Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Molecular mechanisms underlying tumorigenesis of non-small cell lung cancer subtypes Pikor, Larissa A. 2014

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2014_september_pikor_larissa.pdf [ 5.6MB ]
Metadata
JSON: 24-1.0167346.json
JSON-LD: 24-1.0167346-ld.json
RDF/XML (Pretty): 24-1.0167346-rdf.xml
RDF/JSON: 24-1.0167346-rdf.json
Turtle: 24-1.0167346-turtle.txt
N-Triples: 24-1.0167346-rdf-ntriples.txt
Original Record: 24-1.0167346-source.json
Full Text
24-1.0167346-fulltext.txt
Citation
24-1.0167346.ris

Full Text

MOLECULAR MECHANISMS UNDERLYING TUMORIGENESIS OF NON-SMALL CELL LUNG CANCER SUBTYPES  by  Larissa A. Pikor  BScH., BPHE, Queen's University at Kingston, 2009  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY  in  THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES  (Interdisciplinary Oncology)  THE UNIVERSITY OF BRITISH COLUMBIA  (Vancouver)     July 2014  © Larissa Pikor, 2014 ii  Abstract  Lung cancer is the leading cause of cancer death in Canada and worldwide. A late stage of diagnosis in conjunction with a lack of effective treatment options are largely responsible for the poor survival rates of lung cancer. Histological subtypes of lung cancer are known to respond differently to standard therapies, suggesting they are distinct diseases. A better understanding of the molecular alterations and underlying biology of lung cancer subtypes is therefore necessary for the development of novel detection and treatment strategies in order to improve patient prognosis. We hypothesize that lung adenocarcinoma (AC) and squamous cell carcinoma (SqCC) arise through disparate patterns of molecular alterations and that these differences underlie unique biological mechanisms that contribute to subtype development, phenotypes and response to therapy.    In this thesis, I apply multidimensional integrative 'omics approaches to characterize the genomic and epigenomic landscapes of AC and SqCC and elucidate differential patterns of alterations and subtype specific gene disruptions causal to NSCLC and the development of specific subtypes. The integration of DNA copy number, methylation, gene and miRNA expression data on AC and SqCC tumors and patient matched non-malignant tissue identified several subtype specific alterations and revealed unique oncogenic pathways associated with AC and SqCC that can be successfully targeted by existing therapies. By combining genomic analyses with manipulation of candidate genes in vitro and in vivo, we validated the contribution of candidate genes to tumorigenesis and determined the mechanisms through which they contribute to disease pathogenesis. In addition to revealing differentially disrupted genes and pathways we also identified numerous alterations common to both subtypes.    Collectively, this work has further characterized the landscape of molecular alterations that define AC and SqCC, and the mechanisms through which these alterations contribute to subtype tumorigenesis. This work has identified novel candidate genes involved in subtype tumorigenesis as well as miRNAs with potential as diagnostic biomarkers for lung cancer. Taken together, these findings underscore the importance of tailoring treatment strategies to the histological subtype based on the underlying biology of that subtype. iii  Preface  The research in this thesis was conducted with ethics approval from the UBC Research Ethics Board, Certificate Numbers: EDRN H09-00008, CCSRI H09-00934, W81XW-10-1-0634  Portions of Chapter 1 have been published as: 1. [Pikor LA], Ramnarine VR, Lam S, Lam WL (2013). Genetic alterations defining NSCLC subtypes and their therapeutic implications. Lung Cancer 82(2):179-89. 2. Enfield KSS, [Pikor LA], Martinez VD, Lam WL (2012). Mechanistic roles of non-coding RNAs in lung cancer biology and their clinical implications. Genetics Research International 2012:737416   Chapters 2,3 and 4 were co-authored as manuscripts for publication. The following author lists apply for these chapters:  A version of Chapter 2 has been published and presented by Will Lockwood in his thesis. Lockwood WW, Wilson IM, Coe BP, Chari R, [Pikor LA], Thu KL, Yee J, English J, Murray N, Tsao MS, Minna J, Gazdar AF, MacAulay CE, Lam S, Lam WL (2012). Divergent genomic and epigenomic landscapes of lung cancer subtypes underscore the selection of different oncogenic pathways during tumor development. PLoS One 2012;7(5). This chapter presents the necessary description of datasets (genomic, epigenomic and expression) used in subsequent chapters to investigate the molecular biology of lung cancer subtypes.  As this chapter is quite detailed, it was included as a data chapter rather than integrated into the introduction.  I conducted and analyzed dose response experiments, performed SNP6 data analysis and wrote sections of the manuscript pertaining to the experiments I performed.  A version of Chapter 3 has been published. Thu KL, [Pikor LA], Chari R, Wilson IM, MacAulay CE, Gazdar AF, Lam S, Lam WL, Lockwood WW (2011) Disruption of KEAP1/CUL3 E3 ubiquitin ligase complex components is a key mechanism for NFkB pathway activation in lung cancer. Journal of Thoracic Oncology 6(9):1521-9. I was co-first author of this manuscript and iv  participated in the design of the study, performed experiments, data analysis and interpretation of the results, and wrote and edited the manuscript. R Chari and IM Wilson contributed to project design and data analysis. Patient samples were collected by S. Lam and cell lines provided by AF Gazdar. WL Lam and WW Lockwood oversaw the project.  A version of Chapter 4 has been published. [Pikor LA],  Lockwood WW, Thu KL, Vucic EA, Chari R, Gazdar AF, Lam S, Lam WL (2013). YEATS4 is a novel oncogene amplified in non-small cell lung cancer that regulates the p53 pathway. Cancer Research 73(24):7301-12. I designed the study, conducted the experiments, performed data analysis and interpretation and wrote the manuscript. Patient samples were collected by S. Lam, cell lines were provided by AF Gazdar and mouse work was performed by WW Lockwood. KL Thu, EA Vucic and R Chari aided in data analysis and interpretation of results.  A version of Chapter 5 is being prepared for publication. [Pikor LA], Kennett JY, Thu KL, Vucic EA, Lam S, Lam WL (2014). Characterizing miRNA expression in lung adenocarcinoma and squamous cell carcinoma identifies miR-944 and miR-1287 as subtype specific miRNAs. I conceived the experimental design for this project, performed data analysis and interpretation, in vitro experiments and wrote the manuscript. JY Kennett provided flow cytometry technical assistance and KL Thu and EA Vucic aided in data analysis, project design and edited the manuscript.     v  Table of Contents Abstract .......................................................................................................................................... ii Preface ........................................................................................................................................... iii Table of Contents ...........................................................................................................................v List of Tables ................................................................................................................................ xi List of Figures .............................................................................................................................. xii List of Abbreviations ................................................................................................................. xiii Acknowledgements ......................................................................................................................xv Dedication ................................................................................................................................... xvi Chapter 1: Introduction ................................................................................................................1 1.1 Lung cancer ..................................................................................................................... 1 1.2 Lung cancer subtypes/ histopathology of NSCLC.......................................................... 1 1.3 Molecular pathology of lung cancer ............................................................................... 3 1.3.1 Gene expression .......................................................................................................... 3 1.3.2 Copy number alterations ............................................................................................. 4 1.3.3 DNA methylation ........................................................................................................ 4 1.3.4 DNA mutations ........................................................................................................... 5 1.3.5 Non-coding RNAs ...................................................................................................... 5 1.4 Integrative genomics and a systems biology approach to gene discovery ...................... 6 1.5 Thesis theme and rationale.............................................................................................. 7 1.6 Objectives and hypotheses .............................................................................................. 8 1.7 Specific aims and thesis outline ...................................................................................... 8 Chapter 2: Divergent genomic and epigenomic landscapes underscore the selection of different oncogenic pathways during subtype development ....................................................11 2.1 Introduction ................................................................................................................... 11 2.2 Methods......................................................................................................................... 12 2.2.1 DNA samples ............................................................................................................ 12 2.2.2 Tiling path array comparative genomic hybridization .............................................. 13 2.2.3 DNA methylation analysis ........................................................................................ 13 2.2.4 Comparison of subtype alteration frequencies .......................................................... 13 vi  2.2.5 Gene expression microarray analysis........................................................................ 14 2.2.6 Statistical analysis of gene expression data .............................................................. 15 2.2.7 Survival analysis ....................................................................................................... 16 2.2.8 Network identification .............................................................................................. 16 2.2.9 Human lung tissue microarray case selection ........................................................... 16 2.2.10 Immunohistochemical analysis ............................................................................. 17 2.2.11 Trichostatin A dose-response analysis .................................................................. 17 2.3 Results ........................................................................................................................... 18 2.3.1 Assessment of global genomic instability in AC and SqCC..................................... 18 2.3.2 Disparate genetic and epigenetic landscapes characterize lung SqCC and AC ........ 21 2.3.3 Gene disruptions are selected in a subtype specific manner ..................................... 23 2.3.4 Different oncogenic pathways are associated with AC & SqCC .............................. 24 2.3.5 Epigenetically regulated genes complement genetically regulated genes ................ 24 2.3.6 Concerted genetic and epigenetic disruption of subtype- specific genes ................. 29 2.3.7 Subtype specific genes are responsible for AC and SqCC phenotypes .................... 31 2.3.8 Subtype-specific genetic differences are translated to the protein level ................... 33 2.3.9 Subtype-specific genes are associated with distinct clinical characteristics ............. 35 2.3.10 Defining putative treatment strategies tailored to lung cancer subtypes using in silico screening of candidate therapeutic compounds ........................................................... 44 2.4 Discussion ..................................................................................................................... 47 2.5 Conclusions ................................................................................................................... 51 Chapter 3: Genetic disruption of KEAP1/CUL3 E3 ubiquitin ligase complex components is a key mechanisms of NF-κB pathway activation in lung cancer .............................................52 3.1 Introduction ................................................................................................................... 52 3.2 Methods......................................................................................................................... 55 3.2.1 Non-small cell lung cancer (NSCLC) samples ......................................................... 55 3.2.2 Determination of gene dosage and expression levels ............................................... 55 3.2.3 Copy number analysis of external cohorts ................................................................ 56 3.2.4 Multi-dimensional analysis of complex component disruption in the TCGA .......... 56 vii  3.2.5 Western blot assessment of total and phospho- IKBKB and total and phospho- NF-κB protein levels ................................................................................................................... 57 3.2.6 Immunohistochemistry (IHC) staining for IKBKB protein levels ........................... 57 3.2.7 siRNA-mediated complex component knockdowns ................................................. 58 3.2.8 NF-κB target gene analysis ....................................................................................... 58 3.2.9 IKBKB inhibition in NSCLC cell lines .................................................................... 59 3.3 Results ........................................................................................................................... 60 3.3.1 Disruption of E3-ubiquitin ligase complex components in NSCLC ........................ 60 3.3.2 Complex components are differentially altered in AC and SqCC subtypes ............. 63 3.3.3 Other genetic and epigenetic mechanisms of complex component disruption ......... 64 3.3.4 Functional consequences of genetic complex disruption.......................................... 65 3.3.5 Pharmacological inhibition of IKBKB ..................................................................... 68 3.4 Discussion ..................................................................................................................... 70 3.5 Conclusions ................................................................................................................... 73 Chapter 4: YEATS4 is a novel oncogene amplified in non-small cell lung cancer that regulates the p53 pathway ...........................................................................................................74 4.1 Introduction ................................................................................................................... 74 4.2 Methods......................................................................................................................... 75 4.2.1 NSCLC  tumor samples and cell lines ...................................................................... 75 4.2.2 Array comparative genomic hybridization and GISTIC analysis ............................. 76 4.2.3 Gene expression profiling and data integration ........................................................ 76 4.2.4 Analysis of external cohorts...................................................................................... 77 4.2.5 Plasmids and generation of stably expressing YEATS4 lines .................................. 77 4.2.6 Quantitative reverse transcriptase PCR .................................................................... 78 4.2.7 siRNA mediated knockdown of CDKN1A............................................................... 79 4.2.8 Western blot .............................................................................................................. 79 4.2.9 Cell viability.............................................................................................................. 79 4.2.10 Colony formation .................................................................................................. 80 4.2.11 Cellular senescence ............................................................................................... 80 4.2.12 Dose response assay .............................................................................................. 80 viii  4.2.13 Mouse models ....................................................................................................... 81 4.2.14 Pathway analysis and gene set enrichment ........................................................... 81 4.2.15 Survival analysis ................................................................................................... 81 4.3 Results ........................................................................................................................... 82 4.3.1 Recurrently amplified regions in NSCLC................................................................. 82 4.3.2 Identification of YEATS4, the target of 12q15 amplification .................................. 84 4.3.3 YEATS4 is frequently amplified and overexpressed in NSCLC.............................. 84 4.3.4 YEATS4 displays oncogenic properties in vitro and in vivo .................................... 88 4.3.5 YEATS4 suppresses p53 and p21 ............................................................................. 92 4.3.6 YEATS4 alters the sensitivity of cell lines to cisplatin and nutlin ........................... 92 4.3.7 Sensitivity to cisplatin is not mediated solely through the p53-p21 pathway ........... 96 4.3.8 Knockdown phenotypes are independent of p21 signaling in mutant p53 cells ....... 96 4.3.9 Identification of additional cellular networks regulated by YEATS4 ...................... 98 4.4 Discussion ................................................................................................................... 101 Chapter 5: Characterizing miRNA expression in lung adenocarcinoma and squamous cell carcinoma identifies miR-944 and miR-1287 as subtype specific miRNAs ..........................104 5.1 Introduction ................................................................................................................. 104 5.2 Methods....................................................................................................................... 106 5.2.1 NSCLC patients and samples.................................................................................. 106 5.2.2 MicroRNA sequencing and data analysis ............................................................... 106 5.2.3 Identification of subtype specific miRNAs and NSCLC miRNAs ......................... 106 5.2.4 Statistical analysis of miRNA expression data ....................................................... 107 5.2.5 Validation in external cohorts ................................................................................. 108 5.2.6 Integration of DNA copy number and methylation with miRNA expression ........ 108 5.2.7 Target prediction, interaction networks and pathway analysis ............................... 108 5.2.8 Cell culture and manipulation of miR-1287 ........................................................... 109 5.2.9 3'-UTR reporter assays ............................................................................................ 109 5.2.10 qPCR of miRNA expression ............................................................................... 109 5.2.11 Western blotting .................................................................................................. 110 5.2.12 MTT assay .......................................................................................................... 110 ix  5.2.13 Apoptosis, DNA damage and cell proliferation .................................................. 110 5.3 Results ......................................................................................................................... 111 5.3.1 MiRNAs are differentially expressed between NSCLC tumors and non-malignant tissue and accurately segregate profiles based on malignancy. .......................................... 111 5.3.2 Lung AC and SqCC display different patterns of miRNA expression ................... 114 5.3.3 Validation of differentially expressed miRNAs ..................................................... 117 5.3.4 Copy number alterations contribute to aberrant miRNA expression ...................... 119 5.3.5 Predicted targets of subtype miRNAs are associated with distinct pathways ........ 119 5.3.6 Lung cancer prognostic genes are targeted by NSCLC miRNAs ........................... 120 5.3.7 MiR-944 overexpression is highly specific to SqCC .............................................. 124 5.3.8 MiR-1287 inhibits expression of RAD9A by binding its 3'-UTR .......................... 126 5.3.9 Overexpression of miR-1287 enables proliferation of damaged cells in LCCL .... 126 5.4 Discussion ................................................................................................................... 128 Chapter 6: Conclusion ...............................................................................................................133 6.1 Summary ..................................................................................................................... 133 6.1.1 Identify molecular alterations common or specific to lung cancer subtypes. ......... 133 6.1.2 Delineate the biological significance and clinical relevance of alterations. ........... 134 6.2 Conclusions ................................................................................................................. 135 6.3 Significance and clinical implications ........................................................................ 136 6.4 Future directions ......................................................................................................... 137 6.4.1 Mechanisms of YEATS4 mediated tumorigenesis ................................................. 138 6.4.2 Validation of miRNAs as biomarkers for the detection of lung cancer .................. 138 6.4.3 Contribution of miR-944 to SqCC tumorigenesis .................................................. 139 6.4.4 Global assessment of histone modifications in SqCC ............................................ 139 6.4.5 Integration of molecular and immunologic alterations to understand the contribution of the tumor microenvironment to tumorigenesis. .............................................................. 140 References ...................................................................................................................................141 Appendix .....................................................................................................................................153 Appendix A Sample sets and demographic data of cohorts used in this thesis ...................... 153 A.1 Samples sets used in Chapter 2 ............................................................................... 153 x  A.2 Demographic data of BCCRC 261 AC and 92 SqCC with copy number profiles . 153 A.3 Datasets used to determine the prevalence and subtype specificity of YEATS4 amplification and overexpression ....................................................................................... 154 Appendix B List of publications ............................................................................................. 155  xi  List of Tables Table 2.1 Top subtype specific networks associated with genes deregulated by copy number ... 26 Table 2.2. Genes demonstrating concerted genetic and epigenetic disruption in SqCC .............. 30 Table 2.3 Genes demonstrating a significant association with survival ....................................... 36 Table 3.1 Copy Number status of complex components in multiple independent cohorts .......... 62 Table 3.2 Frequency of KEAP1 complex copy number alterations in AC and SqCC ................. 64 Table 4.1 Frequency of YEATS4 gain and amplification across multiple datasets ..................... 87 Table 4.2 Cisplatin and Nutlin IC50s ............................................................................................. 95 Table 4.3 Differentially altered genes following knockdown of YEATS4 used for Ingenuity Pathway Analysis .......................................................................................................................... 99 Table 5.1 Patient demographics for all cases with miRNA sequencing data ............................. 113 Table 5.2 Alteration frequencies of the most frequently deregulated NSCLC miRNAs ............ 113 Table 5.3  AUC results of all BCCRC subtype specific miRNA ............................................... 116 Table 5.4 Subtype  specific and NSCLC miRNAs that validated in the TCGA cohort ............. 118  xii  List of Figures Figure 2.1. Global genetic and epigenetic instability of AC and SqCC ....................................... 20 Figure 2.2. Genomic landscapes of lung AC and SqCC. .............................................................. 22 Figure 2.3. Gene networks involved in the development of SqCC and AC. ................................ 25 Figure 2.4. Epigenetically altered SqCC genes are significantly enriched for SCLC signaling. . 28 Figure 2.5. Subtype specific genes explain AC and SqCC phenotypes........................................ 32 Figure 2.6. Subtype-specific genomic differences are reflected at the protein level. ................... 34 Figure 2.7 CD9 expression is associated with differences in survival in AC and SqCC tumors. 44 Figure 2.8 SqCC cell lines are more sensitive to Trichostatin A than AC cell lines. ................... 46 Figure 3.1 The role of the KEAP1/CUL3 ubiquitin ligase complex ............................................ 53 Figure 3.2 Frequent disruption of the KEAP1 E3-ligase complex and IKBKB in NSCLC. ........ 61 Figure 3.3. Association between gene expression and copy number in clinical lung tumors. ..... 62 Figure 3.4. IKBKB protein expression in NSCLC. ...................................................................... 66 Figure 3.5. Activation of NF-κB targets in tumors with complex components disruption .......... 68 Figure 3.6. Pharmacological inhibition of IKBKB in NSCLC cell lines...................................... 69 Figure 4.1. Recurrent amplifications in NSCLC. ......................................................................... 83 Figure 4.2 YEATS4 is the target of 12q15 amplification. ............................................................. 86 Figure 4.3. Overexpression of YEATS4 induces a malignant phenotype. ................................... 89 Figure 4.4. YEATS4 knockdown impairs growth and induces senescence. ................................ 91 Figure 4.5. YEATS4 alters p21 and p53 protein levels. ............................................................... 94 Figure 4.6. Senescence induced by YEATS4 is dependent on p21 in wildtype p53 cells............ 97 Figure 4.7. Gene networks associated with YEATS4 knockdown. ............................................ 100 Figure 5.1. miRNA expression profiles accurately segregate tumor and non-malignant tissue . 112 Figure 5.2. miRNAs are disrupted in a subtype specific manner that can discriminate between subtypes....................................................................................................................................... 115 Figure 5.3  miRNA-mRNA interaction network of overexpressed SqCC specific miRNA ...... 121 Figure 5.4. Bi-directional miRNAs target genes involved in cellular fate and organization...... 122 Figure 5.5 Predicted interaction between NSCLC miRNAs and lung cancer prognostic genes 123 Figure 5.6 miR-944 overexpression is specific to SqCC and targets ERK signaling. ................ 125 Figure 5. 7. miR-1287 induces a malignant phenotype through the suppression of RAD9A .... 127 xiii  List of Abbreviations 3' UTR  3 prime untranslated region AC   adenocarcinoma aCGH   array comparative genomic hybridization BAC   bacterial artificial chromosome BrdU   bromodeoxyuridine CMAP   Connectivity Map CNA   copy number alteration CS   current smoker BCCA   British Columbia Cancer Agency BH   Benjamini Hochberg COSMIC  Catalogue of Somatic Mutations in Cancer DNA   deoxyribonucleic acid EGFR   Epidermal Growth Factor Receptor Freq   frequency FFPE   formalin fixed paraffin embedded FS   former smoker GEO   Gene Expression Omnibus  GISTIC  Genomic Identification of Significant Targets in Cancer HBEC   human bronchial epithelial cells IHC   immunohistochemistry KEAP1  Kelch-like ECH-associated protein 1  KRAS   v-Ki-ras2 Kirsten rat sarcoma viral oncogene LCCL   lung cancer cell lines MANOVA  multivariate analysis of variance miRNA/miR  microRNA ncRNA  non-coding RNA NF-κB   Nuclear Factor Kappa B NS   never smoker NSCLC  non-small cell lung cancer xiv  OE   overexpressed PCA   proportion of genome altered PCR   polymerase chain reaction PI   propidium iodide RPKM   reads per kilobase of exon model per million mapped reads RT-qPCR  reverse transcription quantitative PCR RNA   ribonucleic acid SCLC   small cell lung cancer SMRT   sub megabase resolution tiling SqCC   squamous cell carcinoma TCGA   The Cancer Genome Atlas TKI   tyrosine kinase inhibitor UE   underexpressed YEATS4  YEATS domain containing 4 xv  Acknowledgements   I would like to acknowledge the contributions of the many members of the Wan Lam Lab, both past and present who contributed to this work, and provided useful insight and discussion, especially the co-authors of each of the manuscript chapters presented herein. In addition, I would also like to acknowledge our collaborators and generous scholarship support from the Interdisciplinary Oncology Program Training Incentive Award, University of British Columbia Graduate Entrance Scholarship, Canadian Institutes of Health Research Frederick Banting and Charles Best Canada Graduate Scholarship Master’s Award, University of British Columbia Four Year Doctoral Fellowship, and the Vanier Canada Graduate Scholarship.   The research presented in this thesis was funded by the following granting agencies: Canadian Institutes of Health Research, Canadian Cancer Society Research Institute, Genome Canada/Genome BC, National Cancer Institute Early Detection Research Network, the Canary Foundation and the US Department of Defense.   I owe particular thanks to my supervisor Dr. Wan Lam and my supervisory committee members Dr. Stephen Lam and Dr. Calum MacAulay for their mentorship, guidance and continued support.  xvi  Dedication   This thesis is dedicated to my family and friends; in particular my Nan who was always my biggest supporter and liked to call me Dr. P. long before it was official, and my childhood friend Kristen who serves as a reminder of why I decided to pursue a career in cancer research, and to live each day to the fullest, appreciate the small things in life and never forget to smile.     1  Chapter 1: Introduction  1.1 Lung cancer  Lung cancer is the leading cause of cancer death in the world, with an estimated 25,500 new cases and 20,200 deaths in Canada in 2013, accounting for 27% of all cancer related deaths- more than breast, prostate and colon cancer combined [1, 2]. Surgery is the best curative option and standard of care for patients with localized disease. However, the majority of patients (>80%) present with locally advanced or metastatic disease for which platinum doublet chemotherapy and radiation are the standard treatment. Despite recent advances in diagnosis and treatment, such as the development of targeted therapies, the 5-year survival rate of lung cancer has failed to improve significantly over the last 30 years, and remains a meager 18% [1, 2]. This modest improvement is largely attributed to two factors; a lack of early detection strategies, and the modest survival benefits and inability of current treatment strategies to cure patients with disseminated disease.    Tobacco smoke exposure is the main etiological factor associated with lung cancer, with smokers having a 14 fold increased risk of developing lung cancer compared to never smokers [3]. However only 10-20% of smokers develop lung cancer, and with the success of smoking cessation programs, there is an increasing percentage of lung cancers (25%) that arise in individuals who never smoked [4, 5]. Other factors known to influence lung cancer risk include carcinogens such as arsenic, radon and asbestos, a family history of lung cancer, second hand smoke and viral infection (Human papillomavirus and Epstein Barr Virus) [4, 6].   1.2 Lung cancer subtypes/ histopathology of NSCLC  Histologically, lung cancer is classified into two broad categories; small-cell lung cancer (SCLC), occurring in approximately 15% of patients and the more prevalent NSCLC, which accounts for roughly 85% of cases [7]. NSCLC itself is a heterogeneous disease comprised of phenotypically diverse and regionally distinct neoplasias that differ in their cell of origin, growth pattern and location within the lung, and can be further subdivided into 3 major histological subtypes: adenocarcinoma (AC), squamous cell carcinoma (SqCC) and large cell carcinoma, of 2  which AC and SqCC are the predominant subtypes, and the focus of this thesis [7]. Within the last few decades, there has been a dramatic shift in the global trends of lung cancer histology, with a steady decline in SCLC and SqCC, making AC the most prevalent subtype of lung cancer. These changes are largely believed to be due to widespread changes in cigarette composition (lower tar and nicotine content) which has led to a change in smoking behavior, with smokers smoking more frequently and inhaling deeper in an attempt to achieve the same effect, causing tobacco carcinogens to be deposited further into the lung periphery.   Adenocarcinoma is the most common histological subtype of lung cancer, accounting for 40-50% of all lung cancer cases and typically arises in the glandular epithelium of the lung periphery from type II pneumocytes or Clara cells [4, 8]. Substantial heterogeneity exists within AC, with >90% of adenocarcinomas consisting of two or more histological subtypes and being classified as mixed subtype [9]. With advances in the field of lung adenocarcinoma, including the development of targeted therapies and the identification of prognostic subsets, there was a need for improvement in histological stratification. Therefore, in 2011 the International Association for the Study of Lung Cancer (IASLC), American Thoracic Society (ATS) and European Respiratory Society (ERS) developed a new classification for AC in which invasive ACs are classified according to the predominant subtype: acinar, lepidic, micropapillary, papillary or solid and have significant differences in prognosis [8].Lepidic tumors are the most indolent subtype.  Acinar and papillary predominant tumors have an intermediate clinical behaviour while solid and micropapillary tumors are associated with poor prognosis suggesting early stage patients with these histologies may benefit from adjuvant therapy [8].    Squamous cell carcinoma comprises 20% of all lung cancer cases, is strongly associated with a history of smoking and develops primarily in the central airways and segmental bronchi [7, 9]. It is believed that SqCC develops through a stepwise transformation of the normal epithelium. This process begins with hyperplasia of basal cells, followed by squamous metaplasia, various degrees of dysplasia (mild, moderate and severe) and then carcinoma in situ (CIS) before eventually developing into a malignant carcinoma [10]. Each stage can be characterized by increased morphological and cytological changes. Well differentiated squamous 3  tumors are characterized by cell keratinization, intracellular bridges and keratin pearl formation, and while less heterogeneous than AC, can also be further sub-classified into basaloid, papillary, clear cell and small cell types [9].   Until recently, NSCLC was treated as a single disease with a "one size fits all" therapeutic approach determined exclusively by disease stage due to the similar therapeutic effects of conventional chemotherapeutic agents. However, evidence from clinical trials has demonstrated that tumor histology influences response rates, toxicity and progression free survival of chemotherapy and targeted drugs such as bevacizumab, pemetrexed and epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKI's) [11, 12], such that histology is now recognized as an important factor in treatment selection [13]. The disparate clinical behaviours of AC and SqCC suggest distinct molecular mechanisms underlie these phenotypic differences and highlight the importance of understanding the biology and molecular origins of lung cancer subtypes.  1.3 Molecular pathology of lung cancer  Lung tumors harbor hundreds to thousands of molecular alterations that result in the activation of oncogenes and inactivation of tumor suppressor genes, leading to the deregulation of fundamental cellular processes and promoting malignancy [14-18]. These include but are not limited to gene and miRNA expression changes, copy number alterations, aberrant DNA methylation, and sequence mutations. In the past two decades, there has been extensive effort to identify and characterize the landscape of driver alterations that contribute to lung tumorignesis in an attempt to better understand the underlying biology of lung tumors and identify novel therapeutic targets [15, 18, 19].   1.3.1 Gene expression   Gene expression signatures have shown the ability to define and distinguish histological subtypes, [20-24] morphological subtypes within AC [20, 21] and SqCC [25] as well as distinguish tumor from non-malignant tissue [26-29], yet their clinical utility is limited due to the lack of overlap between subtype signatures. Interestingly, functional overlap between subtype 4  specific signatures has been observed, suggesting disruption of specific pathways is selected for rather than specific genes. Deregulation of antioxidant proteins, detoxification genes and overexpression of cytokeratins and cytokeratin-regulatory genes (GSTT1, CEL, and PRDX6) often characterize SqCC tumors [20-24], whereas disruption of surfactant-related and small airway-associated genes (SFTPA2, SFTPB, MUC1, and NAPSA) are typically altered in AC [20-24, 30, 31]. These functions are largely associated with the histological properties of the cells of origin from which these subtypes develop, further highlighting the contribution of histology to tumorigenesis. Due to the large number of passenger alterations within tumors, gene expression studies alone are limited in their ability to discriminate genes and pathways implicated in tumorigenesis.  1.3.2 Copy number alterations   DNA copy number alterations (CNAs) are a prominent mechanism of gene disruption in NSCLC [18, 32-39]. Although very few CNAs occur exclusively in a single subtype, many regions are altered at significantly different frequencies between subtypes and therefore deemed regions of subtype specific CNA [33, 34, 36]. For example, a recent analysis of over 2000 tumors identified 13 subtype-specific regions with at least a 25% difference in the frequency of alteration between subtypes [39]. Amidst all copy number studies, the most prominent and consistent difference between subtypes is amplification of 3q in SqCC [32, 33, 35, 38, 40-43]. Given the large size of some amplicons/regions of deletion, integration of gene expression data is useful in identifying of candidate oncogenes or tumor suppressors.   1.3.3 DNA methylation  Epigenetic marks such as DNA methylation are important regulators of somatically heritable changes in gene expression. DNA methylation is a tissue-specific and inherently reversible gene regulatory alteration targeted for chemoprevention, treatment and as potential diagnostic and prognostic biomarkers in malignant and non-malignant tissues [44]. DNA methylation profiling of NSCLC has identified hundreds of aberrantly methylated genes [45-49]. However, to date most genome-wide epigenetic studies lack corresponding gene expression level data, which in the context of determining functional consequences of DNA methylation 5  alterations to lung cancer biology, is limiting. In SqCC, integration of global DNA methylation and expression profiles indicate methylation of HOXA2 and HOXA10 may have prognostic relevance[14]. In AC, aberrantly methylated genes are enriched for cell differentiation, cell cycle regulation, epithelial to mesenchymal transition and RAS and WNT signaling pathways [50].   1.3.4 DNA mutations  Advances in whole genome sequencing technologies have enabled high throughput identification of mutations, copy number aberrations, and structural alterations such as gene fusions and chromosomal rearrangements in a genome-wide, unbiased manner. Sequencing studies have revealed lung cancer to have one of the highest mutation rates of all cancers, as well as immense mutational heterogeneity both within and between patients [17, 51-53]. For example, a single AC tumor was found to have over 50,000 variants, 391 of which affected coding sequences [54]. Frequently mutated genes in lung cancer include TP53, BRAF, ERBB2, KRAS, EGFR, PIK3CA, PTEN, CDKN2A, NF1, FGFR4, KEAP1 and RB1 [15, 55]. In addition to mutations, a number of gene fusions have been identified, including EML4-ALK, KIF5B-RET and multiple ROS1 fusions. The discovery of improved responses and outcomes with EGFR TKIs in lung cancer patients harboring EGFR mutations, and the profound clinical benefit of targeted therapies in other cancers launched the search for additional actionable alterations in lung cancer and marked the beginning of a new era in NSCLC in which NSCLC are further defined by their driver alterations.   