Integrated and Functional Genomic Approaches to Elucidate Differential Genetic Dependencies in

The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters

Citation Wong, Terence. 2018. Integrated and Functional Genomic Approaches to Elucidate Differential Genetic Dependencies in Melanoma. Doctoral dissertation, Harvard University, Graduate School of Arts & Sciences.

Citable link http://nrs.harvard.edu/urn-3:HUL.InstRepos:42014990

Terms of Use This article was downloaded from Harvard University’s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at http:// nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of- use#LAA Integrated and Functional Genomic Approaches to Elucidate Differential Genetic Dependencies in Melanoma

A dissertation presented

by

Terence Cheng Wong

to

The Division of Medical Sciences

in partial fulfillment of the requirements

for the degree of

Doctor of Philosophy

in the subject of

Biological and Biomedical Sciences

Harvard University

Cambridge, Massachusetts

November 2017

© 2017 Terence Cheng Wong

All rights reserved.

Dissertation Advisor: Levi Garraway Terence Cheng Wong

Integrated and Functional Genomic Approaches

to Elucidate Differential Genetic Dependencies in Melanoma

ABSTRACT

Genomic characterization of human cancers over the past decade has generated comprehensive catalogues of genetic alterations in cancer genomes. Many of these genetic events result in molecular or cellular changes that drive cancer cell . In melanoma, a majority of tumors harbor mutations in the BRAF , leading to activation of the MAPK pathway and tumor initiation. The development and use of drugs that target the mutant BRAF and the MAPK pathway have produced significant clinical benefit in melanoma patients.

In addition, recent advances in cancer immunotherapy have led to dramatic and durable responses in tumor types with high mutation rates, including melanoma. However, innate and acquired resistance to targeted therapy and immunotherapy necessitate the discovery, investigation, and pursuit of novel and orthogonal tumor dependencies for the treatment of cancer patients.

In this work, we applied the analysis of functional genomic screening data and cistromic and transcriptomic experimental approaches to characterize SOX10 as a differential genetic dependency in melanoma. To identify novel and potentially actionable lineage-specific differential genetic dependencies across cancer, we performed class-based computational

iii analyses of Project Achilles dependency data. In melanoma, known differential dependencies included BRAF and MAPK1, while SOX10 was the highest ranked novel genetic dependency.

Integrative analysis with additional genomic data uncovered a -dependency relationship for SOX10 in melanoma, which we confirmed, showing that only cell lines that express SOX10 are dependent on SOX10 for cell proliferation. To further investigate SOX10 dependency in melanoma, we characterized the cistrome, the complete set of binding sites in the genome, and transcriptome of SOX10 in melanoma cell lines. The integration of these datasets enabled the determination of SOX10 target and downstream pathways in melanoma, including shared target genes with the well-known oncoprotein . Additional analysis of

SOX10 target genes by MITF expression revealed MITF class-specific SOX10 transcriptional programs in melanoma, characterized by MITF in MITF-high melanoma and FOS-JUN in

MITF-low melanoma.

Taken together, our studies provide new insights into the role and importance of SOX10 in melanoma biology and pave the way for the integration of large-scale datasets to identify and characterize genetic dependencies in cancer.

iv

TABLE OF CONTENTS

PREFACE

Abstract ...... iii

Table of Contents ...... v

Acknowledgements ...... viii

INTRODUCTION: Melanoma, Development, and Genomic Approaches

Summary ...... 2

Melanoma Epidemiology, Genomics, and Treatment ...... 4 Melanoma Epidemiology Genomic Landscape of Melanoma Recent Advances in Melanoma Treatment

SOX10 and Neural Crest Development ...... 17 Neural Crest Development SOX Family of Transcription Factors Role of SOX10 in Neural Crest Development and Melanoma

Functional Genomic Approaches to Study Cancer ...... 32 Genetic Perturbation Reagents Functional Genomic Screens

Genomic Approaches to Study Transcription Factors ...... 38 Characterization of Localization by ChIP-Seq Characterization of Gene Expression by RNA-Seq

Context and Rationale for the Current Work ...... 42

CHAPTER 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Summary ...... 44

Introduction ...... 46

Results ...... 49 Identification of Differential Genetic Dependencies in Tumor Lineages Identification of Differential Genetic Dependencies in Melanoma

v

Model of SOX10 Expression-Based Dependency in Cancer Validation of SOX10 Dependency in Melanoma

Discussion ...... 69

Methods...... 73

Acknowledgements ...... 76

CHAPTER 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Summary ...... 78

Introduction ...... 80

Results ...... 82 Characterization of the SOX10 Cistrome in Melanoma Characterization of the SOX10 Transcriptome in Melanoma Elucidation of SOX10 Target Genes and Dependencies Analysis of SOX10 Target Genes in Melanoma by MITF status

Discussion ...... 123

Methods...... 130

Acknowledgements ...... 137

CHAPTER 3: Investigation of HDAC Inhibitors in Melanoma

Summary ...... 139

Introduction ...... 141

Results ...... 144 Evaluation of HDAC Inhibitors on SOX10 Expression and Activity Combination of HDAC Inhibitors and MAPK Pathway Inhibitors

Discussion ...... 162

Methods...... 165

Acknowledgements ...... 169

vi

CONCLUSION: Integration, Characterization, and Innovation

Summary ...... 171

Power of Functional Genomic Characterization of Cancer ...... 172

Further Characterization of SOX10 in Melanoma ...... 173

Therapeutic Strategies for Targeting SOX10 in Cancer ...... 175

Future Considerations ...... 177

APPENDICES

Integrated Chemical Screen for Modulators of SOX10 Transcriptional Activity ...... 178

Assay Development of an Arrayed ORF Screen for Resistance to Androgen Deprivation in Prostate Cancer ...... 184

REFERENCES ...... 194

vii

ACKNOWLEDGEMENTS

First, I would like to thank Levi Garraway for his mentorship, support, and guidance throughout my graduate career. His welcoming and sincere advising style has been instrumental to my growth as a scientist. Most importantly, I thank him for giving me the freedom to pursue my own interests in and outside of the laboratory.

Next, I would like to thank Cory Johannessen and Jesse Boehm, who have served as additional mentors during my graduate and pre-doctoral training in the Broad Institute Cancer

Program. It has been a pleasure and a privilege to learn from them over the past 8 years.

I would also like to thank current and former members of the Garraway Laboratory, the

Broad Institute Cancer Program, the Broad Institute Office of Academic Affairs, and the Broad

Institute NextGen Association for Postdocs and Graduate Students for being an immense source of knowledge, support, and fun during my time at Broad. I would especially like to thank Luc de

Waal, Quinn Sievers, Tikvah Hayes, Tanaz Sharifnia, Zuzana Tothova, Colles Price, Chloe

Villani, Danielle Kerins, and Angela Florentino. Their friendship has made my time in graduate school so much more rewarding and enjoyable.

I would like to thank the members of my Dissertation Advisory Committee: Joan Brugge,

Myles Brown, David Fisher, and Karen Cichowski. They never failed to offer thoughtful, valuable, and actionable advice and guidance throughout my graduate training.

I would like to thank my friends from Harvard University, Harvey Mudd College, and beyond, including those I have met during my time in Boston. Their boundless support has always kept me moving forward. I would especially like to thank Sapana Thomas, Donavion

Huskey, and Kerri Thomas for always being there and pushing me to be the best person I can be.

viii

I would very much like to thank my partner, Tim Whipple, for his enduring love, care, and support for me over the past 8 years. He has served as a constant source of strength and joy, especially when they are needed most. He has made my time during graduate school more pleasant and fulfilling.

Finally, I would like to thank my family, Mom, Dad, Joni, and Daisy, for their endless encouragement and ceaseless support for everything that I do. Their constant confidence in me has propelled me to achieve things that I had once thought were unattainable. Their strength and determination in life are daily reminders for me to pursue my greatest dreams with the utmost passion and dedication.

ix

DEDICATION

I would like to dedicate this body of work to my parents, Young Wong and Lily Wong, for their tireless and unwavering support for my continued education and life of learning. Their sacrifices during my childhood gave me the opportunity to study and excel at the highest levels.

Without their love and support, I would not be who I am today.

x

Glossary of Terms

Functional genomics: the use of genetic perturbation reagents (e.g., shRNA, ORF, CRISPR) to interrogate the function of genes at a genome-wide level Chromatin immunoprecipitation (ChIP): the enrichment and isolation of genomic DNA fragments physically associated with a protein or histone modification of interest Gene set: a set of genes, typically associated with a specific genomic location, transcription factor binding motif or cis-regulatory element, molecular function, cellular component, biological process, or disease pathway MAPK pathway: a major signaling pathway in melanoma and other cancers that transduces signals from the extracellular environment into the cellular nucleus to drive transcriptional programs relevant for cell proliferation and growth Neural crest: a progenitor cell population transiently present in vertebrate from which many differentiated cell types arise, including chondrocytes, neurons, glial cells, and

List of Abbreviations

ChIP chromatin immunoprecipitation ChIP-seq chromatin immunoprecipitation followed by massively parallel sequencing CCLE Cancer Cell Line Encyclopedia CRISPR clustered regularly interspaced short palindromic repeats GSEA Gene Set Enrichment Analysis MAPK Mitogen-Activated Protein Kinase MSigDB Molecular Signatures Database ORF open reading frame shRNA short-hairpin RNA TCGA The Cancer Genome Atlas

List of Important Genes

ACTB encodes Beta-Actin, a protein involved in cell motility, structure, and motility; major constituent of the cytoskeleton; housekeeping gene

BRAF encodes V-Raf Murine Sarcoma Viral Oncogene Homolog B, a serine/threonine protein kinase involved in regulating the MAPK pathway; downstream of NRAS and upstream of MEK1/2

CDK2 encodes Cyclin-Dependent Kinase 2, a serine/threonine protein kinase involved in cell cycle regulation; catalytic subunit of the cyclin-dependent protein kinase complex, which regulates progression through the cell cycle particularly the G1 to S phase transition; target gene of MITF

xi

DCT encodes , an enzyme involved in melanin production in melanocytes; target gene of MITF

GAPDH encodes Glyceraldehyde-3-Phosphatase Dehydrogenase, a protein involved in carbohydrate metabolism; housekeeping gene

MAP2K1 encodes Mitogen-Activated Protein Kinase Kinase 1 or MAPK/ERK Kinase 1 (MEK1), a protein kinase with a critical role in signal transduction through the MAPK pathway; downstream of BRAF and upstream of ERK1/2

MAP2K2 encodes Mitogen-Activated Protein Kinase Kinase 2 or MAPK/ERK Kinase 2 (MEK2), a protein kinase with a critical role in signal transduction through the MAPK pathway; downstream of BRAF and upstream of ERK1/2

MAPK1 encodes Mitogen-Activated Protein Kinase 1 or Extracellular Signal-Regulated Kinase 2 (ERK2), a protein kinase acting as the integration point for multiple biochemical signals in the MAPK pathway; downstream of MEK1/2

MAPK3 encodes Mitogen-Activated Protein Kinase 3 or Extracellular Signal-Regulated Kinase 1 (ERK1), a protein kinase acting as the integration point for multiple biochemical signals in the MAPK pathway; downstream of MEK1/2

MITF encodes Microphthalmia-Associated Transcription Factor, a transcription factor involved in the regulation of the differentiation and development of melanocytes and the pigment cell-specific transcription of genes; target gene of SOX10

NRAS encodes Neuroblastoma RAS Viral Oncogene Homolog, a protein involved in regulating the MAPK pathway; upstream of BRAF

PAX3 encodes Paired Box 3, a transcription factor involved in development; activates expression of MITF

PMEL also known as SILV, encodes Premelanosome Protein, a transmembrane glycoprotein enriched in melanosomes and involved in the structural organization of premelanosomes; target gene of SOX10

SOX10 encodes SRY (Sex Determining Region Y)-Box 10, a transcription factor involved in the regulation of embryonic development and in the determination of cell fate of neural crest-derived cells; activates expression of MITF

TYR encodes Tyrosinase, an enzyme involved in melanin production in melanocytes; target gene of MITF

xii

INTRODUCTION

Melanoma, Neural Crest Development, and Genomic Approaches

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

SUMMARY

Advanced melanoma accounts for about 1% of all new skin cancer cases, but is associated with the vast majority of skin cancer deaths. If diagnosed early, melanoma can be treated by surgical excision or radiation therapy with a 5-year relative survival rate of 98% for localized disease. However, advanced melanoma with deep invasion or that have spread to lymph nodes has a 5-year relative survival rate of 18% and must be treated with a more comprehensive strategy, including surgery, radiation therapy, chemotherapy, targeted therapy, and immunotherapy. Recent advances in the genomics and tumor microenvironment of melanoma have led to the development of promising therapeutic agents. More than half of metastatic harbor an activating mutation in BRAF, a signaling protein in the MAPK pathway. Small molecule inhibitors targeting BRAF and MEK, including dabrafenib and trametinib, have shown dramatic response rates in tumors with BRAF V600E/K mutation. Due to their high mutation rate and neoantigen load, melanomas are highly immunogenic and are frequently permeated with tumor infiltrating lymphocytes and cytotoxic T cells. Immune checkpoint blockade targeting CTLA-4 and PD-L1, such as ipilimumab, nivolumab, and pembrolizumab, have demonstrated historic rates of response duration with about one third of treated patients living beyond three years. However, acquired and innate resistance to these therapies leaves the vast majority of melanoma patients without a meaningful and durable outcome.

Comprehensive molecular studies of tumors and cancer cell lines have provided many insights into the underlying mechanisms of cancer initiation and progression. Large-scale projects, such as The Cancer Genome Atlas and the Cancer Cell Line Encyclopedia, have generated extensive catalogues of significantly mutated genes and regions with copy number

2

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches alterations. While these genetic markers have proven useful in identifying likely drivers of oncogenesis, it is still unclear which of these alterations and their corresponding genes are essential for cancer cell proliferation and survival and, thus, the best targets for therapeutic development. High-throughput functional genomic techniques have enabled the systematic investigation of all human genes in experimental systems. As one example, Project Achilles has utilized pooled RNA interference screens to evaluate the essentiality of genes in the proliferation, growth, and viability of cancer cells. Subsequent downstream analyses of these large functional genomic datasets have allowed for the elucidation of genetic dependencies by tumor characteristics and genetic features.

Melanocytes, the cell of origin for melanoma, are derived from the neural crest, a transient group of cells in the developing embryo of vertebrates that also gives rise to glial cells and sensory neurons. During embryonic development, SOX10 and other transcription factors play important roles in the formation of neural crest stem cells, maintaining multipotency of neural crest cells, and cell fate specification and differentiation of neural crest-derived cells. In the developmental path toward melanocytes, SOX10 is most well-known and studied as a major transcriptional of the melanocyte lineage master transcription factor, MITF. However,

SOX10 has a role in maintaining multipotency of neural crest stem cells, likely by maintenance of ERBB3 expression and neuregulin 1 signaling. The continued expression of SOX10 in adult human melanocytes and melanoma cells suggest a potential role for SOX10 in conferring multipotent properties to melanocytes that have undergone oncogenic transformation.

3

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

MELANOMA EPIDEMIOLOGY, GENOMICS, AND TREATMENT

Melanoma Epidemiology

Skin cancer is the most commonly diagnosed cancer in the United States. However, skin cancer can be classified into multiple types with varying incidence and mortality rates, including basal cell carcinoma, squamous cell carcinoma, and melanoma. Other rare types of skin cancer include Merkel cell carcinoma, skin lymphoma, and Kaposi sarcoma. The more frequent non- melanoma skin cancers, basal cell carcinoma and squamous cell carcinoma, are very common and are most often found in areas exposed to the sun, including the head, neck, and arms. Most non-melanoma skin cancers are diagnosed early and can be treated and cured with fairly minor surgery or other types of local treatments. Melanoma is less common than other types of skin cancer, but it is more likely to grow and spread to other sites in the body. Invasive melanoma accounts for about 1% of all skin cancer cases, but the vast majority of skin cancer deaths. An estimated 87,110 new cases of melanoma will be diagnosed in the United States in 2017, making melanoma the fifth and sixth leading sites of new cancer cases in men and women, respectively

(Figure I-1). The incidence rate for melanoma has continued to rise over the past four decades, but it appears to be slowing in recent years (Figure I-2). Melanoma is most commonly diagnosed in non-Hispanic whites, with an annual incidence rate of 26 (per 100,000 people), compared to 5 in Hispanics, and 1 in blacks. Incidence rates are higher in women than in men before the age of 50, but this trend is reversed in men and women after the age of 65. These melanoma incidence patterns likely reflect age and sex differences in occupational and recreational exposure to ultraviolet radiation, including exposure to the sun and the use of indoor tanning, as well as differences in human biology and genetics.

4

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-1. Leading sites of new cancer cases in the United States, 2017 estimates. Adapted from American Cancer Society, Cancer Facts & Figures 2017.

Figure I-2. Trends in incidence rates for selected cancers by sex in the United States, 1975-

2013. The incidence rate trend for melanoma (depicted in purple) has risen continuously for men and women over the past 40 years. Adapted from Siegel et al., 2017.

5

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Risk of skin cancer, including melanoma, can be reduced by minimizing skin exposure to intense ultraviolet (UV) radiation by seeking shade and limiting time in the sun, wearing clothing that covers the arms and legs and hats and sunglasses that protect the face, applying broad- spectrum sunscreen that provides protection against UVA and UVB rays (i.e., sun protection factor (SPF) of 30 or higher), and avoiding indoor tanning. Exposure to UV radiation from the sun or indoor tanning devices can cause genetic mutations and poses serious threats for the development of skin cancer. In addition to high exposure to UV radiation, other risk factors for melanoma include a personal or family history of melanoma and the presence of atypical, large, or numerous moles. Skin cancer risk is also increased for individuals with sun sensitivity (e.g., sunburning easily, difficulty tanning, or natural blond or red hair color), a history of excessive sun exposure, diseases or treatments that suppress the immune system, and a past history of skin cancer.

Skin cancer survival rates vary largely by skin cancer type and stage at diagnosis. Almost all cases of basal cell carcinoma and squamous cell carcinoma can be cured by surgery, especially if the tumor is detected and treated early. Melanoma is also highly curable when detected at an early and localized stage, but it is more likely to spread to other parts of the body and result in lower survival in patients with advanced melanoma: the 5-year relative survival rate is 98% for localized disease, 62% for regional disease, and 18% for distant disease (Figure I-3).

Thus, the early detection and diagnosis of melanoma could lead to substantial increases in treatment success, overall survival, and quality of life.

6

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-3. Stage distribution by race (A) and five-year relative survival rates by stage at diagnosis and race (B) for melanoma in the United States, 2006-2012. Fewer cases of melanoma are diagnosed at distant stages, but later stage diagnosis is associated with lower survival. Adapted from Siegel et al., 2017.

Treatment of melanoma is largely determined by stage of disease. For early stage or localized melanoma, the primary tumor and surrounding normal tissue are removed and a sentinel lymph node may be biopsied for further diagnosis. If the sentinel lymph nodes contain cancer, more extensive lymph node surgery may be needed. Advanced melanoma, or melanoma with deep invasion or that have spread to other sites of the body, may be treated with radiation therapy and systemic treatments, including chemotherapy, targeted therapy, and immunotherapy.

The treatment of advanced melanoma has changed immensely in recent years with the Food and

Drug Administration (FDA) approval of several new targeted and immunotherapy drugs, which have demonstrated dramatic and durable shrinkage of tumors in melanoma patients. As such, chemotherapy, which is usually much less effective than newer treatments, has been less frequently used.

7

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Genomic Landscape of Melanoma

The development of melanoma, like all cancers, is influenced by both genetic and environmental factors. As previously mentioned, many environmental risk factors, including UV radiation, can increase an individual’s risk of melanoma. Likewise, genetic risk factors, such as inherited gene mutations, can predispose an individual to familial melanoma or other cancer types. In addition to familial cancer susceptibility genes, many genes have been identified through high-throughput genome sequencing methods to initiate sporadic melanoma.

Although exposure to UV radiation plays a significant role in melanoma development, twin studies estimate 55% of the variation in melanoma liability is due to genetic effects (Shekar et al., 2009). Several genes and hereditary syndromes are associated with familial or inherited melanoma. Over two decades ago, CDKN2A was identified through a combination of linkage studies and a positional cloning approach as the first familial melanoma susceptibility gene

(Hussussian et al., 1994; Kamb et al., 1994). CDKN2A is a major tumor suppressor gene that encodes two , p16 and ARF, from two different reading frames that are involved in cell cycle regulation. Following the identification of CDKN2A, a candidate screening approach of p16 interacting partners identified mutations in CDK4, another gene involved in the cell cycle, as responsible for melanoma development in a small number of families (Puntervoll et al., 2013;

Soufir et al.,1998; Zuo et al., 1996). All identified pathogenic mutations occurred in codon 24, indicating its importance in the function of CDK4 and its role in melanoma development.

The advent of genomic sequencing technologies has enabled the discovery of genes associated with a wide range of tumor types, including melanoma. One of the most significantly mutated genes in melanoma is BRAF, whose encoded protein plays a central role in the MAPK signaling pathway (Davies et al., 2002). Many targeted, exome, and genome sequencing studies

8

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches have confirmed recurrent BRAF missense mutations in over 60% of melanomas that result in a mutated protein, BRAF V600E, with elevated kinase activity and have characterized the landscape of genetic alterations in melanoma (Hodis et al., 2012; Cancer Genome Atlas

Network, 2015) (Figure I-4). Another 25% of melanomas harbor recurrent missense mutations in NRAS, leading to activating mutations in codon 12, 13, and 61, that lock the mutated protein in an active state and drive signaling through the MAPK pathway (van’t Veer et al., 1989; Hodis et al., 2012; Cancer Genome Atlas Network, 2015). The high frequency of BRAF and NRAS mutations in melanoma highlights the importance of this pathway in melanoma development.

9

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-4. Landscape of driver mutations in melanoma. Recurrent and significantly mutated genes in melanoma include BRAF, NRAS, TP53, PTEN, and CDKN2A (p16INK4a). Recurrent and significantly amplified genes in melanoma include CCND1, KIT, CDK4, TERT, and MITF.

Adapted from Hodis et al., 2012.

Other high-throughput genome-scale technologies and analytical approaches have been applied to identify significantly altered genes in melanoma. The integration of copy number and gene expression data identified MITF as a recurrently amplified gene in melanoma tumors and cell lines (Garraway et al., 2005). MITF encodes a transcription factor required for melanocyte development, including the differentiation and survival of melanocytes (Cheli et al., 2010).

Combined ectopic expression of MITF and BRAF V600E transformed primary human

10

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches melanocytes, supporting its function as a melanoma oncogene. The analysis of whole genome sequencing data for non-coding mutations in melanoma identified recurrent mutations in the

TERT promoter in over 70% of melanoma tumors and cell lines (Huang et al., 2013). TERT encodes a reverse transcriptase that has well-established roles in tumorigenesis. These promoter mutations generated de novo transcription factor binding motifs that increased TERT expression.

Together, these and other studies highlight the utility of genome interrogation techniques for the elucidation of melanoma genomes.

Recent Advances in Melanoma Treatment

For most of the past several decades, treatment for cancer patients, including those with melanoma, has consisted of three major treatment modalities: surgery, radiation therapy, and chemotherapy. Due to their restricted fields of effect, surgery and radiation therapy are typically used to treat localized disease. When performed on early stage tumors, surgery and radiation therapy can be highly effective and lead to cures. On the other hand, chemotherapy is administered and functions as a systemic therapy, reaching cells throughout the body and usually causing significant adverse effects in cancer patients. Due to dosage limitations, chemotherapy is rarely curative for solid tumors despite their frequent use in patients with advanced disease.

Fortunately, recent genomic and tumor microenvironment studies of melanoma have led to the discovery of targetable driver genes and immune pathways in melanoma. These molecular and cellular insights have generated new promise and strategies for treating melanoma patients.

The identification of recurrent activating BRAF mutations in melanoma, namely BRAF

V600E, unlocked a new opportunity to target an important biological pathway in melanoma

11

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

(Figure I-5). The missense mutation in BRAF resulted in a constitutively active mutant BRAF protein that could propagate signaling through the MAPK pathway in the absence of upstream growth factor inputs. The location of the mutant residue in the kinase active site and decades of kinase drug development enabled the development of small molecule compounds that could target BRAF. Treatment with these compounds, such as vemurafenib (Zelboraf,

Plexxicon/Genentech) and dabrafenib (Tafinlar, GlaxoSmithKline), blocked MAPK pathway signaling in BRAF mutant melanoma cells and caused regression of BRAF mutant xenografts

(Bollag et al., 2010; GlaxoSmithKline, 2012). Furthermore, treatment of patients with BRAF

V600E melanoma led to dramatic response rates and an improved progression-free survival rate compared to dacarbazine chemotherapy (Chapman et al., 2011; Hauschild et al., 2012).

However, resistance to these targeted drugs inevitably arose within 6 to 18 months and a small percentage of BRAF mutant melanoma patients never benefited from BRAF inhibitors. Genomic profiling of BRAF inhibitor resistant tumors revealed mechanisms of resistance that act primarily to maintain signaling through the MAPK pathway via mutations in NRAS, MAP2K1 (MEK1), and MAP2K2 (MEK2) and amplification of BRAF (Wagle et al., 2011; Van Allen et al., 2014).

In an effort to further reduce signaling through the MAPK pathway, small molecule inhibitors were developed against MEK1 and MEK2 (encoded by MAP2K1 and MAP2K2), which are kinases directly downstream of BRAF in the MAPK pathway. The combination of BRAF inhibitors with MEK inhibitors, such as trametinib (Mekinist, GlaxoSmithKline) and cobimetinib

(Cotellic, Exelixis/Genentech), achieved greater response rates and improved progression-free survival compared to BRAF inhibitor alone (Flaherty et al., 2012; Robert et al., 2015). However, resistance to the combination of BRAF and MEK inhibition still developed in patients, with many tumors exhibiting genetic alterations that serve to reactive the MAPK pathway via

12

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches mutation in MAP2K2 (MEK), amplification of BRAF or splice variants of BRAF (Wagle et al.,

2014). While these studies demonstrated that combination therapy is more effective than single agent therapy, more research must be done to identify other promising melanoma drug targets and to investigate the most safe and effective dosing and scheduling drug combinations.

Figure I-5. Critical proteins in the MAPK pathway in melanoma. The MAPK pathway consists of a hierarchy of proteins involved in signal transduction from extracellular signals to downstream transcriptional outputs: RAS (NRAS), RAF (BRAF and CRAF), MEK1/2, and

ERK1/2. FDA approved drugs against BRAF (e.g., vemurafenib and dabrafenib) and MEK (e.g., trametinib and cobimetinib) bind to and inhibit their target proteins and reduce signaling through the MAPK pathway. Mechanisms of resistance to these targeted agents maintain activated signaling through the pathway despite inhibitor treatment. Adapted from Wood & Luke, 2017.

13

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Basic research into the tumor microenvironment of melanoma have elucidated the role of various immune cell populations and signaling pathways in the immune response to melanoma.

Early efforts for using immunotherapy in melanoma focused on the use of interferon and interleukin cytokines to treat advanced melanoma. However, these treatments exhibited substantial toxicity and modest clinical benefit. The elucidation of immune-regulatory molecules, including cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4), programmed cell-death protein 1 (PD-1), and its ligands, PD-L1 and PD-L2, and the development of monoclonal antibodies against these targets have revolutionized the treatment of advanced melanoma

(Figure I-6). Treatment with anti-CTLA-4 antibodies (ipilimumab) and anti-PD-1 antibodies

(nivolumab and pembrolizumab), commonly referred to as immune-checkpoint blockade, has demonstrated substantial clinical benefits, including considerable response rates (10-15% for anti-CTLA-4 and 30-40% for anti-PD-1 compared to 10% for chemotherapy) and much improved overall survival at two years (20-40% for anti-CTLA-4 and greater than 50% for anti-

PD-1 compared to 20% for chemotherapy) (Hodi et al., 2010; Robert et al., 2011; Robert et al.,

2015; Larkin et al., 2015). Combination immunotherapy (i.e., anti-CTLA-4 and anti-PD-1 antibodies) has shown substantially improved response rates (50-60% for combination) and survival (60% overall survival at 3 years for combination), but with a concomitant increase in adverse effects (Wolchok et al., 2013; Postow et al., 2015; Larkin et al., 2015; Wolchok et al.,

2017). Of note, a majority of melanoma patients have not responded to immune-checkpoint blockade, pointing to innate resistance mechanisms in the cancer genome or the tumor immune microenvironment. Investigations of immunotherapy resistance have revealed genetic and tumor microenvironmental factors that correlate with response, including mutational load, neoantigen load, and expression of cytolytic markers (Snyder et al., 2014; Van Allen et al., 2015). Current

14

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches and future studies of immunotherapies are necessary for the continued evaluation of current and novel immunotherapies, their use as single agents and in combination, as well as their integration with other treatment modalities and the development of personalized treatment regimens to produce the greatest clinical response.

15

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-6. CTLA-4 and PD-1 pathway and checkpoint blockade in immunotherapy.

CTLA-4 is expressed on T cells after activation in the lymph nodes. FDA approved drugs against CTLA-4 (e.g., ipilimumab) suppress negative signals delivered by CTLA-4, which permits sustained T cell activation and proliferation. PD-1 is expressed on T cells in peripheral tissues, including tumors. PD-1 ligands, such as PD-L1 and PD-L2, are expressed on tumor cells in response to inflammatory signals and down-regulate T cell activity through binding to PD-1.

FDA approved drugs against PD-1 (e.g., nivolumab and pembrolizumab) and PD-L1 (e.g., atezolizumab, durvalumab, and avelumab) prevent negative regulation of PD-1 by PD-L1 and may delay T cell exhaustion. Adapted from June et al., 2017.

16

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

SOX10 AND NEURAL CREST DEVELOPMENT

The neural crest is a transient progenitor cell population present during the embryonic development of vertebrates. It contributes to a wide variety of differentiated cell types, including sensory and autonomic ganglia of the nervous system, cartilage and bone of the face, and pigment cells of the skin. Unique to vertebrate embryos, the neural crest represents a key innovation underpinning vertebrate evolution. Due to its multipotency, motility, and ability to give rise to a broad array of cell types, the neural crest serves as an excellent model system of cell behavior and identity.

Neural crest development is believed to be controlled by a set of transcription factors, including SOX10 (Simoes-Costa & Bronner, 2015). SOX10 has been implicated in a number of roles involved in neural crest development, including the formation of neural crest cells, maintenance of neural crest cell multipotency, specification of neural crest cell fates, and differentiation of specific cell types (Kelsh, 2006). A greater understanding of the role of SOX10 in the development of the neural crest and malignant melanoma is of particular significance since

SOX10 mutations underlie several neurocristopathies and SOX10 is required for melanoma cell growth and proliferation.

Neural Crest Development

The neural crest is a group of progenitor cells that arise during embryonic development and contributes to a wide variety of cell types (Figure I-7). These cells are known for their ability to migrate long distances during development and to differentiate into numerous cell types. Present only in vertebrates, the neural crest is thought to be a crucial evolutionary

17

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches innovation in the origin and diversification of vertebrates. Due to their far-reaching migratory ability and developmental plasticity, neural crest cells have been of great interest to developmental biologists for decades and represent an excellent model system of cell identity.

Generally, the neural crest undergoes a standard developmental process that appears to hold true across vertebrate species (Figure I-7). During gastrulation, nascent neural crest cells are induced in the ectodermal germ layer and initially reside in the neural plate border territory between the neural plate and the non-neural ectoderm (Figure I-7B). Signaling pathways, such as FGF, WNT, and BMP, are thought to drive the expression of neural plate border specifier genes, including MSX1, PAX3, TFAP2A, and ZIC1. Activating and inhibitor signals generate a gradient of WNT and BMP activity across the neural plate, resulting in induction of the neural plate border territory (reviewed in Groves & LaBonne, 2014). The gene products of neural plate border specifier genes engage in a series of mutual cross-regulatory interactions that lead to the stabilization of this regulatory state and ensure their continued expression during this stage as well as through later stages of development (Meulemans & Bronner-Fraser, 2004; Khudyakov &

Bronner-Fraser, 2009; Monsoro-Burq et al., 2005; Sato et al., 2005; Nikitina et al., 2008; Bhat et al., 2013).

During neurulation, the neural plate border territory becomes elevated in the neural folds as the neural plate closes to form the neural tube (Figure I-7C). At this stage, the neural plate border specifiers, including PAX3 and MSX1, drive the expression of neural crest specifier genes, including FOXD3, ETS1, SNAI1, SNAI2, and SOX10, which initiates the specification of these neural plate border cells to neural crest cells (Khudyakov & Bronner-Fraser, 2009; Simoes-

Costa et al., 2012; Barembaum & Bronner, 2013). The neural crest specifier genes positively regulate each other, including regulatory interactions between FOXD3 and ETS1, FOXD3 and

18

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

SNAI1/SNAI2, FOXD3 and SOX10, and SOX10 and SNAI2 (Simoes-Costa et al., 2012; Aybar et al., 2003; Dottori et al., 2001; Honore et al., 2003). The process of neural crest specification results in a cell population that exhibits a regulatory state that is distinct from that of cells in the neighboring neural plate.

After closure of the neural tube, neural crest cells undergo epithelial-to-mesenchymal transition (EMT) and delaminate, or separate, from the neural tube (Figure I-7D). Interestingly, the transcription factors that establish neural crest specification (FOXD3, SNAI1, SNAI2, and

SOX10) also drive EMT, indicating that neural crest identity and migration are interconnected.

EMT is a complex process, including the precise regulation of hundreds of genes and the structural remodeling of cellular architecture. The process of EMT involves changes at the cell surface that result in dissolution of adherens junctions, allowing the separation of the neural crest into individual cells. A number of neural crest specifiers, including SOX10, FOXD3, Snail

(SNAI1), and Slug (SNAI2), regulate EMT by repressing expression of cadherins that mediate strong cell-cell interactions, such as E-cadherin (CHD1) and N-cadherin (CHD2) (Dottori et al.,

2001; Cheung et al., 2005; Ferronha et al., 2013). The dissociation of neural crest cells is also regulated by TWIST and ZEB2, which repress E-cadherin (Barriga et al., 2013; Rogers et al.,

2013).

After delamination, neural crest cells undergo migrations along stereotypic pathways to give rise to differentiated cell types (Figure I-7E). Migratory neural crest cells are exposed to different environmental signals as they travel to their destinations and differentiate into their terminal cell types. Several neural crest specifier genes are expressed during neural crest migration, including FOXD3, ETS1, and SOX10. Cis-regulatory analysis has revealed that differential expression of these genes is largely mediated by the use of different enhancers. For

19

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches example, SOX10 is directly regulated at multiple enhancers by PAX3, TFAP2A, FOXD3, and itself (Werner et al., 2007; Dutton et al., 2008; Wahlbuhl et al., 2012). Positive self-regulation allows for continuous expression of genes in the migratory neural crest.

Following their migration to different parts of the embryo, neural crest cells undergo the reverse of EMT, called mesenchymal-to-epithelial transition (MET), incorporate into resident tissues, and differentiate into diverse cell types. The process of neural crest cell diversification centers on neural crest transcription factors, their downstream lineage-specific transcriptional programs, and environmental cues. Neural crest differentiation into melanocytes relies on the transcriptional activity of SOX10 and PAX3, lineage-specification governed by MITF, and signals by WNT and KIT. SOX10 and PAX3 drive melanoma specification by directly activing expression of MITF (Lee et al., 2000; Potterf et al., 2000; Verastegui et al., 2000). In turn, MITF drives the expression of terminal differentiation factors in the melanocyte lineage, including enzymes responsible for melanin synthesis, such as those encoded by DCT, TYR, and PMEL

(Murisier et al., 2007). SOX10 and MITF operate in a feed-forward circuit and cooperate to drive expression of these differentiation factors (Ludwig et al., 2004; Watanabe et al., 1998).

20

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-7. Stages in neural crest development. (A) Schematic dorsal view of a ten-somite stage chicken embryo, showing the neural crest (green) in the vicinity of the midline. (B)

Development of the neural crest begins at the gastrula stage, with the specification of the neural plate border at the edges of the neural plate. (C) As the neural plate closes to form the neural tube, the neural crest progenitors are specified in the dorsal part of the neural folds. (D) After specification, the neural crest cells undergo epithelial-to-mesenchymal transition (EMT) and delaminate from the neural tube. (E) Migratory neural crest cells follow stereotypical pathways to diverse destinations, where they will give rise to distinct derivatives. Adapted from Simoes-

Costa & Bronner, 2015.

21

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

SOX Family of Transcription Factors

SRY-related HMG-box (SOX) genes encode a family of transcription factors involved in the regulation of embryonic development and in the determination of cell fate. SOX family proteins are an evolutionarily conserved group of transcriptional regulators defined by the presence of a highly conserved high mobility group (HMG) domain that mediates DNA binding.

This domain was originally identified in SRY (Sex Determining Region Y), an important factor involved in mammalian male sex determination (Gubbay et al., 1990; Sinclair et al., 1990).

