This is particularly intriguing in the context of type 1 diabetes as it has been suggested that translation of a non-canonical open reading frame in the insulin mRNA may generate an immunogenic peptide contributing to disease onset (Kracht et al.,

2017). Thus, one can imagine the power of this approach to uncover novel

145 mechanisms of translational regulation that may be altered under conditions associated with diabetes.

Another promising extension of my work is the use of TRAP to assess translational regulation during stress conditions in vivo, such as high fat diet feeding. A mouse has been generated to express the GFP-RPL10A transgene in a Cre-driven manner (Liu et al., 2014). Thus, the use of a β cell-specific Cre line will allow for the purification of ribosome-bound mRNA specifically from β cells.

Limiting material may cause this application to be technically challenging, however, the continued advancement of sequencing technologies suggest it will likely be possible in the near future.

5.3 Assessment of RBP functions in β cells

The work presented here includes the first characterization of functional roles for the PCBP family members in β cells. Importantly, I came upon this family of

RBPs through an unbiased search for regulatory motifs in our TRAP screen.

Using in vitro models, I have implicated the members hnRNPK and PCBP1/2 in several aspects of β cell biology, including post-transcriptional control of gene expression, maintenance of cell identity, and glucose-stimulated insulin secretion. Thus, an important next step will be the use of mouse models to genetically delete these factors in vivo, which is ongoing work in the Stoffers’ laboratory. To connect phenotypic findings to PCBP function, a number of approaches will be required, including CLIP-seq and transcriptome/translatome

146 assessments to identify pertinent target genes. Given the multifunctional nature of these RBPs, they may have broad roles in β cells, however, a careful and systematic analysis of their function holds great promise to uncover novel aspects of β cell biology.

Furthermore, there is an extensive array of RBPs in the mammalian genome that control all aspects RNA processing. However, we are only just beginning to understand how this class of proteins may shape critical processes involved in β cell biology. Thus, it will be important to undertake studies aimed at uncovering novel roles for RBPs in β cells. In this work, I have accomplished this by using a de novo motif analysis to narrow our focus on RBPs with specificity for cytosine- rich sequences. Similarly, new screens for translationally regulated genes could also be analyzed to search for enriched regulatory motifs. This bottom-up approach will be strengthened by the increasing number of publically available data sets to globally define RBP targets by eCLIP (Van Nostrand et al., 2016), which will allow for the assessment of gene sets for enrichment of RBP binding, not just motif frequency.

Alternatively, novel roles for RBPs in β cells could be identified by a top-down approach, which would start with screening for RBPs with unique characteristics in β cells. For example, transcriptomic, proteomic, or phospho-proteomic data sets could be mined to identify RBPs with altered abundance or modifications in conditions associated with diabetes. Similarly, relative expression could be

147 assessed to identify RBPs with especially high expression levels in β cells compared to other cell types.

5.4 Expanding the analysis of JUND function and regulation

In β cells, JUND regulates a cohort of genes implicated in redox imbalance and inflammation, and these genes are commonly dysregulated in models of β cell dysfunction. However, the extent to which each of these genes contributes to β cell apoptosis is unknown. For example, NOS2, a nitric oxide synthase, has been implicated in mediating β cell damage in the context of cytokine treatment

(Zumsteg et al., 2000). However, nitric oxide may play beneficial roles in β cells under certain circumstances by neutralizing other oxidant types (Broniowska et al., 2015). Thus, the induction of several pro-oxidant genes by JUND may lead to complex downstream effects on redox homeostasis based on the types and subcellular locations of the generated oxidants. While we believe it is unlikely that the observed pro-apoptotic role for JUND can be entirely ascribed to one particular target gene, close scrutiny of these genes may provide new insights into redox imbalance during metabolic stress.

It should also be noted that the oxidative stress assay used in this work

(CellROX) is not specific to any particular oxidant, but rather detects a range of

ROS types. This was a useful approach to screen for general oxidative stress levels; however, follow-up studies using more specific assays could provide better insight into which oxidants are upregulated during glucolipotoxicity in a

148

JUND-dependent manner. For example, dihydroethidium and spin trapping of nitrogen dioxide can be used to specifically detect superoxide and nitric oxide levels, respectively (Pace and Kalyanaraman, 1993; Zhao et al., 2003). These analyses may also shed light on which JUND targets are contributing to oxidative stress during glucolipotoxicity.

Besides its pro-oxidant role, JUND also regulates several pro-inflammatory genes with connections to islet inflammation, including Ccl2, Cxcl1, and Cxcl2.

Interestingly, these genes have been linked to poor outcomes for islet transplantation (Citro et al., 2012; Piemonti et al., 2002), suggesting that modulation of JUND levels could be a novel approach to advance this clinically relevant application. It is unlikely that the induction of these pro-inflammatory genes contributed to the phenotypes observed in our in vitro and ex vivo systems because their main function involves immune cell recruitment. Nevertheless, we cannot rule out the possibility that these genes also have cell-autonomous roles in β dysfunction. To better assess the relevance of the pro-inflammatory gene signature, however, it will be important to investigate the benefit of JUND depletion or MEK inhibition using in vivo islet transplantation models.

Given its impact on oxidative stress and apoptosis, the role of JUND in β cell compensation during insulin resistance should also be studied. This will require the generation of a conditional null allele for JUND. Crossing this mouse with a β cell-specific Cre line will allow for the in vivo assessment of JUND function in β cells. Based on our ex vivo work, we predict that loss of JUND would dampen 149 oxidative stress in β cells during a high fat diet challenge, thus improving insulin secretion and blood glucose homeostasis. However, it is possible that JUND also regulates insulin secretion independent of its effect on oxidative stress and apoptosis. Thus, a careful and thorough characterization of this in vivo model will be required to assess the impact of JUND deletion on insulin secretion, oxidative stress, cell survival, and proliferation during high fat feeding.

Similar to JUND depletion, I found a significant improvement in islet cell survival during metabolic stress after treatment with the MEK inhibitor trametinib. The application of this finding to in vivo models of diabetes is complicated by the fact that trametinib treatment has been show to improve insulin resistance (Banks et al., 2015). Thus, it will be impossible to discern whether any improvements in insulin secretion or β cell viability seen with systemic trametinib administration can be attributed to MEK inhibition in β cells or an indirect effect due to reduced insulin demand. One approach to more specifically study the role of the

MEK/ERK signaling pathway in β cells in vivo would be to generate a mouse model with a β cell-specific, inducible overexpression of a dominant negative form of MEK1, which blocks ERK phosphorylation.

The connections between the PCBPs and JUND regulation include the presence of a poly(C) stretch in the JUND 3’UTR, binding of the PCBP members to the

JUND mRNA, and loss of PCBPs leading to altered JUND steady-state protein levels. Together, these findings suggest that the post-transcriptional regulation of

150

JUND is directly mediated by the PCBPs via 3’UTR binding. To further support this model, however, it would be pertinent to show that deletion of the JUND poly(C) motif impairs its post-transcriptional regulation. A common approach to address this question is to generate a reporter gene flanked by the JUND UTRs containing deletions of putative regulatory elements, including the poly(C) motif.

Unfortunately, I have been unable to develop a model system to recapitulate the induction of JUND during glucolipotoxicity despite many attempts, thus I could not confidently assess the functionality of the poly(C) motif. Alternatively, genome editing approaches could be used to delete the cytosine-rich sequence element in the endogenous gene. It is unclear whether this would be feasible in Min6 cells as it would require single cell cloning. On the other hand, a mouse model with tissue-specific and inducible deletion of the JUND poly(C) motif could be used to study whether loss of this sequence element abrogates JUND induction in isolated islets treated with glucolipotoxicity. If so, it would be interesting to see if this deletion also impacts β cell viability during in vivo stress conditions, such as high fat feeding.

Lastly, I have established that JUND induction during metabolic stress is a conserved process using isolated mouse and human islets. To extend these findings to the pathogenesis of T2D in humans, it will be critical to evaluate JUND levels in pancreatic tissue samples from diabetic patients. This analysis is typically performed using immunofluorescence staining of pancreatic sections from deceased organ donors. Our ability to perform this experiment is currently

151 limited by the lack of an antibody to confidently detect JUND levels by staining.

Instead, we have relied on Western blot analyses to assess JUND abundance, including for isolated islets from db/db mice. Similarly, we could compare JUND levels in isolated islets from diabetic and non-diabetic human organ donors, however, it is possible that the culturing period after islet isolation would provide time for JUND levels to normalize between groups. Thus, it will be advisable to optimize staining conditions for additional antibodies to determine if JUND levels are altered in islets from diabetic individuals.