1.3.5 Non-coding RNAs  MicroRNAs (miRNAs) are small 18-25nt long non-coding RNAs (ncRNAs) that negatively regulate gene expression post-transcriptionally through transcript degradation or translational repression [56]. A single miRNA is capable of regulating hundreds of protein coding genes, and a gene can similarly be targeted by numerous miRNAs. miRNAs regulate numerous biological process, including but not limited to proliferation, apoptosis, metabolism, epithelial to mesenchymal transition, differentiation and cellular development, and are frequently deregulated in tumorigenesis [56, 57]. The pathogenesis of lung cancer has been associated with the deregulation of several miRNAs, including loss of the well known tumor suppressive miRNA 6  let-7, which targets KRAS and is associated with poor prognosis in lung cancer. Similar to gene expression, miRNA signatures can accurately separate histological subtypes and are thought to be as good or even superior to global mRNA expression profiles in their ability to accurately classify NSCLC subtypes [57]. Array based miRNA profiling studies have shown miR-205 to be a highly specific marker for SqCC, while AC specific miRNAs have been shown to associate with mutation patterns.  For example, miR-155 is upregulated exclusively in AC with wildtype EGFR and KRAS, while miR-21 and miR-25 are upregulated in EGFR mutant AC and miR-495 is up-regulated in KRAS positive AC [58-60]. Based on these and other findings, miRNAs may be just as important to tumour biology and therapeutics as protein coding transcripts.  1.4 Integrative genomics and a systems biology approach to gene discovery  As a gene can be disrupted by a variety of mechanisms, to accurately determine gene disruption status, multiple mechanisms of disruption need to be interrogated. While the application of single dimensional analyses (expression, copy number, or mutation studies alone) are informative for identifying disrupted genes, they often overlook genes disrupted at low frequencies and are not capable of distinguishing causal from passenger events [61]. The integration of multiple dimensions of 'omics data provides a more comprehensive understanding of the genetic mechanisms affecting a tumor as it not only enables the identification of genes with concurrent DNA and expression alterations which are more likely to be driver alterations, but also genes disrupted by multiple mechanisms (indicative of selection) but at low frequencies by any single mechanism [61]. This is particularly relevant for the identification of frequently altered pathways, as signaling pathways are often disrupted at multiple and distinct points in different tumors as opposed to the frequent alteration of a single gene within a given pathway. Large scale, high throughput genomic profiling studies such as that of the National Cancer Institute's The Cancer Genome Atlas (TCGA) project reinforces the concept of assessing multiple 'omics dimensions to characterize tumors and further our understanding of the alterations that underlie their development.   7   Given the survival benefit associated with targeted therapies, the ability to accurately identify driver/actionable alterations within a tumor is critical. However, due to the substantial heterogeneity, high mutational load of lung cancer and tissue specific nature of molecular alterations such as DNA methylation, this can only be truly accomplished through the individual analysis of patient genomes with reference to that patient's matched non-malignant tissue. An individualized approach reduces the probability of overlooking driver genes that are disrupted in only a small subset of tumors, and would otherwise be missed using conventional approaches in which tumor and normal samples are grouped.  Moreover, consideration of multiple dimensions enables the identification of genes with concordant DNA and expression alterations. A multi-dimensional integrative approach on a tumor by tumor basis is therefore ideal for identifying the molecular mechanisms and signaling pathways that contribute to tumorigenesis.  1.5 Thesis theme and rationale  The theme of this thesis is to characterize the genetic and epigenetic alterations and signaling pathways that define and distinguish AC and SqCC in order to improve our understanding of the genetic basis of the major subtypes of NSCLC. As histological subtypes display distinct clinical features and differential responses to treatment, knowledge of the molecular mechanisms underlying these differences will lead to the development of novel subtype specific therapeutic and detection strategies, and potentially alter clinical management and patient prognosis. Previous attempts to discern the genetic differences between AC and SqCC have been limited in their ability to identify unique causal subtype specific disruptions due to small sample sizes, a lack of patient matched non-malignant tissue and single dimension analyses in which tumors are grouped for analysis. As a result, our understanding of the molecular mechanisms that contribute to tumorigenesis and how these differ between subtypes is limited. A genome wide, integrative multi-'omics approach in combination with functional analysis will provide 1) a comprehensive analysis of the recurrent alterations characteristic of AC and SqCC and 2) insight into how these alterations deregulate specific signaling pathways and biological processes and contribute to subtype pathogenesis.    8  1.6 Objectives and hypotheses  The objective of this work is to characterize the genetic alterations, genes and pathways that contribute to tumorigenesis and the differential development of histological subtypes using a multi-dimensional integrative analysis in combination with functional validation.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               This is based on the following hypotheses: 1. AC and SqCC arise through distinct molecular mechanisms, which can be identified through integrative analysis of AC and SqCC genomes. 2. These different patterns of genetic and epigenetic alterations underlie unique biological mechanisms that contribute to subtype development, phenotypes and response to therapy.    1.7 Specific aims and thesis outline  To address the questions of 1) what are the genetic, molecular and biological similarities and differences between AC and SqCC, 2) how these differences contribute to tumorigenesis and subtype biology and 3) whether any of the identified subtype specific alterations have potential therapeutic, prognostic or diagnostic implications, we devised the following specific aims.   Aim 1: Identify DNA and miRNA alterations that are shared or specific to lung cancer subtypes.  Chapters 2 and 5 describe genome wide comparisons of DNA alterations (copy number and methylation) with concordant expression changes and miRNA expression differences between subtypes, respectively. At the time Chapter 2 was submitted for publication, a large scale multi-dimensional integrative analysis and comparison of subtypes on high resolution platforms had yet to be performed. Given the recent observations of subtype specific indications for EGFR TKIs, pemetrexed and bevacuzimab, a comprehensive analysis of the genetic and epigentic alterations and subsequent genes and pathways differentially disrupted between AC and SqCC was warranted. Our integrative analysis identified numerous subtype specific alterations in concordance with hypothesis 1. To date, miRNA profiling studies of lung cancer 9  subtypes have been limited by the use of microarray platforms which depend on an a priori knowledge of miRNAs and as a result are limited to a few hundred miRNAs, and a lack of profiles for patient matched non-malignant tissue. At the time of publication of this thesis, an unbiased, genome wide comparison of miRNA expression between subtypes had yet to be performed. In Chapter 5, we performed miRNA sequencing on  22 SqCC and 66 AC tumor and non-malignant tissue pairs and identified miRNAs differentially altered and expressed between subtypes, as well as a list of miRNAs deregulated in greater than 90% of all cases irrespective of histology. In all chapters of this thesis, genomic findings were validated in multiple independent datasets in order to ensure findings were reproducible and not merely an artifact of our dataset or analysis.   Aim 2: Elucidate genes and pathways involved in subtype tumorigenesis and delineate the biological significance of their disruption.  Upon the identification of subtype specific alterations, I next sought to understand how these alterations contribute to tumorigenesis. This was achieved through in vitro and in vivo manipulation of candidate genes followed by phenotype assessment and interrogation of the pathways and biological processes affected by gene disruption. Chapter 3 is an in depth analysis of one of the subtype specific gene findings from Chapter 2. In this study we assessed the frequency and functional consequences of disruption of our candidate gene (KEAP1) as well as the other components of the E3 ubiquitin ligase complex of which it is a part of. The results from this analysis revealed that disruption of any one complex component was sufficient to impair complex function and that differential patterns of complex components disruption characterize AC and SqCC.   In chapter 4, I focused on the most significant region of amplification in our AC cohort and identified a novel candidate oncogene (YEATS4) that is frequently amplified and overexpressed in both AC and SqCC. Manipulation of this gene in lung cancer cell lines and xenograft models impaired tumor growth, while overexpression in bronchial epithelial cells was sufficient to induce malignant transformation. Downstream analysis revealed a role for this gene 10  in the regulation of the p21-p53 pathway. Although initially identified in AC, our candidate gene was found to be amplified and overexpressed at similar frequencies in both subtypes in multiple independent cohorts, highlighting one of the many shared alterations between subtypes.   Our genome wide analysis of miRNA expression in Chapter 5 identified several subtype specific miRNAs. miRNAs can target hundreds of mRNAs and stringent target prediction of the only AC specific miRNA to validate in the TCGA identified two intriguing predicted targets. The final aim of this chapter was to determine whether these predicted targets are true targets of our AC specific miRNA and assess the effect of miRNA overexpression on tumorigenesis. The findings in this chapter demonstrate the ability of our analysis approach to identify biologically relevant miRNAs involved in tumor biology and subtype development, and provide evidence in support of hypothesis 2.   Aim 3: Determine the clinical relevance of gene/pathway disruption  The final aim of this thesis was to assess the potential clinical relevance of alterations identified in Aim 1 and 2. In Chapter 2 we confirmed in silico findings that suggested SqCC tumors would be sensitive to HDAC inhibitors. Similarly, dose response assays performed in Chapters 3 and 4 demonstrate that gene disruption significantly sensitized cell lines to a IKBKB inhibitor and cisplatin, respectively. The identification of miRNAs ubiquitously altered in NSCLC tumors and the stability of miRNAs in blood highlights the potential of these miRNAs as blood based biomarkers for the detection of lung cancer. Taken together, these findings demonstrate the clinical relevance and potential application of characterizing shared and subtype specific alterations.     11  Chapter 2: Divergent genomic and epigenomic landscapes underscore the selection of different oncogenic pathways during subtype development  2.1 Introduction  Evidence from recent clinical trials has demonstrated that histological subtypes of NSCLC respond differently to both targeted drugs and newly developed chemotherapies, possibly related to differences in cell derivation and pathogenetic origins[12, 62, 63]. One of the most striking examples is the folate antimetabolite pemetrexed, which exhibits superior efficacy and is restricted for use in patients with non-SqCC, presumably due to the higher expression of thymidylate synthase in SqCC tumors [12, 64, 65]. Likewise, numerous studies have associated a higher response rate upon treatment of AC with the EGFR tyrosine kinase inhibitors Gefitinib and Erlotinib, reflecting the higher prevalence of EGFR mutations in this subtype [13, 62]. These discrepancies in tumor biology and clinical response emphasize the need to determine the underlying genetic, epigenetic and metabolic similarities and differences between the NSCLC subtypes in order to define more appropriate avenues for therapeutic intervention.   Initial gene expression profiling studies were able to segregate AC and SqCC tumors into their respective histologic groupings based on multi-gene models; however, critical events in tumorigenesis may be masked by reactive changes when examining expression profiles alone [20, 66]. Conversely, DNA copy number or DNA methylation changes corresponding with gene expression changes are often regarded as evidence of causality. These alterations are critical events driving progression and other cancer phenotypes [67-69]. Since SqCC and AC develop from distinct cell lineages in different regions of the lung, the range of genetic alterations required for tumor initiation may occur in a lineage-restricted manner. For example, the amplification of the lineage survival oncogenes SOX2 and TITF1/NKX2-1 have recently been identified as key events specific to the development of lung SqCC and AC, respectively [18, 70]. However, these genes alone are insufficient to explain the phenotypic diversity of the subtypes, suggesting that the vast majority of genes responsible for their differential development remain unknown. Although subtype specific differences have been observed across all 'omics levels, low genome coverage and/or small sample sizes have been limiting [33-35, 43, 49]. Moreover, gene 12  discovery on its own provides little information regarding tumor biology, and studies interrogating how differentially disrupted genes interact to perturb signaling pathways within subtypes are rare.   In this study, we performed a large-scale analysis of primary NSCLC tumors (261 total;169 AC and 92 SqCC), integrating DNA copy number, methylation and gene expression profiles to identify critical subtype-specific molecular features. The characterization of the genomic and epigenomic landscapes of AC and SqCC revealed an astounding number of differences at the DNA level with subsequent gene expression changes that are selected for during subtype-specific lung tumor development. Importantly, we identified key oncogenic pathways disrupted by these alterations that likely serve as the basis for differential behaviors in tumor biology and clinical outcomes. Lastly, through prognostic analysis and in silico screening of candidate therapeutic compounds using subtype-specific pathway components, we show how these findings may influence our approach to the clinical management of NSCLC.   2.2 Methods  2.2.1 DNA samples Formalin-fixed, paraffin embedded and fresh-frozen tissues were collected from St. Paul’s Hospital, Vancouver General Hospital and Princess Margaret Hospital following approval by the Research Ethics Boards. Hematoxylin and eosin stained sections for each sample were graded by a lung pathologist for use in selecting regions for microdissection. DNA was isolated using standard procedure with proteinase K digestion followed by phenol-chloroform extraction as previously described [71]. All samples were collected under informed, written patient consent and anonymized as approved by the University of British Columbia-British Columbia Cancer Agency Research Ethics Board (REB number H04-60060). Patient information is located in Appendix A2.   13  2.2.2 Tiling path array comparative genomic hybridization Array hybridization was performed as previously described [72-74]. Briefly, equal amounts (200-400 ng) of sample and single male reference genomic DNA were differentially labelled with cyanine-5 dCTP and cyanine-3 dCTP, respectively and cohybridized to the SMRT array v.2, which includes over 32,000 BACs spotted in triplicate and has a resolution of approximately 80kb (BCCRC Array Laboratory, Vancouver, BC). Hybridized arrays were imaged using a charge-coupled device (CCD) camera system and analyzed using SoftWoRx Tracker Spot Analysis software (Applied Precision). Systematic biases were removed from all array data files using a stepwise normalization procedure as previously described [75, 76]. SeeGH software was used to combine replicates and visualize all data as log2 ratio plots [77]. Stringently, all replicate spots with a standard deviation above 0.075 or signal to noise ratios below three were removed from further analysis. The probes were then positioned based on the human March 2006 (hg18) genome assembly, with removal of the X and Y chromosomes. Genomic imbalances (gains and losses) within each sample were identified using aCGH-Smooth [78] with lambda and breakpoint per chromosome settings of 6.75 and 100, respectively.   2.2.3 DNA methylation analysis For 30 AC samples, 30 patient-matched non-malignant lung samples, 13 SqCC samples and 18 non-patient matched bronchial epithelia samples, all of which were fresh frozen (Sample Set # 2, Appendix A1), DNA methylation profiling was performed using the Illumina Human Methylation27 chip. Five hundred nanograms of DNA from each sample was analyzed by this technology. Normalized β-values were obtained and only those with a detection p-value of ≤0.05 were used. When comparing tumor samples (AC/SqCC) and normal non-malignant samples (AC non-malignant parenchyma and bronchial epithelia), probes were deemed aberrantly methylated if the absolute difference between tumor and the average of the appropriate normal samples was ≥0.15.  2.2.4 Comparison of subtype alteration frequencies Regions of differential copy number alteration between AC and SqCC genomes were identified as follows. Each array element was scored as 1 (gain/amplification), 0 (neutral/retention), or -1 14  (loss/deleted) for each individual sample. Values for elements filtered based on quality control criteria were inferred by using neighbouring probes within 10 Mb. Probes were then aggregated into genomic regions if the similarity in copy number status between adjacent probes was at least 90% across all samples from the same subtype. The occurrence of copy number gain/amplification, loss/deletion, and retention at each locus was then compared between AC and SqCC data sets using the Fisher’s exact test. Testing was performed using the R statistical computing environment on a 3 x 2 contingency table as previously described, generating a p-value for each probe [72]. Benjamini-Hochberg multiple hypothesis testing correction based on the number of distinct regions was applied and resulting p-values ≤0.01 were considered significant. Adjacent regions within 1 Mb which matched both the direction of copy number difference and statistical significance were then merged. Finally, regions had to be altered in >20% of samples in a group and the difference between groups >10% to be considered subtype specific.  A similar approach was used for determining subtype-specific DNA methylation alterations. Frequencies of hypermethylation and hypomethylation for each probe were compared using a Fisher's exact test, followed by a Benjamini-Hochberg multiple testing correction. A corrected p-value cut-off of p<0.05 was used to deem a probe differentially methylated between the two groups. Due to the smaller number of probes that passed multiple testing correction, no frequency criteria was applied.  2.2.5 Gene expression microarray analysis  Fresh-frozen lung tumors were obtained from Vancouver General Hospital as described above. Microdissection of tumor cells was performed and total RNA was isolated using RNeasy Mini Kits (Qiagen Inc). Samples were labeled and hybridized to a custom Affymetrix microarray according to the manufacture’s protocols (Affymetrix Inc, Sample Set # 3,Appendix A1). In addition, RNA was obtained from exfoliated bronchial cells of lung cancer free individuals obtained during fluorescence bronchoscopy (Sample Set # 5, Appendix A1) [79]. All individuals were either current or former smokers. Expression profiles were generated for all cases using the Affymetrix U133 Plus 2 platform (Affymetrix Inc). All data was normalized using the Robust 15  Multiarray Average (RMA) algorithm in R [80]. In addition, publically available datasets downloaded from the Gene Expression Omnibus were used: Affymetrix U133 Plus 2 expression data was downloaded for accession numbers GSE3141 (Sample Set # 4, Appendix A1) [81] and GSE8894  (Sample Set #6, Appendix A1) [82].  2.2.6 Statistical analysis of gene expression data  Gene expression probes were mapped to March 2006 (hg18) genomic coordinates and those within the regions of copy number difference between the subtypes were determined. Comparisons between expression levels for AC and SqCC tumors were performed using the Mann-Whitney U test. As the direction of gene expression difference was predicted to match the direction of copy number difference, one tailed p-values were calculated. A Benjamini-Hochberg multiple hypothesis testing correction was applied based on the total number of gene expression probes analyzed for each region. Probes with a corrected p-value ≤ 0.001 were considered significant. If multiple probes mapped to the same gene, the one with the lowest p-value was used. Resulting genes were then mapped to the corresponding probes on the Affymetrix U133 Plus 2 array in order to compare their expression in a second set of NSCLC tumors (GSE3141, Sample Set # 4) against normal bronchial epithelial cells. If multiple probes were present for a gene, the one with the strongest p-value was used. All comparisons were performed using a one-tailed t-test with unequal variances in Excel and genes with a p<0.001 were considered significant. The fold-change for tumors versus normal tissues was then determined in order to determine genes expressed in the direction predicted by copy number.   Principal component analysis was performed using expression data for the three independent tumor data sets (described above, Sample Sets #3, 4, 6) in MATLAB. All genes of interest with probes on the corresponding arrays were used. Briefly, the first and second principal components were generated from the original dataset. In the subsequent validation in secondary datasets, these principal components are then used to weight the expression data for a gene based on the original distribution. The Receiver Operating Characteristic (ROC) area under the curve (AUC) analysis was performed to determine the ability of principle component 1 to separate the AC and 16  SqCC samples into their appropriate histological groups. Calculations were performed using GraphPad Prism software.   Connectivity Map (http://www.broad.mit.edu/cmap/) analysis was performed using the up and downregulated genes specific to each subtype as previously described [83].  2.2.7 Survival analysis Survival analysis was performed using the statistical toolbox in Matlab. Expression data for each gene were sorted and survival times compared between the top 1/3 and bottom 1/3 in expression using a publicly available gene expression microarray dataset with survival data (Sample Set #6, Appendix A1). Two tailed p-values were generated using a Mantel-Cox log test and those < 0.05 were considered significant. Kaplan-Meier plots were then generated for each gene of interest.  2.2.8 Network identification Functional identification of gene networks and canonical signalling pathways was performed using Ingenuity Pathway Analysis program (Ingenuity Systems). AC and SqCC specific gene lists were imported as individual experiments using the Core Analysis tool. The analysis was performed using Ingenuity Knowledge Database with the Affymetrix U133 Plus 2 platform as the reference set and was limited to direct and indirect relationships.  2.2.9 Human lung tissue microarray case selection To determine the expression of ERCC1, KEAP1 and SOX2 in primary NSCLC, we selected 330 NSCLCs (AC, n=220; SqCC, n=110) from surgically resected lung cancer specimens from the Lung Cancer Specialized Program of Research Excellence Tissue Bank at The University of Texas M.D. Anderson Cancer Center. We used archived, formalin-fixed, paraffin-embedded (FFPE) tumor tissue samples placed in tissue micro-array (TMA). The tumor tissue samples were collected between 1997 and 2003, and were histologically analyzed and classified using the 2004 WHO classification system [7]. The characteristics of these TMAs have been previously described in detail [84, 85].  17  2.2.10 Immunohistochemical analysis  The immunohistochemical analysis was done using commercially available antibodies against KEAP1 (dilution1:25; Proteintech), ERCC1 (dilution 1:25; Labvision) and SOX2 (dilution 1:50; R&D system). Immunohistochemical staining was done using an automated stainer (Dako, Inc.) with 5-μm-thick TMA sections from FFPE tissues. Tissue sections were deparaffinized and hydrated. Antigen retrieval was done in pH 6.0 citrate buffer in a decloaking chamber (121°C × 30 seconds, 90°C × 10 seconds) and washed on Tris buffer. Peroxide blocking was done at ambient temperature with 3% H2O2 in methanol. The slides were incubated with primary antibody (KEAP1 and ERCC1 for 60 minutes; SOX2 for 90 minutes) at ambient temperature and washed with Tris buffer, followed by incubation with biotin-labeled secondary antibody for 30 minutes (EnVision Dual Link System-HRP-Dako for KEAP1 and ERCC1; LSAB system-Dako for SOX2). The immunostaining was developed with 0.5% 3,3′- diaminobenzidine, freshly prepared with imidazole-HCl buffer (pH 7.5) containing hydrogen peroxide and an antimicrobial agent (Dako) for 5 minutes, and then the slides were counterstained with hematoxylin, dehydrated, and mounted.  Nuclear ERCC1, cytoplasmic KEAP1, and nuclear SOX2 expressions were quantified using a four-value intensity score (0, 1+, 2+, or 3+) and the percentage (0-100%) of the extent of reactivity. An immunohistochemical expression score was obtained by multiplying the intensity and reactivity extension values (range, 0-300), and these expression scores were used to determine expression levels.  2.2.11 Trichostatin A dose-response analysis  The effect of HDAC inhibitor Trichostatin A, (Cayman Chemicals) on six NSCLC cell lines; three AC (H3255, H1395 and A549) and three SqCC lines (HCC95, HCC15 and H520) was assessed by cell viability assays. Cells were plated in triplicate in 96 well plates at optimal densities for growth (A549 at 2000 cells/well, HCC95, HCC15 and H520 at 3000 cells/well, and H3255 and H1395 at 6000 cells/well). Cells were subjected to a series of 2-fold dilutions of Trichostatin A prepared in cell growth media with DMSO. The experimental concentrations ranged from 100 uM to 109 pM and the final DMSO concentration for treated and untreated 18  (control) cells was 1%. Blank wells contained equal volumes of growth media with 1% DMSO. Cells were incubated for 72 hours at 37°C and then treated with 10 μl of Alamar Blue cell viability reagent (Invitrogen) according to manufacturer’s instructions. The reaction product was quantified by measuring absorbance at 570 nm with reference to 600 nm using an EMax plate reader (Molecular Devices). The response of treated cells was measured as a proportion of the viability of untreated cells, with the mean background subtracted treatment absorbance divided by the mean background subtracted untreated absorbance for each concentration. Dose response curves and IC50 values were generated in Graph Pad v5 using the proportionate response of all 20 drug concentrations. Experiments were repeated in quadruplicate and differences in IC50 values were assessed by a student’s t-test with a p-value <0.05 considered significant.   2.3 Results  2.3.1 Assessment of global genomic instability in AC and SqCC  Based on the differing exposure of cells in the central (SqCC) and peripheral (AC) airways to tobacco carcinogens, which has been linked to the induction of DNA mutations and broad chromosomal instability, we first sought to determine whether global genetic or epigenetic instability was more prevalent in one subtype. Whole genome copy number profiles and DNA methylation profiles were generated and compared for 261 NSCLC FFPE tumors; 169 AC and 92 SqCC (Sample Set #1, Appendix A1) and 30AC, 13SqCC, 30 non-malignant lung parenchyma samples (AC reference) and 18 histologically normal exfoliated bronchial epithelial cell samples (SqCC reference), from patients with NSCLC, respectively (Sample Set #2, Appendix A1).    Unlike the genome, which is identical for most normal cells in the body, the epigenome differs between tissue types [86, 87]. Cancer genomes exhibit global hypomethylation to varying degrees depending on the tissue of origin [88]. DNA methylation profiles are also influenced by mutational profiles within different cancer types, as DNA hyper- and hypomethylation alterations are known to be related to tissue and genetic background [89] as well as smoking 19  behavior [90]. Given the differing mutational spectra of the two NSCLC subtypes and their differing cells of origin, we investigated the overall DNA methylation level of 30 AC and 13 SqCC samples relative to appropriately matched normal cells (exfoliated broncial epithelial cells). Analysis of 27,578 CpG dinucleotides probes within >13000 CpG islands revealed DNA methylation in the bronchial epithelia and SqCC tumors to be slightly lower than in the normal lung or AC tumors (Figure 2.1A). This trend is mirrored and exaggerated in the CpG dinucleotides outside of CpG islands, suggesting that the cells of the central airway are globally hypomethylated relative to the cells of the peripheral airways, whether cancerous or not (Figure 2.1B, Mann Whitney U test, p<0.0001).    The relative genomic instability in AC and SqCC, calculated as the average number of altered probes per sample; gained, lost, and neutral probes, and differentially methylated probes (Tumor βvalue- respective normal β value) was assessed for each tumor using a Mann-Whitney U test. No significant differences in differential methylation at CpG and non CpG island probes or copy number between the two subtypes were found (Figure 2.1C-F), consistent with previous findings [91]. This analysis demonstrates that neither subtype has a proclivity for gain, loss, hypo- or hypermethylation of DNA and therefore any observed differences in alteration frequency at a given locus can be attributed to subtype specific selection of genes within altered regions and not to different degrees of random genomic instability associated with tumor development.     20  Figure 2.1             Figure 2.1. Global genetic and epigenetic instability of AC and SqCC Comparison of DNA methylation averages between AC, histologically normal lung parenchyma, SqCC, and bronchial epithelia for CpG island probes (A) and non-CpG island probes (B). β-value is defined by the methylated signal/total signal for each probe. Average differential methylation levels at CpG islands (C) and non CpG islands (D) for 30AC and 13 SqCC. Percentage of BAC clones of each state calculated for each sample. Box plots illustrate the percentage of clones with gain/amplification, loss/deletion or neutral, in AC (E) and SqCC (F). No significant differences in genetic or epigenetic instability between subtypes was observed, Mann-Whitney U test, p>0.05.  Box plots depict the 25th, median and 75th percentile, whiskers represent the 5th and 95th percentile and dots show those samples outside theses cutoffs. D A F E C B 21  2.3.2 Disparate genetic and epigenetic landscapes characterize lung SqCC and AC  Although NSCLC subtypes exhibit similar levels of genomic instability, if specific genetic pathways are involved in their differential development, differences in the genomic alterations selected during tumorigenesis should be present. To determine if genetic alterations unique to each NSCLC subtype exist, we looked for recurrent regions of aberration in each group. For copy number, samples were grouped by subtype and probes were aggregated into regions based on similar copy number status. The frequency of alteration across autosomes was determined and compared between subtypes using the Fisher’s exact test with multiple testing correction and a cut-off of <0.01 considered significant. To be considered subtype specific, we required regions to be altered in >20% of samples from a subtype group and have a difference between groups of >10%. Figure 2.2 displays the resulting genomic landscapes of AC and SqCC and highlights the corresponding regions of difference between subtypes. 294 regions of copy number disparity between SqCC and AC were identified, 205 of which were SqCC specific, and 89 that were AC specific (For specific regions, see supplemental digital content, available online http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0037775) . Although some regions overlapped, the character of the alteration (i.e. gain vs. loss) was specific to a subtype. Since the alteration status between the subtypes differed strongly, we classified these as subtype-specific copy number alterations. In total, these alterations covered approximately 550 Mbp of the genome, ranging in size from large segments on chromosome arms (64.8 Mbp on 4q) to discrete peaks only kilobases in size (0.05 Mbp in multiple places).   For DNA methylation, tumor methylation levels were compared to the average of available normal reference tissue profiles. The frequency of probe hypermethylation and hypomethylation (tumor–normal ≥ ±0.15) in AC and SqCC samples was compared using the Fisher’s exact test. Following correction for multiple comparisons, 2708 probes corresponding to 2384 genes were found to be differentially methylated (p<0.05) (See supplemental digital content for probes). The SqCC group contained markedly more recurrently hyper- and hypomethylated loci than the AC group, similar to the disparity in the numbers of subtype-specific copy number-regulated genes. In fact, only 8% (217) of the 2708 significant probes were more frequently altered in AC, the rest being more commonly hyper- or hypomethylated in SqCC. 22  Figure 2.2   Figure 2.2. Genomic landscapes of lung AC and SqCC. Alteration frequencies for 169 AC (red) and 92 SqCC (blue) across the human genome. Solid vertical black lines represent chromosome boundaries whereas dotted black lines represent chromosome arm boundaries. The frequency of copy number gain and loss are denoted in the top and middle panel, respectively. The significance of copy number disparity (inverse p-value corrected for multiple comparisons) between subtypes is depicted in the bottom panel. Black lines represent statistically different regions (p<0.01) whereas grey lines are not significant.   23  2.3.3 Gene disruptions are selected in a subtype specific manner   The identification of DNA copy number and methylation disparities between NSCLC subtypes suggests that genes within these regions may be preferentially selected during tumorigenesis and thus responsible for the differential development and pathological characteristics of the subtypes. To identify the potential target genes of these alterations, we integrated DNA copy number and methylation with gene expression levels, with the hypothesis that genes targeted by subtype-specific alterations would 1) be differentially expressed between AC and SqCC, 2) have an expression pattern matching the direction of the alteration and 3) be aberrantly expressed in tumors relative to normal tissue (matching the direction of the alteration), suggesting a role in tumor development. Gene expression profiles were generated for a subset of tumors analyzed by array CGH and DNA methylation (20 SqCC and 29 AC tumors, Sample Set #3, Appendix A1). An independent cohort of 53 SqCC and 58 AC lung tumors and 67 exfoliated bronchial cell samples from cancer-free individuals (Sample Sets #4 and #5, Appendix A1) were used to identify expression differences between tumors and normal cells.   Genes located within each subtype-specific copy number alteration were identified and the expression levels between SqCC and AC and tumor and normal samples compared to determine those that were differentially expressed (Mann Whitney U test, p<0.001, after multiple testing correction). 4669 and 2050 unique genes mapped to the SqCC and AC subtype- specific copy number alterations, respectively. After applying all three criteria for defining candidate subtype-specific copy number alteration regulated targets, 447 SqCC-specific and 71 AC-specific genes, corresponding to a total of 492 unique genes were identified as subtype specific. Expression analysis of the 2384 genes with differential methylation revealed 32 AC candidate genes and 297 SqCC genes that were differentially expressed between subtypes and tumor and normal tissue (See supplemental digital content for lists). In both the copy number and methylation analysis, some genes were disrupted in both subtypes, however their patterns of disruption were specific to a subtype, suggestive of opposing roles (oncogenic vs. tumor suppressive) depending on cellular context and therefore considered to be subtype-specific targets. Genes meeting all our criteria may represent critical alterations driving the development of each subtype.  24  2.3.4 Different oncogenic pathways are associated with AC & SqCC   Cellular pathways and processes specifically disrupted in individual subtypes may reveal key oncogenic mechanisms driving the differential development of AC and SqCC. After identifying the genes differentially altered between subtypes, we next wanted to investigate their biological functions. To discover subtype-related networks of biologically related genes we performed Ingenuity Pathway Analysis (IPA) on the 71 AC and 447 SqCC specific copy number alteration regulated targets. SqCCs exhibited disruptions in gene networks that function in regulating DNA replication, recombination and repair, with additional roles in lymphoid tissue structure and development (Figure 2.3, Table 2.1). Genes involved in the top SqCC network were associated with the binding and modification of histone protein H4, as well as the regulation of the NFKB complex (Figure 2.3b). In contrast, the primary networks in AC displayed functions associated with cell-to-cell signaling, development, and drug metabolism (Table 2.1).  2.3.5 Epigenetically regulated genes complement genetically regulated genes  To determine whether the subtype specific epigenetically-regulated genes carried out functions similar to those subtype-specific genes discovered by the DNA copy number analysis, pathway disruption analysis was performed on the methylation regulated genes. The most significant epigenetically regulated gene network in AC was involved in cell cycle, cell death, and cellular development. This is partly in contrast to the top AC network of copy number regulated genes, which have functions associated with tissue development, but also possess cell signaling and hematological system functions in common. The overall degree of similarity between AC- specific genes that are genetically or epigenetically regulated is quite small, likely due to the low number of AC-specific genes identified (potential reasons for this are discussed below). In contrast, the SqCC gene networks in both analyses are very similar. For example, DNA replication, recombination and repair are highly featured functions of genes identified by both DNA copy number and DNA methylation analyses of SqCC. Genes involved in immunological disease and lymphoid tissue structure and development were also prominent.     25  Figure 2.3   Figure 2.3. Gene networks involved in the development of SqCC and AC.  