Vertebrate genomes contain approximately 20 SOX family members with highly divergent functions in development.

SOX proteins can bind to a consensus motif (ATTGTT) through their HMG domain, which consists of three alpha helices (Badis et al., 2009; Kondoh & Kamachi, 2010). This binding is mediated by the interaction of the HMG domain with the minor groove of DNA

(Remenyi et al., 2003). Binding of SOX proteins to the minor grove causes DNA to bend towards the major groove, which may contribute to their regulatory functions. SOX proteins are classified into groups based on their amino acid sequence similarity in the HMG domain (Figure

I-8). Homology among SOX proteins outside of the HMG domain is only found within each group (Bowles et al., 2000; Schepers et al., 2002).

22

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-8. Grouping and structures of SOX proteins. (A) An unrooted phylogenetic neighbor-joining tree of the high mobility group (HMG) domains in comparison with outgroup

HMG domains of LEF1/TCF7 and HMGB1. (B) Alignment and classification of the SOX HMG domains. The amino acid sequence conservation levels and sequence logos are shown below. (C)

The domain structures of SOX proteins, showing representatives of the groups. The basic structures are highly conserved among members of each group. Adapted from Kamachi &

Kondoh, 2013.

23

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-8 (Continued).

24

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

SOX protein expression and activity is regulated at multiple levels. The expression of

SOX genes is frequently regulated by other SOX proteins. SOX protein function is dose- dependent, indicating that the modulation of SOX protein levels is an important mode of SOX protein regulation. Additionally, SOX protein activity is modulated by post-translational modifications or by interactions with other proteins. Various post-translational modifications, including phosphorylation, acetylation, and sumoylation, have been reported to modulate the activity, stability, and intracellular localization of SOX proteins. These modifications have functional relevance in a variety of molecular functions and biological processes, including embryonic development.

SOX10 proteins generally exhibit their gene regulatory functions only when in complexes with partner transcription factors (Kamachi et al., 2000; Kondoh & Kamachi, 2010).

A functional SOX binding site in the genome is accompanied by a binding site for a partner protein, which is required for SOX-dependent transcriptional regulation. Additionally, binding of a single SOX protein alone to DNA does not lead to transcriptional activation or repression

(Kamachi et al., 2001; Yuan et al., 1995). Different SOX family groups interact with their partner factors in different ways. Some SOX proteins have heterologous partners, while others require homologous dimerization.

The existence of multiple SOX partner factors allows for the step-wise progression of developmental processes. The replacement of a SOX partner factor can result in large-scale changes in target genes, contributing to developmental progression. For example, during melanocyte development from the neural crest, SOX10 pairs with PAX3 to activate expression of MITF, whose gene product then acts as a partner for SOX10 to activate expression of genes

25

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches involved in melanocyte differentiation, such as DCT and TYR (Bondurand et al., 2000; Ludwig et al., 2004; Murisier et al., 2007) (Figure I-9).

Figure I-9. Step-wise progression of developmental regulation by the replacement of SOX partner factors: SOX10, PAX3, and MITF in melanocyte development. First, SOX10 and

PAX3 interact to activate MITF expression. Then, SOX10 and MITF partner together to drive expression of DCT and other genes involved in melanocyte development. Adapted from

Kamachi & Kondoh, 2013.

SOX proteins regulate various developmental processes and strong associations exist between individual SOX groups and specific cell lineages. Members of the same SOX group share similar functions and are often expressed in the same developing tissues (e.g., SOXB1 proteins in the central nervous system, SOXE proteins in the neural crest, SOXF proteins in the vascular system). This overlap creates functional redundancy among group members that co- regulate the same targets and protects developmental processes against deleterious genetic

26

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches variation. Nevertheless, the relative contribution of SOX group members to a developmental process differs with regards to expression timing, expression levels, and protein activity.

As an example of the association between SOX groups and distinct cell lineages, SOXE proteins are expressed and active during neural crest development. SOX9 is initially expressed in neural crest cells in the dorsal neural tube (Cheung & Briscoe, 2003). As these cells migrate away from the neural tube, SOX10 expression is activated in a SOX9-dependent manner with

ETS1 and MYB (Betancur et al., 2010). At first, SOX10 interacts with MEF2C in the cranial neural crest, PAX3 in melanocytes, and POU3F1/2 in Schwann cells (Agarwal et al., 2011;

Bondurand et al., 2000; Kuhlbrodt et al., 1998). These SOX10-partner complexes activate the expression of a second set of SOX10 partners, such as MITF in melanocytes, leading to the terminal differentiation of neural crest cells (Bondurand et al., 2000).

Role of SOX10 in Neural Crest Development and Melanoma

The SOXE group of SOX family transcription factors includes SOX8, SOX9, and

SOX10, and is characterized in part by a C-terminal transcriptional activation domain. SOX10 has been intensely studied with respect to its involvement in Waardenburg-Shah syndrome or

Waardenburg syndrome Type 4 (WS4), a disorder that combines the features of (cochlear deafness and pigmentary defects) and Hirschsprung’s disease (enteric aganglionosis) (Read, 2000; Shah et al., 1981; Badner & Chakravarti, 1990; Parisi & Kapur,

2000). WS4 is genetically heterogeneous: WS4A (OMIM #277580) is caused by mutation in

EDNRB, WS4B (OMIM #613265) is caused by mutation in EDN3, and WS4C (OMIM #613266) is caused by mutation in SOX10. Heterozygous SOX10 mutations identified in patients with

27

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Waardenburg-Shah syndrome have been predicted to result in loss of function, suggesting that the pathologic mechanism in WS4C is haploinsufficiency (Pingault et al., 1998). SOX10 mutations are also associated with a more severe dysmyelination syndrome combined with

Waardenburg-Shah syndrome, called Peripheral Demyelinating Neuropathy, Central

Dysmyelination, Waardenburg Syndrome, and Hirschsprung Disease (PCWH) (OMIM

#609136), that result in truncated SOX10 mutant proteins with dominant-negative activity (Inoue et al., 1999; Pingault et al., 2000; Inoue et al., 2002; Inoue et al., 2004). It has been proposed that all nonsense and frameshift mutations that cause premature termination of translation of SOX10 mRNA transcripts generate truncated SOX10 proteins with potent dominant-negative activity, while the more severe disease is realized only when the mutant mRNAs escape the nonsense-mediated decay (NMD) pathway (Inoue et al., 2004). Homozygous mutant phenotypes in mouse and zebrafish models have shown a severe embryonic lethal defect comprising failure of oligodendrocyte differentiation and reduction or absence of many neural crest-derived cell types, including melanocytes (Southard-Smith et al., 1998; Kapur, 1999). These findings in animal models likely explain the lack of reported homozygous mutant phenotypes in humans.

The expression of SOX10 is important throughout neural crest development (Figure I-

10). Studies in animal models have implicated SOX10 in the formation of the neural crest, showing that knockdown of SOX10 eliminates early neural crest markers, while overexpression of SOX10 results in expansion of these markers and delamination of neural crest cells from the neural tube (Aoki et al., 2003; Honore et al., 2003; McKeown et al., 2005; Cheung & Briscoe,

2003). SOX10 also has a role in maintaining the multipotency of neural crest cells and in inhibiting the differentiation of neural cell fates (Kim et al., 2003). Additionally, SOX10 acts as a master transcriptional regulator of fate specification in melanocytes by directly activating

28

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches expression of MITF, a well-characterized transcription factor in melanocyte development

(Steingrimsson et al., 2004; Vance & Goding, 2004). Interestingly, SOX10 has been shown to regulate DCT, a target gene of MITF involved in melanocyte differentiation, in mouse cells, but not in zebrafish models, providing uncertainty into the role of SOX10 in the differentiation step of neural crest cells into melanocytes (Jiao et al. 2004; Ludwig et al., 2004; Dutton et al., 2001).

29

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-10. SOX10 and neural crest development. Neural crest development occurs in a series of sequential steps, some of which involve SOX10 function. Early neural crest cells, which are transiently multipotent, are induced from ectodermal precursors. They generate a series of partially restricted neural crest cells by a progressive fate restriction mechanism. These neural crest cells express receptors, including ERBB3. Selection of individual fates from these partially restricted neural crest cells requires extracellular ligands. The resultant extracellular signaling acts in combination with neural crest cell transcription factors, including SOX10, to activate transcription of fate-specific transcription factors (e.g., MITF). The fate-specific transcription factors regulate target genes to generate the differentiated phenotype. Adapted from Kelsh, 2006.

30

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Recent studies in cell culture and animal models have investigated SOX10 function in melanoma. Using a mouse model for giant congenital nevi, Shakhova et al. show that Sox10 haploinsufficiency counteracts Nras(Q61K)-driven congenital and melanoma formation.

Moreover, they also show that SOX10 is crucial for the maintenance of melanoma cells and that knockdown of SOX10 leads to reduced cell proliferation and survival (Shakhova et al., 2012). In a separate study, Cronin et al., confirm that SOX10 is required for the proliferation of melanoma cells and that SOX10 haploinsufficiency reduces melanoma initiation in the Grm1(Tg) transgenic mouse model (Cronin et al., 2013). They go on to demonstrate that knockdown of

SOX10 results in cell cycle arrest, altered cellular morphology, and induced senescence.

Additionally, Shakhova et al. also show that a related factor, SOX9, is up-regulated in melanoma upon loss of SOX10. Furthermore, they demonstrate that SOX9 binds to the SOX10 promoter and induces down-regulation of SOX10 expression, revealing a functionally antagonistic feedback loop between SOX9 and SOX10 (Shakhova et al., 2015). Together, these studies provide evidence for the necessary role of SOX10 in the initiation and maintenance of melanoma.

31

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

FUNCTIONAL GENOMIC APPROACHES TO STUDY CANCER

The investigation of the and gene function has been enabled over the past decade by advances in large-scale genomic technologies. Massively parallel sequencing has allowed for the rapid and high resolution characterization of the human genome and exome at the single nucleotide level, providing scientists and physicians with information regarding single point mutations, small insertions and deletions, and large chromosomal rearrangements. SNP arrays have enabled the study of copy number alterations, including duplications and deletions, and RNA-sequencing has facilitated precise measurements of gene expression at the transcript level. At the epigenetic level, bisulfite sequencing has permitted the characterization of DNA methylation sites in the genome and ChIP-sequencing has allowed for the genomic localization of histone modifications and chromatin associated proteins. In addition to these techniques to describe the structural genome at a genome-wide level, functional genetic and genomic reagents and assays have been developed in the past decade to enable systematic characterization of gene function across the genome and in various model systems. Among these, shRNA, ORF, and

CRISPR-Cas9 reagents and their genome-wide applications have enjoyed wide-spread use and have become part of the molecular biologist’s toolkit.

Genetic Perturbation Reagents

Short-hairpin RNAs (shRNAs) are synthetic RNA constructs produced via stable transduction of a cell with lentivirus carrying an shRNA-encoding plasmid. Once the RNA is expressed and takes the shape of a hairpin, shRNAs resemble endogenous microRNAs

(miRNAs), which are further processed by large complexes in the nucleus and the cytoplasm,

32

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Drosha and Dicer, respectively. These processed shRNAs can be loaded into the RNA-induced silencing complex (RISC), recognize and hybridize to mRNA transcripts with perfect or near perfect sequence identity, and lead to mRNA cleavage and degradation or translational repression. This process is commonly referred to as RNA interference (RNAi). shRNA reagents have been used to selectively inhibit the expression of specific genes at the transcript level. Due to their similar mechanism of action as miRNAs, shRNAs often have “off-target” effects on the expression of non-target genes. Regardless, robust and consistent knockdown of gene expression can be achieved with excellent shRNA reagents.

Open reading frames (ORFs) are synthetic expression constructs for the expression of wild-type or mutant genes. As with shRNAs, reliable expression of ORFs can be attained by stable transduction of cells with lentivirus carrying an ORF-containing plasmid. Once the ORF is integrated into the genome, ORF expression is driven by engineered promoters and may incorporate a protein tag (e.g., GFP). ORF expression can be experimentally regulated with the use of inducible systems (e.g., doxycycline/tetracycline, IPTG, etc.) or tissue-specific promoters, allowing for precise ORF expression. In cancer research applications, ORFs can resemble wild- type or mutant gene sequences, allowing for the evaluation of novel or aberrant genetic alterations observed in human tumors.

Clustered regularly interspaced short palindromic repeats (CRISPR) serve as part of the bacterial immune system, consisting of stored genetic information from previously encountered bacteriophages. Cas9 is an enzyme produced by bacteria that uses transcribed CRISPR RNA to recognize and cut DNA with sequence similarity. To target and cut specific sites in the genome,

Cas9 and CRISPR guide RNA (gRNA) constructs can be stably expressed in cells via lentiviral delivery. While initially developed to cut DNA and ultimately lead to deleterious interruption of

33

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches a gene’s open reading frame, the CRISPR-Cas9 system has been modified to perform other functions that take advantage of the site-specific targeting activity of CRISPR, including the use of a catalytically-dead Cas9 fused to either an activating or repressive domain to activate

(CRISPRa) or inhibit (CRISPRi) gene expression. Additional CRISPR techniques include the use of other Cas proteins that take advantage of their smaller size or their ability to cut RNA.

Furthermore, the popularity of CRISPR-Cas9 is likely a result of its highly modular experimental system, where the CRISPR gRNA for nearly any genomic can be easily designed and introduced into cells. In contrast, previous gene editing techniques, such as nucleases

(ZFNs) and transcription activator-like effector nucleases (TALENs) required laborious protein design and delivery. Overall, CRISPR-Cas9 and related methods have revolutionized gene editing and gene targeting in basic and translational research.

Functional Genomic Screens

Genetic perturbation reagents, such as shRNAs, ORFs, and CRISPR-Cas9, have enabled the investigation of gene function in experimental systems via genetic knockdown, overexpression, or other functional changes. These tools are useful when studying the function of genes that harbor mutations, copy number changes, or other genetic alterations in cancer. For example, one could determine the essentiality of a gene in a particular phenotype or pathway via shRNA-mediated knockdown or CRISPR-Cas9-mediated knockout of the gene of interest.

Conversely, the ability of a gene or genetic variant to rescue a particular phenotype or drug treatment could be tested via overexpression of a wild-type or mutant ORF. The versatility of these genetic perturbation reagents allows for the functional evaluation of genes in a myriad of ways.

34

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Many institutions, including Broad Institute of MIT and Harvard, have combined the utility of genetic perturbation reagents and the power of high-throughput assays to perform systematic, large-scale functional genomic screens. These genome-wide screens enable the functional interrogation of all genes in the human genome in a multitude of different systems.

Screens can be performed in arrayed format, where reagents are introduced in separate wells in a high-throughput manner, or in pooled format, where reagents are labeled (e.g., by DNA barcodes), assays are performed in bulk, and screening data are deconvoluted by computational algorithms (Figure I-11). Arrayed screens require precise assay optimization and assay execution in small wells, which typically restrict the duration of arrayed screens due to limitations in cell culture growth area. However, the assay readout process for arrayed screens is relatively simple and straightforward. Conversely, the readout of pooled screens involves high- throughput characterization (e.g., sequencing, flow cytometry, imaging, etc.) of cells that have survived a selection process (e.g., growth competition, drug treatment, cell state transition, etc.) and subsequent deconvolution by advanced computational methods. Due to their assay in bulk, pooled screens can scale up easily and can accommodate larger screening libraries.

35

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-11. General workflow for functional genomics screening using CRISPR-Cas9 in pooled vs. arrayed approaches. In pooled screening, perturbation libraries are delivered to a single batch of cells before selection for specific phenotypes are performed. The output of pooled screens is derived from high-throughput characterization (such as sequencing) of the selected vs. control cell populations. In arrayed screening, perturbation reagents are delivered to discrete populations of cells. Phenotypes are identified based on well location rather than through selection. The output of arrayed screens is a ranked phenotypic measure for each perturbation reagent evaluated in the screen. Adapted from Agrotis & Ketteler, 2015.

36

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

The success of functional genomic screens, like all other high-throughput screens, relies on consistent reagents, a robust model system and assay design (including positive controls), and a reliable assay readout (i.e., high dynamic range and large signal-to-noise ratio). Reagents must exhibit stable expression and activity in experiments and should be comparable among perturbations within the same experiment. The cell culture model system should be highly relevant to the biological question being asked (e.g., mutation background, response to drug, in vitro vs. in vivo, etc.) and the assay must feature reagents or conditions that will serve as robust positive controls. Finally, the assay readout must be sensitive enough to detect significant differences or changes in the experimental system (i.e., high dynamic range) and to differentiate between a strong, positive hit and a background or negative control. When designed well and according to these parameters, functional genomic screens have the potential to identify

37

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

GENOMIC APPROACHES TO STUDY TRANSCRIPTION FACTORS

Transcription factors are proteins associated with DNA with roles in regulating gene expression. They typically contain DNA-binding domains that allow for direct contact with

DNA, as well as other domains that may facilitate interactions with other transcription factors or chromatin-associated proteins to activate or repress transcription. Their roles in normal and disease biology have been studied extensively in cell culture and animal models. These studies have revealed tissue-specific transcription factors, such as the estrogen and the in breast and prostate tissues, respectively, as well as transcription factors with widespread effects on gene expression and cellular processes, such as TP53 and MYC.

Characterization of Transcription Factor Localization by ChIP-Seq

Developments in biochemical and sequencing technologies have enabled the comprehensive characterization of transcription factor localization sites and patterns in cells.

Chromatin immunoprecipitation (ChIP) involves the “crosslinking” of cells (i.e., the formation of covalent bonds between macromolecules in physical contact, such as transcription factors and

DNA), the fragmentation of nuclear DNA by sonication, the “pulldown” or extraction of a protein of interest with bound DNA fragments using protein-specific antibodies and magnetic purification (i.e., immunoprecipitation), and the “decrosslinking” and enrichment of DNA fragments that were initially bound by the protein of interest (Figure I-12). Important considerations for ChIP experiments include the expression level and form of the protein.

Proteins can be profiled using endogenous protein (i.e., normal protein with natural expression in cells) or ectopically expressed proteins with or without a peptide tag for isolation. A key control

38

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches for ChIP experiments is the input control, which consists of DNA fragments that have not been enriched by immunoprecipitation and contain just the “input” DNA (i.e., global population of

DNA fragments across the genome). The enriched DNA fragments may then be further characterized by a variety of analytical methods. For the quantification of a limited number of specific and known genomic loci, ChIP-enriched DNA fragments can be analyzed by quantitative polymerase chain reaction (qPCR), which will amplify and quantify specific genomic sequences using given oligonucleotide primers. While ChIP-qPCR is highly sensitive and can detect specific enrichment of genomic DNA regions, this approach requires prior knowledge of the genomic loci under investigation and the procurement of specific primers. To assay the full set of genomic localization sites for a protein of interest, ChIP-enriched DNA fragments must be analyzed by more high-throughput technologies, such as microarray hybridization (ChIP-chip) and massively parallel sequencing (ChIP-seq). For ChIP-chip, DNA fragments are labeled with a fluorophore and hybridized to a microarray along with a control set of DNA fragments (i.e., input control) labeled with a different fluorophore. Fluorescent signals are then measured and converted to gene expression levels based on the position of probes on the microarray. As sequencing costs have fallen over the years, massively parallel sequencing has been increasingly used to systematically characterize ChIP-enriched DNA in an unbiased fashion. For ChIP-seq, sequencing libraries are prepared from ChIP-enriched DNA fragments and input control samples, normalized and pooled to enable multiplexed assessment, and sequenced. Sequencing reads are deconvoluted by computational algorithms, assigned to original

ChIP samples, and quantified to determine enrichment at genomic loci. Thus, ChIP-seq is a powerful tool to investigate the complete set of binding sites of a specific transcription factor across the genome in an unbiased and systematic manner.

39

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Figure I-12. Overview of chromatin immunoprecipitation (ChIP) and downstream analysis.

First, DNA and chromatin-associated proteins are crosslinked by formaldehyde or ultraviolet radiation. Next, chromatin is sheared into 100-500 bp chromatin fragments using sonication or enzymatic digestion. Then, chromatin fragments associated with the protein of interest are selectively enriched using appropriate antibodies coupled to magnetic or agarose beads. After washing and reversal of crosslinks, DNA fragments are purified. The resulting ChIP-enriched

DNA fragments can then be quantified for specific genomic loci by quantitative PCR (ChIP- qPCR) or be systematically evaluated at a genome-wide level by microarray (ChIP-chip) or massively parallel sequencing (ChIP-seq). Adapted from Song et al., 2015.

40

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

Characterization of Gene Expression by RNA-Seq

Developments in sequencing technology have also enabled the rapid and comprehensive characterization of gene expression throughout the genome. RNA can be isolated from biological systems, converted to complementary DNA (cDNA) by reverse transcription (RT), and assessed by low- or high-throughput approaches to quantify gene expression. For a small number of genes, RT-qPCR can be used to evaluate gene expression using specific primers and fluorescent reagents. For transcriptional profiling at a genome-wide level, systematic measurements of gene expression can be performed using microarrays or massively parallel sequencing (RNA-seq) similar to the techniques used in ChIP-chip or ChIP-seq described above. Thus, microarray and

RNA-seq techniques provide rapid, precise, and quantitative assessments of gene expression on a global level. When combined with genetic perturbation reagents (e.g., shRNAs or CRISPR-

Cas9) targeting specific transcription factors, transcriptional profiling can also yield comprehensive characterization of genes under regulatory control of transcription factors.

The integration of genomic localization and expression data from ChIP-seq and RNA-seq experiments, respectively, allow for the powerful characterization of transcription factor target genes and downstream pathways. Overlap of gene lists bound by and regulated by transcription factors can produce a set of high confidence target genes. Furthermore, additional analyses of molecular and cellular pathways can lead to a more complete understanding of transcription factor function.

41

Introduction: Melanoma, Neural Crest Development, and Genomic Approaches

CONTEXT AND RATIONALE FOR THE CURRENT WORK

Recent advances in the genomic and tumor microenvironmental characterization of human melanoma have provided a greater understanding of the molecular and cellular drivers of this disease. The discovery of activating mutations in BRAF and the immunosuppressive environments in melanomas have led to the development of targeted therapy and immune checkpoint blockade for the treatment of advanced melanoma. These therapeutic strategies have produced immense clinical benefit in melanoma patients, including dramatic reductions in tumor burden and extended durations of response, respectively. However, the inevitable acquisition of resistance and the lack of wide-spread response to these therapies necessitate the discovery of novel tumor dependencies and therapeutic approaches for the treatment of melanoma.

In the following work, we describe our integrative approach to identify, characterize, and target novel differential genetic dependencies in melanoma. First, we leverage existing genomic datasets to discover differential genetic dependencies across tumor types, including melanoma.

We identify SOX10 as a lineage-specific genetic dependency in melanoma and validate a

SOX10 gene expression-dependency model in cancer cells. Then, we characterize the complete set of SOX10 binding sites and regulated genes in the melanoma genome. We determine the set of core SOX10 target genes in melanoma, identify shared target gene sets between SOX10 and

MYC, and describe differences in SOX10 localization in melanoma by MITF status. Finally, we evaluate the use of HDAC inhibitors as a therapeutic strategy in melanoma, including their effects on SOX10 expression and combination with MAPK pathway inhibitors. Taken together, these studies aim to highlight the generation and analysis of original and existing large-scale genomic datasets to broaden our biological understanding of melanoma and to discover new therapeutic approaches to improve the treatment of melanoma patients.

42

CHAPTER 1

Identification of Differential Genetic Dependencies in Tumor Lineages

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

SUMMARY

Over the past decade, major advances in the understanding of cancer etiology has centered on the sequencing and analysis of cancer genomes from human patient samples or from human cancer cell lines. These comprehensive studies have expanded our catalog of cancer genes, including oncogenes and tumor suppressor genes, and have led to the development of targeted therapeutic agents. However, while these projects have increased our knowledge base of the genetic alterations in cancer, it remains to be determined whether these genes and the mutations that they harbor are required for the initiation and maintenance of human tumors.

Functional genomic approaches, such as shRNA and CRISPR-Cas9 loss-of-function screens, have enabled the comprehensive characterization of essential genes in cancer cell lines. The analysis of these functional genomic datasets by tumor lineages or by mutation patterns can provide insight into molecular or cellular pathways that are required for cancer cell growth and proliferation in specific tumor contexts. In particular, the discovery of tumor lineage-specific genetic dependencies can identify previously unknown cancer dependencies that may be targeted by therapeutic agents, which can be combined with existing treatment regimens for long-lasting responses in cancer patients.

In this study, we analyzed the Project Achilles dataset of shRNA loss-of-function genetic screens in over 500 human cancer cell lines for differential genetic dependencies by tumor lineage. DEMETER dependency scores representing the on-target viability effects of shRNAs targeting nearly all 20,000 genes in the human genome were analyzed iteratively for each tumor lineage against all other tumor lineages. This two-class comparison analysis identified differential dependencies in multiple tumor types, including melanoma, prostate cancer, breast cancer, colorectal cancer, gastric cancer, pancreatic cancer, liver cancer, head and neck cancer,

44

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages kidney cancer, medulloblastoma, acute myeloid leukemia (AML), B-cell acute lymphocytic leukemia (B-ALL), multiple myeloma, and others. Many of the top differential dependencies have been previously identified by Project Achilles or in prior studies, including CTNNB1 in colorectal cancer, KRAS in pancreatic cancer, and PAX8 in ovarian cancer (Cheung et al., 2011).

However, potentially novel genetic dependencies include FOXA1 and SUZ12 in prostate and breast cancers, BCL2L1, CSNK1A1, and PLK1 in colorectal and gastric cancers, and BCL2L1 and HNF1B in liver and kidney cancers. The top ranked differential genetic dependencies in the melanoma lineage were BRAF, SOX10, and MAPK1. A majority of melanoma cell lines, including those profiled in Project Achilles, harbor an activating BRAF mutation (BRAF

V600E) which sensitizes cells to MAPK pathway inhibitors (e.g., dabrafenib, trametinib, etc.) and imparts dependency on BRAF and MAPK1. SOX10 is a transcription factor with important roles in the development of melanocytes, the cell of origin of melanoma, from the neural crest.

Integrative analysis of dependency and gene expression data revealed a model of gene expression-dependency for SOX10 in cancer. This genetic dependency model was confirmed in cell population doubling assays, in which genetic knockdown of SOX10 led to decreased cell proliferation in cell lines expressing SOX10. Additionally, SOX10 knockdown was evaluated in combination with MAPK pathway inhibitor treatment in melanoma cell lines. Surprisingly, combined inhibition of these orthogonal dependencies did not result in synergistic killing of melanoma cells. Instead, knockdown of SOX10 may lead to a senescent cell state that protects melanoma cells from the negative effects of MAPK pathway inhibitors. Taken together, these studies highlight the utility of integrative and functional genomic approaches to identify differential genetic dependencies in cancer, including SOX10 as a unique vulnerability in melanoma.

45

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

INTRODUCTION

Over the past few decades, traditional genetic techniques and large-scale sequencing- based technologies have enabled the characterization of cancer genomes at an impactful and unprecedented pace. Early studies of oncogenes and tumor suppressor genes identified major driver genes associated with cancer, including TP53, KRAS, MYC, and others. These genes were altered in the genomes of cancer cells at high frequency, in various ways, and in multiple tumor types. For example, TP53 has been shown to harbor point mutations and deletions, with mutation in nearly all high-grade serous ovarian carcinoma, in 30% of melanoma cases, and no observed genetic alterations in rhabdomyosarcoma. Conversely, the BCR-ABL1 fusion gene resulting from the Philadelphia (the reciprocal translocation between chromosome 9 and ) is observed in nearly all cases of chronic myeloid leukemia (CML) and was initially discovered by observing abnormalities in the karyotypes of CML patients. The advent of massively parallel sequencing technology, such as the sequencing-by-synthesis method pioneered and promoted by Illumina, allowed for the rapid and efficient characterization of cancer genomes with resolution at the single nucleotide level. These nucleotide sequencing technologies and complementary approaches for characterizing the transcriptome (microarrays and RNA-seq), proteome (mass spectrometry), DNA methylome (bisulfite sequencing), metabolome (mass spectrometry), and other large-scale cellular features have enabled comprehensive characterization of the structural genomics and related aspects of cancer. Many institutional and government groups, including individual labs and The Cancer Genome Atlas

(TCGA), have published collaborative studies on the genomics of glioblastoma, ovarian cancer, colorectal cancer, and many others (Cancer Genome Atlas Research Network, 2008; Cancer

Genome Atlas Research Network, 2011; Cancer Genome Atlas Network, 2012).

46

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

While these comprehensive projects have catalogued the vast majority of point mutations, copy number alterations, and chromosomal rearrangements in these tumor types, it is generally unclear which of these genetic alterations are required for tumor initiation and maintenance.

Computational approaches have been developed to assess whether genetic alterations are driver events (i.e., critical for the formation and/or propagation of cancer phenotypes) or passenger events (i.e., altered as a consequence of genomic instability or other mutagenic processes but has no role in tumor development) in cancer genomes. Many of these techniques employ sophisticated algorithms that compare the frequency and significance of the observed genetic alterations to that of a background model of changes in the genome. These computational pipelines, such as MutSigCV for significantly mutated genes and GISTIC for significantly amplified or deleted regions, have proven useful in nominating likely driver genes from large catalogues of cancer genes (Lawrence et al., 2013; Beroukhim et al., 2007). However, genes put forward by these computational studies must be evaluated in experimental systems, including cell culture and animal models of cancer, to confirm their functional roles in tumor development.

Examples of functional experiments include the mutation, knockdown, or overexpression of cancer genes in cell culture, in xenograft models, or in genetically engineered animal models.

These assays are critical for the characterization of genetic alterations in cancer and for the identification of potential therapeutic targets in tumors.

The advent of genetic perturbation technologies, including shRNAs, ORFs, and CRISPR-

Cas9, has revolutionized the functional study of cancer genes. The stable expression of short- hairpin RNAs (shRNAs) or open reading frames (ORFs) in cancer cell can lead to the knockdown or overexpression of specific genes, respectively. Meanwhile, the introduction of clustered regularly interspaced short palindromic repeats (CRISPR) guide RNAs with Cas9 into

47

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages cancer cells can result in DNA cutting and subsequent genetic knockout of targeted genes. These powerful genetic tools enable the versatile and controlled experimentation of specific genes in various model systems. When coupled with advanced characterization techniques and scaled to the level of the genome, these genetic perturbation reagents allow for the systematic, large-scale interrogation of gene function (Luo et al., 2008; Cheung et al., 2011; Cowley et al., 2014;

Tsherniak et al., 2017). Developments in biotechnology have permitted the synthesis of large libraries of functional genetic reagents. In addition, high-throughput technologies, such as massively parallel sequencing or flow cytometry, enable the quantitative readout of experimental assays. Ultimately, the convergence of these technologies has propelled the use of functional genomic screens to reveal new insights into biology and disease.

48

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

RESULTS

The comprehensive structural characterization of cancer genomes has generated large catalogues of altered genes in cancer, including those that harbor point mutations or copy number alterations. However, the functional role and significance of many of these cancer- associated genes remain unknown and must be confirmed by functional studies in experimental models. In addition, the inevitability of therapeutic resistance to targeted therapies and the lack of widespread response to immunotherapies necessitate the identification and characterization of novel therapeutic targets and approaches for cancer treatment. Despite the increased categorization and treatment of cancer patients by genetic mutations and other genomic markers, the majority of oncology clinical trials and drug indications are determined by tumor lineage. In addition, many tissues possess tissue-specific biology that may make them amenable to more specific, tissue-targeted treatment. Thus, the identification of lineage-specific dependencies may lead to new biological and therapeutic insights for cancer.

Identification of Differential Genetic Dependencies in Tumor Lineages

To identify lineage-specific genetic dependencies across tumor lineages, we analyzed the

Broad Institute Project Achilles dataset of comprehensive loss-of-function genetic screens in over 500 cancer cell lines (Tsherniak et al., 2017). Project Achilles utilized genome-scale pooled shRNA libraries to investigate genes that are essential for cancer cell growth and proliferation.

Cancer cell lines were transduced with lentivirus encoding shRNAs targeting all human genes

(about 5 shRNAs per gene) and propagated for 16 population doublings. Genomic DNA was isolated and shRNA abundance was quantified by multiplexed massively parallel sequencing.

49

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages shRNA-level dependency scores were calculated by log fold change of shRNA abundance at the late time-point vs. the early time-point. Gene-level dependency scores were generated using

ATARIS or DEMETER, two algorithms developed at Broad Institute to collapse shRNA-level dependency scores to gene-level dependency scores. ATARIS enriches for RNAi reagents whose phenotypic effects are correlated across multiple samples (Shao et al., 2013). DEMETER segregates on-target (i.e., gene) and off-target (i.e., seed) effects of RNAi reagents (Tsherniak et al., 2017). By calculating the difference in means of ATARIS or DEMETER dependency scores for cell lines in a tumor lineage (e.g., melanoma) vs. all other cell lines (e.g., non-melanoma) and ranking genes by their differential dependency, we generated lists of differential genetic dependencies according to tumor lineage. While similar analyses have been performed by

Project Achilles using previous iterations of data (Cheung et al., 2011), our analysis utilized the most recent dataset by Project Achilles, including shRNA dependency data for over 500 cell lines, to investigate lineage-specific differential genetic dependencies in cancer.

This analysis of differential dependencies by tumor lineage was performed for tumor types in the Project Achilles dataset with sufficient numbers of cell lines for analysis (i.e., tumor lineages with dependency data for at least six cell lines) (Table 1-1). These lineage comparison analyses uncovered well-established dependencies and frequently altered genes in cancer, including CTNNB1 in colorectal cancer and KRAS in pancreatic cancer, as well as dependencies recently published by Project Achilles, including PAX8 in ovarian cancer (Table 1-2) (Figure 1-

1) (Cheung et al., 2011). In addition to these genetic vulnerabilities, our analysis revealed previously unidentified putative differential dependencies across different tumor types, including

FOXA1 and SUZ12 in prostate and breast cancers, BCL2L1, CSNK1A1, and PLK1 in colorectal and gastric cancers, BCL2L1 and HNF1B in liver and kidney cancers, and others (Table 1-2)

50

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

(Figure 1-1). These genes represent known oncogenes as well as tissue-specific transcription factors that may be important targets for therapeutic intervention. For example, FOXA1 is transcription factor involved in embryonic development, establishment of tissue-specific gene expression, and regulation of gene expression in differentiated tissues. As a pioneer transcription factor, it can open compacted chromatin to enable DNA binding of other transcription factors.

Previous studies have shown that FOXA1 can facilitate binding of the and the androgen receptor to chromatin in breast cancer and prostate cancer cells and is required for ER and AR transcriptional activity (Hurtado et al., 2011; Robinson et al., 2014). In addition, the shared putative differential genetic dependencies across tumor types allude to the shared developmental origins of the corresponding tissues. For example, colorectal, gastric, and liver cancers all harbor putative dependencies on CTNNB1 and BCL2L1. These genetic dependencies may reflect genes that are important for the development of the gastrointestinal tract and may have continued essential functions in tumors derived from these tissues.

While our analyses nominated many putative differential genetic dependencies in tumor lineages, the analysis of primary screening data can lead to false positive findings and must be validated in experimental models. Additionally, genetic dependencies may be enriched for greater essentiality (i.e., more negative DEMETER scores) in certain lineages, but these dependencies are usually not lineage-exclusive.

51

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Table 1-1. Number of cell lines by tumor lineage in Project Achilles.

Lineage Lines Lung NSCLC 93 Glioblastoma 35 Melanoma 35 Breast 34 Ovarian 33 Colon 25 Renal 24 Lung SCLC 23 Esophageal 21 Gastric 21 AML 18 Pancreas 18 Head and Neck 15 Endometrial 12 Multiple Myeloma 12 Ewings 9 Liver 8 Prostate 8 DLBCL 7 Medulloblastoma 7 Rhabdomyosarcoma 6 Rhabdoid 6 B-ALL 5 Osteosarcoma 4 Rhabdoid 4 Bladder 3 Merkel 3 T-ALL 2 CML 2 Cervix 2 Lung Mesothelioma 2 Meningioma 2 ATL 1 Fibroscarcoma 1 Leiomyosarcoma 1

52

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Table 1-2. Top 30 differential genetic dependencies across tumor types in Project Achilles.

ENDO, endometrium.