5.5 Translational regulation mediated by an hnRNPK-DDX3X complex

To expand our understanding of the factors regulating JUND translation in an hnRNPK-dependent manner, I searched the hnRNPK interactome using publicly available mass spectrometry data sets. Consistent with its highly multifunctional nature, hnRNPK has a broad array of binding partners (Mikula et al., 2015). To narrow my search, I capitalized on the fact that JUND is known to have a highly structured 5’UTR that requires helicase activity for efficient translation (Hartman et al., 2006), leading me to screen three helicase genes. Of these, I found that only DDX3X was required for efficient translation of JUND in β cells. I propose a model in which hnRNPK and DDX3X form a complex that recruits the translation pre-initiation complex (PIC) to the JUND mRNA. Further experimentation will be required to clarify several mechanistic details of this model.

First, it is currently unclear whether the hnRNPK-DDX3X complex regulates translation in a cap-dependent manner. Several studies, along with my own data, 152 indicate that there is a global suppression of translation in β cells during glucolipotoxicity (Hatanaka et al., 2014). This is likely attributed both to the phosphorylation of eIF2α and the induction of 4E-BP1 levels, which sequesters eIF4E and reduces cap-dependent translation (Yamaguchi et al., 2008). Thus,

JUND may “escape” translational suppression during glucolipotoxicity via cap- independent translation mediated by the hnRNPK-DDX3X complex. The relative requirement for the 5’ cap in translation initiation is often assessed by inserting

5’UTR sequences into a bicistronic reporter assay. Given my challenges to model JUND translation using reporter systems, however, it is unclear whether this approach will be useful to assess cap-independent mechanisms.

One difference between the effect of hnRNPK and DDX3X on JUND translation is that hnRNPK is only required for the induction of JUND during metabolic stress but does not impact JUND regulation under baseline conditions. On the other hand, loss of DDX3X impairs JUND translation both under baseline and stress conditions. One possible explanation for this discrepancy is that DDX3X can interact with an additional factor or factors under baseline conditions that can functionally substitute for hnRNPK, and this unknown factor may be lost or inactivated under stress conditions. For example, DDX3X interacts with PABP

(Soto-Rifo et al., 2012), which binds to the 3’UTR of the JUND mRNA via the polyA tail. Thus, under baseline conditions, PABP and hnRNPK may play redundant roles in JUND translation via their interactions with DDX3X. During stress conditions, however, PABP may be inactivated or reduced, leading to the

153 dependency on hnRNPK for efficient JUND translation. PABP can be regulated by its expression level, proteolytic cleavage, post-translational modification, and the expression of inhibitory factors, such as PAIP2 (McKinney et al., 2013; Walsh et al., 2013). However, it is currently unknown if any change in PABP expression or function occurs during glucolipotoxicity.

While I chose DDX3X as a candidate based on its known helicase function, it is unclear whether this enzymatic activity of DDX3X is required for JUND translation. Indeed, some DEAD-box proteins regulate translation of specific mRNAs via recruitment of the PIC, which is independent of their helicase activity

(Marintchev, 2013). Thus, it will be worthwhile to mutate the functional domains of DDX3X to further examine how this protein regulates JUND translation.

Additionally, DDX3X has been implicated in the proper selection of translation initiation sites (Guenther et al., 2018). Interestingly, I have found that loss of

DDX3X in Min6 cells leads to the appearance of an additional faint band for

JUND on western blot, which may represent translation from an alternative start codon. To expand on this observation, ribosome profiling could be used to globally map translation initiation sites in control and DDX3X-depleted β cells.

Additionally, the precise mechanism by which hnRNPK may impact DDX3X function deserves closer examination. I considered the possibility that hnRNPK recruits DDX3X to the JUND mRNA in a stress-dependent manner, but I saw no change in the hnRNPK-DDX3X interaction during glucolipotoxicity by co- immunoprecipitation experiments. Instead, I found that DDX3X interacts with the 154

18S ribosomal RNA, a component of the PIC, in an hnRNPK- and stress- dependent manner. However, it is unclear how hnRNPK regulates the interaction between DDX3X and the PIC. To better understand the components of this complex and how they change under metabolic stress, it would be informative to perform mass spectrometry on hnRNPK- and DDX3X-interacting factors under control and glucolipotoxic conditions.

I have implicated DDX3X in regulating JUND translation in β cells, but whether this RNA helicase impacts β cell function or viability is unknown. DDX3X likely regulates a range of targets in β cells, thus it is unclear whether loss of this factor will have any overlapping phenotypes with those we have observed for JUND depletion. Given its links to cell cycle progression, apoptosis, and innate immunity in other cell types (Ariumi, 2014), however, a broader examination of

DDX3X function in β cells is warranted. Anecdotally, CRISPR-mediated depletion of DDX3X in Min6 cells leads to a significant reduction in cell number over time.

This may be attributable to a reduction in cell proliferation as DDX3X has previously been shown to enhance translation of the mRNA encoding cyclin E1

(Lai et al., 2010). To better define the role of DDX3X in β cells, loss-of-function studies should be performed in primary islets or in vivo to assess the impact on cell survival, proliferation, and insulin secretion. Furthermore, a broad examination of genes regulated by DDX3X should be carried out using TRAP- seq or ribosome profiling to provide mechanistic insight into how this RNA helicase impacts β cell biology. 155

BIBLIOGRAPHY

Accili, D., Talchai, S.C., Kim-Muller, J.Y., Cinti, F., Ishida, E., Ordelheide, A.M., Kuo, T., Fan, J., and Son, J. (2016). When β-cells fail: lessons from dedifferentiation. Diabetes, Obesity and Metabolism 18, 117–122.

Agarwal, S.K., Guru, S.C., Heppner, C., Erdos, M.R., Collins, R.M., Park, S.Y., Saggar, S., Chandrasekharappa, S.C., Collins, F.S., Spiegel, A.M., et al. (1999). Menin interacts with the AP1 transcription factor JunD and represses JunD-activated transcription. Cell 96, 143–152.

American Diabetes Association (2018). Economic Costs of Diabetes in the U.S. in 2017. Diabetes Care 41, 917–928.

Ariumi, Y. (2014). Multiple functions of DDX3 RNA helicase in gene regulation, tumorigenesis, and viral infection. Front Genet 5, 423.

Bailey, T.L., Boden, M., Buske, F.A., Frith, M., Grant, C.E., Clementi, L., Ren, J., Li, W.W., and Noble, W.S. (2009). MEME SUITE: tools for motif discovery and searching. Nucleic Acids Research 37, W202–W208.

Banks, A.S., McAllister, F.E., Camporez, J.P.G., Zushin, P.-J.H., Jurczak, M.J., Laznik- Bogoslavski, D., Shulman, G.I., Gygi, S.P., and Spiegelman, B.M. (2015). An ERK/Cdk5 axis controls the diabetogenic actions of PPARγ. Nature 517, 391–395.

Bolger, T.A., and Wente, S.R. (2011). Gle1 is a multifunctional DEAD-box protein regulator that modulates Ded1 in translation initiation. Journal of Biological Chemistry 286, 39750–39759.

Bomsztyk, K., Denisenko, O., and Ostrowski, J. (2004). hnRNP K: One protein multiple processes. Bioessays 26, 629–638.

Brissova, M., Shiota, M., Nicholson, W.E., Gannon, M., Knobel, S.M., Piston, D.W., Wright, C.V.E., and Powers, A.C. (2002). Reduction in pancreatic transcription factor PDX-1 impairs glucose-stimulated insulin secretion. J. Biol. Chem. 277, 11225–11232.

Broniowska, K.A., Oleson, B.J., McGraw, J., Naatz, A., Mathews, C.E., and Corbett, J.A. (2015). How the location of superoxide generation influences the β-cell response to nitric oxide. Journal of Biological Chemistry 290, 7952–7960.

Butler, A.E., Janson, J., Bonner-Weir, S., and Ritzel, R. (2003). β-Cell deficit and increased β-cell apoptosis in humans with type 2 diabetes. Diabetes.

Cagnol, S., and Chambard, J.-C. (2009). ERK and cell death: Mechanisms of ERK-induced cell death - apoptosis, autophagy and senescence. FEBS Journal 277, 2–21.

Campbell, J.E., Ussher, J.R., Mulvihill, E.E., Kolic, J., Baggio, L.L., Cao, X., Liu, Y., Lamont, B.J., Morii, T., Streutker, C.J., et al. (2016). TCF1 links GIPR signaling to the control of beta cell function and survival. Nature Medicine 22, 84–90.