Ingenuity Pathway Analysis identified biologically related networks from the subtype specific genes deregulated by subtype-specific copy number alterations. The top resultant gene networks for each subtype are displayed; (A) AC network related to HNF4 signaling. (B) SqCC network of potential interactions between multiple histone regulating genes. Solid lines denote direct interactions, dotted lines represent indirect gene interactions. Network components highlighted in red are upregulated whereas those in green are downregulated. Those not highlighted are used to display relationships. Molecules are represented as follows; corkscrews represent enzymes, y-shaped molecules are transmembrane receptors, barbells are regulators of transcription, thimble-shaped molecules are transporters, kinases are triangular, and circular molecules encompass all other gene products. A B 26  Table 2.1 Top subtype specific networks associated with genes deregulated by copy number Network Molecules in Network Score Focus Molecules Top Functions AC 1 APEH, ATP13A3, BRIP1, C22ORF28, CDC40, CDC42EP3, CNBP, CTDSPL2, DAPK2, DLL3, GTF2H3, GTPBP3, HNF4A, IGSF8, ITGA10, ITGB1, JAG2, LCMT2, LSG1, MAPK14, MIB1, MPHOSPH9, NAT13, OPA1, PARP, PARP11, PITPNB, PLDN, SMAD2, SNX5, STAT4, TGFB1, UBA5, UMPS, WHSC1L1 40 19 Tissue Development, Cell-To-Cell Signaling and Interaction, Hematological System Development and Function AC 2 ATP2C1, beta-estradiol, BMP4, BRWD1, Ca2+, CD9, CD53, CD160, CENPM, cyclic AMP, EIF3J, EIF4A2, FYTTD1, GJA1, GJA5, hCG, IL15, MAPT, MIR21, MRPL3, MYO1B, NR3C1, NUDT17, PTPRU, S1PR5, SGPP1, SHBG, TCEB1, TIPIN, TM2D2, TTBK1, TTBK2, UBR2, UBXN7, VEGFA 30 16 Drug Metabolism, Endocrine System Development and Function, Lipid Metabolism AC 3 ADAP1, ASH2L, BAG4, Beta ark, C16ORF53, CSDA, DENND4A, DPY30, DUSP3, DUSP9, ELANE, FSH, Histone h3, HIVEP1, LSP1, LY96, MAPK1, MLL3 ,NFkB (complex), NOTCH2NL, PCYT1A, PDC, PDZK1, Pka, PPAP2A, PPP1R2, PPP1R9A, PRLH, PTPN7, RGS5, RIOK3, RLN2, S1PR3, SETD1A, ST8SIA1 17 10 Cardiovascular System Development and Function, Organismal Development, Tissue Morphology SqCC 1 ACTL6A, ASF1B, BAG4, Basc, BLM, BRD4, CD3, CHAF1A, CUL3, DCUN1D1, ECSIT, Histone h4, Importin alpha, Importin beta, KPNA1, KPNA4, LGALS7, LTBR, MIB1, MSH6, NFkB (complex), NUP62, PCYT1A, PNKD, RECQL, RELB, RFC2, RFC4, RNF7, SAE1, SETD8, SH3RF1, SIAH2, SLC25A4, TNFRSF13C 44 29 DNA Replication, Recombination, and Repair, Cell Cycle, Lymphoid Tissue Structure and Development SqCC 2 Alcohol group acceptor phosphotransferase, Alpha tubulin, ATF5, CAMK2N2, CaMKII, CLINT1, COPS7A, CSDA, CSTA, DNAJB11, FBXO45, FOXL2, HDL, HNRNPM, Hsp70, Hsp90, IL11, IL6ST, MAPK1, MAPK9, MAPK12, OTX1, Pak, PAK2, POMC, PPM1F, PPP5C, Sapk, SORBS2, STAR, STAT, TNPO1, TNPO2, TOM1L1, Ubiquitin 34 25 Nervous System Development and Function, Organismal Development, Endocrine System Development and Function SqCC 3 ACTR2, ARHGAP26, Arp2/3, CLDND1, CYFIP2, DOT1L, EFNA5, EHBP1, EHD3, FCHO2, FXR1, H1FX, Histone H1, Histone h3,histone-lysine N-methyltransferase, KLHL24, LARP1B, LRRC58, MIR124, Nfat, NMD3, P38 MAPK, Pka, PLEKHA2, PPM1L, PTMS, Rac, RANBP1, Ras homolog, SETD7, SH3BP1, SMC4, UHRF1, WHSC1L1, XPO1 29 25 Cell Morphology, Cell-To-Cell Signaling and Interaction, Cellular Development 27    Of particular interest was the enrichment of aberrantly methylated genes in the small cell lung cancer signaling pathway (genes known to be deregulated in small cell lung cancer) (Figure 2.4a). This was the most significantly enriched canonical pathway affected by DNA methylation alterations and is of interest because both of these lung cancers (SCLC and SqCC) arise in the central airways with similar exposure to cigarette smoke carcinogens. E2F1 is among the hypomethylated and overexpressed genes represented in this pathway, and is known to be overexpressed in SCLC and to drive expression of EZH2 which is also overexpressed in SCLC [92, 93]. To explore this pathway further, we investigated whether EZH2 was more highly expressed in SqCC than AC tumors. As expected, we found that EZH2 was expressed at a significantly higher level in SqCC tumors than AC tumors (Mann Whitney U test p<0.0001), demonstrating the biological consequence of E2F1 disruption (Figure 2.4B). These findings could reflect the key role of EZH2 and the polycomb group, a protein complex involved in DNA methylation [92] in the pathogenesis of SqCC.                 28  Figure 2.4     Figure 2.4. Epigenetically altered SqCC genes are significantly enriched for SCLC signaling. C B A 29  Figure 2.4. Epigenetically altered SqCC genes are enriched for SCLC signaling.  (A) SCLC signaling components are altered at levels of the pathway by DNA methylation in SqCC. Hypomethylated and overexpressed genes are shown in red, hypermethylated and underexpressed genes are shown in green. (B) EZH2 expression in 58 AC tumors and 53 SqCC tumors. As predicted, EZH2 expression is higher in SqCC tumors compared to AC tumors,  Mann-Whitney U test (p<0.0001). (C) FHIT is significantly more methylated in SqCC (n=13) and AC (n=30) tumors (Mann-Whitney U test, p<0.05).    2.3.6 Concerted genetic and epigenetic disruption of subtype- specific genes  In order to determine if both DNA copy number and DNA methylation aberrations simultaneously disrupted any genes, we combined the subtype-specific gene lists derived using the two analytical approaches described above. None of the 71 genes altered by copy number in AC or the 32 genes associated with DNA methylation overlapped. This was not surprising given the observed lack of similarity at the level of function/network analysis, and could be due in part to the extensive heterogeneity within AC. Combining the 447 copy-number associated genes with the 297 DNA methylation genes from SqCC yielded an overlap of 38 genes (Table 2.2). These genes exhibit frequent concurrent genetic, epigenetic, and subsequent gene-expression alterations that discriminate them from AC tumors. Notably, loss of the well-known 3p tumor suppressor gene (TSG) FHIT which is associated with smoking  and has been investigated as a potential biomarker for centrally-occurring lung cancers (Figure 2.4c), and amplification of the recently identified lineage specific oncogene BRF2 were among these genes, highlighting their importance in the development of SqCC. In addition to genes previously associated with lung cancer, TSGs and oncogenes known to be deregulated in other cancer types such as PRDM2 and SIAH2 were also altered at the genetic, epigenetic and gene expression levels. Interestingly, seven genes (ATP2C1, PCYT1A, ZWILCH, CENTB2, BAG4, PARP11 and CSDA) were disrupted by gene-dosage in one subtype and DNA methylation in the other. Such multidimensional disruption is indicative of strong selective pressures, suggesting these genes may play a pivotal role in tumor biology.    30  Table 2.2. Genes demonstrating concerted genetic and epigenetic disruption in SqCC Gene SqCC Copy Number Status SqCC Methylation Status CFI Loss and Underexpressed Hypermethylated and Underexpressed CYFIP2 Loss and Underexpressed Hypermethylated and Underexpressed FHIT Loss and Underexpressed Hypermethylated and Underexpressed GLRB Loss and Underexpressed Hypermethylated and Underexpressed LRP2BP Loss and Underexpressed Hypermethylated and Underexpressed LTC4S Loss and Underexpressed Hypermethylated and Underexpressed OTUD4 Loss and Underexpressed Hypermethylated and Underexpressed PRDM2 Loss and Underexpressed Hypermethylated and Underexpressed SGPP2 Loss and Underexpressed Hypermethylated and Underexpressed ARMC8 Gained and Overexpressed Hypomethylated and Overexpressed BAG4 Gained and Overexpressed Hypomethylated and Overexpressed BRF2 Gained and Overexpressed Hypomethylated and Overexpressed C3orf26 Gained and Overexpressed Hypomethylated and Overexpressed CENTB2 Gained and Overexpressed Hypomethylated and Overexpressed CSDA Gained and Overexpressed Hypomethylated and Overexpressed EXOSC5 Gained and Overexpressed Hypomethylated and Overexpressed FBXO27 Gained and Overexpressed Hypomethylated and Overexpressed KLK13 Gained and Overexpressed Hypomethylated and Overexpressed MGC2408 Gained and Overexpressed Hypomethylated and Overexpressed MRPL51 Gained and Overexpressed Hypomethylated and Overexpressed NCK1 Gained and Overexpressed Hypomethylated and Overexpressed PCYT1A Gained and Overexpressed Hypomethylated and Overexpressed PDXP Gained and Overexpressed Hypomethylated and Overexpressed PPAN Gained and Overexpressed Hypomethylated and Overexpressed RAD51AP1 Gained and Overexpressed Hypomethylated and Overexpressed RECQL Gained and Overexpressed Hypomethylated and Overexpressed RELB Gained and Overexpressed Hypomethylated and Overexpressed RNF185 Gained and Overexpressed Hypomethylated and Overexpressed RPL35A Gained and Overexpressed Hypomethylated and Overexpressed RPS27A Gained and Overexpressed Hypomethylated and Overexpressed RSRC1 Gained and Overexpressed Hypomethylated and Overexpressed SAE1 Gained and Overexpressed Hypomethylated and Overexpressed SFRS10 Gained and Overexpressed Hypomethylated and Overexpressed SHOX2 Gained and Overexpressed Hypomethylated and Overexpressed SIAH2 Gained and Overexpressed Hypomethylated and Overexpressed SMEK2 Gained and Overexpressed Hypomethylated and Overexpressed SNRPB Gained and Overexpressed Hypomethylated and Overexpressed SNRPD2 Gained and Overexpressed Hypomethylated and Overexpressed    31  2.3.7 Subtype specific genes are responsible for AC and SqCC phenotypes  Next, we sought to confirm that the genes differentially disrupted at the genomic and epigenomic level are responsible for the different biological characteristics of AC and SqCC. Since they are regulated by subtype specific alterations, we hypothesized that the expression levels of these genes should be able to accurately segregate NSCLC tumors into distinct AC and SqCC groups. As predicted, when using the expression values for the 49 NSCLC tumors from our data set, principle component analysis with the 778 unique genetically and/or epigenetically deregulated genes clearly delineated distinct subtype specific clusters (Figure 2.5a). A receiver operating characteristic (ROC) area under the curve (AUC) value of 0.9690 (P<0.0001) confirmed that principle component 1 was a strong discriminator of the subtypes (Figure 2.5a). Given that these genes were uncovered based on differences between the subtypes, this finding was not surprising. Therefore, to validate the role of these genes in subtype development, we applied the same analysis to two independent publically available sample sets. The first consisted of 111 (58 AC and 53 SqCC, Sample Set #3) and the second of 138 clinical lung tumors (62 AC and 76 SqCC, Sample Set #6) [82]. Strikingly, this analysis was also able to separate the AC and SqCC samples with a substantial accuracy (ROC AUC values of 0.9076 and 0.9442, P<0.0001, respectively) (Figure 2.5b and c). Validation of our subtype specific genes in these large, independent panels of NSCLC tumors from separate institutions provides further evidence that the genes regulated by subtype specific genomic and epigenomic disruptions are responsible for driving the differential development of AC and SqCC. Furthermore, our results highlight the impact of this novel integrative genome, epigenome and transcriptome analysis in identifying robust target genes that can be used as biomarkers of disease.         32  Figure 2.5    Figure 2.5. Subtype specific genes explain AC and SqCC phenotypes.    A B C 33  Figure 2.5. Subtype specific genes explain AC and SqCC phenotypes. Principal components analysis of all genes demonstrating expression differences between subtypes as a result of genetic and/or epigenetic alterations in: (A) 49 NSCLC tumors (29 AC, 20 SqCC) used for gene discovery in this study, publically available data from (B) 111 NSCLC tumors (58 AC and 53 SqCC, Sample Set #3-Duke,) and (C) 138 NSCLC tumors (62 AC and 76 SqCC, Sample Set #6- Samsung,). Red circles indicate AC samples, blue circles indicate SqCC samples. Strong separation of tumors along principal component 1 is observed in all sets, demonstrating the contribution of these genes to the differential phenotypes. The respective ROC curves for each dataset using the respective principle component 1 values for each sample are shown on the right. AUC values of 0.9690, 0.9076 and 0.9442 for A, B and C, respectively, suggest the gene expression signature is accurately discriminates subtypes.   2.3.8 Subtype-specific genetic differences are translated to the protein level  To confirm that the genome and transcriptome differences between subtypes affects the relative protein levels of the identified genes, we performed immunohistochemical (IHC) analysis on an independent panel of >200 lung tumors. Protein levels for three subtype specific genes with available antibodies validated for IHC were analyzed: ERCC1 (inactivated in AC), KEAP1 (inactivated in AC) and SOX2 (activated in SqCC) (Figure 2.6). Average protein levels for all three genes were significantly different between subtypes in the direction predicted by our integrative analysis (Figure 2.6a, d, and g, p<0.001, unpaired t test with Welch's correction). The average nuclear ERCC1 expression and cytoplasmic KEAP1 expression were significantly lower in AC tumors (ERCC1:43.45±5.389, N = 175 and KEAP1:126.5 ±4.179, N = 184) compared to SqCC tumors (ERCC1: 79.99±9.095, N = 106 and KEAP1:160.9 ±5.401, N = 110) consistent with these genes being inactivated in AC. Conversely, nuclear levels of SOX2, a SqCC lineage specific oncogene were significantly higher in SqCC (206.5±8.839, N = 106) than AC (70.39± 6.342, N = 170). These data demonstrate that the genomic, epigenomic and expression differences between subtypes are translated to the protein level, supporting the hypothesis that these changes have a functional consequence on the phenotypes of AC and SqCC.  34  Figure 2.6    Figure 2.6. Subtype-specific genomic differences are reflected at the protein level.  Immunohistochemical analysis of protein levels for ERCC1 (A–C), KEAP1 (D–F) and SOX2 (G–I) in squamous and adenocarcinoma lung tumors. Average immunohistochemical protein expression levels for each subtype are plotted ± SEM of each group. Representative microphotographs showing tumoral cells (arrows) with higher levels of expression of nuclear ERCC1 (B and C), cytoplasmic KEAP1 (E and F) and nuclear SOX2 (H and I) in squamous cell carcinomas (B, E and H) compared to lung adenocarcinomas (C, F, and I). Images are of samples reflecting the average protein expression for each group (ERCC1: SqCC = ~80, AC =~43; KEAP1: SqCC = ~161, AC = ~126; SOX2: SqCC = ~207, AC = ~70). Magnification 200x. * and ** = p<0.001 and p<0.0001, two-tailed unpaired t test with Welch’s correction, respectively.  35  2.3.9 Subtype-specific genes are associated with distinct clinical characteristics   In an attempt to elucidate the influence of our subtype specific genes on the clinical characteristics of AC and SqCC we determined the survival associations using a Mantel-Cox log rank test for each of the 778 subtype specific genes in AC, SqCC and NSCLC as a whole in sample set #6 which has overall survival information available for 138 cases (62 AC and 76 SqCC). Since these genes are responsible for defining the distinct biology of these diseases, we reasoned that their expression should only correlate with clinical features in one subtype and not the other subtype or NSCLC (AC + SqCC) as a whole. 131 AC and 46 SqCC specific genes had significant (p <0.05) associations with overall survival (Table 2.3). Remarkably, the associations were completely specific to an individual subtype as no genes were correlated with survival in the same manner across both subtypes. Six genes (DSG2, PLAC2, ATP9A, TPM4, CD9 and PSMD11) were significantly associated with survival in both subtypes; however, they displayed opposing survival patterns, with low expression associated with poor survival in one subtype and high expression with poor survival in the other (Figure 2.7). Thus, although associated with survival in both subtypes, these genes exhibit distinct subtype-specific associations. Interestingly, low levels of CD9 expression have been previously implicated in the poor prognosis of lung cancer patients [94]. However, we now show that this association is subtype specific, with low levels of CD9 correlated with poor prognosis in SqCC and high levels with poor prognosis in AC (Figure 2.7). Importantly, only eight genes that were associated with survival in one of the subtypes were also significant when analyzing NSCLC as a whole, providing further evidence to the importance of treating the subtypes as separate disease entities and underscoring the potential clinical relevance of subtype specific alterations. 36  Table 2.3 Genes demonstrating a significant association with survival   AC (n=62) NSCLC (n=138) SqCC (n=76) Probe Set ID Gene Symbol P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) 235053_at --- 0.0144 16.9405 49.44151 0.17068 24.50015 49.4357 0.53886 39.27745 Undefined 238277_at --- 0.03738 33.81425 Undefined 0.71096 49.43922 45.26797 0.3616 51.27054 34.00295 235288_at --- 0.00853 24.33483 Undefined 0.51286 51.27393 Undefined 0.25026 51.27537 34.00406 235772_at --- 0.00036 21.33842 Undefined 0.2415 34.00001 42.80249 0.30358 45.27746 31.14234 240969_at --- 0.14949 24.50005 Undefined 0.57797 51.27273 39.00354 0.01258 Undefined 23.63333 1559360_at --- 0.37749 49.4449 27.2 0.0207 Undefined 29.13404 0.03775 Undefined 28.8752 240527_at --- 0.38033 49.44523 Undefined 0.3844 55.36737 39.00058 0.04488 Undefined 23.63478 238580_at --- 0.31036 33.8017 42.806 0.10849 Undefined 28.10062 0.02382 Undefined 28.10747 225318_at --- 0.01158 21.33602 Undefined 0.00396 24.50067 Undefined 0.13721 28.86847 Undefined 236010_at --- 0.02674 33.80404 Undefined 0.85402 39.26732 42.80081 0.32808 39.26707 31.14145 238281_at --- 0.00073 19.83424 Undefined 0.30314 28.86707 42.80193 0.12523 Undefined 31.13674 1560625_s_at --- 0.35725 39.60002 24.5 0.10421 49.43598 28.36731 0.00727 Undefined 28.1056 202394_s_at ABCF3 0.04375 Undefined 24.13333 0.73065 49.43428 51.27379 0.24315 34.00969 Undefined 212895_s_at ABR 0.73575 42.83241 28.36667 0.95439 42.8039 39.26733 0.03852 17.0087 Undefined 1554390_s_at ACTR2 0.0024 Undefined 22.57662 0.41497 42.80818 34.00004 0.42825 31.13862 51.27873 209321_s_at ADCY3 0.37955 39.00968 24.50383 0.07318 39.0035 Undefined 0.01947 14.46667 Undefined 205268_s_at ADD2 0.0208 24.33711 Undefined 0.05083 34.00914 51.26791 0.18153 34.00443 Undefined 223145_s_at AKIRIN2 0.0042 Undefined 19.83365 0.39962 45.26702 39.27015 0.32392 39.60958 Undefined 224982_at AKT1S1 0.03429 Undefined 21.33333 0.26164 Undefined 33.80263 0.42188 Undefined 39.27398 236626_at ALG1 0.28678 33.80035 55.3923 0.38036 Undefined 34.00052 0.04702 Undefined 28.1 1559640_at ANKFN1 0.03854 24.1401 Undefined 0.20954 28.36686 51.2675 0.6613 Undefined 39.6 208074_s_at AP2S1 0.02846 Undefined 24.33346 0.0408 Undefined 31.1347 0.55993 Undefined 39.27003 206632_s_at APOBEC3B 0.47521 49.4339 33.80355 0.10034 39.60002 Undefined 0.01601 28.87687 Undefined 37    AC (n=62) NSCLC (n=138) SqCC (n=76) Probe Set ID Gene Symbol P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) 223665_at ARPM1 0.0041 55.37556 16.93581 0.03967 Undefined 28.86709 0.81962 33.8 31.90115 202872_at ATP6V1C1 0.04422 39.60013 21.33399 0.60693 45.27069 33.80082 0.79845 45.27323 51.29902 216129_at ATP9A 0.0126 27.20467 Undefined 0.85486 51.26829 42.80025 0.01832 Undefined 28.1 220488_s_at BCAS3 0.0401 27.21179 55.39337 0.37157 28.36738 39.60042 0.75453 39.60025 33.80295 229177_at C16orf89 0.02698 29.14723 55.37189 0.53278 31.90096 45.2675 0.7107 39.27301 34.00405 219114_at C3orf18 0.02553 20.30096 Undefined 0.14933 28.86671 42.80078 0.80837 45.273 Undefined 220218_at C9orf68 0.03902 29.13582 Undefined 0.80718 49.44241 39.60069 0.21869 Undefined 31.13333 1553693_s_at CBR4 0.10425 33.57563 Undefined 0.94272 39.26718 29.13397 0.0402 Undefined 23.63478 202047_s_at CBX6 0.03347 24.33355 Undefined 0.00724 24.3334 Undefined 0.20137 28.86884 Undefined 1553886_at CCDC108 0.2563 27.2001 49.43969 0.37501 51.26785 42.80266 0.00775 Undefined 14.4692 226723_at CCDC23 0.84863 49.46235 39.00205 0.0259 Undefined 28.3667 0.00063 Undefined 21.00575 201005_at CD9 0.02904 Undefined 22.56753 0.53697 39.00059 51.2691 0.03468 28.87073 Undefined 235117_at CHAC2 0.02397 49.43635 19.83681 0.85201 49.44391 51.26889 0.52374 39.27526 Undefined 218566_s_at CHORDC1 0.0101 Undefined 19.1687 0.50217 42.8012 29.13335 0.14139 31.90955 51.28971 208925_at CLDND1 0.0018 Undefined 22.57001 0.4322 Undefined 45.2716 0.67328 39.61269 Undefined 230609_at CLINT1 0.03465 24.3392 Undefined 0.57718 39.27082 49.43703 0.05689 39.27138 17.00452 216295_s_at CLTA 0.00431 Undefined 24.13618 0.75029 45.26681 39.60227 0.15425 23.63333 Undefined 206158_s_at CNBP 0.02958 Undefined 22.56982 0.86081 49.43538 51.27565 0.36265 31.13497 Undefined 221676_s_at CORO1C 0.00315 Undefined 21.33395 0.57564 39.60005 51.27283 0.30983 28.1 51.28618 217889_s_at CYBRD1 0.01382 Undefined 29.13997 0.90114 45.26687 51.27134 0.25109 28.1 51.26834 223385_at CYP2S1 0.68764 28.36735 29.13414 0.16294 28.86693 Undefined 0.02874 17 Undefined 226745_at CYP4V2 0.3883 Undefined 39.61238 0.04372 Undefined 39.00442 0.6401 28.10529 39.27164 1555301_a_at DIP2A 0.88767 Undefined 55.38224 0.08636 Undefined 31.13481 0.02417 Undefined 31.14654 202514_at DLG1 0.00817 Undefined 24.33561 0.65233 Undefined 51.27229 0.75011 34.00781 Undefined 38    AC (n=62) NSCLC (n=138) SqCC (n=76) Probe Set ID Gene Symbol P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) 213707_s_at DLX5 0.02338 20.30062 42.80028 0.44252 33.57163 33.80075 0.23382 34.00095 Undefined 209187_at DR1 0.11236 Undefined 39.00756 0.4124 Undefined 39.26692 0.04499 Undefined 23.63478 1553105_s_at DSG2 0.00731 Undefined 21.33333 0.57155 Undefined 51.27133 0.01533 31.13771 Undefined 204455_at DST 0.02439 Undefined 24.5 0.47177 55.36794 45.27146 0.94676 39.60211 51.30379 227103_s_at ECE2 0.03216 Undefined 24.50024 0.61781 45.2674 39.26885 0.27278 23.63333 51.28227 219787_s_at ECT2 0.04794 Undefined 27.20217 0.96564 39.60003 45.27416 0.9103 39.27167 31.9 227540_at EEFSEC 0.12983 33.80965 55.37667 0.08076 33.80235 55.36909 0.03373 23.63478 Undefined 208264_s_at EIF3J 0.01706 Undefined 19.83377 0.45044 Undefined 33.80004 0.78744 39.26773 33.80045 210213_s_at EIF6 0.01797 Undefined 21.33505 0.24002 42.80152 28.86696 0.64135 34.00444 28.86746 1555274_a_at EPT1 0.00473 Undefined 24.13482 0.79192 55.36992 51.26965 0.37729 39.27466 51.27142 224576_at ERGIC1 0.13149 Undefined 28.36758 0.0231 Undefined 29.1339 0.59119 39.27742 33.80355 218481_at EXOSC5 0.01974 49.43491 21.33395 0.31024 49.43857 33.80179 0.96215 45.27775 33.8022 215133_s_at FAM153A/B 0.10274 Undefined 22.56667 0.03727 Undefined 23.63367 0.07532 Undefined 31.13787 211623_s_at FBL 0.0136 Undefined 22.56795 0.26212 55.37016 33.8006 0.69323 Undefined 51.2858 225737_s_at FBXO22 0.02196 Undefined 19.16994 0.60894 39.60087 33.80066 0.83565 34.00913 39.27489 228220_at FCHO2 0.00771 42.80089 20.3 0.53555 33.80002 34.00015 0.11278 17.00242 51.27968 218880_at FOSL2 0.03719 42.80044 21.3341 0.11227 Undefined 31.1347 0.86562 34.00245 39.27724 219170_at FSD1 0.48502 39.00755 24.5 0.19504 33.80163 Undefined 0.02367 28.10747 Undefined 205384_at FXYD1 0.02207 24.50712 Undefined 0.11755 28.36939 51.26727 0.62167 33.81213 45.27641 217398_x_at GAPDH 0.00599 Undefined 21.33395 0.29288 Undefined 33.80167 0.94368 Undefined Undefined 225161_at GFM1 0.00928 Undefined 19.1687 0.21403 Undefined 39.26803 0.52363 34.01822 31.13856 218473_s_at GLT25D1 0.02424 Undefined 20.3 0.73697 45.27217 49.44348 0.24779 33.80179 51.26929 235678_at GM2A 0.23687 42.81498 19.83333 0.61717 42.80336 Undefined 0.0219 22.5 Undefined 214431_at GMPS 0.0474 Undefined 27.2023 0.43364 49.43901 33.80168 0.29773 Undefined 31.90064 39    AC (n=62) NSCLC (n=138) SqCC (n=76) Probe Set ID Gene Symbol P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) 210761_s_at GRB7 0.01647 Undefined 28.37302 0.33651 45.27118 33.80045 0.23033 39.60983 Undefined 213911_s_at H2AFZ 0.01466 Undefined 24.50124 0.94486 49.43829 39.60273 0.48193 39.27751 33.80045 206194_at HOXC4 0.05237 24.33779 Undefined 0.02445 24.33492 Undefined 0.18757 22.50201 39.60027 217805_at ILF3 0.00013 Undefined 24.33919 0.06694 Undefined 33.80253 0.80721 34.00397 28.1 201389_at ITGA5 0.02179 Undefined 20.3 0.09358 42.80019 28.86712 0.79504 33.80209 39.27456 239695_at JAK1 0.0454 29.13355 Undefined 0.48185 34.00013 42.80262 0.68832 31.9 45.27641 202417_at KEAP1 0.05737 Undefined 27.2 0.91753 45.26791 49.4404 0.04781 39.6006 Undefined 213208_at KIAA0240 0.43129 24.50348 39.62181 0.02893 22.50118 55.36901 0.03511 28.10483 Undefined 226328_at KLF16 0.01845 Undefined 22.56667 0.79878 39.00306 45.27031 0.06192 33.81203 Undefined 221986_s_at KLHL24 0.03703 21.33523 Undefined 0.02229 24.13362 Undefined 0.11447 28.87458 51.27145 225267_at KPNA4 0.03218 Undefined 24.3359 0.64512 Undefined 39.60009 0.49027 45.27188 33.80006 34031_i_at KRIT1 0.08207 42.80441 33.57252 0.92975 39.6014 39.60017 0.04527 28.87493 Undefined 209008_x_at KRT8 0.0298 Undefined 22.56667 0.18157 55.37109 28.1 0.81381 39.27259 28.1 216952_s_at LMNB2 0.01254 Undefined 12.46667 0.91158 55.37024 Undefined 0.29715 34.00048 Undefined 240936_at LOC100287290 0.0142 22.56894 Undefined 0.92029 45.27692 39.00242 0.36183 51.27015 31.90132 229187_at LOC283788 0.28723 28.36682 Undefined 0.20841 Undefined 39.60002 0.027 Undefined 34.00403 202736_s_at LSM4 0.00474 Undefined 24.34862 0.71931 39.60217 33.80062 0.45598 34.0006 Undefined 224656_s_at LUZP6/MTPN 0.00987 Undefined 29.13705 0.60054 Undefined 34.00004 0.09512 39.61736 Undefined 242838_at MAP6D1 0.3869 42.81631 Undefined 0.05706 33.56682 Undefined 0.03499 34.00369 Undefined 200712_s_at MAPRE1 0.00286 Undefined 12.46868 0.40758 49.43642 34.00003 0.29926 17 39.26873 204825_at MELK 0.02125 55.37232 22.5686 0.53408 49.43576 33.80078 0.58503 51.30354 Undefined 224725_at MIB1 0.00118 Undefined 19.17769 0.3179 55.36692 33.80026 0.62303 34.0042 Undefined 231975_s_at MIER3 0.00445 Undefined 28.37263 0.0452 Undefined 33.80143 0.40189 39.60096 34.00044 224784_at MLLT6 0.58994 29.13854 39.00102 0.06093 31.90083 49.43717 0.00102 17 Undefined 40    AC (n=62) NSCLC (n=138) SqCC (n=76) Probe Set ID Gene Symbol P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) 223086_x_at MRPL51 0.00434 Undefined 21.33395 0.45839 42.80053 33.80126 0.5265 39.60044 51.27282 204331_s_at MRPS12 0.00082 Undefined 24.13416 0.45616 49.43749 39.27097 0.09454 34.02697 Undefined 213380_x_at MST1P9 0.81968 22.57145 33.80789 0.04798 Undefined 33.80322 0.00081 Undefined 22.50201 205455_at MST1R 0.69461 24.50304 33.8026 0.19801 Undefined 39.00408 0.01155 Undefined 34.0004 226856_at MUSTN1 0.03969 21.33717 49.43585 0.62207 42.80249 51.26764 0.4065 Undefined 51.27849 228846_at MXD1 0.00339 Undefined 19.83333 0.01847 Undefined 22.50033 0.6546 Undefined 39.63075 212462_at MYST4 0.10006 28.37113 Undefined 0.01361 28.86671 Undefined 0.64988 39.60051 Undefined 217745_s_at NAA50 0.03094 Undefined 28.36976 0.89708 49.43981 45.27204 0.8809 39.60787 33.8019 204725_s_at NCK1 0.0232 Undefined 29.14091 0.45144 45.26871 Undefined 0.08955 31.13754 Undefined 208969_at NDUFA9 0.01447 49.43586 21.33395 0.52469 42.8016 33.80242 0.82327 39.26987 39.60899 219396_s_at NEIL1 0.04625 27.22196 55.37556 0.20662 33.56729 49.43815 0.40991 31.1358 Undefined 218036_x_at NMD3 0.00253 Undefined 24.33426 0.24537 Undefined 33.80111 0.73953 45.2738 39.27156 205129_at NPM3 0.00325 Undefined 24.5 0.16436 Undefined 31.90049 0.80558 51.27316 33.80546 207740_s_at NUP62 0.03487 Undefined 24.33683 0.18146 55.37659 31.90033 0.95715 34.00119 39.27398 215952_s_at OAZ1 0.03153 Undefined 28.37069 0.83043 45.26775 51.26712 0.14832 31.90753 Undefined 1567245_at OR5J2 0.55999 39.00086 Undefined 0.29774 39.00273 24.33406 0.00638 Undefined 14.13568 220669_at OTUD4 0.28954 39.61079 27.2 0.00609 Undefined 31.13369 0.03617 Undefined 31.90056 207634_at PDCD1 0.05585 28.36714 Undefined 0.03491 31.13376 Undefined 0.48531 39.62476 Undefined 202212_at PES1 0.00027 Undefined 19.16994 0.1771 Undefined 39.26986 0.76431 34.00568 33.80504 201600_at PHB2 0.00212 Undefined 12.46749 0.26775 49.43491 31.90047 0.58147 39.6005 33.8036 226846_at PHYHD1 0.85006 49.47308 33.80646 0.12844 Undefined 33.80158 0.02709 Undefined 31.90501 204297_at PIK3C3 0.01335 Undefined 24.14054 0.14753 45.26707 29.13394 0.59912 Undefined 33.8 202522_at PITPNB 0.00512 42.80212 11.6 0.44106 42.80204 31.90069 0.61958 34.00213 39.27623 219584_at PLA1A 0.1342 24.13645 Undefined 0.01409 28.1017 Undefined 0.30541 39.28778 51.273 41    AC (n=62) NSCLC (n=138) SqCC (n=76) Probe Set ID Gene Symbol P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) 244374_at PLAC2 0.0089 24.33367 Undefined 0.9082 39.00066 34.00004 0.02157 Undefined 31.14163 201411_s_at PLEKHB2 0.02874 Undefined 27.20319 0.15134 Undefined 29.13356 0.37245 33.80401 51.27426 217841_s_at PPME1 0.02953 Undefined 28.36886 0.13805 Undefined 39.26963 0.69668 Undefined 33.80063 201979_s_at PPP5C 0.01205 Undefined 27.20299 0.12645 Undefined 33.80025 0.83953 39.6096 33.80045 203056_s_at PRDM2 0.03028 29.13442 Undefined 0.26908 31.13423 45.27053 0.26807 34.00161 28.10422 200707_at PRKCSH 0.0038 Undefined 27.20319 0.94375 39.60022 45.26917 0.38645 33.80392 45.27737 206445_s_at PRMT1 0.01499 Undefined 27.20435 0.14026 Undefined 33.80131 0.29468 Undefined 33.80116 201267_s_at PSMC3 0.00053 Undefined 29.14001 0.09155 Undefined 34 0.97923 39.60516 39.2773 208777_s_at PSMD11 0.03359 Undefined 28.36795 0.91385 49.43483 55.36745 0.03752 34.01822 Undefined 204748_at PTGS2 0.12298 Undefined 29.13333 0.03916 Undefined 31.90193 0.15653 Undefined 28.87493 222981_s_at RAB10 0.03622 49.44053 19.83377 0.98263 49.43854 51.27597 0.58024 34.00554 Undefined 202483_s_at RANBP1 0.00028 Undefined 19.16994 0.29456 55.37286 39.27082 0.69617 39.60904 33.8021 1552482_at RAPH1 0.85588 28.37231 39.00947 0.24756 Undefined 39.60089 0.0052 Undefined 34.00609 205091_x_at RECQL 0.01306 Undefined 29.14178 0.65705 45.26784 39.60024 0.56676 45.27521 39.27466 205205_at RELB 0.00024 Undefined 12.46826 0.0613 42.80261 28.10416 0.42162 31.90878 51.28158 220334_at RGS17 0.9146 49.44602 29.13333 0.19007 49.43744 28.36678 0.00478 Undefined 21.00575 202129_s_at RIOK3 0.01461 Undefined 29.13465 0.93382 42.80218 34.00016 0.14088 23.63333 51.29828 203022_at RNASEH2A 0.01996 Undefined 27.20329 0.91781 39.60158 39.60411 0.86278 39.27691 31.1415 204208_at RNGTT 0.03237 Undefined 29.13469 0.12068 Undefined 33.80109 0.63752 Undefined 51.28576 234243_at RPF1 0.01228 19.83474 Undefined 0.91676 39.26867 42.80095 0.10884 39.26785 22.5 200022_at RPL18 0.04181 Undefined 27.20692 0.02266 Undefined 28.10041 0.63984 Undefined Undefined 230695_s_at RSPH9 0.21092 33.81375 33.56745 0.28838 33.80153 33.57037 0.04879 31.13333 Undefined 230378_at SCGB3A1 0.59364 33.80097 55.38584 0.09802 31.13412 55.36957 0.00959 28.88063 Undefined 238017_at SDR16C5 0.24231 42.81232 24.33362 0.03907 Undefined 34.00241 0.39325 Undefined 34.00689 42    AC (n=62) NSCLC (n=138) SqCC (n=76) Probe Set ID Gene Symbol P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) 57703_at SENP5 0.00984 Undefined 22.56849 0.88425 49.43804 39.6003 0.4657 34.00108 31.13936 205637_s_at SH3GL3 0.03148 22.56814 Undefined 0.06137 29.13336 Undefined 0.99496 51.2827 45.27023 214437_s_at SHMT2 0.00877 Undefined 24.5004 0.45624 49.43966 33.80165 0.71676 39.27181 31.14163 227791_at SLC9A9 0.41764 33.80973 55.38054 0.05746 28.36677 Undefined 0.04263 28.10127 Undefined 233759_s_at SMEK2 0.04774 Undefined 29.13447 0.57193 39.0003 51.27423 0.11077 39.27633 Undefined 202690_s_at SNRPD1 0.02425 Undefined 27.20323 0.42348 55.36969 33.80201 0.75276 51.27334 39.60742 200826_at SNRPD2 0.00192 Undefined 24.13475 0.0512 Undefined 31.90002 0.20704 Undefined 28.86667 208608_s_at SNTB1 0.00764 Undefined 24.50541 0.11671 Undefined 33.56895 0.31805 39.27284 Undefined 234005_x_at STK36 0.4891 Undefined 29.13392 0.04193 Undefined 33.80256 0.2361 Undefined 31.90206 200870_at STRAP 0.01224 Undefined 22.56753 0.88866 55.37286 51.27355 0.30975 39.60044 Undefined 217834_s_at SYNCRIP 0.00745 Undefined 27.20213 0.09284 49.43463 31.13375 0.83177 45.29099 51.2942 217839_at TFG 0.00023 Undefined 11.6 0.10494 Undefined 33.8009 0.78479 Undefined 51.28579 207332_s_at TFRC 0.00253 Undefined 19.1687 0.93278 42.80181 51.26735 0.50376 39.6108 33.8 203235_at THOP1 0.54289 39.60594 27.20092 0.56371 39.60616 49.43707 0.03206 34.01432 Undefined 1552522_at TIGD4 0.08266 24.14291 49.43914 0.46013 Undefined 34.00004 0.0356 Undefined 34.00185 224413_s_at TM2D2 0.04324 42.80067 28.36803 0.56527 45.2677 33.80005 0.60526 31.9124 39.60534 236430_at TMED6 0.03763 24.50075 55.37892 0.39659 34.00011 55.37017 0.62181 33.80344 39.62128 225766_s_at TNPO1 0.00087 Undefined 24.1406 0.05921 Undefined 31.13516 0.82285 Undefined 33.80204 217960_s_at TOMM22 0.00386 Undefined 24.34083 0.42893 Undefined 39.60017 0.71248 39.60655 39.27185 201512_s_at TOMM70A 0.00499 Undefined 19.1687 0.20854 Undefined 39.26963 0.88668 34.0042 39.27171 213011_s_at TPI1 0.0212 Undefined 22.56767 0.26103 Undefined 31.13388 0.79194 Undefined Undefined 212481_s_at TPM4 0.02596 Undefined 22.57022 0.63006 55.37739 39.27207 0.02658 17 Undefined 236020_s_at TRUB1 0.0006 24.33376 Undefined 0.01604 29.13403 Undefined 0.76752 31.13333 39.60318 1557073_s_at TTBK2 0.65558 55.46281 39.00417 0.72421 39.60268 51.27788 0.03372 31.13436 Undefined 43    AC (n=62) NSCLC (n=138) SqCC (n=76) Probe Set ID Gene Symbol P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) P Value Median Survival (Low) Median Survival (High) 211337_s_at TUBGCP4 0.0071 Undefined 12.46667 0.68162 55.37233 51.27741 0.20628 34.00365 Undefined 214007_s_at TWF1 0.02118 Undefined 11.60108 0.7189 42.80104 39.60032 0.09158 31.13565 Undefined 222601_at UBA6 0.04354 39.60059 22.56767 0.11044 Undefined 33.80161 0.46036 51.29911 45.27548 227413_at UBLCP1 0.01242 42.80171 24.13996 0.11705 Undefined 34.00001 0.80738 45.26912 Undefined 225655_at UHRF1 0.02933 Undefined 27.20192 0.44667 Undefined 39.60386 0.7462 39.27203 33.80063 216775_at USP53 0.37925 27.20326 33.80577 0.05508 51.26675 29.13395 0.00277 Undefined 17 229369_at VSIG2 0.01852 20.30096 Undefined 0.30675 31.90264 45.26791 0.60017 39.27796 34.00913 219060_at WDYHV1 0.02266 Undefined 24.33353 0.68566 49.44061 39.267 0.19027 39.60983 Undefined 208775_at XPO1 0.0335 Undefined 29.13592 0.84698 Undefined 51.27005 0.79137 34.01252 33.80342 207757_at ZFP2 0.00983 21.33607 Undefined 0.65835 31.90012 42.80369 0.17654 39.26719 34.01328 241793_at ZMYND17 0.44512 33.56949 28.36771 0.55572 49.43697 27.20028 0.02879 Undefined 22.50407 232117_at ZNF471 0.16299 Undefined 22.56779 0.01829 Undefined 33.57015 0.02274 Undefined 22.5 238454_at ZNF540 0.02493 28.37374 Undefined 0.15849 31.13421 42.80043 0.81279 33.81307 34.00037 218349_s_at ZWILCH 0.01393 Undefined 27.20322 0.70174 42.80079 39.26778 0.44344 34.01252 51.27434 44  Figure 2.7    Figure 2.7 CD9 expression is associated with differences in survival in AC and SqCC tumors.  The prognostic value of CD9 expression levels was evaluated in 53 SqCC tumors and 58 AC tumors. Poor prognosis is significantly associated (Mantel-Cox log test, p<0.05) with low CD9 levels in SqCC (A) and  high CD9 in AC (B). Survival of the 1/3 lowest CD9 expressers is shown in red, and the top 1/3 is shown in blue.    2.3.10 Defining putative treatment strategies tailored to lung cancer subtypes using in silico screening of candidate therapeutic compounds  Lastly, after defining and validating our AC and SqCC specific cancer genes, we applied these findings to define potential treatment strategies tailored to each lung cancer subtype. Using our subtype specific genes, we queried the Connectivity Map (CMAP) database to identify compounds that could "reverse" the expression direction of each subtype signature. CMAP consists of thousands of gene expression profiles from different cancer cell lines treated with a vast collection of small molecules [83]. By comparing the subtype specific signatures of up and down regulated genes with preexisting response signatures, CMAP identifies small molecules whose effects on gene expression changes are positively or negatively correlated. Negative correlation scores imply that the matched molecules have a mode of action that can reverse the expression direction of query genes, and therefore serve as potential therapeutic compounds. Using this in silico screening approach, we identified numerous instances (cell line/treatment B A 45  combination) that were significantly correlated with both our AC and SqCC specific gene signatures. SqCC had an expression signature that was negatively correlated with multiple HDAC and PI3K/ mTOR inhibitors including trichostatin A, vorinostat (also known as SAHA), LY-294002 and MS-275 - all HDAC inhibitors as well as quinostatin, sirolimus (also known as Rapamycin) and wortmannin - all PI3K/mTOR inhibitors. These findings were interesting given that the alteration of histone modifying enzymes was the major network disrupted in SqCC and we observed concerted disruption of PRC2 components, which are responsible for de novo methylation. In addition, PIK3CA activation (mutation and/or amplification) is known to occur more frequently in SqCC than AC [95] and many downstream components of this pathway were also altered specifically in SqCC. Conversely, CMAP analysis for AC was not very informative, as none of the negatively correlated molecules shared the same functions.    To confirm the results of the CMAP analysis, we treated a panel of six NSCLC cell lines (three AC and three SqCC) with the HDAC inhibitor Trichostatin A, the most significant negatively correlated HDAC inhibitor from the SqCC analysis. Importantly, we selected the available cell lines that best represented their respective clinical tumor subtypes by performing principle component analysis with the subtype-specific genes using publically available gene expression profiles for a large panel of NSCLC cell lines (Figure 2.8A). As predicted by the in silico analysis, SqCC cell lines were on average, five times more sensitive to Trichostatin A than AC cell lines (SqCC avg. IC50 =69nM, AC avg. IC50 = 346 nM, Mann-Whitney U Test, p < 0.05) (Figure 2.8B), validating the clinical relevance of subtype specific alterations.      46  Figure 2.8       Figure 2.8 SqCC cell lines are more sensitive to Trichostatin A than AC cell lines.  (A) Selection of cell lines for in vitro assays by PCA analysis of two lung cancer cell line datasets.  Red indicates AC cell lines, blue SqCC cell lines and green NSCLC cell lines. (B) Dose-response analysis of  the HDAC inhibitor Trichostatin A on the relative viability of AC (A549, H3255 and H1395) and SqCC (H520, HCC15 and HCC95) cell lines. Each curve was generated from the average data points from four separate experiments. Vertical error bars represent SEM. Student's t test p =0.0002; all AC replicate IC50 values vs. all SqCC replicate IC50 values. (C) Table with the average IC50 and SEM for each cell line tested derived from four separate experiments. Replicate experiments were highly repeatable. A B p= 0.0002 47  2.4 Discussion  The emergence of tumor cells from normal precursors is thought to involve a complex interplay between genetics and cell lineage [96]. Due to the different cell types involved as well as the attributes of an individual cell’s local environment, it is logical to assume different mechanisms are required for tumorigenesis of each lung cancer subtype. Previous studies suggest that distinct patterns of DNA alteration exist for AC and SqCC; however, the specific genes responsible for the different tumor phenotypes are largely unknown [33-35, 38]. At the time of publication, this study provide the first comprehensive investigation of the key genetic and epigenetic alterations distinguishing AC and SqCC lung tumors at the gene level. Through the integration of whole-genome DNA copy number, DNA methylation, and gene expression data, we identified 778 genes altered in a subtype-specific manner. These genes are associated with distinct gene networks, providing insight into the signaling pathways that contribute to subtype tumorigenesis. Furthermore, subtype-type specific changes were found to be correlated with clinical outcomes and revealed novel putative treatment strategies for SqCC.   While no difference in the percentage of AC or SqCC genomes altered by copy number was observed, SqCC tumors were found to be more hypomethylated, suggesting that the epigenetic machinery is highly deregulated in SqCC (Figure 2.1). There is precedent for this finding, as altered global methylation is thought to be a consequence of exposure to the carcinogens found in tobacco smoke [89, 90, 97]. Global hypomethylation, such as that caused by cigarette smoke, is also known to be associated with chromosomal instability. We identified a greater number of  subtype specific alterations linked to both DNA copy number and DNA methylation in SqCC than AC. The reason for this is unclear, but may be indicative of similar selective pressures in the SqCC tumors that facilitate the development of recurrent alterations, whereas increased cellular and/or genetic heterogeneity in AC due to the different histological subtypes results in a greater diversity of alterations. Heterogeneity of clinical characteristics may also contribute to this discrepancy, as lung cancer in non-smokers are more likely to appear as AC tumors, and cigarette smoke can contribute to specific genetic or epigenetic alterations [4, 79]. Although 22.5% of our AC tumors were from never smokers, no significant differences in copy number were identified between AC tumors from ever and never smokers (data not shown).  48    The identification of subtype-specific copy number and methylation alterations with concordant expression changes demonstrate that different genetic pathways are involved in the pathogenesis of AC and SqCC. Previously identified lineage specific oncogenes including SOX2 and BRF2 were identified, validating our approach [70, 98]. Although some of the regions and genes altered by copy number have previously been shown to be important in NSCLC development, our findings suggest their newfound importance to a specific lung cancer subtype. For example, previously identified oncogenes NOTCH3 and FOXM1 were gained and overexpressed specifically in SqCC while the tumor suppressor KEAP1 was deleted and underexpressed specifically in AC [99-101]. This is the first report suggesting these previously established lung cancer-associated genes are actually involved in subtype-specific tumorigenesis.     A gene network-based analysis of our subtype specific alterations revealed additional insights into the differential oncogenic mechanisms driving the pathogenesis of AC and SqCC. The top SqCC gene network associated with subtype specific copy number alterations was associated with DNA replication, recombination and repair, while SqCC specific genes altered by methylation were associated with the small cell signaling pathway. Of particular interest was the finding that the transcription factor E2F1,one of the deregulated components of this pathway, exhibited SqCC-specific hypomethylation and overexpression. E2F1 is upregulated in SCLC tumors [93], and suppresses apoptosis and induces expression of EZH2, an oncogenic polycomb histone-methyltransferase [92]. Our observation that EZH2 expression is significantly higher in SqCC than AC, along with the preferential disruption of polycomb group proteins in SqCC further supports the relevance of this pathway to SqCC, especially given the identification of SqCC specific deregulation of numerous histone modifying enzymes by copy number alterations. Histones are fundamental building blocks of eukaryotic chromatin and are involved in a myriad of cellular processes, including replication, repair, recombination and chromosome segregation [102-104]. Recently, global alterations of histone modification patterns have been reported in human cancers, with alterations occurring more frequently in SqCC than AC, consistent with our findings [105, 106]. Our data suggest that direct deregulation of histone modification enzymes including ASF1B, PRMT1, SAE1, SET8, CHAF1A and UHRF1 may drive this phenomenon and play a key role during the development of lung SqCC. As histone modifications also play an 49  essential role in DNA replication, there may be a synergistic effect between the histone modifying genes and replication/recombination associated genes that contribute to tumor development.   