Oral Presentations

-1-

95.

Comprehensive Molecular Characterization of Papillary Renal Cell Carcinoma

W. Marston Linehan The Cancer Genome Atlas Research Network, National Institutes of Health, Bethesda, Maryland Background: Papillary renal cell carcinoma (PRCC), the second most common type of RCC, is a heterogeneous disease made up of a number of different types of renal cancer, including those with indolent, often multifocal presentation as well as solitary tumors with an aggressive, highly lethal phenotype. Little is known about the genetic basis of sporadic papillary RCC and there are no effective forms of therapy for patients with advanced disease. Methods: We performed comprehensive molecular characterization of 161 surgically resected primary papillary renal cell carcinomas using whole-exome sequencing, messenger RNA, microRNA, copy number, methylation and proteomic analyses. Integrative analysis was performed to correlate the molecular features with stage and survival. Results: We determined that Type 1 and Type 2 PRCC represented distinctly different types of renal cancer characterized by specific genetic alterations, and that Type 2 PRCC could be further classified into at least three individual subtypes based on molecular differences and patient survival. Alterations in MET were associated with Type 1 tumors, while CDKN2A silencing, SETD2 mutations, TFE3 fusions, and increased expression of the NRF2ARE pathway were identified in Type 2 tumors. We found a CpG island methylator phenotype (CIMP) in a distinct subset of Type 2 PRCC characterized by early onset, poor survival, and germline or somatic mutation of the fumarate hydratase (FH) gene. Conclusions: Integrative analysis confirms different biological entities characterized by distinct genetic features for the Type 1 and Type 2 histologic subtypes of PRCC, and reveals an important role for the MET (Type 1) and NRF2ARE (Type 2) pathways.

-2-

38.

High-Throughput Somatic Variant Impact Phenotyping Using Gene Expression Signatures

Angela N. Brooks1,2, Alice H. Berger1,2, Xiaoyun Wu2, Larson Hogstrom2, Itay Tirosh2, Federica Piccioni2, Mukta Bagul2, Cong Zhu2, Yashaswi Shretha2, David Root2, Pablo Tamayo2, Ryo Sakai3, Bang Wong2, Ted Natoli2, David Lahr2, Atanas Kamburov2, Aravind Subramanian2, Gad Getz2, Todd Golub1,2, Matthew Meyerson1,2, Jesse Boehm2 1Division

of Medical Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts; 2Cancer Program, Broad Institute, Cambridge, Massachusetts; 3Department of Electrical Engineering, KU Leuven, Belgium Cancer genome sequencing efforts have led to the rapid identification of thousands of cancer-associated somatic mutations; however, there is a significant bottleneck in determining the functional impact of these variants. In addition, infrequently observed variants, even in well-characterized genes, pose a challenge in distinguishing impactful from passenger mutations. Understanding mutation function is critical for our knowledge of cancer biology and to more rapidly determine targeted treatment strategies based on individual tumor genetic profiles. We report a high-throughput approach for expression-based variant impact phenotyping called e-VIP. We applied this method to study ~450 somatic mutations identified in primary lung adenocarcinomas. The e-VIP approach compares gene expression changes upon introduction of wild-type versus mutant allele cDNAs in cell lines to disentangle functional mutations from likely inert mutations. We further classify alleles as gain-offunction or loss-of-function through differences in the signature strength between wild-type and mutant alleles. e-VIP correctly classifies known functional mutations in genes such as KRAS, EGFR, and RIT1 and predicts functional effects of never-characterized mutations. We characterized rare mutations in clinically-actionable oncogenes such as EGFR and unexpected dominant mutations in the transcription factor MAX and the phosphatase subunit PPP2R1A, among others. We observed an enrichment of loss-of-function missense mutations in known tumor suppressor genes such as STK11, KEAP1, and FBXW7. Most genes assayed also harbored variants that are likely inert, further underscoring the importance of characterizing individual variant alleles. Orthogonal functional approaches including an EGFR inhibitor resistance screen and a pooled tumor formation assay, were used as validation. In principle, e-VIP can characterize any genetic variant, independent of prior knowledge of gene function, and should significantly advance the pace of functional characterization of variants identified from genome sequencing studies.