Rank MELANOMA PROSTATE BREAST OVARY ENDO 1 BRAF SUZ12 CCM2 ABCE1 GAPDH 2 SOX10 FOXA1 UBR4 UBC CDK2 3 MAPK1 COPB1 ABCB7 NCAPD3 SRSF2 4 ARNT MBNL1 PIK3CA UBR4 UTP6 5 MED1 SLC33A1 HSPD1 THG1L CWC22 6 TBCE GATA2 FOXA1 RPF2 HNRNPL 7 TXNDC17 HNF1B KDM1A ZC3H6 MCM3AP 8 ZEB2 MED14 COPZ1 WDR61 RPAP1 9 HLA-J SMARCA4 SUZ12 PIWIL3 PLK1 10 PELI3 HSPA8 GATA3 RIMKLB RPF2 11 RABGAP1 CCND1 GRPEL1 PARD6B FARP1 12 INCA1 MMS22L COPB1 SRPR SSBP4 13 PPP2R2A SFSWAP XRCC6 VPRBP ANKRD11 14 NCAPD3 FOXK1 ADSL PAX8 WDR61 15 DLG1 EZH2 TUBA1B RAC1 PRKRA 16 OSTCP2 CDK4 SMC2 FAM32A MMS22L 17 SRSF2 BCOR CSNK1A1 SOX1 ZC3H18 18 RCL1 EEPD1 PNPT1 RPS27 SMC5 19 IPO7 PSMC4 CAPZB OSTCP2 TAF8 20 CCNK PSMD12 PSMA4 MCM7 PUF60 21 PROP1 CCDC3 NEDD8 APOL6 ASNS 22 CLEC12B SRSF3 TFRC LUM PDCD11 23 INTS7 UBR5 HCCS RSL1D1 RPS12 24 ACLY PLK1 MCM7 PARD3 TAF4 25 ECM2 HOXC6 SNRPD1 GSPT1 SUPT16H 26 INSL6 SMC5 PPP2CA PSMB6 GAK 27 TOMM40 UBE2L3 NDUFS7 TMC6 UGT2B10 28 KEAP1 CMPK1 GPN3 EXOC3L1 POLR2C 29 CIRH1A DNAJC11 HAUS6 DDIT4L EIF5 30 OGDH YEATS2 SOD2 PSMB5 PSMB5

53

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Table 1-2 (Continued). LARGE, large intestine (colon).

Rank LARGE STOMACH OESOPHAGUS PANCREAS LIVER 1 CTNNB1 CTNNB1 RPF2 KRAS MANBA 2 BCL2L1 CSNK1A1 SMC1A CASP8AP2 BCL2L1 3 PLK1 BCL2L1 PSMC4 SCAP HNF1B 4 ADSL MCM7 RPN2 ATP6V1B2 RPN2 5 SUZ12 NDUFB8 WDR12 NUP214 STAG2 6 CSNK1A1 ATP6V0C UTP6 SMC1A HAUS1 7 WNK1 ZC3H18 RRS1 FZR1 SLC33A1 8 CYFIP1 PIK3CA RBBP4 MED14 RHEB 9 ISCA2 SMC1A MED14 PAFAH1B1 CFLAR 10 MED12 PLK1 RSL1D1 EEF2 CTNNB1 11 PSMB6 MED12 GAR1 NUPL1 MRPL38 12 ICE2 POLR2D PDCD11 RPP40 ARF4 13 PSMD2 ADSL NUPL1 PSMA4 EP300 14 VARS SRSF3 WDR36 COPE EGFR 15 PSMC4 SMC3 POLR2D NUP205 GGCX 16 GAR1 PSMA4 PWP2 PRPF18 ATP6V1B2 17 GATA6 FPGS PSMD6 PSMA2 FIP1L1 18 MMS22L ETF1 NOP58 MED1 COX6C 19 TUBG1 COPE MAK16 HSPE1 GBF1 20 CCM2 CLTC NIFK WNK1 ACSM6 21 NDC80 KRAS NUP205 RAB6A PSMC1 22 APC CDK13 FTSJ3 PSMC2 TIMM44 23 RPL5 SNRPF PMPCB RACGAP1 MAX 24 DHPS EIF2B3 SRPR TP53 DHPS 25 EIF2B3 PAFAH1B1 PPP1R12A PPP1R12A SERPINB8 26 FOXB1 RAD21 SMC3 DHPS NIPBL 27 EEF2 ZFR CENPT EIF2S2 SNRNP25 28 BOLA3 VCP NOL7 VPS28 SLC25A30 29 HNF1B CCNK EIF2S3 GPN3 FKBP1A 30 HSPA8 CAB39 PES1 PSMB1 YAP1

54

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Table 1-2 (Continued). KIDNEY, renal cell carcinoma, GBM, glioblastoma; MEDULLO, medulloblastoma; UPPER, upper aerodigestive tract (head and neck).

Rank KIDNEY GBM MEDULLO UPPER 1 HNF1B KLHL32 BCL2L1 MED12 2 ITGAV PPP1R12A GAPDH RPN2 3 FAM32A INTS7 MAX EGFR 4 PAX8 KTI12 ATP6V0C TBCE 5 TDP1 FASTKD2 SUZ12 SMC1A 6 CENPE AIG1 FLNA COPA 7 COPG1 WDR61 KIAA1958 PSMC4 8 BCL2L1 CLEC12B MASP1 EEF1A1 9 STAM EPB41L5 CCND2 NUPL1 10 SIRPD DENND4A TYMS PSMC2 11 PARD6B NRN1L SEPHS2 PPP1R12A 12 PROP1 ITGAV DENND4A MED14 13 BTNL3 SPOPL TAF1D PSMD6 14 RPN2 CCT3 MYBL2 RPF2 15 PABPN1 ICA1L ACTL6A ATP6V1B2 16 CFLAR TXNDC17 TFR2 URI1 17 PRKCD TNS4 APC POLR2D 18 DYNC1I1 CABIN1 CPSF4 NUP54 19 SST APOF IGF1R WDR12 20 PARD3 CMTM8 SIRPD OSTCP2 21 GPR125 RAB7B NOP58 ANLN 22 MDM4 FAIM UBR5 GPN3 23 MTMR11 CPSF3L FUBP3 CHMP2A 24 B3GNT7 KIAA0825 ACSM6 SRPR 25 MGME1 SMEK1 ZC3H18 EDF1 26 NCAPD3 SCFD1 COL5A2 UNC13B 27 B4GALT4 PUM1 CSE1L MCM7 28 ARAP2 SLC7A11 STRN4 CCT8 29 LMBR1L FRMD3 RNF40 ABCE1 30 OTUB1 MED22 NXF3 PGS1

55

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Table 1-2 (Continued). AML, acute myeloid leukemia; B-ALL, B-cell acute lymphoid leukemia; MM, multiple myeloma; DLBCL, diffuse large B-cell lymphoma.

Rank AML B-ALL MM DLBCL 1 MED1 MSI2 IRF4 SUZ12 2 MYB CPA6 RELB MOCS2 3 CBFB ATF5 STRN4 ECM2 4 TACR2 KCNJ18 AFF1 FARP1 5 CCM2 EBF1 EP300 MBNL1 6 SMARCB1 APC NFKB1 NPTXR 7 BARHL1 STRN4 KCNJ18 POU2F2 8 STRN4 BARHL1 RAD21 MAX 9 MED14 OGDH WDR61 HOXD10 10 RCCD1 CLSTN3 MAX B4GALT4 11 ACLY SUZ12 MOCS2 EBF1 12 ZEB2 ACOT13 FBXW11 KLHL32 13 CPA6 CXCR4 MED1 CYTL1 14 OGDH AFF1 MYH9 SUB1 15 SH2D1A BTK INSL6 ZNF513 16 MED19 MDM4 BARHL1 LYZ 17 FRAT1 CLTB NIPBL TAF11 18 MSI2 OR2H1 ACLY INHBC 19 AFF1 BLVRB DLST MYD88 20 DBX1 MBOAT7 KLHL32 TMEM237 21 ZMIZ1 MAP3K8 ROCK1 UBQLN3 22 OPA1 ACY1 CCDC144NL WBSCR28 23 GALNT8 SOD2 OR2H1 CLINT1 24 CYTL1 CLTA GABPA ARID1B 25 C1orf116 RAB15 MTX1 CCND3 26 RAB15 BLNK RING1 WDR90 27 MED9 NR2C2AP XRCC6 CBLL1 28 PSMA7 ACTL6A TRERF1 NPFFR2 29 ZC3H18 DHX38 ARPC1B MTX1 30 HAS3 SEC22C DGCR6 RAB25

56

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Table 1-2 (Continued). EWING, Ewing sarcoma; OSTEO, osteosarcoma; RMYOS, rhabdomyosarcoma; RHABD, malignant rhabdoid tumor.

Rank EWING OSTEO RMYOS RHABD 1 STRN4 PRDX1 CDK16 MDM4 2 TAF1D CHMP2A PELI3 SUZ12 3 SATL1 GTF3A FLNA HOXC6 4 CCND1 ZNF189 SMARCA4 MCHR1 5 BARHL1 ARF4 ACTL6A TACR2 6 RAB15 NOL7 GLYR1 CLEC7A 7 ATP6V1G3 KIAA1958 ZEB2 B4GALT4 8 AFAP1L1 CUL1 BARHL1 TXNDC17 9 WSB2 SUB1 IFT140 INCA1 10 HOXC6 EIF2S3 EXOSC6 HDAC11 11 C20orf203 SNRPF CABIN1 KAT5 12 KLRF1 TXN SDCBP DKC1 13 SLC5A3 CDH8 SLC25A4 AKNA 14 BHLHE41 SNRNP25 CBLL1 TDP1 15 CDK4 PES1 ECM2 IFT140 16 TMEM86B EXOC3L1 FASTKD2 MTMR11 17 CXCR4 RBBP4 VSX1 UBE2D3 18 KLHL32 MTOR SLC16A12 RAB35 19 HSD17B1 LSM5 APITD1-CORT FAM32A 20 TRERF1 RCL1 ABT1 RIMKLB 21 TPM3 ASNS OTUD4 SPINK2 22 GFI1 HNRNPU MCHR1 NCAPD3 23 WIPI2 TXNDC17 KIAA1958 TAF5L 24 TMEM97 NOM1 PRM3 KIAA1958 25 GFRA3 KLHL32 SATL1 ZNF235 26 HDAC11 MPHOSPH6 CABS1 PEX2 27 RAB7B RILPL1 FAM96B ELAVL2 28 INCA1 SPCS2 DENND4A PLEKHA1 29 SRSF2 HEATR1 TMEM86B KHDRBS1 30 JAK2 DNAJA3 UBN1 SERBP1

57

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Figure 1-1. Known and novel differential genetic dependencies in various tumor lineages.

DEMETER dependency scores were derived from pooled shRNA loss-of-function screens in cancer cell lines (Project Achilles). Negative DEMETER dependency scores indicate greater dependency.

58

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Identification of Differential Genetic Dependencies in Melanoma

To identify differential genetic dependencies in melanoma, we compared dependency scores for melanoma cell lines and non-melanoma cell lines. This analysis generated a ranked list of genetic dependencies with BRAF, SOX10, and MAPK1 as the top ranked genes (Table 1-2)

(Figure 1-2). A majority of melanoma cell lines, including those in Project Achilles, harbor mutations in the BRAF gene. These mutations increase BRAF protein activity and propagate signaling through the MAPK pathway, which includes MEK1 and MEK2 (encoded by MAP2K1 and MAP2K2, respectively), and ERK1 and ERK2 (encoded by MAPK3 and MAPK1) (Figure I-

5). Mutations in BRAF have been linked to dependency on the MAPK pathway and sensitivity to

MAPK pathway inhibitors, including small molecule inhibitors of BRAF (e.g., vemurafenib and dabrafenib) and MEK (e.g., trametinib) (Chapman et al., 2011; Hauschild et al., 2012; Flaherty et al., 2012; Robert et al., 2015). Thus, our finding that BRAF and MAPK1 are differential genetic dependencies in melanoma is supported by previous studies in melanoma and characterization of the MAPK pathway, and represents excellent examples of bona fide genetic dependencies identified using Project Achilles dependency data.

59

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Figure 1-2. Differential genetic dependencies in melanoma. DEMETER dependency scores were derived from pooled shRNA loss-of-function screens in cancer cell lines (Project Achilles).

Negative DEMETER dependency scores indicate greater dependency.

In addition, other putative differential genetic dependencies in melanoma, including

ARNT, HLA-J, PELI3, and ZEB2, have previously identified associations with melanoma. A meta-analysis of genome-wide association studies for melanoma susceptibility has identified associated variants near the ARNT gene, which encodes HIF-1-beta and plays important roles in the cellular response to hypoxia with HIF-1-alpha (Macgregor et al., 2011; Amos et al., 2011).

HLA-J is a transcribed pseudogene from the human leukocyte antigen (HLA) major histocompatibility complex (MHC) gene complex on chromosome 6p that is most related to

60

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

HLA-A and HLA-G, and may modulate response to immune-based therapy (Ragoussis et al.,

1989; Messer et al., 1992). PELI3 is an E3 ubiquitin ligase in the Pellino E3 ubiquitin protein ligase family that mediates lysine-63 (K63)-linked polyubiquitination of substrates (Moynagh,

2014). The protein product of PELI3, Pellino 3, has been shown to activate c-Jun, a major transcription factor in melanoma, in reporter assays (Jensen & Whitehead, 2003). Finally, ZEB2 encodes a well-known factor involved in the epithelial-to-mesenchymal transition (EMT). A recent study identified a transcriptional network that involved ZEB2, MITF, and ZEB1, that controlled melanomagenesis (Denecker et al., 2014). Thus, the analysis of functional genomic data can identify putative genetic dependencies with strong associations to cancer phenotypes.

Model of SOX10 Expression-Based Dependency in Cancer

Our analysis identified SOX10 as a novel differential genetic dependency in melanoma, with the majority of melanoma cell lines exhibiting a lower SOX10 DEMETER dependency score compared to non-melanoma cell lines (Figure 1-2). While SOX10 was a strong dependency in many melanoma cell lines, SOX10 dependency was also observed in two glioblastoma cell lines: LN229 and LN464. LN464 harbors an activating BRAF mutation and has a UV mutation signature, indicating that it is incorrectly annotated and might be a melanoma cell line (personal communication). SOX10 is a transcription factor with important roles in neural crest development, including the formation of neural crest stem cells, maintaining multipotency of neural crest cells, and cell fate specification and differentiation of neural crest- derived cells. This prior knowledge of SOX10 biology led us to investigate the Cancer Cell Line

Encyclopedia (CCLE) dataset of genomic data across over 1,000 cell lines (Barretina et al.,

2012; https://portals.broadinstitute.org/ccle). Analysis of the SOX10 data in CCLE revealed that

61

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

SOX10 mRNA is expressed at high levels in many melanoma cell lines, and in a few breast cancer and glioblastoma cell lines (Figure 1-3).

Figure 1-3. SOX10 mRNA expression in cell lines across tumor types in the Cancer Cell

Line Encyclopedia (CCLE). SOX10 is highly expressed in a majority of melanoma cell lines and in a few cell lines of the breast cancer and lineages. Data from CCLE.

The observation that melanoma cell lines are both dependent on SOX10 and express

SOX10 at high levels led us to a model of genetic dependency in which SOX10 gene expression is linked to SOX10 genetic dependency. In melanoma cell lines and across all cancer cell lines,

62

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

SOX10 expression is highly correlated with SOX10 dependency (Figure 1-4). This high expression of SOX10 in melanoma is likely maintained from melanocytes, the cell of origin of melanoma, and may represent a unique dependency in melanoma compared to other tumor lineages. SOX10 may represent a “master regulator” in melanoma and drives expression of a suite of genes that are important for the continued growth and proliferation of melanoma cells. In contrast, the expression of cancer initiating or promoting genes in other tumor lineages may be driven by other linage-specific master regulator genes, such as PAX8 in ovarian cancer and

FOXA1 in breast and prostate cancers (Cheung et al., 2011; Tsherniak et al., 2017). Thus, these master regulator genes are crucial for tumor formation and maintenance, and they represent novel potential targets for cancer therapy that are orthogonal to current means of treatment. In addition, the systematic investigation of SOX10 dependency may reveal new biological and therapeutic insights for melanoma.

63

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Figure 1-4. SOX10 dependency is correlated with SOX10 expression across cancer. SOX10 mRNA expression data from CCLE; SOX10 dependency data (DEMETER) from Project

Achilles.

Validation of SOX10 Dependency in Melanoma

All stated findings of differential genetic dependencies, including that of SOX10 in melanoma, were computationally derived from pooled shRNA loss-of-function screening data.

To validate the model of SOX10 dependency in melanoma, we generated cell lines with stable knockdown of SOX10 and evaluated their growth dynamics in population doubling assays. We obtained the shRNA reagents targeting SOX10 in the Project Achilles screens from the Broad

Institute Genomic Perturbation Platform. These constitutively active shRNA constructs, which also contain a puromycin resistance cassette, were packaged into lentivirus via transfection with delta 8.9 and VSVg plasmids in 293T cells. Lentivirus harvested from 293T cells were aliquoted and frozen before being used in experiments. A panel of melanoma cell lines were transduced with lentivirus containing SOX10 shRNA constructs and selected for successful integration and

64

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages expression by puromycin selection. Knockdown of SOX10 mRNA and protein were confirmed by reverse transcriptase-quantitative PCR (RT-qPCR) and immunoblot, respectively (Figure 1-

5). Cell lines stably transduced with SOX10 shRNA were seeded for population doubling assays

72 hours post-infection. Cells were counted every 3-4 days and were reseeded if possible.

SOX10 knockdown led to reduced cell number and population doubling in all cell lines expressing SOX10, including melanoma and non-melanoma cell lines (Figure 1-6).

Furthermore, the knockdown of SOX10 was associated with inhibition of cell growth: the return of SOX10 protein levels was correlated with increased cell proliferation in CHL1 cells (Figure

1-7). These experimental findings confirm our correlative model of SOX10 expression and

SOX10 dependency derived from the Project Achilles screening data. In addition, the negative growth phenotype resulting from SOX10 knockdown is observed in melanoma cell lines regardless of mutation status in BRAF or NRAS, two frequently and significantly mutated genes in melanoma representing prominent nodes in the MAPK pathway (Hodis et al., 2012; Cancer

Genome Atlas Network, 2015). This result indicates that SOX10 dependency in melanoma is independent and orthogonal of the dependency of melanoma cells on the MAPK pathway and that genetic or chemical inhibition of SOX10 may be combined with MAPK pathway inhibitors

(e.g., dabrafenib and trametinib) for additive or synergistic effects in cell culture and in patients.

65

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Figure 1-5. shRNA reagents targeting SOX10 reduce SOX10 mRNA and protein levels in representative cell lines. Melanoma cell lines were stably transduced with shRNA reagents targeting SOX10. RNA and protein were isolated from cells and SOX10 RNA and protein levels were quantified by RT-qPCR and immunoblot, respectively.

66

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Figure 1-6. Knockdown of SOX10 leads to reduced cell growth and proliferation only in cell lines expressing SOX10. Melanoma cell lines were stably transduced with shRNA reagents targeting SOX10 and propagated in cell culture. Cells were counted every 4-6 days.

67

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Figure 1-7. shRNAs targeting SOX10 reduce SOX10 protein levels in population doubling assays. Return of SOX10 expression is observed at Day 9 for CHL1, correlating with cell growth. There were insufficient numbers of cells at Day 9 for MEWO to perform immunoblot.

68

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

DISCUSSION

In this study, we have utilized the Project Achilles pooled shRNA loss-of-function screening dataset to reveal lineage-specific genetic dependencies across a range of tumor lineages and further validated SOX10 as a top differential genetic dependency in melanoma.

Additionally, we have confirmed a gene expression-dependency model for SOX10 in cancer, suggesting a useful biomarker for future targeting of SOX10. Finally, we have evaluated the combination of SOX10 knockdown with MAPK pathway inhibitor treatment in melanoma cells.

These findings highlight the power and utility of large-scale functional genomic screens and their subsequent analyses to uncover novel biological insights and potential therapeutic targets. While structural genomic characterization studies, such as whole exome and genome sequencing of patient tumors, have generated catalogues of cancer-associated genes, functional genomic screens provide an additional and crucial level of data to inform our understanding of the underlying genetic basis of cancer. In addition to confirming the essentiality of mutated driver genes in various tumor types, such as BRAF in melanoma and KRAS in pancreatic cancer, our analysis identified cancer essential genes that are not recurrently mutated or altered in human cancer, including SOX10 in melanoma, SUZ12 in prostate and breast cancers, and HNF1B in liver and kidney cancers. Pending validation in functional experiments, these genes represent tumor dependencies with previously unrecognized roles in cancer.

Despite the promise and value of functional genomic screens, they possess inherent shortcomings. For example, shRNAs are notorious for their “off-target” or seed effects (i.e., the unintended knockdown of non-target genes due to sequence similarity in the region of the shRNA that is analogous to the seed sequence in miRNAs). Because shRNAs utilize the same cellular machinery as miRNAs, these off-target effects must be parsed out and removed

69

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages computationally by sophisticated algorithms, such as ATARIS and DEMETER (Shao et al.,

2013; Tsherniak et al., 2017). Additionally, the power of functional genomic screens relies on large sample sizes (i.e., large number of genetic perturbation reagents and large number of cell lines). Ideally, such screens would employ the use of cell lines with identical or very similar genetic backgrounds to minimize the confounding of experimental conclusions by inherent genetic variation or somatic mutation. For example, the presence of BRAF mutations in a majority of melanoma cell lines resulted in the identification of BRAF and MAPK1 as differential genetic dependencies in melanoma, but should more precisely be characterized as differential genetic dependencies in BRAF-mutant cell lines. Due to this intrinsic genetic heterogeneity in cell lines, class-based comparisons of functional genomic screening data require sufficient number of cell lines in each class. Computational analysis of this problem has determined that at least 8 cell lines are needed for such analyses of Project Achilles data, limiting the number of lineages that can be analyzed in this manner (Cheung et al., 2011).

We have described SOX10 dependency as part of a gene expression-dependency relationship, where SOX10 expression serves as biomarker for SOX10 dependency in cancer.

Previous analyses of Project Achilles data have described similar expression-driven models of genetic dependency, as well as additional dependency classes, including mutation-driven,

CYCLOPS (i.e., genes for which hemizygous copy number loss is predictive of dependency), and paralog deficiency (i.e., functional loss of one paralog is associated with dependency on another paralog) (Shao et al., 2013; Tsherniak et al., 2017). Expression-driven dependencies include lineage-specifying transcription factors, such as SPDEF in prostate and breast, NKX2-1 in lung, and PAX8 in ovary. Many of these dependencies are transcription factors and act as master regulators in the specification and survival of a particular tissue lineage.

70

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

In addition to the previously discussed SOX10 expression-dependency relationship, we also determined that SOX10 is a genetic dependency in SOX10-expressing melanoma cell lines regardless of BRAF or NRAS mutation status. This finding is of clinical importance because there are currently no targeted therapies for NRAS mutant or BRAF WT/NRAS WT melanomas.

Patients with tumors not harboring BRAF mutation cannot attempt therapeutic regimens that include BRAF and MEK inhibitors, which have shown considerable response and progression- free survival rates in patients with BRAF-mutant tumors. Thus, SOX10 dependency represents a potential therapeutic target in thisa melanoma patient population.

SOX10 has major roles in neural crest development, including maintaining the multipotency of neural crest cells (Kim et al., 2003; Kelsh, 2006). Multipotency describes the ability of a cell to generate more than one distinct cell type. Multipotent cells are also capable of self-renewal (i.e., generating multipotent cells of equivalent potential). Thus, the maintenance of

SOX10 expression in melanoma cells from neural-crest derived melanocytes may provide melanoma cells with multipotent capabilities, including transcriptional and morphologic plasticity, that allow them to more easily adapt to molecular and cellular challenges as they undergo oncogenic transformation. Although not assessed in our study, the migratory abilities of melanoma cells in humans may be dependent on SOX10 as SOX10 is involved in the delamination of neural crest cells from the neural plate (McKeown et al., 2005; Cheung &

Briscoe, 2003). Additional studies of SOX10 function in melanoma cell invasion and metastasis are necessary to further elucidate the role of SOX10 in these cancer phenotypes.

In the adult human, SOX10 expression is restricted to melanocytes and glial cells

(including oligodendrocytes and Schwann cells), all of which originate from the neural crest.

Additionally, SOX10 mutations in humans, while not observed in cancer, have been well-

71

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages characterized in genetic diseases, including Waardenburg syndrome Type 4C (WS4C) and

PCWH. These syndromes are characterized by deafness, pigment abnormalities, enteric aganglionosis, as well as severe demyelination (in the case of PCWH). Furthermore, animal models of homozygous SOX10 mutation are embryonic lethal (Southard-Smith et al., 1998;

Kapur, 1999). Thus, systemic therapeutic targeting of SOX10 may lead to undesirable effects in normal melanocytes and glial cells at the least, and severe impairment of the nervous system at the most. However, targeted delivery of SOX10 therapeutics to melanoma cells would spare glial cells from unintended consequences of systemic therapy.

72

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

METHODS

Computational analysis of Project Achilles data

All genetic dependency data, including shRNA, ATARIS, and DEMETER datasets, were obtained from Project Achilles: https://portals.broadinstitute.org/achilles. Gene-level dependencies were determined from ATARIS and DEMETER scores for shRNA knockdown screens (Shao et al., 2013; Tsherniak et al., 2017). Differential dependency scores were determined by calculating the difference in means of DEMETER scores for melanoma cell lines and non-melanoma cell lines, and statistical significance was determined by student’s t-test.

Cell lines and reagents

All cell lines were obtained from and were identity confirmed with DNA fingerprinting by the

Broad Institute Genomics Platform. SKMEL19, COLO679, WM983B, UACC62, RVH421,

COLO741, SKMEL28, A375, COLO205, CHL1, WM793, SF295, and HCT116 were grown in

RPMI medium, 10% FBS. G361, WM88, WM2664, A2058, SKMEL24, HS294T, IGR39,

LOXIMVI, LN229, and LN464 were grown in DMEM medium, 10% FBS. MEWO and

SKMEL5 were grown in EMEM medium, 10% FBS. Antibodies for immunoblot were obtained from Abcam (SOX10 rabbit monoclonal, ab155279 [EPR4007]) and Sigma-Aldrich (vinculin mouse monoclonal, V9131).

73

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

RNA interference, lentivirus production, and lentiviral transduction

All shRNA constructs (shSOX10-1, shSOX10-2, shSOX10-3, shSOX10-4, shSOX10-5, shLUC) were obtained from the Broad Institute Genetic Perturbation Platform: https://portals.broadinstitute.org/gpp/public/. 293T cells were transfected with pLKO.1 plasmid

(encoding shRNA and puromycin-resistance gene), delta 8.9 plasmid, and VSV-G plasmid using

FuGENE transfection reagent. Lentivirus was harvested from culture supernatant 72 hours post- transfection and frozen at -80 °C. Melanoma cells were transduced with lentivirus using polybrene in a 30 min spin-infection. Lentivirus was removed 24 hours post-transduction and cells were selected for stable transduction using puromycin for 24 hours. Stable transduction and expression of shRNA constructs was confirmed by qRT-PCR and immunoblot for mRNA and protein knockdown, respectively.

Clone ID shSOX10-1 TRCN0000018984 shSOX10-2 TRCN0000018985 shSOX10-3 TRCN0000018986 shSOX10-4 TRCN0000018987 shSOX10-5 TRCN0000018988 shLUC TRCN0000072243

Immunoblot analysis

Cells were washed once with cold PBS and lysed passively with cold lysis buffer (1% NP40, protease inhibitor cocktail (Roche), phosphatase inhibitor cocktail sets I and II (CalBioChem)).

74

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

Protein lysates were quantified (Bradford assay), normalized, denatured (95 °C), analyzed by

SDS gel electrophoresis on 4-20% Tris-Glycine gels (Invitrogen). Resolved proteins were transferred to nitrocellulose or PVDF membranes, blocked in LiCOR blocking buffer, and probed with primary antibodies. After appropriate incubation with the appropriate secondary antibody, proteins were imaged and quantified using an Odyssey CLx scanner (LiCOR).

Population doubling and drug treatment assay

Cells transduced by shRNA and selected by puromycin were seeded for population doubling assay 72 hours post-infection. Cells were plated in triplicate in 12-well plates, counted every 3-4 days, and reseeded to maintain growth if possible. Stable knockdown of SOX10 protein was confirmed by immunoblot throughout experiments.

75

Chapter 1: Identification of Differential Genetic Dependencies in Tumor Lineages

ACKNOWLEDGEMENTS

SOX10 knockdown project and experimental design were done in collaboration with Cory

Johannessen. Barbara Weir provided clarification and assistance with Project Achilles data analysis.

76

CHAPTER 2

Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

SUMMARY

Previous studies of SOX10 have described its crucial role in the development of the neural crest, a progenitor cell population from which melanocytes and other cell types derive. As a transcription factor, SOX10 recognizes and binds to DNA to regulate the expression of target genes, including MITF, which has well-established functions in melanocyte development.

Transcription factors typically exhibit thousands of localization sites throughout the genome and can control the up- or down-regulation of suites of genes driving molecular functions in the cell.

Due to their vast influences on gene expression, transcription factors are involved in the majority of normal cellular processes and many transcription factors are aberrantly expressed or activated in cancer. Experimental approaches to investigate transcription factors include chromatin immunoprecipitation and transcriptional profiling, which are used to describe the genes that are bound and regulated by transcription factors, respectively. When applied at a genome-wide scale, these techniques can reveal the complete set of bound and regulated genes for a transcription factor of interest.

In this study, we characterized the cistrome and transcriptome of SOX10 in a panel of melanoma cell lines with high and low expression of MITF (MITF-high and MITF-low melanoma). SOX10 chromatin immunoprecipitation (ChIP) was optimized and confirmed by quantitative PCR (ChIP-qPCR) of known SOX10 binding sites before SOX10 ChIP-enriched

DNA fragments were comprehensively profiled by massively parallel sequencing (ChIP-seq).

SOX10 ChIP-seq peaks exhibited high sequence conservation, enrichment in promoter and 5’

UTR regions, and enrichment of SOX binding motifs. Of note, motif analysis revealed an enriched inverted repeat motif in SOX10 ChIP-peaks, suggesting a model of SOX10 binding to the melanoma genome as a dimer. In parallel, SOX10 knockdown was optimized and confirmed

78

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma by reverse transcription-qPCR (RT-qPCR) of known SOX10 target genes before RNA transcripts were profiled by RNA-sequencing (RNA-seq). Genes bound or regulated by SOX10 were determined by appropriate thresholds using scores generated by BETA or GFOLD, respectively. Integration of these gene lists determined the core set of SOX10 target genes.

Additionally, gene set analysis of the bound and regulated gene lists revealed shared target gene sets between SOX10 and MYC, an oncogene with well-established roles in cell cycle progression and cellular transformation. Analysis of differential SOX10 ChIP peaks in MITF- high and MITF-low melanoma cell lines led to the identification of distinct SOX10 transcriptional programs in these two classes of melanoma. SOX10 ChIP peaks in MITF-high melanoma are characterized by binding motifs for MITF and other basic helix-loop-helix

(bHLH) transcription factors, while those in MITF-low melanoma exhibit binding motifs for

FOS-JUN family and other basic (bZIP) transcription factors. Taken together, these studies comprehensively characterize the cistrome and transcriptome of SOX10, providing a core set of SOX10 target genes, in melanoma and describe common and distinct SOX10 transcriptional programs in melanoma

79

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

INTRODUCTION

Transcription factors are proteins that normally reside in the cell nucleus and regulate the expression of genes. They typically contain multiple protein domains, including a DNA-binding domain that allows for direct contact with DNA, a ligand-binding domain that receives inputs from the cell interior or the extracellular environment, and an activating or repressive domain that carries out its gene regulatory function. Transcription factors are generally classified into groups that share similar functional domains. Examples of DNA-binding domains include: high mobility group (HMG), basic helix-loop-helix (bHLH), basic leucine zipper (bZIP), etc. Each distinct DNA-binding domain possesses a preferential binding motif, a short sequence of DNA with shared nucleotide identity and biochemical characteristics.

Transcription factors play important roles in healthy organisms, including embryonic development, response to intracellular and extracellular signals, and cell cycle regulation.

Because of their central role in maintaining cellular identity and tissue homeostasis, abnormal expression or activity of transcription factors may lead to disease phenotypes, including cancer.

Several transcription factors have been implicated in cancer, including the estrogen receptor (ER, encoded by ESR1), the androgen receptor (AR, encoded by AR), Erg and other ETS family members, ETV1, ETV4, and ETV5 (encoded by ERG, ETV1, ETV4, and ETV5, respectively),

TCF3 and TCF4 (encoded by TCF7L1 and TCF7L2, respectively), the alpha (encoded by RARA), c-Myc and its family members, L-Myc and N-Myc (encoded by MYC,

MYCL, and MYCN, respectively), and (encoded by TP53). With the exception of c-Myc and p53, most transcription factors are expressed and regulate transcriptional programs in a tissue- specific manner. For example, ESR1 is only expressed in normal and malignant cells of the female reproductive system. Thus, transcription factors operant in normal tissue may maintain

80

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma their expression and activity levels after transformation into malignant tissue. Furthermore, since transcription factors regulate the expression of a suite of genes and those genes may be required for cell growth and survival, transcription factors may also be required for tumor development.

The study of transcription factors has been facilitated in recent years by the development of genome-scale characterization methods, including ChIP-seq and RNA-seq. ChIP-seq combines antibody-based enrichment of genomic loci bound by a protein of interest with massively parallel sequencing to systematically determine the complete set of localization sites for a transcription factor or histone modification throughout the genome. RNA-seq, when augmented by genetic perturbation reagents such as shRNA or CRISPR-Cas9, can yield precise and quantitative measurements of gene expression regulation by transcription factors. The integration of genomic localization and expression data from ChIP-seq and RNA-seq experiments allow for the powerful characterization of transcription factor target genes and pathways.

81

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

RESULTS

Our studies identified SOX10 as a major differential genetic dependency in melanoma cell lines that express SOX10 and determined that this gene expression-dependency relationship exists regardless of BRAF or NRAS mutation status. Due to the established role of SOX10 as a key transcriptional regulator of the development of neural crest-derived cells, including melanocytes, we sought to comprehensively characterize the SOX10 transcriptional program in melanoma, including the set of SOX10 localization sites in the melanoma genome and the set of genes that are transcriptionally regulated by SOX10. Transcriptional and phenotypic studies of melanoma tumors and cell lines have described the existence of two major classes of melanoma:

MITF-high melanoma, which are more proliferative and sensitive to MAPK pathway inhibitors, and MITF-low melanoma, which are more invasive and resistant to MAPK pathway inhibitors

(Konieczkowski et al., 2014; Muller et al., 2014; Wellbrock & Arozarena, 2015). To enable the downstream analysis of SOX10 target genes in MITF-high and MITF-low melanoma, we selected cell lines for profiling to be inclusive of high and low levels of MITF RNA expression

(Table 2-1). The majority of these cells were characterized for both SOX10 binding and transcriptional regulation to maximize the overlap and potential integration of data.

82

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Table 2-1. Melanoma cell lines characterized by SOX10 ChIP-seq and shSOX10 RNA-seq.

Cell Line SOX10 exp MITF exp ChIP-seq RNA-seq WM983B SOX10+ MITF-high profiled profiled SKMEL19 SOX10+ MITF-high profiled profiled G361 SOX10+ MITF-high profiled profiled RVH421 SOX10+ MITF-high -- profiled WM2664 SOX10+ MITF-high -- profiled COLO679 SOX10+ MITF-high profiled profiled UACC62 SOX10+ MITF-high profiled profiled WM88 SOX10+ MITF-high profiled profiled A2058 SOX10+ MITF-low profiled profiled SKMEL24 SOX10+ MITF-low profiled profiled HS294T SOX10+ MITF-low profiled profiled WM793 SOX10+ MITF-low profiled profiled IGR39 SOX10- -- profiled -- LOXIMVI SOX10- -- profiled --

Characterization of the SOX10 Cistrome in Melanoma

In an effort to characterize the cistrome, the complete set of genomic binding sites, of

SOX10 in melanoma, we optimized and performed chromatin immunoprecipitation followed by massively parallel sequencing (ChIP-seq) of genomic regions bound by SOX10 in a panel of melanoma cell lines (Table 2-1). Optimization steps included refining the sonication duration to generate optimal chromatin fragments for subsequent sequencing and determining the ideal concentration of SOX10 antibody for immunoprecipitation. Successful ChIP-mediated enrichment of SOX10-bound loci was confirmed by ChIP-qPCR using known SOX10 binding sites. These ChIP-enriched DNA samples were selected for sequencing library preparation, sequencing, and analysis. All optimization steps and SOX10 ChIP experiments were performed in our laboratory, while the library preparation, sequencing, and initial data processing were performed by the Dana-Farber Cancer Institute Center for Functional Cancer Epigenetics

83

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

(CFCE) and the Dana-Farber Cancer Institute Molecular Biology Core Facility (MBCF).

Downstream analysis of SOX10 ChIP-seq data and integration with transcriptional regulation data were performed in collaboration with DFCI CFCE.