The top gene network detected as perturbed in AC tumors contained genes mainly involved in regulating tissue development and cell-to-cell signaling and known to be targeted by the transcription factor HNF4a. HNF4a regulates a large set of genes in a cell-specific manner and is necessary for cell differentiation and maintenance of a differentiated epithelial phenotype [107]. In other carcinomas, deregulation of HNF4a leads to increased cellular proliferation, progression and dedifferentiation [108-110]. This suggests that HNF4a may act as a tumor suppressor in epithelial carcinogenesis [107]. Interestingly, although HNF4a itself was not affected, we found that numerous downstream targets of this gene are downregulated specifically in AC. This may have the same net affect as inactivation of HNF4a itself and lead to increased cellular proliferation during AC tumorigenesis.   Interestingly, we identified numerous genomic regions that showed opposite patterns of alteration in each lung cancer subtype. For example, a discrete alteration spanning 2.4 Mbp on chromosome bands 8p12-11.23 was commonly gained in SqCC and lost in AC, while PARP11 was upregulated in SqCC by DNA hypomethylation and downregulated in AC by copy number loss, implying that these regions/genes may play opposite roles during the development of the individual NSCLC subtypes, acting as TSGs in AC and as oncogenes in SqCC. Importantly, these differentially altered genes may be indicative of disparate clinical outcomes depending on which subtype they are disrupted in. CD9 was one of six genes that displayed opposite patterns of alteration (gained/overexpressed in SqCC and copy loss/underexpression AC) along with differential survival; high expression of this gene was correlated with favorable survival in SqCC, whereas low expression was associated with good survival in AC (Figure 2.7). Together, these results indicate that the genes involved in defining clinical characteristics are largely exclusive to individual NSCLC subtypes and influenced by the acquisition of distinct genetic alterations during tumor development, underscoring the importance of separating AC and SqCC when assessing genes involved in predicting patient prognosis and other clinical outcomes. This 50  information will become particularly important as targeted therapeutic strategies based around these genes develop. For example, since activated MEK1 and MEK2 phosphorylate and activate ERK (MAPK1) [111], the differential deregulation of MAPK1 in AC (inactivated) and SqCC (activated) tumors may be an important consideration in determining the efficacy of MEK inhibitors in lung cancer subtypes.    The specific alterations selected during the development of each subtype may influence treatment outcomes and therefore play a role in clinical management. To demonstrate how our genomic findings can be used to define treatment strategies tailored to lung cancer subtypes and attempt to elucidate novel subtype specific treatment strategies, we performed CMAP analysis on our AC and SqCC specific gene signatures to identify compounds that can potentially reverse the expression of these genes. While the results for AC were uninformative, CMAP analysis of SqCC genes identified numerous HDAC and PI3K/mTOR inhibitors as compounds that could potentially induce a gene expression signature negatively correlated with that associated with SqCC. The HDAC inhibitor result was remarkable as the alteration of histone modifying enzymes was the most prominent network disrupted in this subtype, providing a biological basis for this finding. Furthermore, cancer cells with elevated activity of E2F1 have been shown to be highly susceptible to HDAC inhibitor induced cell death and HDAC inhibitors such as SAHA have been shown to suppress the activity of EZH2 [112, 113]. As E2F1 and EZH2 are both upregulated in SqCC (Figure 2.4b & c), this data suggests that treatment with HDAC inhibitors, in conjunction with standard chemotherapy, could be a promising avenue for disease treatment. In addition, since PIK3CA activation (mutation and/or amplification) is known to occur more frequently in SqCC the finding of multiple PI3K/mTOR inhibitors as potential therapeutics for SqCC is logical [95, 114]. Together, this data demonstrates the potential to use information about the underlying molecular biology of cancer subtypes to make informed decisions about clinical management strategies, and suggests that HDAC and PI3K/mTOR inhibitors, in combination with current treatment regimes, may provide a novel treatment tailored to lung SqCC.  51  2.5 Conclusions  Fundamental differences in tumor biology may be a primary factor determining the differential outcomes and response to therapies of lung cancer subtypes. A better understanding of the molecular mechanisms underlying subtype development is therefore essential to improving the poor prognosis of lung cancer. Our high-resolution integrative analysis of NSCLC genomes and epigenomes delineated novel tumor subtype-specific genetic and epigenetic alterations responsible for driving the differential pathogenesis and phenotypes of AC and SqCC. The specific genes and networks identified in this study provide essential starting points for elucidating mechanisms of tumor differentiation and developing tailored therapeutics for lung cancer treatment. More generally, our results confirm at the molecular level that these lung cancer subtypes are distinct disease entities and highlight how biological differences between AC and SqCC can influence patient outcome and response to therapy. When designing new treatment strategies and testing new drugs in clinical trials, these subtype differences as well as the biological pathways should be taken into account.   52  Chapter 3: Genetic disruption of KEAP1/CUL3 E3 ubiquitin ligase complex components is a key mechanisms of NF-κB pathway activation in lung cancer  3.1 Introduction  Our multi-dimensional integrative analysis of AC and SqCC genomes identified for the first time, subtype specific patterns of alteration of several previously established lung cancer associated genes, highlighting their newfound importance in subtype tumorigenesis. Among these genes was Kelch-like ECH-associated protein 1 (KEAP1), a substrate adaptor protein that binds substrates to an E3-ubiquitin ligase complex comprised of Cullin 3 (CUL3) and Ring box 1 (RBX1) and was found to be preferentially lost in AC(Figure 3.1). The ubiquitin-proteasome pathway plays an essential role in maintaining normal cellular functions by controlling the abundance of several proteins and preventing undesired downstream effects. The most well characterized substrate of the KEAP1/CUL3 ubiquitin E3 ligase complex is NRF2 [115-117]. In response to oxidative stress, NRF2 stimulates transcription of cytoprotective genes that scavenge harmful reactive molecules, preventing cellular damage [118]. Interestingly, lung specific Keap1 knockout in mice was shown to protect against cell damage caused by cigarette smoke by enabling Nrf2 accumulation and increased expression of its target genes [119]. NRF2 has also been implicated in cancer cell resistance to chemotherapeutics by its activation of drug-metabolizing and drug-efflux proteins [118].   NF-κB is a transcription factor that acts as a critical regulator of genes implicated in cell proliferation and survival, angiogenesis, epithelial to mesenchymal transition as well as inflammatory and immune responses [120-122]. The NF-κB pathway is activated in over 60% of lung cancers, however, the genetic mechanisms underlying its activation remain largely unknown [120, 123-126]. In the cytoplasm, NF-κB is bound by inhibitory proteins (I-kappaB), but upon stimulation, the kinase IKBKB phosphorylates IκB, releasing its inhibition, and enabling NF-κB translocation to the nucleus where it exerts its effects (Figure 3.1) [15]. Recently, Lee et al. implicated the KEAP1 E3-ligase complex in the regulation of NF-κB signaling, by demonstrating that KEAP1 binds IKBKB drawing it to the E3-ligase complex for ubiquitination and degradation [101]. As IKBKB is known to promote tumorigenicity through 53  phosphorylation-mediated inhibition of tumor suppressors and upregulation of NF-κB signaling, these findings further implicate the dysregulation of KEAP1 and the E3-ligase complex in NSCLC tumorigenesis and suggest disruption of this complex may underlie the high frequency of NF-κB activation in lung cancer.     Figure 3.1   Figure 3.1 The role of the KEAP1/CUL3 ubiquitin ligase complex  (A) When complex components are intact, KEAP1 facilitates binding of NRF2 or IKBKB which promotes their ubiquitination. This complex prevents accumulation of IKBKB and subsequent NF-κB activation. (B) Disruption of any complex component compromises function leading to stabilization and accumulation of IKBKB, and aberrant activation of NF-κB. A B 54   Inactivating somatic mutations, loss of heterozygosity and hypermethylation of KEAP1 have been reported at varying frequencies (3-41%) in lung tumors and cell lines, and low KEAP1 expression is associated with poor patient outcome [84, 127-132]. However, the moderate frequency of KEAP1 gene disruption alone is not sufficient to explain the high (>60%) frequency of NF-κB activation observed in lung cancer [124]. The ability of the E3-ligase complex to ubiquitinate IKBKB was found to be most efficient when all three complex components were expressed and intact, suggesting disruption of even a single component compromises function. Although somatic DNA alterations have been observed in the genes encoding some of these complex components, the frequency of genetic and/or epigenetic disruption of complex components and whether complex component gene disruptions are a key mechanism of NF-κB activation in lung cancer is unknown [101]. Given the importance of the NF-κB pathway in lung cancer and the lack of inquiry into the role of the other complex components in lung cancer, we sought to determine the subtype specific patterns and biological effect of component disruption.   We hypothesize that somatic disruptions of CUL3 (2q36.2) and RBX1 (22q13.2), in addition to KEAP1 (19p13.2) occur frequently in lung tumors, representing a prominent genetic mechanism that may be responsible for IKBKB accumulation and stimulation of NF-κB, and that genetic disruption of any one E3-ubiquitin ligase complex component is sufficient to result in tumorigenic NF-κB activation due to loss of complex function and subsequent accumulation of IKBKB [15, 101, 133]. In this study, we 1) investigated whether these complex components and IKBKB (8p11.21) exhibit gene dosage and expression alterations and the frequencies at which they occur in multiple independent tumor cohorts, 2) investigated whether the complex components display subtype specific disruption, and 3) assessed the functional consequence of complex disruption on NF-κB activity, as these genetic events may be significant contributors to the NF-κB activation commonly observed in lung cancer.  55  3.2 Methods   3.2.1 Non-small cell lung cancer (NSCLC) samples   261 lung tumors (169 adenocarcinomas (AC) and 92 squamous cell carcinomas (SqCC)) were accrued from Vancouver General Hospital (Vancouver) and Princess Margaret Hospital (Toronto) following ethics approval with patient consent (Chapter 2 and Sample Set #1, Appendix A1). Tissue sections were microdissected with the guidance of lung pathologists. Matched non-malignant lung tissue was also obtained for a subset of the primary tumors collected. DNA for all 261 samples was extracted using standard phenol-chloroform procedures. RNA was extracted from tumor and matched non-malignant tissues using RNeasy Mini Kits (Qiagen Inc.). NSCLC cell lines (H1650, HCC827, H3255, H358, H23, HCC95, H2347, and H2122) were obtained from ATCC or the laboratory of AFG and cultivated as previously described [134]. These cell lines were fingerprinted to confirm their identity [61]. Human bronchial epithelial cells (HBEC-KT) were provided by Dr. John Minna (University of Texas Southwestern Medical Center at Dallas) and maintained as previously described [98]. Primary non-malignant human bronchial epithelial (NHBE) lung cells were obtained from Lonza.  3.2.2 Determination of gene dosage and expression levels Copy number status (gain, loss, or neutral) for the KEAP1, CUL3, RBX1, and IKBKB loci was determined for each tumor sample and 63 NSCLC cell lines by array comparative genomic hybridization (array CGH) using the whole genome tiling path array (SMRT v.2, BCCRC Array Laboratory, Vancouver, BC) as previously described in Chapter 2 and [76, 135, 136]. Gene expression levels of KEAP1, CUL3, RBX1, and IKBKB were determined using custom Agilent gene expression microarrays in 35 AC and 13 SqCC lung tumors and corresponding matched non-malignant tissues. Genes were classified as over or underexpressed if the fold change in mRNA expression levels in tumors relative to matched non-malignant tissues was greater or less than 2 fold. Gene expression for KEAP1, CUL3, RBX1, and IKBKB was also assessed in an additional, distinct cohort of 49 NSCLC (29 AC and 20 SqCC) tumors with matched CGH profiles using custom Affymetrix arrays to determine the contribution of copy number on gene expression (Sample Set #3, Appendix A1). The association between copy number and gene 56  expression was assessed by segregating these 49 NSCLC tumors into those with and without copy number alterations for each gene as previously described [76]. Expression levels were compared in both groups using a U test with a p-value less than 0.05 considered significant. The probe with the highest median intensity across all tumor samples was assessed for each gene.  3.2.3 Copy number analysis of external cohorts Publically available NSCLC data was downloaded to further explore the frequency and subtype specificity of genomic disruption at the KEAP1, CUL3, RBX1, and IKBKB loci. Affymetrix SNP 250K data for 383 matched tumor non-malignant NSCLC pairs were accessed from the dbGaP Genotypes and Phenotypes database (Study Accession: phs000144.v1.p1). Affymetrix SNP 6.0 array profiles were obtained for 232 NSCLC tumors and matched non-malignant tissue (155 SqCC, 77 AC, GSE25016) and 54 NSCLC cell lines, 27 of which overlapped with cell lines profiled on the SMRT array, from the Wellcome Trust Sanger Institute CGP Data Archive (sanger.ac.uk/genetics/CGP/Archive/). Copy number profiles were generated using Partek Genomics Suite software. For tumor profiles, the patient matched non-malignant sample was used as a baseline for defining copy number alterations, whereas SNP profiles derived from 72 cytogenetically normal HapMap individuals were used as a reference for the cell line data [137]. In total, 570 tumors and 90 NSCLC cell lines (combination of lines profiled on the SNP and SMRT array) were analyzed for copy number alterations. The Broad Institute's Tumorscape database (www.broadinstitute.org/tumorscape) was also accessed to investigate copy number status at these four gene loci [138].  3.2.4 Multi-dimensional analysis of complex component disruption in the TCGA  Level 3 Affymetrix SNP6 copy number, HM27K DNA methylation and whole exome sequencing data was downloaded for all available AC and SqCC tumors. Analysis of DNA methylation data was limited to those cases with profiles of the matched non-malignant tissue (54 SqCC and 48 AC). Genes were identified as aberrantly methylated if the delta beta value (normal Beta value subtracted from the tumor Beta value) was greater than ± 0.15.   57  3.2.5 Western blot assessment of total and phospho- IKBKB and NF-κB protein levels   IKBKB protein levels were assessed in 8 NSCLC cell lines with various combinations of KEAP1, CUL3, or RBX1 genomic loss or IKBKB genomic gain using antibodies from Cell Signaling (IKBKB #2678, p-IKBKB #2697, NF-κB: #4764, p-NF-κB: #3033, and GAPDH #2118). Western blots were performed following standard procedures as previously described [98]. Cells were washed with cold PBS and lysed in RIPA buffer with complete protease inhibitor cocktail (Roche, Basel Switzerland). Protein lysates were quantified using the BCA assay (Fisher Scientific). Lysates were diluted and boiled for electrophoresis then transferred to a polyvinylidene membrane. Membranes were blocked in 5% skim milk or 5% BSA in Tris buffered saline containing Tween 20 (TBS-T) (according to the manufacturer's instructions) and then incubated with primary antibody (1:1000) at 4°C overnight. Following three washes in TBS-T, membranes were incubated with HRP conjugated secondary antibody (Cell Signaling, cat. #7074, 1:20000) for 1 hour at room temperature. Antibody binding was visualized by enhanced chemiluminescence (Thermo Scientific) after three washes in TBS-T.  3.2.6 Immunohistochemistry (IHC) staining for IKBKB protein levels  5 µm thick sections were cut from 13 formalin fixed, paraffin embedded tumor specimens with various states of genomic disruption to the E3-ubiquitin ligase complex genes or IKBKB,based on copy number profiles. IHC to determine protein expression of IKBKB was performed as previously described [98]. Briefly, slides were deparaffinized in xylene and rehydrated with graded ethanol washes. Antigen retrieval was performed using a decloaking chamber with sodium citrate buffer pH 6.0, after which endogenous peroxidase activity was blocked using 3% H2O2 for 30 minutes at room temperature. Sections were blocked with goat serum for 3 hours at room temperature and then incubated overnight at 4°C with 32 µg/ml of anti-IKBKB mouse monoclonal primary antibody (EMD4 Biosciences, cat. OP134, San Diego, CA, USA). Prior to incubation with an anti-mouse HRP-streptavidin conjugated secondary antibody (DAKO, cat. K4000), four five minute washes in TBS-T were performed to remove unbound primary antibody. Detection of antibody binding was assessed using diaminobenzidine (Sigma Aldrich, cat. D4293). Slides were counterstained with hematoxylin for visualization. Intensity of staining was scored using a 0-3+ system based on the consensus of 3 observers (KT, LP, JCE). The mean 58  staining intensity for each tumor section was judged as follows:  0 - no staining, 1 - weak intensity, 2 - moderate intensity, 3 - strong staining.  3.2.7 siRNA-mediated complex component knockdowns On-Target plus SMART pool siRNAs targeting KEAP1, RBX1, CUL3, and a non-targeting control pool of siRNAs were purchased from Thermo Scientific. One day prior to transfection, HBEC-KT cells were plated in regular growth media (antibiotic free) at a density of 200,000 cells/well in six well plates. Transfections were performed using an siRNA concentration of 100 nm according to the Thermo Scientific DharmaFECT siRNA transfection protocol. DharamaFECT 1 transfection reagent (Dharmacon) was used at a concentration of 0.2 µl/100 µl media. After 12 hours, transfection media was replaced with regular growth media. RNA was harvested from cells at 48 hours post transfection using the Trizol method (Invitrogen) and protein lysates were prepared using RIPA buffer 72 hours post transfection. Knockdown efficiencies were measured by qPCR with the following TaqMan assays from Applied Biosystems and using 18S rRNA as an endogenous control:  Hs99999901_s1 (18S), Hs00202227_m1 (KEAP1), Hs00180183_m1 (CUL3), and Hs00360274_m1 (RBX1). Western blots were performed as above to measure total and phospho-IKBKB and NF-κB protein levels.   3.2.8 NF-κB target gene analysis   Expression of nine genes transcriptionally controlled by NF-κB (CCND1, CXCR4, MMP2, TRAF1, MMP9, BCL2L11, CXCL13, PTGS2 and CXCL12) - as annotated in the Ingenuity Pathway Analysis database (Ingenuity® Systems) were analyzed in 48 NSCLC tumors with deregulated KEAP1, CUL3, RBX1 or IKBKB expression levels (2-fold or greater). Wilcoxon signed-rank tests were used to compare NF-κB target gene expression in tumors versus their matched non-malignant lung tissues. NF-κB target genes were considered upregulated if expression levels were significantly elevated in tumors relative to matched non-malignant tissues (Wilcoxon p <0.05).  59  3.2.9  IKBKB inhibition in NSCLC cell lines   Cell viability assays were performed to measure the effect of IKBKB inhibition by a cell permeable, competitive ATP inhibitor, IKK-2 inhibitor IV (Calbiochem cat. #401481) on five NSCLC cell lines (H3255, H2122, H23, HCC95 and H1650). Cells were plated in triplicate in 96 well plates at optimal densities for growth (H2122 and H23 at 2000 cells/well, H1650 and HCC95 at 3000 cells/well, and H3255 at 5000 cells/well) and subjected to a series of 2-fold dilutions of IKBKB inhibitor prepared in cell growth media and DMSO. The experimental inhibitor concentrations ranged from 500 µM to 244 nM and the final DMSO concentration for treated and untreated (control) cells was 1%. Blank wells contained equal volumes of growth media with 1% DMSO. Cells were incubated for 72 hours at 37°C and then treated with 10µl of Alamar Blue cell viability reagent (Invitrogen) according to manufacturer's instructions. The reaction product was quantified by measuring absorbance at 570 nm with reference to 600 nm using an EMax plate reader (Molecular Devices). The average absorbance readings for blank wells were subtracted from all treatment and control wells and technical replicates were averaged. The response of treated cells was measured as a proportion of the viability of untreated cells, with the mean background subtracted treatment absorbance divided by the mean background subtracted untreated absorbance for each inhibitor concentration. Dose response curves and IC50 values were generated in Graph Pad v5 using the proportionate response of all 12 drug concentrations. Experiments were repeated in triplicate and differences in IC50 values were determined using a student's t-test with a p-value < 0.05 considered significant.    60  3.3 Results  3.3.1 Disruption of E3-ubiquitin ligase complex components in NSCLC   261 tumors were screened for DNA copy number alterations at the KEAP1, CUL3, and RBX1 loci. A significant proportion of tumor samples (103 of 261, 39%) showed genomic loss in at least one of the complex associated genes (Figure 3.2A). KEAP1 was the most frequently disrupted complex component, undergoing genomic loss in 23% of lung tumors analyzed (Figure 3.2A). Strikingly, 69% (71 of 103) of the tumors harboring copy number alterations had alterations affecting only one of the genes assessed. Gene expression analysis revealed aberrant expression in at least one of the complex component genes in 40% of tumors analyzed (19 of 48) and of those with aberrant expression, 84% (16 of 19) had only one of the genes affected (Figure 3.2A). At the expression level, RBX1 was the most frequently altered complex component,  underexpressed in 21% of tumors analyzed (Figure 3.2A). We also detected a high frequency of IKBKB DNA copy number gain and mRNA overexpression (23% and 35% of tumors, respectively) (Figure 3.2A). When IKBKB status was taken into account, the frequency of NSCLC disruption at any of the KEAP1, CUL3, RBX1 and/or IKBKB loci rose to 54% (141 of 261 tumors, Figure 3.2A) at the gene copy number level and 63% (30 of 48 tumors, Figure 3.2A) at the expression level. Even with the inclusion of IKBKB, the majority of tumors exhibited complex or IKBKB genetic disruption at only one gene locus (Figure 3.2B and 3.2C). Frequent genetic disruption of E3-ubiquitin ligase complex components or IKBKB (73%) was also evident in NSCLC cell lines (Table 3.1).   To determine whether complex component and IKBKB gene dosage alterations are regulating gene expression, we integrated DNA copy number and gene expression data for the complex genes, and IKBKB in an additional set of 49 NSCLC tumors. KEAP1, RBX1 and CUL3, expression was significantly lower in tumors with genomic loss compared to those without loss (U test, p=0.00076, p=0.00116, and p=0.00339, respectively, Figure 3.3A-C), whereas IKBKB gene expression was elevated in tumors with gain compared to those without (U test, p=0.0143, Figure 3.3D). These findings demonstrate that dosage alterations affect mRNA expression levels, and therefore, likely contribute to E3-ligase complex disruption. 61  Figure 3.2  Figure 3.2 Frequent disruption of the KEAP1 E3-ligase complex and IKBKB in NSCLC.  (A) Summary of DNA copy number and gene expression alterations in NSCLC tumors. (B) Copy number analysis of 261 lung tumors revealed frequent loss of KEAP1, RBX1, and CUL3, as well as frequent gain of IKBKB (141/261, 54%). Vertical columns indicate individual tumor samples and only samples with ≥1 alterations are shown. (C) mRNA expression profiles for 48 lung tumors revealed frequent underexpression of complex components and overexpression of IKBKB (30/48, 62.5%). Expression was considered altered if tumor/matched non-malignant tissue was changed >2 fold. (D) KEAP1, CUL3, and IKBKB exhibit significant differences in copy number alteration patterns between AC and SqCC (Fisher's exact test, p < 0.05). 62   Table 3.1 Copy Number status of complex components in multiple independent cohorts Gene Cell Lines (n=90) dbGAP  (n=383) GSE25016 (n=232) KEAP1  39% loss 23% loss 18.5% loss CUL3 17% loss 3% loss 18.5% loss RBX1 39% loss 11% loss 10% loss IKBKB 28% gain 11% gain 23% gain Any 73% 34% 52%    Figure 3.3  Figure 3.3. Association between gene expression and copy number in clinical lung tumors.  Box and whisker plots demonstrating KEAP1 (A), RBX1 (B), CUL3 (C) and IKBKB (D) expression levels in tumors with and without genetic disruption. Tumors were grouped based on copy number status for each gene (gain, loss, neutral) and expression levels were compared between groups using a U test with a p-value < 0.05 considered significant.  Whiskers show the min and max, while boxes illustrate the 25th, median and 75th percentile. 63   In addition to the tumor data generated using array comparative genomic hybridization, we analyzed copy number profiles derived from publically available SNP array data from the dbGaP Genotypes and Phenotypes database (n=383) and GSE25016 (n=232) both of which had matched non-malignant tissue for all cases, and a panel of 90 NSCLC cell lines. Consistent with our findings, frequent genomic disruption to the E3-ubiquitin ligase complex components and IKBKB was observed in all external cohorts (Table 3.1). 73% of cell lines harboured DNA alterations encompassing at least one complex component or IKBKB, whereas 34% and 52% of tumors had disruption. We further interrogated the copy number status of KEAP1, CUL3, RBX1, and IKBKB genes in the Broad Institute's Tumorscape database [138]. This revealed that CUL3 was significantly deleted in 12% of all 3131 tumors in the database and in 13% percent of all NSCLC specimens (n=733), of which 5.4% had focal CUL3 deletions. Similarly, IKBKB was significantly amplified in 22% of the all tumors and 28% of NSCLC specimens, of which 12% contained focal IKBKB DNA amplifications. KEAP1 and RBX1 were not significantly deleted in the Tumorscape database. Together, these results support our observation that KEAP1 complex components and IKBKB undergo frequent copy number alterations in NSCLC.    3.3.2 Complex components are differentially altered in AC and SqCC subtypes  As distinct patterns of DNA alterations exist for AC and SqCC, we sought to determine whether complex component disruption displays subtype specific patterns of alteration. While both subtypes of NSCLC showed high frequency of complex component and IKBKB gene disruption, the pattern of gene disruption differed between these subtypes (Figure 3.2D and Table 3.2). KEAP1 loss appears to be the main mechanism of complex disruption in lung AC, accounting for 64% of cases with complex disruption, and is more prevalent in AC than in the SqCC subtype (Fisher's exact test, p = 0.0075, Table 3.2). In contrast, CUL3 loss and IKBKB gain occurred more often in the SqCC than AC subtype of lung tumors (Fisher's exact test, p = 4.926 x 10-6 and p = 8.446 x 10-7, respectively, Table 3.2). Gain of IKBKB was the most frequent alteration in SqCC, occurring in 57% of cases, followed closely by CUL3 loss in 44% of cases with complex component disruption.   64   To confirm these subtype specific findings, we interrogated complex component and IKBKB copy number in The Cancer Genome Atlas (TCGA) and GSE25016 cohorts, both of which have CN data for AC and SqCC. The same trends in component disruption were observed across all three datasets; with KEAP1 and RBX1 loss being more prevalent in AC and CUL3 loss and IKBKB gain more frequent in SqCC (Table 3.2). Statistically significant differences between the frequency of component alteration was observed for all components in the GSE25016 dataset and for KEAP1 and CUL3 in the TCGA (Table 3.2). Analysis of external cohorts corroborated the subtype specific patterns of alteration we observed and demonstrate for the first time that the mechanisms of KEAP1/CUL3 complex disruption are subtype specific.   Table 3.2 Frequency of KEAP1 complex copy number alterations in AC and SqCC  Frequency in AC Frequency in SqCC Fishers Exact Test  Gene BCCRC TCGA GSE BCCRC TCGA GSE BCCRC TCGA GSE KEAP1 50/169 (29.6%) 48/277 (17.3%) 27/77 (35.1%) 10/92 (10.9%) 20/201 (10%) 16/155 (10.3%) 0.0075 0.02427 1.1x10-5 CUL3 13/169 (7.7%) 4/277 (1.4%) 6/77 (7.8%) 28/92 (30.4%) 27/201 (13.4%) 37/155 (24%) 4.93x10-6 1.36x10-7 0.00233 RBX1 31/169 (18.3%) 29/277 (10.5%) 14/77 (18.2%) 8/92 (8.7%) 11/201 (5.5%) 9/155 (5.8%) 0.24 0.06495 0.0047 IKBKB 20/169 (11.8%) 44/277 (16.2%) 11/77 (14.3%) 39/92 (42.4%) 46/201 (22.9%) 42/155 (27.1%) 8.45x10-7 0.05248  0.0312 ANY 78/169 (45.6%) 99/277 (35.7%) 41/77 (53.2%) 64/92 (69.6%) 78/201 (38.8%) 79/155 (51%)      3.3.3 Other genetic and epigenetic mechanisms of complex component disruption  Given our hypothesis, we focused on measuring gene dosage alterations that could account for disruption of the E3-ubiquitin ligase complex and its downstream consequences. However, as mutations and hypermethylation of KEAP1 have been described in lung cancer and these events are known to downregulate KEAP1 expression [84, 128, 129, 132], we next sought to determine whether IKBKB and complex components were frequently altered by other genetic or epigenetic mechanisms. To assess the contribution of DNA methylation and gene mutation to complex disruption, we analyzed 408 tumors (230 AC and 178 SqCC) for which mutation data 65  was available and 102 tumors with matched non-malignant tissue (54 SqCC and 48 AC) and HM27K methylation profiling from the TCGA. KEAP1 was the most frequently mutated component, mutated in 17% (40/230) of AC and 12% (22/178) of SqCC. All other complex components were infrequently mutated; 3.4% for CUL3, 0.5% for RBX1 and 2.2% for IKBKB. Interestingly, mutations in CUL3 were significantly more frequent in SqCC ( 6% vs. 1.5%, Fisher's Test p=0.011), further supporting the notion that CUL3 is preferentially disrupted in SqCC. None of the complex components or IKBKB were hypo- or hypermethylated in any of the samples assessed. The rarity of aberrant methylation across all complex components and the lack of mutations in CUL3, RBX1 and IKBKB further supports the notion that copy number is the primary genetic mechanism through which these genes are deregulated. The high frequency of KEAP1 mutations, and mutual exclusivity with copy loss underscores the selective inactivation of KEAP1and highlights its importance in lung tumor biology.  3.3.4 Functional consequences of genetic complex disruption    Since IKBKB protein levels would be directly affected by E3-complex function, we assessed IKBKB protein levels in NSCLC cell lines with and without complex disruption to measure the consequence of complex disruption. Immunoblotting for IKBKB in non-malignant human bronchial epithelial (NHBE) cells and a panel of 8 NSCLC cell lines revealed elevated expression levels in lines with genomic loss of KEAP1, CUL3, RBX1, or gain of IKBKB (Figure 3.4A). No correlation between the number of disrupted components and the levels of IKBKB were observed. To directly assess the effect of complex component integrity on IKBKB and NF-κB activity, we performed siRNA mediated knockdowns of KEAP1, CUL3 and RBX1 in non-malignant, bronchial epithelial cells. We achieved at least 80% knockdown for all three genes and observed an increase in phospho-IKBKB and phospho- NF-κB levels in the knockdowns relative to the non-targeting control, providing evidence that complex disruption directly regulates NF-κB signaling  (Figure 3.4B).      66  Figure 3.4.      Figure 3.4. IKBKB protein expression in NSCLC.   (A) Western blot depicting IKBKB protein expression in eight cell lines with varying degrees of genetic disruption to KEAP1 E3-ligase complex components and/or IKBKB (as determined from copy number profiles of 90 NSCLC cell lines). NHBE is a non-malignant human bronchial epithelial lung line that provides a baseline for IKBKB expression. The number of disrupted genes for each line is indicated above the cell line. (B) Western blot depicting the effects of transient siRNA knockdown of KEAP1, RBX1, and CUL3, on total and phospho- IKBKB and NF-κB protein levels. (C) qRT-PCR analysis of three NFkB transcriptional targets upon siRNA knockdown of individual complex components.   A B C 67   In an attempt to determine whether genetic disruption of complex components affects the relative protein levels of IKBKB, we performed immunohistochemistry (IHC) on a panel of 13 lung tumors with various states of complex component disruption. Staining for IKBKB revealed protein expression in both complex disrupted and undisrupted tumors; however, the vast inter- and intratumor heterogeneity in staining and potential non-specific antibody binding limited our ability to identify significant correlations between IKBKB protein levels and E3-complex genetic disruption (data not shown, available as Supplemental Digital Content online at http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3164321/).    As we were unable to discriminate differences in IKBKB protein levels between complex disrupted and non disrupted tumors by IHC, we sought to determine whether loss and underexpression of  E3-ligase complex components activates NF-κB signaling. Expression of nine well characterized transcriptional targets of NF-κB were analyzed in the 48 NSCLC tumor pairs for which complex component disruption could be determined. Of the nine genes, four - BCL2L11, CXCL13, MMP9, and TRAF1 were significantly up-regulated in tumors with underexpression of a complex component or overexpression of IKBKB as compared to their matched non-malignant lung tissue (Figure 3.5). Assessment of the same NF-κB target genes following siRNA knockdown confirmed these genes to be overexpressed relative to non-targeting controls (Figure 3.4C), supporting our hypothesis that complex disruption leads to activation of NF-κB signaling.   68  Figure 3.5  Figure 3.5. Activation of NF-κB targets in tumors with complex components disruption  (A-D) Box plots demonstrating four examples of NF-κB target genes (BCL2L11, CXCL13, MMP9, TRAF1) that are significantly upregulated in tumors with complex component disruption, relative to matched non-malignant lung tissues (Wilcoxon sign rank test, p ≤ 0.001). Copy number status of the NF-κB target gene loci was determined in these tumors to ensure that overexpression was not due to dosage changes of the target genes. Whiskers show the min and max values, while boxes illustrate the 25th, median and 75th percentile.   3.3.5 Pharmacological inhibition of IKBKB    Overexpression of IKBKB can contribute to a malignant phenotype through activation of the NF-κB signaling pathway and consequential effects on cell growth. We hypothesized cells dependent on this pathway would be more sensitive to IKBKB inhibition than those without complex alterations and normal IKBKB protein levels. IKBKB inhibition experiments revealed H1650, an AC line without genomic loss of any complex component, had reduced sensitivity (as measured by cell viability) to IKBKB inhibition (IC50 13.41) than H2122 and H23 (IC50 8.19 and 6.59, respectively) which have multiple components altered (student's t-test, p < 0.05) (Figure B A C D 69  3.6). H3255 and HCC95 showed insensitivity to inhibition compared to H1650 (student's t-test, p < 0.05), however these two cell lines have the highest IKBKB protein expression levels of the cell lines we tested, which likely contributed to their relative resistance to this competitive inhibitor. Replicate experiments were highly reproducible, with similar trends in sensitivity observed across all replicates.   Figure 3.6    Figure 3.6. Pharmacological inhibition of IKBKB in NSCLC cell lines   Dose response assays were performed to measure the effect of IKBKB inhibition on viability of 5 NSCLC cell lines. H2122, H23, H3255, and HCC95 harboured genetic alterations to either IKBKB and/or one or more complex components while H1650 was not altered at the DNA level.  Error bars represent the standard error of the mean of replicate experiments. IKBKB protein expression as measured by western blot and the IC50s of each cell line are shown in the table. 70  3.4 Discussion  The NF-κB pathway is aberrantly activated in the majority of lung cancers and is essential in KRAS driven mouse models of lung tumorigenesis [123-126]. NF-κB signaling contributes to tumorigenesis via its promotion of cell proliferation and survival [120-122]. In order for NF-κB to become active, inhibition by IκB must be released. This is achieved through phosphorylation of IκB by the kinase, IKBKB [139]; hence, IKBKB has a critical role in NF-κB activation [101, 139]. In fact, constitutive IKBKB activity has been postulated to drive the aberrant NF-κB activation observed in cancer [121]. Despite what is known about the cascade of protein signaling events that result in NF-κB activation, the genetic mechanisms responsible for the aberrant activation of NF-κB signaling in lung cancer are not well understood. In this study, we hypothesized that genetic disruption and loss of function of the KEAP1 E3-ubiquitin ligase complex, which regulates IKBKB protein levels, is a major mechanism of IKBKB accumulation and consequential NF-κB activation in lung cancer.   These results provide evidence that somatic E3-ligase complex disruption is a prominent genetic mechanism of NF-κB activation in lung cancer that compromises the ability of cells to degrade the NF-κB activator, IKBKB. We discovered a remarkably high frequency of both genetic disruption and gene expression changes for the genes encoding E3-ligase protein components (KEAP1, RBX1, and CUL3) as well as the gene encoding the complex's oncogenic substrate, IKBKB. We found that genetic disruption of the complex genes alters mRNA expression and results in elevated IKBKB protein levels, demonstrating the consequence of complex disruption. Moreover, we demonstrated evidence of NF-κB activation in complex compromised lung tumors and showed the importance of IKBKB protein expression in driving the lung cancer phenotype.   Although genetic and epigenetic disruption of KEAP1 has been reported in lung cancer before, to our knowledge, this is the first study to comprehensively characterize somatic gene dosage alterations to the CUL3, RBX1, and IKBKB loci in a large cohort of clinical lung tumors. The strikingly high frequency of copy number and gene expression alterations observed in our study highlights the importance of these E3-ubiquitin ligase complex components and also 71  IKBKB in lung cancer. The recurrent nature of DNA copy number alterations at the complex component loci and their effects on gene expression are strong evidence that these genes are targeted for dosage alterations as opposed to passengers of alterations targeting other genes. In addition, the high proportion of disrupted lung tumors observed to have genetic alterations affecting a single component only, at both the copy number (67%, Figure 3.2B) and gene expression levels (73%, Figure 3.2C), suggests that disruption of only a single complex component is sufficient to compromise complex function and promote NF-κB signaling through abnormal IKBKB accumulation. Interestingly, we observed differential complex component disruption patterns in AC and SqCC subtypes (Figure 3.2D). Although E3-ubiquitin ligase complex disruption occurs in both subtypes, the differences in the component genes preferentially altered suggests that complex disruption is achieved by different means.    Examination of IKBKB protein levels in NSCLC cell lines revealed high expression in lines harboring genetic disruption to at least one complex component or IKBKB, whereas the non-malignant lung line (NHBE) and a line without genetic disruption (H1650) showed very low or undetectable levels. The E3-ligase complex was considered to be genetically intact in H1650 as neither underexpression of KEAP1, RBX1, and CUL3, or overexpression of IKBKB relative to NHBE cells was observed. This suggests there are no genetic or epigenetic alterations affecting the complex components or IKBKB in this cell line and the observed IKBKB levels were consistent with H1650 having a functioning E3-ligase complex, supporting our hypothesis. Therefore, in addition to affecting gene expression, copy number losses of the loci coding for complex components and gains of IKBKB appear to influence IKBKB protein expression. A trend towards higher IKBKB expression in lines with more complex components/IKBKB alterations was not evident, suggesting genetic disruption of a single component is sufficient to result in loss of complex function and IKBKB accumulation (Figure 3.4A). This finding is consistent with the observation that the majority of tumors exhibiting complex disruption have only one complex component altered, further supporting the idea that single component disruption is sufficient to produce an oncogenic effect.  