Campeau, E., Ruhl, V.E., Rodier, F., Smith, C.L., Rahmberg, B.L., Fuss, J.O., Campisi, J., Yaswen, P., Cooper, P.K., and Kaufman, P.D. (2009). A versatile viral system for expression and depletion of proteins in mammalian cells. PLoS ONE 4, e6529.

156

Chaudhury, A., Chander, P., and Howe, P.H. (2010a). Heterogeneous nuclear ribonucleoproteins (hnRNPs) in cellular processes: Focus on hnRNP E1's multifunctional regulatory roles. Rna 16, 1449–1462.

Chaudhury, A., Hussey, G.S., Ray, P.S., Jin, G., Fox, P.L., and Howe, P.H. (2010b). TGF-|[beta]|- mediated phosphorylation of hnRNP E1 induces EMT via transcript-selective translational induction of Dab2 and ILEI. Nature Cell Biology 12, 286–293.

Chen, H.-C., Lin, W.-C., Tsay, Y.-G., Lee, S.-C., and Chang, C.-J. (2002). An RNA helicase, DDX1, interacting with poly(A) RNA and heterogeneous nuclear ribonucleoprotein K. J. Biol. Chem. 277, 40403–40409.

Cinti, F., Bouchi, R., Kim-Muller, J.Y., Ohmura, Y., Rodrigo Sandoval, P., Masini, M., Marselli, L., Suleiman, M., Ratner, L.E., Marchetti, P., et al. (2015). Evidence of β-cell Dedifferentiation in Human Type 2 Diabetes. J. Clin. Endocrinol. Metab. jc.2015–2860–12.

Citro, A., Cantarelli, E., Maffi, P., Nano, R., Melzi, R., Mercalli, A., Dugnani, E., Sordi, V., Magistretti, P., Daffonchio, L., et al. (2012). CXCR1/2 inhibition enhances pancreatic islet survival after transplantation. J. Clin. Invest. 122, 3647–3651.

Claiborn, K.C., Sachdeva, M.M., Cannon, C.E., Groff, D.N., Singer, J.D., and Stoffers, D.A. (2010). Pcif1 modulates Pdx1 protein stability and pancreatic β cell function and survival in mice. J. Clin. Invest. 120, 3713–3721.

Cnop, M., Abdulkarim, B., Bottu, G., Cunha, D.A., Igoillo-Esteve, M., Masini, M., Turatsinze, J.-V., Griebel, T., Villate, O., Santin, I., et al. (2014). RNA Sequencing Identifies Dysregulation of the Human Pancreatic Islet Transcriptome by the Saturated Fatty Acid Palmitate. Diabetes 63, 1978– 1993.

Çelik, H., Sajwan, K.P., Selvanathan, S.P., Marsh, B.J., Pai, A.V., Saygideger Kont, Y., Han, J., Minas, T.Z., Rahim, S., Erkizan, H.V., et al. (2015). Ezrin binds to DEAD-box RNA helicase DDX3 and regulates its function and protein level. Molecular and Cellular Biology 107, MCB.00332–15– 18.

Del Guerra, S., Lupi, R., Marselli, L., Masini, M., Bugliani, M., Sbrana, S., Torri, S., Pollera, M., Boggi, U., Mosca, F., et al. (2005). Functional and Molecular Defects of Pancreatic Islets in Human Type 2 Diabetes. Diabetes 54, 727–735.

Delépine, M., Nicolino, M., Barrett, T., Golamaully, M., Lathrop, G.M., and Julier, C. (2000). EIF2AK3, encoding translation initiation factor 2-alpha kinase 3, is mutated in patients with Wolcott-Rallison syndrome. Nature Genetics 25, 406–409.

Dever, T.E., and Green, R. (2012). The Elongation, Termination, and Recycling Phases of Translation in Eukaryotes. Cold Spring Harb Perspect Biol 4, a013706–a013706.

Doliba, N.M., Liu, Q., Li, C., Chen, P., Liu, C., Naji, A., and Matschinsky, F.M. (2017). Inhibition of cholinergic potentiation of insulin secretion from pancreatic islets by chronic elevation of glucose and fatty acids: Protection by casein kinase 2 inhibitor. Molecular Metabolism 6, 1240–1253.

Dor, Y., and Glaser, B. (2013). Beta-Cell Dedifferentiation and Type 2 Diabetes. N Engl J Med 368, 572–573.

Doyle, J.P., Dougherty, J.D., Heiman, M., Schmidt, E.F., Stevens, T.R., Ma, G., Bupp, S., 157

Shrestha, P., Shah, R.D., Doughty, M.L., et al. (2008). Application of a Translational Profiling Approach for the Comparative Analysis of CNS Cell Types. Cell 135, 749–762.

Doyle, M.J., and Sussel, L. (2007). Nkx2.2 regulates beta-cell function in the mature islet. Diabetes 56, 1999–2007.

Eiring, A.M., Harb, J.G., Neviani, P., Garton, C., Oaks, J.J., Spizzo, R., Liu, S., Schwind, S., Santhanam, R., Hickey, C.J., et al. (2010). miR-328 Functions as an RNA Decoy to Modulate hnRNP E2 Regulation of mRNA Translation in Leukemic Blasts. Cell 140, 652–665.

Elsner, M., Gehrmann, W., and Lenzen, S. (2010). Peroxisome-Generated Hydrogen Peroxide as Important Mediator of Lipotoxicity in Insulin-Producing Cells. Diabetes 60, 200–208.

Engin, F., Nguyen, T., Yermalovich, A., and Hotamisligil, G.S. (2014). Aberrant islet unfolded protein response in type 2 diabetes. Scientific Reports 4.

Evans Molina, C., Hatanaka, M., and Mirmira, R.G. (2013). Lost in translation: endoplasmic reticulum stress and the decline of β-cell health in diabetes mellitus. Diabetes, Obesity and Metabolism 15, 159–169.

Evans, J.R., Mitchell, S.A., Spriggs, K.A., Ostrowski, J., Bomsztyk, K., Ostarek, D., and Willis, A.E. (2003). Members of the poly (rC) binding protein family stimulate the activity of the c-myc internal ribosome entry segment in vitro and in vivo. Oncogene 22, 8012–8020.

Feng, Z., Ma, J., and Hua, X. (2017). Epigenetic regulation by the menin pathway. Endocr Relat Cancer 24, T147–T159.

Gao, T., McKenna, B., Li, C., Reichert, M., Nguyen, J., Singh, T., Yang, C., Pannikar, A., Doliba, N., Zhang, T., et al. (2014). Pdx1 Maintains Beta Cell Identity and Function by Repressing an Alpha Cell Program. Cell Metabolism 19, 259–271.

Gehrmann, W., Elsner, M., and Lenzen, S. (2010). Role of metabolically generated reactive oxygen species for lipotoxicity in pancreatic β-cells. Diabetes, Obesity and Metabolism 12, 149– 158.

Geissler, R., Golbik, R.P., and Behrens, S.-E. (2012). The DEAD-box helicase DDX3 supports the assembly of functional 80S ribosomes. Nucleic Acids Research 40, 4998–5011.

Gelain, D.P., Dalmolin, R.J.S., Belau, V.L., Moreira, J.C.F., Klamt, F., and Castro, M.A.A. (2009). A systematic review of human antioxidant genes. Front Biosci (Landmark Ed) 14, 4457–4463.

Gerald, D., Berra, E., Frapart, Y.M., Chan, D.A., Giaccia, A.J., Mansuy, D., Pouysségur, J., Yaniv, M., and Mechta-Grigoriou, F. (2004). JunD Reduces Tumor Angiogenesis by Protecting Cells from Oxidative Stress. Cell 118, 781–794.

Gerstberger, S., Hafner, M., and Tuschl, T. (2014). A census of human RNA-binding proteins. Nature Publishing Group 15, 829–845.

Grassilli, E., Pisano, F., Cialdella, A., Bonomo, S., Missaglia, C., Cerrito, M.G., Masiero, L., Ianzano, L., Giordano, F., Cicirelli, V., et al. (2016). A novel oncogenic BTK isoform is overexpressed in colon cancers and required for RAS-mediated transformation. Oncogene 35, 4368–4378.

158

Guan, B.J., Krokowski, D., Majumder, M., Schmotzer, C.L., Kimball, S.R., Merrick, W.C., Koromilas, A.E., and Hatzoglou, M. (2014). Translational control during endoplasmic reticulum stress beyond phosphorylation of the translation initiation factor eIF2α. Journal of Biological Chemistry 289, 12593–12611.