In order to optimize the ChIP protocol for the enrichment of genomic loci endogenously bound by SOX10, we performed a series of experiments evaluating the sonication and immunoprecipitation steps. First, genomic DNA from the panel of selected cell lines was fragmented using a water bath sonicator for a range of sonication periods: 10 min, 12.5 min, 15 min, 17.5 min, and 20 min. The resulting DNA fragments were analyzed by agarose gel electrophoresis. The vast majority of cell lines produced optimally sized chromatin fragments for subsequent sequencing in the 100-500 bp range at 15 min of sonication (Figure 2-1). Thus, a sonication step of 15 min was used for all ChIP experiments. To determine the optimal antibody concentration for the immunoprecipitation step, SOX10 ChIP was performed using increasing concentrations of SOX10 antibody: 1 μg, 2 μg, and 4 μg. In addition, 4 μg of IgG was included as a non-specific binding control. The resulting DNA fragments were assayed for enrichment at known SOX10 binding sites: MITF, ERBB3, and MIA (Bondurand et al., 2000; Potterf et al.,

2000; Verastegui et al., 2000; Prasad et al., 2011; Graf et al., 2014). Genomic regions serving as negative controls included ACTB, GAPDH, and RNA28S5. Maximal enrichment at known

SOX10 binding sites and minimal enrichment at negative control regions was observed using 2

μg SOX10 antibody (Figure 2-2). Our ChIP-qPCR assays revealed slight enrichment at ACTB and GAPDH, suggesting that these general housekeeping genes may in fact be bound by SOX10 and may not be the best negative control regions for SOX10 ChIP.

84

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-1. Optimization of sonication for ChIP-seq. Chromatin was sheared by sonication for increasing durations and analyzed by agarose gel electrophoresis.

Figure 2-2. Optimization of antibody concentration for ChIP-seq. Crosslinked chromatin fragments were enriched by increasing amounts of SOX10 antibody or IgG and ChIP-enriched

DNA fragments were quantified by qPCR.

85

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Histone modifications are chemical changes on histone proteins that regulate gene expression and are associated with distinct chromatin states. In particular, acetylated lysine 27 on histone H3 (H3K27ac) marks active promoters and strong enhancers across human cell types

(Mikkelsen et al., 2007; Ernst et al., 2011). Furthermore, H3K27ac marks co-localize with

BRD4, which is disproportionately found at “super enhancers” (Whyte et al., 2013; Knoechel et al., 2014). To investigate the potential role of SOX10 in establishing or maintaining enhancers in the melanoma genome, H3K27ac ChIP was performed in parallel with SOX10 ChIP in the same panel of melanoma cell lines.

After optimization of the sonication and immunoprecipitation steps, we performed

SOX10 ChIP and H3K27ac ChIP in a panel of MITF-high and MITF-low cell lines and assayed the resulting DNA fragments for enrichment at known SOX10 binding sites by qPCR. Melanoma cell lines were crosslinked by formaldehyde treatment and lysed. Cell lysates were mechanically homogenized, sonicated using a water bath sonicator, then incubated with SOX10 or H3K27ac antibody and magnetic beads or left untreated (i.e., input control). Chromatin fragments bound by SOX10 or H3K27ac were separated by magnetism, decrosslinked by heat, and quantified for known SOX10 binding sites (MITF, ERBB3, and MIA) by qPCR. Melanoma cell lines expressing high levels of MITF mRNA (i.e., MITF-high melanoma) exhibited enrichment of SOX10 and

H3K27ac at the MITF locus (Figure 2-3), consistent with the model of SOX10-mediated regulation of MITF expression as well as the correlation of H3K27ac marks and active gene expression. While melanoma cell lines with low MITF mRNA expression (i.e., MITF-low melanoma) did not show enrichment of SOX10 or H3K27ac at the MITF locus (Figure 2-4), both MITF-high and MITF-low melanoma cell lines demonstrated enrichment of SOX10 at

ERBB3 and MIA loci (Figure 2-3, Figure 2-4), suggesting that these SOX10 localization sites

86

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma are common across melanoma regardless of MITF expression status. Notably, SOX10 was not enriched at these known SOX10 binding sites in melanoma cell lines that do not express SOX10

(i.e., biological negative controls), indicating that our SOX10 antibody for SOX10 ChIP is specific for SOX10 protein in melanoma cells (Figure 2-5).

Figure 2-3. Enrichment of SOX10 and H3K27ac at known SOX10 binding sites in MITF- high melanoma cell lines. SOX10 ChIP-enriched DNA fragments were quantified by qPCR for known SOX10 binding sites (MITF, ERBB3, MIA), housekeeping genes (ACTB, GAPDH), and negative control regions (RNA28S5, RHO, CHEK2).

87

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-4. Enrichment of SOX10 and H3K27ac at known SOX10 binding sites in MITF- low melanoma cell lines. SOX10 ChIP-enriched DNA fragments were quantified by qPCR for known SOX10 binding sites (MITF, ERBB3, MIA), housekeeping genes (ACTB, GAPDH), and negative control regions (RNA28S5, RHO, CHEK2).

88

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-5. SOX10 and H3K27ac are not enriched at known SOX10 binding sites in melanoma cell lines that do not express SOX10. SOX10 ChIP-enriched DNA fragments were quantified by qPCR for known SOX10 binding sites (MITF, ERBB3, MIA), housekeeping genes

(ACTB, GAPDH), and negative control regions (RNA28S5, RHO, CHEK2).

To comprehensively determine the complete set of SOX10 and H3K27ac localization sites in the melanoma genome, we performed massively parallel sequencing of SOX10 and

H3K27ac ChIP-enriched DNA fragments from melanoma cell lines. SOX10 and H3K27ac ChIP

DNA samples were selected for sequencing based on their confirmed enrichment of known

SOX10 binding sites by qPCR. ChIP and input control samples were processed and sequencing libraries were prepared by the Dana-Farber Cancer Institute Center for Functional Cancer

Epigenetics (DFCI CFCE) and multiplexed sequencing was performed by the Dana-Farber

Cancer Institute Molecular Biology Core Facility (DFCI MBCF). Raw sequencing reads were mapped to the genome using Bowtie and peaks were called using MACS. SOX10 ChIP peaks were analyzed for sequence conservation by PhastCons, genomic feature annotation by CEAS, and motif enrichment by SeqPos. Samples from melanoma cell lines that express SOX10 generated approximately 20,000 SOX10 ChIP peaks, while samples from melanoma cell lines

89

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma that do not express SOX10 (i.e., biological negative controls) produced approximately 2,000

SOX10 ChIP peaks, likely representing experimental noise from ChIP experiments. SOX10

ChIP peaks in melanoma cell lines that express SOX10 exhibited high levels of sequence conservation around peak regions, suggesting that these genomic loci have evolutionary and functional significance (Figure 2-6). In addition, these SOX10 ChIP peaks were enriched in promoter and 5’UTR regions compared to the normal genomic distribution of genomic elements, matching the typical binding patterns of transcription factors to gene promoters and enhancers in

5’UTR regions to regulate gene expression (Figure 2-7). Finally, motif analysis of SOX10 ChIP peaks for recurrent transcription factor binding motifs identified SOX family binding motifs

(ACAA(A/T)G) as enriched in SOX10 ChIP peaks, indicating that these peak regions contain

SOX binding motifs that are shared across SOX transcription factor family members (Figure 2-

8). SOX10 ChIP peaks in negative control cell lines that do not express SOX10 (e.g., IGR39 and

LOXIMVI) did not exhibit sequence conservation, were not enriched in promoter and 5’UTR regions, and were not enriched for SOX family binding motifs. Taken together, these observations provide evidence for these SOX10 ChIP peak regions as bona fide SOX10 localization sites in melanoma.

90

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-6. SOX10 ChIP peaks demonstrate high sequence conservation only in melanoma cell lines that express SOX10 (left) compared to cell lines that do not express SOX10

(right). SOX10 ChIP-seq peaks were analyzed for sequence conservation by PhastCons.

Figure 2-7. SOX10 ChIP peaks are enriched in promoter and 5’UTR regions in melanoma cell lines that express SOX10. SOX10 ChIP-seq peaks were analyzed for genomic feature annotation and distribution by CEAS.

91

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-8. SOX10 ChIP peaks are enriched for SOX family binding motifs. Motif analysis was performed using SeqPos.

In addition to the enrichment of known SOX transcription factor family member motifs in SOX10 ChIP peaks, we also observed a high scoring de novo inverted repeat motif for SOX10

ChIP peaks in melanoma cell lines that express SOX10 (Figure 2-8). This inverted repeat motif consisted of two SOX family motifs (ACAA(A/T)G) separated by four nucleotides

(ACAA(A/T)GNNNNC(T/A)TTGT). This finding immediately suggested that SOX10 may bind to the genome as a dimer. In the SOX family of transcription factors, SOX10 is most related to

92

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma other members of the subfamily of SOXE proteins, SOX8 and SOX9. Previous studies have demonstrated via biochemical methods that SOX10 can bind to DNA as a monomer or as a dimer via cooperative binding with other SOXE transcription factors (Peirano & Wegner, 2000;

Schlierf et al., 2002). Additionally, bioinformatics approaches have identified palindromic SOX binding sites enriched in regulatory regions in melanoma cells (Huang et al., 2015). To determine whether certain genes may be bound in a differential manner by SOX10 (i.e., as a monomer or as a dimer), we performed additional motif analysis to identify peaks that exhibit monomer or dimer motifs. Peaks were filtered by those that contained either a monomer or dimer motif and also occurred within 10 kb of a transcriptional start site. From this data, genes bound by SOX10 were classified as possessing monomer motif(s) only, dimer motif(s) only, or monomer or dimer motifs (i.e., genes may harbor multiple SOX10 peaks). More than half (50-

65%) of peaks with a SOX10 motif possessed a dimer motif, less than half (30-50%) had a monomer motif, and very few genes (less than 5%) exhibited both monomer and dimer motifs

(Figure 2-9). These findings suggest that SOX10 localization sites and their associated genes are likely preferentially bound by SOX10 as a monomer or as a dimer, but not both. Furthermore, these different binding modes may lead to distinct effects on SOX10 transcriptional activity or may be associated with unique transcriptional programs or cell states. For example, MITF and its target genes, DCT and BCL2A1, only possess dimer motifs, while other putative SOX10 target genes, including ERBB3, ETV1, and ETV6, exhibit monomer motifs, suggesting that the MITF- related transcriptional program may be regulated by SOX10 dimer binding. To our knowledge, our study was the first to characterize SOX10 genomic localization sites in a panel of melanoma cell lines and to demonstrate that SOX10 binds to the genome as a monomer and as a dimer at a genome-wide level.

93

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-9. Majority of genes bound by SOX10 possess monomer or dimer binding motifs, but not both. SOX10 ChIP peaks in SKMEL19, WM983B, COLO679, HS294T, and WM793 cells were analyzed for the presence of monomer or dimer SOX motifs occurring within 10 kb of a transcriptional start site. Genes associated with SOX10 peaks were categorized by the presence of monomer only, dimer only, or monomer and dimer motifs.

The SOX10 ChIP-seq experiments characterized the complete set of SOX10 localization sites in the genome of melanoma cell lines. These peaks occur throughout the genome in coding and non-coding genomic regions and are likely associated with genes and their expression. To determine the set of genes that are bound by SOX10, we evaluated our catalogue of SOX10 ChIP peaks using BETA and identified genes that are recurrently associated with SOX10 ChIP peaks across melanoma cell lines. SOX10 ChIP peaks were analyzed by the Binding and Expression

Target Analysis (BETA) algorithm, which assigns an associated gene and score to each peak.

94

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

BETA infers the direct target genes from ChIP-seq data based on peak intensity and proximity to gene transcriptional start sites. Each peak is assigned a BETA score, where a higher BETA score indicates a higher likelihood of being a true target gene. To identify the core SOX10 cistrome in melanoma, we evaluated genes based on their SOX10 ChIP BETA scores across our panel of 10 profiled melanoma cell lines that express SOX10. A gene was considered bound by SOX10 in a given cell line if its BETA score was greater than 0.6 in that cell line. This threshold was chosen due to the scoring of MITF as a known target of SOX10 in MITF-high melanoma cell lines.

Genes were scored across all profiled cell lines for shared or unique binding by SOX10. Over

12,000 genes are bound by SOX10 in at least one cell one, while less than 3,000 genes are shared among 5 out of 10 profiled cell lines (Figure 2-10). However, these genes may be bound only in certain subsets of melanoma and may not represent a consensus list of common SOX10-bound genes in melanoma. Thus, we applied a threshold of 8 out of 10 cell lines to define the core

SOX10 cistrome in melanoma, which generated a list of 774 genes. This list of genes includes

SOX10, known SOX10 target genes (S100A1, S100B, ERBB3, and MITF), and novel candidate

SOX10 target genes with known roles in melanoma and oncogenesis (BRD2, BRD4, DUSP4,

DUSP6, ETV4, ETV5, FOSL1, JUNB, and SNAI2). Many of these genes are bound by SOX10 in all 10 profiled cell lines (including ERBB3, ETV4, FOSL1, JUNB, and SOX10), while some genes, such as MITF, were only bound by SOX10 in 8 cell lines. This distinction may reflect true biology (i.e., MITF is only highly expressed in MITF-high melanoma) or may be due to lower detection limits for some of the profiled cell lines. In addition, these findings suggest that SOX10 regulates the expression of a wide range of cancer-associated genes and may partially the dependency on SOX10 in melanoma.

95

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-10. SOX10 exhibits diverse binding sites among melanoma cell lines, but possesses a smaller set of shared bound genes in melanoma. Genes were scored for their binding by

SOX10 (i.e., BETA > 0.6) across profiled cell lines. Over 12,000 genes are bound by SOX10 in at least one cell line, but less than 3,000 genes are shared among half of the profiled cell lines.

To investigate the molecular pathways of genes bound by SOX10, we compared the list of 774 genes bound by SOX10 in at least 8 cell lines to the curated gene sets in the Molecular

Signatures Database (MSigDB) using a hypergeometric test. The top scoring gene set included targets of SMAD2 or SMAD3 (Table 2-2). The SMAD2 and SMAD3 proteins are principally involved in signaling through the TGF-beta pathway. These SMAD binding sites, which were originally identified by ChIP-chip in keratinocytes, were significantly enriched in ETS and

TFAP2 binding elements and knockdown of either ETS1 or TFAP2A resulted in overall alteration of TGF-beta-induced transcription (Koinuma et al., 2009). Remarkably, TFAP2A is a neural plate border specifier gene that is expressed during the early stages of neural crest development. Also, TFAP2A and other neural plate border specifiers drive the expression of neural crest specifier genes, including ETS1 and SOX10. Thus, the finding that SOX10 and

96

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

SMAD2/3 share target genes may point to an even more upstream level of regulation during development, where the shared targets of SOX10 and SMAD2/3 may be relevant for multiple stages of neural crest development.

Table 2-2. MSigDB gene sets with significant overlap with 774 genes bound by SOX10.

# Genes in # Genes in Gene Set Name Gene Set (K) Overlap (k) k/K p-value FDR q-value KOINUMA_TARGETS_OF_SMAD2_OR_SMAD3 824 74 0.0898 5.17E-39 2.45E-35 BLALOCK_ALZHEIMERS_DISEASE_UP 1691 100 0.0591 7.56E-37 1.79E-33 DANG_BOUND_BY_MYC 1103 77 0.0698 4.71E-33 7.43E-30 MARTENS_TRETINOIN_RESPONSE_DN 841 66 0.0785 2.46E-31 2.84E-28 BENPORATH_SOX2_TARGETS 734 62 0.0845 3.01E-31 2.84E-28 KRIEG_HYPOXIA_NOT_VIA_KDM3A 770 59 0.0766 1.56E-27 1.23E-24 GRAESSMANN_APOPTOSIS_BY_DOXORUBICIN_DN 1781 89 0.05 2.01E-27 1.36E-24 GOBERT_OLIGODENDROCYTE_DIFFERENTIATION_DN 1080 68 0.063 1.29E-26 7.63E-24 GRADE_COLON_CANCER_UP 871 60 0.0689 1.46E-25 7.13E-23 MARSON_BOUND_BY_FOXP3_UNSTIMULATED 1229 71 0.0578 1.51E-25 7.13E-23 BUYTAERT_PHOTODYNAMIC_THERAPY_STRESS_UP 811 57 0.0703 9.36E-25 4.03E-22 GRAESSMANN_APOPTOSIS_BY_DOXORUBICIN_UP 1142 67 0.0587 1.64E-24 6.47E-22 HSIAO_HOUSEKEEPING_GENES 389 39 0.1003 1.11E-22 4.02E-20 BENPORATH_MYC_MAX_TARGETS 775 53 0.0684 1.55E-22 5.23E-20 KRIGE_RESPONSE_TO_TOSEDOSTAT_6HR_UP 953 58 0.0609 4.36E-22 1.38E-19 MARSON_BOUND_BY_FOXP3_STIMULATED 1022 60 0.0587 4.86E-22 1.44E-19 KRIGE_RESPONSE_TO_TOSEDOSTAT_24HR_UP 783 52 0.0664 1.46E-21 4.06E-19 SMID_BREAST_CANCER_BASAL_UP 648 47 0.0725 3.86E-21 1.01E-18 ZWANG_CLASS_1_TRANSIENTLY_INDUCED_BY_EGF 516 42 0.0814 7.25E-21 1.80E-18 WAKABAYASHI_ADIPOGENESIS_PPARG_RXRA_BOUND_8D 882 52 0.059 2.66E-19 6.28E-17

Other significantly overlapping gene sets included targets in embryonic stem cells, genes up-regulated in brains of patients with Alzheimer’s disease, and genes down-regulated during differentiation of oligodendroglial precursor cells (Table 2-2). SOX2 and SOX10 are both SOX transcription factor family members and may share target genes or similar binding motifs. Additionally, SOX10 targets in melanoma may be enriched for stem cell-like phenotypes,

97

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma such as the maintenance of multipotency in neural crest development, and these targets may be similar to those for SOX2 in stem cells. As previously mentioned, SOX10 has been shown to play a major role in the development of neural crest-derived cells, including oligodendrocytes and other glial cells. Thus, SOX10 likely binds to and regulates the expression of genes with altered regulation during normal oligodendrocyte differentiation or disease phenotypes in

Alzheimer’s disease. In particular, the up-regulation of a suite of genes involved in maintaining the multipotency of oligodendroglial precursors may contribute to Alzheimer’s disease pathology by preventing the preservation of myelinated axons.

Two of the top 15 gene sets included MYC targets or bound genes (Table 2-2). This finding indicates that SOX10 may bind to genes that are also bound by MYC. Since MYC is a well-established oncogene with demonstrated roles in cell cycle progression and cellular transformation, this observation suggests that SOX10 may mediate similar oncogenic phenotypes in melanoma. Furthermore, the existence of shared target genes suggests that SOX10 dependency may be partially explained by MYC and its downstream targets.

Characterization of the SOX10 Transcriptome in Melanoma

Comprehensive genomic localization studies, such as ChIP-seq, can only define the full set of localization sites in the genome for a transcription factor. While binding of transcription factors and other chromatin-associated proteins to the genome is associated with activity, true target genes must also demonstrate changes in gene expression when the transcription factor in question is perturbed. To elucidate the set of genes whose expression is regulated by SOX10, we optimized and performed transcriptional profiling after shRNA-mediated knockdown of SOX10 or luciferase (LUC) control in the same panel of melanoma cell lines profiled by SOX10 ChIP-

98

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma seq (Table 2-1). The major optimization step involved determining the optimal shRNA transduction duration to observe significant knockdown of SOX10 target genes. Successful knockdown of SOX10-regulated genes was confirmed by reverse transcription qPCR (RT-qPCR) of known SOX10 target genes and immunoblot of SOX10 protein levels. These RNA samples were selected for library preparation, sequencing, and analysis. All optimization steps and

SOX10 knockdown experiments were performed in our laboratory, while the library preparation, sequencing, and initial data processing were performed by the Dana-Farber Cancer Institute

Molecular Biology Core Facility (DFCI MBCF). Downstream analysis of shSOX10 RNA-seq data and integration with SOX10 genomic localization data were performed in collaboration with the Dana-Farber Cancer Institute Center for Functional Cancer Epigenetics (DFCI CFCE).

In order to optimize the shRNA knockdown protocol to identify transcriptionally regulated genes of SOX10, we performed a series of experiments to evaluate the time-course and degree of knockdown. SKMEL19 melanoma cells were transduced with lentivirus stably expressing an shRNA targeting SOX10 (shSOX10-1 or shSOX10-5) or luciferase (shLUC), as previously discussed, and RNA lysates were collected over a range of infection periods: 24 h, 48 h, and 72 h. The resulting RNA samples were converted to complementary DNA (cDNA) by reverse transcription and quantified by qPCR for mRNA expression levels of SOX10, MITF, a known SOX10 target gene, PMEL and CDK2, known MITF target genes, and ACTB and

GAPDH, negative control housekeeping genes. shRNA-mediated knockdown of SOX10 was sufficient to reduce MITF mRNA levels within 48 h (Figure 2-11). Notably, decreases in PMEL mRNA levels did not occur until 72 h. Thus, using the 48 h time-point allowed for the discovery of SOX10 target genes without detecting downstream effects of transcriptional regulation. The

99

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

48 h time-point was confirmed in several cell lines before being applied to subsequent shSOX10

RNA-seq experiments.

Figure 2-11. SOX10 knockdown leads to reduction in MITF mRNA levels in 48 hours.

SKMEL19 cells were stably transduced with SOX10 shRNAs. RNA was isolated at 24 h, 48 h, and 72 h, then quantified by RT-qPCR for SOX10, MITF, PMEL, CDK2, ACTB, and GAPDH.

100

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

After optimization of the shRNA knockdown protocol, we performed shRNA-mediated knockdown of SOX10 in a panel of MITF-high and MITF-low melanoma cell lines and assayed the resulting RNA for decreases in expression levels for known SOX10 target genes by RT- qPCR and for reduction in SOX10 protein levels by immunoblot. Melanoma cell lines were transduced with lentivirus stably expressing an shRNA targeting SOX10 (shSOX10-1 or shSOX10-5) or luciferase (shLUC), and RNA and protein lysates were isolated at 48 h post- infection. The resulting RNA samples were converted to complementary DNA (cDNA) by reverse transcription and quantified by qPCR for mRNA expression of SOX10, MITF, PMEL and

CDK2 (MITF target genes), and ACTB and GAPDH (negative control housekeeping genes).

SOX10 knockdown in MITF-high melanoma cell lines led to reduced SOX10, MITF, PMEL, and

CDK2 mRNA expression levels, consistent with the model of SOX10-mediated regulation of

MITF expression (Figure 2-12). Additionally, SOX10 knockdown in MITF-high melanoma cell lines also resulted in reduced SOX10 protein levels (Figure 2-13).

101

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-12. shRNA-mediated knockdown of SOX10 leads to reduced mRNA expression of

SOX10, MITF, and MITF target genes PMEL and CDK2. Melanoma cells were stably transduced with SOX10 shRNAs. RNA was isolated at 48 h, then quantified by RT-qPCR for expression of SOX10, MITF, PMEL, CDK2, ACTB, and GAPDH.

Figure 2-13. shRNA-mediated knockdown of SOX10 leads to reduced protein levels of

SOX10. Melanoma cells were stably transduced with SOX10 shRNAs. Protein was isolated at

48 h, then quantified by immunoblot for SOX10 and beta-actin.

102

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

To comprehensively determine the complete set of genes transcriptionally regulated by

SOX10, we performed transcriptional profiling after shRNA-mediated knockdown of SOX10 in a panel of melanoma cell lines. SOX10 knockdown RNA samples were selected for RNA- sequencing based on their confirmed knockdown of SOX10 protein and of MITF mRNA expression. shSOX10 and shLUC RNA samples were processed, sequencing libraries were prepared, and multiplexed sequencing was performed by the Dana-Farber Cancer Institute

Molecular Biology Core Facility (DFCI MBCF). Raw sequencing reads were mapped to the genome and raw read counts were generated by STAR. Raw read counts were normalized to fragments per kilobase of transcript per million (FPKM) expression values by Cufflinks. Raw sequencing reads were further analyzed by VIPER. All samples exhibited appropriate read distribution across genomic features (less than 5% intronic reads), removal of rRNA sequences

(less than 1% rRNA reads), and coverage over the gene body (even distribution of mapped reads over gene body), indicative of successful sequencing library preparation. In addition, principal component analysis (PCA) and hierarchical clustering of shSOX10 and shLUC RNA-seq samples demonstrated that transcriptional profiles corresponding to shSOX10-1, shSOX10-5, and shLUC samples in the same cell line were more similar to each other than to other transcriptional profiles of their respective perturbation. This observation most likely reflects the limited changes in global transcriptional profiles by SOX10 knockdown. That is, SOX10 only regulates the expression of a restricted number of genes and its knockdown does not have transcriptional effects on the vast majority of genes in the genome. Finally, shRNA-mediated knockdown of SOX10 resulted in reduced mRNA expression levels of SOX10 and MITF in

MITF-high melanoma cell lines (Figure 2-14), confirming previous observations and the validity of our transcriptional profiling data.

103

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-14. Knockdown of SOX10 leads to reduced mRNA expression levels of SOX10 and

MITF in RNA-seq data. Melanoma cells were stably transduced with SOX10 shRNAs. RNA was isolated at 48 h, then prepared for RNA-sequencing. Raw sequencing reads were aligned using STAR and normalized FPKM values were generated using Cufflinks. LFC values were calculated between shSOX10 and shLUC samples for each cell line.

The SOX10 knockdown RNA-seq experiments characterized the complete set of gene transcripts with altered expression in shSOX10 and shLUC conditions. To determine the list of genes that are regulated by SOX10, we analyzed the SOX10 knockdown RNA-seq data using

GFOLD and identified genes whose expression is recurrently changed after SOX10 knockdown across melanoma cell lines. SOX10 knockdown RNA-seq data were analyzed by the Generalized

Fold Change (GFOLD) algorithm, which assigns a GFOLD score to each gene. GFOLD generates biologically meaningful rankings of differentially expressed genes, especially for experiments when only a single biological replicate is available. Each gene is assigned a GFOLD score, with positive scores indicating up-regulation and negative scores indicating down- regulation. To identify the core set of genes regulated by SOX10 in melanoma, we evaluated genes based on their GFOLD scores across our panel of profiled melanoma cell lines. A gene was considered regulated by SOX10 if the absolute value of its GFOLD score was greater than

0.5 in at least half of our SOX10 knockdown samples. Using this threshold, we generated a list

104

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma of 1,522 genes, many of which are linked to melanoma and cancer. This gene list included

SOX10, known SOX10 target genes (MITF and ERBB3), genes involved in melanocyte development (DCT, MLANA, and TYR), genes previously implicated in melanoma (BCL2A1,

CDK2, ETV1, ETV5, FOS, JUN, JUNB, NR4A1, and NR4A3), and genes with well-established links to cancer (AURKA, AURKB, BCL2, BRCA1, BRCA2, CDK1, CDK6, EZH2, MCL1, and

MYC). These findings point to the important role of SOX10 in maintaining cell fate transcriptional programs in normal melanocytes and also to the potential role of SOX10 in regulating a variety of oncogenic phenotypes, including cell cycle progression, apoptosis, and

DNA damage repair. Since these genes are only shown to be transcriptionally altered after

SOX10 knockdown, it is premature to call them SOX10 target genes. Instead, some of these genes may themselves be targets of SOX10 target genes, and the integration of binding and transcriptional data are required to delineate the hierarchy of gene regulation downstream of

SOX10.

To investigate the molecular pathways of genes regulated by SOX10, we analyzed the

SOX10 knockdown RNA-seq data to identify gene sets that are enriched in genes that are positively regulated by SOX10 (i.e., genes that exhibit higher expression in shLUC compared to shSOX10) using Gene Set Enrichment Analysis (GSEA). This analysis, which included RNA- seq data for 12 cell lines, was performed with the collection of curated gene sets in MSigDB.

GSEA determines the degree to which a gene set is overrepresented at the top or bottom of a ranked list of genes (i.e., the input dataset). GSEA results are ranked by normalized enrichment score (NES), which accounts for differences in gene set size and in correlations between gene sets and the input dataset, allowing for the comparison of GSEA results across gene sets. The top two gene sets ranked by NES were gene sets containing MYC targets that are up-regulated in

105

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma lymphoma cells expressing MYC (SCHUMACHER_MYC_TARGETS_UP) or positively correlated with MYC amplification in small cell lung cancer cell lines

(KIM_MYC_AMPLIFICATION_TARGETS_UP) (Table 2-3) (Figure 2-15). Two other gene sets in the top 20 gene sets were also associated with MYC targets

(KIM_MYCN_AMPLIFICATION_TARGETS_UP,

SCHLOSSER_MYC_TARGETS_AND_SERUM_RESPONSE_UP). This overlap of SOX10 regulated genes and MYC target genes suggests that SOX10 and MYC share transcriptional targets in melanoma and may regulate similar programs important for cancer phenotypes. Taken together with our previous findings of the significant overlap between genes bound by SOX10 and gene sets of MYC target genes (Table 2-2), these results point toward an overlapping and convergent role for these two transcription factors in melanoma.

106

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Table 2-3. Gene sets enriched in genes regulated by SOX10.

NAME SIZE NES NOM p-val FDR q-val SCHUHMACHER_MYC_TARGETS_UP 78 -1.9504278 0 0.047830332 KIM_MYC_AMPLIFICATION_TARGETS_UP 195 -1.884819 0 0.11882981 EPPERT_PROGENITOR 129 -1.8673564 0.002070393 0.113550834 BENPORATH_ES_1 368 -1.863339 0 0.09293893 KEGG_CYSTEINE_AND_METHIONINE_METABOLISM 34 -1.863037 0.00203666 0.07473156 PYEON_CANCER_HEAD_AND_NECK_VS_CERVICAL_UP 181 -1.8531091 0 0.07533491 ZHENG_GLIOBLASTOMA_PLASTICITY_UP 245 -1.8517749 0 0.06573913 DUTERTRE_ESTRADIOL_RESPONSE_6HR_UP 220 -1.8310931 0.002188184 0.0837505 NIKOLSKY_BREAST_CANCER_6P24_P22_AMPLICON 20 -1.829031 0 0.077495515 HU_ANGIOGENESIS_DN 37 -1.8268988 0 0.07192532 LIANG_HEMATOPOIESIS_STEM_CELL_NUMBER_LARGE_VS_TINY_UP 43 -1.8219231 0 0.07208674 FERREIRA_EWINGS_SARCOMA_UNSTABLE_VS_STABLE_UP 158 -1.8209684 0.002070393 0.06773892 XU_RESPONSE_TO_TRETINOIN_AND_NSC682994_DN 15 -1.8174652 0 0.06646357 BROWN_MYELOID_CELL_DEVELOPMENT_DN 124 -1.8148905 0 0.064237304 DAIRKEE_CANCER_PRONE_RESPONSE_BPA 51 -1.8137465 0 0.061513882 KEGG_PROTEIN_EXPORT 23 -1.8131454 0 0.058302797 NAKAMURA_CANCER_MICROENVIRONMENT_DN 44 -1.8086209 0.002132196 0.059975438 UDAYAKUMAR_MED1_TARGETS_UP 132 -1.8053193 0.004301075 0.05985047 KIM_MYCN_AMPLIFICATION_TARGETS_UP 90 -1.7997993 0 0.06335095 SCHLOSSER_MYC_TARGETS_AND_SERUM_RESPONSE_UP 46 -1.779975 0 0.086467944

Figure 2-15. MYC target gene sets are enriched for genes regulated by SOX10. SOX10 knockdown RNA-seq data were analyzed by GSEA to determined gene sets enriched in SOX10- regulated genes.

107

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

In addition, three other top scoring gene sets enriched in SOX10 regulated genes include genes up-regulated in hematopoietic lineage committed progenitor cells vs. mature cells

(EPPERT_PROGENITOR), genes overexpressed in embryonic stem cells

(BENPORATH_ES_1), and genes up-regulated in neural stem cells with double knockout of

TP53 and PTEN and associated with an undifferentiated state with high renewal potential

(ZHENG_GLIOBLASTOMA_PLASTICITY_UP) (Table 2-3). All of these gene sets are related to a stem/progenitor cell-like state that is characterized by cellular plasticity and the acquisition of alternate cellular states, which is reminiscent to the role that SOX10 plays in maintaining the multipotency of neural crest cells and their subsequent fate specification during neural crest development. In addition, the latter gene set (ZHENG_GLIOBLASTOMA_PLASTICITY_UP) is linked to increased MYC protein levels and its associated signature, and serves to highlight the connection between SOX10 and MYC and their role in the differentiation and renewal of neural crest stem cells (Zheng et al., 2008).

Elucidation of SOX10 Target Genes and Melanoma Dependencies

Our characterization of the SOX10 cistrome and transcriptome in melanoma yielded lists of genes and gene sets with potential significance in melanomagenesis. The examination of individual genes revealed that genes bound or regulated by SOX10 play major roles in normal melanocyte development as well as common oncogenic phenotypes and pathways. Additionally, the analysis of significantly enriched gene sets added further evidence to the model that SOX10 target genes are involved in major oncogenic programs, including those shared by MYC.

108

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

To further investigate the overlap in SOX10 and MYC target genes in melanoma, we analyzed the gene set overlap and enrichment results from the bound and regulated genes analyses. First, we intersected the set of genes bound by SOX10 with the genes in the MYC bound gene sets to generate a list of 79 genes bound by SOX10 and MYC. Next, we identified the genes in the leading edge (i.e., the core genes that account for the gene set’s enrichment signal) of the MYC target gene sets from GSEA, resulting in a list of 144 genes regulated by

SOX10 and MYC. Then, the overlap of these two lists revealed a small set of 4 genes that are bound and regulated by SOX10 and MYC in melanoma: HSPD1, HSPE1, RCC1, and SRM

(Figure 2-16). Of these genes, HSPE1 exhibited the greatest correlation with SOX10 with respect to expression and dependency in melanoma cell lines (Figure 2-17). HSPD1 and HSPE1 encode heat shock proteins with demonstrated roles in tumor promotion. Our SOX10 ChIP-seq and SOX10 knockdown RNA-seq data confirm that SOX10 binds to these gene loci and regulate their expression in melanoma cell lines (Figure 2-18, Figure 2-19). As important MYC target genes and with established functions as heat shock proteins, HSPD1 and HSPE1 may represent major downstream effectors of SOX10 dependency in melanoma. Taken together, these results put forth a model of shared SOX10 and MYC target genes and the potential for SOX10 to coordinate similar oncogenic transcriptional programs as MYC.

109

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-16. Identification of genes bound and regulated by SOX10 and MYC in melanoma.

Figure 2-17. HSPE1 expression and dependency is correlated with that of SOX10 in melanoma.

110

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-18. SOX10 binds to the MYC, HSPD1, and HSPE1 gene loci.

Figure 2-19. SOX10 knockdown leads to reduced expression of MYC, HSPD1, and HSPE1.

111

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

The separate characterization of SOX10 genomic localization sites and transcriptionally regulated genes after SOX10 knockdown generated the set of SOX10 bound genes and SOX10 regulated genes in melanoma. While these datasets are crucial to our understanding of SOX10 in melanoma biology and development, the integration of binding and expression data would produce the most significant and impact set of SOX10 target genes. To determine the set of true

SOX10 target genes (i.e., genes bound and regulated by SOX10), we intersected the list of genes bound by SOX10 and the list of genes regulated by SOX10. For the set of SOX10 bound genes, we used a threshold of BETA > 0.6 in at least 6 of 10 SOX10 ChIP-profiled cell lines, which yielded 1,897 genes. For the set of SOX10 regulated genes, we used a threshold of |GFOLD| >

0.5 in at least half of SOX10 knockdown RNA-seq samples, which yielded 1,522 genes. The overlap of these gene lists revealed three biologically relevant sets of SOX10-associated genes:

1) SOX10 target genes (bound and regulated), including DCT, ERBB3, ETV1, ETV5, EZH2,

MITF, and SOX10; 2) SOX10 bound-only genes, including ARID1A, ARID2, FOSB, FOSL1, and

JUND; and 3) SOX10 regulated-only genes, including AURKA, AURKB, BCL2, BCL2A1,

BRCA1, BRCA2, CDK1, CDK2, CDK6, FOS, FOSL2, JUN, MLANA, MYC, and TYR. The

SOX10 target genes include known SOX10 target genes, including SOX10, MITF, and ERBB3.

However, our analysis nominates many putative SOX10 target genes, including DCT, which is expressed in normal melanocytes, and EZH2, whose gene product is involved in epigenetic regulation in cancer. The SOX10 bound-only genes include two additional noteworthy genes involved in epigenetics, ARID1A and ARID2, the latter of which has been observed to be significantly mutated in human melanoma. While these two genes may not be transcriptionally regulated by SOX10, the binding of SOX10 to these genes and EZH2 implicate SOX10 in the regulation of the melanoma epigenome. Finally, the SOX10 regulated-only genes include many

112

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma well-known oncogenes, including AURKA, AURKB, BCL2, CDK1, CDK6, and MYC, and two genes with major roles in DNA damage repair, BRCA1 and BRCA2. While SOX10 may not bind directly to these genes in melanoma cells, these genes are likely downstream of a SOX10 target gene. For example, the SOX10 regulated-only genes also include CDK2, MLANA, and TYR, which are known targets of MITF, a SOX10 target gene. Thus, SOX10 regulated-only genes likely represent a set of genes that are two transcriptional levels below SOX10, but may have important functional associations with SOX10 dependency in melanoma.