72   We have conclusively demonstrated elevated IKBKB protein expression in NSCLC cell lines with complex disruption, however, measuring this effect directly in tumor tissue sections was not a straightforward task due to the extent of heterogeneity in tumor staining intensity across and within individual tumors which likely reflects the heterogeneous nature of lung tumor specimens. Due to this innate tumor heterogeneity, unlike cell lines, we were unable to conclude whether or not there was a significant correlation between E3-ligase complex disruption and IKBKB protein levels in vivo. Examples of tumor staining are available in Supplemental Digital Content 6, available online at http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3164321/.   To investigate the direct consequence of complex disruption on IKBKB accumulation and NF-κB activity, we performed siRNA knockdowns on the individual complex coding genes (KEAP1, RBX1, and CUL3). Consistent with our hypothesis, we observed elevated levels of activated IKBKB and NF-κB as well as elevated expression of NF-κB target genes upon complex component disruption in HBEC cells (Figure 3.4B and C). Since phospho-NF-κB is an indicator of active NF-κB signaling, these results clearly illustrate the functional consequence of E3-ligase complex disruption. Our work provides evidence to support the hypothesis that genetic loss of the complex component encoding genes causes downregulation in their expression, and that loss of expression of these genes results in increased levels of activated IKBKB and NF-κB.    A number of reports have detailed the critical role of IKBKB protein in driving NF-κB activation, and the importance of IKBKB to cancer cell viability is emphasized by the development of IKBKB inhibitors as a strategy for tempering NF-κB signaling [121, 133, 139]. We found that IKBKB inhibition reduced NSCLC cell viability and that cells without complex or IKBKB disruption, which we hypothesized to be less dependent on IKBKB expression for growth, were indeed more insensitive to IKBKB inhibition, as were cells with high endogenous levels of IKBKB protein (Figure 3.6). In addition to cell experiments to verify the importance of E3-ligase complex disruption in lung cancer, we analyzed the expression levels of several NF-κB target genes in tumors with complex disruption to measure its effect on NF-κB activity. Despite the possibility that other mechanisms could also contribute to the transcription of the NF-κB target genes, we observed a significant increase in the expression of NF-κB target genes in 73  complex compromised tumors (Figure 3.5). Together, these findings support our hypotheses and demonstrate the biological significance of complex disruption and subsequent IKBKB overexpression in lung cancer biology.  3.5 Conclusions  Our analyses have revealed remarkably frequent genetic disruption and aberrant expression of not only KEAP1, but all members of the KEAP1 E3-ubiquitin ligase complex and IKBKB in lung cancer. For the first time we show that AC and SqCC acquire copy number alterations to different components of the E3-complex or IKBKB which suggests the genetic mechanisms of complex disruption that promote NF-κB activation may be subtype specific. Moreover, we have shown that IKBKB protein expression is elevated in NSCLCs with genetic loss of KEAP1, CUL3 or RBX1 or gain of IKBKB, and that knockdown of complex components leads to an accumulation of active IKBKB and NF-κB, demonstrating the functional consequence and significance of complex disruption. We have also provided evidence of NF-κB activity, a downstream effect of IKBKB accumulation, in complex disrupted tumors and cell lines. Collectively, our findings suggest that prominent genetic disruption to the E3-ubiquitin ligase complex and its oncogenic substrate, IKBKB, play a major role in driving the aberrant NF-κB activation that is characteristic of lung tumorigenesis but that this activation occurs via different mechanisms depending on the subtype.  74  Chapter 4: YEATS4 is a novel oncogene amplified in non-small cell lung cancer that regulates the p53 pathway   4.1 Introduction  Within the last decade, characterization of lung cancer genomes has revealed a number of genes critical to tumorigenesis, resulting in significant changes to lung cancer treatment and a subsequent increase in progression free and overall survival for a subset of these patients. These successes have prompted a search for additional driver alterations, and have identified a number of recurrently mutated genes including TP53,EGFR, CDKN2A, PTEN, NRAS, BRAF, PIK3CA, DDR2, KEAP1and NRF2 as well as gene fusions encompassing RET and ROS tyrosine kinases [15, 55, 140-142]. Despite these discoveries, approximately 50% of lung cancers harbor no known targetable alterations, highlighting the need for a better understanding of the biology underlying lung tumorigenesis [55, 142].    In addition to somatic mutations, copy number alterations such as recurrent amplifications and deletions occur in almost all lung cancers [18, 76]. DNA amplification directly contributes to oncogene activation and the promotion of tumorigenesis, particularly for tumors driven by oncogene addiction. Oncogenes amplified at the DNA level therefore make ideal therapeutic targets as unlike loss of function tumor suppressor genes (TSG), they have the potential to be targeted directly. In NSCLC, recurrent amplifications of several regions activate known oncogenes. These include; 1q21.2 (ARNT), 3q26.3-q27 (PIK3CA & SOX2), 5p15.33 (TERT), 7p11.2 (EGFR), 7q31.1(MET), 8p12 (FGFR1) 8q24.21 (MYC), 12q14.1 (CDK4), 14q13.3 (NKX2-1) [18, 70, 95, 122, 143-145]. In chapter 2, numerous regions of frequent amplification were identified, some of which were highly specific to a subtype like 3q and others such as 12q and 20q that were frequently altered in both AC and SqCC. In an attempt to identify novel oncogenes involved in NSCLC tumorigenesis, we integrated DNA copy number and gene expression data in order to identify candidate driver genes within highly recurrent amplicons.   75   Our approach was based on the rationale that oncogenes selectively amplified and biologically relevant to NSCLC tumor biology would: i) span regions of frequent high level amplification, ii) undergo frequent overexpression and iii) exert pro-tumorigenic functions in vitro and in vivo. Our analysis identified a recurrent amplicon at 12q15, within which we identified the candidate oncogene YEATS4/GAS41 (YEATS domain containing 4, glioma-amplified sequence 41). In vivo and in vitro functional assays were performed to characterize the biologic effects and investigate the oncogenic mechanism of YEATS4 in lung tumorigenesis. Based on the frequency of YEATS4 amplification and overexpression in NSCLC tumors and cell lines, its role in viability, anchorage independent growth, senescence and tumor formation, we propose that YEATS4 is novel candidate oncogene in lung cancer.     4.2 Methods  4.2.1 NSCLC  tumor samples and cell lines  261 formalin-fixed paraffin embedded and fresh-frozen lung tumors (169 AC and 92 SqCC) were obtained under informed, written consent with approval from the University of British Columbia-BC Cancer Research and University of Toronto Ethics Board from patients undergoing surgical resection at the Vancouver General Hospital and the Princess Margaret Hospital in Toronto [146]. Tissue sections were micro-dissected with the guidance of lung pathologists and matched non-malignant lung tissue obtained for a subset of the primary tumors. DNA was extracted using standard phenol-chloroform procedures. RNA was extracted from tumor and matched non-malignant normal tissue using RNeasy Mini Kits (Qiagen) or Trizol reagent (Invitrogen). Quality and quantity of genomic material was assessed using a NanoDrop 1000 spectrophotometer and by gel electrophoresis and/or by Agilent 2100 Bioanalyzer. Demographic information for this cohort is summarized in Appendix A2. NSCLC cell lines H1993, H1355, H226, A549 were obtained from American Type Culture Collection and HCC4011 from Dr. Adi Gazdar (University of Texas Southwestern Medical Center at Dallas) and fingerprinted to confirm their identity [147]. All lines were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum and 0.1% Penicillin-Streptomycin (Invitrogen). 76  Immortalized normal human bronchial epithelial cells (HBEC) with (HBEC-KT53) and without p53 knockdown (HBEC-KT), courtesy of Dr. John Minna, were cultured in K-SFM media supplemented with 50ng/μl BPE and 5 ng/μl EGF (Invitrogen). Demographic data for the panel of cell lines used in this study can be found at http://edrn.jpl.nasa.gov/ecas/data/dataset/urn:edrn:UTSW_MutationData  4.2.2 Array comparative genomic hybridization and GISTIC analysis  Copy number profiles were generated for 261 NSCLC tumors using whole-genome tiling path array comparative genomic hybridization (aCGH), and were processed as previously described in Chapter 2. Probes were mapped to the March 2006 (Hg18) genomic coordinates, X and Y chromosomes removed and aCGH-Smooth was used to segment and smooth log2 ratio values [78]. The corresponding segments and ratio values were analyzed using the GISTIC algorithm [148] and gene pattern software (http://www.broadinstitute.org/cancer/software/genepattern/ )  to identify regions of significant amplification across samples. Amplification threshold of 0.8, join segment size of 2, qv threshold 0.05 and removal of the X chromosome were the settings applied for analysis.  4.2.3 Gene expression profiling and data integration  Gene expression profiles were generated using custom Agilent microarrays for a subset (35 AC and 13SqCC)  of the 261 tumors which had sufficient quantity and quality material for both tumor and matched non-malignant tissue. Data was normalized using the Robust Multiarray Average algorithm in R [80]. Genes were classified as over- or underexpressed if the mRNA fold change in tumors relative to matched non-malignant was greater or less than 2-fold. Mann-Whitney U tests with Benjamini Hochberg correction p<0.05 were used to compare expression of 12q15 genes between tumor and non-malignant tissue in 83 AC  pairs (EDRN cohort) and determine whether increased gene dosage resulted in increased gene expression. A Spearman's correlation conducted using MATLAB software was used to determine the strength of the correlation between copy number and expression, with a coefficient >0.55 considered significant.   77  4.2.4 Analysis of external cohorts Amplification of YEATS4 was assessed in six publically available datasets. Affymetrix SNP 6 data from GSE25016, Early Detection Research Network (EDRN), the Broad Institute and TCGA were downloaded and segmented using Partek Genomics Suite Copy Number, Paired Analysis Workflow, using the matched non-malignant profiles as a copy number baseline for each tumor and the following parameters; signal to noise >0.3, minimum of 50 markers per segment, p-value threshold of 10-7 for the statistical difference between intensities of adjacent segments and for significance of deviation of intensities in tumor tissue from intensities in non-malignant lung. An additional dataset with array CGH profiles from Memorial Sloan Kettering was segmented using the break point algorithm FAÇADE [149]. Data from the Sanger Institute's Cancer Cell Line Project was used to investigate the prevalence of YEATS4 amplification in human cancer cell lines. Copy gain was defined as 2.3 to 5 copies and amplification as greater than 5 copies. YEATS4 expression was also assessed in the EDRN and TCGA data sets. The details of all datasets used in this chapter are listed in Appendix A3.  4.2.5 Plasmids and generation of stably expressing YEATS4 lines Lentiviral short hairpin RNA constructs targeting YEATS4 were purchased from Open Biosystems. Lentiviral production and infection were performed as previously described [98]. Five independent pLKO.1 lentiviral shRNA constructs targeting YEATS4; shY4-1 (TRCN0000013143) shY4-2 (TRCN0000013144)  shY4-3 (TRCN0000013145), shY4-4 (TRCN0000013146)  and shY4- 5 (TRCN0000013147) were tested for their ability to reduce YEATS4 mRNA and protein expression. NSCLC cell lines were transfected with constructs containing a single shRNA targeting YEATS4 (shY4) or an empty vector control (PLKO). YEATS4 knockdown was confirmed by quantitative real-time PCR (RT-qPCR). shY4-1 resulted in the greatest degree of knockdown and was used for all subsequent experiments. Stably transfected lines were maintained in growth media supplemented with puromycin, 0.8 μg/ml (A549), 1 μg/ml (H1355, HCC4011, H226) or 1.5 μg/ml (H1993).    78  Ectopic expression of YEATS4 was achieved using Invitrogen's Ultimate ORF (clone IOH2880) following manufacturer's instructions. The YEATS4 gene insert was shuttled from a pENTRTM221 entry vector to the lentiviral destination vector pLenti6.3/V5-DEST by LR recombination. Lentiviral vector viral stock was produced by transfecting H293FT cells using Invitrogen's ViraPower TMHiPerformTM Lentiviral Expression System. HBEC-KT and HBEC-KT53 lines seeded in 6 well plates were transformed with virus expressing YEATS4 or an empty vector (EV) and stable transformants were selected following two weeks of treatment with 2μg/ml Blasticidin. Transfected lines were maintained in growth media containing 2μg/ml blasticidin. YEATS4 expression was confirmed by qRT-PCR and western blotting.  4.2.6 Quantitative reverse transcriptase PCR RT-qPCR was performed on SDS7900HT (Applied Biosystems) using the ∆∆Ct method with 18S rRNA expression levels used as a reference for normalization. Validation of 18S as an appropriate reference gene for lung cancer is described in [150]. Reverse transcription was performed using the High Capacity cDNA Reverse Transcription kit (Applied Biosystems) according to manufacturer's instructions. TaqMan gene expression assays and master mix used were: YEATS4 (Hs00232423_m1), 18S (Hs99999901_s1) and Gene Expression Master Mix (4369016). RT-qPCR was  performed on a reaction volume of 15μl containing 0.2 ng of cDNA and using default thermal cycling conditions (2 mins at 50°C, 10 min at 95°C and 40 cycles of 15sec at 95°C and 1 min at 60°C). Data was analyzed using 7500 Fast System Software v1.4 with auto calibration and outliers removed. Samples were analyzed with reference to their matched control, with an RQ of greater than 2 used to define overexpression. Expression array findings were validated in a cohort of 59AC pairs for which RNA was available for both the tumor and matched non-malignant tissue. Of the 59 AC pairs analyzed, 35 of the sample pairs were profiled by expression array. A Pearson correlation was used to determine the concordance between array data and RT-qPCR results.   79  4.2.7 siRNA mediated knockdown of CDKN1A On-Target plus SMART pool siRNA targeting CDKN1A (L-003471-00) and a non-targeting control (NTC) pool (D-001810-10) were purchased from Thermo Scientific. siRNA knockdown of CDKN1A was performed on H1993, H1355 and H226 PLKO and shY4 cell lines as described in Chapter 3. One day prior to infection, cells were seeded in six well plates at a density of 125,000 cells/well for senescence and 200,000 cells/well for lysates and collected 48 and 72 hours post transfection, respectively. Knockdown efficiency was assessed by western blot.  4.2.8 Western blot  Protein lysates were collected and western blots performed as previously described [151]. Membranes were incubated with primary antibodies against YEATS4 (W-21, Santa Cruz), MDM2 (HDM2-232, Santa Cruz), p53 (ab7757 AbCam), p21 (#2946 Cell Signaling), p14 (#2407 Cell Signaling), phospho p53 (#9284 Cell Signaling), cleaved PARP (#9541 Cell Signaling), pRb807/811 (Cell Signaling# #9308), p27/kip1 (#2552 Cell Signaling) and GAPDH or B-Actin as loading controls (#2118 and #4970 respectively, Cell Signaling). Following primary antibody incubation, membranes were incubated with anti-mouse or anti-rabbit HRP conjugated secondary antibodies (Cell Signaling #7074, 7076) and visualized by enhanced chemiluminescence.   4.2.9 Cell viability  MTT assays were used to assess cell viability following shRNA knockdown and overexpression of YEATS4. Cells were seeded in triplicate in 96 well plates at optimal densities for growth (2000 cells/well for H1993, HBEC-KT, HBEC KT-53, 2500 cells/well for H1355, H226, A549 and 4000 cells/well for HCC4011). Media only wells were plated to serve as absorbance blanks to normalize wells. Viability was measured over five consecutive days by the addition of 10ul MTT reagent (every 24 hours) and incubated for an additional four hours, followed by the addition of 100ul of 20% SDS to solubilize the MTT dye.  Plates were quantified by spectrophotometry (EMax plate reader, Molecular Devices) at 570nm with reference to 650nm. Experiments were performed in triplicate.  80  4.2.10 Colony formation Anchorage independent growth was assessed in all transfected cell lines by the soft agar method. Single cell suspensions were prepared in growth medium with 10% FBS and 0.3% low melting point agarose, and seeded in triplicate at 1000 cells/well in 12 well plates and cultured for 14 days at 37°C. Colonies were stained with MTT and counted, with colony formation is reported as the mean ± SEM normalized to the average of the control (PLKO) of triplicate experiments.  4.2.11 Cellular senescence Beta- galactosidase (βgal) activity at pH 6 was used as a measure of cellular senescence. Cells were plated in triplicate in 6 well plates at concentrations such that 24 hours later they were roughly 50% confluent and processed according to the manufacturer's instructions (Senescent Cells Histochemical staining kit, Sigma). Cells were visualized by a phase contrast microscope and images taken of three areas within each well. The percent of senescent cells was calculated and differences between PLKO and shY4 were assessed by a student's t-test with a p-value <0.05 considered significant.  4.2.12 Dose response assay To measure the effect of YEATS4 manipulation on cisplatin and nutlin sensitivity, cell viability assays were performed as described in Chapter 3. Cells were plated in triplicate in 96 well plates at optimal densities for growth (H226, A549, HBEC KT and KT53 at 2500 cells/well, H1993 and H1355 at 4000 cells/well, and HCC4011 at 6000 cells/well) and subjected to a series of 20, 2-fold dilutions of cisplatin or nutlin inhibitor prepared in cell growth media with final concentration of 1% DMSO. The experimental inhibitor concentrations ranged from 333 µM to 635 pM for cisplatin and 100µm to 191pm for nutlin. Experiments were repeated in triplicate and differences in IC50 values were determined using a student's t-test with a p-value < 0.05 considered significant.    81  4.2.13 Mouse models Tumor forming ability of H1993 and H1355 shY4 and PLKO cells was assessed in Crl:nu-foxn1nu mice. Subcutaneous flank injections of 3x106 cells in 50μl of PBS were injected into 6-week old mice obtained from The Jackson Laboratory. Tumor size and volume was measured by palpation 11 days following injection and every 6-8 days thereafter. Experiments were terminated once tumors reached a volume of 400 mm3 or became ulcerated. Tumor burden between mice injected with YEATS4 KD (n=10) cells and those expressing and empty vector (PLKO) (n=10) was assessed using a student's T-test.  4.2.14  Pathway analysis and gene set enrichment Following knockdown, gene expression profiles for H1993, H1355 and H226 shY4 and PLKO cells were generated using Illumina HT12 expression microarrays and normalized using BRB array tools. A list of over- and underexpressed genes (2-fold difference) for each knockdown line relative to empty vector controls were generated. Genes altered in all lines in the same direction were considered for target gene analysis using Ingenuity Pathway Analysis software. A pre-ranked gene set enrichment analysis was performed on fold change data for all genes for each cell line using the C3 transcription factor gene set to determine which transcription factor target genes were affected by YEATS4 knockdown.   4.2.15 Survival analysis  Survival analysis was performed using a Mantel-Cox log test in Matlab with p-values <0.05 considered significant. Director's Challenge data was sorted by YEATS4 expression and survival times between the top and bottom tertiles of expression compared and Kaplan-Meier plots generated.       82  4.3 Results  4.3.1 Recurrently amplified regions in NSCLC  Copy number profiles for 169 AC and 92 SqCC were generated using aCGH as described in Chapter 2. Significant regions of high level amplification (log2 ratio >0.8) were identified using the Genomic Identification of Significant Targets in Cancer (GISTIC) algorithm which calculates significance scores by considering both the amplitude and frequency of copy number alterations [148]. GISTIC analysis of all 261 samples (NSCLC) identified 3 significant regions of focal amplification; 7p11.2 (q=0.00075), 8p12 (q= 0.036) and 12q15 (q=0.036). Subtype specific analysis revealed 2 regions of amplification across the 169 AC tumors; 12q15 (q= 4.5x10-5) and 20q13.33 (q=0.017) and 6 regions across the 92 SqCC tumors; 1p34.2 (q= 0.044), 3q27.1 (q=1.4 x 10-10), 7p11.2 (q=0.029), 8p11.23 (q=0.0042), 8p12 (q=0.0042)  and 14q13.3 (q=0.03) (Fig. 4.1A-C). Amplification of these regions have been previously described in NSCLC indicating our tumors display patterns of alteration characteristic of lung cancer [18, 55, 152, 153].    While none of the regions identified were common between all three analyses, all of the regions identified in NSCLC were also significant in a subtype specific manner (Figure 4.1A-C). Further examination of these amplicons revealed that known oncogenes EGFR and BRF2, both of which are known to be preferentially amplified in SqCC, [114] [98] were driving selection of the 7p11.2 and 8q12 amplicons, respectively. Intriguingly, the primary target of 12q15 amplification, which is believed to be MDM2- a ubiquitin ligase that targets TP53 for proteasomal degradation, and when overexpressed results in aberrant p53 inactivation, was excluded from both the focal and wide peak boundaries. The exclusion of MDM2 from this focal region suggested that a gene other than MDM2 may be driving selection of this amplicon. This combined with the fact that all other regions harbored known oncogenes 7p11.2 (EGFR), 8p11.23 (FGFR1), 8p12 (BRF2) 14q13.3 (NKX2-1), 20q13.3 (EEF1A) or are known to be subtype specific regions of amplification (1p34.2 and 3q in SqCC) [18, 55] prompted us to further explore the 12q15 amplicon.  83  Figure 4.1    Figure 4.1. Recurrent amplifications in NSCLC.  GISTIC plots for (A) 261 NSCLC, (B) 169 AC and (C) 92 SqCC. Chromosomes are depicted as rows and chromosome numbers are indicated. Red peaks indicate frequently amplified regions and the green vertical line indicates the false discovery rate threshold (q=0.05). Peaks extending beyond this line indicates a significant region. X-axis indicates the GISTIC score scale. Genomic coordinates and the genes located within the 12q15 amplicon are shown below.      84  4.3.2 Identification of YEATS4, the target of 12q15 amplification  The peak amplified region of 12q15 spanned a 432 kb interval (68,030,736-68,462,888) and contained 7 genes; LYZ, YEATS4, FRS2, CCT2, LRRC10, BEST3, RAB3IP, none of which have been previously implicated in lung tumorigenesis. Based on the notion that selectively amplified oncogenes would demonstrate elevated expression, we integrated copy number and gene expression data for adenocarcinoma tumors and matched non-malignant tissue. Due to the limited size of our dataset with both copy number and expression data, identification of the 12q15 driver gene was performed in the largest dataset available (EDRN, n=83). Of the 7 genes within the amplicon, only YEATS4 was both gained/amplified and concomitantly overexpressed in lung tumors relative to matched non-malignant tissues (Figure. 4.2A-C). While YEATS4 has not been previously described in lung cancer, it is a well-established oncogene in cancers of neural origin [154, 155] and frequently amplified in liposarcomas [156].  4.3.3 YEATS4 is frequently amplified and overexpressed in NSCLC  YEATS4 was amplified in 18% (47/261) and overexpressed in 31% (15/48) of cases from our cohort. While 12q15 was not significant in the GISTIC analysis of our 92 SqCC cases, to conclusively determine whether amplification of YEATS4 was specific to AC, we compared copy number and expression data for both subtypes. Although no statistical difference in YEATS4 copy number or expression was observed between subtypes (Figure. 4.2D-F), AC tumors had a higher number of copies and greater fold change in expression compared to SqCC tumors. This suggests that while copy gain is a frequent event in both subtypes, it is likely a broader amplification event that occurs at a lower amplitude in SqCC relative to AC, which is why 12q15 failed to be identified by GISTIC in SqCC tumors. Analysis of external datasets with both AC and SqCC data supported our findings, with gain /amplification and overexpression occurring at similar frequencies in both data sets (Table 4.1), indicating that amplification and overexpression of YEATS4 is not subtype specific.     85   To gain further insight into the prevalence of YEATS4 amplification, we investigated YEATS4 copy number and expression in publically available NSCLC tumor datasets. YEATS4 was gained (2.3-5 copies) or amplified (> 5 copies) at various frequencies across the five datasets, ranging from 5-22% and 0.4-5% respectively (Table 4.1). A broader analysis of 508 human cancer cell lines revealed YEATS4 copy gain/amp in 43/128 (33.6%) of lung cancer cell lines and in 122/508 (24%) of all cancer cell lines (Table 4.1). Expression analysis of the EDRN and TCGA data sets, revealed YEATS4 was overexpressed at comparable frequencies to our dataset; 18% (15/83) and 33% (14/42), respectively (Table 4.1). Taken together, these results show YEATS4 is frequently gained and overexpressed in NSCLC, irrespective of subtype.   To validate array findings and verify YEATS4 is upregulated at the transcript level, we assessed YEATS4 expression by quantitative reverse transcriptase PCR (RT-qPCR) in a panel of 59 lung ACs relative to matched non-malignant tissue and in 18 NSCLC cell lines (2 SqCC and 16 AC) with reference to an immortalized normal human bronchial epithelial (HBEC) line. 15/59 (25.4%) tumors and 8/18 (44.4%) cell lines showed a two-fold or greater increase in YEATS4 expression relative to their matched control. Moreover, analysis of the 35 AC samples with expression data revealed a strong correlation between array findings and PCR results (r=0.75, P<0.001, Pearson Correlation, data not shown), validating array findings and confirming frequent overexpression of YEATS4. Western blotting of cell lines with and without YEATS4 amplification revealed increased YEATS4 expression in lines with amplification, demonstrating that amplification drives overexpression at both the mRNA and protein level (Figure. 4.2G). Multivariate analysis of YEATS4 revealed no significant associations with age, sex, stage, smoking status or race. Survival analysis of the Director's challenge expression datasets [157] revealed a trend towards poorer survival in patients with YEATS4 amplification, however failed to reach statistical significance (p>0.05, data not shown).       86  Figure 4.2   Figure 4.2 YEATS4 is the target of 12q15 amplification.   (A) Comparison of mRNA expression in 83 AC tumors and matched non-malignant tissue from the EDRN (p=0.0092). (B) YEATS4 expression between tumors with gain/amplification and tumors with neutral copy number (p<0.0001). (C) Spearman's correlation of copy number and expression for tumors with copy number alterations of YEATS4 (r=0.59, p=0.009). Comparison of YEATS4 copy number (D), expression (E) and fold change (F) between AC and SqCC. Expression values for all plots are in log2 units. (G) Immunoblot of YEATS4 in NSCLC lines with and without amplification of 12q15. GAPDH was used as a loading control.87  Table 4.1 Frequency of YEATS4 gain and amplification across multiple datasets       Data Set  Histology  # Samples with CN  Gain (n)  (2.3-5 copies)  Freq Gain (%)  Amplification (n)  (>5 copies) Freq Amp (%) Samples with Expression (n) Samples with OE Freq OE (%) BCCRC  AC & SqCC  169,92 28,19 16.5,20.1 5,2 3.0, 2.2 35,13 11, 4 31,30 EDRN  AC  83 18 21.7 4 4.8 83 15 18 GSE25016 AC & SqCC  77,155 10,22 13, 14.2 0,1 0, 0.6 N/A N/A N/A dbGAP  AC  354 18 5.1 4 1.1 N/A N/A N/A MSKCC  AC  199 21 10.6 N/A N/A N/A N/A N/A Broad  NSCLC  473 28 5.9 8 1.7 N/A N/A N/A TCGA AC & SqCC 277,201 19,22 6.9,10.9 15,6 5.4,3.0 17,25 8, 6 32, 35 Sanger LCCL NSCLC/SCLC 128 43 33.6 1 0.78 N/A N/A N/A Sanger cell lines All cancers 508 122 24 8 1.5 N/A N/A N/A           88  4.3.4 YEATS4 displays oncogenic properties in vitro and in vivo  YEATS4 encodes a protein found in a number of multi-subunit protein complexes involved in chromatin modification and transcriptional regulation and has also been shown to be involved in the regulation of TP53. To assess its oncogenic potential, YEATS4 was stably transfected into two immortalized HBEC lines; HBEC-KT and HBEC-KT53 (KT-YEATS and KT53-YEATS). Empty vector transfected cells were used as controls (KT-EV and KT53-EV). YEATS4 gene and protein expression was confirmed by qPCR and western blot (Fig. 4.3A-B). Relative to controls, ectopic expression of YEATS4 had no effect on viability and failed to induce anchorage independent growth in HBECs (data not shown), indicating that in normal, immortalized cells, YEATS4 overexpression alone is incapable of inducing colony formation. However, a dramatic inhibition of senescence in overexpressing cells relative to controls was observed in both lines (Student's t-test, p<0.05) (Fig. 3C-D), suggesting elevated YEATS4 expression is capable of inducing a phenotype associated with malignant transformation. 89  Figure 4.3  Figure 4.3. Overexpression of YEATS4 induces a malignant phenotype.  Ectopic expression of YEATS4 increases (A) mRNA expression (mean± SEM of triplicate replicates) and (B) protein levels relative to EV controls. GAPDH was used as a loading control. (C) β-Gal staining for cellular senescence in EV and YEATS4 expressing HBECs. Cells stained blue indicate senescence. Original magnification, 10x. (D) Quantification of cellular senescence in YEATS4 and control cells. The mean of the proportion of senescent cells (senescent cells/total cells) for YEATS4 and EV lines is shown for triplicate experiments ± SEM. ** p<0.01, Student's t-test.      90   Complimentary knockdown experiments using lentiviral shRNAs were performed in lung cancer cell lines with (H1993, H1355, H226) and without (A549, HCC4011) YEATS4 amplification and various p53 backgrounds. Empty vector transfected cells were used as controls (PLKO) and knockdown was confirmed by qPCR and western blotting (Figure 4.4A-B). Knockdown significantly decreased cell viability in H1993 and H1355 (p=0.0127 and  p=0.0172, respectively), both of which harbour YEATS4 amplification and mutant p53 (Figure 4.4C), but had no effect on A549, HCC4011 or H226 lines (p = 0.428, p = 0.45 and p = 0.49, respectively) which do not harbor YEATS4 amplification (A549 &H4011), or have YEATS4 amplification with wild type (wt) p53 (H226) (Figure 4.4C). Similarly, knockdown resulted in a significant decrease in anchorage-independent colony formation in H1993 (p = 7.26 x10-6) and H1355 (p = 6.06 x 10-10) cells, but not in A549 (p = 0.97), H4011 (p = 0.21) or H226 (p = 0.74) cells, indicating wt p53 may abrogate the effect of YEATS4 knockdown on viability and colony formation in lines with amplification (Figure 4.4D). A significant increase in senescence was observed in all three lines with amplification; H1993 (p = 5.71 x 10-6), H1355 (p = 0.0012) and H226 (p = 1.21 x10-13) as well as moderate increase in A549 (p = 2.99x 10-7). No difference in senescence was observed in HCC4011 (p = 0.06) (Figure 4.4E). The finding that A549 cells showed a modest increase in senescence is not surprising given the role of YEATS4 in the p53 pathway (discussed below) and the wt p53 background of this line, which enables pathway activation and cellular senescence.    To explore the oncogenic potential of YEATS4 in vivo, tumor formation in SCID mice was examined by subcutaneous flank injections of H1993 and H1355 control and shY4 cells. Tumor formation was significantly reduced in shY4 cells of both cell lines at all time points (Figure 4.4F,G). Our results demonstrate that knockdown of YEATS4 in cell lines with amplification effectively inhibits tumor growth, with a significant inhibition in viability, tumor and anchorage independent growth and increased cellular senescence, strongly supporting YEATS4 as an oncogene in NSCLC.     91  Figure 4.4   Figure 4.4. YEATS4 knockdown impairs growth and induces senescence.   92  Figure 4.4. YEATS4 knockdown impairs growth and induces senescence.  shRNA targeting YEATS4  significantly reduces (A) mRNA expression and (B) protein levels in all cell lines relative to controls (PLKO). GAPDH was used as a loading control. (C) Viability of cell lines with knockdown (shY4) relative to controls as measured by MTT. (D) Colony formation ability of shY4 cell lines relative to controls. (E) Quantification of cellular senescence based of β-Gal staining. Values reported as mean ± SEM of triplicate experiments.* p<0.05, ** p<0.01, *** p<0.001, Student's t-test of shY4 cells relative to PLKO. (F,G)  Effect of YEATS4 knockdown on tumor growth in mice injected with H1993 or H1355 PLKO and shY4 cells. Error bars indicate SEM of each group of 10 mice, * p<0.05   4.3.5 YEATS4 suppresses p53 and p21  Inactivation of the p53 pathway is one of the most frequent alterations in lung cancer, with somatic mutations occurring in approximately 50% of all cases [114, 158]. p53 is a key tumor suppressor that regulates cell cycle, DNA repair, apoptosis, and senescence and inhibits aberrant proliferation and the propagation of damaged cells. A study by Park et al, showed that under normal, unstressed conditions, YEATS4 binds to and inhibits the promoters of p14 and p21, subsequently repressing the p53 tumor suppressor pathway [159]. To assess whether this interaction occurs in NSCLC, we assessed these proteins in cell lines with YEATS4 manipulation. Upon YEATS4 knockdown, p21 and p53 protein levels were increased, with the greatest increases in expression of p21 and p53 observed in cell lines harbouring YEATS4 amplification or wt p53 respectively (Figure 4.5A). No change in p14 levels was observed upon knockdown. Overexpressing lines showed a modest reduction of p21 and p14 as well as a reduction of p53 levels in HBEC-KT (Figure 4.5B). MDM2 levels remained unchanged following knockdown or overexpression of YEATS4, indicating that the observed changes in p21, p14 and p53 were a direct result of YEATS4 manipulation.   4.3.6 YEATS4 alters the sensitivity of cell lines to cisplatin and nutlin  To determine whether the downstream effects of YEATS4 manipulation alters cellular sensitivity to chemotherapy, cell lines were treated with serial dilutions of cisplatin, a commonly 93  prescribed first line chemotherapy for lung cancer patients that crosslinks DNA triggering apoptosis, or nutlin, a cis-imidazoline analog that inhibits the interaction of p53 and MDM2, stabilizing p53. Based on the observed effects on p53 and p21 protein levels following manipulation of YEATS4 expression and the notion that cells with YEATS4 amplification may be dependent on YEATS4 for growth and survival, we hypothesized that HBEC-KT/KT53-Y cells would be more resistant to treatment, while shY4 cells harbouring YEATS4 amplification would be more sensitive.    As expected, HBEC-KT-YEATS and HBEC-KT53-YEATS lines were significantly more resistant to both cisplatin and nutlin than their control counterparts (Figure 4.5C; Table 4.2). Differences in sensitivity were less consistent in the lung cancer cell lines, likely due to the fact these cell lines harbour numerous genomic alterations which could influence drug sensitivities. While H1993 shY4 cells were significantly more sensitive to cisplatin (IC50 PLKO:11.45 vs. shY4:8.65) (Figure 4.5D) supporting our hypothesis, knockdown in both H1355 and H226, showed the opposite trend resulting in greater resistance relative to controls (Table 4.2). As anticipated, A549 and HCC4011 shY4 cells showed no difference in sensitivity. As specimens with mutant p53 are resistant to nutlin, only A549 and H226 were treated. Similar to the cisplatin results, A549 shY4 cells showed no significant difference in sensitivity to nutlin (PLKO: 7.58 vs. shY4:6.91, p=0.84), while H226 shY4 cells were unexpectedly significantly more resistant (PLKO:3.27, shY4:4.33, p=0.033) (Table 4.2). Analysis of lung cancer cell line IC50 data from the Sanger drug sensitivity project failed to reveal a significant association between YEATS4 amplification and response to cisplatin or nutlin. However, based on the fact that transformed bronchial epithelial cells which harbour minimal genetic alterations were significantly more resistant to cisplatin and nutlin following overexpression of YEATS4, and  H1993 shY4 cells (which harbor the greatest amplification of YEATS4) were more sensitive to cisplatin compared to controls, we feel this data supports the notion that YEATS4  alters the in vitro sensitivity of lung cells to cisplatin and nutlin.    94  Figure 4.5  Figure 4.5. YEATS4 alters p21 and p53 protein levels.   95  Figure 4.5. YEATS4 alters p21 and p53 protein levels.  (A) Knockdown of YEATS4 increases expression of p21 in cell lines with YEATS4 amplification and increases p53 in all lines that express p53. (B) Overexpression of YEATS4 reduces p14 and p21 levels in both HBEC lines, and p53 only in the HBEC KT line. Dose-response curves of HBEC KT (C) and H1993 (D) cells treated with 2-fold dilutions of cisplatin for 72 hours. Viability is shown as a proportion of treated cells against untreated controls (mean ± SEM of triplicate experiments). (E) Immunoblot of PLKO and shY4 cell lines treated with 40μM of Cisplatin for 0, 24 or 48 hours. Cisplatin treatment induces apoptosis as measured by the increase in cleaved PARP, p53 and phosphorylated p53 (Ser15). GAPDH was used as a loading control for all blots.   Table 4.2 Cisplatin and Nutlin IC50s   Cisplatin Nutlin  Cell Line YEATS4 IC50 SEM t Test IC50 SEM t Test Trend H1993 PLKO 11.45 1.11      H1993 shY4 8.649 0.488 0.004    Sensitive H1355 PLKO 9.111 0.491      H1355 shY4 15.91 0.905 1.6E-06    Resistant H226 PLKO 5.788 0.276  3.266 0.130   H226 shY4 9.965 0.716 0.0003 4.325 0.376 0.033 Resistant in both A549 PLKO 10.88 0.452  7.58 0.316   A549 shY4 11.4 0.705 0.204 6.908 0.123 0.084 Not significant H4011 PLKO 8.952 0.326      H4011 shY4 10.32 0.566 0.055    Not significant HBEC KT EV 11.09 1.472  16.55 1.405   HBEC KT YEATS 17.32 0.696 0.007 26.01 2.696 0.023 Resistant in both HBEC KT53 EV 15.41 1.612  19.63 1.367   HBEC KT53 YEATS 20.38 0.718 0.029 24.85 1.385 0.034 Resistant in both   96  4.3.7 Sensitivity to cisplatin is not mediated solely through the p53-p21 pathway  To gain further insight into the potential mechanisms of altered sensitivity to cisplatin, cell lines were treated with 40μM cisplatin and protein lysates collected at 0, 24 and 48 hours post treatment. As expected, cisplatin treatment resulted in an increase in p53, p53 Ser15 phosphorylation (a marker of stabilization), p21 and induced apoptosis as measured by cleaved PARP in HBECs. However, no differences between HBEC-EV and HBEC-YEATS cells were observed for any of the proteins examined (data not shown). In shY4 cells with amplification, treatment with cisplatin led to a greater induction of p53 and phospho-p53 (Ser15), and in H226 also led to a significant increase in p21 levels relative to control cells (Figure 4.5E). As no significant differences in protein levels were observed, despite a significant increase in resistance following OE, our results suggests that while the p53-p21 signaling pathway may be involved, resistance is likely mediated through the interaction of YEATS4 with other signaling pathways.   4.3.8 Knockdown phenotypes are independent of p21 signaling in mutant p53 cells  To explore the effect of increased p21 expression on the observed phenotypes, siRNA knockdown of CDKN1A was performed on shY4 and PLKO cells for cell lines with YEATS4 amplification. Knockdown of CDKN1A showed no effect on viability or colony formation in any of the lines (data not shown), but significantly altered senescence levels in the presence of wt p53 (Figure 4.6A). CDKN1A siRNA reduced senescence in both H226 shY4 and PLKO cells relative to non-targeting control siRNA treated cells, such that the percent of senescent H226 shY4-p21 cells was similar to H226 PLKO-NTC (Figure 4.6A). These experiments suggest that in a wildtype p53 background, the increase in senescence following YEATS4 knockdown occurs in a p53 dependent manner and is the direct result of increased p21expression. As CDKN1A knockdown failed to rescue viability, colony formation and senescence in cell lines with mutant p53, these findings further support the notion that the phenotypes observed following YEATS4 knockdown are not solely due to changes in p53-p21 signaling. Based on these findings, and the prominent role of Rb in senescence, we next investigated whether the increased senescence following YEATS4 knockdown could be due to altered Rb signaling. Modest reductions in Rb Ser807/811 phosphorylation following YEATS4 knockdown were observed, however in mutant p53 cell lines this does not appear to be due to increased levels of p27 (Figure 4.6B). 