-3-

32.

CoMEt: A Statistical Approach to Identify Combinations of Mutually Exclusive Alterations in Cancer

Mark D.M. Leiserson1, Hsin-Ta Wu1, Fabio Vandin1, Benjamin J. Raphael1 1Department

of Computer Science and Center for Computational Molecular Biology, Brown University, Providence, Rhode Island Identifying driver mutations in cancer genomes is a significant challenge due to the mutational heterogeneity of tumors. This mutational heterogeneity arises because driver mutations target genes in signaling and regulatory pathways, each of which can be perturbed in numerous ways. We introduce Combinations of Mutually Exclusive Alterations (CoMEt), an algorithm to identify combinations of candidate driver mutations de novo, without any prior biological knowledge (e.g. pathways or protein interactions). CoMEt searches for combinations of mutations that exhibit mutual exclusivity, a pattern frequently observed for mutations in cancer pathways. CoMEt uses an exact statistical test for mutual exclusivity that is less biased toward high frequency alterations than previous approaches and more sensitive in detecting combinations of lower frequency alterations. We compute the exact test using a novel tail enumeration procedure and also derive a binomial approximation. CoMEt simultaneously identifies collections of one or more combinations of mutually exclusive alterations, consistent with the observation of multiple hallmarks of cancer. CoMEt summarizes over multiple possible collections with high scores. Finally, CoMEt also enables simultaneous analysis of subtype-specific mutations. We show that CoMEt outperforms other mutual exclusivity approaches on simulated and real data. We apply CoMEt to hundreds of samples from four different TCGA cancer types: gastric cancer (STAD), glioblastoma (GBM) and acute myeloid leukemia (AML), and breast cancer (BRCA). We identify multiple mutually exclusive sets within each cancer type. These include the RTK/RAS pathway in gastric cancer; the Rb and p53 signaling pathways in GBM; and a collection containing multiple kinases, including FLT3 and RAS genes, in AML. Many of these collections overlap known pathways, but others reveal novel putative cancer genes. In addition, we analyze subtype-specific mutations in four molecular subtypes of breast cancer and three molecular subtypes of gastric cancer. We identify several pathways that are enriched for mutations in specific subtypes including the PI(3)K/AKT signaling pathway in the Luminal A subtype of BRCA and the strong exclusive module containing CDH1, the fusion gene ARHGAP26-CLDN18, and amplification on EPHB3 in the genomically stable subtype of STAD. CoMEt analysis also reveals subtle relationships between subtype-specific mutations and mutations in different pathways. These findings provide testable hypotheses for experimental validation.

-4-

91.

Decoding Breast Cancer with Quantitative Radiomics and Radiogenomics: Imaging Phenotypes in Breast Cancer Risk Assessment, Diagnosis, Prognosis, and Response to Therapy