Guenther, U.-P., Weinberg, D.E., Zubradt, M.M., Tedeschi, F.A., Stawicki, B.N., Zagore, L.L., Brar, G.A., Licatalosi, D.D., Bartel, D.P., Weissman, J.S., et al. (2018). The helicase Ded1p controls use of near-cognate translation initiation codons in 5′ UTRs. Nature 1–20.

Guo, Z., Kozlov, S., Lavin, M.F., Person, M.D., and Paull, T.T. (2010). ATM activation by oxidative stress. Science 330, 517–521.

Gutiérrez, G.D., Bender, A.S., Cirulli, V., Mastracci, T.L., Kelly, S.M., Tsirigos, A., Kaestner, K.H., and Sussel, L. (2017). Pancreatic β cell identity requires continual repression of non-β cell programs. J. Clin. Invest. 127, 244–259.

Habelhah, H., Shah, K., Huang, L., Ostareck-Lederer, A., Burlingame, A.L., Shokat, K.M., Hentze, M.W., and Ronai, Z. (2001). ERK phosphorylation drives cytoplasmic accumulation of hnRNP-K and inhibition of mRNA translation. Nature Cell Biology 3, 325–330.

Hai, T., and Curran, T. (1991). Cross-family dimerization of transcription factors Fos/Jun and ATF/CREB alters DNA binding specificity. Proceedings of the National Academy of Sciences 88, 3720–3724.

Han, J., Song, B., Kim, J., Kodali, V.K., Pottekat, A., Wang, M., Hassler, J., Wang, S., Pennathur, S., Back, S.H., et al. (2015). Antioxidants complement the requirement for protein chaperone function to maintain beta cell function and glucose homeostasis. Diabetes db141357.

Harding, H.P., Novoa, I., Zhang, Y., Zeng, H., Wek, R., Schapira, M., and Ron, D. (2000). Regulated translation initiation controls stress-induced gene expression in mammalian cells. Molecular Cell 6, 1099–1108.

Hartman, T.R., Qian, S., Bolinger, C., Fernandez, S., Schoenberg, D.R., and Boris-Lawrie, K. (2006). RNA helicase A is necessary for translation of selected messenger RNAs. Nat Struct Mol Biol 13, 509–516.

Hatanaka, M., Anderson-Baucum, E., Lakhter, A., Kono, T., Maier, B., Tersey, S.A., Tanizawa, Y., Evans-Molina, C., Mirmira, R.G., and Sims, E.K. (2017). Chronic high fat feeding restricts islet mRNA translation initiation independently of ER stress via DNA damage and p53 activation. Scientific Reports 7, 102–111.

Hatanaka, M., Maier, B., Sims, E.K., Templin, A.T., Kulkarni, R.N., Evans-Molina, C., and Mirmira, R.G. (2014). Palmitate Induces mRNA Translation and Increases ER Protein Load in Islet β-Cells via Activation of the Mammalian Target of Rapamycin Pathway. Diabetes 63, 3404– 3415.

Heiman, M., Schaefer, A., Gong, S., Peterson, J.D., Day, M., Ramsey, K.E., Suárez-Fariñas, M., Schwarz, C., Stephan, D.A., Surmeier, D.J., et al. (2008). A Translational Profiling Approach for the Molecular Characterization of CNS Cell Types. Cell 135, 738–748.

Hetz, C. (2012). The unfolded protein response: controlling cell fate decisions under ER stress and beyond. Nature Publishing Group 13, 89–102.

159

Hetz, C., Chevet, E., and Harding, H.P. (2013). Targeting the unfolded protein response in disease. Nature Publishing Group 12, 703–719.

Huang, C.-J., Lin, C.-Y., Haataja, L., Gurlo, T., Butler, A.E., Rizza, R.A., and Butler, P.C. (2007). High Expression Rates of Human Islet Amyloid Polypeptide Induce Endoplasmic Reticulum Stress–Mediated β-Cell Apoptosis, a Characteristic of Humans With Type 2 but Not Type 1 Diabetes. Diabetes 56, 2016–2027.

Huang, D.W., Sherman, B.T., and Lempicki, R.A. (2009). Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4, 44–57.

Hussey, G.S., Chaudhury, A., Dawson, A.E., Lindner, D.J., Knudsen, C.R., Wilce, M.C.J., Merrick, W.C., and Howe, P.H. (2011). Identification of an mRNP Complex Regulating Tumorigenesis at the Translational Elongation Step. Molecular Cell 41, 419–431.

Hutchins, E.J., and Szaro, B.G. (2013). c-Jun N-Terminal Kinase Phosphorylation of Heterogeneous Nuclear Ribonucleoprotein K Regulates Vertebrate Axon Outgrowth via a Posttranscriptional Mechanism. Journal of Neuroscience 33, 14666–14680.

Ingolia, N.T., Ghaemmaghami, S., Newman, J.R.S., and Weissman, J.S. (2009). Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling. Science 324, 218–223.

Ingolia, N.T., Lareau, L.F., and Weissman, J.S. (2011). Ribosome Profiling of Mouse Embryonic Stem Cells Reveals the Complexityand Dynamics of Mammalian Proteomes. Cell 147, 789–802.

Itoh, K., Chiba, T., Takahashi, S., Ishii, T., Igarashi, K., Katoh, Y., Oyake, T., Hayashi, N., Satoh, K., Hatayama, I., et al. (1997). An Nrf2/small Maf heterodimer mediates the induction of phase II detoxifying enzyme genes through antioxidant response elements. Biochem. Biophys. Res. Commun. 236, 313–322.

Itoh, N., and Okamoto, H. (1980). Translational control of proinsulin synthesis by glucose. Nature 283, 100–102.

Jackson, R.J., Hellen, C.U.T., and Pestova, T.V. (2010). The mechanism of eukaryotic translation initiation and principles of its regulation. Nature Publishing Group 11, 113–127.

Ježek, P., Dlasková, A., and Plecitá-Hlavatá, L. (2012). Redox homeostasis in pancreatic β cells. Oxidative Medicine and Cellular Longevity 2012, 932838–16.

Ji, X., Wan, J., Vishnu, M., Xing, Y., and Liebhaber, S.A. (2013). CP Poly(C) Binding Proteins Act as Global Regulators of Alternative Polyadenylation. Molecular and Cellular Biology 33, 2560– 2573.

Ji, X., Park, J.W., Bahrami-Samani, E., Lin, L., Duncan-Lewis, C., Pherribo, G., Xing, Y., and Liebhaber, S.A. (2016). αCP binding to a cytosine-rich subset of polypyrimidine tracts drives a novel pathway of cassette exon splicing in the mammalian transcriptome. Nucleic Acids Research 44, 2283–2297.

Jimenez-Moreno, C.M., Herrera-Gomez, I. de G., Lopez-Noriega, L., Lorenzo, P.I., Cobo- Vuilleumier, N., Fuente-Martin, E., Mellado-Gil, J.M., Parnaud, G., Bosco, D., Gauthier, B.R., et al. (2015). A Simple High Efficiency Intra-Islet Transduction Protocol Using Lentiviral Vectors. Curr Gene Ther 15, 436–446. 160

Jin, Y., Wang, L., Qu, S., Sheng, X., Kristian, A., Maelandsmo, G.M., Pallmann, N., Yuca, E., Tekedereli, I., Gorgulu, K., et al. (2015). STAMP2 increases oxidative stress and is critical for prostate cancer. EMBO Mol Med 7, 315–331.

Jonas, J.C., Sharma, A., Hasenkamp, W., Ilkova, H., Patanè, G., Laybutt, R., Bonner-Weir, S., and Weir, G.C. (1999). Chronic hyperglycemia triggers loss of pancreatic beta cell differentiation in an animal model of diabetes. J. Biol. Chem. 274, 14112–14121.

Jonsson, J., Carlsson, L., Edlund, T., and Edlund, H. (1994). Insulin-promoter-factor 1 is required for pancreas development in mice. Nature 371, 606–609.

Juliana, C.A., Yang, J., Rozo, A.V., Good, A., Groff, D.N., Wang, S.-Z., Green, M.R., and Stoffers, D.A. (2017). ATF5 regulates β-cell survival during stress. Proc. Natl. Acad. Sci. U.S.a. 114, 1341–1346.

Kahn, S.E., Hull, R.L., and Utzschneider, K.M. (2006). Mechanisms linking obesity to insulin resistance and type 2 diabetes. Nature 444, 840–846.