97  Figure 4.6  Figure 4.6. Senescence induced by YEATS4 is dependent on p21 in wildtype p53 cells.  (A) Quantification of  percent senescent cells following siRNA mediated knockdown of p21 in PLKO and shY4 cells. * p<0.05, Student's t-test. Values reported as mean ± SEM of triplicate experiments. (B)  Knockdown of YEAST4 leads to reduced  Rb Ser807/811 phosphorylation and p27 in p53 mutant cells. B-actin was used as a loading control.      98  4.3.9 Identification of additional cellular networks regulated by YEATS4  In an attempt to gain a better understanding of other pathways YEATS4 is involved in, we performed expression profiling on shY4 and PLKO cells for the three cell lines with YEATS4 amplification. To identify significantly enriched pathways/networks and gene sets affected by YEATS4 knockdown, Ingenuity Pathway Analysis (IPA) and Gene Set Enrichment Analysis (GSEA) were performed. A total of 32 genes (27 overexpressed and 5 underexpressed, Table 4.3) were differentially expressed between knockdown and control cells across all three cell lines. Due to the small number of input genes, none of the significantly enriched canonical pathways passed multiple testing correction. However, network analysis, which assesses regulatory relationships existing between genes and proteins, identified two networks associated with pro-tumorigenic functions; (1) cancer and  (2) cell death, survival, cell cycle and cell morphology. These networks were centered around known targets or binding partners of YEATS4 including p53, CDKN1A and MYC (Figure 4.7), further supporting our in vitro findings. Pre-ranked GSEA revealed significant enrichment of a number of transcription factor gene sets including MYCN, which has been shown to be a binding partner of YEATS4 and all 6 serum response factor (SRF) gene sets. SRF is a ubiquitously expressed transcription factor implicated in cell proliferation, differentiation, metastasis and clinically associated with castration-resistant prostate cancer [160, 161]. Interestingly, PDLIM7 which contains a serum response element and is transcribed upon induction of SRF, was shown to inhibit p53 and p21 through the inhibition of MDM2 self ubiquitination. While neither MYCN, SRF or PDLIM7 were differentially disrupted at the mRNA level following knockdown, our downstream analysis suggests the target genes of these two transcription factors could be involved in YEATS4 mediated tumorigenesis and warrant investigation in future studies to elucidate additional mechanisms through which YEATS4 promotes tumorigenesis.        99  Table 4.3 Differentially altered genes following knockdown of YEATS4 used for Ingenuity Pathway Analysis Gene Direction of change YEATS4 Underexpressed MUC1 Underexpressed SCNN1A Underexpressed ST3GAL5 Underexpressed KLHL5 Underexpressed CRABP2 Underexpressed LOC283267 Underexpressed TINP1 Underexpressed FNDC3A Underexpressed LOC100128196 Underexpressed NSA2 Underexpressed ACOT7 Underexpressed S100A4 Underexpressed NKD2 Underexpressed MRPL35 Underexpressed SULT1A3 Underexpressed ISCA1 Underexpressed LOC100132658 Underexpressed DENND2D Underexpressed MGC42367 Underexpressed LIN7C Underexpressed LOC400061 Underexpressed SLC1A6 Underexpressed DDX12 Underexpressed RRAD Underexpressed PPIL5 Underexpressed TNFSF12 Underexpressed CAMTA1 Overexpressed RRAGD Overexpressed TMEM158 Overexpressed TAGLN Overexpressed GLIPR1 Overexpressed      100  Figure 4.7  Figure 4.7. Gene networks associated with YEATS4 knockdown.  Ingenuity Pathway Analysis was used to identify biologically related gene networks from the 34 genes with differential expression across all three cell lines (H1993, H1355, H226) following knockdown. One of the top networks- cell death, survival, cell cycle and cell morphology related to p53, CDKN1A, TNF and MYC signaling is displayed. Solid lines denote direct interactions while dotted lines represent indirect relationships. Components highlighted in red are upregulated while those in green are downregulated.  101  4.4 Discussion  While single dimensional genomic analyses have been instrumental in cancer gene discovery, this type of analysis often overlooks genes disrupted at low frequencies, and is unlikely to distinguish causal from passenger events. The integration of multiple parallel genomic dimensions enables the identification of genes with concurrent DNA and expression alterations, which are likely selected for due to their roles in driving cancer phenotypes [61]. Towards this end, we integrated copy number and gene expression data in an attempt to identify novel oncogenes important in lung tumorigenesis. While our analysis revealed gains/ amplifications in a number of regions previously reported in NSCLC, the amplicon at 12q15 was the only one without a candidate driver gene located within the amplicon boundaries and was therefore the only regions we pursued further. Integration of expression and copy number data for the 7 genes located within 12q15 identified YEATS4 as the candidate target gene of this amplicon.   First identified and isolated in the glioblastoma multiforme cell line TX3868, YEATS4 is a highly conserved nuclear protein essential for cell viability that is frequently amplified in gliomas, astrocytomas and liposarcomas [154, 156, 162]. A member of a protein family characterized by the presence of an N-terminal YEATS domain, YEATS4 shares high homology with transcription factor family members AF-9 and ENL [163]. Like other family members, YEATS4 is involved in chromatin modification and transcriptional regulation through its incorporation into multi-subunit complexes; specifically the human TIP60/TRRAP and SRCAP complexes [164, 165], which mediate the incorporation of an H2A variant histone protein into nucleosomes, altering chromatin structure and controlling transcriptional regulation.   In addition to its role in transcriptional regulation, yeast two hybrid screens have revealed a number of YEATS4 binding partners. These include MYC, MYCN, TACC1, TACC2, NuMa, AF10, PFDN1 and KIAA1009 [166-170]. Analysis of expression data before and after YEATS4 knockdown showed no effect on expression of any binding partners, suggesting that YEATS4 does not control the expression of its binding partners at the mRNA level. To date, the majority of work surrounding YEATS4 has focused primarily on the identification of YEATS4 binding 102  partners with only a few studies having explored the phenotypic effects of YEATS4 amplification, none of which have been performed in lung [159, 170].    Our study is the first to show gain/amplification and overexpression of YEATS4 in NSCLC and the first to implicate amplification of YEATS4 in lung cancer tumorigenesis. We observed frequent gain/amplification of YEATS4 in multiple independent tumor cohorts in addition to our own, as well as a strong correlation between gain and overexpression in both tumors and cell lines (Figure 4.2). Analysis of the catalogue of somatic mutations in cancer (COSMIC) revealed YEATS4 is rarely mutated in lung (0.23%) or any cancer type (0.17%) suggesting that DNA amplification is the predominant mechanism of activation. In addition to the genomic evidence supporting selection of YEATS4 in NSCLC, we demonstrate the oncogenic potential of YEATS4 both in vitro and in vivo (Figures 4.3&4.4). Ectopic expression resulted in a significant reduction in senescence, suggesting overexpression of YEATS4 is sufficient to induce phenotypic changes characteristic of malignant transformation, (Figure 4.3) while knockdown in cell lines with amplification and mutant p53 showed reduced viability and colony formation along with an increase in senescence, consistent with oncogenic function. While wt p53 abrogates the effects on viability and colony formation on YEATS4 knockdown lines with amplification, a significant increase in senescence was still observed. In addition to these phenotypic effects, we also demonstrated that YEATS4 inhibits p21 thereby repressing p53 activity, consistent with the findings of Park and Roeder who demonstrated this interaction in unstressed conditions [159]. siRNA-mediated knockdown of CDKN1A failed to rescue viability, colony formation and senescence in mutant p53 backgrounds suggesting the phenotypic effects of YEATS4 amplification occur through a mechanism other than p21.   MDM2, an E3 ubiquitin ligase, is the major negative regulator of p53, mediating its ubiquitination and subsequent degradation [171, 172]. Overexpression results in inactivation of p53 and is a common mechanism of p53 inactivation in cancer. MDM2 is frequently amplified and overexpressed in human cancers including lung cancer, and is largely considered to be the driver gene of the 12q15 amplicon [18, 173]. We were therefore intrigued to discover that despite being frequently gained in our dataset, MDM2 did not fall within the boundaries of the 103  12q15 amplicon identified in our cohort. This led us to postulate that an alternative oncogene was being selected in this region. When looking at high resolution copy number profiles, while the majority of cases showed identical copy number for both YEATS4 and MDM2, a small number of cases (3/83) had more copies of YEATS4 than MDM2, suggesting YEATS4 is selected as the target of amplification in these samples and that amplification of YEATS4 is not merely a passenger event of MDM2 amplification. Of note, 4/83 cases had higher level gain/amplification of MDM2 relative to YEATS4. For cancers with amplification of 12q15 spanning both YEATS4 and MDM2, these genes may work synergistically to suppress p53, however further experimentation is required to investigate this hypothesis. Along with the many tumor promoting effects of YEATS4, of immediate clinical interest is our discovery of a YEATS4 dependent mechanism of reduced cisplatin and nutlin sensitivity, which appears to occur at least in part through inhibition of p21 and subsequent suppression of the p53 pathway.    In summary, we have shown that 12q15 is frequently amplified in both AC and SqCC and suggest that YEATS4, given its multiple oncogenic functions, is a novel oncogene involved in lung tumorigenesis. Our findings imply that MDM2 is not the sole driver gene targeted by amplification of 12q15 and that suppression of the p53 pathway can be achieved through amplification of YEATS4 via inhibition p21. For the first time, we demonstrate that YEATS4 expression alters cisplatin and nutlin sensitivity. Additional investigation into the signaling pathways altered as a result of YEATS4 amplification will provide further insight into the mechanism underlying YEATS4-mediated tumorigenesis and its potential clinical relevance.   104  Chapter 5: Characterizing miRNA expression in lung adenocarcinoma and squamous cell carcinoma identifies miR-944 and miR-1287 as subtype specific miRNAs   5.1 Introduction  The human genome is comprised of less than 2% protein coding genes, however more than 90% of the genome is transcribed, suggesting that the majority of the transcriptome is comprised of non-coding RNAs (ncRNAs) - transcripts that lack an open reading frame and as such do not encode a protein [174-176]. This by no means implies that ncRNAs lack function, but rather highlights the importance of looking beyond protein coding genes in order to improve our knowledge of normal and disease biology. While some are known to play important roles in the regulation of gene expression, splicing, epigenetic control, chromatin structure and nuclear transport, the function of most ncRNAs remains unknown [177, 178]. miRNAs are the most thoroughly investigated class of ncRNAs and are often located at chromosomal breakpoint regions, fragile sites, and minimal regions of loss of heterozygosity or amplification, making miRNA loci highly susceptible to genomic alterations and subsequently, de-regulated expression [179, 180]. To date, over 2000 human miRNAs have been identified [181] and a single miRNA is capable of affecting multiple protein coding genes. It is believed that over one third of the genome is regulated by at least one miRNA [182]. With roles in cellular functions from proliferation and apoptosis to cellular development and epithelial to mesenchymal transition, it is not surprising that miRNA deregulation has been linked to human diseases including cancer.    MicroRNAs (miRNAs) are small 18-25 nucleotide non-coding RNAs that negatively regulate gene expression post-transcriptionally through transcript degradation or translational repression [56]. miRNAs are known to be frequently deregulated in cancer as well as pre-invasive lesions and circulate in bodily fluids (blood, sputum, urine etc.) with substantial stability, making them ideal candidates for non-invasive, early detection biomarkers. While protein coding genes remain the primary focus of current genomic and proteomic studies, the deregulation of several miRNAs (let-7a, miR-34, -125, -126 -221, -222, to name a few) have 105  been implicated in lung tumorigenesis [183-187], and in the past decade miRNA expression profiles have been associated with lung cancer prognosis, disease progression, survival, and outcome prediction, as well as subtype discrimination [58, 183, 188, 189]. However, these studies are limited by the use of microarray profiling which restricts the number of miRNAs assessed, small sample sizes, or the absence of patient matched non-malignant tissue. Moreover, expression profiling studies aimed at distinguishing differences between AC and SqCC have used minimal statistical criteria to identify differentially expressed miRNA, rarely including tumor/non-malignant fold change criteria and typically just correlating miRNA expression with clinical features. As such, an understanding of the recurrently altered miRNAs defining each subtype and how these miRNAs contribute to subtype specific tumorigenesis through the genes and pathways they affect is lacking. We hypothesize that similar to DNA level alterations, the differential deregulation of miRNAs and their consequential effects on genes and pathways may contribute to the disparate clinical phenotypes observed in AC and SqCC.   In this study we performed a genome wide, unbiased analysis of miRNA expression patterns in NSCLC subtypes. miRNA sequencing was performed on a cohort of 88 NSCLC tumors with matched non-malignant tissue (66 AC and 22 SqCC) and detailed clinical information. We identified miRNA panels capable of distinguishing tumor from non-malignant tissue (n=85) and AC from SqCC (n=47), and validated our findings in independent, external datasets. Network and pathway analysis was performed on the most robust mRNA targets of our validated miRNAs based on in silico target prediction algorithms. Consistent with our observations of protein coding genes (Chapter 2), pathway analysis of target genes disrupted by SqCC specific miRNAs revealed disruption of multiple histone modifying enzymes. In vitro analysis of miR-1287, an AC specific miRNA, confirmed RAD9A as a biological target and implicated miR-1287 in AC tumorigenesis. Overexpression of miR-1287 in HBECs and lung cancer cell lines resulted in reduced apoptosis and DNA damage repair, validating the ability of our approach to identify biologically relevant miRNAs. To our knowledge, this is the first study to 1) perform a genome wide unbiased comparison of miRNA expression profiles in AC and SqCC, and 2) identify the pathways and functions disrupted by these subtype specific miRNAs.   106  5.2 Methods  5.2.1 NSCLC patients and samples 88 fresh-frozen lung tumors (66AC and 22SqCC) with matched non-malignant tissue were obtained from treatment naïve patients undergoing surgical resection with curative intent (BCCRC cohort). Samples were collected from the Tumor Tissue Repository of British Columbia Cancer Agency under informed, written patient consent. Tissue sections were microdissected with the guidance of a lung pathologist to obtain >70% tumor cell content. Total RNA was extracted using Trizol reagent (Invitrogen) and then size fractionated to isolate small RNAs including miRNA.   5.2.2 MicroRNA sequencing and data analysis miRNA sequencing libraries for the 88 fresh-frozen tumor pairs were constructed, bar-coded for multiplex sequencing and sequenced on the Illumina HiSeq 2000 platform using a plate-based protocol developed by the British Columbia Genome Sciences Center (BCGSC) [190]. Raw sequence reads were separated into individual samples based on their assigned indexes, adapter sequences removed and reads trimmed based on quality control metrics. High quality reads were aligned to the NCBI GRCh37 reference genome and miRBase v18 using the BWA algorithm. Reads were normalized as reads per kilobase of exon model per million mapped reads (RPKM).  5.2.3 Identification of subtype specific miRNAs and NSCLC miRNAs Expression levels for identical miRNAs from different genomic locations were summed, leaving 1372 unique miRNAs for examination. miRNAs with read counts <1 were considered undetected/not expressed and miRNAs with undetectable expression in all samples within a group were excluded from further analysis, resulting in 916 miRNAs for all tumors and normals, 859 miRNA in AC and 753 miRNAs in SqCC for further analysis. The fold change in miRNA expression between tumor and normal pairs was calculated for each miRNA in each sample, and a 2-fold change in expression was used as a cutoff to define over- and underexpression. miRNAs without 2-fold over- or underexpression in any of the samples within a subtype were removed, leaving 513 miRNAs in AC and 419 miRNAs in SqCC.  107  To be considered subtype specific, miRNAs had to meet three criteria:  i) differential expression in tumors versus non-malignant tissue, as defined by a Wilcoxon Sign Rank with Benjamini-Hochberg (BH) multiple testing correction p<0.05, ii) over- or underexpressed in at least 25% of samples within the subtype and a difference in frequency of aberrant expression >15% between subtypes, and iii) differential tumor expression and frequency of alteration between AC and SqCC as defined by permutation and Fisher's Exact tests with BH correction p<0.05. NSCLC miRNA were defined as i) differentially altered in tumors versus non-malignant tissue based on a Wilcoxon Sign Rank with Benjamini-Hochberg (BH) multiple testing correction p<0.05, ii) over- or underexpressed in at least 25% of samples and <15% difference in frequency between subtypes OR over-/underexpression in >50% of both subtypes, and iii) no difference in expression or frequency of alteration between AC vs SqCC, based on permutation and Fisher's Exact tests with BH correction p>0.05.  5.2.4 Statistical analysis of miRNA expression data Unsupervised hierarchical clustering using Ward's method was performed on tumor samples only (n=88), and non-malignant samples only (n=88) using Partek Genomics Suite software. A Fisher's Exact test was performed to assess the distribution of subtypes in tumor and non-malignant profiles, with a p<0.05 considered significant.   Principal component analysis was performed using miRNA sequencing data for AC and SqCC tumors in Matlab. All 47 subtype specific miRNAs were used to generate the principal components. Receiver Operating Characteristic (ROC) area under the curve (AUC) analysis was performed in GraphPad Prism Software on subtype specific miRNAs to determine the ability of these miRNAs to accurately discriminate AC from SqCC (p<0.05).   To identify miRNAs associated with clinical features (gender, smoking status, stage and histology), a multivariate analysis of variance (MANOVA) on the 88 tumors with miRNA sequencing data was performed in R. A p<0.05 was considered significant. MANOVA results were used to confirm that the variance in subtype specific miRNA expression was in fact due to histology. 108  5.2.5 Validation in external cohorts Level 3 miRNA sequencing data was obtained from the TCGA for use as an external cohort for validation. Expression profiles were processed as described for 'Level 3 data' in the TCGA data compendium (2011 Cancer Genome Atlas Network). Data was downloaded for all patients with tumor and matched non-malignant tissue, which at the time included 35 AC and 35 SqCC. Subtype specific and NSCLC miRNAs were assessed using the same criteria applied in our initial discovery cohort, and only those passing all criteria were considered validated.   5.2.6 Integration of DNA copy number and methylation with miRNA expression Affymetrix SNP 6.0 copy number and Illumina HM27K DNA methylation profiles were available for 77 (62 AC, 15SqCC) and 58 (45 AC, 13 SqCC) cases with miRNA sequencing, respectively. miRNA copy number and methylation status were integrated with expression data in order to determine the frequency of concerted DNA and expression alterations. Array data was processed as described in Chapter 2 and standard thresholds were used to define copy gain (>2.3 copies), loss (<1.7 copies), DNA hypomethylation (∆β <-0.15) and hypermethylation (∆β>0.15). miRNAs were considered to be controlled by genetic or epigenetic mechanisms if concerted DNA disruption (copy number or DNA methylation changes) and miRNA expression changes were observed in a concordant direction (e.g. copy loss and underexpression) in >20% of cases.   5.2.7 Target prediction, interaction networks and pathway analysis miRNAs identified as NSCLC or subtype specific were input into the microRNA Data Integration Portal v.2 (http://ophid.utoronto.ca/mirDIP), which integrates 13 microRNA target prediction algorithms and six microRNA prediction databases to predict miRNA-transcript (mRNA) interactions [191, 192]. For this study, we used stringent miRNA target prediction criteria by considering only predictions that were supported by at least six sources. Interactions between miRNAs and their predicted mRNA targets were then visualized as networks using NAViGaTOR v.2.14 (http://ophid.utoronto.ca/navigator) [193, 194]. Interaction networks were generated for subtype specific miRNAs as well as NSCLC miRNAs disrupted in >90% of cases. Pathway analysis was performed on all predicted mRNA targets (miRTarBase v3.5) of miRNAs disrupted in a subtype specific manner using Ingenuity Pathway Analysis. 109  5.2.8 Cell culture and manipulation of miR-1287 Immortalized human bronchial epithelial cells courtesy of Dr. John Minna were maintained in KSFM supplemented with 5ng/μl EGF and 50ng/μl BPE. AC cell lines A549 and H1993 were obtained from the American Type Culture Collection and cultured and maintained in RPMI supplemented with 10% FBS and 0.1% penicillin-streptomycin (Invitrogen). HBECs and lung cancer cell lines were transfected with 20nM of miR-1287 mimic or a non-targeting control (Thermo Scientific) according to manufacturer's instructions. miR-1287 expression was verified by qRT-PCR 48 hours post-transfection and target gene expression by western blot 72 hours post-transfection.  5.2.9 3'-UTR reporter assays A commercially available RAD9A 3'-UTR dual luciferase reporter plasmid (pEZX-RAD9A-UTR) was purchased from GeneCopoeia. HBECs and lung AC cell lines were transfected with miR-1287 mimic or the non-targeting control for 24 hours and then the cells were transfected with the reporter vector. Luciferase assays were performed 48 hours later using the Luc-Pair miR Luciferase Assay kit (GeneCopoeia) and luminescence was measured on a luminometer. Firefly luciferase activity was normalized to Renilla luminescence to account for differences in transfection efficiency.   5.2.10 qPCR of miRNA expression Quantitative real-time PCR analysis of miR-1287 was performed using the miRNA-specific TaqMan MicroRNA Assay Kit (Applied Biosystems) and an Applied Biosystems 7500 Fast Real Time PCR system. 10 nanograms of sample RNA was converted to cDNA using the TaqMan MicroRNA Reverse Transcription kit and 10μg of cDNA was combined with TaqMan Universal PCR Master Mix without AmpErase Uracil N-glycosylase and miRNA specific primers as per the manufacturer's instructions (Applied Biosystems). U6 small nuclear RNA was used as an endogenous control to normalize cDNA input.  110  5.2.11 Western blotting Protein lysates were collected and western blots performed as previously described [151]. Membranes were incubated with primary antibodies against RAD9A (ab70810, AbCam at 1:1000) and TEAD3 (#13224, Cell Signaling at 1:1000) and GAPDH as a loading control (#2118, Cell Signaling at 1:50,000). Following primary antibody incubation, membranes were incubated with anti-rabbit HRP conjugated secondary antibodies (Cell Signaling #7076) and visualized by enhanced chemiluminescence.   5.2.12 MTT assay MTT assays were performed as previously described [195] to assess cell viability following miR-1287 overexpression. Cells were seeded in triplicate in 96 well plates at optimal densities for growth (1000 cells/well for A549 and 1500 cells/well for H1993 and HBECs). 48 hours post transfection, viability was measured over five consecutive days and quantified by spectrophotometry. Experiments were performed in triplicate.  5.2.13 Apoptosis, DNA damage and cell proliferation Apoptosis, DNA damage and cell proliferation were assessed by measurement of propidium iodide (PI) and Annexin staining, phospho-ɣH2AX and 5-bromo-2'-deoxyuridine (BrdU) incorporation, respectively (FITC Annexin V Apoptosis Detection Kit and Apoptosis, DNA Damage and Cell proliferation kit, BD Pharmigen). Cells were seeded in 6-well plates at a density of 1.5x105 for A549 and 2x105 for HBEC and H1993 and 24 hours after seeding they were transfected with miR-1287 mimic or non-targeting control. 48 hours after transfection, cells were treated with 10uM cisplatin for 4 hours, stained with BrdU and then allowed to recover for 24 hours. Following recovery, cells were trypsinized, washed, and processed as per the manufacturer's instructions. Flow cytometry was performed using a FACSCanto II flow cytometer and analyzed with FACSDiva software (BD Biosciences). Results are reported as the average percent of positive staining cells ± SEM of triplicate experiments.   111  5.3 Results  5.3.1 MiRNAs are differentially expressed between NSCLC tumors and non-malignant tissue and accurately segregate profiles based on malignancy.  Expression levels for 1372 unique miRNAs generated by miRNA sequencing of 176 fresh-frozen NSCLC tumors and matched non-malignant tissues (66AC and 22 SqCC pairs, Table 5.1) were examined. To assess the similarity of miRNA expression profiles across tumors and matched non-malignant tissue, unsupervised hierarchical clustering was performed on 916 miRNAs with detectable expression. Clustering revealed two distinct clusters; one comprised of all tumors and the other of all non-malignant tissues and two tumor samples, demonstrating that miRNA expression strongly differentiates tumors from normals (Figure 5.1A).    Based on the accuracy of clustering, we next sought to determine which miRNAs were differentially expressed between tumors and matched non-malignant tissue. 215 miRNAs met our criteria of being differentially altered in tumors vs. matched non-malignant tissue (Sign Rank with Benjamini-Hochberg correction p<0.05), and aberrantly expressed (2 fold or greater between tumor and normal) in at least 25% of samples. However, only 85 miRNAs had no significant difference in expression or frequency of alteration between subtypes (Fishers Exact test, BH p>0.05) and were deemed "true" NSCLC miRNAs. 18 miRNAs (12 OE, 6 UE, Table 5.2) were altered in greater than 90% of all tumors, and principal component analysis of these miRNAs confirmed their ability to accurately discriminate tumors from non-malignant tissue (Figure 5.1B), highlighting the potential of these 18 miRNAs as biomarkers for NSCLC detection. Interestingly, PCA revealed tumors cluster more closely together than non-malignant tissues, suggesting there is greater inter-individual variance in miRNA expression in non-malignant tissues than tumors (Figure 5.1B).   112  Figure 5.1  Figure 5.1. miRNA expression profiles accurately segregate tumor and non-malignant tissue Clustering of 176 miRNA expression profiles (66 AC and 22 SqCC with matched non malignant tissue) revealed two distinct clusters associated with malignancy, one comprised of tumor samples (blue) and the other of all non-malignant tissues (red) and two tumor samples (A). Principal component analysis using expression of the 18 miRNAs altered in >90% of all cases accurately separated tumor and non-malignant tissue (B). Blue dots represent non-malignant tissue while red dots represent tumor samples.    A. B. 113   Table 5.1 Patient demographics for all cases with miRNA sequencing data Feature AC (n=66) SqCC (n=22) Total (n=88) Stage    I 41(62.1%) 7 (31.8%) 48 (54.5%) II 17 (25.8%) 10 (45.5%) 27 (30.7%) III 5 (7.6%) 4 (13.6%) 8 (9.1%) IV 2 (3.0%) 1 (4.5%) 3 (3.4%) Sex    Female 44 (66.7%) 7 (31.8%) 51 (58%) Male 22(33.3%) 15 (68.2%) 37 (42%) Average Age 67 (45-90) 71 (58-86) 68 (45-90) Smoking Pack Years 47 48.6 47.4 Years Quit 5.5 10.2 6.64    Table 5.2 Alteration frequencies of the most frequently deregulated NSCLC miRNAs    AC SqCC All Samples miRNA Alteration Freq OE Freq UE Freq OE Freq UE Freq OE Freq UE hsa-mir-210 OE 100% 0% 95% 0% 99% 0% hsa-mir-96 OE 98% 0% 100% 0% 99% 0% hsa-mir-130b OE 95% 0% 100% 0% 97% 0% hsa-mir-183 OE 94% 0% 100% 0% 95% 0% hsa-mir-345 OE 94% 2% 95% 0% 94% 1% hsa-mir-877 OE 94% 2% 95% 0% 94% 1% hsa-mir-331 OE 94% 0% 91% 5% 93% 1% hsa-mir-182 OE 94% 0% 86% 0% 92% 0% hsa-mir-708 OE 89% 0% 100% 0% 92% 0% hsa-mir-141 OE 92% 0% 86% 0% 91% 0% hsa-mir-193b OE 89% 3% 95% 0% 91% 2% hsa-mir-301b OE 89% 0% 95% 0% 91% 0% hsa-mir-144 UE 0% 98% 0% 100% 0% 99% hsa-mir-30a UE 0% 97% 0% 100% 0% 98% hsa-mir-451a UE 0% 97% 0% 91% 0% 95% hsa-mir-143 UE 0% 97% 0% 82% 0% 93% hsa-mir-486 UE 0% 94% 0% 86% 0% 92% hsa-mir-101 UE 0% 89% 0% 95% 0% 91% 114  5.3.2  Lung AC and SqCC display different patterns of miRNA expression  AC and SqCC develop through distinct patterns of genomic alterations affecting unique protein coding genes [146]. Therefore, we next sought to investigate whether miRNAs display subtype specific patterns of expression that can discriminate subtypes. Unsupervised hierarchical clustering of tumor and non malignant tissue independently revealed similar results; one of the clusters contained primarily AC samples, while the other cluster contained a mix of both AC and SqCC (Figure 5.2A-B). The distribution of AC and SqCC was significantly different between clusters in both tumors (Fisher's Exact test, p=0.0063) and non-malignant tissue (p=0.0097) suggesting that histological subtypes influence miRNA expression in tumors and non-malignant lung tissue. Analysis of AC and SqCC non-malignant tissue revealed 17 differentially expressed miRNAs (8 OE and 9 UE, permutation test BH p<0.05 and fold change >2), several of which have been implicated in cancer and two (miR-107 and miR-429) in NSCLC [196, 197].   To identify miRNAs most likely involved in subtype specific tumorigenesis, we applied a series of statistical tests and fold change criteria to the 916 miRNAs with detectable expression. 47 miRNAs met all of our criteria and were deemed subtype specific; 12 OE and 1 UE in AC and 25 OE and 4 UE in SqCC (Table 5.3). The majority of subtype specific miRNAs were overexpressed (37 vs. 5) and SqCC specific (29 vs. 13, Figure 5.2C&D). The increased number of SqCC specific miRNAs relative to AC specific miRNAs mirrors the findings in Chapter 2, in which a greater number of SqCC specific alterations were identified. In addition, we identified 5 miRNAs (miR-203, -326, -375,-378a and -4662a) with bi-directional disruption (OE in one subtype and UE in the other or vice versa), suggesting that while these miRNAs are disrupted in both subtypes, they display subtype specific patterns of alteration. Area under the curve (AUC) analysis of individual miRNAs revealed 43/47 (91%) subtype specific miRNAs were able to distinguish between subtypes (p<0.05, Table 5.3), with 13 miRNAs having an AUC >0.8. Two miRNAs, miR-944 (OE in SqCC), and miR-375 (OE in AC and UE in SqCC), had AUCs over 0.9 (0.9669 and 0.9001 respectively) suggesting that these miRNAs may be the most robust in discriminating subtypes (Figure 5.2 E&F). Interestingly, 4 of the 5 miRNAs with the highest AUCs (miR-944, -30b, -152 and -4652) were all overexpressed in SqCC indicating that miRNAs overexpressed in SqCC may be the most useful for differentiating subtypes.  115  Figure 5.2  Figure 5.2. miRNAs are disrupted in a subtype specific manner that can discriminate between subtypes Hierarchical clustering of tumors (A) and non-malignant tissue (B) identifies two clusters, one comprised of primarily AC and the other of a mix of AC and SqCC. Venn diagram illustrating the differentially expressed miRNA in AC and SqCC relative to matched non-malignant tissue. Overexpressed miRNA are depicted in  (C) and underexpressed miRNA in (D). AUC curves of  miR-944 (E) and miR-375 (F), the two subtype specific miRNA most accurate at discriminating AC and SqCC tumors. 116  Table 5.3  AUC results of all BCCRC subtype specific miRNA  miRNA Alteration AUC pValue hsa-mir-1251 AC OE 0.6873 0.008792 hsa-mir-1287 AC OE 0.7989 <0.0001 hsa-mir-26b AC OE 0.7018 0.00477 hsa-mir-3189 AC OE 0.8695 <0.0001 hsa-mir-320b AC OE 0.7163 0.002492 hsa-mir-320d AC OE 0.8457 <0.0001 hsa-mir-4724 AC OE 0.7924 <0.0001 hsa-mir-4728 AC OE 0.8526 <0.0001 hsa-mir-489 AC OE 0.7225 0.001865 hsa-mir-491 AC OE 0.7211 0.001991 hsa-mir-551a AC OE 0.7255 0.001865 hsa-mir-5698 AC OE 0.8354 <0.0001 hsa-mir-204 AC UE 0.6784 0.0126 hsa-mir-326 AC OE/SqCC UE 0.6419 0.04719 hsa-mir-375 AC OE/SqCC UE 0.9001 <0.0001 hsa-mir-203 AC UE/SqCC OE 0.7287 0.001386 hsa-mir-378a AC UE/SqCC OE 0.7658 0.000202 hsa-mir-4662a AC UE/SqCC OE 0.8119 <0.0001 hsa-mir-105 SqCC OE 0.8233 <0.0001 hsa-mir-1227 SqCC OE 0.836 <0.0001 hsa-mir-1237 SqCC OE 0.7803 <0.0001 hsa-mir-1295a SqCC OE 0.6798 0.01193 hsa-mir-1343 SqCC OE 0.7107 0.003206 hsa-mir-152 SqCC OE 0.8691 <0.0001 hsa-mir-15b SqCC OE 0.7679 0.00018 hsa-mir-1910 SqCC OE 0.7321 0.001172 hsa-mir-3619 SqCC OE 0.6102 0.1232 hsa-mir-3682 SqCC OE 0.7417 0.000724 hsa-mir-378b SqCC OE 0.8281 <0.0001 hsa-mir-4488 SqCC OE 0.6674 0.01924 hsa-mir-4497 SqCC OE 0.6921 0.0072 hsa-mir-4652 SqCC OE 0.8561 <0.0001 hsa-mir-4713 SqCC OE 0.7314 0.001212 hsa-mir-4778 SqCC OE 0.6646 0.02131 hsa-mir-4787 SqCC OE 0.7094 0.003411 hsa-mir-651 SqCC OE 0.6784 0.0126 hsa-mir-663a SqCC OE 0.6977 0.005702 hsa-mir-767 SqCC OE 0.8058 <0.0001 hsa-mir-873 SqCC OE 0.7066 0.003858 hsa-mir-876 SqCC OE 0.7163 0.002492 hsa-mir-93 SqCC OE 0.6236 0.08374 hsa-mir-942 SqCC OE 0.5634 0.3754 hsa-mir-944 SqCC OE 0.9669 <0.0001 hsa-mir-135a SqCC UE 0.7693 0.000166 hsa-mir-181d SqCC UE 0.5909 0.2034 hsa-mir-30b SqCC UE 0.75 0.000473 hsa-mir-338 SqCC UE 0.6908 0.007626 117  5.3.3 Validation of differentially expressed miRNAs  Although a number of miRNA profiling studies have been performed in lung cancer, miRNA sequencing data is still quite limited. In order to validate the findings from our study and ensure we proceed with only the most robust targets, we downloaded miRNA sequencing data for 219 lung AC and 367 SqCC from the Cancer Genome Atlas (TCGA). We limited the validation cohort to only those tumors with patient matched non-malignant tissue and current or former smokers in order to replicate our original analysis; resulting in 33AC and 35 SqCC pairs for validation. Of our 85 NSCLC miRNAs, 76 were measurable in TCGA cohorts and 51% of these were validated as common to NSCLC. Of the 18 miRNAs that were altered in >90% of all NSCLC cases, 16 (89%) validated in the TCGA, with the other 2 (miR-301b, and -101) identified as subtype specific miRNAs in SqCC. Impressively, of the 16 miRNAs that validated as NSCLC miRNAs, all were altered in >70% of cases, and 3 (miR-210, -130, -183) were altered in >90%, suggesting this panel of miRNAs are robustly and recurrently altered in NSCLC (Table 5.4). Despite identifying a greater number of subtype specific miRNAs in the TCGA set (63 vs. 47), likely because of the increased SqCC cohort size, subtype specific miRNA validation in the TCGA dataset was surprisingly low, with only 27% of the 33 assessed miRNAs validating. All miRNAs, both subtype specific and NSCLC (n=48) that validated in the TCGA dataset are shown in Table 5.4.        118  Table 5.4 Subtype  specific and NSCLC miRNAs that validated in the TCGA cohort    Frequency of Alteration in BCCRC  Frequency of Alteration in TCGA miRNA Alteration  AC OE  AC UE  SqCC OE  SqCC UE  AC OE  AC UE  SqCC OE  SqCC UE miR-1287 AC OE 50.0% 1.5% 9.1% 9.1% 48.5% 0% 11.4% 22.9% miR-326 AC OE/SqCC UE 30.3% 12.1% 9.1% 50.0% 60.6% 6.1% 5.7% 65.7% miR-375 AC OE/ SqCC UE 65.2% 6.1% 13.6% 68.2% 69.7% 9.1% 8.6% 71.4% miR-1227 SqCC OE 22.7% 1.5% 59.1% 0% 9.1% 0% 42.9% 0% miR-1910 SqCC OE 15.2% 0% 54.5% 0% 0% 0% 42.9% 0% miR-651 SqCC OE 30.3% 15.2% 68.2% 0% 33.3% 0% 57.1% 11.4% miR-944 SqCC OE 31.8% 12.1% 95.5% 0% 39.4% 12.1% 85.7% 8.6% miR-1295a SqCC OE 12.1% 0% 40.9% 0% 9.1% 0% 45.7% 0% miR-338 SqCC UE 1.5% 66.7% 0% 100.0% 0% 54.5% 2.9% 85.7% let-7c NSCLC UE 1.5% 72.7% 0% 50.0% 0% 63.6% 2.9% 60.0% miR-1247 NSCLC UE 3.0% 77.3% 0% 59.1% 9.1% 63.6% 11.4% 51.4% miR-126 NSCLC UE 4.5% 50.0% 9.1% 40.9% 0% 51.5% 5.7% 68.6% miR-144 NSCLC UE 0.0% 98.5% 0% 100.0% 0% 87.9% 2.9% 80.0% miR-223 NSCLC UE 1.5% 74.2% 0% 50.0% 3.0% 36.4% 8.6% 42.9% miR-451a NSCLC UE 0% 97.0% 0% 90.9% 3.0% 81.8% 2.9% 85.7% miR-486 NSCLC UE 0% 93.9% 0% 86.4% 0% 81.8% 2.9% 85.7% miR-1180 NSCLC OE 40.9% 10.6% 27.3% 0% 51.5% 0% 51.4% 0% miR-1301 NSCLC OE 87.9% 0% 86.4% 0% 78.8% 0% 71.4% 0% miR-130b NSCLC OE 95.5% 0% 100.0% 0% 93.9% 0% 97.1% 0% miR-141 NSCLC OE 92.4% 0% 86.4% 0% 84.8% 0% 71.4% 2.9% miR-148a NSCLC OE 54.5% 1.5% 72.7% 0% 45.5% 0% 31.4% 8.6% miR-17 NSCLC OE 43.9% 0% 54.5% 0% 48.5% 0% 40.0% 0% miR-18a NSCLC OE 68.2% 0% 77.3% 0% 45.5% 6.1% 57.1% 2.9% miR-193b NSCLC OE 89.4% 3.0% 95.5% 0% 84.8% 0% 88.6% 2.9% miR-199a NSCLC OE 50.0% 3.0% 59.1% 0% 39.4% 0% 31.4% 8.6% miR-199b NSCLC OE 40.9% 3.0% 50.0% 9.1% 42.4% 0% 31.4% 5.7% miR-200c NSCLC OE 63.6% 0% 68.2% 0% 45.5% 0% 54.3% 0% miR-20a NSCLC OE 34.8% 3.0% 45.5% 0% 45.5% 6.1% 42.9% 0% miR-214 NSCLC OE 83.3% 3.0% 90.9% 4.5% 45.5% 6.1% 31.4% 11.4% miR-219 NSCLC OE 57.6% 6.1% 68.2% 0% 51.5% 3.0% 57.1% 0% miR-301a NSCLC OE 84.8% 0% 81.8% 0% 78.8% 0% 71.4% 0% miR-3127 NSCLC OE 59.1% 3.0% 59.1% 0% 81.8% 0% 71.4% 0% miR-324 NSCLC OE 80.3% 0% 81.8% 0% 72.7% 0% 57.1% 2.9% miR-335 NSCLC OE 31.8% 4.5% 40.9% 0% 33.3% 9.1% 40.0% 5.7% miR-3607 NSCLC OE 75.8% 1.5% 59.1% 9.1% 93.9% 0% 57.1% 8.6% miR-3677 NSCLC OE 45.5% 7.6% 54.5% 4.5% 69.7% 0% 71.4% 2.9% miR-4326 NSCLC OE 77.3% 0% 90.9% 0% 48.5% 0% 62.9% 0% miR-484 NSCLC OE 56.1% 1.5% 59.1% 0% 30.3% 3.0% 34.3% 5.7% miR-505 NSCLC OE 60.6% 1.5% 72.7% 0% 81.8% 0% 54.3% 2.9% miR-625 NSCLC OE 60.6% 4.5% 77.3% 0% 75.8% 0% 54.3% 2.9% miR-654 NSCLC OE 42.4% 19.7% 50.0% 9.1% 48.5% 6.1% 40.0% 5.7% miR-744 NSCLC OE 68.2% 0% 77.3% 0% 57.6% 0% 65.7% 0% miR-758 NSCLC OE 40.9% 15.2% 54.5% 4.5% 42.4% 3.0% 31.4% 8.6% miR-760 NSCLC OE 65.2% 0% 0% 77.3% 57.6% 0% 85.7% 0% miR-766 NSCLC OE 77.3% 1.5% 81.8% 0.0% 81.8% 0% 57.1% 5.7% miR-874 NSCLC OE 69.7% 3.0% 90.9% 4.5% 51.5% 0% 45.7% 8.6% miR-877 NSCLC OE 93.9% 1.5% 95.5% 0% 66.7% 0% 74.3% 0% miR-940 NSCLC OE 80.3% 3.0% 100.0% 0% 66.7% 0% 74.3% 0% 119  5.3.4 Copy number alterations contribute to aberrant miRNA expression  To determine whether differential expression of any of the validated miRNAs could be the result of recurrent DNA level alterations, we examined copy number status and DNA methylation levels of the 48 differentially expressed miRNAs. Due to the lack of miRNAs represented on the HM27K Illumina Methylation platform (only 110 miRNAs have probes), only 4 of our validated miRNAs (miR-17, -18a, 20a and -219) were measurable, and none were frequently hypo- or hypermethylated in any of the cases. 62/66 AC and 15/22 SqCC samples with miRNA sequencing data also had copy number profiles. A number of the validated miRNAs mapped to similar genomic locations, including 1q24.3, 12p13.31, 13q31.3, 14q32.31, 16p13 and 17q11.2, of which 1q, 14q32.31 and 17q11.2 are known to be recurrently amplified in NSCLC [37-39]. 9 miRNAs (miR-130b, -141, -148a, -301a, -484, -651, -940, -944 and -4326) had copy number alterations and concordant miRNA expression changes affecting >20% of cases, however, only miR-141 and miR-301a had a statistically significant association between copy number and expression (Mann Whitney U test, p<0.05). Together, our results suggest that while copy number alterations occur, they are not the driving force of over- or underexpression for the majority of the NSCLC and subtype specific miRNAs we identified.  5.3.5 Predicted targets of subtype miRNAs are associated with distinct pathways  In an attempt to elucidate the signaling pathways and biological processes disrupted by our validated subtype specific miRNAs, we identified their predicted mRNA targets using miRDIP with stringent filtering criteria (prediction by 6 algorithms). The 5 OE SqCC miRNAs were predicted to target 849 unique genes, including known tumor suppressors and oncogenes HIF1a, EYA4, BCL2 and IGF1R. Visualization of miRNA-mRNA interactions in NAViGaTOR showed a number of mRNAs to be targets of multiple miRNAs, with 14 predicted to be targets of 4 SqCC miRNAs (Figure 5.3). Pathway analysis of all SqCC predicted targets revealed enrichment of target genes in the epithelial to mesenchymal transition, molecular mechanisms of cancer, TGFB, PTEN and VEGF signaling pathways as well as post-translational modification, cell morphology and cellular movement and DNA replication, recombination and repair functions. The networks of these functions were centered around CUL3, JNK and histone proteins, respectively. 