Maryellen Giger, Hui Li, Karen Drukker, Yuan Ji, Yitan Zhu, Charles Perou, Carl Jaffe, Justin Kirby, Erich Huang, John Freyman, Elizabeth Morris, Elizabeth Burnside, and the TCIA Breast Cancer Group Purpose: To demonstrate, using the TCGA TCIA breast cancer dataset, the role of quantitative radiomics in characterizing the molecular subtypes of breast cancer and associating the magenetic resonance imaging (MRI) computer-extracted image phenotypes (CEIP) with genomic data. Understanding of the potentially correlative or complimentary relationships between quantitative image phenotypes, cancer subtypes, and genomic data of breast tumors is expected to allow for improved prognostic assessment and subsequently more effective cancer treatment plans. Method and Materials: Analyses were performed on the TCGA breast cases that possessed corresponding MRI studies. MRI-based phenotyping analysis included 3D lesion segmentation based on a fuzzy c-means clustering algorithm and the computerized feature extraction yielding quantitative characteristics from the six phenotypic categories of size, shape, morphology, enhancement texture, kinetics, and variance kinetics. Correlative and classification analyses were conducted. The performance of the image-based phenotypes and genomic data in distinguishing between molecular subtypes of breast cancer was evaluated using ROC analysis with area under the ROC curve (AUC) as the figure of merit. Results: MR images were available on 91 TCGA breast cases. After identification of the lesion locations by TCIA radiologists, the quantitative analysis was automatically conducted and evaluation conducted in tasks including receptor status (e.g., ER, PR), molecular subtype (e.g., Luminal A, Luminal B, Basal-like), risk of recurrence (e.g., PAM50, Mammaprint), and genomic data. Multiple linear regression analyses demonstrated statistically significant Pearson correlations (0.5-0.55) between MRI tumor signatures and multi-gene assay recurrence scores. Important MR phenotypes included tumor size and enhancement texture patterns characterizing tumor heterogeneity. Use of the MRI signatures in the tasks of distinguishing between good and poor prognosis in terms of levels of recurrence yielded AUC values (standard error) of 0.83 (0.07), 0.77 (0.06), 0.80 (0.07), and 0.75 (0.08) for MammaPrint, Oncotype DX, PAM50 Risk of Relapse Subtype (ROR-S), and PAM50 ROR-P (subtype+proliferation), respectively. Significant associations were also identified between the MRI phenotypes (such as tumor size, shape, margin, enhancement texture, blood flow kinetics) and molecular features involved in multiple regulation layers (including DNA mutation, miRNA expression, protein expression, pathway gene expression and copy number variation). Conclusion: The results from this study indicate that quantitative MRI analysis shows promise as a means for highthroughput image-based phenotyping to yield quantitative predictive models of breast cancer for precision medicine and patient treatment strategies. This project is funded in part by the University of Chicago Dean Bridge Fund and in part with federal funds from the National Cancer Institute, National Institutes of Health, under Contract No. HHSN261200800001E. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.

-5-

102.

Integrated Genomic Characterization of Pheochromocytoma and Paraganglioma

Matthew D. Wilkerson1, Katherine L. Nathanson2, Karel Pacak3, The Cancer Genome Atlas Pheochromocytoma and Paraganglioma Analysis Working Group 1Lineberger

Comprehensive Cancer Center, Department of Genetics, The University of North Carolina at Chapel Hill, Chapel Hill, North Carolina; 2Abramson Cancer Center, Department of Medicine, Division of Translational Medicine and Human Genetics, University of Pennsylvania, Philadelphia, Pennsylvania; 3National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, Maryland In recent years, we have seen important advances in understanding the molecular basis of pheochromocytoma and paraganglioma (PCC/PGL), particular in relationship to inherited susceptibility. Nevertheless, specific understanding of the somatic genomic alterations, in hereditary and non-hereditary PCC/PGL including malignant ones, is still very limited. The Cancer Genome Atlas (TCGA) consortium has conducted a coordinated effort to characterize the molecular basis of PCC/PGL and have collected a cohort of 184 PCC/PGL (excluding head and neck PGL). At least 30% of patients were attributed with a germline mutation in a known familial PCC/PGL susceptibility gene, thus making PCC/PGL the tumor type with the greatest rate of germline mutations in the TCGA. Some of the familial susceptibility genes (NF1, RET, and VHL) were also observed to be somatically mutated and in similar expression subtypes. Most notably, we have identified the first RNA fusion genes (and several species of fusion genes) in PCC/PGL, demonstrating for the first time that inter-chromosomal translocation and gene fusion is a method of molecular pathogenesis in this disease. In particular, we report the novel fusion genes UBTF-MAML3 and TCF4MAML3, which are activating based on over-expression properties, are recurrent (5% of cases), found in tumors that had no other driving event, occur in exactly one gene expression subtype, and associate with poor patient outcome. We identified new somatically mutated driver genes, such as CSDE1 which is also coordinated with transcript splicing alterations. Lastly, our integration of these abundant germline and somatic alterations across mRNA, miRNA and methylation platforms has enabled a characterization of vastly divergent patterns of molecular pathogenesis in PCC/PGL.