Kahvejian, A., Svitkin, Y.V., Sukarieh, R., M'Boutchou, M.-N., and Sonenberg, N. (2005). Mammalian poly(A)-binding protein is a eukaryotic translation initiation factor, which acts via multiple mechanisms. Genes & Development 19, 104–113.

Kaneto, H., Xu, G., Song, K.H., Suzuma, K., Bonner-Weir, S., Sharma, A., and Weir, G.C. (2001). Activation of the hexosamine pathway leads to deterioration of pancreatic beta-cell function through the induction of oxidative stress. J. Biol. Chem. 276, 31099–31104.

Kapeli, K., and Yeo, G.W. (2012). Genome-wide approaches to dissect the roles of RNA binding proteins in translational control: implications for neurological diseases. Front Neurosci 6, 144.

Kearse, M.G., and Wilusz, J.E. (2017). Non-AUG translation: a new start for protein synthesis in eukaryotes. Genes & Development 31, 1717–1731.

Keene, J.D. (2007). RNA regulons: coordination of post-transcriptional events. Nat Rev Genet 8, 533–543.

Kharroubi, I., Ladrière, L., Cardozo, A.K., Dogusan, Z., Cnop, M., and Eizirik, D.L. (2004). Free Fatty Acids and Cytokines Induce Pancreatic β-Cell Apoptosis by Different Mechanisms: Role of Nuclear Factor-κB and Endoplasmic Reticulum Stress. Endocrinology 145, 5087–5096.

Kiledjian, M., Wang, X., and Liebhaber, S.A. (1995). Identification of two KH domain proteins in the alpha-globin mRNP stability complex. Embo J. 14, 4357–4364.

Kim, D., Pertea, G., Trapnell, C., Pimentel, H., Kelley, R., and Salzberg, S.L. (2013). TopHat2: accurate alignment of transcriptomes in the presence of insertions, deletions and gene fusions. Genome Biol 14, R36.

Kimura, Y., Nagata, K., Suzuki, N., Yokoyama, R., Yamanaka, Y., Kitamura, H., Hirano, H., and Ohara, O. (2010). Characterization of multiple alternative forms of heterogeneous nuclear ribonucleoprotein K by phosphate-affinity electrophoresis. Proteomics 10, 3884–3895.

Kinoshita, E., Kinoshita-Kikuta, E., Matsubara, M., Yamada, S., Nakamura, H., Shiro, Y., Aoki, Y., Okita, K., and Koike, T. (2008). Separation of phosphoprotein isotypes having the same number of phosphate groups using phosphate-affinity SDS-PAGE. Proteomics 8, 2994–3003. 161

Kinoshita, E., Kinoshita-Kikuta, E., Takiyama, K., and Koike, T. (2006). Phosphate-binding tag, a new tool to visualize phosphorylated proteins. Mol. Cell Proteomics 5, 749–757.

Kitamura, T. (2013). The role of FOXO1 in β-cell failure and type 2 diabetes mellitus. Nat Rev Endocrinol 9, 615–623.

Knoch, K.-P., Bergert, H., Borgonovo, B., Saeger, H.-D., Altkrüger, A., Verkade, P., and Solimena, M. (2004). Polypyrimidine tract-binding protein promotes insulin secretory granule biogenesis. Nature Cell Biology 6, 207–214.

Knott, S.R.V., Maceli, A.R., Erard, N., Chang, K., Marran, K., Zhou, X., Gordon, A., Demerdash, El, O., Wagenblast, E., Kim, S., et al. (2014). A Computational Algorithm to Predict shRNA Potency. Molecular Cell 56, 796–807.

Komar, A.A., Mazumder, B., and Merrick, W.C. (2012). A new framework for understanding IRES- mediated translation. Gene 502, 75–86.

Konheim, Y.L., and Wolford, J.K. (2003). Association of a promoter variant in the inducible cyclooxygenase-2 gene (PTGS2) with type 2 diabetes mellitus in Pima Indians. Hum. Genet. 113, 377–381.

Kracht, M.J.L., van Lummel, M., Nikolic, T., Joosten, A.M., Laban, S., van der Slik, A.R., van Veelen, P.A., Carlotti, F., de Koning, E.J.P., Hoeben, R.C., et al. (2017). Autoimmunity against a defective ribosomal insulin gene product in type 1 diabetes. Nature Medicine 23, 501–507.

Kreuz, S., and Fischle, W. (2016). Oxidative stress signaling to chromatin in health and disease. Epigenomics 8, 843–862.

Kuersten, S., Radek, A., Vogel, C., and Penalva, L.O.F. (2013). Translation regulation gets its “omics” moment. WIREs RNA 4, 617–630.

Lai, M.-C., Chang, W.-C., Shieh, S.-Y., and Tarn, W.-Y. (2010). DDX3 regulates cell growth through translational control of cyclin E1. Molecular and Cellular Biology 30, 5444–5453.

Laurent, G., Solari, F., Mateescu, B., Karaca, M., Castel, J., Bourachot, B., Magnan, C., Billaud, M., and Mechta-Grigoriou, F. (2008). Oxidative Stress Contributes to Aging by Enhancing Pancreatic Angiogenesis and Insulin Signaling. Cell Metabolism 7, 113–124.

Laursen, L.S., Chan, C.W., and ffrench-Constant, C. (2011). Translation of myelin basic protein mRNA in oligodendrocytes is regulated by integrin activation and hnRNP-K. J. Cell Biol. 192, 797–811.

Laybutt, D.R., Preston, A.M., Åkerfeldt, M.C., Kench, J.G., Busch, A.K., Biankin, A.V., and Biden, T.J. (2007). Endoplasmic reticulum stress contributes to beta cell apoptosis in type 2 diabetes. Diabetologia 50, 752–763.

Leahy, J.L. (2005). Pathogenesis of Type 2 Diabetes Mellitus. Archives of Medical Research 36, 197–209.

Lee, A.S.Y., Kranzusch, P.J., and Cate, J.H.D. (2015). eIF3 targets cell-proliferation messenger RNAs for translational activation or repression. Nature 522, 111–114.

Lenzen, S., Drinkgern, J., and Tiedge, M. (1996). Low antioxidant enzyme gene expression in 162 pancreatic islets compared with various other mouse tissues. Free Radical Biology and Medicine 20, 463–466.

Liao, Y., Smyth, G.K., and Shi, W. (2014). featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930.

Liepelt, A., Mossanen, J.C., Denecke, B., Heymann, F., De Santis, R., Tacke, F., Marx, G., Ostareck, D.H., and Ostareck-Lederer, A. (2014). Translation control of TAK1 mRNA by hnRNP K modulates LPS-induced macrophage activation. Rna 20, 899–911.

Linder, P., and Jankowsky, E. (2011). From unwinding to clamping - the DEAD box RNA helicase family. Nature Publishing Group 12, 505–516.

Lipinski, B. (2001). Pathophysiology of oxidative stress in diabetes mellitus. J. Diabetes Complicat. 15, 203–210.

Liu, J., Krautzberger, A.M., Sui, S.H., Hofmann, O.M., Chen, Y., Baetscher, M., Grgic, I., Kumar, S., Humphreys, B.D., Humphreys, B., et al. (2014). Cell-specific translational profiling in acute kidney injury. J. Clin. Invest. 124, 1242–1254.

Lu, P.D., Jousse, C., Marciniak, S.J., Zhang, Y., Novoa, I., Scheuner, D., Kaufman, R.J., Ron, D., and Harding, H.P. (2004). Cytoprotection by pre-emptive conditional phosphorylation of translation initiation factor 2. Embo J. 23, 169–179.

Lubelsky, Y., and Ulitsky, I. (2018). Sequences enriched in Alu repeats drive nuclear localization of long RNAs in human cells. Nature 555, 107–111.

Lukong, K.E., Chang, K.-W., Khandjian, E.W., and Richard, S. (2008). RNA-binding proteins in human genetic disease. Trends in Genetics 24, 416–425.

Macejak, D.G., and Sarnow, P. (1991). Internal initiation of translation mediated by the 5' leader of a cellular mRNA. Nature 353, 90–94.

Maciag, A. (2004). Mutant K-rasV12 increases COX-2, peroxides and DNA damage in lung cells. Carcinogenesis 25, 2231–2237.

Maedler, K., Størling, J., Sturis, J., Zuellig, R.A., Spinas, G.A., Arkhammar, P.O.G., Mandrup- Poulsen, T., and Donath, M.Y. (2004). Glucose- and interleukin-1beta-induced beta-cell apoptosis requires Ca2+ influx and extracellular signal-regulated kinase (ERK) 1/2 activation and is prevented by a sulfonylurea receptor 1/inwardly rectifying K+ channel 6.2 (SUR/Kir6.2) selective potassium channel opener in human islets. Diabetes 53, 1706–1713.