120   Interestingly, miRNA-mRNA interactions of the two validated, bi-directionally altered miRNAs miR-326 and miR-375, revealed a large number (37/112) and significant enrichment of genes involved in cellular fate and organization (Fisher's Exact Test, p=2.2x10-12, Figure 5.4). Ingenuity Pathway Analysis of the 112 target genes further supported this finding, with tissue development, cell morphology, assembly and organization and cellular function and maintenance ranking as the top functions of these genes. These findings suggest that miR-326 and miR-375 have differing roles within AC and SqCC and may play a role in maintaining cell function and lineage, providing insight into the reason for their differential disruption.   5.3.6 Lung cancer prognostic genes are targeted by NSCLC miRNAs  To assess the prognostic implications of the NSCLC miRNAs altered in >90% of cases, we aligned the list of predicted mRNA targets for each miRNA from miRDIP to a curated list of 1,066 lung cancer prognostic genes [198]. Of the 1,066 prognostic genes, 42 (4%) were predicted targets of the 16 validated miRNAs (Figure 5.5). The majority of miRNAs were highly connected to multiple prognostic genes. The RNA bindings protein QKI which regulates pre-mRNA splicing and mRNA stability was the most targeted mRNA, predicted to be a target of 5 unique miRNAs, highlighting the potential importance of this gene to NSCLC (Figure 5.6).  Conversely, miR-141 and miR-182, both of which are frequently overexpressed were connected to the most prognostic genes (n=9). This data emphasizes the biological and prognostic relevance of these NSCLC miRNAs and further supports the potential of these miRNAs as biologically based biomarkers for lung cancer detection. 121  Figure 5.3  Figure 5.3  miRNA-mRNA interaction network of overexpressed SqCC specific miRNA122  Figure 5.3 miRNA-mRNA interaction networks of overexpressed SqCC specific miRNAs Predicted mRNA targets of overexpressed SqCC specific miRNAs were identified using miRDIP and miRNA-mRNA interactions visualized in NAViGaTOR. Grey lines indicate miRNA-mRNA interactions and predicted targets are depicted as circular nodes with colouring corresponding to Gene Ontology terms associated with gene function. SqCC specific miRNA were predicted to target 849 unique genes, but only those genes targeted by multiple miRNAs are displayed.    Figure 5.4  Figure 5.4. Bi-directional miRNAs target genes involved in cellular fate and organization miRNA-mRNA interactions of the bi-directionally altered miRNA miR-326 and miR-375. Target genes are coloured based on their Gene Ontology functions and were significantly enriched for genes involved in cellular fate and organization (Fisher's test p<0.0001).123  Figure 5.5  Figure 5.5 Predicted interaction between NSCLC miRNAs and lung cancer prognostic genes miRNAs frequently disrupted in NSCLC were input into miRDIP to identify predicted targets. The network of miRNA-mRNA interactions was visualized in NAViGaTOR, but restricted to those targets with known prognostic significance in lung cancer (n=1066). In total, the network is comprised of 13 miRNAs and 42 prognostic target genes with most miRNAs well connected to prognostic genes. miR-141 and miR-182 are the most highly connected to prognostic gene targets. Connections between miRNAs and mRNAs are illustrated by grey lines. Predicted targets are depicted as elliptical nodes with colouring corresponding to Gene Ontology terms associated with gene function. Vertical mRNA nodes represent genes frequently upregulated in the signature while wide nodes represent downregulated genes. Fuzzy green clusters depict all other predicted mRNA targets not associated with prognostic genes. 124  5.3.7 MiR-944 overexpression is highly specific to SqCC  Of the miRNAs regulated by copy number, miR-944 was the most frequently altered; gained and overexpressed in over 50% of SqCC cases and overexpressed without any known DNA alteration in another 40% of cases. miR-944 is lowly expressed in both AC and SqCC non-malignant tissue (1.35 ± 0.23 RPKM in AC and 1.42 ± 0.34 RPKM in SqCC, Student's t test p=0.87), with no significant difference in expression between AC tumors and non-malignant tissue (7.7 ± 3.99 RPKM, Student's t- test, p=0.11, Figure 5.6A). However, in SqCC tumors, miR-944 expression is dramatically increased (258.8 ± 75.13 RPKM,) relative to both AC tumors (Student's t-test p=0.0014) and SqCC non-malignant tissue (Student's t-test p<0.001, Figure 5.6A), with an average fold change of 172.8. A similar pattern of expression was seen in samples from the TCGA (Figure 5.6B). Given this dramatic difference in expression, it is not surprising miR-944 was the best miRNA to accurately distinguish between subtypes (BCCRC: AUC =0.967, p<0.0001, Figure 5.2E, TCGA: AUC=0.858, p<0.001, data not shown).    While miR-944 may be a useful marker for subtype discrimination given its subtype specificity and frequent overexpression in SqCC, we were curious as to the target genes and pathways predicted to be disrupted by this miRNA, and the contribution of miR-944 to SqCC tumorigenesis. miR-944 was predicted by miRDIP to target 305 unique mRNAs, including the tumor suppressors APC, EYA4 and NF1. Pathway analysis of miR-944 predicted targets revealed multiple networks associated with cellular development and cell death and survival, that were centered around ERK signaling, although ERK itself was not a predicted target (Figure 5.6C). miR-944 targets were also significantly enriched for the immune related functions of NF-κB and TGFβ signaling. The high frequency of miR-944 overexpression and the predicted disruption of key signaling pathways suggests that miR-944 may play a pivotal role SqCC tumorigenesis    125  Figure 5.6  Figure 5.6 miR-944 overexpression is specific to SqCC and targets ERK signaling. miR-944 expression in AC and SqCC tumors and non-malignant tissue in the BCCRC (A) and TCGA (B) cohorts. Blue bars represent non-malignant tissue and red bars tumors. Error bars depict the standard error of the mean. (C) ERK signaling components are downregulated (shown in green) by the predicted targets of miR-944. Solid lines represent direct interactions while dotted lines represent indirect gene interactions. Corkscrews represent enzymes, three pronged shapes are kinases, thimble shapes are transporters, cylinders are ion channels and circular molecules encompass all other gene products.    A.B.1101001000AC SqCCAverage RPKM (log10)1101001000AC SqCCAverage RPKM (log10)C.126  5.3.8 MiR-1287 inhibits expression of RAD9A by binding its 3'-UTR  miR-1287 which was overexpressed in 50% of AC, was the only AC specific miRNA to validate in the TCGA. Target prediction of miR-1287 identified 5 predicted targets, 3 of which (RAD9A, TEAD3 and SLC8A1) were targeted exclusively in AC. Both RAD9A, a component of the 9-1-1 cell cycle checkpoint complex which plays a central role in DNA damage repair and cell cycle arrest, and TEAD3, a transcription factor involved in the Hippo signaling pathway have been implicated in tumorigenesis. Transfection of miR-1287 mimic oligonucleotides significantly reduced RAD9A but not TEAD3 protein levels compared to non-targeting control oligonucleotides (Figure 5.7A). To determine whether RAD9A is a direct target of miR-1287, a RAD9A 3'UTR dual luciferase reporter construct was co-transfected into cell lines with miR-1287 mimic or control oligonucleotides and luciferase activity measured 48 hours post-transfection. Overexpression of miR-1287 significantly reduced luciferase activity relative to the control olignonucleotides, confirming miR-1287 directly inhibits RAD9A (Figure 5.7B). Taken together, these results strongly suggest that miR-1287 post-transcriptionally inhibits RAD9A by directly binding the 3'-UTR of RAD9A mRNA and preventing translation.   5.3.9 Overexpression of miR-1287 enables proliferation of damaged cells in LCCL  To investigate whether overexpression of miR-1287 is biologically relevant to the development of a malignant phenotype, we overexpressed miR-1287 in HBEC, A549 and H1993 cell lines (Figure 5.7C). In HBECs, treatment of transfected cells with 10uM of cisplatin for four hours reduced proliferation and apoptosis, suggesting these cells may be undergoing senescence (Figure 5.7D and E). Conversely, overexpression in LCCLs had the opposite effect, inducing apoptosis and increasing proliferation relative to controls (Figure 5.7D and E). A greater proportion of miR-1287 OE cells were positive for both phospho ɣH2AX and BrdU in LCCLs, suggesting that in these lines, miR-1287 inhibits DNA damage repair through suppression of RAD9A, enabling proliferation of damaged cells. Induction of phenotypic changes in non-malignant HBECs and AC cell lines suggests that overexpression of miR-1287 is associated with malignant changes.    127  Figure 5.7    Figure 5. 7. miR-1287 induces a malignant phenotype through the suppression of RAD9A (A) Western blot analysis of TEAD3 and RAD9A following treatment with 20nM control or miR-1287 oligonucleotides. (B) Luciferase assays of pEZX-RAD9A-3'UTR reporter in indicated cells transfected with control olignonucleotides (black) or miR-1287 oligonucleotides (grey). (C) Levels of miR-1287 expression following transfection measured by qRT-PCR. Representative images of the effect of miR-1287 overexpression on apoptosis (D) and DNA damage repair (pH2AX) and proliferation (BrdU incorporation) following treatment with cisplatin. (E) Error bars represent standard error of the mean of triplicate experiments, *p<0.05, ***p<0.0001.   128  5.4 Discussion  Within the last decade, ncRNAs especially miRNAs, have emerged as key players in tumorigenesis due to their involvement in numerous biological processes. While several studies have shown that miRNA signatures can distinguish AC and SqCC, the majority of these studies focus only on miRNAs that are differentially expressed between tumor tissues, with little use of non-malignant tissue as a baseline [199-201]. A lack of investigation into the biological roles of these differentially altered miRNAs has resulted in a limited understanding of how miRNAs contribute to subtype tumorigenesis. In this study, we performed miRNA sequencing on a cohort of AC and SqCC tumors with patient matched non-malignant tissue in an attempt to generate the first unbiased and comprehensive analysis of miRNA deregulation in NSCLC subtypes and to identify subtype specific miRNAs that may contribute to the distinct clinical phenotypes of AC and SqCC. As smoking is known to lead to different patterns of genomic alterations and has been shown to influence miRNA expression patterns (Vucic et al., Unpublished), we limited our analysis to current and former smokers in order to eliminate the confounding effects of smoking.    Validation of profiling results is essential to ensure robust findings and eliminate potential artifacts. The TCGA is the largest public repository of lung AC and SqCC miRNA expression data, and the only one generated by sequencing. Despite hundreds of tumor profiles, fewer than 80 patient matched non-malignant tissues have profiles, significantly limiting the number of available cases for validation. Nevertheless, we used this cohort to validate the 85 NSCLC and 47 subtype specific miRNAs we identified. The overall validation rate of our subtype specific and NSCLC miRNAs was surprisingly low (48/132, 36%). Analysis of the TCGA data revealed lower detection sensitivity (Vucic et al., unpublished), which could partially explain the poor validation rates, especially in AC where heterogeneity is known to be a major factor. Discrepancies in collection of non-malignant tissue and microdissection between our cohort and the TCGA's may also contribute to this disparity. Despite a low validation rate, we feel confident that the miRNAs that did validate represent the most robust candidates for further examination. As additional sequencing data emerges, validation in other cohorts will provide further insight into the most recurrently altered and differentially expressed miRNAs between subtypes. 129    The identification of miRNAs that were aberrantly expressed in greater than 90% of cases suggests that these miRNAs are important in lung tumor biology, regardless of subtype. The high validation rate of these miRNAs in the TCGA (16/18), and the fact that several of these miRNAs have been previously implicated in tumorigenesis, confirms the importance of these miRNAs to lung tumor biology. Notably, miR-126, -130b, -144, -451a and -486 were identified in meta-analyses of miRNA profiling studies comparing tumor tissue to non-malignant tissue, further validating the prevalence of their disruption in NSCLC [202, 203]. These findings in conjunction with the observation that a number of lung cancer prognostic genes were predicted targets of these aberrantly expressed miRNAs (Figure 5.6), provides additional support to the relevance of these miRNAs in tumor biology and suggests they may be biologically relevant biomarkers. As miRNAs are known to circulate throughout the body with substantial stability, analysis of these miRNAs in blood samples and large sample cohorts to determine their potential as prognostic and/or diagnostic blood based biomarkers of lung cancer is warranted.   Hierarchical clustering of tumor and non-malignant tissue miRNA sequencing profiles revealed miRNA expression is associated with histology, in agreement with previous studies [199, 201, 204, 205]. Not surprisingly, it also suggests that heterogeneity amongst subtypes is present. The identification of subtype specific miRNAs demonstrates that like protein coding genes, disruption of distinct miRNAs underlie tumorigenesis of AC and SqCC. Despite a larger AC cohort, the majority of subtype specific miRNAs identified were SqCC specific, mimicking our findings in Chapter 2. The greater prevalence of overexpressed subtype specific miRNAs, suggests that most deregulated miRNAs function to suppress tumor suppressor genes. Of the 9 validated subtype specific miRNAs, 4 (miR-326 OE in AC and UE in SqCC, -338 UE in SqCC, -375 OE in AC and UE in SqCC and -944 OE in SqCC) have been previously identified as subtype specific, and all in the same subtype and same direction as we observed [204-207].    miR-375 is the only subtype specific miRNA to be implicated in tumorigenesis, let alone lung cancer. Loss of miR-375 expression has been observed in multiple cancer types and is associated with suppression of cancer hallmarks through the targeting of oncogenes including IFG1R, PDK1 and JAK2 [207]. While underexpression of miR-375 is most common, in breast 130  and prostate cancer, miR-375 is up-regulated relative to normal tissues and alters proliferation [208]. Interestingly, studies of esophageal cancer observed a similar pattern of subtype specific expression, with AC tumors having a 6-fold higher expression than SqCC tumors [209], however stable introduction of miR-375 into A549 lung AC cells failed to induce a significant inhibition in growth [210]. Taken together, these findings suggest that miR-375 functions as both a tumor suppressive and oncogenic miRNA and that its function is likely context dependent, supporting our observation of its bi-directional alteration in NSCLC subtypes.   To date, miRNA profiling studies have identified numerous miRNAs differentially expressed between AC and SqCC, and correlated these with clinical features such as survival and prognosis. However, none of these studies have assessed the functional consequences of this disruption. To provide insights into the differential oncogenic mechanisms disrupted by subtype specific miRNAs, we performed network and pathway analysis on miRDIP predicted targets of subtype specific miRNAs. The top SqCC gene networks were associated with post-translational modification and DNA replication, recombination and repair, corroborating the findings in Chapter 2 and further supporting the relevance of disruption to histone modifying enzymes in SqCC. In Chapter 3, analysis of the KEAP1-CUL3 E3 ubiquitin ligase complex revealed this complex to be differently altered between subtypes, with loss of CUL3 occurring preferentially in SqCC [151]. Interestingly, the most significant SqCC network disrupted by SqCC specific miRNAs was centered around multiple CUL proteins, specifically CUL2, CUL3 and CUL5. All three of these CUL proteins are components of different E3 ubiquitin ligase complexes that ubiquitinate key tumor suppressors and oncogenes including JAK2, TP53, HIF1a and NRF2 [115, 117, 211-213]. As we demonstrated, loss of the Cullin component disrupts the function of the complex, leading to disruption of substrate regulation and promotion of tumorigenesis [151]. Mutation and copy loss of CUL3 in SqCC has been shown by us and others, however, our findings suggest that multiple cullin proteins may be underexpressed in SqCC due to aberrant miRNA expression. Validation of these proteins as direct targets of the predicted miRNA will be essential in further establishing the role of cullin proteins in SqCC tumorignesis.  131   miR-944 was by far the most frequently deregulated SqCC specific miRNA we  identified, overexpressed in 96% and 86% of the BCCRC and TCGA cohorts, respectively.  Examination into the genomic coordinates of miR-944 revealed it to be located at 3q28. Amplification of 3q in SqCC is one of the most well characterized and consistent copy number differences between AC and SqCC [41], and offers an explanation into the frequent copy gain, overexpression and subtype specificity of miR-944. Interestingly, the miR-944 transcript is located within the TP63 transcript, a widely used immunohistochemical marker to define lung SqCC. This overlap with TP63 may explain why miR-944 was so accurate in discriminating subtypes and supports the notion that miR-944 could be a useful diagnostic biomarker for SqCC. The identification of disrupted ERK signaling by miR-944 predicted targets highlights a potential mechanism through which miR-944 may contribute to SqCC tumorigenesis. Given the high frequency of miR-944 disruption in SqCC, investigation into its biological relevance may provide novel insight into mechanisms of SqCC pathogenesis.   With only one AC specific miRNA validating, pathway analysis would be uninformative, especially given that we identified only 5 predicted targets. As two of the predicted targets have been implicated in tumorigenesis but little is known about miR-1287, we explored the role of miR-1287 in the pathogenesis of lung AC. We have shown a tumor promoting role of miR-1287, with overexpression in LCCLs promoting the survival and proliferation of damaged, genetically unstable cells. Conversely, overexpression in HBECs resulted in a reduction of apoptosis and proliferation, suggesting that non-malignant cells may induce a senescent state as a tumor suppressive mechanism, however this needs to be further explored. Moreover, we validated RAD9A as a target of miR-1287. RAD9A is a cell cycle checkpoint protein, and together with RAD1 and HUS1 forms the 9-1-1 complex, which plays a central role in activation of DNA damage induced checkpoints [214, 215]. In response to DNA damage, RAD17 loads the 9-1-1 complex around DNA lesions, facilitating phosphorylation and activation of CHK1 kinase- the key event that determines cell survival following genotoxic stress [214, 215]. RAD9A is the key component of the 9-1-1 complex due to the presence of a nuclear localization sequence [215]. Single nucleotide polymorphisms within RAD9A are associated with an increased susceptibility for lung AC, further supporting a role for this gene in lung AC [216]. The identification of a 132  malignant phenotype associated with miR-1287 overexpression confirms that our analysis approach accurately identifies miRNAs involved in subtype tumor biology. Importantly, our findings highlight the contribution of miRNA dysregulation to lung cancer biology and suggests that analysis of the other subtype specific miRNAs, the functions of which are largely unknown, will provide further insight into the molecular mechanisms underlying AC and SqCC.    While our analysis identified numerous subtype specific and NSCLC miRNAs, we recognize that the stringent criteria we used may have excluded true subtype specific miRNAs. For example, miR-205 and miR-21 which have been reported to be overexpressed in SqCC and NSCLC respectively in numerous studies [60, 188, 202], failed to meet our criteria. Despite having significantly different expression between subtypes, miR-205 was not differentially altered and therefore was not considered subtype specific. Similarly miR-21, despite being frequently overexpressed in both subtypes (68% in AC and 77% in SqCC), was found to have significantly different expression between subtypes. The goal of this work was to identify miRNAs that were either significantly different between subtypes, or extremely similar. Therefore, although our list of miRNAs may not be comprehensive, we have shown that our criteria identifies biologically relevant miRNAs that may have use as biomarkers for the detection of lung cancer.    In conclusion, our study provides the first comprehensive sequencing analysis of miRNA expression in lung AC and SqCC. Collectively, our study confirms that similar to protein coding genes, miRNA deregulation occurs in both a subtype specific and non specific manner and that these disruption patterns can reveal novel insight into the biology and molecular mechanisms underlying tumorigenesis and contributing to their disparate clinical phenotypes. We have identified ubiquitously deregulated miRNAs that could be promising biomarkers for the detection of lung cancer as well as subtype specific miRNAs that accurately discriminate subtypes. For the first time, we illustrate the oncogenic pathways and functions associated with subtype specific miRNA deregulation, and implicate miR-1287 and its target RAD9A in AC tumorigenesis. Our findings underscore the importance of miRNAs to lung cancer biology and highlight the need to look beyond protein coding gene alterations.  133  Chapter 6: Conclusion  6.1 Summary   Lung cancer is the leading cause of cancer related mortality, accounting for more deaths than breast, prostate and colon cancer combined [1, 2]. The five year survival rate of lung cancer has failed to improve significantly over the past decades, from 14% in 1994 to 18% in 2008, emphasizing the need for novel diagnostic and therapeutic strategies [2]. Genomic analyses of tumor genomes, especially recent characterization using high throughput sequencing technologies have identified numerous driver alterations contributing to tumorigenesis. However, driver alterations for up to 50% of cases remain to be elucidated, highlighting the need to improve our understanding of the molecular mechanisms underlying lung tumorigenesis.   Histological subtypes of lung cancer display disparate clinical phenotypes and responses to therapy, suggesting they are distinct diseases that arise through divergent molecular mechanisms. The goal of this work was to characterize the landscape of molecular alterations in AC and SqCC through the integration of multiple 'omics dimensions, and elucidate the mechanisms through which frequently deregulated genes and signaling pathways common to NSCLC or altered in a subtype specific manner contribute to tumorigenesis. Our analysis revealed several molecular alterations specific to a given subtype. Importantly, the findings from this work provide insights into novel potential therapeutic strategies for SqCC and identified potential diagnostic biomarkers for lung cancer treatment and prognosis.  6.1.1 Identify molecular alterations common or specific to lung cancer subtypes.  NSCLC is a heterogeneous disease comprised of regionally distinct and phenotypically diverse tumors. Fundamental differences in tumor biology are thought to underlie the phenotypic differences of these tumors, and although previous studies have identified distinct patterns of alterations, the specific genes and signaling pathways responsible for the disparate clinical phenotypes are only just beginning to be identified. Therefore in Chapter 2, we integrated genome wide copy number and methylation with gene expression profiles for a cohort of 169 AC and 92 SqCC tumors to characterize the landscape of molecular alterations with concordant 134  expression changes specific to each subtype as these are most likely to be causal in tumorigenesis. Our analysis revealed 294 regions of copy number disparity of which 205 were specific to SqCC and 89 were AC specific as well as 2384 differentially methylated genes. After excluding genes without differential expression between tumors and non-malignant tissue, we identified 778 genetically or epigenetically deregulated subtype specific genes. From these genes we identified key oncogenic pathways disrupted in each subtype that likely serve as the basis for their differential biology and clinical outcomes. Downregulation of HNF4α target genes was the most common pathway specific to AC while SqCC tumors demonstrated disruption of multiple histone modifying enzymes and E2F1. In addition, several genes previously implicated in NSCLC such as ERCC1 and KEAP1 were found to display subtype specific patterns of alteration.    Within the past decade, miRNAs have emerged as critical regulators of gene expression, implicated in normal biological processes as well as tumorigenesis [56]. To date, miRNA profiling studies of  histological subtypes have been exclusively array based, limiting the number of miRNAs for interrogation. In Chapter 5 we performed miRNA sequencing on 66 AC and 22 SqCC tumors with matched non-malignant tissue. We identified miRNAs recurrently altered in all cases regardless of subtype as well as 48 subtype specific miRNAs; 29 of which were SqCC specific, 13 that were AC specific and 5 that were disrupted in both subtypes, but in opposing directions. Similar to protein coding genes, subtype specific miRNAs were associated with distinct pathways and functions in each subtype and their expression accurately discriminates subtypes. Together these profiling studies revealed novel subtype specific patterns of alterations and provide insight into the specific genes that contribute to subtype tumorigenesis, further demonstrating that AC and SqCC develop through distinct pathways.   6.1.2 Delineate the biological significance and clinical relevance of alterations.  Despite the existence of numerous profiling studies examining the pattern of alterations defining AC and SqCC, few studies have investigated the function and molecular mechanisms through which these subtype specific alterations contribute to subtype development and tumorigenesis. In chapters 3, 4 and 5 we assessed the clinical relevance of subtype specific 135  alterations by manipulating candidate genes identified by integrative analysis or miRNA sequencing in vitro and in vivo to delineate the biological significance of their disruption and elucidate the mechanism by which they contribute to disease pathogenesis.    Specifically, we show that the complex components of the KEAP1-CUL3 E3 ubiquitin ligase as well as its substrate IKBKB are differentially disrupted in AC and SqCC, with KEAP1 loss characterizing AC tumors and CUL3 loss and IKBKB gain being preferential in SqCC. Importantly, disruption of a single complex component was sufficient to impair function, leading to the accumulation and aberrant activation of NF-κB signaling. In chapter 4 we identified 12q13-15 and more specifically YEATS4 as amplified and overexpressed in >20% of NSCLCs. Overexpression of YEATS4 abrogated senescence while attenuation of YEATS4 in lung cancer cell lines reduced proliferation, colony formation and tumor growth and induced cellular senescence. Furthermore we demonstrate that YEATS4 acts as a negative regulator of the p21-p53 pathway by inhibiting p21 and that YEATS4 expression affected the cellular response to cisplatin with increased levels associated with resistance and lower levels with sensitivity. Finally in chapter 5, we demonstrate the ability of our analysis approach to identify biologically relevant miRNAs by showing the effect of miR-1287 overexpression in normal immortalized bronchial epithelial cells and lung cancer cell lines. In addition to the identification of subtype specific miRNAs, we identified a panel of 16 miRNAs recurrently aberrantly expressed in tumors (>90%) regardless subtype and suggest that these miRNAs may be useful biomarkers for the detection of lung cancer.  6.2 Conclusions  We hypothesized that 1) AC and SqCC arise through distinct molecular mechanisms, which can be identified through integrative analysis of AC and SqCC genomes and 2) that these different patterns of genetic and epigenetic alterations underlie unique biological mechanisms that contribute to subtype development, phenotypes and response to therapy. Our characterization of AC and SqCC genomes using an integrative 'omics approach revealed a number of subtype specific alterations with concordant expression changes as well as differentially expressed miRNAs. These differences were found to affect distinct oncogenic 136  pathways and accurately separate AC and SqCC samples, providing further evidence that these subtype specific genes are responsible for driving the differential development of AC and SqCC. These findings, support our hypotheses and are consistent with previous reports, but importantly they provide new insight into specific genes underlying subtype tumorigenesis. Functional analysis of frequently deregulated subtype specific and non-specific genes revealed the phenotypes associated with their disruption, and the molecular mechanisms through which they promote tumor development and progression. Through the characterization of these mechanisms, we were able to identify drugs whose sensitivity was altered by gene disruption, highlighting the clinical relevance of these genes. In addition to subtype specific genes, genes and miRNAs frequently disrupted across both subtypes were identified. This suggest that certain biological functions or cellular processes are essential to all lung tumors, regardless of subtype. Collectively, the findings from this thesis confirm that while certain molecular alterations are shared amongst histological subtypes, distinct alterations underlie their differential development, and clinical phenotypes.  6.3 Significance and clinical implications  The success of EGFR TKIs in EGFR mutant lung AC, and the profound benefit of targeted therapies such as trastuzumab and imatinib in breast cancer and chronic myeloid leukemia, respectively demonstrated the potential survival benefit of targeted therapies and launched the search for additional actionable alterations in lung cancer. As a result, the landscape of genomic alterations defining AC and SqCC, has been extensively studied [15, 17, 18, 33-36, 39, 51, 55, 153]. While a subset of the genes involved in lung cancer have been well characterized, including NKX2-1, PTEN, CDKN1A, PIK3CA, SOX2, KRAS and EGFR, the majority of genes responsible for tumorigenesis remain unknown. This is exemplified by the findings that roughly 50% of NSCLC tumors harbour no known driver gene/targetable alteration [142]. Moreover, the majority of studies describing the patterns of alterations in NSCLC subtypes focus primarily on gene identification, with little emphasis on elucidating the pathways and molecular mechanisms underlying subtype development. Therefore one of the most important findings from this thesis is the identification of novel subtype specific genes and uncovering the molecular mechanisms through which these genes are involved in tumorigenesis. 137   Diverse patterns of genomic alterations suggests that AC and SqCC are distinct diseases.  These findings have had significant impact on the clinical management on NSCLC, with the one size fits all approach no longer considered practical. Targeted therapies significantly improve survival of patients harbouring the specific alteration compared to standard chemotherapy; however patients will inevitably recur, such that targeted therapies are administered without curative intent. The development of novel, effective treatment approaches is desperately needed, especially for SqCC in which there are currently no targeted therapies approved for clinical use. This thesis identified several patterns of gene disruption associated with treatment sensitivity, the most clinically relevant of which was the finding that SqCC tumors/cell lines are more sensitive to HDAC inhibition than AC, offering a novel treatment strategy for this subtype with few effective treatment options. PI3K inhibitors were also identified as potentially effective in SqCC and the TCGAs characterization of SqCC tumors found the PI3K pathway was altered in 47% of cases [55], further supporting the role of this pathway in SqCC tumorigenesis. While the dependence of SqCC tumors on these pathways and specific alterations has yet to be functionally defined, our analysis has identified new therapeutic avenues worth exploring. Overall, this thesis illustrates the molecular differences that underlie AC and SqCC and highlights the importance of not only stratifying patients based on histology but also underlying tumor biology.  6.4 Future directions  The findings of this work have generated several additional research avenues worth pursuing: i) understanding the mechanism of YEATS4 mediated tumorigenesis ii) discerning the contribution of miR-944 to SqCC tumorigenesis , iii) validation of the most frequently deregulated miRNAs as biomarkers for the early detection of lung cancer, iv) analysis of histone modifications in SqCC and v) integration of genomics data with features of the tumor microenvironment. Future work investigating these issues will be essential to further define the mechanisms underlying lung tumorigenesis and assess the potential of candidates as robust biomarkers for diagnosis or novel therapeutic targets.  138  6.4.1 Mechanisms of YEATS4 mediated tumorigenesis     Our analysis of YEATS4 revealed it contributes to tumorigenesis by altering numerous cellular processes. While we show that YEATS4 negatively regulates the p53-p21 pathway, in p53 mutant cell lines, knockdown of CDKN1A had no effect on viability, senescence or anchorage independent growth, and cisplatin sensitivity could not be associated with differences in p21 or p53, suggesting that these phenotypes are mediated through other pathways. Analysis of differentially expressed genes following knockdown identified the targets of MYC and SRF transcription factors to be significantly enriched. These findings provide a logical starting point to further investigate the mechanisms of YEATS4 mediated tumorigenesis. Elucidation of these mechanisms will be important in determining the clinical relevance of YEATS4 amplification in NSCLC.  6.4.2 Validation of miRNAs as biomarkers for the detection of lung cancer  NSCLC patient survival improves dramatically to over 70% when tumors are detected at the earliest localized stage, emphasizing that the greatest survival benefits for lung cancer will be achieved through early detection [1]. Due to the time consuming and invasive nature of current early detection techniques such as bronchoscopy and low-dose computed tomography, as well as a lack of criteria (other than smoking) for identifying individuals at risk of lung cancer, early detection strategies are not widely available to the general public. Non-invasive diagnostic biomarkers therefore hold immense promise in improving lung cancer survival. miRNAs circulate in the blood with substantial stability, making them attractive targets for biomarkers. The identification of 16 miRNAs that were frequently disrupted in our NSCLC cases, validated in the TCGA and previously reported by others, suggests they may have potential as biomarkers. There are a number of criteria for robust, highly sensitive and specific biomarkers; however for this panel of miRNAs, the first step should involve analysis in blood samples to determine whether they are detectable, and the concordance between tumor and blood expression. Successful candidates should then be assessed in blood samples from patients with and without cancer as well as those from patients with benign lung disease to assess the ability of these miRNAs to discriminate between benign and malignant disease. Finally, validation in multiple 139  large independent cohorts will be essential for determining the clinical feasibility of remaining candidates.  6.4.3 Contribution of miR-944 to SqCC tumorigenesis  miR-944 expression was frequently deregulated in SqCC (2 fold overexpression in >95%) and highly specific for SqCC tumors, accurately separating AC from SqCC with significant accuracy in both our datasets and the TCGA (AUC values of 0.9669 and 0.858, p<0.0001, respectively). Predicted targets of miR-944 include known tumor suppressors EYA4 and NF1. Together, these findings provide strong evidence supporting a role for miR-944 in SqCC tumorigenesis. In order to determine the biological relevance of miR-944 overexpression, its function and mechanism of action should be characterized in lung cancer cell lines and immortalized human bronchial epithelial cells by ectopic expression. There are currently no targeted therapies approved for clinical use in SqCC. The ubiquitous disruption of miR-944 could therefore have significant clinical implications if it is found that miR-944 does in fact contribute to SqCC development.   6.4.4 Global assessment of histone modifications in SqCC  Although this thesis integrated multiple 'omics dimensions (copy number, DNA methylation, gene and miRNA expression), several other mechanisms of gene disruption exist. These include sequence mutations (point mutations, insertions and deletions) which can alter gene function, gene fusions which create new oncogenic proteins, histone modifications which affect chromatin structure and accessibility and ncRNAs (long non-coding, antisense RNAs) which modulate gene expression. Post-translational modification of histone tails by acetylation and methylation play crucial roles in regulating chromatin structure and gene expression [104]. The polycomb repressive complex 2 (PRC2) is a histone methyltransferase comprised of numerous polycomb proteins that methylates histone H3 on Lysine 27 (H3K27me, me2 or me3), a mark that transcriptionally silences chromatin [217]. Components of this complex including EZH2- the catalytic subunit, were found to be frequently deregulated in SqCC likely leading to aberrant patterns of gene expression. High throughput approaches for the analysis of histone modifications are widely available, and a large public repository of histone modification data is 140  available through ENCODE [218]. Given the prominent disruption of histone modifying enzymes in SqCC, interrogation of the polycomb group proteins and histone modifications throughout the genome would likely reveal novel insights into SqCC pathogenesis.   6.4.5 Integration of molecular and immunologic alterations to understand the contribution of the tumor microenvironment to tumorigenesis.  The immune system plays a key role in the early elimination of tumors, and its importance in tumorigenesis is becoming increasingly appreciated, with immune evasion now considered a hallmark of cancer [219]. The tumor microenvironment is a complex milieu in which cancer cells, stromal cells and tumor infiltrating leukocytes interact. Targeting immune cells within the microenvironment is a growing field, and immunotherapies for the treatment of lung cancers have shown clinical activity in NSCLC suggesting this immune modulation may be an effective treatment in lung cancer [220-222]. However, preliminary data show that only a subset of patients respond to PD-1 blockade. While EGFR mutations were recently found to activate the PD-1 pathway contributing to the immune escape of these tumors [223], it is largely unknown whether specific genomic subsets of lung tumors activate this or other immune pathways and may confer sensitivity to immunotherapies. The identification of immune related signaling pathways disrupted by subtype specific alterations suggests that the immune cell compartments and activation/suppression of these cells within the tumor microenvironement may differ between subtypes. Integration of genomics data with tumor microenvironment profiles will provide insight into how genomic alterations influence the immune compartment and may reveal novel therapeutic strategies for NSCLC.        141  References 1. Siegel, R., et al., Cancer statistics, 2014. CA Cancer J Clin, 2014. 64(1): p. 9-29. 2. Statistics, C.C.S.s.A.C.o.C., Canadian Cancer Statistics 2013. 2013. 3. Sato, M., et al., A translational view of the molecular pathogenesis of lung cancer. J Thorac Oncol, 2007. 2(4): p. 327-43. 4. Sun, S., J.H. Schiller, and A.F. Gazdar, Lung cancer in never smokers--a different disease. Nat Rev Cancer, 2007. 7(10): p. 778-90. 5. Hecht, S.S., Tobacco carcinogens, their biomarkers and tobacco-induced cancer. Nat Rev Cancer, 2003. 3(10): p. 733-44. 6. Hajdu, S.I., Much overlooked causes of lung cancer. Ann Clin Lab Sci, 2011. 41(1): p. 97-101. 7. Travis, W.D., et al., Pathology and genetics of tumours of the lung, pleura, thymus and heart. World Health Organization classification of tumours2004:  8. Travis, W.D., et al., International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society International Multidisciplinary Classification of Lung Adenocarcinoma. Journal of Thoracic Oncology, 2011. 6(2): p. 244-285. 9. Travis, W.D., Pathology of lung cancer. Clin Chest Med, 2002. 23(1): p. 65-81, viii. 10. Wistuba, II and A.F. Gazdar, Lung cancer preneoplasia. Annu Rev Pathol, 2006. 1: p. 331-48. 11. Johnson, D.H., et al., Randomized phase II trial comparing bevacizumab plus carboplatin and paclitaxel with carboplatin and paclitaxel alone in previously untreated locally advanced or metastatic non-small-cell lung cancer. J Clin Oncol, 2004. 22(11): p. 2184-91. 12. Scagliotti, G., et al., The differential efficacy of pemetrexed according to NSCLC histology: a review of two Phase III studies. Oncologist, 2009. 14(3): p. 253-63. 13. Langer, C.J., et al., The evolving role of histology in the management of advanced non-small-cell lung cancer. J Clin Oncol, 2010. 