To identify SOX10 target genes that are also genetic dependencies in melanoma, we determined the overlap of genes that are bound by SOX10 (SOX10 ChIP-seq), are transcriptionally regulated by SOX10 (shSOX10 RNA-seq), and are genetic dependencies in melanoma (Project Achilles RNAi or CRISPR-Cas9). We generated two sets of melanoma genetic dependencies using two different functional genomic datasets from Project Achilles:

DEMETER dependency scores from RNAi screens (Tsherniak et al., 2017) and CERES dependency scores from CRISPR-Cas9 screens (Meyers et al., 2017). As previously discussed,

DEMETER segregates on- and off-target effects of RNAi reagents, such as shRNA. Genome targeting by CRISPR-Cas9 reagents elicits a gene-independent anti-proliferative response due to an increased number of target loci and resulting DNA breaks (Aguirre et al., 2016). CERES is a computational method that estimates gene-level dependency from CRISPR-Cas9 essentiality screens while accounting for the off-target effect of copy number on proliferation (Tsherniak et al., 2017). For the overlap, we used identical thresholds for genes bound by SOX10 (BETA > 0.6 in at least 6 of 10 cell lines) and regulated by SOX10 (|GFOLD| > 0.5 in at least half of shSOX10 samples) as before and a threshold of DEMETER < -0.5 (1,000 genes) or CERES < -0.5 (1,599 genes) in at least half of melanoma cell lines profiled in the respective Project Achilles screens.

113

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

The intersection of these gene sets revealed a short list of genes that are SOX10 targets and genetic dependencies in melanoma, including SOX10, NR4A1, HSPD1, HSPE1, MITF, and

RCC1 (Figure 2.20). NR4A1 encodes an orphan that acts as a transcription factor. Genome-wide drug resistance screens have identified NR4A1 as a gene that can confer resistance to MAPK pathway inhibitors (including BRAF and MEK inhibitors) in melanoma cells (Johannessen et al., 2013). HSPD1, HSPE1, and RCC1 were identified as shared bound genes of SOX10 and MYC, further underscoring the importance of these shared genes and their molecular functions in melanomagenesis. This analysis highlights the convergence of genomic localization data, transcriptional regulation data, and gene essentiality data to reveal biological insights into the SOX10 transcriptional program and SOX10 dependency in melanoma.

Figure 2-20. Identification of SOX10 target genes that are also genetic dependencies in melanoma. Genetic dependencies were determined by DEMETER scores (RNAi shRNA) or

CERES scores (CRISPR-Cas9).

114

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Analysis of SOX10 Target Genes in Melanoma by MITF Status

Studies of melanoma tumors and cell lines have described two major transcriptional states in melanoma: 1) MITF-high melanoma – melanomas that express high levels of MITF, have a more proliferative phenotype, and are sensitive to MAPK pathway inhibitors; and 2)

MITF-low melanoma – melanomas that express low levels of MITF (and high levels of AXL), have a more invasive phenotype, and are resistant to MAPK pathway inhibitors (Konieczkowski et al., 2014; Muller et al., 2014; Wellbrock & Arozarena, 2015). According to our analysis of

Project Achilles dependency data, SOX10 is a genetic dependency in all melanoma cell lines that express SOX10, regardless of MITF expression level. Thus, our SOX10 gene expression- dependency model suggests that SOX10 is active in the melanoma cell and binds to its target genes in MITF-high and MITF-low melanoma to impart a dependency in all melanoma cells that express SOX10. MITF is a known target gene of SOX10 and MITF is an established lineage oncogene important for melanoma cell survival (Lee et al., 2000; Potterf et al., 2000; Verastegui et al., 2000). However, it remains unclear which SOX10 target genes contribute to SOX10 dependency in MITF-low melanoma and whether other genes may play vital roles in addition to

MITF in MITF-high melanoma.

To investigate differential patterns of SOX10 localization in melanoma cells by MITF expression, we performed supervised analyses of our SOX10 ChIP-seq data by MITF status in collaboration with the Dana-Farber Cancer Institute Center for Functional Cancer Epigenetics

(DFCI CFCE). DESeq, which infers differential signals in data with two classes based on the negative binomial distribution, was used to cluster SOX10 peaks by enrichment in MITF-high or

MITF-low cell lines as classified by their baseline MITF expression (Figure 2-21). There were

676 SOX10 peaks differentially enriched in MITF-high melanoma cell lines and 136 SOX10

115

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma peaks differentially enriched in MITF-low melanoma cell lines. These sets of peaks corresponded to 562 genes for MITF-high melanoma and 127 genes for MITF-low melanoma.

The genes associated with SOX10 peaks in MITF-high melanoma included many genes involved in melanocyte development and growth, such as MITF, DCT, TYR, PMEL, and MLANA (Figure

2-22). To explore molecular pathways linked to these genes, we queried these genes for significant overlap with gene sets in MSigDB. Gene sets with significant overlap with differentially SOX10 bound genes in MITF-high melanoma included gene sets associated with oligodendrocyte differentiation, Alzheimer’s disease, , and melanomagenesis

(Table 2-4). It is encouraging that this analysis identifies gene sets associated with melanoma and uveal melanoma, as MITF-high melanoma are characteristic of human melanoma.

Oligodendrocytes, like melanocytes, are derived from the neural crest and SOX10 is expressed throughout the development of glial cells, including oligodendrocytes in the central nervous system and Schwann cells in the peripheral nervous system (Kuhlbrodt et al., 1998). In MITF- high melanoma, SOX10 may be down-regulating genes associated with oligodendrocyte differentiation to maintain a multipotent, neural crest cell-like state in melanoma. Meanwhile gene sets with significant overlap with differentially SOX10 bound genes in MITF-low melanoma included gene sets associated with breast cancer, estrogen receptor targets, and endocrine therapy resistance, and targets of SMAD2 or SMAD3 (Table 2-5). In addition to melanoma, SOX10 expression in cancer has also been observed in glioblastoma and breast cancer cell lines (Figure 1-3). Histopathological studies have demonstrated SOX10 expression by immunohistochemistry in 12-40% of breast carcinomas, with positive labeling in up to 66% of basal-like, triple-negative breast cancer (TNBC) (Cimino-Mathews et al., 2013; Ivanov et al.,

2013). These significant gene set overlaps point to known facets of SOX10 biology unrelated to

116

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma melanomagenesis and may suggest shared SOX10 transcriptional programs across development and lineages. Additionally, the enrichment of SMAD2 or SMAD3 targets in MITF-low melanoma may point to the previously mentioned relationship among SMAD2/3, ETS1,

TFAP2A, and SOX10 in neural crest development, and highlights the potential shared targets between SMAD2/3 and SOX10.

117

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-21. Heatmap of differentially enriched SOX10 peaks in MITF-high and MITF-low melanoma cell lines. SOX10 ChIP-seq peaks were clustered by cell line according to MITF status using DESeq.

Figure 2-22. Overlap of genes associated with differentially enriched SOX10 peaks in

MITF-high and MITF-low melanoma cell lines.

118

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Table 2-4. Gene sets with significant overlap with differentially SOX10 bound genes in

MITF-high melanoma.

# Genes in # Genes in Gene Set Name Gene Set (K) Overlap (k) k/K p-value FDR q-value GOBERT_OLIGODENDROCYTE_DIFFERENTIATION_DN 1080 61 0.0565 2.69E-26 1.27E-22 DACOSTA_UV_RESPONSE_VIA_ERCC3_DN 855 54 0.0632 1.24E-25 2.94E-22 MEISSNER_BRAIN_HCP_WITH_H3K4ME3_AND_H3K27ME3 1069 52 0.0486 1.16E-19 1.83E-16 BENPORATH_SUZ12_TARGETS 1038 44 0.0424 1.18E-14 1.40E-11 DACOSTA_UV_RESPONSE_VIA_ERCC3_COMMON_DN 483 30 0.0621 1.58E-14 1.49E-11 BENPORATH_ES_WITH_H3K27ME3 1118 45 0.0403 3.63E-14 2.86E-11 GRAESSMANN_APOPTOSIS_BY_DOXORUBICIN_DN 1781 58 0.0326 5.75E-14 3.89E-11 ONKEN_UVEAL_MELANOMA_UP 783 36 0.046 3.23E-13 1.91E-10 DODD_NASOPHARYNGEAL_CARCINOMA_UP 1821 57 0.0313 4.87E-13 2.56E-10 RODRIGUES_THYROID_CARCINOMA_ANAPLASTIC_DN 537 29 0.054 1.43E-12 6.76E-10 REACTOME_DEVELOPMENTAL_BIOLOGY 396 25 0.0631 1.77E-12 7.60E-10 BLALOCK_ALZHEIMERS_DISEASE_DN 1237 44 0.0356 4.49E-12 1.77E-09 WANG_MLL_TARGETS 289 21 0.0727 7.75E-12 2.82E-09 SCHAEFFER_PROSTATE_DEVELOPMENT_48HR_DN 428 25 0.0584 9.57E-12 3.23E-09 ONKEN_UVEAL_MELANOMA_DN 526 27 0.0513 2.71E-11 8.54E-09 CHEN_METABOLIC_SYNDROM_NETWORK 1210 42 0.0347 3.02E-11 8.94E-09 BLALOCK_ALZHEIMERS_DISEASE_UP 1691 51 0.0302 3.44E-11 9.57E-09 PEREZ_TP53_TARGETS 1174 41 0.0349 4.37E-11 1.15E-08 DAIRKEE_TERT_TARGETS_DN 124 14 0.1129 7.43E-11 1.79E-08 KEGG_MELANOGENESIS 102 13 0.1275 7.57E-11 1.79E-08

119

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Table 2-5. Gene sets with significant overlap with differentially SOX10 bound genes in

MITF-low melanoma.

# Genes in # Genes in Gene Set Name Gene Set (K) Overlap (k) k/K p-value FDR q-value KOINUMA_TARGETS_OF_SMAD2_OR_SMAD3 824 15 0.0182 9.18E-10 4.34E-06 CHARAFE_BREAST_CANCER_LUMINAL_VS_MESENCHYMAL_DN 460 11 0.0239 1.17E-08 2.78E-05 MEISSNER_BRAIN_HCP_WITH_H3K4ME3_AND_H3K27ME3 1069 15 0.014 2.95E-08 4.65E-05 CHARAFE_BREAST_CANCER_LUMINAL_VS_BASAL_DN 455 10 0.022 1.22E-07 1.45E-04 PLASARI_TGFB1_SIGNALING_VIA_NFIC_1HR_DN 106 6 0.0566 1.95E-07 1.85E-04 GOZGIT_ESR1_TARGETS_DN 781 12 0.0154 3.03E-07 2.39E-04 DURAND_STROMA_S_UP 297 8 0.0269 5.19E-07 3.09E-04 RUTELLA_RESPONSE_TO_HGF_VS_CSF2RB_AND_IL4_UP 408 9 0.0221 5.23E-07 3.09E-04 GRAESSMANN_APOPTOSIS_BY_DOXORUBICIN_DN 1781 17 0.0095 7.92E-07 4.16E-04 NUYTTEN_EZH2_TARGETS_UP 1037 13 0.0125 9.35E-07 4.42E-04 CREIGHTON_ENDOCRINE_THERAPY_RESISTANCE_3 720 11 0.0153 1.03E-06 4.42E-04 BROWNE_HCMV_INFECTION_24HR_DN 148 6 0.0405 1.39E-06 5.49E-04 ZHENG_BOUND_BY_FOXP3 491 9 0.0183 2.40E-06 8.20E-04 DAVICIONI_TARGETS_OF_PAX_FOXO1_FUSIONS_UP 255 7 0.0275 2.43E-06 8.20E-04 BERTUCCI_MEDULLARY_VS_DUCTAL_BREAST_CANCER_DN 169 6 0.0355 3.01E-06 9.49E-04 BUYTAERT_PHOTODYNAMIC_THERAPY_STRESS_UP 811 11 0.0136 3.22E-06 9.52E-04 REACTOME_GLYCOSAMINOGLYCAN_METABOLISM 111 5 0.045 6.56E-06 1.77E-03 BIOCARTA_GATA3_PATHWAY 16 3 0.1875 6.72E-06 1.77E-03 THUM_SYSTOLIC_HEART_FAILURE_UP 423 8 0.0189 7.10E-06 1.77E-03 CHICAS_RB1_TARGETS_CONFLUENT 567 9 0.0159 7.64E-06 1.81E-03

120

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

To investigate whether specific transcriptional programs are enriched in these sets of differential SOX10 peaks by MITF status, we performed motif analysis in collaboration with the

DFCI CFCE. Motifs enriched in the MITF-high SOX10 peaks included SOX (SOX2, SOX3,

SOX4, SOX9, SOX10, and SOX15), MITF, and other basic helix-loop-helix (bHLH) transcription factors (USF2, PIF5, and SPCH) (Figure 2-23). This finding is expected due to the high expression of MITF in these cell lines and the relatedness in binding motif consensus among bHLH transcription factors. Conversely, MITF and bHLH binding motifs were absent in motifs enriched in MITF-low SOX10 peaks, which included SOX (SOX2, SOX3, SOX4, and

SOX10), FOS-JUN family members (Fra1 (FOSL1), FOSL2, and JUN/AP-1), and other basic leucine zipper (bZIP) transcription factors (BATF and ATF3) (Figure 2-23). FOS-JUN family members are important for the propagation of MAPK pathway signaling in melanoma cells. A recent study characterizing proliferative (MITF-high) and invasive (MITF-low) melanoma cells has revealed that the genomic regulatory regions underlying these phenotypic states are regulated by MITF and AP-1, respectively (Verfaillie et al., 2015). The enrichment of these motifs in

SOX10 peaks in their corresponding melanoma transcriptional classes indicated that SOX10 may play a role in maintaining an oncogenic transcriptional program in melanoma regardless of phenotypic state defined by MITF and AP-1.

121

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Figure 2-23. Significantly enriched motifs in differentially enriched SOX10 peaks in MITF- high (left) and MITF-low (right) melanoma. SOX10 ChIP peaks were clustered across cell lines by MITF expression level (MITF-high vs. MITF-low). Motif analysis was performed on differentially enriched peaks in both classes to identify significantly enriched transcription factor binding motifs.

122

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

DISCUSSION

In this study, we have characterized the complete cistrome and transcriptome of SOX10 in melanoma, including generating lists of genes bound by and/or regulated by SOX10. SOX10 target genes include genes involved in melanocyte differentiation (e.g., MITF and DCT) and oncogenic phenotypes (e.g., ETV1, ETV5, and EZH2). Additionally, our gene set analyses of

SOX10-bound and SOX10-regulated genes have uncovered a convergence on MYC target genes, indicating that SOX10 and MYC regulate common target genes and, likely, downstream molecular and cellular pathways. Finally, differential analysis of SOX10-bound genes in MITF- high and MITF-low melanoma has revealed distinct transcriptional programs in these two classes of melanoma that are likely governed by different transcription factors.

Our genome-wide profiling of SOX10 localization sites in melanoma cell lines is the largest study of its kind performed to date. Fufa et al. have previously characterized SOX10 genomic localization in mouse melanocytes and described how SOX10 predominantly engages open chromatin regions and binds to distal regulatory elements, including melanocyte enhancers

(Fufa et al., 2015). In addition to generating a comprehensive catalogue of SOX10 binding sites in the genome of human melanoma cell lines, our utilization of endogenously expressed SOX10 protein in our ChIP experiments enable natural interpretation of our data as we did not alter the cell models in any way that would have modulated SOX10 expression or activity (e.g., ectopic expression and/or use of protein tag). While SOX10 protein levels were high enough in these cell lines to successfully perform SOX10 ChIP and identify SOX10 binding sites, genomic loci bound less frequently or less tightly by SOX10 may have been missed in our experiments. Future experiments could optimize and use more initial chromatin to increase confidence for SOX10 bound loci. As an example, we had included two normal melanocyte cell lines in our SOX10

123

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

ChIP studies. While these experiments yielded SOX10 peaks in known SOX10 target genes

(e.g., MITF), the low levels of input chromatin prevented the identification of other SOX10 peaks in the genome. Characterization of the SOX10 cistrome in normal melanocytes and cells throughout neural crest development would allow for the comparison of SOX10 binding sites and patterns throughout development and oncogenic transformation, leading to the potential determination of cancer-specific SOX10 target genes and their relationship with developmental processes (e.g., multipotency, fate specification, etc.).

In our SOX10 ChIP-qPCR optimization experiments, we used ACTB and GAPDH as negative control loci for SOX10 binding. However, we consistently observed minor enrichment of SOX10 at these seemingly neutral genomic loci, suggesting that SOX10 binds to these

“housekeeping” genes at low levels. Meanwhile, our transcriptional profiling data indicate that

SOX10 likely does not regulate expression of ACTB and GAPDH as their expression levels do not change upon knockdown of SOX10. Thus, this observation may reflect true binding of

SOX10 to these and potentially other “housekeeping” genes, but additional transcription factors or regulatory proteins may have more dominant roles in the regulation of these genes in cells.

Our motif analysis of SOX10 peaks across melanoma cell lines identified a previously unidentified inverted repeat motif enriched in SOX10 peaks. This binding motif resembles two

SOX family consensus sequences arranged in a palindromic manner and suggests that the binding of SOX10 to DNA may occur as a monomer and as a dimer. Previous biochemical studies have demonstrated the binding of SOXE family members to DNA as dimers in cell-free in vitro settings and computational studies have described palindromic dimer sequences for SOX family members in the genome (Peirano & Wegner, 2000; Schlierf et al., 2002; Huang et al.,

2015). However, no study has shown the direct binding of SOX10 as a dimer to the melanoma

124

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma genome. Additionally, it is not known whether dimer binding to DNA by SOX10 occurs through homodimerization with other SOX10 proteins or through heterodimerization with other SOX family members. Additionally, differential SOX10 binding modalities (i.e., dimer vs. monomer) may have significant consequences on target gene expression. Further studies are required to elucidate the potential regulation of SOX10 transcriptional activity by monomer or dimer binding to DNA.

Gene set analyses of SOX10-bound and SOX10-regulated genes separately identified

MYC target genes as enriched in these gene lists. MYC is a well-established oncogenic transcription factor with demonstrated roles in cell cycle progression and cellular transformation.

The observation that SOX10 and MYC share a subset of target genes indicates that part of

SOX10 dependency in melanoma may be explained by the regulation of MYC- and cancer- associated genes by SOX10. Future experiments must be performed to determine whether

SOX10 and MYC are direct transcriptional partners at these target genes. Nevertheless, their binding to genomic loci may not be dependent on one another and may allow for multiple layers of transcriptional regulation in melanoma cells. Additionally, these observations put forward a model of tumor dependencies where lineage-specific genetic dependencies, especially those involving master transcriptional regulators (e.g., SOX10, FOXA1, PAX8), may share target genes and downstream pathways with common oncogenic proteins and transcription factors (e.g.,

MYC). Additional study and analysis of these relationships can shed light on potential interaction and inter-relatedness of oncogenic pathways in human cancer.

The generation of both binding and transcriptional data for SOX10 allowed for the investigation of SOX10-associated genes by broad classes: bound-only, regulated-only, bound and regulated (i.e., true direct target genes). Notably, the list of direct target genes includes well-

125

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma known oncogenes, including ETV1, ETV5, and EZH2. Studies in our laboratory and others have demonstrated a role for ETV proteins, namely ETV1, ETV4, and ETV5, in cancer, including melanoma and prostate cancer (Jane-Valbuena et al., 2010; Pop et al., 2014). These transcription factors promote many oncogenic phenotypes, including cell proliferation, motility, and invasion.

In addition, EZH2, the catalytic component of the polycomb-repressive complex 2 (PRC2), is required for proliferation and invasion of melanoma cell lines (Zingg et al., 2015). The role of

EZH2 and PRC2 in modifying and altering the chromatin landscape of melanoma genomes may also contribute to SOX10-associated cancer phenotypes in melanoma. These and other oncogenic targets of SOX10 may be working independently or in concert to promote oncogenesis in melanoma cells.

While the list of direct SOX10 target genes may be of most immediate interest, the

SOX10 bound-only and regulated-only gene lists also provided new insights into melanoma biology. The gene products of two SOX10 bound-only genes, ARID1A and ARID2, have established roles in the epigenetic regulation of gene expression, primarily through their participation in the BAF and PBAF SWI/SNF complexes, respectively (Kadoch & Crabtree,

2015; Hodges et al., 2016). These large multi-subunit complexes exert control in gene regulation, cell lineage specification, and organismal development through their chromatin remodeling ability. Additionally, sequencing studies have identified nonsense mutations in ARID2, ARID1A, and ARID1B in human melanoma (Hodis et al., 2012; Cancer Genome Atlas Network, 2015).

Thus, while ARID1A and ARID2 did not experience expression changes after SOX10 knockdown, these observations, together with the newly discovered transcriptional regulatory relationship between SOX10 and EZH2, support a model of SOX10 involvement in the epigenetic regulation of melanoma.

126

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Additionally, many SOX10-regulated genes have clear associations with cancer, including well-known oncogenes AURKA, AURKB, BCL2, CDK1, CDK6, and MYC, and two genes involved in DNA damage repair, BRCA1 and BRCA2. The protein products of these oncogenes have established roles in cellular processes relevant to oncogenesis, including the segregation of in mitosis (AURKA and AURKB), the inhibition of apoptosis

(BCL2), and cell cycle progression (CDK1, CDK6, and MYC). While these genes may not be directly bound by SOX10, they may lie downstream of SOX10 and may be coordinately regulated in melanoma. Interestingly, our experiments were sensitive enough to identify target genes of SOX10 target genes. For example, a few MITF target genes, including CDK2, MLANA, and TYR, were considered regulated-only (i.e., regulated but not bound) by SOX10. Meanwhile,

DCT, another MITF target gene, was classified as a SOX10 target gene, confirming earlier studies demonstrating SOX10 regulation of DCT expression in model systems (i.e., bound and regulated). Thus, our cistrome and transcriptome datasets can reveal additional insights into the regulatory gene networks present in melanoma.

Our integrative analyses of SOX10 target genes and melanoma genetic dependencies from RNAi- and CRISPR-based screens identified lists of genes that were largely non- overlapping. This discrepancy may be attributed to the experimental and computational differences between these types of functional genomic screens. RNAi reagents, such as shRNAs, knockdown gene expression with variable efficiency depending on the degree of sequence similarity to the target sequence, while CRISPR reagents more reliably knockout gene function by introducing indels into genomic target loci that permanently alter protein expression. In addition, the computational approaches for processing and analyzing RNAi and CRISPR essentiality screens (DEMETER and CERES, respectively) differ with regards to the off-target

127

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma effects that are computationally removed and the scale of dependencies that are inferred. For example, DEMETER generates relative dependency scores across cell lines, while CERES provides absolute dependency scores using defined essential genes, such as housekeeping genes.

Thus, the list of melanoma genetic dependencies identified using one method will likely differ from the other method. Most significantly, CRISPR screens may be susceptible to nominating genes that are required for cell viability only when they are completely knocked out, while RNAi screens may provide a lower threshold for gene knockdown to identify genes essential for cell viability. On the other hand, DEMETER may identify genes that are only relatively essential in a class of cell lines compared to another class, but the genes may not reflect actual molecular dependencies in cancer cells. The use of these and other types of experimental and computational approaches will allow for a more thorough understanding of SOX10 dependency in melanoma.

The melanoma cell lines included in our SOX10 cistrome and transcriptome profiling studies were intentionally chosen to be representative of the two major transcriptional states in human melanoma: MITF-high and MITF-low melanoma. Aside from MITF expression, these melanoma classes are also characterized by phenotypic and drug sensitivity differences: MITF- high melanoma possess a proliferate phenotype and are sensitive to MAPK pathway inhibitors, while MITF-low melanoma have an invasive phenotype and are resistant to MAPK pathway inhibitors (Konieczkowski et al., 2014; Muller et al., 2014; Wellbrock & Arozarena, 2015). Our differential analysis of SOX10 peaks in MITF-high and MITF-low melanoma revealed differentially enriched peaks and corresponding differences in transcription factor binding motifs present in those class-specific peaks. As expected, SOX10 peaks in MITF-high melanoma were enriched for MITF and other bHLH transcription factor binding motifs. Additionally, the genes associated with these peaks were enriched for melanoma and related gene sets, including

128

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma oligodendrocyte differentiation. As previously mentioned, SOX10 has well-established roles in neural crest development, including the differentiation of glial cells (i.e., oligodendrocytes and

Schwann cells). On the other hand, SOX10 peaks in MITF-low melanoma were enriched for

FOS-JUN family members and other bZIP transcription factor binding motifs. The associated genes were enriched for breast cancer and related gene sets, including endocrine therapy resistance. Histopathological studies have demonstrated the expression of SOX10 in breast cancer, especially triple-negative breast cancer (Cimino-Mathews et al., 2013; Ivanov et al.,

2013). Thus, in addition to describing distinct SOX10 transcriptional programs in MITF-high and MITF-low melanoma, our findings point to a model of SOX10 transcriptional regulation characterized by overlaps between tissue types (i.e., melanoma vs. breast cancer). Differences in

SOX10 binding in MITF-high and MITF-low melanoma may be attributed to different partner factors for SOX10 in these two contexts. Additional experiments to characterize SOX10 binding partners in melanoma cell lines will further elucidate this model of differential SOX10 regulation in melanoma.

129

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

METHODS

Cell lines and reagents

All cell lines were obtained from and were identity confirmed with DNA fingerprinting by the

Broad Institute Genomics Platform. SKMEL19, COLO679, WM983B, UACC62, RVH421, and

WM793 were grown in RPMI medium, 10% FBS. G361, WM88, WM2664, A2058, SKMEL24,

HS294T, IGR39, and LOXIMVI were grown in DMEM medium, 10% FBS. Antibodies for immunoblot were obtained from Abcam (SOX10 rabbit monoclonal, ab155279 [EPR4007]),

Santa Cruz (SOX10 mouse monoclonal, sc-271163; SOX10 goat polyclonal, sc-17342), and Cell

Signaling (β-actin mouse monoclonal, #3700 8H10D10).

Chromatin immunoprecipitation (ChIP)

Antibodies for ChIP were obtained from Abcam (SOX10 rabbit monoclonal, ab155279

[EPR4007]) and Santa Cruz (rabbit IgG, sc-2027). Cells were grown in 10-cm tissue culture dishes until near confluency (approximately 10 million cells per SOX10 ChIP). Growth medium was removed and replaced with 1% formaldehyde PBS solution for 10 min at room temperature.

Cells were washed with cold PBS, then removed from tissue culture dish with cell lifter in 500

μL cold PBS supplemented with protease inhibitor. Cells were centrifuged at 4 °C to remove

PBS and resuspended in 350 μL lysis buffer (1% SDS, 10 mM EDTA, 50 mM Tris-HCl pH 8.1) supplemented with protease inhibitor. Cells were sonicated in a water bath sonicator at high amplitude (30 sec ON, 30 sec OFF) for 15 min to generate chromatin fragments of 200-500 bp.

Samples were centrifuged at 8 °C and supernatants were transferred to new tubes. To prepare

130

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma magnetic beads, 10 μL Dynabeads A (Invitrogen 10002D) and 10 μL Dynabeads G (Invitrogen

10004D) were washed with 500 μL cold PBS supplemented with 5 mg/mL BSA and diluted with

300 μL cold PBS supplemented with 5 mg/mL BSA. SOX10 antibody (2 μg per ChIP) was added to beads and rotated at 4 °C for 6 h. For ChIP, 320 μL supernatant and 110 μL beads (in cold dilution buffer) were added to 1600 μL dilution buffer (1% Triton, 2 mM EDTA, 150 mM

NaCl, 20 mM Tris-HCl pH 8.1) and rotated at 4 °C for 16 h. After overnight incubation, beads were washed with cold washing RIPA buffer (50 mM HEPES pH 7.6, 1 mM EDTA, 0.7% sodium deoxycholate, 1% NP40, 0.5 M LiCl) and with cold TE buffer. Beads and input samples were diluted in 100 μL decrosslinking buffer (1% SDS, 0.1 M NaHCO3). Samples were incubated at 65 °C for 6 h, then stored at 4 °C. The resulting ChIP-enriched was used in subsequent experiments (ChIP-qPCR and ChIP-seq).

To determine sonication efficiency: After sonication and centrifugation, supernatant samples were decrosslinked at 65 °C in a temperature block for 16 h. DNA was extracted by MinElute

PCR Purification Kit and run on 1% agarose gel.

ChIP-quantitative PCR (ChIP-qPCR)

ChIP-enriched DNA fragments were evaluated for enrichment at genomic loci by quantitative

PCR using Power SYBR Green Master Mix from Invitrogen per the manufacturer’s instructions.

Genomic loci, including known SOX10 binding sites (MITF, ERBB3, and MIA), housekeeping genes (ACTB and GAPDH), and negative control regions (RNA28S5, RHO, and CHEK2), were assayed using the following qPCR primers:

131

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

Genomic Locus Forward Primer (5’-3’) Reverse Primer (5’-3’)

MITF (promoter) CAAAGGGGCATTCTGCTATT TCAGATCAAGGCCAATTCAC

ERBB3 (intron) CCATCCCACCCTCAGTAGAC CTACCCTCATCCTGCCTCTC

MIA (promoter) TGGGCTGTTTCTGGTAATCA CACCTTGGAATTTCCTGTGC

ACTB (intron) ACGCCTCCGACCAGTGTT GCCCAGATTGGGGACAAA

GAPDH (exon) GCTCTCTGCTCCTCCTGTTC TAGCCTCCCGGGTTTCTC

RNA28S5 (exon) CTGGGTATAGGGGCGAAAGAC GGCCCCAAGACCTCTAATCAT

RHO (intron) TGGGTGGTGTCATCTGGTAA GGATGGAATGGATCAGATGG

CHEK2 (3’ UTR) AACCTCGCTATGCTCCCTTC CAGGTTTCCCAGGATGTCAC

ChIP-sequencing (ChIP-seq)

SOX10 ChIP DNA samples were confirmed for successful enrichment of known SOX10 binding sites (MITF, ERBB3, and MIA) compared to housekeeping genes (ACTB and GAPDH) and negative control regions (RNA28S5, RHO, and CHEK2) by quantitative PCR. SOX10 ChIP and input samples were evaluated for DNA quality and sequencing libraries were generated using the

KAPA Hyper Prep Kit from KAPA Biosystems. The resulting sequencing libraries were quantified, evaluated, and pooled before being sequenced on an Illumina NextSeq 500 platform.

All ChIP-seq sample preparation was performed by the Dana-Farber Cancer Institute Center for

Functional Cancer Epigenetics (DFCI CFCE) and all sequencing was performed by the Dana-

Farber Cancer Institute Molecular Biology Core Facility (DFCI MBCF).

132

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

ChIP-seq data quality control and analysis

SOX10 ChIP-seq data were processed using ChiLin (Qin et al., 2016). Raw sequencing reads were mapped to the genome using Bowtie (Langmead et al., 2009) and peaks were called using

MACS (Zhang et al., 2008). SOX10 ChIP peaks were analyzed for sequencing conservation by

PhastCons (Siepel et al., 2005), genomic feature annotation by CEAS (Shin et al., 2009), and motif enrichment by SeqPos (He et al., 2010; Liu et al., 2011). Analysis of SOX10 peaks for differentially enriched peaks in MITF-high and MITF-low melanoma was performed using

DESeq (Anders & Huber, 2010; Anders et al., 2013; Love et al., 2014). Putative genes associated with SOX10 peaks were inferred using BETA (Wang et al., 2013). All ChIP-seq data quality control and analysis were performed by the DFCI CFCE.

Immunoblot analysis

Cells were washed once with cold PBS and lysed passively with cold lysis buffer (1% NP40, protease inhibitor cocktail (Roche), phosphatase inhibitor cocktail sets I and II (CalBioChem)).

Protein lysates were quantified (Bradford assay), normalized, denatured (95 °C), analyzed by

SDS gel electrophoresis on 4-20% Tris-Glycine gels (Invitrogen). Resolved proteins were transferred to nitrocellulose or PVDF membranes, blocked in LiCOR blocking buffer, and probed with primary antibodies. After appropriate incubation with the appropriate secondary antibody, proteins were imaged and quantified using an Odyssey CLx scanner (LiCOR).

133

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

RNA purification, cDNA synthesis, and quantitative PCR

RNA was isolated from cells 48 hours post-transduction with lentivirus encoding shSOX10-1, shSOX10-5 or shLUC (control) using the RNeasy Mini Kit from Qiagen (#74104) per the manufacturer’s instructions. cDNA was synthesized from isolated RNA using the SuperScript III

First-Strand Synthesis SuperMix for qRT-PCR from Invitrogen (#11752-050) per the manufacturer’s instructions. cDNA was quantified by quantitative PCR using Power SYBR

Green PCR Master Mix from Invitrogen (#4367659) per the manufacturer’s instructions and the following qPCR primers:

Gene Forward Primer (5’-3’) Reverse Primer (5’-3’)

SOX10 CCTCACAGATCGCCTACACC CATATAGGAGAAGGCCGAGTAGA

MITF CATTGTTATGCTGGAAATGCTAGAA GGCTTGCTGTATGTGGTACTTGG

PMEL AGGTGCCTTTCTCCGTGAG AGCTTCAGCCAGATAGCCACT

CDK2 CCAGGAGTTACTTCTATGCCTGA TTCATCCAGGGGAGGTACAAC

ACTB ACGCCTCCGACCAGTGTT GCCCAGATTGGGGACAAA

GAPDH GCTCTCTGCTCCTCCTGTTC TAGCCTCCCGGGTTTCTC

134

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

RNA-sequencing (RNA-seq)

RNA from shSOX10-1, shSOX10-5, and shLUC samples were evaluated for RNA quality using the Agilent Bioanalyzer and sequencing libraries were prepared using the TruSeq Stranded mRNA Library Prep Kit from Illumina. The resulting sequencing libraries were quantified, evaluated, and pooled before being sequenced using an Illumina NextSeq 500 platform. The single-end 75bp sequencing runs produced approximately 400 million reads per sequencing lane.

All RNA sample preparation and sequencing were performed by the DFCI MBCF.

RNA-seq data quality control and analysis

STAR was used to align the raw sequencing reads to the hg19 assembly of the human reference genome and to generate raw read counts per gene (Dobin et al., 2013). Cufflinks was used to generate normalized fragments per kilobase of transcript per million (FPKM) expression values

(Trapnell et al., 2012). Data were further analyzed using the VIPER software package

(https://bitbucket.org/cfce/viper/) to determine read distribution over genomic features, quantify abundance of ribosomal RNA, and coverage of genebodies. All RNA-seq quality control was performed by the DFCI MBCF. GFOLD was used to generate generalized fold change (GFOLD) values from RNA-seq data and to compare gene expression across samples (Feng et al., 2012).

All RNA-seq data analysis was performed by the DFCI CFCE.

135

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

MSigDB overlap analysis of localization data

MSigDB was used to compute overlaps between a provided gene set and the gene sets included in the MSigDB collections (http://software.broadinstitute.org/gsea/msigdb/index.jsp). Our provided gene set included all genes associated with SOX10 peaks in at least 8 of 10 cell lines.

Overlap was computed with MSigDB Collection C2: Curated Gene Sets, which include gene sets curated from various sources.

Gene set enrichment analysis (GSEA) of transcriptome data

Normalized FPKM values for shSOX10-1, shSOX10-5, and shLUC RNA-seq samples were analyzed by GSEA to identify gene sets significantly enriched in shSOX10 vs. shLUC and vice versa (Subramanian et al., 2005).

136

Chapter 2: Characterization of the SOX10 Cistrome and Transcriptome in Melanoma

ACKNOWLEDGEMENTS

Xiaoyang Zhang provided guidance and assistance with ChIP-seq protocol optimization. Henry

Long and Prakash Rao (DFCI CFCE) assisted with ChIP-seq experimental setup and sequencing.

Henry Long, Jingyu Fan, and Xintao Qiu performed initial ChIP-seq data processing (ChiLin) and analysis, including motif analysis and differential peak analysis. Zachary Herbert (DFCI

MBCF) assisted with RNA-seq experimental setup, sequencing, and initial RNA-seq data processing (VIPER). Henry Long, Jingyu Fan, and Xintao Qiu also helped with RNA-seq data analysis (GFOLD).

137

CHAPTER 3

Investigation of HDAC inhibitors in Melanoma

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

SUMMARY

The eukaryotic genome exists as chromatin in the cell nucleus. Chromatin, composed of

DNA and associated proteins, namely histones, is highly dynamic and undergoes chemical and structural changes throughout the cell cycle and development. Alterations to chromatin can produce substantial and wide-spread changes in gene expression. Epigenetic mechanisms that affect the structure and expression of the genome include DNA methylation, histone modifications, and chromatin remodeling. One type of chromatin modification, histone acetylation, is generally associated with increased DNA accessibility, recruitment of transcription factors, and active gene expression. The acetylation of histones is performed by histone acetyltransferases (HATs) and is reversed by histone deacetylases (HDACs). These protein enzymes act in concert with other chromatin-associated proteins and transcription factors to modulate the expression of the genome. Small molecule compounds have been developed to target and inhibit HDAC proteins, resulting in wide-spread changes in gene expression. Based on improved response and survival rates, HDAC inhibitors have been approved for the treatment of cutaneous T-cell lymphoma, peripheral T-cell lymphoma, and multiple myeloma. However, due to their somewhat indiscriminate and extensive effects on gene expression, treatment with

HDAC inhibitors exhibit multiple adverse effects in patients, including anemia, thrombocytopenia, and leukopenia.