Maghzal, G.J., Krause, K.-H., Stocker, R., and Jaquet, V. (2012). Detection of reactive oxygen species derived from the family of NOX NADPH oxidases. Free Radical Biology and Medicine 53, 1903–1918.

Makeyev, A.V., and Liebhaber, S.A. (2002). The poly(C)-binding proteins: a multiplicity of functions and a search for mechanisms. Rna 8, 265–278.

Mamane, Y., Petroulakis, E., LeBacquer, O., and Sonenberg, N. (2006). mTOR, translation initiation and cancer. Oncogene 25, 6416–6422.

Marintchev, A. (2013). Roles of helicases in translation initiation: a mechanistic view. Biochim. 163

Biophys. Acta 1829, 799–809.

Maris, C., Dominguez, C., and Allain, F.H.T. (2005). The RNA recognition motif, a plastic RNA- binding platform to regulate post-transcriptional gene expression. FEBS Journal 272, 2118–2131.

Martin, S.K., Carroll, R., Benig, M., and Steiner, D.F. (1994). Regulation by glucose of the biosynthesis of PC2, PC3 and proinsulin in (ob/ob) mouse islets of Langerhans. FEBS Letters 356, 279–282.

McKinney, C., Yu, D., and Mohr, I. (2013). A new role for the cellular PABP repressor Paip2 as an innate restriction factor capable of limiting productive cytomegalovirus replication. Genes & Development 27, 1809–1820.

Mechta-Grigoriou, F., Gerald, D., and Yaniv, M. (2001). The mammalian Jun proteins: redundancy and specificity. Oncogene 20, 2378–2389.

Mikula, M., Rubel, T., Karczmarski, J., Statkiewicz, M., Bomsztyk, K., and Ostrowski, J. (2015). Beads-free protein immunoprecipitation for a mass spectrometry-based interactome and posttranslational modifications analysis. Proteome Science 1–7.

Morgan, D., Oliveira-Emilio, H.R., Keane, D., Hirata, A.E., Santos da Rocha, M., Bordin, S., Curi, R., Newsholme, P., and Carpinelli, A.R. (2006). Glucose, palmitate and pro-inflammatory cytokines modulate production and activity of a phagocyte-like NADPH oxidase in rat pancreatic islets and a clonal beta cell line. Diabetologia 50, 359–369.

Morgenstern, J.P., and Land, H. (1990). Advanced mammalian gene transfer: high titre retroviral vectors with multiple drug selection markers and a complementary helper-free packaging cell line. Nucleic Acids Research 18, 3587–3596.

Mukhopadhyay, N.K., Kim, J., Cinar, B., Ramachandran, A., Hager, M.H., Di Vizio, D., Adam, R.M., Rubin, M.A., Raychaudhuri, P., De Benedetti, A., et al. (2009). Heterogeneous nuclear ribonucleoprotein K is a novel regulator of androgen receptor translation. Cancer Research 69, 2210–2218.

Naarmann, I.S., Harnisch, C., Flach, N., Kremmer, E., Kühn, H., Ostareck, D.H., and Ostareck- Lederer, A. (2008). mRNA silencing in human erythroid cell maturation: heterogeneous nuclear ribonucleoprotein K controls the expression of its regulator c-Src. J. Biol. Chem. 283, 18461– 18472.

Navid, F., and Colbert, R.A. (2017). Causes and consequences of endoplasmic reticulum stress in rheumatic disease. Nature Publishing Group 13, 25–40.

Neelankal John, A., Ram, R., and Jiang, F.-X. (2018). RNA-Seq Analysis of Islets to Characterise the Dedifferentiation in Type 2 Diabetes Model Mice db/db. Endocr. Pathol. 15, 297–15.

Nguyen, G.T., Green, E.R., and Mecsas, J. (2017). Neutrophils to the ROScue: Mechanisms of NADPH Oxidase Activation and Bacterial Resistance. Front. Cell. Infect. Microbiol. 7, 339–24.

Oh, S., Flynn, R.A., Floor, S.N., Purzner, J., Martin, L., Do, B.T., Schubert, S., Vaka, D., Morrissy, S., Li, Y., et al. (2016). Medulloblastoma-associated DDX3 variant selectively alters the translational response to stress. Oncotarget 7, 28169–28182.

Oshima, H., Taketo, M.M., and Oshima, M. (2006). Destruction of pancreatic beta-cells by 164 transgenic induction of prostaglandin E2 in the islets. J. Biol. Chem. 281, 29330–29336.

Ostareck, D.H., Ostareck-Lederer, A., Shatsky, I.N., and Hentze, M.W. (2001). Lipoxygenase mRNA silencing in erythroid differentiation: The 3'UTR regulatory complex controls 60S ribosomal subunit joining. Cell 104, 281–290.

Ostareck, D.H., Ostareck-Lederer, A., Wilm, M., Thiele, B.J., Mann, M., and Hentze, M.W. (1997). mRNA silencing in erythroid differentiation: hnRNP K and hnRNP E1 regulate 15-lipoxygenase translation from the 3' end. Cell 89, 597–606.

Ostareck-Lederer, A., and Ostareck, D.H. (2012). Control of mRNA translation and stability in haematopoietic cells: The function of hnRNPs K and E1/E2. Biology of the Cell 96, 407–411.

Ostareck-Lederer, A., Ostareck, D.H., Cans, C., Neubauer, G., Bomsztyk, K., Superti-Furga, G., and Hentze, M.W. (2002). c-Src-mediated phosphorylation of hnRNP K drives translational activation of specifically silenced mRNAs. Molecular and Cellular Biology 22, 4535–4543.

Ostrowski, J., Kawata, Y., Schullery, D.S., Denisenko, O.N., Higaki, Y., Abrass, C.K., and Bomsztyk, K. (2001). Insulin alters heterogeneous nuclear ribonucleoprotein K protein binding to DNA and RNA. Proceedings of the National Academy of Sciences 98, 9044–9049.

Pace, M.D., and Kalyanaraman, B. (1993). Spin trapping of nitrogen dioxide radical from photolytic decomposition of nitramines. Free Radical Biology and Medicine 15, 337–342.

Pacher, P., Beckman, J.S., and Liaudet, L. (2007). Nitric oxide and peroxynitrite in health and disease. Physiol. Rev. 87, 315–424.

Page, K.A., and Reisman, T. (2013). Interventions to Preserve Beta-Cell Function in the Management and Prevention of Type 2 Diabetes. Curr Diab Rep 13, 252–260.

Paneni, F., Osto, E., Costantino, S., Mateescu, B., Briand, S., Coppolino, G., Perna, E., Mocharla, P., Akhmedov, A., Kubant, R., et al. (2013). Deletion of the Activated Protein-1 Transcription Factor JunD Induces Oxidative Stress and Accelerates Age-Related Endothelial Dysfunction. Circulation 127, 1229–1240.

Parsyan, A., Svitkin, Y., Shahbazian, D., Gkogkas, C., Lasko, P., Merrick, W.C., and Sonenberg, N. (2011). mRNA helicases: the tacticians of translational control. Nature Publishing Group 12, 235–245.

Passmore, L.A., Schmeing, T.M., Maag, D., Applefield, D.J., Acker, M.G., Algire, M.A., Lorsch, J.R., and Ramakrishnan, V. (2007). The Eukaryotic Translation Initiation Factors eIF1 and eIF1A Induce an Open Conformation of the 40S Ribosome. Molecular Cell 26, 41–50.

Pause, A., Belsham, G.J., Gingras, A.C., Donzé, O., Lin, T.A., Lawrence, J.C., and Sonenberg, N. (1994). Insulin-dependent stimulation of protein synthesis by phosphorylation of a regulator of 5'-cap function. Nature 371, 762–767.

Pelechano, V., Wei, W., and Steinmetz, L.M. (2015). Widespread Co-translational RNA Decay Reveals Ribosome Dynamics. Cell 161, 1400–1412.

Persaud, S.J., Muller, D., Belin, V.D., Kitsou-Mylona, I., Asare-Anane, H., Papadimitriou, A., Burns, C.J., Huang, G.C., Amiel, S.A., and Jones, P.M. (2007). The role of arachidonic acid and its metabolites in insulin secretion from human islets of langerhans. Diabetes 56, 197–203. 165

Peyot, M.-L., Pepin, E., Lamontagne, J., Latour, M.G., Zarrouki, B., Lussier, R., Pineda, M., Jetton, T.L., Madiraju, S.R.M., Joly, E., et al. (2010). Beta-cell failure in diet-induced obese mice stratified according to body weight gain: secretory dysfunction and altered islet lipid metabolism without steatosis or reduced beta-cell mass. Diabetes 59, 2178–2187.