28(36): p. 5311-20. 14. Heller, G., et al., Genome-wide CpG island methylation analyses in non-small cell lung cancer patients. Carcinogenesis, 2013. 34(3): p. 513-21. 15. Ding, L., et al., Somatic mutations affect key pathways in lung adenocarcinoma. Nature, 2008. 455(7216): p. 1069-75. 16. Liu, P., et al., Identification of somatic mutations in non-small cell lung carcinomas using whole-exome sequencing. Carcinogenesis, 2012. 33(7): p. 1270-6. 17. Seo, J.S., et al., The transcriptional landscape and mutational profile of lung adenocarcinoma. Genome Res, 2012. 22(11): p. 2109-19. 18. Weir, B.A., et al., Characterizing the cancer genome in lung adenocarcinoma. Nature, 2007. 450(7171): p. 893-8. 19. Vogelstein, B., et al., Cancer genome landscapes. Science, 2013. 339(6127): p. 1546-58. 20. Bhattacharjee, A., et al., Classification of human lung carcinomas by mRNA expression profiling reveals distinct adenocarcinoma subclasses. Proc Natl Acad Sci U S A, 2001. 98(24): p. 13790-5. 21. Garber, M.E., et al., Diversity of gene expression in adenocarcinoma of the lung. Proc Natl Acad Sci U S A, 2001. 98(24): p. 13784-9. 142  22. McDoniels-Silvers, A.L., et al., Differential gene expression in human lung adenocarcinomas and squamous cell carcinomas. Clin Cancer Res, 2002. 8(4): p. 1127-38. 23. McDoniels-Silvers, A.L., et al., Differential expression of critical cellular genes in human lung adenocarcinomas and squamous cell carcinomas in comparison to normal lung tissues. Neoplasia, 2002. 4(2): p. 141-50. 24. Nacht, M., et al., Molecular characteristics of non-small cell lung cancer. Proc Natl Acad Sci U S A, 2001. 98(26): p. 15203-8. 25. Wilkerson, M.D., et al., Lung squamous cell carcinoma mRNA expression subtypes are reproducible, clinically important, and correspond to normal cell types. Clin Cancer Res, 2010. 16(19): p. 4864-75. 26. Hou, J., et al., Gene expression-based classification of non-small cell lung carcinomas and survival prediction. PLoS One, 2010. 5(4): p. e10312. 27. Yamagata, N., et al., A training-testing approach to the molecular classification of resected non-small cell lung cancer. Clin Cancer Res, 2003. 9(13): p. 4695-704. 28. Virtanen, C., et al., Integrated classification of lung tumors and cell lines by expression profiling. Proc Natl Acad Sci U S A, 2002. 99(19): p. 12357-62. 29. Rohrbeck, A., et al., Gene expression profiling for molecular distinction and characterization of laser captured primary lung cancers. J Transl Med, 2008. 6: p. 69. 30. Beer, D.G., et al., Gene-expression profiles predict survival of patients with lung adenocarcinoma. Nat Med, 2002. 8(8): p. 816-24. 31. Giordano, T.J., et al., Organ-specific molecular classification of primary lung, colon, and ovarian adenocarcinomas using gene expression profiles. Am J Pathol, 2001. 159(4): p. 1231-8. 32. Balsara, B.R. and J.R. Testa, Chromosomal imbalances in human lung cancer. Oncogene, 2002. 21(45): p. 6877-83. 33. Pei, J., et al., Genomic imbalances in human lung adenocarcinomas and squamous cell carcinomas. Genes Chromosomes Cancer, 2001. 31(3): p. 282-7. 34. Sy, S.M., et al., Distinct patterns of genetic alterations in adenocarcinoma and squamous cell carcinoma of the lung. Eur J Cancer, 2004. 40(7): p. 1082-94. 35. Tonon, G., et al., High-resolution genomic profiles of human lung cancer. Proc Natl Acad Sci U S A, 2005. 102(27): p. 9625-30. 36. Yakut, T., et al., Assessment of molecular events in squamous and non-squamous cell lung carcinoma. Lung Cancer, 2006. 54(3): p. 293-301. 37. Massion, P.P., et al., Genomic copy number analysis of non-small cell lung cancer using array comparative genomic hybridization: implications of the phosphatidylinositol 3-kinase pathway. Cancer Res, 2002. 62(13): p. 3636-40. 38. Petersen, I., et al., Patterns of chromosomal imbalances in adenocarcinoma and squamous cell carcinoma of the lung. Cancer Res, 1997. 57(12): p. 2331-5. 39. Staaf, J., et al., Landscape of somatic allelic imbalances and copy number alterations in human lung carcinoma. Int J Cancer, 2013. 132(9): p. 2020-31. 40. Bass, A.J., et al., SOX2 is an amplified lineage-survival oncogene in lung and esophageal squamous cell carcinomas. Nature Genetics, 2009. 41(11): p. 1238-U105. 41. Bjorkqvist, A.M., et al., DNA gains in 3q occur frequently in squamous cell carcinoma of the lung, but not in adenocarcinoma. Genes Chromosomes Cancer, 1998. 22(1): p. 79-82. 143  42. Cancer Genome Atlas Research, N., et al., Comprehensive genomic characterization of squamous cell lung cancers. Nature, 2012. 489(7417): p. 519-25. 43. Luk, C., et al., Molecular cytogenetic analysis of non-small cell lung carcinoma by spectral karyotyping and comparative genomic hybridization. Cancer Genet Cytogenet, 2001. 125(2): p. 87-99. 44. Baylin, S.B., The cancer epigenome: its origins, contributions to tumorigenesis, and translational implications. Proc Am Thorac Soc, 2012. 9(2): p. 64-5. 45. Tessema, M. and S.A. Belinsky, Mining the epigenome for methylated genes in lung cancer. Proc Am Thorac Soc, 2008. 5(8): p. 806-10. 46. Heller, G., C.C. Zielinski, and S. Zochbauer-Muller, Lung cancer: from single-gene methylation to methylome profiling. Cancer Metastasis Rev, 2010. 29(1): p. 95-107. 47. Shiraishi, M., et al., HOX gene clusters are hotspots of de novo methylation in CpG islands of human lung adenocarcinomas. Oncogene, 2002. 21(22): p. 3659-62. 48. Zochbauer-Muller, S., et al., Aberrant promoter methylation of multiple genes in non-small cell lung cancers. Cancer Res, 2001. 61(1): p. 249-55. 49. Toyooka, S., et al., Mutational and epigenetic evidence for independent pathways for lung adenocarcinomas arising in smokers and never smokers. Cancer Res, 2006. 66(3): p. 1371-5. 50. Selamat, S.A., et al., Genome-scale analysis of DNA methylation in lung adenocarcinoma and integration with mRNA expression. Genome Res, 2012. 22(7): p. 1197-211. 51. Imielinski, M., et al., Mapping the hallmarks of lung adenocarcinoma with massively parallel sequencing. Cell, 2012. 150(6): p. 1107-20. 52. Liu, J., et al., Genome and transcriptome sequencing of lung cancers reveal diverse mutational and splicing events. Genome Res, 2012. 22(12): p. 2315-27. 53. Kandoth, C., et al., Mutational landscape and significance across 12 major cancer types. Nature, 2013. 502(7471): p. 333-9. 54. Lee, W., et al., The mutation spectrum revealed by paired genome sequences from a lung cancer patient. Nature, 2010. 465(7297): p. 473-7. 55. Hammerman, P.S., et al., Comprehensive genomic characterization of squamous cell lung cancers. Nature, 2012. 489(7417): p. 519-25. 56. Bartel, D.P., MicroRNAs: genomics, biogenesis, mechanism, and function. Cell, 2004. 116(2): p. 281-97. 57. Calin, G.A. and C.M. Croce, MicroRNA signatures in human cancers. Nat Rev Cancer, 2006. 6(11): p. 857-66. 58. Lebanony, D., et al., Diagnostic assay based on hsa-miR-205 expression distinguishes squamous from nonsquamous non-small-cell lung carcinoma. J Clin Oncol, 2009. 27(12): p. 2030-7. 59. Dacic, S., et al., miRNA expression profiling of lung adenocarcinomas: correlation with mutational status. Mod Pathol, 2010. 23(12): p. 1577-82. 60. Seike, M., et al., MiR-21 is an EGFR-regulated anti-apoptotic factor in lung cancer in never-smokers. Proc Natl Acad Sci U S A, 2009. 106(29): p. 12085-90. 61. Chari, R., et al., Integrating the multiple dimensions of genomic and epigenomic landscapes of cancer. Cancer Metastasis Rev, 2010. 29(1): p. 73-93. 62. Mok, T.S., et al., Gefitinib or carboplatin-paclitaxel in pulmonary adenocarcinoma. N Engl J Med, 2009. 361(10): p. 947-57. 144  63. Broet, P., et al., Prediction of clinical outcome in multiple lung cancer cohorts by integrative genomics: implications for chemotherapy selection. Cancer Res, 2009. 69(3): p. 1055-62. 64. Igawa, S., et al., Pemetrexed for previously treated patients with non-small cell lung cancer and differences in efficacy according to thymidylate synthase expression. Chemotherapy, 2012. 58(4): p. 313-20. 65. Ceppi, P., et al., Squamous cell carcinoma of the lung compared with other histotypes shows higher messenger RNA and protein levels for thymidylate synthase. Cancer, 2006. 107(7): p. 1589-96. 66. Thomas, R.K., B. Weir, and M. Meyerson, Genomic approaches to lung cancer. Clin Cancer Res, 2006. 12(14 Pt 2): p. 4384s-4391s. 67. Feinberg, A.P. and B. Tycko, The history of cancer epigenetics. Nat Rev Cancer, 2004. 4(2): p. 143-53. 68. Hyman, E., et al., Impact of DNA amplification on gene expression patterns in breast cancer. Cancer Res, 2002. 62(21): p. 6240-5. 69. Pollack, J.R., et al., Microarray analysis reveals a major direct role of DNA copy number alteration in the transcriptional program of human breast tumors. Proc Natl Acad Sci U S A, 2002. 99(20): p. 12963-8. 70. Bass, A.J., et al., SOX2 is an amplified lineage-survival oncogene in lung and esophageal squamous cell carcinomas. Nat Genet, 2009. 41(11): p. 1238-42. 71. Garnis, C., et al., Chromosome 5p aberrations are early events in lung cancer: implication of glial cell line-derived neurotrophic factor in disease progression. Oncogene, 2005. 24(30): p. 4806-12. 72. Coe, B.P., et al., Differential disruption of cell cycle pathways in small cell and non-small cell lung cancer. Br J Cancer, 2006. 94(12): p. 1927-35. 73. Baldwin, C., et al., Multiple microalterations detected at high frequency in oral cancer. Cancer Res, 2005. 65(17): p. 7561-7. 74. Lockwood, W.W., et al., Whole genome tiling path array CGH analysis of segmental copy number alterations in cervical cancer cell lines. Int J Cancer, 2007. 120(2): p. 436-43. 75. Khojasteh, M., et al., A stepwise framework for the normalization of array CGH data. BMC Bioinformatics, 2005. 6: p. 274. 76. Lockwood, W.W., et al., DNA amplification is a ubiquitous mechanism of oncogene activation in lung and other cancers. Oncogene, 2008. 27(33): p. 4615-24. 77. Chi, B., et al., SeeGH--a software tool for visualization of whole genome array comparative genomic hybridization data. BMC Bioinformatics, 2004. 5: p. 13. 78. Jong, K., et al., Breakpoint identification and smoothing of array comparative genomic hybridization data. Bioinformatics, 2004. 20(18): p. 3636-7. 79. Chari, R., et al., Effect of active smoking on the human bronchial epithelium transcriptome. BMC Genomics, 2007. 8: p. 297. 80. Irizarry, R.A., et al., Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics, 2003. 4(2): p. 249-64. 81. Bild, A.H., et al., Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature, 2006. 439(7074): p. 353-7. 145  82. Lee, E.S., et al., Prediction of recurrence-free survival in postoperative non-small cell lung cancer patients by using an integrated model of clinical information and gene expression. Clin Cancer Res, 2008. 14(22): p. 7397-404. 83. Lamb, J., et al., The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science, 2006. 313(5795): p. 1929-35. 84. Solis, L.M., et al., Nrf2 and Keap1 abnormalities in non-small cell lung carcinoma and association with clinicopathologic features. Clin Cancer Res, 2010. 16(14): p. 3743-53. 85. Yuan, P., et al., Sex determining region Y-Box 2 (SOX2) is a potential cell-lineage gene highly expressed in the pathogenesis of squamous cell carcinomas of the lung. PLoS One, 2010. 5(2): p. e9112. 86. Doi, A., et al., Differential methylation of tissue- and cancer-specific CpG island shores distinguishes human induced pluripotent stem cells, embryonic stem cells and fibroblasts. Nat Genet, 2009. 41(12): p. 1350-3. 87. Choi, S.H., et al., Changes in DNA methylation of tandem DNA repeats are different from interspersed repeats in cancer. Int J Cancer, 2009. 125(3): p. 723-9. 88. Hoffmann, M.J. and W.A. Schulz, Causes and consequences of DNA hypomethylation in human cancer. Biochem Cell Biol, 2005. 83(3): p. 296-321. 89. Yang, H.H., et al., Influence of genetic background and tissue types on global DNA methylation patterns. PLoS One, 2010. 5(2): p. e9355. 90. Vaissiere, T., et al., Quantitative analysis of DNA methylation profiles in lung cancer identifies aberrant DNA methylation of specific genes and its association with gender and cancer risk factors. Cancer Res, 2009. 69(1): p. 243-52. 91. Petersen, I., et al., Core classification of lung cancer: correlating nuclear size and mitoses with ploidy and clinicopathological parameters. Lung Cancer, 2009. 65(3): p. 312-8. 92. Wu, Z.L., et al., Polycomb protein EZH2 regulates E2F1-dependent apoptosis through epigenetically modulating Bim expression. Cell Death Differ, 2010. 17(5): p. 801-10. 93. Eymin, B., et al., Distinct pattern of E2F1 expression in human lung tumours: E2F1 is upregulated in small cell lung carcinoma. Oncogene, 2001. 20(14): p. 1678-87. 94. Higashiyama, M., et al., Reduced motility related protein-1 (MRP-1/CD9) gene expression as a factor of poor prognosis in non-small cell lung cancer. Cancer Res, 1995. 55(24): p. 6040-4. 95. Yamamoto, H., et al., PIK3CA mutations and copy number gains in human lung cancers. Cancer Res, 2008. 68(17): p. 6913-21. 96. Garraway, L.A. and W.R. Sellers, Lineage dependency and lineage-survival oncogenes in human cancer. Nat Rev Cancer, 2006. 6(8): p. 593-602. 97. Hillemacher, T., et al., Global DNA methylation is influenced by smoking behaviour. Eur Neuropsychopharmacol, 2008. 18(4): p. 295-8. 98. Lockwood, W.W., et al., Integrative genomic analyses identify BRF2 as a novel lineage-specific oncogene in lung squamous cell carcinoma. PLoS Med, 2010. 7(7): p. e1000315. 99. Dang, T.P., et al., Constitutive activation of Notch3 inhibits terminal epithelial differentiation in lungs of transgenic mice. Oncogene, 2003. 22(13): p. 1988-97. 100. Wang, I.C., et al., Transgenic expression of the forkhead box M1 transcription factor induces formation of lung tumors. Oncogene, 2008. 27(30): p. 4137-49. 146  101. Lee, D.F., et al., KEAP1 E3 ligase-mediated downregulation of NF-kappaB signaling by targeting IKKbeta. Mol Cell, 2009. 36(1): p. 131-40. 102. Strahl, B.D. and C.D. Allis, The language of covalent histone modifications. Nature, 2000. 403(6765): p. 41-5. 103. Esteller, M., Epigenetics in cancer. N Engl J Med, 2008. 358(11): p. 1148-59. 104. Esteller, M., Cancer epigenomics: DNA methylomes and histone-modification maps. Nat Rev Genet, 2007. 8(4): p. 286-98. 105. Barlesi, F., et al., Global histone modifications predict prognosis of resected non small-cell lung cancer. J Clin Oncol, 2007. 25(28): p. 4358-64. 106. Van Den Broeck, A., et al., Loss of histone H4K20 trimethylation occurs in preneoplasia and influences prognosis of non-small cell lung cancer. Clin Cancer Res, 2008. 14(22): p. 7237-45. 107. Lazarevich, N.L. and D.I. Fleishman, Tissue-specific transcription factors in progression of epithelial tumors. Biochemistry (Mosc), 2008. 73(5): p. 573-91. 108. Grigo, K., et al., HNF4 alpha orchestrates a set of 14 genes to down-regulate cell proliferation in kidney cells. Biol Chem, 2008. 389(2): p. 179-87. 109. Watt, A.J., W.D. Garrison, and S.A. Duncan, HNF4: a central regulator of hepatocyte differentiation and function. Hepatology, 2003. 37(6): p. 1249-53. 110. Lucas, B., et al., HNF4alpha reduces proliferation of kidney cells and affects genes deregulated in renal cell carcinoma. Oncogene, 2005. 24(42): p. 6418-31. 111. Gollob, J.A., et al., Role of Raf kinase in cancer: therapeutic potential of targeting the Raf/MEK/ERK signal transduction pathway. Semin Oncol, 2006. 33(4): p. 392-406. 112. Zhao, Y., et al., Inhibitors of histone deacetylases target the Rb-E2F1 pathway for apoptosis induction through activation of proapoptotic protein Bim. Proc Natl Acad Sci U S A, 2005. 102(44): p. 16090-5. 113. Orzan, F., et al., Enhancer of Zeste 2 (EZH2) is up-regulated in malignant gliomas and in glioma stem-like cells. Neuropathol Appl Neurobiol, 2011. 37(4): p. 381-94. 114. Herbst, R.S., J.V. Heymach, and S.M. Lippman, Lung cancer. N Engl J Med, 2008. 359(13): p. 1367-80. 115. Cullinan, S.B., et al., The Keap1-BTB protein is an adaptor that bridges Nrf2 to a Cul3-based E3 ligase: oxidative stress sensing by a Cul3-Keap1 ligase. Mol Cell Biol, 2004. 24(19): p. 8477-86. 116. Furukawa, M. and Y. Xiong, BTB protein Keap1 targets antioxidant transcription factor Nrf2 for ubiquitination by the Cullin 3-Roc1 ligase. Mol Cell Biol, 2005. 25(1): p. 162-71. 117. Kobayashi, A., et al., Oxidative stress sensor Keap1 functions as an adaptor for Cul3-based E3 ligase to regulate proteasomal degradation of Nrf2. Mol Cell Biol, 2004. 24(16): p. 7130-9. 118. Hayes, J.D. and M. McMahon, NRF2 and KEAP1 mutations: permanent activation of an adaptive response in cancer. Trends Biochem Sci, 2009. 34(4): p. 176-88. 119. Blake, D.J., et al., Deletion of Keap1 in the lung attenuates acute cigarette smoke-induced oxidative stress and inflammation. Am J Respir Cell Mol Biol, 2010. 42(5): p. 524-36. 120. Karin, M., et al., NF-kappaB in cancer: from innocent bystander to major culprit. Nat Rev Cancer, 2002. 2(4): p. 301-10. 147  121. Nakanishi, C. and M. Toi, Nuclear factor-kappaB inhibitors as sensitizers to anticancer drugs. Nat Rev Cancer, 2005. 5(4): p. 297-309. 122. Weiss, J., et al., Frequent and focal FGFR1 amplification associates with therapeutically tractable FGFR1 dependency in squamous cell lung cancer. Sci Transl Med, 2010. 2(62): p. 62ra93. 123. Okamoto, T., T. Sanda, and K. Asamitsu, NF-kappa B signaling and carcinogenesis. Curr Pharm Des, 2007. 13(5): p. 447-62. 124. Tang, X., et al., Nuclear factor-kappaB (NF-kappaB) is frequently expressed in lung cancer and preneoplastic lesions. Cancer, 2006. 107(11): p. 2637-46. 125. Meylan, E., et al., Requirement for NF-kappaB signalling in a mouse model of lung adenocarcinoma. Nature, 2009. 462(7269): p. 104-7. 126. Basseres, D.S., et al., Requirement of the NF-kappaB subunit p65/RelA for K-Ras-induced lung tumorigenesis. Cancer Res, 2010. 70(9): p. 3537-46. 127. Ohta, T., et al., Loss of Keap1 function activates Nrf2 and provides advantages for lung cancer cell growth. Cancer Res, 2008. 68(5): p. 1303-9. 128. Singh, A., et al., Dysfunctional KEAP1-NRF2 interaction in non-small-cell lung cancer. PLoS Med, 2006. 3(10): p. e420. 129. Wang, R., et al., Hypermethylation of the Keap1 gene in human lung cancer cell lines and lung cancer tissues. Biochem Biophys Res Commun, 2008. 373(1): p. 151-4. 130. Kohno, T., et al., A catalog of genes homozygously deleted in human lung cancer and the candidacy of PTPRD as a tumor suppressor gene. Genes Chromosomes Cancer, 2010. 131. Suzuki, T., J. Maher, and M. Yamamoto, Select Heterozygous Keap1 Mutations Have a Dominant-negative Effect on Wild-type Keap1 In Vivo. Cancer Research, 2010: p. epub. 132. Kan, Z., et al., Diverse somatic mutation patterns and pathway alterations in human cancers. Nature, 2010. 466(7308): p. 869-73. 133. Bivona, T.G., et al., FAS and NF-kappaB signalling modulate dependence of lung cancers on mutant EGFR. Nature, 2011. 471(7339): p. 523-6. 134. Garnis, C., et al., High resolution analysis of non-small cell lung cancer cell lines by whole genome tiling path array CGH. Int J Cancer, 2006. 118(6): p. 1556-64. 135. Ishkanian, A.S., et al., A tiling resolution DNA microarray with complete coverage of the human genome. Nat Genet, 2004. 36(3): p. 299-303. 136. Wong, K.K., et al., A comprehensive analysis of common copy-number variations in the human genome. Am J Hum Genet, 2007. 80(1): p. 91-104. 137. McCarroll, S.A., et al., Integrated detection and population-genetic analysis of SNPs and copy number variation. Nat Genet, 2008. 40(10): p. 1166-74. 138. Beroukhim, R., et al., The landscape of somatic copy-number alteration across human cancers. Nature, 2010. 463(7283): p. 899-905. 139. Schmid, J.A. and A. Birbach, IkappaB kinase beta (IKKbeta/IKK2/IKBKB)--a key molecule in signaling to the transcription factor NF-kappaB. Cytokine Growth Factor Rev, 2008. 19(2): p. 157-65. 140. Kohno, T., et al., KIF5B-RET fusions in lung adenocarcinoma. Nat Med, 2012. 18(3): p. 375-7. 141. Takeuchi, K., et al., RET, ROS1 and ALK fusions in lung cancer. Nat Med, 2012. 18(3): p. 378-81. 148  142. Pao, W. and K.E. Hutchinson, Chipping away at the lung cancer genome. Nat Med, 2012. 18(3): p. 349-51. 143. Hirsch, F.R., et al., Evaluation of HER-2/neu gene amplification and protein expression in non-small cell lung carcinomas. Br J Cancer, 2002. 86(9): p. 1449-56. 144. Okudela, K., et al., PIK3CA mutation and amplification in human lung cancer. Pathol Int, 2007. 57(10): p. 664-71. 145. Onozato, R., et al., Activation of MET by gene amplification or by splice mutations deleting the juxtamembrane domain in primary resected lung cancers. J Thorac Oncol, 2009. 4(1): p. 5-11. 146. Lockwood, W.W., et al., Divergent genomic and epigenomic landscapes of lung cancer subtypes underscore the selection of different oncogenic pathways during tumor development. PLoS One, 2012. 7(5): p. e37775. 147. Gazdar, A.F., et al., Lung cancer cell lines as tools for biomedical discovery and research. J Natl Cancer Inst, 2010. 102(17): p. 1310-21. 148. Beroukhim, R., et al., Assessing the significance of chromosomal aberrations in cancer: methodology and application to glioma. Proc Natl Acad Sci U S A, 2007. 104(50): p. 20007-12. 149. Coe, B.P., et al., FACADE: a fast and sensitive algorithm for the segmentation and calling of high resolution array CGH data. Nucleic Acids Res, 2010. 38(15): p. e157. 150. Chari, R., et al., A sequence-based approach to identify reference genes for gene expression analysis. BMC Med Genomics, 2010. 3: p. 32. 151. Thu, K.L., et al., Genetic disruption of KEAP1/CUL3 E3 ubiquitin ligase complex components is a key mechanism of NF-kappaB pathway activation in lung cancer. J Thorac Oncol, 2011. 6(9): p. 1521-9. 152. Zhao, X., et al., Homozygous deletions and chromosome amplifications in human lung carcinomas revealed by single nucleotide polymorphism array analysis. Cancer Res, 2005. 65(13): p. 5561-70. 153. Job, B., et al., Genomic aberrations in lung adenocarcinoma in never smokers. PLoS One, 2010. 5(12): p. e15145. 154. Fischer, U., et al., Cloning of a novel transcription factor-like gene amplified in human glioma including astrocytoma grade I. Hum Mol Genet, 1997. 6(11): p. 1817-22. 155. Munnia, A., et al., Expression, cellular distribution and protein binding of the glioma amplified sequence (GAS41), a highly conserved putative transcription factor. Oncogene, 2001. 20(35): p. 4853-63. 156. Barretina, J., et al., Subtype-specific genomic alterations define new targets for soft-tissue sarcoma therapy. Nat Genet, 2010. 42(8): p. 715-21. 157. Shedden, K., et al., Gene expression-based survival prediction in lung adenocarcinoma: a multi-site, blinded validation study. Nat Med, 2008. 14(8): p. 822-7. 158. Petitjean, A., et al., Impact of mutant p53 functional properties on TP53 mutation patterns and tumor phenotype: lessons from recent developments in the IARC TP53 database. Hum Mutat, 2007. 28(6): p. 622-9. 159. Park, J.H. and R.G. Roeder, GAS41 is required for repression of the p53 tumor suppressor pathway during normal cellular proliferation. Mol Cell Biol, 2006. 26(11): p. 4006-16. 149  160. Medjkane, S., et al., Myocardin-related transcription factors and SRF are required for cytoskeletal dynamics and experimental metastasis. Nat Cell Biol, 2009. 11(3): p. 257-68. 161. Prencipe, M., et al., Identification of transcription factors associated with castration-resistance: is the serum responsive factor a potential therapeutic target? Prostate, 2013. 73(7): p. 743-53. 162. Zimmermann, K., et al., Targeted disruption of the GAS41 gene encoding a putative transcription factor indicates that GAS41 is essential for cell viability. J Biol Chem, 2002. 277(21): p. 18626-31. 163. Schulze, J.M., A.Y. Wang, and M.S. Kobor, YEATS domain proteins: a diverse family with many links to chromatin modification and transcription. Biochem Cell Biol, 2009. 87(1): p. 65-75. 164. Cai, Y., et al., The mammalian YL1 protein is a shared subunit of the TRRAP/TIP60 histone acetyltransferase and SRCAP complexes. J Biol Chem, 2005. 280(14): p. 13665-70. 165. Doyon, Y., et al., Structural and functional conservation of the NuA4 histone acetyltransferase complex from yeast to humans. Mol Cell Biol, 2004. 24(5): p. 1884-96. 166. Piccinni, E., et al., Direct interaction of Gas41 and Myc encoded by amplified genes in nervous system tumours. Acta Biochim Pol, 2011. 58(4): p. 529-34. 167. Lauffart, B., et al., Interaction of the transforming acidic coiled-coil 1 (TACC1) protein with ch-TOG and GAS41/NuBI1 suggests multiple TACC1-containing protein complexes in human cells. Biochem J, 2002. 363(Pt 1): p. 195-200. 168. Harborth, J., K. Weber, and M. Osborn, GAS41, a highly conserved protein in eukaryotic nuclei, binds to NuMA. J Biol Chem, 2000. 275(41): p. 31979-85. 169. Debernardi, S., et al., The MLL fusion partner AF10 binds GAS41, a protein that interacts with the human SWI/SNF complex. Blood, 2002. 99(1): p. 275-81. 170. Schmitt, J., et al., GAS41 amplification results in overexpression of a new spindle pole protein. Genes Chromosomes Cancer, 2012. 51(9): p. 868-80. 171. Marine, J.C. and G. Lozano, Mdm2-mediated ubiquitylation: p53 and beyond. Cell Death Differ, 2010. 17(1): p. 93-102. 172. Manfredi, J.J., The Mdm2-p53 relationship evolves: Mdm2 swings both ways as an oncogene and a tumor suppressor. Genes Dev, 2010. 24(15): p. 1580-9. 173. Momand, J., et al., The MDM2 gene amplification database. Nucleic Acids Res, 1998. 26(15): p. 3453-9. 174. Birney, E., et al., Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature, 2007. 447(7146): p. 799-816. 175. Ponting, C.P. and T.G. Belgard, Transcribed dark matter: meaning or myth? Hum Mol Genet, 2010. 19(R2): p. R162-8. 176. Carninci, P., et al., The transcriptional landscape of the mammalian genome. Science, 2005. 309(5740): p. 1559-63. 177. Hung, T. and H.Y. Chang, Long noncoding RNA in genome regulation: prospects and mechanisms. RNA Biol, 2010. 7(5): p. 582-5. 178. Taft, R.J., et al., Non-coding RNAs: regulators of disease. J Pathol, 2010. 220(2): p. 126-39. 179. Zhang, L., et al., microRNAs exhibit high frequency genomic alterations in human cancer. Proc Natl Acad Sci U S A, 2006. 103(24): p. 9136-41. 150  180. Calin, G.A., et al., Human microRNA genes are frequently located at fragile sites and genomic regions involved in cancers. Proc Natl Acad Sci U S A, 2004. 101(9): p. 2999-3004. 181. Kozomara, A. and S. Griffiths-Jones, miRBase: annotating high confidence microRNAs using deep sequencing data. Nucleic Acids Res, 2014. 42(Database issue): p. D68-73. 182. Iorio, M.V. and C.M. Croce, MicroRNAs in cancer: small molecules with a huge impact. J Clin Oncol, 2009. 27(34): p. 5848-56. 183. Yanaihara, N., et al., Unique microRNA molecular profiles in lung cancer diagnosis and prognosis. Cancer Cell, 2006. 9(3): p. 189-98. 184. Takamizawa, J., et al., Reduced expression of the let-7 microRNAs in human lung cancers in association with shortened postoperative survival. Cancer Res, 2004. 64(11): p. 3753-6. 185. Garofalo, M., et al., miR-221&222 regulate TRAIL resistance and enhance tumorigenicity through PTEN and TIMP3 downregulation. Cancer Cell, 2009. 16(6): p. 498-509. 186. Gallardo, E., et al., miR-34a as a prognostic marker of relapse in surgically resected non-small-cell lung cancer. Carcinogenesis, 2009. 30(11): p. 1903-9. 187. Jiang, L., et al., Hsa-miR-125a-3p and hsa-miR-125a-5p are downregulated in non-small cell lung cancer and have inverse effects on invasion and migration of lung cancer cells. BMC Cancer, 2010. 10: p. 318. 188. Markou, A., et al., Prognostic value of mature microRNA-21 and microRNA-205 overexpression in non-small cell lung cancer by quantitative real-time RT-PCR. Clin Chem, 2008. 54(10): p. 1696-704. 189. Yu, S.L., et al., MicroRNA signature predicts survival and relapse in lung cancer. Cancer Cell, 2008. 13(1): p. 48-57. 190. Comprehensive molecular portraits of human breast tumours. Nature, 2012. 490(7418): p. 61-70. 191. McGuffin, M.J. and I. Jurisica, Interaction techniques for selecting and manipulating subgraphs in network visualizations. IEEE Trans Vis Comput Graph, 2009. 15(6): p. 937-44. 192. Pio, G., et al., Integrating microRNA target predictions for the discovery of gene regulatory networks: a semi-supervised ensemble learning approach. BMC Bioinformatics, 2014. 15 Suppl 1: p. S4. 193. Shirdel, E.A., et al., NAViGaTing the micronome--using multiple microRNA prediction databases to identify signalling pathway-associated microRNAs. PLoS One, 2011. 6(2): p. e17429. 194. Brown, K.R., et al., NAViGaTOR: Network Analysis, Visualization and Graphing Toronto. Bioinformatics, 2009. 25(24): p. 3327-9. 195. Lockwood, W.W., et al., Integrative genomics identified RFC3 as an amplified candidate oncogene in esophageal adenocarcinoma. Clin Cancer Res, 2012. 18(7): p. 1936-46. 196. Takahashi, Y., et al., MiR-107 and MiR-185 can induce cell cycle arrest in human non small cell lung cancer cell lines. PLoS One, 2009. 4(8): p. e6677. 197. Zhu, W., et al., Expression of miR-29c, miR-93, and miR-429 as potential biomarkers for detection of early stage non-small lung cancer. PLoS One, 2014. 9(2): p. e87780. 151  198. Zhu, C.Q., et al., Understanding prognostic gene expression signatures in lung cancer. Clin Lung Cancer, 2009. 10(5): p. 331-40. 199. Landi, M.T., et al., MicroRNA expression differentiates histology and predicts survival of lung cancer. Clin Cancer Res, 2010. 16(2): p. 430-41. 200. Lu, Y., et al., MicroRNA profiling and prediction of recurrence/relapse-free survival in stage I lung cancer. Carcinogenesis, 2012. 33(5): p. 1046-54. 201. Bishop, J.A., et al., Accurate classification of non-small cell lung carcinoma using a novel microRNA-based approach. Clin Cancer Res, 2010. 16(2): p. 610-9. 202. Vosa, U., et al., Meta-analysis of microRNA expression in lung cancer. Int J Cancer, 2013. 132(12): p. 2884-93. 203. Guan, P., et al., Meta-analysis of human lung cancer microRNA expression profiling studies comparing cancer tissues with normal tissues. J Exp Clin Cancer Res, 2012. 31: p. 54. 204. Hamamoto, J., et al., Identification of microRNAs differentially expressed between lung squamous cell carcinoma and lung adenocarcinoma. Mol Med Rep, 2013. 8(2): p. 456-62. 205. Zhang, Y.K., et al., miRNAs expression profiling to distinguish lung squamous-cell carcinoma from adenocarcinoma subtypes. J Cancer Res Clin Oncol, 2012. 138(10): p. 1641-50. 206. Lazar, V., et al., Integrated molecular portrait of non-small cell lung cancers. BMC Med Genomics, 2013. 6: p. 53. 207. Yan, J.W., J.S. Lin, and X.X. He, The emerging role of miR-375 in cancer. Int J Cancer, 2013. 208. Jonsdottir, K., et al., Validation of expression patterns for nine miRNAs in 204 lymph-node negative breast cancers. PLoS One, 2012. 7(11): p. e48692. 209. Mathe, E.A., et al., MicroRNA expression in squamous cell carcinoma and adenocarcinoma of the esophagus: associations with survival. Clin Cancer Res, 2009. 15(19): p. 6192-200. 210. Nishikawa, E., et al., miR-375 is activated by ASH1 and inhibits YAP1 in a lineage-dependent manner in lung cancer. Cancer Res, 2011. 71(19): p. 6165-73. 211. Tan, M., et al., SAG/ROC2/RBX2 is a HIF-1 target gene that promotes HIF-1 alpha ubiquitination and degradation. Oncogene, 2008. 27(10): p. 1404-11. 212. Nie, L., et al., Notch-induced Asb2 expression promotes protein ubiquitination by forming non-canonical E3 ligase complexes. Cell Res, 2011. 21(5): p. 754-69. 213. Sato, Y., et al., Degradation of phosphorylated p53 by viral protein-ECS E3 ligase complex. PLoS Pathog, 2009. 5(7): p. e1000530. 214. Ishikawa, K., H. Ishii, and T. Saito, DNA damage-dependent cell cycle checkpoints and genomic stability. DNA Cell Biol, 2006. 25(7): p. 406-11. 215. Sierant, M.L., N.E. Archer, and S.K. Davey, The Rad9A checkpoint protein is required for nuclear localization of the claspin adaptor protein. Cell Cycle, 2010. 9(3): p. 548-56. 216. Maniwa, Y., et al., His239Arg SNP of HRAD9 is associated with lung adenocarcinoma. Cancer, 2006. 106(5): p. 1117-22. 217. Margueron, R. and D. Reinberg, The Polycomb complex PRC2 and its mark in life. Nature, 2011. 469(7330): p. 343-9. 152  218. Zhou, V.W., A. Goren, and B.E. Bernstein, Charting histone modifications and the functional organization of mammalian genomes. Nat Rev Genet, 2011. 12(1): p. 7-18. 219. Hanahan, D. and R.A. Weinberg, Hallmarks of cancer: the next generation. Cell, 2011. 144(5): p. 646-74. 220. Hall, R.D., J.E. Gray, and A.A. Chiappori, Beyond the standard of care: a review of novel immunotherapy trials for the treatment of lung cancer. Cancer Control, 2013. 20(1): p. 22-31. 221. Zielinski, C., et al., Rationale for targeting the immune system through checkpoint molecule blockade in the treatment of non-small-cell lung cancer. Ann Oncol, 2013. 24(5): p. 1170-9. 222. Decoster, L., I. Wauters, and J.F. Vansteenkiste, Vaccination therapy for non-small-cell lung cancer: review of agents in phase III development. Ann Oncol, 2012. 23(6): p. 1387-93. 223. Akbay, E.A., et al., Activation of the PD-1 pathway contributes to immune escape in EGFR-driven lung tumors. Cancer Discov, 2013. 3(12): p. 1355-63.    153  Appendix Appendix A  Sample sets and demographic data of cohorts used in this thesis  A.1 Samples sets used in Chapter 2   A.2 Demographic data of BCCRC 261 AC and 92 SqCC with copy number profiles               Sample Set Cohort # Samples # AC # SqCC # Non-Neoplasitic Lung Tissue Assay Type 1 BC Cancer Agency - Vancouver 261 169 92 0 Copy Number - BCCRC Whole Genome Tiling Path Array CGH 2 BC Cancer Agency - Vancouver 92 30 13 48 DNA Methylation - Illumina HumanMethylation27 chip 3 BC Cancer Agency - Vancouver 49 29 20 0 Gene Expression - Custom Affymetrix 4 GEO Duke University – GSE3141 111 58 53 0 Gene Expression - Affymetrix GeneChip Human Genome U133 Plus 2.0 Array 5 BC Cancer Agency – Vancouver 0 0 0 67 Gene Expression - Affymetrix GeneChip Human Genome U133 Plus 2.0 Array 6 GEO Samsung Medical Center -GSE8894 138 62 76 0 Gene Expression - Affymetrix GeneChip Human Genome U133 Plus 2.0 Array  AC (n=169) SqCC (n=92) Stage I 76 (44.9%) 32 (34.7%)  II 40 (23.6%) 32 (34.7%)  III 22 (13%) 14 (15.2%)  IV 27 (16%) 10 (10.9%) Sex Male 63 (37.3%) 66 (71.7%)  Female 106 (62.7%) 26 (28.3%) Age Median 66 69  Age Range 35-90 48-88 Smoking Status Current smoker 48 (28.4%) 30 (32.6%)  Former smoker 80 (47.3%) 61 (66.3%)  Never smoker 38 (22.5%)   N/A 3 (1.8%) 1 (1.1%) 154  A.3 Datasets used to determine the prevalence and subtype specificity of YEATS4 amplification and overexpression DataSet Samples Platform Institute Website 1 169 AC & 92 SqCC tumors aCGH BCCRC http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE31798 2 35 AC & 13 SqCC with matched non-malignant tissue Custom Affymetrix expression arrays BCCRC http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE31799 3 59 AC with matched non-malignant tissue qRT-PCR BCCRC N/A 4 18 NSCLC cell lines qRT-PCR BCCRC N/A 5 83 AC & matched non-malignant tissue SNP 6,  Illumina WG-6 v3 BeadChip Arrays Early Detection Research Network (EDRN)/Canary Foundation http://edrn.nci.nih.gov/science-data 6 155 SqCC & 77AC with matched non-malignant tissue SNP 6 Max Plank Institute (GSE25016) http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE25016 7 354 AC with matched non-malignant tissue SNP 6 Broad Institute- Database of genotypes and phenotypes http://www.broadinstitute.org/cgibin/cancer/publications/pub_paper.cgi?mode=view&paper_id=169 8 199 AC with matched non-malignant tissue Agilent 44K aCGH Memorial Sloan Kettering Cancer Center http://cbio.mskcc.org/Public/lung_array_data/ 9 508 cell lines from multiple human cancers SNP6 Sanger Cell Line Project http://www.sanger.ac.uk/genetics/CGP/CellLines/ 10 277 AC & 201 SqCC; 25 AC & 17 SqCC with non-malignant tissue SNP6 and RNA seq The Cancer Genome Atlas https://tcga-data.nci.nih.gov/tcga/ 155  Appendix B  List of publications  This Appendix lists all of the publications I contributed to during my degree that were either published, accepted, currently in submission or prepared for submission. In total there are 26 publications. First or co-first authorships are underlined.  1. Thu KL, Pikor L, Kennett J, Alvarez C, Lam WL (2010) Methylation analysis by DNA immunoprecipitation. Journal of Cellular Physiology 222: 522-31 2. Raj Chari, Kim M. Lonergan, Larissa A. Pikor, Bradley P. Coe, Chang Qi Zhu, Timothy H. W. Chan, Calum MacAulay, Ming-Sound Tsao, Stephen Lam, Raymond T. Ng, Wan L. Lam (2010)  A sequenced-based approach to identify reference genes for gene expression analysis. BMC Medical Genomics 3:32,1-11. 3. Katey S. S. Enfield, Greg L Stewart, Larissa Pikor, Carlos E. Alvarez, Stephen Lam, Wan L. Lam and Raj Chari (2011) MircroRNA gene dosage alterations and drug response in lung cancer. Journal of Biomedicine and Biotechnology 2011:474632,1-15 4. Pikor LA, Enfield K, Heryet C, Lam WL (2011). DNA extraction from paraffin embedded material. Journal of Visualized Experiments Mar 26;(49):2763  5. Thu KL, Pikor LA, Chari R, Wilson IM, MacAulay CE, Gazdar AF, Lam S, Lam WL, Lockwood WW (2011) Disruption of KEAP1/CUL3 E3 ubiquitin ligase complex components is a key mechanism for NFkB pathway activation in lung cancer. Journal of Thoracic Oncology 6(9):1521-9.  6. WW Lockwood, K Thu, L Lin, L Pikor, R Chari, WL Lam, DG Beer (2012) Integrative genomics identified RFC3 as an amplified candidate oncogene in esophageal adenocarcinoma. Clinical Cancer Research 1;18(7): 1936-46 7. Enfield KSS, Pikor LA, Martinez VD, Lam WL (2012). Mechanistic roles of non-coding RNAs in lung cancer biology and their clinical implications. Genetics Research International 2012:737416, 1-16  8. Lockwood WW, Wilson IM, Coe BP, Chari R, Pikor LA, Thu KL, Yee J, English J, Murray N, Tsao MS, Minna J, Gazdar AF, MacAulay CE, Lam S, Lam WL (2012). Divergent genomic and epigenomic landscapes of lung cancer subtypes underscore the selection of different oncogenic pathways during tumor development. PLoS One 2012;7(5) e37775 9. Pikor L, Thu K, Vucic E, Lam W (2013). The detection and implication of genome instability in cancer. Cancer Metastasis Review;32 (3-4):341-52 10. Crea F, Sun L, Pikor L, Lam WL, Helgason CD (2013). Mutational analysis of Polycomb genes in solid tumors identifies PHC3 amplification as a possible cancer-driving genetic alteration. British Journal of Cancer 109(6):1699-702 11. Pikor LA, Ramnarine VR, Lam S, Lam WL (2013). Genetic alterations defining NSCLC subtypes and their therapeutic implications. Lung Cancer 82(2)179-189 12. Martinez VD, Vucic EA, Pikor LA, Thu KL, Hubaux R, Lam WL (2013). Frequent concerted genetic mechanisms disrupt multiple components of the NRF2 inhibtor KEAP1/CUL3/RBX1 E3-ubiquitin ligase complex in thyroid cancer. Molecular Cancer 12:124,1-6 156  13. Pikor LA, Lockwood WW, Chari R, Vucic EA, Lam S, Lam WL (2013). YEATS4 is a novel oncogene amplified in non-small cell lung cancer that regulates the p53 pathway. Cancer Research 73(24):7301-12.  14. Thu KL, Radulovich N,  Becker-Santos DD, Pikor LA, Pusic A, Lockwood WW, Tsao MS, Lam WL (2014). SOX15 is a novel tumor suppressor in pancreatic cancer with a role in Wnt/β-catenin signaling. Oncogene 33(3):379-88 15. Martinez VD, Vucic EA, Thu KL, Pikor LA, Lam S, Lam WL (2013) Disruption of KEAP1 /CUL3/RBX1 E3 ubiquitin ligase complex components by multiple genetic mechanisms is associated with poor prognosis in head and neck cancer. Head & Neck, In press.  16. Crea F, Watahiki A, Quagliata L, Xue H, Pikor L, Parolia A,Wang Y, LinD, Lam WL, Farrar WL, Isogai T, Morant R, Castori-Eppenberger S, Chi KN, Wang Y, Helgason CD (2014) Identification of a long non-coding RNA as a novel biomarker and potential therapeutic targetd for metastatic prostate cancer. Oncotrarget, In press.  17. Vucic EA, Thu KL, Pikor LA, Enfield KSS, Yee J, English JC, MacAulay CE, Lam S, Jurisica I, Lam WL (2014) Smoking status impacts microRNA mediated prognosis in lung adenocarcinoma biology. Molecular Cancer, Under Review. 18. Martinez VD, Vucic EA, Thu KL, Pikor LA, Hubaux R, Lam WL (2014) Unique pattern of component gene disruption in the NRF2 inhibitor KEAP1/CUL3/RBX1 E3-ubiquiting ligase complex in serous ovarian cancer. BioMed Research International, Under Review.  19. Hubaux R, Thu KL, Vucic EA, Pikor LA, Kung SHY, Martinez VD, Mosslemi M, Becker-Santos DD, Gazdar AF, Lam S, Lam WL (2014) Microtubule affinity-regulating kinase 2 contributes to cisplatin sensitivity through modulation of the DNA damage response in non-small cell lung cancer. Oncotarget. Under review. 20. Crea F, Quagliata L, Frumento P,  Azad A, Xue H, Pikor LA, Watahiki A, Morant R, Castori-Eppenberger S, Wang Y, Parolia A, Lam WL, Gleave M, Chi KN, Wang YZ, Helgason CD (2014) Integrated analysis of the prostate cancer small-nucleolar transcriptome reveals snoRNA55 as a novel diagnostic and prognostic biomarker. Annals of Oncology. Under review.   Manuscripts in preparation.  21. Pikor LA, Chari R, Kennett JY, Solis LM, Valencia I, Wistuba I, Gazdar AF, Lam S, Lam WL (2014) SIRPA is a candidate tumor suppressor in lung adenocarcinoma that induces senescence mediated by RB and is associated with EGFR mutation. In preparation.  22. Pikor LA, Thu KL, Vucic EA, Lam S, Lam WL (2014). miRNA sequencing identifies miRNAs that differentiate between adenocarcinoma and squamous cell carcinoma and distinguish tumors from non-malignant tissue and benign lesions. 23. Thu KL, Radulovich N, Becker-Santos DD, Pikor LA, Lam WL, Tsao MS (2014) SOX15 and other SOX family members are important mediators of tumorigenesis in multiple cancer types. OncoScience. In preparation. 24. Becker-Santos DD, Thu KL, Pikor LA, Vucic EA, MacAulay CE, Jurisica I, Robinson WP, Lam S, Lam WL (2014). miRNA expression patterns in human non-small cell lung cancer and fetal lung: a comparative study. In preparation for Journal of Thoracic Oncology 157  25. Conway E, Pikor LA, Lam S, Lam WL (2014). Macrophages, inflammation and cancer. In preparation. 26. Mosslemi M, Thu KL, Vucic EA, Pikor LA, Ng RT, MacAulay CE, Lam WL (2014). Development of a Multi-dimensional Integrative Tumor gene Ranking Algorithm (MITRA) for the identification of candidate genes in cancer. In preparation      

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0167346/manifest

Comment

Related Items