In this study, we investigated the effects of HDAC inhibitors on SOX10 expression and the combination of HDAC inhibitors with MAPK pathway inhibitors in melanoma. Treatment of melanoma cells with HDAC inhibitors resulted in dramatic decreases in SOX10 protein levels and mRNA expression of SOX10, MITF, and downstream genes (e.g., PMEL and CDK2).

Further experiments with broad and selective HDAC inhibitors demonstrated that these effects

139

Chapter 3: Investigation of HDAC Inhibitors in Melanoma are time-dependent and dose-dependent, and are primarily controlled by HDAC1 and HDAC2.

HDAC inhibitor treatment led to reduced cell viability across melanoma cell lines regardless of

SOX10 expression, suggesting that the reduction in SOX10 levels may not be the major downstream effector of HDAC inhibitor-mediated killing in melanoma cells. Additional experiments combining HDAC inhibitors with BRAF and MEK inhibitors (dabrafenib and trametinib, respectively) demonstrated that low doses of HDAC inhibitors may be synergistic with low doses of BRAF and MEK inhibitors in a subset of melanoma cell lines. Furthermore, the systematic investigation of combinatorial drug dosing and scheduling of these inhibitors revealed that the combination of low doses of HDAC inhibitors with normal doses of BRAF and

MEK inhibitors was more effective at killing melanoma cells than the MAPK pathway inhibitors alone, suggesting that low, non-toxic doses of HDAC inhibitors may be beneficial when combined with standard of care treatment in melanoma patients. Taken together, these studies demonstrate the utility of HDAC inhibitors in regulating SOX10 and its downstream genes, and the promise of combinatorial treatment of melanoma patients with HDAC and MAPK pathway inhibitors.

140

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

INTRODUCTION

The eukaryotic genome exists as chromatin in the cell nucleus. Chromatin is comprised of DNA and associated proteins, including histones. Histone proteins form multimeric complexes containing 8 subunits around which DNA is wrapped and compacted in the nucleus.

The packaging of DNA as chromatin is highly dynamic throughout the cell cycle and development. The active modification and remodeling of chromatin by transcription factors and protein complexes lead to structural and transcriptional changes in the genome, collectively termed the epigenome.

The epigenome represents an additional layer of genome regulation. Common mechanisms of epigenetic regulation include the methylation of DNA at cytosine residues, the post-translational modification of histone tails (e.g., acetylation, methylation, etc.), and physical changes mediated by chromatin remodeling complexes (e.g., PRC2 and SWI/SNF). Methylation of cytosine residues at CpG dinucleotides do not interfere with DNA base-pairing, but have significant impacts on the binding of transcription factors (e.g., CTCF) and the recruitment of chromatin-associated protein complexes (e.g., Sin3A HDAC complex). While generally not methylated in normal tissues, CpG islands, short stretches of DNA with high concentration of

CpG dinucleotides, can become aberrantly methylated in cancer. In addition, the abnormal methylation of genomic regions near gene promoters may result in altered gene expression in cancer. For example, the promoter hyper-methylation of tumor suppressor genes is an alternate mechanism to silence gene expression to promote tumorigenesis.

Histone tails can be post-translationally modified in various ways, including acetylation, methylation, phosphorylation, and others. In addition to the type of modification, the specific site or residue of modification also determines functional consequences. For example, methylation of

141

Chapter 3: Investigation of HDAC Inhibitors in Melanoma lysine 4 on histone 3 (H3K4me) is associated with transcriptional activation, while methylation of lysine 9 on histone 3 (H3K9me) is linked with transcriptional repression and silent heterochromatin. Acetylation of lysine residues on histone H3 and H4 are generally associated with transcriptional activation. This activity is mediated by two primary mechanisms: 1) acetylation of lysine residues neutralizes the positive charge at that site and leads to the weakening of electrostatic interactions with the negatively-charged DNA backbone, and 2) acetylated lysine residues serve as binding sites for transcription factors that can recognize them, resulting in changes in gene expression. Acetylation of histones is performed by histone acetyltransferases (HATs) and reversed by histone deactylases (HDACs). Due to the association of acetylated histones with transcriptional activation, HAT activity is linked to active gene expression, while HDAC activity is associated with repressed gene expression.

HDAC proteins are enzymes that recognize and remove acetyl groups from lysine residues on histones. Because they have an easily identifiable binding pocket, small molecule inhibitors have been developed against HDAC proteins. These compounds inhibit HDAC activity, leading to reduced lysine acetylation of histone H3 and altered gene expression. Due to the non-specific effects of HDAC activity across the genome, HDAC inhibition results in global changes in gene expression.

HDAC inhibitors have been approved by the U.S. Food and Drug Administration for the treatment of hematological malignancies: vorinostat (SAHA) for cutaneous T-cell lymphoma, belinostat for peripheral T-cell lymphoma, and panobinostat for multiple myeloma. As single agents or in combination with other drugs, treatment with HDAC inhibitors have led to improved response and survival rates in patients. However, use of these drugs also result in multiple adverse effects, including anemia (low red blood cell count), thrombocytopenia (low platelet

142

Chapter 3: Investigation of HDAC Inhibitors in Melanoma count), and leukopenia (low white blood cell count). These negative side effects may be attributed to non-specific and widespread effects of HDAC inhibitors on gene expression, especially when administered as systemic therapy. Additionally, these HDAC inhibitors target multiple HDAC proteins, including HDAC1, HDAC2, HDAC3, and HDAC6. With such a broad target space, these compounds likely have many off-target effects on gene expression as well as acetylation of unintended protein targets. The investigation of more selective HDAC inhibitors with sustained efficacy may provide clinical benefit without undesirable adverse effects in cancer patients.

143

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

RESULTS

Evaluation of HDAC Inhibitors on SOX10 Expression and Activity

In this study, we have established SOX10 as a differential genetic dependency in melanoma and that it binds to and regulates hundreds of target genes in melanoma cells, some of which contribute to common oncogenic phenotypes. Targeting and inhibition of SOX10 expression or activity could lead to substantial negative effects on the growth and viability of melanoma cells and could produce meaningful benefit in human melanoma patients. Since

SOX10 is a transcription factor and transcription factors are historically challenging to target with small molecule compounds, we investigated a known and robust observation that treatment with HDAC inhibitors (e.g., vorinostat, panobinostat, and trichostatin A) reduces SOX10 protein levels in melanoma cell lines (Yokoyama et al., 2008; Johannessen et al., 2013). To confirm this finding in our laboratory, we evaluated the effects of HDAC inhibitors on SOX10 protein levels in melanoma cell lines. Treatment of cell lines expressing SOX10 with 10 μM vorinostat

(SAHA) resulted in near complete reduction of SOX10 protein levels at 24 h (Figure 3-1).

144

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Figure 3-1. Vorinostat treatment reduces SOX10 protein levels in cancer cell lines. Cell lines were treated with 10 μM vorinostat (SAHA) over 24 h. Protein was isolated and quantified by immunoblot for SOX10 and beta-actin.

Published and unpublished studies, including those by our laboratory, have shown that broad or pan-HDAC inhibitors, including vorinostat, panobinostat, entinostat, and trichostatin A

(TSA), robustly reduce SOX10 RNA and protein levels in melanoma cell lines. However, it is unclear which individual or combination of HDAC proteins is responsible for this effect since these pan-HDAC inhibitors block several HDAC proteins: HDAC1, HDAC2, HDAC3, and

HDAC6. Furthermore, clinical use of vorinostat, approved by the FDA for the treatment of cutaneous T-cell lymphoma, can lead to a wide array of adverse effects in patients, including fatigue, diarrhea, nausea, thrombocytopenia, and anemia. Since vorinostat targets multiple

HDAC proteins, HDAC inhibitors that are more selective in their HDAC targets may result in fewer or less severe side effects in patients. To determine if selective HDAC inhibitors can also reduce SOX10 RNA and protein levels, melanoma cell lines were treated with a panel of selective HDAC inhibitors provided by the Broad Institute Stanley Center Medicinal Chemistry group (Table 3-1) and assayed for SOX10 RNA and protein levels. Treatment of G361,

WM983B, and A375 cells with three compounds, CI-994 (HDAC1/2/3 inhibitor), Merck60 and

BRD2492 (HDAC1/2 inhibitors), at 20 μM robustly reduced SOX10 protein levels at 72 h, but

145

Chapter 3: Investigation of HDAC Inhibitors in Melanoma not at 24 h (Figure 3-2). The reduction in SOX10 levels correlated with on-target activity of the

HDAC inhibitors, demonstrated by increased acetylation of histone H3. Identical treatment of

SKMEL19 and A2058 cells was lethal at 72 h, suggesting that HDAC inhibitors may have differential effects in melanoma cell lines.

Table 3-1. Broad and selective HDAC inhibitors.

146

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Figure 3-2. Selective HDAC inhibitors reduce SOX10 protein levels at 72 h. G361 cells were treated with HDAC inhibitors at 20 μM for 24 h or 72 h. Protein was isolated and quantified by immunoblot for SOX10, acetylated H3, acetylated tubulin, and vinculin.

To further investigate the mechanism of SOX10 protein reduction by HDAC inhibitors, we performed a series of experiments to evaluate the effects of broad and selective HDAC inhibitors on SOX10 gene expression and protein levels. First, G361 cells were treated with 10

μM of vorinostat (SAHA), Merck60, or BRD2492 for a period of 0 to 7 days. As previously observed, vorinostat reduced SOX10 protein within 24 h, while Merck60 and BRD2492 reduced

SOX10 protein within 72 h (Figure 3-3). Reduction in SOX10 protein was correlated with increase in histone H3 acetylation. Cells were nonviable beyond the time of SOX10 protein reduction and were not available for investigation. Next, G361 cells were treated with increasing concentrations of vorinostat (SAHA), CI-994, Merck60, or BRD2492 for 72 h. Treatment with 2

μM of Merck60 sufficiently reduced SOX10 protein, while greater concentrations of 10 μM were

147

Chapter 3: Investigation of HDAC Inhibitors in Melanoma needed of CI-994 or BRD2492 to observe similar effects (Figure 3-4). Vorinostat was lethal at these doses for 72 h treatment. These findings supporting the time-dependent and dose- dependent reduction of SOX10 protein by broad and selective HDAC inhibitors indicate that

HDAC inhibitors have an on-target effect against SOX10.

Figure 3-3. Vorinostat, Merck60, and BRD2492 reduce SOX10 protein levels in a time- dependent manner. G361 cells were treated with HDAC inhibitors at 10 μM over 7 days.

Protein was isolated and quantified by immunoblot for SOX10, acetylated H3, and beta-actin.

Figure 3-4. Vorinostat, CI-994, Merck60, and BRD2492 reduce SOX10 protein levels in a dose-dependent manner. G361 cells were treated with HDAC inhibitors over 72 h. Protein was isolated and quantified by immunoblot for SOX10, acetylated H3, and beta-actin.

148

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

HDAC inhibitors bind to HDAC proteins and inhibit their activity. The HDAC proteins implicated in reducing SOX10 protein levels have transcriptional (HDAC1, HDAC2, HDAC3) or protein modifying activity (HDAC6). Thus, the observed HDAC inhibitor-mediated reduction of SOX10 protein levels may be the result of changes at the transcriptional level (i.e., mRNA expression) or at the protein level (e.g., protein stability or degradation). To elucidate whether

HDAC inhibitors regulate SOX10 at the mRNA or protein level, we performed a set of experiments to investigate the interaction between HDAC inhibitors, SOX10 protein levels, and the proteasome or downstream SOX10 target genes. First, cell lines were treated with vorinostat alone or in combination with MG132, a proteasome inhibitor, for 16 h. Simultaneous co- treatment with MG132 resulted in partial, but not complete, rescue of HDAC inhibitor-mediated reduction of SOX10, suggesting that proteasomal degradation is not the major mechanism through which HDAC inhibitors regulate SOX10 levels (Figure 3-5). Next, SKMEL19 cells were treated with vorinostat or Merck60 in varying concentrations and durations. Treatment with

2 μM Merck60 reduced SOX10 mRNA expression levels at 72 h, as well as mRNA levels for

MITF and CDK2 (Figure 3-6). In addition, increasing concentrations of Merck60 resulted in a dose-dependent reduction in SOX10, MITF, and CDK2 mRNA expression (Figure 3-6).

Additional experiments with vorinostat treatment reiterated the observations made with

Merck60, showing that 2 μM vorinostat can reduce SOX10, MITF, and CDK2 mRNA expression within 24 h (Figure 3-7). Taken together, these results confirm the on-target effects of HDAC inhibitors on SOX10 levels and indicate that the mechanism of HDAC inhibitor-mediated regulation is largely at the transcriptional level.

149

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Figure 3-5. Proteasome inhibitor MG132 does not fully rescue vorinostat-mediated reduction of SOX10 protein levels. Cells were treated with 10 μM vorinostat (SAHA) and 10

μM MG132. Protein was isolated and quantified by immunoblot for SOX10 and beta-actin.

Figure 3-6. Merck60 treatment reduces mRNA expression of SOX10, MITF, and CDK2 in a time-dependent and dose-dependent manner. SKMEL19 cells were treated with Merck60 for increasing times or increasing doses. RNA was isolated and quantified by RT-qPCR for SOX10,

MITF, PMEL, CDK2, ACTB, and GAPDH.

150

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Figure 3-7. Vorinostat treatment reduces mRNA expression of SOX10, MITF, and CDK2 within 24 h. SKMEL19 cells were treated with vorinostat (SAHA) for increasing duration or increasing doses. RNA was isolated and quantified by RT-qPCR for SOX10, MITF, PMEL,

CDK2, ACTB, and GAPDH.

Our experiments have demonstrated that extended treatment with broad or selective

HDAC inhibitors leads to negative viability effects in melanoma cell lines, with much shorter time-frames for the broad HDAC inhibitor, vorinostat. In addition, we have shown that one way in which HDAC inhibitors exert their effects in melanoma cells is through their regulation of

SOX10 gene expression. However, it remains unclear whether SOX10 is a specific or the most significant target of HDAC inhibitor-mediated transcriptional regulation and killing in melanoma cells. To determine whether broad and selective HDAC inhibitors kill melanoma cell lines in a

SOX10-specific manner, we performed drug treatment experiments in a panel of melanoma cell lines with varying levels of SOX10 expression. The 96 h cell viability dose-curves for all tested

HDAC inhibitors (vorinostat, CI-994, Merck60, and BRD2492) did not exhibit significant differences between melanoma cell lines that express SOX10 compared to those that do not

151

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

(Figure 3-8). These results suggest that the cell killing effects of HDAC inhibitors is not primarily mediated by their regulation of SOX10 mRNA expression. The SOX10 gene may be one of the major transcriptional targets of HDAC proteins, but there are likely hundreds or thousands more genes whose expression would be perturbed by HDAC inhibitor treatment.

Despite their non-specific effects in melanoma cells, HDAC inhibitors could still hold clinical benefit in combination with other therapeutic targets and modalities.

Figure 3-8. HDAC inhibitors reduce melanoma cell viability regardless of SOX10 expression. Cells were treated with HDAC inhibitors over a range of doses for 4 days. Cell viability was quantified by CellTiter-Glo.

152

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Combination of HDAC Inhibitors and MAPK Pathway Inhibitors

One of the most dramatic recent advances in the treatment of melanoma patients is the use of targeted agents against the MAPK pathway. BRAF inhibitors (e.g., vemurafenib, dabrafenib) and MEK inhibitors (e.g., trametinib) have demonstrated strong negative effects in melanoma cell line models as well as in melanoma patients. However, innate and acquired resistance to these inhibitors necessitates the exploration of additional therapeutic avenues that may produce additive or synergistic killing effects in melanoma cells. To evaluate the therapeutic potential of combinatorial treatment with HDAC inhibitors with MAPK pathway inhibitors

(MAPKi) in melanoma, we performed a set of related drug combination experiments in melanoma cell lines. First, simultaneous, short-term (96 h) treatment of vorinostat or Merck60 with combined dabrafenib and trametinib did not show synergistic killing in MAPKi-sensitive

(SKMEL24, WM88) or MAPKi-resistant (A2058, IGR39, RPMI7951) melanoma cell lines

(Figure 3-9). Next, to determine if pre-treatment with an HDAC inhibitor can sensitize MAPKi- resistant melanoma cell lines (RPMI7951, IGR39, A2058, WM793) to MAPK pathway inhibitors, melanoma cells were treated sequentially with two different schedules: 1) 2 μM vorinostat or Merck60 for 3 days, then 100 nM dabrafenib + 10 nM trametinib for 3 days; 2) 2

μM vorinostat or Merck60 for 6 days, then 100 nM dabrafenib + 10 nM trametinib for 4 days.

These experiments generated mixed results across the tested MAPKi-resistant cell lines, with some combinations showing additivity or synergy in cell killing (Figure 3-10). Compound on- target effects were confirmed by RT-qPCR: down-regulation of SOX10, MITF, and ERBB3 mRNA expression by HDAC inhibitors and down-regulation of CCND1, ETV5, DUSP6, and

SPRY2 by MAPK pathway inhibitors. Next, to determine if pre-treatment with low doses of

HDAC inhibitors can sensitize MAPKi-sensitive and MAPKi-resistant melanoma cell lines to

153

Chapter 3: Investigation of HDAC Inhibitors in Melanoma low doses of MAPK pathway inhibitors, a panel of melanoma cell lines was treated with 1 μM vorinostat for 3 days, then 10 nM dabrafenib + 1 nM trametinib for 4 days. If cells were viable afte 7 days of treatment, cells were counted, reseeded, and treated with the same drug schedule for another 7 days. The sequential combination of low dose vorinostat and low dose dabrafenib + trametinib showed strong synergy (Excess over Bliss (EOB) > 0.3) in WM793, SKMEL19,

UACC62, and G361, slight synergy (0.1 < EOB < 0.2) in WM983B, RVH421, and WM115, and additivity (0 < EOB < 0.1) in WM2664, A2058, and RPMI7951 (Figure 3-11). These results suggest that low dose vorinostat treatment may be synergistic in some melanoma cell models, but not all.

154

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Figure 3-9. Simultaneous, short-term combination treatment of HDAC inhibitors and

MAPK pathway inhibitors did not produce synergistic killing in melanoma cell lines. Cells were treated with increasing doses of HDAC inhibitor (vorinostat/SAHA or Merck60) and increasing doses of dabrafenib and trametinib for 96 h. Cell viability was quantified by CellTiter-

Glo and Excess over Bliss scores were calculated.

155

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Figure 3-10. Pre-treatment with HDAC inhibitors can sensitize some MAPKi-resistant melanoma cell lines to MAPK pathway inhibitors. Cells were treated with 2 μM HDAC inhibitor (vorinostat/SAHA or Merck60) for 3 or 6 days, then 100 nM dabrafenib + 10 nM trametinib for 3 or 4 days. Cells were counted every 3-4 days.

156

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Figure 3-11. Pre-treatment with low dose of HDAC inhibitors can sensitize some MAPKi- sensitive and MAPKi-resistant melanoma cell lines to low doses of MAPK pathway inhibitors. Cells were treated with 1 μM vorinostat (SAHA) for 3 days, then 10 nM dabrafenib +

1 nM trametinib for 4 days. Cells were counted at 7 days, then passaged and counted again if possible.

157

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

To determine whether variable dosing and duration of treatment with HDAC inhibitors and MAPK pathway inhibitors can kill melanoma cell lines more effectively than the standard clinical regimen of MAPK pathway inhibitors alone (dabrafenib + trametinib), melanoma cells were treated with high or low doses of vorinostat (10 μM or 1 μM) and/or high or low doses of dabrafenib + trametinib (100 nM, 10 nM or 10 nM, 1 nM) in pulsed (1 day) or chronic (7 days for vorinostat, 4 days for dabrafenib + trametinib) schedules. The combination of dosing and duration variables resulted in 25 different treatment conditions (Table 3-2). Overall, chronic high dose vorinostat treatment led to cell killing with or without MAPK pathway inhibitor treatment

(Figure 3-12). Notably, chronic low dose vorinostat treatment was equivalent to pulsed high dose vorinostat in all cell lines, suggesting that lower doses of vorinostat or other HDAC inhibitors may be therapeutically effective in patients. Furthermore, the combinations of chronic low dose vorinostat or pulsed high dose vorinostat with chronic high dose dabrafenib + trametinib were more effective than chronic high dose dabrafenib + trametinib alone. These results suggest that vorinostat should be further investigated in combination with conventional

MAPK pathway-targeted therapies in melanoma patients.

158

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Table 3-2. Combination dosing and duration schedule of vorinostat (SAHA) and dabrafenib (DAB) + trametinib (TRA).

159

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Figure 3-12. Combination treatment of vorinostat and dabrafenib + trametinib in MAPKi- sensitive melanoma cell lines. SKMEL19 (A) and WM983B (B) cells were treated with vorinostat and/or dabrafenib + tramenitib using a variable dosing and duration schedule (Table 3-

2). Standard treatment (100 nM dabrafenib + 10 nM trametinib) is marked in red. Additionally, similar results were observed in WM2664 and UACC62 cells.

160

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Figure 3-12 (Continued).

161

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

DISCUSSION

In this study, we have investigated the effects of HDAC inhibitors on SOX10 levels and the potential for combination of HDAC inhibitors with MAPK pathway inhibitors for the treatment of melanoma. We confirmed previous findings that HDAC inhibitors reduce SOX10 protein levels and extended those observations with selective HDAC inhibitors and downstream targets of SOX10, including MITF and its target genes. In addition, we demonstrated that the

HDACi-mediated reduction in SOX10 was time-dependent and dose-dependent, indicating that these results represent on-target effects on SOX10 expression. While HDAC inhibitor treatment killed melanoma cell lines regardless of SOX10 expression, we showed through variable drug dosing and duration schedules that low-dose HDAC inhibitors can be combined with typical doses of BRAF and MEK inhibitors for greater killing in melanoma cells.

Our experiments with broad and selective HDAC inhibitors revealed that HDAC1 and

HDAC2 are the HDAC proteins responsible for the observed reduction in SOX10 levels. To determine whether HDAC1 or HDAC2 is the primary HDAC protein involved, we attempted shRNA-mediated knockdown experiments against the individual HDAC genes. However, our reagents were not sufficient to produce reliable knockdown of their intended targets in order to obtain conclusive results. Future experiments with better shRNA or CRISPR-Cas9 reagents may provide additional data to determine whether HDAC1 and/or HDAC2 are required for sustained

SOX10 expression. Also, HDAC1 and HDAC2 are highly related proteins and may be functionally redundant, regulating the expression of a similar set of transcriptional targets. Thus, the combined inhibition of HDAC1 and HDAC2 may be necessary to reduce SOX10 levels in melanoma cells.

162

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Generally, HDAC proteins act to remove acetyl groups from lysines on histone tails, leading to stronger interactions between histones and DNA and the loss of acetyl groups as binding sites for transcription factors, and ultimately to repressed gene expression. Thus, HDAC inhibitors would result in reduced HDAC activity, the maintenance of acetylated histone marks, and active gene expression. Our observations of SOX10 expression changes in melanoma cell lines after treatment with HDAC inhibitors do not match the general pattern of HDAC-mediated gene regulation. Thus, the most parsimonious model would include an intermediate protein or transcription factor whose expression is reduced by HDAC activity (per the traditional HDAC model) and which negatively regulates SOX10 expression. HDAC inhibitor treatment would relieve negative regulation of HDACs on this protein or transcription factor, allowing it to reduce

SOX10 expression in melanoma cells. Additional experiments to identify the transcriptional targets of HDAC inhibitors in melanoma cell lines coupled with experiments to characterize the set of transcription factors that regulate SOX10 expression would provide insights into this gene regulatory hierarchy.

HDAC inhibitor treatment of melanoma cell lines resulted in wide-spread killing regardless of SOX10 expression, suggesting that SOX10 is not the major effector of HDAC inhibitor-mediated cell killing (Figure 3-8). This observation disqualifies HDAC inhibitors as a

SOX10-expressing melanoma-specific therapy, but HDAC inhibitors can still provide much benefit in melanoma treatment plans. In fact, the widespread transcriptional effects of HDAC inhibitors may be an advantage when used in combination with other therapies, such as targeted therapy and immunotherapy. For example, treatment with HDAC inhibitors may shift melanoma cells, such as those that are innately resistant to MAPK pathway inhibitors, into an alternate transcriptional state that makes them more susceptible or sensitive to MAPK pathway inhibitors.

163

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Additionally, the extensive changes in gene expression resulting from HDAC inhibitor treatment may lead to the expression and presentation of neo-antigens on the tumor cell surface that result in activation of an immune response facilitated by immunotherapy agents.

Our combinatorial drug treatment experiments of HDAC inhibitors and BRAF and MEK inhibitors in melanoma cell lines provide evidence for the continued study of these combinations in cancer treatment. While treatment with the typical doses of BRAF and MEK inhibitors led to cell killing, the addition of low doses of HDAC inhibitors resulted in additional killing (Figure

3-12). As mentioned previously, treatment with HDAC inhibitors in human patients results in a number of hematologic-related adverse effects, including anemia, thrombocytopenia, and leukopenia. Lower doses of HDAC inhibitors, and potentially more selective HDAC inhibitors, would likely decrease the risk of toxic effects in patients while still providing clinical benefit.

164

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

METHODS

Cell lines and reagents

All cell lines were obtained from and were identity confirmed with DNA fingerprinting by the

Broad Institute Genomics Platform. SKMEL19, WM983B, UACC62, RVH421, and WM793 were grown in RPMI medium, 10% FBS. G361, WM88, WM2664, A2058, SKMEL24, IGR39,

LOXIMVI, RPMI7951, and LN464 were grown in DMEM medium, 10% FBS. C32 was grown in EMEM medium, 10% FBS. Antibodies for immunoblot were obtained from Abcam (SOX10 rabbit monoclonal, ab155279 [EPR4007]), EMD Millipore (acetyl-histone H3 rabbit polyclonal,

06-599), Cell Signaling (β-actin mouse monoclonal, #3700 8H10D10), and Sigma-Aldrich

(vinculin mouse monoclonal, V9131). Vorinostat, dabrafenib, and trametinib were obtained from

Selleck Chemicals. Selective HDAC inhibitors, including Merck60, BRD2492, and CI-994, were obtained from the Broad Institute Stanley Center for Psychiatric Research.

Immunoblot analysis

Cells were washed once with cold PBS and lysed passively with cold lysis buffer (1% NP40, protease inhibitor cocktail (Roche), phosphatase inhibitor cocktail sets I and II (CalBioChem)).

Protein lysates were quantified (Bradford assay), normalized, denatured (95 °C), analyzed by

SDS gel electrophoresis on 4-20% Tris-Glycine gels (Invitrogen). Resolved proteins were transferred to nitrocellulose or PVDF membranes, blocked in LiCOR blocking buffer, and probed with primary antibodies. After appropriate incubation with the appropriate secondary antibody, proteins were imaged and quantified using an Odyssey CLx scanner (LiCOR).

165

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

RNA purification, cDNA synthesis, and quantitative PCR

RNA was isolated from cells 48 hours post-transduction with lentivirus encoding shSOX10-1, shSOX10-5 or shLUC (control) using the RNeasy Mini Kit from Qiagen (#74104) per the manufacturer’s instructions. cDNA was synthesized from isolated RNA using the SuperScript III

First-Strand Synthesis SuperMix for qRT-PCR from Invitrogen (#11752-050) per the manufacturer’s instructions. cDNA was quantified by quantitative PCR using Power SYBR

Green PCR Master Mix from Invitrogen (#4367659) per the manufacturer’s instructions and the following qPCR primers:

Gene Forward Primer (5’-3’) Reverse Primer (5’-3’)

SOX10 CCTCACAGATCGCCTACACC CATATAGGAGAAGGCCGAGTAGA

MITF CATTGTTATGCTGGAAATGCTAGAA GGCTTGCTGTATGTGGTACTTGG

PMEL AGGTGCCTTTCTCCGTGAG AGCTTCAGCCAGATAGCCACT

CDK2 CCAGGAGTTACTTCTATGCCTGA TTCATCCAGGGGAGGTACAAC

ACTB ACGCCTCCGACCAGTGTT GCCCAGATTGGGGACAAA

GAPDH GCTCTCTGCTCCTCCTGTTC TAGCCTCCCGGGTTTCTC

166

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Cell viability assay (short-term assays)

Cells were seeded in 96-well plates and treated with vorinostat, CI-994, Merck60, or BRD2492 in increasing doses with or without dabrafenib and trametinib in increasing doses (10:1 molar ratio for dabrafenib and trametinib) for 4 days. Cell viability was quantified by CellTiter-Glo per the manufacturer’s instructions. Synergy was calculated using Excess over Bliss (EoB): % inhibition (combined) – (% inhibition (HDACi) + % inhibition (BRAFi+MEKi) – [% inhibition

(HDACi) * % inhibition (BRAFi+MEKi)]).

Population doubling assays (pretreatment assay #1)

Cells were seeded in 12-well plates and treated with vorinostat or Merck60 at 2 μM for 3 or 6 days, then with 100 nM dabrafenib + 10 nM trametinib for 3 or 4 days. Cells were counted every

3-4 days. Synergy was calculated using Excess over Bliss (EoB): % inhibition (combined) – (% inhibition (HDACi) + % inhibition (BRAFi+MEKi) – [% inhibition (HDACi) * % inhibition

(BRAFi+MEKi)]).

Population doubling assays (pretreatment assay #2)

Cells were seeded in 12-well plates and treated with vorinostat at 1 μM for 3 days, then with 10 nM dabrafenib + 1 nM trametinib for 4 days. Cells were counted every 7 days. Synergy was calculated using Excess over Bliss (EoB): % inhibition (combined) – (% inhibition (HDACi) +

% inhibition (BRAFi+MEKi) – [% inhibition (HDACi) * % inhibition (BRAFi+MEKi)]).

167

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

Population doubling assays (variable dosing and duration schedule assay)

Cells were seeded in 6 well plates and treated with vorinostat at 1 μM or 10 μM for 1 day or 7 days, then with 100 nM dabrafenib + 10 nM trametinib or 10 nM dabrafeinib + 1 nM trametinib for 1 day or 4 days. Cells were imaged using a light microscope.

168

Chapter 3: Investigation of HDAC Inhibitors in Melanoma

ACKNOWLEDGEMENTS

Initial discussions regarding HDAC inhibitor experiments were conducted with Cory

Johannessen. Selective HDAC inhibitors were provided by Florence Wagner (Broad Institute

Stanley Center for Psychiatric Research).

169

CONCLUSION

Integration, Characterization, and Innovation

Conclusion: Integration, Characterization, and Innovation

SUMMARY

Taken together, the work presented here illustrates the power and utility of integrating large genome-scale datasets to uncover novel biological insights and potential therapeutic strategies for the treatment of cancer. Large functional genomic screening projects, such as

Project Achilles, have enabled the systematic interrogation of gene function across a multitude of model systems. When combined with additional genomic datasets, including those that characterize cancer cell lines and patient tumors, the analysis of functional genomic screens can reveal previously unknown facets of cancer biology. Moreover, other large-scale approaches to characterize cancer-associated genes and proteins may provide additional knowledge of cancer dependencies. The investigation and comparison of cancer cells with corresponding normal cells, related lineages, and developmental stages will put our functional studies of cancer dependencies into the context of normal human biology and development. Finally, the targeting of most tumor vulnerabilities remains a challenge despite decades of drug discovery and development. Novel approaches and therapeutic modalities present promising opportunities for cancer treatment.

171

Conclusion: Integration, Characterization, and Innovation

Power of Functional Genomic Characterization of Cancer

Our utilization and analysis of functional genomic data from Project Achilles enabled the discovery of differential genetic dependencies across a range of tumor types. The identified dependencies included known vulnerabilities, including those associated with driver mutations in cancer (e.g., BRAF in melanoma, KRAS in pancreatic cancer, etc.) and previously unknown dependencies (e.g., SOX10 in melanoma). These types of analyses substantially supplement and complement the comprehensive genomic studies that characterize cancer cell lines and patient tumors at the structural level, such as the Cancer Cell Line Encyclopedia and The Cancer

Genome Atlas. The integration of structural and functional data types allow for a more cohesive and unified understanding of cancer.

In addition to the identification of novel and potentially targetable genetic dependencies in cancer, our studies also highlight the opportunity to discover previously unknown connections between lineage-specific dependencies and common oncogenic dependencies. For example, our gene set analyses revealed shared target genes between SOX10 and MYC. This overlap reveals a potential link between tissue-specific master regulators important for the development of normal and malignant tissues and well-characterized genes and pathways associated with cancer. In addition, the analysis of shared differential genetic dependencies across tumor lineages, may also uncover major genes and pathways in normal tissues and development.

172

Conclusion: Integration, Characterization, and Innovation

Further Characterization of SOX10 in Melanoma

Our comprehensive characterization of the SOX10 cistrome and transcriptome in melanoma cell lines has contributed to a greater understanding of SOX10 biology, transcriptional activity, and dependency in melanoma. Prior to this work, SOX10 had primarily been studied in the context of neural crest development and in differentiated glial cells, including Schwann cells and oligodendrocytes. Our investigation of SOX10 dependency and transcriptional programs in melanoma adds to the growing field of SOX10 biology in human development and disease.

In addition to the cistromic and transcriptomic study of SOX10 in melanoma, further investigation of SOX10 localization and transcriptional regulation in related cells and tissues would allow for comparative assessments of SOX10 activity throughout development and the adult organism. SOX10 ChIP-seq and SOX10 knockdown RNA-seq experiments in normal melanocytes would enable the discovery of SOX10 target genes involved in oncogenic transformation from melanocytes to melanoma cells. Parallel experiments in differentiated cell types derived from the neural crest, including glial cells, would provide insights into fate-specific

SOX10 target genes and transcriptional programs. Finally, the characterization of SOX10 binding and transcriptional targets in neural crest cells throughout development, from the formation of the neural crest through fate specification and migration to terminal tissues, would facilitate unprecedented understanding of differential SOX10 targets and downstream pathways throughout embryonic development.

Additionally, the role of SOX10 in melanoma can be further characterized in parallel and complementary ways. Our studies have focused on the target genes, effectors, and pathways that lie downstream of SOX10 in melanoma biology. While these studies have added to our foundational knowledge of SOX10 biology, there remains much to learn regarding upstream

173

Conclusion: Integration, Characterization, and Innovation regulatory molecules and pathways leading to SOX10 in melanoma. These types of studies would have direct impact and application to the potential future targeting of SOX10 in cancer.

SOX10 interacts with and binds to DNA with partner transcription factors, such as PAX3 and MITF. However, the full set of SOX10 partner factors has not been fully characterized, especially in melanoma. To generate the complete catalogue of SOX10 interaction partners, melanoma cells could be subjected to SOX10 co-immunoprecipitation experiments coupled with high-throughput protein characterization by mass spectrometry. These studies could also incorporate additional components, including SOX10 and partner factor interactions on DNA vs. interactions not associated with DNA binding. In addition to determining the list of SOX10 interaction partners, such studies could also provide insights into the regulation of SOX10 activity, such as whether SOX10 interacts with its partner factors before or during DNA binding.

SOX10, like other proteins and transcription factors, are dynamically regulated by post- translational modifications that may affect protein stability, activity, and localization. For example, sumoylation of SOX10 at three sumoylation consensus sites do not affect SOX10 cellular localization, but represses its transcriptional activity at two target genes, MITF and

GJB1, and modulates its synergy with PAX3 and EGR2 at these promoters (Girard & Goossens,

2006). The characterization of other post-translational modifications to SOX10, such as phosphorylation and acetylation, would provide insights into mechanisms regulating SOX10 protein levels and transcriptional activity.

174

Conclusion: Integration, Characterization, and Innovation

Therapeutic Strategies for Targeting SOX10 in Cancer

Our studies demonstrating the essentiality of SOX10 in melanoma strongly support its continued investigation as a therapeutic target in human melanoma. The widespread dependency on SOX10 across melanoma cell lines, including those that may not respond to MAPK pathway inhibitors, demonstrate the promise of targeting SOX10 as a therapeutic strategy in melanoma.

Additionally, SOX10 dependency may be an orthogonal and equally essential vulnerability in melanoma compared to MAPK and other pathway dependencies, leading to potentially dramatic responses in combination therapy approaches.

While the targeting of SOX10 alone may provide clinical benefit, combinatorial inhibition of SOX10 dependency and other melanoma dependencies may prove more beneficial.