Pi, J., Bai, Y., Zhang, Q., Wong, V., Floering, L.M., Daniel, K., Reece, J.M., Deeney, J.T., Andersen, M.E., Corkey, B.E., et al. (2007). Reactive Oxygen Species as a Signal in Glucose- Stimulated Insulin Secretion. Diabetes 56, 1783–1791.

Piemonti, L., Leone, B.E., Nano, R., Saccani, A., Monti, P., Maffi, P., Bianchi, G., Sica, A., Peri, G., Melzi, R., et al. (2002). Human pancreatic islets produce and secrete MCP-1/CCL2: relevance in human islet transplantation. Diabetes 51, 55–65.

Piro, S., Anello, M., Di Pietro, C., Lizzio, M.N., Patan egrave, G., Rabuazzo, A.M., Vigneri, R., Purrello, M., and Purrello, F. (2002). Chronic exposure to free fatty acids or high glucose induces apoptosis in rat pancreatic islets: Possible role of oxidative stress. Metabolism 51, 1340–1347.

Poitout, V., and Robertson, R.P. (2008). Glucolipotoxicity: Fuel Excess and β-Cell Dysfunction. Endocrine Reviews 29, 351–366.

Poole, L.B. (2015). The basics of thiols and cysteines in redox biology and chemistry. Free Radical Biology and Medicine 80, 148–157.

Prentki, M., and Nolan, C.J. (2006). Islet β cell failure in type 2 diabetes. J. Clin. Invest. 116, 1802–1812.

Puri, S., Folias, A.E., and Hebrok, M. (2015). Plasticity and dedifferentiation within the pancreas: development, homeostasis, and disease. Cell Stem Cell 16, 18–31.

Quoyer, J., Longuet, C., Broca, C., Linck, N., Costes, S., Varin, E., Bockaert, J., Bertrand, G., and Dalle, S. (2010). GLP-1 mediates antiapoptotic effect by phosphorylating Bad through a beta- arrestin 1-mediated ERK1/2 activation in pancreatic beta-cells. Journal of Biological Chemistry 285, 1989–2002.

Rahier, J., Guiot, Y., Goebbels, R.M., Sempoux, C., and Henquin, J.C. (2008). Pancreatic β-cell mass in European subjects with type 2 diabetes. Diabetes, Obesity and Metabolism 10, 32–42.

Ray, D., Kazan, H., Cook, K.B., Weirauch, M.T., Najafabadi, H.S., Li, X., Gueroussov, S., Albu, M., Zheng, H., Yang, A., et al. (2014). A compendium of RNA-binding motifs for decoding gene regulation. Nature 499, 172–177.

Ray, P.D., Huang, B.-W., and Tsuji, Y. (2012). Reactive oxygen species (ROS) homeostasis and redox regulation in cellular signaling. Cellular Signalling 24, 981–990.

Retz, K.C., and Steele, W.J. (1980). Ribosome turnover in rat brain and liver. Life Sci. 27, 2601– 2604.

Richter, J.D., and Coller, J. (2015). Pausing on Polyribosomes: Make Way for Elongation in Translational Control. Cell 163, 292–300.

Robertson, R.P., Harmon, J., Tran, P.O., Tanaka, Y., and Takahashi, H. (2003). Glucose toxicity in beta-cells: type 2 diabetes, good radicals gone bad, and the glutathione connection. Diabetes 52, 581–587. 166

Robertson, R.P., Zhou, H., Zhang, T., and Harmon, J.S. (2007). Chronic oxidative stress as a mechanism for glucose toxicity of the beta cell in Type 2 diabetes. Cell Biochem Biophys 48, 139–146.

Robinson, M.D., McCarthy, D.J., and Smyth, G.K. (2010). edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 26, 139–140.

Rochette, L., Zeller, M., Cottin, Y., and Vergely, C. (2014). Diabetes, oxidative stress and therapeutic strategies. Biochim. Biophys. Acta 1840, 2709–2729.

Sachdeva, M.M., Sachdeva, M.M., Claiborn, K.C., Claiborn, K.C., Khoo, C., Khoo, C., Yang, J., Yang, J., Groff, D.N., Groff, D.N., et al. (2009). Pdx1 (MODY4) regulates pancreatic beta cell susceptibility to ER stress. Proc. Natl. Acad. Sci. U.S.a. 106, 19090–19095.

Sakai, K., Matsumoto, K., Nishikawa, T., Suefuji, M., Nakamaru, K., Hirashima, Y., Kawashima, J., Shirotani, T., Ichinose, K., Brownlee, M., et al. (2003). Mitochondrial reactive oxygen species reduce insulin secretion by pancreatic beta-cells. Biochem. Biophys. Res. Commun. 300, 216– 222.

Sanjana, N.E., Shalem, O., and Zhang, F. (2014). Improved vectors and genome-wide libraries for CRISPR screening. Nat Meth 11, 783–784.

Sanz, E., Yang, L., Su, T., Morris, D.R., McKnight, G.S., and Amieux, P.S. (2009). Cell-type- specific isolation of ribosome-associated mRNA from complex tissues. Proc. Natl. Acad. Sci. U.S.a. 106, 13939–13944.

Sastry, L., Johnson, T., Hobson, M.J., Smucker, B., and Cornetta, K. (2002). Titering lentiviral vectors: comparison of DNA, RNA and marker expression methods. Gene Ther 9, 1155–1162.

Sataranatarajan, K., Lee, M.-J., Mariappan, M.M., and Feliers, D. (2008). PKCδ regulates the stimulation of vascular endothelial factor mRNA translation by angiotensin II through hnRNP K. Cellular Signalling 20, 969–977.

Scarl, R.T., Lawrence, C.M., Gordon, H.M., and Nunemaker, C.S. (2017). STEAP4: its emerging role in metabolism and homeostasis of cellular iron and copper. J. Endocrinol. 234, R123–R134.

Scheuner, D., Mierde, D.V., Song, B., Flamez, D., Creemers, J.W.M., Tsukamoto, K., Ribick, M., Schuit, F.C., and Kaufman, R.J. (2005). Control of mRNA translation preserves endoplasmic reticulum function in beta cells and maintains glucose homeostasis. Nature Medicine 11, 757– 764.

Schieber, M., and Chandel, N.S. (2014). ROS function in redox signaling and oxidative stress. Curr. Biol. 24, R453–R462.

Schullery, D.S., Ostrowski, J., Denisenko, O.N., Stempka, L., Shnyreva, M., Suzuki, H., Gschwendt, M., and Bomsztyk, K. (1999). Regulated interaction of protein kinase Cdelta with the heterogeneous nuclear ribonucleoprotein K protein. J. Biol. Chem. 274, 15101–15109.

Sendoel, A., Dunn, J.G., Rodriguez, E.H., Naik, S., Gomez, N.C., Hurwitz, B., Levorse, J., Dill, B.D., Schramek, D., Molina, H., et al. (2017). Translation from unconventional 5' start sites drives tumour initiation. Nature 541, 494–499.

Sharma, P.R., Mackey, A.J., Dejene, E.A., Ramadan, J.W., Langefeld, C.D., Palmer, N.D., 167

Taylor, K.D., Wagenknecht, L.E., Watanabe, R.M., Rich, S.S., et al. (2015). An Islet-Targeted Genome-Wide Association Scan Identifies Novel Genes Implicated in Cytokine-Mediated Islet Stress in Type 2 Diabetes. Endocrinology 156, 3147–3156.

Shaw, J.E., Sicree, R.A., and Zimmet, P.Z. (2010). Global estimates of the prevalence of diabetes for 2010 and 2030. Diabetes Research and Clinical Practice 87, 4–14.

Shimabukuro, M., Zhou, Y.T., Levi, M., and Unger, R.H. (1998). Fatty acid-induced beta cell apoptosis: a link between obesity and diabetes. Proceedings of the National Academy of Sciences 95, 2498–2502.

Short, J.D., and Pfarr, C.M. (2002). Translational regulation of the JunD messenger RNA. J. Biol. Chem. 277, 32697–32705.

Sidrauski, C., McGeachy, A.M., Ingolia, N.T., and Walter, P. (2015). The small molecule ISRIB reverses the effects of eIF2α phosphorylation on translation and stress granule assembly. Elife.