-6-

58.

A Multi-Cancer Gene Signature Associated With Stromal Activation

Zhenqiu Liu1, Ann E. Walts2, Beth Y. Karlan3,4, Sandra Orsulic3,4 1Biostatistics

and Bioinformatics Research Center, 2Department of Pathology and Laboratory Medicine, 3Women’s Cancer Program, Samuel Oschin Comprehensive Cancer Institute, CedarsSinai Medical Center, Los Angeles, California; 4Department of Obstetrics and Gynecology, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, California The presence of cancer cells induces a reaction in the surrounding stroma similar to fibrosis and wound healing after injury. Activated stroma (also known as reactive or desmoplastic stroma) is morphologically different from diseasefree stroma and is characterized by the increased presence of activated myofibroblasts and altered extracellular matrix. Despite recent advances toward understanding the key roles of activated stroma in the initiation, progression and recurrence of different cancers, it is currently unknown if different cancer types share a common pattern of stromal cell recruitment and activation across their respective microenvironments. Previously, we identified a TGFβregulated collagen remodeling stromal gene signature associated with metastasis, recurrence and poor survival in ovarian cancer and showed that inhibition of one of the signature genes, COL11A1, is effective in reducing ovarian cancer growth and dissemination in a mouse model. Using TCGA datasets from multiple primary cancer types, we now show that a highly conserved set of COL11A1 co-expressed genes is present in all epithelial cancers examined, including cancers of the breast, ovary, lung, pancreas, stomach, urinary bladder, colon, thyroid, cervix, head and neck, and prostate, but not in non-epithelial malignancies such as leukemias, gliomas and sarcomas. In any given epithelial cancer, this gene signature is typically associated with the mesenchymal molecular subtype which often has the worst prognosis of the molecular subtypes of that cancer. Although many of the individual signature genes are present in myofibroblasts and mesenchymal stem cells and are enriched in benign processes such as wound healing, fibrosis and tissue homeostasis, others are cancer-specific and could serve as biomarkers of malignant stromal activation and/or highly selective therapeutic targets within the tumor microenvironment.

-7-

104.

Integrated Analysis of TCGA Identifies Targets and Patient Populations for Antibody-Drug Conjugates

Wenyan Zhong1, Keith Ching2, Tao Xie2, Jeremy Myers1, Marc Damelin1, Puja Sapra1, Kim Arndt1, Jadwiga R. Bienkowska2, Paul A. Rejto2 1Oncology

Research, Pfizer Worldwide R&D, Pearl River, New York; 2Oncology Research, Pfizer Worldwide R&D, La Jolla, California Antibody-drug conjugates (ADCs) are a promising class of therapeutics for the treatment of cancer. The strategy selectively targets tumor cells by attaching cytotoxic agents to an antibody that recognizes antigen preferentially expressed on the cell surface of tumor tissues compared to normal tissues. We describe an integrated computational approach for ADC target identification and patient selection employing The Cancer Genome Atlas (TCGA) and GeneTissue Expression (GTEx) data. Scores were defined for each target to capture its expression in normal tissues (NormScore) and tumor tissues (TumorScore). NormScore was calculated as the number of normal tissues in which the median target expression level is above defined expression threshold in that tissue. TumorScore is calculated as the rank product of differential expression between tumor and normal tissue (fold change) and prevalence (% samples above defined expression threshold in tumor tissues). We then developed a customized target selection criteria tailored to various type of payload class using these scores along with other factors such as cell surface expression. For example, Microtubule inhibitor payloads may require higher target expression than DNA damage payloads due to differences in potency. Our approach successfully predicted ADC targets currently in clinical trials (eg. HER2, GPNMB, and STEAP1) in addition to novel ADC targets. While IHC based target expression measurement is the primary biomarker used for patient selection for ADC therapeutics, RNASeq from large tumor panels in TCGA is an excellent data resource for estimating the prevalence of target expression and predicting patient populations to guide clinical biomarker assay development and clinical trial design. These examples demonstrate how Pfizer incorporates integrated TCGA analysis from target identification through to patient selection.