The use of improved genetic perturbation reagents, such as CRISPR-Cas9, against SOX10 will facilitate a better understanding of SOX10 dependency and its combination with current therapeutic strategies in melanoma, including MAPK inhibitors. In addition, the rapid adoption of immunotherapy as a standard treatment for melanoma patients warrants its investigation with

SOX10-directed therapies in animal models of melanoma.

As a transcription factor, SOX10 will be challenging to directly target with current therapeutic modalities. Its lack of an obvious binding pocket or enzymatic active site and its primarily nuclear localization preclude the targeting of SOX10 with small molecule inhibitors and monoclonal antibodies, respectively. Instead, therapeutic targeting of SOX10 will need to take advantage of new and promising strategies in drug development. Antisense oligonucleotides

(ASOs) and other nucleic acid-based drug molecules may circumvent the difficulty in directly binding to SOX10 protein by targeting SOX10 mRNA transcripts before translation into protein.

Meanwhile, gene therapy-based approaches, such as CRISPR-Cas9, could directly target SOX10

175

Conclusion: Integration, Characterization, and Innovation in the genome and prevent transcription in the first place. These and other therapies will likely face many challenges, including specific targeting to melanoma cells and unwanted off-target effects as a result of systemic treatment.

If direct targeting of SOX10 is not a tractable option, upstream regulators of SOX10 expression or protein activity could be pursued. For example, our studies on HDAC inhibitors demonstrate the promise of targeting chromatin-modifying enzymes to achieve desired effects on

SOX10 expression. In addition, the sumoylation and other as yet undiscovered post-translational modifications of SOX10 may be targeted via the enzymes that carry out those protein modifications.

176

Conclusion: Integration, Characterization, and Innovation

FUTURE CONSIDERATIONS

The integration of large datasets, the characterization of genetic dependencies, and the innovation of novel therapeutic approaches lay the groundwork for promising and insightful developments in basic and translational cancer research. The convergence of molecular biology, functional genomics, computational biology, novel technologies, and therapeutic science has enabled and will continue to propel the discovery and application of scientific understanding of cancer for the ultimate benefit of patients. Our work describes the integration and analysis of multiple datasets to uncover new biological and therapeutic insights in cancer. The approaches used in these and other studies will be crucial to our future understanding and treatment of cancer.

177

APPENDIX 1

Integrated Chemical Screen for Modulators of SOX10 Transcriptional Activity

178

INTRODUCTION: CONNECTIVITY MAP

While HDAC inhibitors may prove useful in targeting SOX10, other chemical approaches to extinguish SOX10 and/or selectively kill SOX10+ melanoma lines may be beneficial. Most chemical screening projects begin with libraries of hundreds of thousands of compounds, such as the Broad Institute’s Diversity of Synthesis (DOS) Library. However, these projects are very expensive and require a lot of labor and time. Instead, we embarked on a path toward identifying SOX10 inhibitors using pre-existing compound data. The Broad Institute

Connectivity Map (CMAP) is a large database of gene expression signatures as a result of compound or genetic perturbations in a number of representative cancer cell lines (Lamb et al.,

2006). To simplify and reduce the cost of generating these signatures, CMAP utilizes the L1000 platform to characterize cell states. L1000 experimentally assays transcript levels for approximately 1,000 genes, then imputes the values for all other genes using an empirically derived algorithm. Signatures derived from compound treatment or genetic perturbations

(shRNA or ORF) can be queried in the CMAP database and connections can be made based on correlation or anti-correlation (https://clue.io/cmap).

RESULTS

As a transcription factor, SOX10 is considered an “undruggable” target with current drug development approaches. However, upstream or downstream effectors of SOX10 may be targeted with compounds. To identify small molecules that may modulate SOX10 gene expression, protein stability, or transcriptional activity, we performed a computational “in silico” screen for compounds whose gene expression signature correlates with a SOX10 knockdown

179

signature in a SOX10-expressing melanoma cell line (A375) in the Broad Institute Connectivity

Map (CMAP). CMAP contains imputed genome-wide expression data based on nearly 1000 landmark genes (L1000) for 20,000 compounds and 21,500 genetic perturbations (shRNA or

ORF) in 8 cancer cell lines (Lamb et al., 2006). CMAP compound signatures were queried for those that correlate with the SOX10 Consensus Gene Signature (CGS), a weighted correlation of the three SOX10 shRNA signatures, resulting in a set of 155 compounds with a rankpoint (scaled percentile rank) > 90 (range: -100 to 100). These compounds represent potential inhibitors of

SOX10 in melanoma cells. Of these compounds, only 81 were available through the Broad

Institute Therapeutics Platform and met requirements for purity, concentration, and volume.

After obtaining permission to use these compounds from the corresponding project leads, 80 compounds were requested in 384-well format in 4 doses (1:5 dilution from 10 μM for final concentrations in treated wells of 0.08, 0.4, 2, and 10 μM). With assistance from the Broad

Institute Therapeutics Platform, this plate of compounds was screened in triplicate across 20 cell lines: 5 SOX10+ MITF-high melanoma, 5 SOX10+ MITF-low melanoma, and 3 SOX10- melanoma (Table A1-1). Cell viability was assayed by CellTiter-Glo (CTG) after 96-h compound treatment. CTG values were normalized within each plate (to DMSO) and area under the curve (AUC) values were calculated for each compound in each plate as a measure of sensitivity (smaller AUC indicates greater sensitivity to given compound). AUC values for each compound were averaged across replicates of the same cell line and replicates with high variance as compared to DMSO controls were removed.

180

Table A1-1. Melanoma cell lines assayed in compound screen.

Cell Line SOX10 exp MITF exp WM983B SOX10+ MITF-high SKMEL19 SOX10+ MITF-high RVH421 SOX10+ MITF-high COLO679 SOX10+ MITF-high UACC62 SOX10+ MITF-high A2058 SOX10+ MITF-low SKMEL24 SOX10+ MITF-low HS294T SOX10+ MITF-low WM793 SOX10+ MITF-low A375 SOX10+ MITF-low IGR39 SOX10- -- RPMI7951 SOX10- -- LOXIMVI SOX10- --

To investigate potential differential sensitivity in various classes of cell lines, three metrics were calculated for each class comparison: delta AUC (AUCmean_Class1 – AUCmean_Class2), z-score ([AUC – AUCmean] / SD), and robust z-score ([AUC – AUCmedian] / |AUC –

AUCmedian|median). As a positive control, vemurafenib (BRAF inhibitor) was included in the screen. Cell lines previously known to be sensitive to vemurafenib exhibited increased sensitivity to vemurafenib in this screen. Additionally, MITF+ melanoma cell lines also showed increased sensitivity to vemurafenib, which confirms previous studies performed in our laboratory.

However, there were not enough SOX10- melanoma cell lines to confidently identify compounds with differential sensitivity between SOX10+ and SOX10- melanoma. Intriguingly, MAPK inhibitor-resistant (or MITF-) melanoma cell lines showed an enhanced sensitivity to sirolimus

(mTOR inhibitor) based on AUC values (Figure A1-1). The dose curves of sirolimus-treated cells indicate that sirolimus might not be killing these cells, but may impart a cytostatic effect on

181

MAPK inhibitor-sensitive but not in MAPK inhibitor-resistant melanoma cells. Further validation of these and additional MAPK (dabrafenib, trametinib) and mTOR (everolimus, temsirolimus, KU-0063794, AZD8055) inhibitors over a 28-point dose curve (over 4-log range) confirmed the initial finding that rapamycin-based mTOR inhibitors (sirolimus, everolimus, and temsirolimus) do not kill melanoma cell lines, while melanoma cell lines exhibit sensitivity to

KU-0063794 and AZD8055 regardless of MITF status (Figure A1-2).

Figure A1-1. MITF-low melanoma cell lines are not sensitive to vemurafenib, but are more sensitive to sirolimus compared to MITF-high melanoma cell lines. Cells were treated with small molecules, including vemurafenib and sirolimus, in a compound screen for 4 days. Cell viability was quantified by CellTiter-Glo.

182

Figure A1-2. Rapamycin-based mTOR inhibitors (sirolimus, everolimus, and temsirolimus) do not kill MITF-low melanoma cell lines. Cells were treated with increasing doses of BRAF inhibitor, MEK inhibitor, or mTOR inhibitor for 4 days. Cell viability was quantified by

CellTiter-Glo.

183

APPENDIX 2

Assay Development of an Arrayed ORF Screen for Resistance to Androgen Deprivation in Prostate Cancer

184

INTRODUCTION: PROSTATE CANCER AND FUNCTIONAL GENOMIC SCREENS

Prostate Cancer Epidemiology, Treatment, and Genomics

Prostate cancer is the most frequently diagnosed cancer and second-leading cause of cancer death in men in the United States, with 161,360 estimated new cases and 26,590 estimated deaths in 2017 (Siegel et al., 2017). The majority of prostate cancers are discovered in the local or regional stages, with a 5-year relative survival rate approaching 100%. However, prostate cancers diagnosed in the distant stage (metastasis) have a 5-year relative survival rate of

28%. Well-established risk factors for prostate cancer include increasing age, African ancestry, and a family history of disease. About 60% of all prostate cancer cases are diagnosed in men 65+ years of age, and 97% occur in men 50+. African American men have the highest documented prostate cancer incidence rates in the world. Recent studies suggest that a diet high in processed meat or dairy foods may be a risk factor, and obesity appears to increase risk of aggressive prostate cancer.

For two decades, population-based screening programs for prostate cancer using the prostate-specific antigen (PSA) test has allowed for the early detection of prostate tumors.

Individuals at high risk for prostate cancer (African American men or men with a close relative diagnosed with prostate cancer before age 65) are often advised to begin screening at an earlier age or to be screened more frequently than those at average risk. However, the use of the PSA test has been controversial and recent studies have failed to conclusively indicate whether or not routine PSA screening is beneficial. It is unclear if PSA screening reduces deaths from prostate cancer and screening may also lead to overdiagnosis and subsequent overtreatment, which may cause harm. Advances in detecting prostate cancer and distinguishing between patients who may

185

require treatment compared to those who can undergo active surveillance is urgently needed to reduce the burden of overdiagnosis and overtreatment.

Prostate cancer arises from cells in the prostate gland. The vast majority (95%) of prostate cancers are adenocarcinomas that express the androgen receptor (AR). The most significant other subtype is neuroendocrine prostate cancer, representing < 2% of prostate cancers, which lack AR expression. AR signaling is a critical signaling pathway for the proliferation and survival of prostate cancer cells. The growth of the majority of prostate cancers is driven by male sex hormones, called androgens. Some prostate tumors grow very slowly and may not cause symptoms for years. In other cases, prostate cancer can metastasize to other sites in the body, primarily to bone. Treatment for localized prostate cancer includes surgical excision of the prostate (radical prostatectomy) or radiation therapy. In cases of advanced prostate cancer, these treatments are usually followed with androgen deprivation therapy (ADT) in the form of

LHRH agonists, GnRH antagonists, or non-steroidal anti-androgens (e.g., bicalutamide), which initially will reduce tumor burden and circulating prostate-specific antigen (PSA) to low or undetectable levels. However, the disease will recur in the vast majority of cases as castration- resistant prostate cancer (CRPC) for which there are limited therapeutic options.

Androgen deprivation therapy (ADT) is most commonly used to treat advanced prostate cancer. Most individuals with this diagnosis initially respond very well to these treatments, with tumor shrinkage or slowed tumor growth. Unfortunately, in most cases, the cancers eventually stop responding to ADT and the disease progresses to castration-resistant prostate cancer

(CRPC), which has very poor prognosis and limited treatment options. While the most frequently used ADTs reduce androgen levels, they do not completely eliminate these hormones in the body. The revelation that a more complete blockade of androgen production would be a more

186

effective treatment led to the development of a new anti-androgen therapy, abiraterone (Zytiga), which was approved by the U.S. Food and Drug Administration (FDA) in April 2011 for the treatment of metastatic CRPC due to prolonged overall survival. In addition, the FDA approved enzalutamide (Xtandi, formerly MDV3100) in August 2012 for the treatment of metastatic

CRPC because it significantly prolonged survival. Thus, abiraterone and enzalutamide are new treatment options for patients with metastatic CRPC, but continued research is needed to assess the potential of these therapies as a treatment for earlier stage prostate cancer. Moreover, some metastatic CRPC patients never respond to either abiraterone or enzalutamide, so further research is essential to meet this medical need.

The progression of prostate cancer from normal epithelium to prostatic intraepithelial neoplasia (PIN), adenocarcinoma, and metastasis, and associated genetic, molecular, and cellular changes have been previously described. Alternative genetic pathways of prostate tumorigenesis have been proposed suggesting that prostate cancer is a collection of heterogeneous diseases, each with unique genetic and cellular aberrations that distinguish different subtypes of prostate cancer. The inherent heterogeneity of prostate cancer and the inability of ADT to eradicate all prostate cancer subclones results in the invariable progression to CRPC.

Known mechanisms of resistance to ADT include: 1) reactivation of AR signaling via AR mutation, AR amplification, AR overexpression, AR splice variants, up-regulation of AR nuclear coactivators, and de novo synthesis of androgens; 2) activation of oncogenic signaling pathways;

3) stimulation of anti-apoptotic survival mechanisms; 4) interactions between tumor cells and their microenvironment; and 5) impaired drug delivery to cancer cells. Drug resistance in metastatic CRPC is multifactorial and complex due to the existence of subpopulations of cancer cells (tumor heterogeneity) with unique genetic and cellular milieu within prostate tumors. The

187

ultimate success of prostate cancer treatment will likely emerge from combinatorial therapies targeting multiple pathways of resistance to treatment.

Like all cancers, prostate cancer is a genetic (and epigenetic) disease of the cell that leads to uncontrolled growth that can spread to other parts of the body. In decades past, genetic and molecular studies have identified key genetic regulators of prostate tumorigenesis, including

MYC, TP53, PTEN, and NKX3-1. More recently, aberrations in EZH2, a component of the

Polycomb group complex, and chromosomal rearrangements involving ETS family members

(most notably, TMPRSS2-ERG) have been identified in prostate tumors. With the advent of genome-scale technologies, prostate cancer has been characterized with respect to somatic copy number alterations (SCNAs), point mutations, small insertions/deletions, and structural rearrangements. Recurrent somatic alterations in prostate tumors include mutations in SPOP,

TP53, PTEN, FOXA1, CDKN1B, MED12 and PIK3CA, deletion of PTEN, RB1, TP53, NKX3-1,

CDKN1B and CHD1, amplification of MYC, and genomic rearrangements resulting in ETS- family fusions (e.g., TMPRSS2-ERG). While each of these genetic events have been observed in prostate tumors and may play a role in prostate cancer initiation or progression, it is unknown whether there is a temporal sequence associated with these events or whether there is a causal or cooperative relationship among them. Future structural and functional genomic studies will undoubtedly shed light on the molecular evolution and genetic drivers of prostate cancer.

Identification of Resistance Mechanisms by Functional Genomic Screens

Systematic studies of resistance have been bolstered in recent years with the generation of genome-scale libraries of functional genomic reagents: short-hairpin RNAs (shRNAs) for

188

loss-of-function and open-reading frames (ORFs) for gain-of-function perturbations. These libraries can be cloned into mammalian expression vectors that can be delivered to cells via lentiviral-mediated transduction. Functional screens can be performed with arrayed libraries in short-term high-throughput assays allowing for rapid determination of hits or with pooled libraries in long-term batch assays subsequently deconvoluted by deep sequencing.

Arrayed screens using kinase and genome ORF collections have been successfully performed in our laboratory to identify genes and pathways that can confer resistance to targeted therapies (Johannessen et al., 2010; Johannessen et al., 2013). For example, the CCSB/Broad

Institute kinase ORF collection was used to perform a kinase ORF in A375 BRAF V600E melanoma cells to investigate resistance to a selective BRAF inhibitor, PLX4720. COT

(MAP3K8) was identified as a potent driver of resistance to RAF inhibition in BRAF V600E melanoma cell lines (Johannessen et al., 2010). Additional experiments validated the initial finding and revealed that COT overexpression leads to RAF inhibitor resistance via reactivation of the MAPK pathway. More recently, the CCSB/Broad Institute genome ORF collection was used to systematically elucidate resistance to multiple inhibitors of the MAPK pathway, including RAF, MEK, and ERK inhibitors (Johannessen et al., 2013). These studies revealed a cyclic AMP-dependent melanocytic signaling network not previously associated with drug resistance. The genome ORF collection allowed for the identification of transcription factors, G- protein coupled receptors (GPCRs), and other classes of proteins that could confer resistance to

MAPK pathway inhibitors, including transcription factors downstream of the MAPK and cyclic

AMP pathways (e.g., FOS, ETV1, and MITF). Thus, arrayed kinase and genome ORF screens have the power to identify biologically relevant mechanisms of resistance to cancer therapy.

189

RESULTS

Assay Development for an Arrayed, Genome-Scale ORF Screen for Resistance to

Androgen Deprivation in Prostate Cancer

In anticipation of an arrayed, genome-scale ORF screen to explore resistance mechanisms to androgen deprivation in prostate cancer, we performed optimization experiments in collaboration with the Broad Institute Genetic Perturbation Platform. The objective of the screen is to identify genes whose overexpression is sufficient to promote growth of androgen-dependent prostate cancer cells in the absence of androgens. Androgen depletion will be modeled by substituting fetal bovine serum (FBS) with charcoal-stripped FBS (CSS) in the cell culture growth media. CSS has been treated with activated carbon to remove non-polar, lipophilic molecules, such as hormones, cytokines, and certain growth factors. R1881, a synthetic androgen, will be used as a positive control for androgen stimulation.

Of the seven established prostate cancer cell lines, the LNCaP cell line is the only truly androgen-dependent cell line and thus suitable for the proposed “sufficiency” resistance screen.

When grown in media supplemented with CSS (CSS media), LNCaP cells do not proliferate, but remain viable for an extended period of time (Figure A2-1). R1881 dilution experiments determined that the optimal dose of R1881 to stimulate maximal growth of LNCaP cells in CSS media is 1 pM R1881 (Figure A2-2). Therefore, LNCaP cells and 1 pM R1881 were selected for use in the proposed ORF screen.

190

Figure A2-1. LNCaP cells exhibit reduced proliferation in CSS media. Cells were seeded in

6-well plates and grown in CSS or FBS media. Cells were counted every 7-8 days.

Figure A2-2. 1 pM R1881 is sufficient to promote growth of LNCaP cells in CSS media.

LNCaP cells were grown in 96-well plates in increasing concentrations of R1881. Cell viability was assayed by CellTiter-Glo.

191

The CCSB/Broad Institute’s genome ORF collection will be used to screen ~16,000

ORFs for resistance genes. This genome ORF collection is subcloned into the pLX304 expression vector containing a 3’ V5 tag and a blasticidin resistance gene. The pLX304-ORF-

V5-Blast constructs are packaged in lentiviral particles and arrayed in 384-well plates for ease of high-throughput screening.

Experiments have been performed to determine the optimal amount of polybrene and blasticidin to be used during the infection and selection stages of the assay. Polybrene is a cationic polymer commonly used to increase the efficiency of retroviral and lentiviral infection of mammalian cells. Blasticidin is an antibiotic used to select cells that have been stably transfected with the pLX304-ORF-V5-Blast construct, which includes a blasticidin resistance gene (Blast). Optimization experiments have indicated that 4 μg/ml polybrene is non-toxic to

LNCaP cells and 15 μg/ml blasticidin is sufficient to kill uninfected LNCaP cells (data not shown). Additionally, virus plates containing control ORFs have been utilized to determine the amount of virus from the genome ORF collection to optimally infect cells. Experiments using increasing titers of virus indicated that virus volumes of 3-4 μl resulted in > 85% infection efficiency (data not shown).

To identify robust positive control ORFs, a comprehensive literature and database search for genes that promote androgen-independent growth, interact with AR, or are altered in castration-resistant prostate cancer generated a list of testable genes. ORFs corresponding to these genes were stably transfected in LNCaP cells grown in CSS media and assayed for viability. ORFs for CDK6 and CDK4 promoted androgen-independent growth compared to a control (eGFP) ORF (Figure A2-3).

192

Figure A2-3. CDK6 and CDK4 ORFs promote growth of LNCaP cells in CSS media.

LNCaP cells were seeded in 96-well plates in CSS media and stably transduced with lentivirally- encoded ORFs. Cell viability was assayed by CellTiter-Glo.

Major issues were encountered during assay development, including the relatively small signal generated by the positive control ORFs (CDK6 and CDK4) and differential signal across a single plate (i.e., outer wells have higher signal than inner wells). As a result, the arrayed genome-scale ORF screen for resistance mechanisms to androgen deprivation in prostate cancer was not performed.

193

REFERENCES

Agarwal P, Verzi MP, Nguyen T, Hu J, Ehlers ML, McCulley DJ, Xu SM, Dodou E, Anderson JP, Wei ML, Black BL. The MADS box transcription factor MEF2C regulates melanocyte development and is a direct transcriptional target and partner of SOX10. Development. 2011. 138(12):2555-65.

Aguirre AJ, Meyers RM, Weir BA, Vazquez F, Zhang CZ, Ben-David U, Cook A, Ha G, Harrington WF, Doshi MB, Kost-Alimova M, Gill S, Xu H, Ali LD, Jiang G, Pantel S, Lee Y, Goodale A, Cherniack AD, Oh C, Kryukov G, Cowley GS, Garraway LA, Stegmaier K, Roberts CW, Golub TR, Meyerson M, Root DE, Tsherniak A, Hahn WC. Genomic Copy Number Dictates a Gene-Independent Cell Response to CRISPR/Cas9 Targeting. Cancer Discov. 2016. 6(8):914-29.

American Cancer Society. Cancer Facts & Figures 2017. 2017.

Amos CI, Wang LE, Lee JE, Gershenwald JE, Chen WV, Fang S, Kosoy R, Zhang M, Qureshi AA, Vattathil S, Schacherer CW, Gardner JM, Wang Y, Bishop DT, Barrett JH; GenoMEL Investigators, MacGregor S, Hayward NK, Martin NG, Duffy DL; Q-Mega Investigators, Mann GJ, Cust A, Hopper J; AMFS Investigators, Brown KM, Grimm EA, Xu Y, Han Y, Jing K, McHugh C, Laurie CC, Doheny KF, Pugh EW, Seldin MF, Han J, Wei Q. Genome-wide association study identifies novel loci predisposing to cutaneous melanoma. Hum Mol Genet. 2011. 20(24):5012-23.

Anders S, Huber W. Differential expression analysis for sequence count data. Genome Biol. 2010. 11(10):R106.

Anders S, McCarthy DJ, Chen Y, Okoniewski M, Smyth GK, Huber W, Robinson MD. Count- based differential expression analysis of RNA sequencing data using R and Bioconductor. Nat Protoc. 2013. 8(9):1765-86.

Aoki Y, Saint-Germain N, Gyda M, Magner-Fink E, Lee YH, Credidio C, Saint-Jeannet JP. Sox10 regulates the development of neural crest-derived melanocytes in Xenopus. Dev Biol. 2003. 259(1):19-33.

Aybar MJ, Nieto MA, Mayor R. Snail precedes slug in the genetic cascade required for the specification and migration of the Xenopus neural crest. Development. 2003. 130(3):483-94.

Badis G, Berger MF, Philippakis AA, Talukder S, Gehrke AR, Jaeger SA, Chan ET, Metzler G, Vedenko A, Chen X, Kuznetsov H, Wang CF, Coburn D, Newburger DE, Morris Q, Hughes TR, Bulyk ML. Diversity and complexity in DNA recognition by transcription factors. Science. 2009. 324(5935):1720-3.

Badner JA, Chakravarti A. Waardenburg syndrome and Hirschsprung disease: evidence for pleiotropic effects of a single dominant gene. Am J Med Genet. 1990. 35(1):100-4.

194

Barembaum M, Bronner ME. Identification and dissection of a key mediating cranial neural crest specific expression of transcription factor, Ets-1. Dev Biol. 2013. 382(2):567-75.

Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, Wilson CJ, Lehár J, Kryukov GV, Sonkin D, Reddy A, Liu M, Murray L, Berger MF, Monahan JE, Morais P, Meltzer J, Korejwa A, Jané-Valbuena J, Mapa FA, Thibault J, Bric-Furlong E, Raman P, Shipway A, Engels IH, Cheng J, Yu GK, Yu J, Aspesi P Jr, de Silva M, Jagtap K, Jones MD, Wang L, Hatton C, Palescandolo E, Gupta S, Mahan S, Sougnez C, Onofrio RC, Liefeld T, MacConaill L, Winckler W, Reich M, Li N, Mesirov JP, Gabriel SB, Getz G, Ardlie K, Chan V, Myer VE, Weber BL, Porter J, Warmuth M, Finan P, Harris JL, Meyerson M, Golub TR, Morrissey MP, Sellers WR, Schlegel R, Garraway LA. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature. 2012. 483(7391):603-7.

Barriga EH, Maxwell PH, Reyes AE, Mayor R. The hypoxia factor Hif-1α controls neural crest chemotaxis and epithelial to mesenchymal transition. J Cell Biol. 2013. 201(5):759-76.

Beroukhim R, Getz G, Nghiemphu L, Barretina J, Hsueh T, Linhart D, Vivanco I, Lee JC, Huang JH, Alexander S, Du J, Kau T, Thomas RK, Shah K, Soto H, Perner S, Prensner J, Debiasi RM, Demichelis F, Hatton C, Rubin MA, Garraway LA, Nelson SF, Liau L, Mischel PS, Cloughesy TF, Meyerson M, Golub TA, Lander ES, Mellinghoff IK, Sellers WR. Assessing the significance of chromosomal aberrations in cancer: methodology and application to glioma. Proc Natl Acad Sci U S A. 2007. 104(50):20007-12.

Bhat N, Kwon HJ, Riley BB. A gene network that coordinates preplacodal competence and neural crest specification in zebrafish. Dev Biol. 2013. 373(1):107-17.

Bollag G, Hirth P, Tsai J, Zhang J, Ibrahim PN, Cho H, Spevak W, Zhang C, Zhang Y, Habets G, Burton EA, Wong B, Tsang G, West BL, Powell B, Shellooe R, Marimuthu A, Nguyen H, Zhang KY, Artis DR, Schlessinger J, Su F, Higgins B, Iyer R, D'Andrea K, Koehler A, Stumm M, Lin PS, Lee RJ, Grippo J, Puzanov I, Kim KB, Ribas A, McArthur GA, Sosman JA, Chapman PB, Flaherty KT, Xu X, Nathanson KL, Nolop K. Clinical efficacy of a RAF inhibitor needs broad target blockade in BRAF-mutant melanoma. Nature. 2010. 467(7315):596-9.

Bondurand N, Pingault V, Goerich DE, Lemort N, Sock E, Le Caignec C, Wegner M, Goossens M. Interaction among SOX10, PAX3 and MITF, three genes altered in Waardenburg syndrome. Hum Mol Genet. 2000. 9(13):1907-17.

Bowles J, Schepers G, Koopman P. Phylogeny of the SOX family of developmental transcription factors based on sequence and structural indicators. Dev Biol. 2000. 227(2):239-55.

Cancer Genome Atlas Network. Comprehensive molecular characterization of human colon and rectal cancer. Nature. 2012. 487(7407):330-7.

Cancer Genome Atlas Network. Genomic Classification of Cutaneous Melanoma. Cell. 2015. 161(7): 1681-1696.

195

Cancer Genome Atlas Research Network. Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature. 2008. 455(7216):1061-8.

Cancer Genome Atlas Research Network. Integrated genomic analyses of ovarian carcinoma. Nature. 2011. 474(7353):609-15.

Chapman PB, Hauschild A, Robert C, Haanen JB, Ascierto P, Larkin J, Dummer R, Garbe C, Testori A, Maio M, Hogg D, Lorigan P, Lebbe C, Jouary T, Schadendorf D, Ribas A, O'Day SJ, Sosman JA, Kirkwood JM, Eggermont AM, Dreno B, Nolop K, Li J, Nelson B, Hou J, Lee RJ, Flaherty KT, McArthur GA; BRIM-3 Study Group. Improved survival with vemurafenib in melanoma with BRAF V600E mutation. N Engl J Med. 2011. 364(26):2507-16.

Cheli Y, Ohanna M, Ballotti R, Bertolotto C. Fifteen-year quest for microphthalmia-associated transcription factor target genes. Pigment Cell Melanoma Res. 2010. 23(1):27-40.

Cheung HW, Cowley GS, Weir BA, Boehm JS, Rusin S, Scott JA, East A, Ali LD, Lizotte PH, Wong TC, Jiang G, Hsiao J, Mermel CH, Getz G, Barretina J, Gopal S, Tamayo P, Gould J, Tsherniak A, Stransky N, Luo B, Ren Y, Drapkin R, Bhatia SN, Mesirov JP, Garraway LA, Meyerson M, Lander ES, Root DE, Hahn WC. Systematic investigation of genetic vulnerabilities across cancer cell lines reveals lineage-specific dependencies in ovarian cancer. Proc Natl Acad Sci U S A. 2011. 108(30):12372-7.

Cheung M, Briscoe J. Neural crest development is regulated by the transcription factor Sox9. Development. 2003. 130(23):5681-93.

Cheung M, Chaboissier MC, Mynett A, Hirst E, Schedl A, Briscoe J. The transcriptional control of trunk neural crest induction, survival, and delamination. Dev Cell. 2005. 8(2):179-92.

Cimino-Mathews A, Subhawong AP, Elwood H, Warzecha HN, Sharma R, Park BH, Taube JM, Illei PB, Argani P. Neural crest transcription factor Sox10 is preferentially expressed in triple- negative and metaplastic breast carcinomas. Hum Pathol. 2013. 44(6):959-65.

Cowley GS, Weir BA, Vazquez F, Tamayo P, Scott JA, Rusin S, East-Seletsky A, Ali LD, Gerath WF, Pantel SE, Lizotte PH, Jiang G, Hsiao J, Tsherniak A, Dwinell E, Aoyama S, Okamoto M, Harrington W, Gelfand E, Green TM, Tomko MJ, Gopal S, Wong TC, Li H, Howell S, Stransky N, Liefeld T, Jang D, Bistline J, Hill Meyers B, Armstrong SA, Anderson KC, Stegmaier K, Reich M, Pellman D, Boehm JS, Mesirov JP, Golub TR, Root DE, Hahn WC. Parallel genome-scale loss of function screens in 216 cancer cell lines for the identification of context-specific genetic dependencies. Sci Data. 2014. 1:140035.

Cronin JC, Watkins-Chow DE, Incao A, Hasskamp JH, Schönewolf N, Aoude LG, Hayward NK, Bastian BC, Dummer R, Loftus SK, Pavan WJ. SOX10 ablation arrests cell cycle, induces senescence, and suppresses melanomagenesis. Cancer Res. 2013. 73(18):5709-18.

Davies H, Bignell GR, Cox C, Stephens P, Edkins S, Clegg S, Teague J, Woffendin H, Garnett MJ, Bottomley W, Davis N, Dicks E, Ewing R, Floyd Y, Gray K, Hall S, Hawes R, Hughes J,

196

Kosmidou V, Menzies A, Mould C, Parker A, Stevens C, Watt S, Hooper S, Wilson R, Jayatilake H, Gusterson BA, Cooper C, Shipley J, Hargrave D, Pritchard-Jones K, Maitland N, Chenevix-Trench G, Riggins GJ, Bigner DD, Palmieri G, Cossu A, Flanagan A, Nicholson A, Ho JW, Leung SY, Yuen ST, Weber BL, Seigler HF, Darrow TL, Paterson H, Marais R, Marshall CJ, Wooster R, Stratton MR, Futreal PA. Mutations of the BRAF gene in human cancer. Nature. 2002. 417(6892):949-54.

Denecker G, Vandamme N, Akay O, Koludrovic D, Taminau J, Lemeire K, Gheldof A, De Craene B, Van Gele M, Brochez L, Udupi GM, Rafferty M, Balint B, Gallagher WM, Ghanem G, Huylebroeck D, Haigh J, van den Oord J, Larue L, Davidson I, Marine JC, Berx G. Identification of a ZEB2-MITF-ZEB1 transcriptional network that controls melanogenesis and melanoma progression. Cell Death Differ. 2014. 21(8):1250-61.

Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, Gingeras TR. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013. 29(1):15-21.

Dottori M, Gross MK, Labosky P, Goulding M. The winged-helix transcription factor Foxd3 suppresses interneuron differentiation and promotes neural crest cell fate. Development. 2001. 128(21):4127-38.

Dutton JR, Antonellis A, Carney TJ, Rodrigues FS, Pavan WJ, Ward A, Kelsh RN. An evolutionarily conserved intronic region controls the spatiotemporal expression of the transcription factor Sox10. BMC Dev Biol. 2008. 8:105.

Dutton KA, Pauliny A, Lopes SS, Elworthy S, Carney TJ, Rauch J, Geisler R, Haffter P, Kelsh RN. Zebrafish colourless encodes and specifies non-ectomesenchymal neural crest fates. Development. 2001. 128(21):4113-25.

Feng J, Meyer CA, Wang Q, Liu JS, Shirley Liu X, Zhang Y. GFOLD: a generalized fold change for ranking differentially expressed genes from RNA-seq data. Bioinformatics. 2012. 28(21):2782-8.

Ferronha T, Rabadán MA, Gil-Guiñon E, Le Dréau G, de Torres C, Martí E. LMO4 is an essential in the Snail2-mediated epithelial-to-mesenchymal transition of neuroblastoma and neural crest cells. J Neurosci. 2013. 33(7):2773-83.

Flaherty KT, Infante JR, Daud A, Gonzalez R, Kefford RF, Sosman J, Hamid O, Schuchter L, Cebon J, Ibrahim N, Kudchadkar R, Burris HA 3rd, Falchook G, Algazi A, Lewis K, Long GV, Puzanov I, Lebowitz P, Singh A, Little S, Sun P, Allred A, Ouellet D, Kim KB, Patel K, Weber J. Combined BRAF and MEK inhibition in melanoma with BRAF V600 mutations. N Engl J Med. 2012. 367(18):1694-703.

Fufa TD, Harris ML, Watkins-Chow DE, Levy D, Gorkin DU, Gildea DE, Song L, Safi A, Crawford GE, Sviderskaya EV, Bennett DC, Mccallion AS, Loftus SK, Pavan WJ. Genomic analysis reveals distinct mechanisms and functional classes of SOX10-regulated genes in melanocytes. Hum Mol Genet. 2015. 24(19):5433-50.

197

Garraway LA, Widlund HR, Rubin MA, Getz G, Berger AJ, Ramaswamy S, Beroukhim R, Milner DA, Granter SR, Du J, Lee C, Wagner SN, Li C, Golub TR, Rimm DL, Meyerson ML, Fisher DE, Sellers WR. Integrative genomic analyses identify MITF as a lineage survival oncogene amplified in malignant melanoma. Nature. 2005. 436(7047):117-22.

Girard M, Goossens M. Sumoylation of the SOX10 transcription factor regulates its transcriptional activity. FEBS Lett. 2006. 580(6):1635-41.

GlaxoSmithKline. BRF113683 Clinical Study Report. Data on File, May 2012.

Graf SA, Busch C, Bosserhoff AK, Besch R, Berking C. SOX10 promotes melanoma cell invasion by regulating melanoma inhibitory activity. J Invest Dermatol. 2014. 134(8):2212-2220.

Groves AK, LaBonne C. Setting appropriate boundaries: fate, patterning and competence at the neural plate border. Dev Biol. 2014. 389(1):2-12.

Gubbay J, Collignon J, Koopman P, Capel B, Economou A, Münsterberg A, Vivian N, Goodfellow P, Lovell-Badge R. A gene mapping to the sex-determining region of the mouse Y chromosome is a member of a novel family of embryonically expressed genes. Nature. 1990. 346(6281):245-50.

Hauschild A, Grob JJ, Demidov LV, Jouary T, Gutzmer R, Millward M, Rutkowski P, Blank CU, Miller WH Jr, Kaempgen E, Martín-Algarra S, Karaszewska B, Mauch C, Chiarion-Sileni V, Martin AM, Swann S, Haney P, Mirakhur B, Guckert ME, Goodman V, Chapman PB. Dabrafenib in BRAF-mutated metastatic melanoma: a multicentre, open-label, phase 3 randomised controlled trial. Lancet. 2012. 380(9839):358-65.

He HH, Meyer CA, Shin H, Bailey ST, Wei G, Wang Q, Zhang Y, Xu K, Ni M, Lupien M, Mieczkowski P, Lieb JD, Zhao K, Brown M, Liu XS. Nucleosome dynamics define transcriptional enhancers. Nat Genet. 2010. 42(4):343-7.

Hodges C, Kirkland JG, Crabtree GR. The Many Roles of BAF (mSWI/SNF) and PBAF Complexes in Cancer. Cold Spring Harb Perspect Med. 2016. 6(8).

Hodi FS, O'Day SJ, McDermott DF, Weber RW, Sosman JA, Haanen JB, Gonzalez R, Robert C, Schadendorf D, Hassel JC, Akerley W, van den Eertwegh AJ, Lutzky J, Lorigan P, Vaubel JM, Linette GP, Hogg D, Ottensmeier CH, Lebbé C, Peschel C, Quirt I, Clark JI, Wolchok JD, Weber JS, Tian J, Yellin MJ, Nichol GM, Hoos A, Urba WJ. Improved survival with ipilimumab in patients with metastatic melanoma. N Engl J Med. 2010. 363(8):711-23.