Somers, J., Pöyry, T., and Willis, A.E. (2013). A perspective on mammalian upstream open reading frame function. Int. J. Biochem. Cell Biol. 45, 1690–1700.

Song, B., Scheuner, D., Ron, D., Pennathur, S., and Kaufman, R.J. (2008). Chop deletion reduces oxidative stress, improves beta cell function, and promotes cell survival in multiple mouse models of diabetes. J. Clin. Invest. 118, 3378–3389.

Soto-Rifo, R., Rubilar, P.S., Limousin, T., de Breyne, S., cimo, D.D.E., and Ohlmann, T.E.O. (2012). DEAD-box protein DDX3 associates with eIF4F to promote translation of selected mRNAs. Embo J. 31, 3745–3756.

Spriggs, K.A., Bushell, M., and Willis, A.E. (2010). Translational regulation of gene expression during conditions of cell stress. Molecular Cell 40, 228–237.

Spriggs, K.A., Stoneley, M., Bushell, M., and Willis, A.E. (2008). Re-programming of translation following cell stress allows IRES-mediated translation to predominate. Biology of the Cell 100, 27–38.

Stoffers, D.A., Zinkin, N.T., Stanojevic, V., Clarke, W.L., and Habener, J.F. (1997a). Pancreatic agenesis attributable to a single nucleotide deletion in the human IPF1 gene coding sequence. Nature Genetics 15, 106–110.

Stoffers, D.A., Ferrer, J., Clarke, W.L., and Habener, J.F. (1997b). Early-onset type-ll diabetes mellitus (MODY4) linked to IPF1. Nature Genetics 17, 138–139.

Sweeney, T.R., Abaeva, I.S., Pestova, T.V., and Hellen, C.U.T. (2013). The mechanism of translation initiation on Type 1 picornavirus IRESs. Embo J. 33, 76–92.

Tabatabaie, T., Vasquez-Weldon, A., Moore, D.R., and Kotake, Y. (2003). Free radicals and the pathogenesis of type 1 diabetes: beta-cell cytokine-mediated free radical generation via cyclooxygenase-2. Diabetes 52, 1994–1999.

Talchai, C., Xuan, S., Lin, H.V., Sussel, L., and Accili, D. (2012). Pancreatic β cell dedifferentiation as a mechanism of diabetic β cell failure. Cell 150, 1223–1234.

The ENCODE Project Consortium, Consortium, T.E.P., data analysis coordination, O.C., data 168 production, D.P.L., data analysis, L.A., group, W., scientific management, N.P.M., steering committee, P.I., Boise State University and University of North Carolina at Chapel Hill Proteomics groups (data production and analysis), Broad Institute Group (data production and analysis), et al. (2012). An integrated encyclopedia of DNA elements in the human genome. Nature 488, 57–74.

Thorel, F., Népote, V., Avril, I., Kohno, K., Desgraz, R., Chera, S., and Herrera, P.L. (2010). Conversion of adult pancreatic alpha-cells to beta-cells after extreme beta-cell loss. Nature 464, 1149–1154.

Tiedge, M., Lortz, S., Drinkgern, J., and Lenzen, S. (1997). Relation between antioxidant enzyme gene expression and antioxidative defense status of insulin-producing cells. Diabetes 46, 1733– 1742.

Touriol, C., Bornes, S., Bonnal, S., Audigier, S., Prats, H., Prats, A.-C., and Vagner, S. (2012). Generation of protein isoform diversity by alternative initiation of translation at non-AUG codons. Biology of the Cell 95, 169–178.

Uruno, A., and Motohashi, H. (2011). The Keap1-Nrf2 system as an in vivo sensor for electrophiles. Nitric Oxide 25, 153–160.

Uruno, A., Furusawa, Y., Yagishita, Y., Fukutomi, T., Muramatsu, H., Negishi, T., Sugawara, A., Kensler, T.W., and Yamamoto, M. (2013). The Keap1-Nrf2 system prevents onset of diabetes mellitus. Molecular and Cellular Biology 33, 2996–3010.

Van Nostrand, E.L., Pratt, G.A., Shishkin, A.A., Gelboin-Burkhart, C., Fang, M.Y., Sundararaman, B., Blue, S.M., Nguyen, T.B., Surka, C., Elkins, K., et al. (2016). Robust transcriptome-wide discovery of RNA-binding protein binding sites with enhanced CLIP (eCLIP). Nat Meth 1–9.

Ventoso, I., Kochetov, A., Montaner, D., Dopazo, J., and Santoyo, J. (2012). Extensive Translatome Remodeling during ER Stress Response in Mammalian Cells. PLoS ONE 7, e35915.

Waggoner, S.A., Johannes, G.J., and Liebhaber, S.A. (2009). Depletion of the poly(C)-binding proteins alphaCP1 and alphaCP2 from K562 cells leads to p53-independent induction of cyclin- dependent kinase inhibitor (CDKN1A) and G1 arrest. J. Biol. Chem. 284, 9039–9049.

Walsh, D., Mathews, M.B., and Mohr, I. (2013). Tinkering with translation: protein synthesis in virus-infected cells. Cold Spring Harb Perspect Biol 5, a012351–a012351.

Wang, H., Vardy, L.A., Tan, C.P., Loo, J.M., Guo, K., Li, J., Lim, S.G., Zhou, J., Chng, W.J., Ng, S.B., et al. (2010). PCBP1 Suppresses the Translationof Metastasis-Associated PRL-3 Phosphatase. Cancer Cell 18, 52–62.

Weaver, J.R., Grzesik, W., and Taylor-Fishwick, D.A. (2014). Inhibition of NADPH oxidase-1 preserves beta cell function. Diabetologia 58, 113–121.

Wellen, K.E., Fucho, R., Gregor, M.F., Furuhashi, M., Morgan, C., Lindstad, T., Vaillancourt, E., Gorgun, C.Z., Saatcioglu, F., and Hotamisligil, G.S. (2007). Coordinated Regulation of Nutrient and Inflammatory Responses by STAMP2 Is Essential for Metabolic Homeostasis. Cell 129, 537– 548.

Xia, Y., and Zweier, J.L. (1997). Superoxide and peroxynitrite generation from inducible nitric oxide synthase in macrophages. Proceedings of the National Academy of Sciences 94, 6954– 169

6958.

Yagishita, Y., Fukutomi, T., Sugawara, A., Kawamura, H., Takahashi, T., Pi, J., Uruno, A., and Yamamoto, M. (2014). Nrf2 protects pancreatic β-cells from oxidative and nitrosative stress in diabetic model mice. Diabetes 63, 605–618.

Yamaguchi, S., Ishihara, H., Yamada, T., Tamura, A., Usui, M., Tominaga, R., Munakata, Y., Satake, C., Katagiri, H., Tashiro, F., et al. (2008). ATF4-Mediated Induction of 4E-BP1 Contributes to Pancreatic β Cell Survival under Endoplasmic Reticulum Stress. Cell Metabolism 7, 269–276.

Zhao, H., Kalivendi, S., Zhang, H., Joseph, J., Nithipatikom, K., Vásquez-Vivar, J., and Kalyanaraman, B. (2003). Superoxide reacts with hydroethidine but forms a fluorescent product that is distinctly different from ethidium: potential implications in intracellular fluorescence detection of superoxide. Free Radical Biology and Medicine 34, 1359–1368.

Zhou, J., Ye, S., Fujiwara, T., Manolagas, S.C., and Zhao, H. (2013a). Steap4 Plays a Critical Role in Osteoclastogenesis in Vitro by Regulating Cellular Iron/Reactive Oxygen Species (ROS) Levels and cAMP Response Element-binding Protein (CREB) Activation. Journal of Biological Chemistry 288, 30064–30074.

Zhou, J., Wan, J., Gao, X., Zhang, X., Jaffrey, S.R., and Qian, S.-B. (2015). Dynamic m6A mRNA methylation directs translational control of heat shock response. Nature 526, 591–594.

Zhou, P., Zhang, Y., Ma, Q., Gu, F., Day, D.S., He, A., Zhou, B., Li, J., Stevens, S.M., Romo, D., et al. (2013b). Interrogating translational efficiency and lineage-specific transcriptomes using ribosome affinity purification. Proceedings of the National Academy of Sciences 110, 15395– 15400.

Zumsteg, U., Frigerio, S., and Holländer, G.A. (2000). Nitric oxide production and Fas surface expression mediate two independent pathways of cytokine-induced murine beta-cell damage. Diabetes 49, 39–47.

170