-8-

57.

Mutation Hotspots Associate with Gene Expression, Signaling Pathways, Protein Domains, and Drug Response

William Poole, Theo Knijnenburg, Brady Bernard, Ilya Shmulevich Institute for Systems Biology, Seattle, Washington Overview: The distribution of mutations in cancer genes is nonrandom. Oncogenes are recurrently mutated at the same amino acid positions. Tumor suppressor genes, on the other hand, often have protein-truncating alterations that form non-uniform patterns of mutations. The TCGA provides a significant opportunity to identify these hotspots of mutations in both well-known and novel cancer genes. Statistical associations with molecular data, such as gene expression, protein expression and drug response, can elucidate the functional consequences of the mutation hotspots. Approach: We have developed a novel multiscale clustering algorithm that uses gene mutation data to detect ‘hotspots’ of single-nucleotide protein-affecting mutations. The approach begins by fitting multiple mixture models each representing a different length scale. These multiscale models are combined using a greedy algorithm, which aims to find the set of non-overlapping clusters that minimize the Akaike information criterion. This methodology allows for the discovery of mutation hotspots of widely differing sizes; clusters range from individual to hundreds of amino acids. We have applied this approach to all mutation data in 11 cancers from TCGA. We performed a largescale statistical analysis to associate these mutation hotspots with tens of thousands of molecular features in TCGA, including gene, protein (RPPA) and clinical phenotypes. Additionally, we employed annotations of protein domains and drug response measurements in cancer cell lines to further establish the functional importance of the mutation hotspots. Results: The uncovered mutation hotspots led to a novel ranking of genes, which is different from gene rankings based on commonly employed methods for identification of significantly mutated genes. Mutation hotspots are significantly enriched with annotated protein domains. Additionally, our approach highlighted specific regions of interest in genes, as these regions had a stronger statistical association with pathway-level expression signatures when compared to mutations found across the entire gene. We interpret these findings as the identification of functional regions within these genes. This hypothesis is further supported by the observation that sensitivity to anticancer drugs in cell lines is better explained by mutation hotspots versus mutations in the entire gene.

-9-

4.

Integrated Molecular Characterization of Uterine Carcinosarcoma

Co-Chairs: Rehan Akbani, Douglas A. Levine The Cancer Genome Atlas Research Network, UCS group Uterine carcinosarcoma (UCS) is a rare tumor that is found in less than 5% of all uterine cancers. The median age of patients affected by this disease is 65 years and patients often present with vaginal bleeding. The overall 5-year survival rate for UCS is approximately 35% with a median overall survival of approximately 24 months, which is much worse than endometrial carcinoma (UCEC) with a median overall survival of more than 60 months. UCS is an aggressive disease and TCGA’s mission is to better understand the molecular characteristics of this disease, with an overarching goal to improve treatment options. We performed an integrated genomic, epigenomic, transcriptomic, and proteomic characterization of 57 uterine carcinosarcomas (UCSs) using array- and sequencing-based technologies. UCSs have extensive copy number alterations, poor unsupervised clustering, and highly recurrent somatic mutations. Nearly all (91%) cases had TP53 mutations and frequent mutations were also found in PTEN, PIK3CA, PPP2R1A, FBXW7, and KRAS. Transcriptome sequencing identified a strong EMT gene signature in subset of 17 (30%) cases. Corresponding decreases in mir200 family expression were apparent in cases with EMT signatures and were generally under epigenetic control. UCS had the largest range of EMT signature among the different tumor types studied and shared proteomic features with both, gynecological and non-epithelial tumors. Our results indicate that UCS tumors share many features with serous-like endometrial carcinomas including frequent TP53 mutations and extensive somatic copy number alterations, though with greater EMT features. Despite having mixed histology, the tumors demonstrated similar clonality to other common solid tumors suggesting a homogeneous cellular population at the molecular level. These data taken together suggest that some UCS tumors develop from an endometrioid lineage, though most are thought to de-differentiate from a serous precursor accounting for their clinical aggressiveness and poor response to treatment. Multiple somatic mutations and copy number alterations in genes that are therapeutic targets have been identified.