Hodis E, Watson IR, Kryukov GV, Arold ST, Imielinski M, Theurillat JP, Nickerson E, Auclair D, Li L, Place C, Dicara D, Ramos AH, Lawrence MS, Cibulskis K, Sivachenko A, Voet D, Saksena G, Stransky N, Onofrio RC, Winckler W, Ardlie K, Wagle N, Wargo J, Chong K, Morton DL, Stemke-Hale K, Chen G, Noble M, Meyerson M, Ladbury JE, Davies MA,

198

Gershenwald JE, Wagner SN, Hoon DS, Schadendorf D, Lander ES, Gabriel SB, Getz G, Garraway LA, Chin L. A landscape of driver mutations in melanoma. Cell. 2012. 150(2):251-63.

Honoré SM, Aybar MJ, Mayor R. Sox10 is required for the early development of the prospective neural crest in Xenopus embryos. Dev Biol. 2003. 260(1):79-96.

Huang FW, Hodis E, Xu MJ, Kryukov GV, Chin L, Garraway LA. Highly recurrent TERT promoter mutations in human melanoma. Science. 2013. 339(6122):957-9.

Huang YH, Jankowski A, Cheah KS, Prabhakar S, Jauch R. SOXE transcription factors form selective dimers on non-compact DNA motifs through multifaceted interactions between dimerization and high-mobility group domains. Sci Rep. 2015. 5:10398.

Hurtado A, Holmes KA, Ross-Innes CS, Schmidt D, Carroll JS. FOXA1 is a key determinant of estrogen receptor function and endocrine response. Nat Genet. 2011. 43(1):27-33.

Hussussian CJ, Struewing JP, Goldstein AM, Higgins PA, Ally DS, Sheahan MD, Clark WH Jr, Tucker MA, Dracopoli NC. Germline p16 mutations in familial melanoma. Nat Genet. 1994. 8(1):15-21.

Inoue K, Khajavi M, Ohyama T, Hirabayashi S, Wilson J, Reggin JD, Mancias P, Butler IJ, Wilkinson MF, Wegner M, Lupski JR. Molecular mechanism for distinct neurological phenotypes conveyed by allelic truncating mutations. Nat Genet. 2004. 36(4):361-9.

Inoue K, Shilo K, Boerkoel CF, Crowe C, Sawady J, Lupski JR, Agamanolis DP. Congenital hypomyelinating neuropathy, central dysmyelination, and Waardenburg-Hirschsprung disease: phenotypes linked by SOX10 mutation. Ann Neurol. 2002. 52(6):836-42.

Inoue K, Tanabe Y, Lupski JR. deficiencies in both the central and the peripheral nervous systems associated with a SOX10 mutation. Ann Neurol. 1999. 46(3):313-8.

Ivanov SV, Panaccione A, Nonaka D, Prasad ML, Boyd KL, Brown B, Guo Y, Sewell A, Yarbrough WG.0020 Diagnostic SOX10 gene signatures in salivary adenoid cystic and breast basal-like carcinomas. Br J Cancer. 2013. 109(2):444-51.

Jané-Valbuena J, Widlund HR, Perner S, Johnson LA, Dibner AC, Lin WM, Baker AC, Nazarian RM, Vijayendran KG, Sellers WR, Hahn WC, Duncan LM, Rubin MA, Fisher DE, Garraway LA. An oncogenic role for ETV1 in melanoma. Cancer Res. 2010. 70(5):2075-84.

Jensen LE, Whitehead AS. Pellino3, a novel member of the Pellino protein family, promotes activation of c-Jun and Elk-1 and may act as a scaffolding protein. J Immunol. 2003. 171(3):1500-6.

Jiao Z, Mollaaghababa R, Pavan WJ, Antonellis A, Green ED, Hornyak TJ. Direct interaction of Sox10 with the promoter of murine Dopachrome Tautomerase (Dct) and synergistic activation of Dct expression with Mitf. Pigment Cell Res. 2004. 17(4):352-62.

199

Johannessen CM, Boehm JS, Kim SY, Thomas SR, Wardwell L, Johnson LA, Emery CM, Stransky N, Cogdill AP, Barretina J, Caponigro G, Hieronymus H, Murray RR, Salehi-Ashtiani K, Hill DE, Vidal M, Zhao JJ, Yang X, Alkan O, Kim S, Harris JL, Wilson CJ, Myer VE, Finan PM, Root DE, Roberts TM, Golub T, Flaherty KT, Dummer R, Weber BL, Sellers WR, Schlegel R, Wargo JA, Hahn WC, Garraway LA. COT drives resistance to RAF inhibition through MAP kinase pathway reactivation. Nature. 2010. 468(7326):968-72.

Johannessen CM, Johnson LA, Piccioni F, Townes A, Frederick DT, Donahue MK, Narayan R, Flaherty KT, Wargo JA, Root DE, Garraway LA. A melanocyte lineage program confers resistance to MAP kinase pathway inhibition. Nature. 2013. 504(7478):138-42.

June CH, Warshauer JT, Bluestone JA. Is autoimmunity the Achilles' heel of cancer immunotherapy? Nat Med. 2017. 23(5):540-547.

Kadoch C, Crabtree GR. Mammalian SWI/SNF chromatin remodeling complexes and cancer: Mechanistic insights gained from human genomics. Sci Adv. 2015. 1(5):e1500447.

Kamachi Y, Kondoh H. Sox proteins: regulators of cell fate specification and differentiation. Development. 2013. 140(20):4129-44.

Kamachi Y, Uchikawa M, Kondoh H. Pairing SOX off: with partners in the regulation of embryonic development. Trends Genet. 2000. 16(4):182-7.

Kamachi Y, Uchikawa M, Tanouchi A, Sekido R, Kondoh H. Pax6 and SOX2 form a co-DNA- binding partner complex that regulates initiation of lens development. Genes Dev. 2001. 15(10):1272-86.

Kamb A, Shattuck-Eidens D, Eeles R, Liu Q, Gruis NA, Ding W, Hussey C, Tran T, Miki Y, Weaver-Feldhaus J, et al. Analysis of the p16 gene (CDKN2) as a candidate for the chromosome 9p melanoma susceptibility locus. Nat Genet. 1994. 8(1):23-6.

Kapur RP. Early death of neural crest cells is responsible for total enteric aganglionosis in Sox10(Dom)/Sox10(Dom) mouse embryos. Pediatr Dev Pathol. 1999. 2(6):559-69.

Kelsh RN. Sorting out Sox10 functions in neural crest development. Bioessays. 2006. 28(8):788- 98.

Khudyakov J, Bronner-Fraser M. Comprehensive spatiotemporal analysis of early chick neural crest network genes. Dev Dyn. 2009. 238(3):716-23.

Kim J, Lo L, Dormand E, Anderson DJ. SOX10 maintains multipotency and inhibits neuronal differentiation of neural crest stem cells. Neuron. 2003. 38(1):17-31.

Knoechel B, Roderick JE, Williamson KE, Zhu J, Lohr JG, Cotton MJ, Gillespie SM, Fernandez D, Ku M, Wang H, Piccioni F, Silver SJ, Jain M, Pearson D, Kluk MJ, Ott CJ, Shultz LD, Brehm

200

MA, Greiner DL, Gutierrez A, Stegmaier K, Kung AL, Root DE, Bradner JE, Aster JC, Kelliher MA, Bernstein BE. An epigenetic mechanism of resistance to targeted therapy in T cell acute lymphoblastic leukemia. Nat Genet. 2014. 46(4):364-70.

Koinuma D, Tsutsumi S, Kamimura N, Taniguchi H, Miyazawa K, Sunamura M, Imamura T, Miyazono K, Aburatani H. Chromatin immunoprecipitation on microarray analysis of Smad2/3 binding sites reveals roles of ETS1 and TFAP2A in transforming growth factor beta signaling. Mol Cell Biol. 2009. 29(1):172-86.

Kondoh H, Kamachi Y. SOX-partner code for cell specification: Regulatory target selection and underlying molecular mechanisms. Int J Biochem Cell Biol. 2010. 42(3):391-9.

Konieczkowski DJ, Johannessen CM, Abudayyeh O, Kim JW, Cooper ZA, Piris A, Frederick DT, Barzily-Rokni M, Straussman R, Haq R, Fisher DE, Mesirov JP, Hahn WC, Flaherty KT, Wargo JA, Tamayo P, Garraway LA. A melanoma cell state distinction influences sensitivity to MAPK pathway inhibitors. Cancer Discov. 2014. 4(7):816-27.

Kuhlbrodt K, Herbarth B, Sock E, Hermans-Borgmeyer I, Wegner M. Sox10, a novel transcriptional modulator in glial cells. J Neurosci. 1998. 18(1):237-50.

Lamb J, Crawford ED, Peck D, Modell JW, Blat IC, Wrobel MJ, Lerner J, Brunet JP, Subramanian A, Ross KN, Reich M, Hieronymus H, Wei G, Armstrong SA, Haggarty SJ, Clemons PA, Wei R, Carr SA, Lander ES, Golub TR. The Connectivity Map: using gene- expression signatures to connect small molecules, genes, and disease. Science. 2006. 313(5795):1929-35.

Langmead B, Trapnell C, Pop M, Salzberg SL. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 2009. 10(3):R25.

Larkin J, Chiarion-Sileni V, Gonzalez R, Grob JJ, Cowey CL, Lao CD, Schadendorf D, Dummer R, Smylie M, Rutkowski P, Ferrucci PF, Hill A, Wagstaff J, Carlino MS, Haanen JB, Maio M, Marquez-Rodas I, McArthur GA, Ascierto PA, Long GV, Callahan MK, Postow MA, Grossmann K, Sznol M, Dreno B, Bastholt L, Yang A, Rollin LM, Horak C, Hodi FS, Wolchok JD. Combined Nivolumab and Ipilimumab or Monotherapy in Untreated Melanoma. N Engl J Med. 2015. 373(1):23-34.

Lawrence MS, Stojanov P, Polak P, Kryukov GV, Cibulskis K, Sivachenko A, Carter SL, Stewart C, Mermel CH, Roberts SA, Kiezun A, Hammerman PS, McKenna A, Drier Y, Zou L, Ramos AH, Pugh TJ, Stransky N, Helman E, Kim J, Sougnez C, Ambrogio L, Nickerson E, Shefler E, Cortés ML, Auclair D, Saksena G, Voet D, Noble M, DiCara D, Lin P, Lichtenstein L, Heiman DI, Fennell T, Imielinski M, Hernandez B, Hodis E, Baca S, Dulak AM, Lohr J, Landau DA, Wu CJ, Melendez-Zajgla J, Hidalgo-Miranda A, Koren A, McCarroll SA, Mora J, Crompton B, Onofrio R, Parkin M, Winckler W, Ardlie K, Gabriel SB, Roberts CWM, Biegel JA, Stegmaier K, Bass AJ, Garraway LA, Meyerson M, Golub TR, Gordenin DA, Sunyaev S, Lander ES, Getz G. Mutational heterogeneity in cancer and the search for new cancer-associated genes. Nature. 2013. 499(7457):214-218.

201

Lee M, Goodall J, Verastegui C, Ballotti R, Goding CR. Direct regulation of the Microphthalmia promoter by Sox10 links Waardenburg-Shah syndrome (WS4)-associated hypopigmentation and deafness to WS2. J Biol Chem. 2000. 275(48):37978-83.

Liu T, Ortiz JA, Taing L, Meyer CA, Lee B, Zhang Y, Shin H, Wong SS, Ma J, Lei Y, Pape UJ, Poidinger M, Chen Y, Yeung K, Brown M, Turpaz Y, Liu XS. Cistrome: an integrative platform for transcriptional regulation studies. Genome Biol. 2011. 12(8):R83.

Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014. 15(12):550.

Ludwig A, Rehberg S, Wegner M. Melanocyte-specific expression of dopachrome tautomerase is dependent on synergistic gene activation by the Sox10 and Mitf transcription factors. FEBS Lett. 2004. 556(1-3):236-44.

Luo B, Cheung HW, Subramanian A, Sharifnia T, Okamoto M, Yang X, Hinkle G, Boehm JS, Beroukhim R, Weir BA, Mermel C, Barbie DA, Awad T, Zhou X, Nguyen T, Piqani B, Li C, Golub TR, Meyerson M, Hacohen N, Hahn WC, Lander ES, Sabatini DM, Root DE. Highly parallel identification of essential genes in cancer cells. Proc Natl Acad Sci U S A. 2008. 105(51):20380-5.

Macgregor S, Montgomery GW, Liu JZ, Zhao ZZ, Henders AK, Stark M, Schmid H, Holland EA, Duffy DL, Zhang M, Painter JN, Nyholt DR, Maskiell JA, Jetann J, Ferguson M, Cust AE, Jenkins MA, Whiteman DC, Olsson H, Puig S, Bianchi-Scarrà G, Hansson J, Demenais F, Landi MT, Dębniak T, Mackie R, Azizi E, Bressac-de Paillerets B, Goldstein AM, Kanetsky PA, Gruis NA, Elder DE, Newton-Bishop JA, Bishop DT, Iles MM, Helsing P, Amos CI, Wei Q, Wang LE, Lee JE, Qureshi AA, Kefford RF, Giles GG, Armstrong BK, Aitken JF, Han J, Hopper JL, Trent JM, Brown KM, Martin NG, Mann GJ, Hayward NK. Genome-wide association study identifies a new melanoma susceptibility locus at 1q21.3. Nat Genet. 2011. 43(11):1114-8.

McKeown SJ, Lee VM, Bronner-Fraser M, Newgreen DF, Farlie PG. Sox10 overexpression induces neural crest-like cells from all dorsoventral levels of the neural tube but inhibits differentiation. Dev Dyn. 2005. 233(2):430-44.

Messer G, Zemmour J, Orr HT, Parham P, Weiss EH, Girdlestone J. HLA-J, a second inactivated class I HLA gene related to HLA-G and HLA-A. Implications for the evolution of the HLA-A- related genes. J Immunol. 1992. 148(12):4043-53.

Meulemans D, Bronner-Fraser M. Gene-regulatory interactions in neural crest evolution and development. Dev Cell. 2004. 7(3):291-9.

Meyers RM, Bryan JG, McFarland JM, Weir BA, Sizemore AE, Xu H, Dharia NV, Montgomery PG, Cowley GS, Pantel S, Goodale A, Lee Y, Ali LD, Jiang G, Lubonja R, Harrington WF, Strickland M, Wu T, Hawes DC, Zhivich VA, Wyatt MR, Kalani Z, Chang JJ, Okamoto M, Stegmaier K, Golub TR, Boehm JS, Vazquez F, Root DE, Hahn WC, Tsherniak A.

202

Computational correction of copy number effect improves specificity of CRISPR-Cas9 essentiality screens in cancer cells. Nat Genet. 2017. 49(12):1779-1784.

Monsoro-Burq AH, Wang E, Harland R. Msx1 and Pax3 cooperate to mediate FGF8 and WNT signals during Xenopus neural crest induction. Dev Cell. 2005. 8(2):167-78.

Moynagh PN. The roles of Pellino E3 ubiquitin ligases in immunity. Nat Rev Immunol. 2014. 14(2):122-31.

Müller J, Krijgsman O, Tsoi J, Robert L, Hugo W, Song C, Kong X, Possik PA, Cornelissen- Steijger PD, Geukes Foppen MH, Kemper K, Goding CR, McDermott U, Blank C, Haanen J, Graeber TG, Ribas A, Lo RS, Peeper DS. Low MITF/AXL ratio predicts early resistance to multiple targeted drugs in melanoma. Nat Commun. 2014. 5:5712.

Murisier F, Guichard S, Beermann F. The tyrosinase enhancer is activated by Sox10 and Mitf in mouse melanocytes. Pigment Cell Res. 2007. 20(3):173-84.

Nikitina N, Sauka-Spengler T, Bronner-Fraser M. Dissecting early regulatory relationships in the lamprey neural crest gene network. Proc Natl Acad Sci U S A. 2008. 105(51):20083-8.

Parisi MA, Kapur RP. Genetics of Hirschsprung disease. Curr Opin Pediatr. 2000. 12(6):610-7.

Peirano RI, Wegner M. The glial transcription factor Sox10 binds to DNA both as monomer and dimer with different functional consequences. Nucleic Acids Res. 2000. 28(16):3047-55.

Pingault V, Bondurand N, Kuhlbrodt K, Goerich DE, Préhu MO, Puliti A, Herbarth B, Hermans- Borgmeyer I, Legius E, Matthijs G, Amiel J, Lyonnet S, Ceccherini I, Romeo G, Smith JC, Read AP, Wegner M, Goossens M. SOX10 mutations in patients with Waardenburg-Hirschsprung disease. Nat Genet. 1998. 18(2):171-3.

Pingault V, Guiochon-Mantel A, Bondurand N, Faure C, Lacroix C, Lyonnet S, Goossens M, Landrieu P. Peripheral neuropathy with hypomyelination, chronic intestinal pseudo-obstruction and deafness: a developmental "neural crest syndrome" related to a SOX10 mutation. Ann Neurol. 2000. 48(4):671-6.

Pop MS, Stransky N, Garvie CW, Theurillat JP, Hartman EC, Lewis TA, Zhong C, Culyba EK, Lin F, Daniels DS, Pagliarini R, Ronco L, Koehler AN, Garraway LA. A small molecule that binds and inhibits the ETV1 transcription factor oncoprotein. Mol Cancer Ther. 2014. 13(6):1492-502.

Postow MA, Chesney J, Pavlick AC, Robert C, Grossmann K, McDermott D, Linette GP, Meyer N, Giguere JK, Agarwala SS, Shaheen M, Ernstoff MS, Minor D, Salama AK, Taylor M, Ott PA, Rollin LM, Horak C, Gagnier P, Wolchok JD, Hodi FS. Nivolumab and ipilimumab versus ipilimumab in untreated melanoma. N Engl J Med. 2015. 372(21):2006-17.

203

Potterf SB, Furumura M, Dunn KJ, Arnheiter H, Pavan WJ. Transcription factor hierarchy in Waardenburg syndrome: regulation of MITF expression by SOX10 and PAX3. Hum Genet. 2000. 107(1):1-6.

Prasad MK, Reed X, Gorkin DU, Cronin JC, McAdow AR, Chain K, Hodonsky CJ, Jones EA, Svaren J, Antonellis A, Johnson SL, Loftus SK, Pavan WJ, McCallion AS. SOX10 directly modulates ERBB3 transcription via an intronic neural crest enhancer. BMC Dev Biol. 2011. 11:40.

Puntervoll HE, Yang XR, Vetti HH, Bachmann IM, Avril MF, Benfodda M, Catricalà C, Dalle S, Duval-Modeste AB, Ghiorzo P, Grammatico P, Harland M, Hayward NK, Hu HH, Jouary T, Martin-Denavit T, Ozola A, Palmer JM, Pastorino L, Pjanova D, Soufir N, Steine SJ, Stratigos AJ, Thomas L, Tinat J, Tsao H, Veinalde R, Tucker MA, Bressac-de Paillerets B, Newton- Bishop JA, Goldstein AM, Akslen LA, Molven A. Melanoma prone families with CDK4 germline mutation: phenotypic profile and associations with MC1R variants. J Med Genet. 2013. 50(4):264-70.

Qin Q, Mei S, Wu Q, Sun H, Li L, Taing L, Chen S, Li F, Liu T, Zang C, Xu H, Chen Y, Meyer CA, Zhang Y, Brown M, Long HW, Liu XS. ChiLin: a comprehensive ChIP-seq and DNase-seq quality control and analysis pipeline. BMC Bioinformatics. 2016. 17(1):404.

Ragoussis J, Bloemer K, Pohla H, Messer G, Weiss EH, Ziegler A. A physical map including a new class I gene (cda12) of the human major histocompatibility complex (A2/B13 haplotype) derived from a monosomy 6 mutant cell line. Genomics. 1989. 4(3):301-8.

Read AP. Waardenburg syndrome. Adv Otorhinolaryngol. 2000. 56:32-8.

Reményi A, Lins K, Nissen LJ, Reinbold R, Schöler HR, Wilmanns M. Crystal structure of a POU/HMG/DNA ternary complex suggests differential assembly of Oct4 and Sox2 on two enhancers. Genes Dev. 2003. 17(16):2048-59.

Robert C, Karaszewska B, Schachter J, Rutkowski P, Mackiewicz A, Stroiakovski D, Lichinitser M, Dummer R, Grange F, Mortier L, Chiarion-Sileni V, Drucis K, Krajsova I, Hauschild A, Lorigan P, Wolter P, Long GV, Flaherty K, Nathan P, Ribas A, Martin AM, Sun P, Crist W, Legos J, Rubin SD, Little SM, Schadendorf D. Improved overall survival in melanoma with combined dabrafenib and trametinib. N Engl J Med. 2015. 372(1):30-9.

Robert C, Long GV, Brady B, Dutriaux C, Maio M, Mortier L, Hassel JC, Rutkowski P, McNeil C, Kalinka-Warzocha E, Savage KJ, Hernberg MM, Lebbé C, Charles J, Mihalcioiu C, Chiarion- Sileni V, Mauch C, Cognetti F, Arance A, Schmidt H, Schadendorf D, Gogas H, Lundgren- Eriksson L, Horak C, Sharkey B, Waxman IM, Atkinson V, Ascierto PA. Nivolumab in previously untreated melanoma without BRAF mutation. N Engl J Med. 2015. 372(4):320-30.

Robert C, Thomas L, Bondarenko I, O'Day S, Weber J, Garbe C, Lebbe C, Baurain JF, Testori A, Grob JJ, Davidson N, Richards J, Maio M, Hauschild A, Miller WH Jr, Gascon P, Lotem M, Harmankaya K, Ibrahim R, Francis S, Chen TT, Humphrey R, Hoos A, Wolchok JD. Ipilimumab

204

plus dacarbazine for previously untreated metastatic melanoma. N Engl J Med. 2011. 364(26):2517-26.

Robinson JL, Hickey TE, Warren AY, Vowler SL, Carroll T, Lamb AD, Papoutsoglou N, Neal DE, Tilley WD, Carroll JS. Elevated levels of FOXA1 facilitate androgen receptor chromatin binding resulting in a CRPC-like phenotype. Oncogene. 2014. 33(50):5666-74.

Rogers CD, Saxena A, Bronner ME. Sip1 mediates an E-cadherin-to-N-cadherin switch during cranial neural crest EMT. J Cell Biol. 2013. 203(5):835-47.

Sato T, Sasai N, Sasai Y. Neural crest determination by co-activation of Pax3 and Zic1 genes in Xenopus ectoderm. Development. 2005. 132(10):2355-63.

Schepers GE, Teasdale RD, Koopman P. Twenty pairs of sox: extent, homology, and nomenclature of the mouse and human sox transcription factor gene families. Dev Cell. 2002. 3(2):167-70.

Schlierf B, Ludwig A, Klenovsek K, Wegner M. Cooperative binding of Sox10 to DNA: requirements and consequences. Nucleic Acids Res. 2002. 30(24):5509-16.

Shah KN, Dalal SJ, Desai MP, Sheth PN, Joshi NC, Ambani LM. White forelock, pigmentary disorder of irides, and long segment Hirschsprung disease: possible variant of Waardenburg syndrome. J Pediatr. 1981. 99(3):432-5.

Shakhova O, Cheng P, Mishra PJ, Zingg D, Schaefer SM, Debbache J, Häusel J, Matter C, Guo T, Davis S, Meltzer P, Mihic-Probst D, Moch H, Wegner M, Merlino G, Levesque MP, Dummer R, Santoro R, Cinelli P, Sommer L. Antagonistic cross-regulation between Sox9 and Sox10 controls an anti-tumorigenic program in melanoma. PLoS Genet. 2015. 11(1):e1004877.

Shakhova O, Zingg D, Schaefer SM, Hari L, Civenni G, Blunschi J, Claudinot S, Okoniewski M, Beermann F, Mihic-Probst D, Moch H, Wegner M, Dummer R, Barrandon Y, Cinelli P, Sommer L. Sox10 promotes the formation and maintenance of giant congenital naevi and melanoma. Nat Cell Biol. 2012. 14(8):882-90.

Shao DD, Tsherniak A, Gopal S, Weir BA, Tamayo P, Stransky N, Schumacher SE, Zack TI, Beroukhim R, Garraway LA, Margolin AA, Root DE, Hahn WC, Mesirov JP. ATARiS: computational quantification of gene suppression phenotypes from multisample RNAi screens. Genome Res. 2013. 23(4):665-78.

Shekar SN, Duffy DL, Youl P, Baxter AJ, Kvaskoff M, Whiteman DC, Green AC, Hughes MC, Hayward NK, Coates M, Martin NG. A population-based study of Australian twins with melanoma suggests a strong genetic contribution to liability. J Invest Dermatol. 2009. 129(9):2211-9.

Shin H, Liu T, Manrai AK, Liu XS. CEAS: cis-regulatory element annotation system. Bioinformatics. 2009. 25(19):2605-6.

205

Siegel RL, Miller KD, Jemal A. Cancer Statistics, 2017. CA Cancer J Clin. 2017. 67(1):7-30.

Siepel A, Bejerano G, Pedersen JS, Hinrichs AS, Hou M, Rosenbloom K, Clawson H, Spieth J, Hillier LW, Richards S, Weinstock GM, Wilson RK, Gibbs RA, Kent WJ, Miller W, Haussler D. Evolutionarily conserved elements in vertebrate, insect, worm, and yeast genomes. Genome Res. 2005. 15(8):1034-50.

Simões-Costa M, Bronner ME. Establishing neural crest identity: a gene regulatory recipe. Development. 2015. 142(2):242-57.

Simões-Costa MS, McKeown SJ, Tan-Cabugao J, Sauka-Spengler T, Bronner ME. Dynamic and differential regulation of stem cell factor FoxD3 in the neural crest is encrypted in the genome. PLoS Genet. 2012. 8(12):e1003142.

Sinclair AH, Berta P, Palmer MS, Hawkins JR, Griffiths BL, Smith MJ, Foster JW, Frischauf AM, Lovell-Badge R, Goodfellow PN. A gene from the human sex-determining region encodes a protein with homology to a conserved DNA-binding motif. Nature. 1990. 346(6281):240-4.

Snyder A, Makarov V, Merghoub T, Yuan J, Zaretsky JM, Desrichard A, Walsh LA, Postow MA, Wong P, Ho TS, Hollmann TJ, Bruggeman C, Kannan K, Li Y, Elipenahli C, Liu C, Harbison CT, Wang L, Ribas A, Wolchok JD, Chan TA. Genetic basis for clinical response to CTLA-4 blockade in melanoma. N Engl J Med. 2014. 371(23):2189-2199.

Song C, Zhang S, Huang H. Choosing a suitable method for the identification of replication origins in microbial genomes. Front Microbiol. 2015. 6:1049.

Soufir N, Avril MF, Chompret A, Demenais F, Bombled J, Spatz A, Stoppa-Lyonnet D, Bénard J, Bressac-de Paillerets B. Prevalence of p16 and CDK4 germline mutations in 48 melanoma- prone families in France. The French Familial Melanoma Study Group. Hum Mol Genet. 1998. 7(2):209-16.

Southard-Smith EM, Kos L, Pavan WJ. Sox10 mutation disrupts neural crest development in Dom Hirschsprung mouse model. Nat Genet. 1998. 18(1):60-4.

Steingrímsson E, Copeland NG, Jenkins NA. Melanocytes and the microphthalmia transcription factor network. Annu Rev Genet. 2004. 38:365-411.

Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP. Gene set enrichment analysis: a knowledge- based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005. 102(43):15545-50.

Sun C, Wang L, Huang S, Heynen GJ, Prahallad A, Robert C, Haanen J, Blank C, Wesseling J, Willems SM, Zecchin D, Hobor S, Bajpe PK, Lieftink C, Mateus C, Vagner S, Grernrum W, Hofland I, Schlicker A, Wessels LF, Beijersbergen RL, Bardelli A, Di Nicolantonio F,

206

Eggermont AM, Bernards R. Reversible and adaptive resistance to BRAF(V600E) inhibition in melanoma. Nature. 2014. 508(7494):118-22.

Trapnell C, Roberts A, Goff L, Pertea G, Kim D, Kelley DR, Pimentel H, Salzberg SL, Rinn JL, Pachter L. Differential gene and transcript expression analysis of RNA-seq experiments with TopHat and Cufflinks. Nat Protoc. 2012. 7(3):562-78.

Tsherniak A, Vazquez F, Montgomery PG, Weir BA, Kryukov G, Cowley GS, Gill S, Harrington WF, Pantel S, Krill-Burger JM, Meyers RM, Ali L, Goodale A, Lee Y, Jiang G, Hsiao J, Gerath WFJ, Howell S, Merkel E, Ghandi M, Garraway LA, Root DE, Golub TR, Boehm JS, Hahn WC. Defining a Cancer Dependency Map. Cell. 2017. 170(3):564-576.

Van Allen EM, Miao D, Schilling B, Shukla SA, Blank C, Zimmer L, Sucker A, Hillen U, Foppen MHG, Goldinger SM, Utikal J, Hassel JC, Weide B, Kaehler KC, Loquai C, Mohr P, Gutzmer R, Dummer R, Gabriel S, Wu CJ, Schadendorf D, Garraway LA. Genomic correlates of response to CTLA-4 blockade in metastatic melanoma. Science. 2015. 350(6257):207-211.

Van Allen EM, Wagle N, Sucker A, Treacy DJ, Johannessen CM, Goetz EM, Place CS, Taylor- Weiner A, Whittaker S, Kryukov GV, Hodis E, Rosenberg M, McKenna A, Cibulskis K, Farlow D, Zimmer L, Hillen U, Gutzmer R, Goldinger SM, Ugurel S, Gogas HJ, Egberts F, Berking C, Trefzer U, Loquai C, Weide B, Hassel JC, Gabriel SB, Carter SL, Getz G, Garraway LA, Schadendorf D; Dermatologic Cooperative Oncology Group of Germany (DeCOG). The genetic landscape of clinical resistance to RAF inhibition in metastatic melanoma. Cancer Discov. 2014. 4(1):94-109. van 't Veer LJ, Burgering BM, Versteeg R, Boot AJ, Ruiter DJ, Osanto S, Schrier PI, Bos JL. N- ras mutations in human cutaneous melanoma from sun-exposed body sites. Mol Cell Biol. 1989. 9(7):3114-6.

Vance KW, Goding CR. The transcription network regulating melanocyte development and melanoma. Pigment Cell Res. 2004. 17(4):318-25.

Verastegui C, Bille K, Ortonne JP, Ballotti R. Regulation of the microphthalmia-associated transcription factor gene by the Waardenburg syndrome type 4 gene, SOX10. J Biol Chem. 2000. 275(40):30757-60.

Verfaillie A, Imrichova H, Atak ZK, Dewaele M, Rambow F, Hulselmans G, Christiaens V, Svetlichnyy D, Luciani F, Van den Mooter L, Claerhout S, Fiers M, Journe F, Ghanem GE, Herrmann C, Halder G, Marine JC, Aerts S. Decoding the regulatory landscape of melanoma reveals TEADS as regulators of the invasive cell state. Nat Commun. 2015. 6:6683.

Wagle N, Emery C, Berger MF, Davis MJ, Sawyer A, Pochanard P, Kehoe SM, Johannessen CM, Macconaill LE, Hahn WC, Meyerson M, Garraway LA. Dissecting therapeutic resistance to RAF inhibition in melanoma by tumor genomic profiling. J Clin Oncol. 2011. 29(22):3085-96.

207

Wagle N, Van Allen EM, Treacy DJ, Frederick DT, Cooper ZA, Taylor-Weiner A, Rosenberg M, Goetz EM, Sullivan RJ, Farlow DN, Friedrich DC, Anderka K, Perrin D, Johannessen CM, McKenna A, Cibulskis K, Kryukov G, Hodis E, Lawrence DP, Fisher S, Getz G, Gabriel SB, Carter SL, Flaherty KT, Wargo JA, Garraway LA. MAP kinase pathway alterations in BRAF- mutant melanoma patients with acquired resistance to combined RAF/MEK inhibition. Cancer Discov. 2014. 4(1):61-8.

Wahlbuhl M, Reiprich S, Vogl MR, Bösl MR, Wegner M. Transcription factor Sox10 orchestrates activity of a neural crest-specific enhancer in the vicinity of its gene. Nucleic Acids Res. 2012. 40(1):88-101.

Wang S, Sun H, Ma J, Zang C, Wang C, Wang J, Tang Q, Meyer CA, Zhang Y, Liu XS. Target analysis by integration of transcriptome and ChIP-seq data with BETA. Nat Protoc. 2013. 8(12):2502-15.

Watanabe A, Takeda K, Ploplis B, Tachibana M. Epistatic relationship between Waardenburg syndrome genes MITF and PAX3. Nat Genet. 1998. 18(3):283-6.

Wellbrock C, Arozarena I. Microphthalmia-associated transcription factor in melanoma development and MAP-kinase pathway targeted therapy. Pigment Cell Melanoma Res. 2015. 28(4):390-406.

Werner T, Hammer A, Wahlbuhl M, Bösl MR, Wegner M. Multiple conserved regulatory elements with overlapping functions determine Sox10 expression in mouse embryogenesis. Nucleic Acids Res. 2007. 35(19):6526-38.

Whyte WA, Orlando DA, Hnisz D, Abraham BJ, Lin CY, Kagey MH, Rahl PB, Lee TI, Young RA. Master transcription factors and mediator establish super-enhancers at key cell identity genes. Cell. 2013. 153(2):307-19.

Wolchok JD, Chiarion-Sileni V, Gonzalez R, Rutkowski P, Grob JJ, Cowey CL, Lao CD, Wagstaff J, Schadendorf D, Ferrucci PF, Smylie M, Dummer R, Hill A, Hogg D, Haanen J, Carlino MS, Bechter O, Maio M, Marquez-Rodas I, Guidoboni M, McArthur G, Lebbé C, Ascierto PA, Long GV, Cebon J, Sosman J, Postow MA, Callahan MK, Walker D, Rollin L, Bhore R, Hodi FS, Larkin J. Overall Survival with Combined Nivolumab and Ipilimumab in Advanced Melanoma. N Engl J Med. 2017. 377(14):1345-1356.

Wolchok JD, Kluger H, Callahan MK, Postow MA, Rizvi NA, Lesokhin AM, Segal NH, Ariyan CE, Gordon RA, Reed K, Burke MM, Caldwell A, Kronenberg SA, Agunwamba BU, Zhang X, Lowy I, Inzunza HD, Feely W, Horak CE, Hong Q, Korman AJ, Wigginton JM, Gupta A, Sznol M. Nivolumab plus ipilimumab in advanced melanoma. N Engl J Med. 2013. 369(2):122-33.

Wood K, Luke J. The Biology and Therapeutic Approach to BRAF-Mutant Cutaneous Melanoma. Am J Hematol Oncol. 2017. 13(1):4-10.

208

Yokoyama S, Feige E, Poling LL, Levy C, Widlund HR, Khaled M, Kung AL, Fisher DE. Pharmacologic suppression of MITF expression via HDAC inhibitors in the melanocyte lineage. Pigment Cell Melanoma Res. 2008. 21(4):457-63.

Yuan H, Corbi N, Basilico C, Dailey L. Developmental-specific activity of the FGF-4 enhancer requires the synergistic action of Sox2 and Oct-3. Genes Dev. 1995. 9(21):2635-45.

Zhang Y, Liu T, Meyer CA, Eeckhoute J, Johnson DS, Bernstein BE, Nusbaum C, Myers RM, Brown M, Li W, Liu XS. Model-based analysis of ChIP-Seq (MACS). Genome Biol. 2008. 9(9):R137.

Zheng H, Ying H, Yan H, Kimmelman AC, Hiller DJ, Chen AJ, Perry SR, Tonon G, Chu GC, Ding Z, Stommel JM, Dunn KL, Wiedemeyer R, You MJ, Brennan C, Wang YA, Ligon KL, Wong WH, Chin L, DePinho RA. p53 and Pten control neural and glioma stem/progenitor cell renewal and differentiation. Nature. 2008. 455(7216):1129-33.

Zingg D, Debbache J, Schaefer SM, Tuncer E, Frommel SC, Cheng P, Arenas-Ramirez N, Haeusel J, Zhang Y, Bonalli M, McCabe MT, Creasy CL, Levesque MP, Boyman O, Santoro R, Shakhova O, Dummer R, Sommer L. The epigenetic modifier EZH2 controls melanoma growth and metastasis through silencing of distinct tumour suppressors. Nat Commun. 2015. 6:6051.

Zuo L, Weger J, Yang Q, Goldstein AM, Tucker MA, Walker GJ, Hayward N, Dracopoli NC. Germline mutations in the p16INK4a binding domain of CDK4 in familial melanoma. Nat Genet. 1996. 12(1):97-9.

209