- 10 -

31.

Characterization of Tumor-Infiltrating T-lymphocytes in TCGA Cancers

Linghua Wang, Liu Xi, Kyle Covington, Richard A. Gibbs, David A. Wheeler Human Genome Sequencing Center, Baylor College of Medicine, Houston, Texas Tumor-infiltrating lymphocytes (TILs) are a type of white blood cells recruited into the tumor in an attempt to kill the tumor cells. TILs reflect the host’s anti-tumor immune response and it has been demonstrated in multiple tumor types that TILs can be a valuable predictor of prognosis. The use of TILs as an adoptive cell transfer therapy has recently shown great promise in the treatment of human cancers including metastatic melanoma and colorectal cancer. To date, however, many properties of TILs are not fully understood. To gain a better understanding of the anti-tumor immune response, we developed a pipeline to characterize expression of T-cell receptor (TCR) repertoire and other T-cell makers in patients’ tumors using the RNA-seq data. We have applied this approach to cutaneous squamous cell carcinomas and the TCGA colorectal cancer samples. The expression of TCR-β and TCR-α genes in the variable region, joining region, and constant region was evident in over a third of patients, allowing further assessment of the state of TCR repertoire. Further analysis of the sequencing reads aligned at the breakpoints of TCR rearrangement suggested a polyclonal profile. Compared to the TCR repertoire of normal lymphocytes, the repertoire was limited in some of the patients with depletion of a portion of the TCR repertoire, suggesting the possibility a specific response to tumor antigen was being mounted in those patients. We are developing methods to quantify the levels of TILs with the expression data as part of the TCGA pan-can project. In summary, our preliminary results revealed novel and important details about the biology of anti-tumor T-cell responses.

- 11 -

43.

Global Analysis of Somatic Structural Alterations and Their Impact on Gene Expression in Diverse Human Cancers

Babak Alaeimahabadi, Erik Larsson Institute of Biomedicine, The Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden Tumor genomes are mosaics of somatic structural variations (SVs) that may contribute to the activation of oncogenes or inactivation of tumor suppressors, for example, by altering gene copy-number. However, while gene copy-number variation accounts for some of transcriptional variability seen between tumors, most mRNA changes remain unexplained. Notably, there are multiple other ways in which SVs can alter transcription, but the overall impact of these mechanisms on tumor transcriptional output has not been systematically studied. Moreover, the overall structural basis of copy-number changes in cancer is poorly known. Here, we use whole-genome sequencing (WGS) data from TCGA to map SVs across >500 tumors and >15 cancers, and investigate the relationship between SVs, copy-number alterations, and mRNA levels. We use known chromosomal breakpoints from copy-number data to carefully evaluate and optimize tools and parameters for WGSbased SV detection. We find that ~20% of copy number alterations can be clarified structurally and that most copynumber amplifications are due to tandem duplications. We also find that some seemingly simple copy-number alterations have a more complex structural basis involving composite events on different chromosomes. We observe frequent swapping of strong and weak promoters in the context of gene fusions, and find that these events have a measurable global impact on mRNA levels. Notably, many of these fusions are due to short-range events, visible in copy-number data as small copy number segments (10%) but are virtually never observed in normal samples or ENCODE cell lines ( 100 events) and one with less (p