Epigenomic Promoter Alterations Amplify Gene Isoform and Immunogenic Diversity in Gastric Adenocarcinoma

Published OnlineFirst March 20, 2017; DOI: 10.1158/2159-8290.CD-16-1022

RESEARCH ARTICLE

Aditi Qamra1,2, Manjie Xing3,4, Nisha Padmanabhan3, Jeffrey Jun Ting Kwok5, Shenli Zhang3, Chang Xu3, Yan Shan Leong6, Ai Ping Lee Lim1, Qianqao Tang7, Wen Fong Ooi1, Joyce Suling Lin1, Tannistha Nandi1, Xiaosai Yao1, Xuewen Ong3, Minghui Lee3, Su Ting Tay3, Angie Tan Lay Keng3, Erna Gondo Santoso7, Cedric Chuan Young Ng7, Alvin Ng3,4, Apinya Jusakul3, Duane Smoot8, Hassan Ashktorab9, Sun Young Rha10, Khay Guan Yeoh11,12, Wei Peng Yong13, Pierce K.H. Chow14,15, Weng Hoong Chan16, Hock Soo Ong16, Khee Chee Soo15, Kyoung-Mee Kim17, Wai Keong Wong16, Steven G. Rozen3,18, Bin Tean Teh3,6,7,18, Dennis Kappei6, Jeeyun Lee19, John Connolly5,20, and Patrick Tan1,3,6,18,21

ABSTRACT Promoter elements play important roles in isoform and cell type–specific expression. We surveyed the epigenomic promoter landscape of gastric adenocarcinoma, analyzing 110 chromatin profiles (H3K4me3, H3K4me1, H3K27ac) of primary gastric cancers, gastric cancer lines, and nonmalignant gastric tissues. We identified nearly 2,000 promoter alterations (somatic promoters), many deregulated in various epithelial malignancies and mapping frequently to alternative promoters within the same gene, generating potential pro-oncogenic isoforms (RASA3). Somatic promoter– associated N-terminal peptides displaying relative depletion in tumors exhibited high-affinity MHC binding predictions and elicited potent T-cell responses in vitro, suggesting a mechanism for reducing tumor antigenicity. In multiple patient cohorts, gastric cancers with high somatic promoter usage also displayed reduced T-cell cytolytic marker expression. Somatic promoters are enriched in PRC2 occupancy, display sensitivity to EZH2 therapeutic inhibition, and are associated with novel cancer-associated transcripts. By generating tumor-specific isoforms and decreasing tumor antigenicity, epigenomic promoter alterations may thus drive intrinsic tumorigenesis and also allow nascent cancers to evade host immunity.

SIGNIFICANCE: We apply epigenomic profiling to demarcate the promoter landscape of gastriccancer . Many tumor-specific promoters activate different promoters in the same gene, some generating pro-oncogenic isoforms. Tumor-specific promoters also reduce tumor antigenicity by causing relative depletion of immunogenic peptides, contributing to cancer immunoediting and allowing tumors to evade host immune attack. Cancer Discov; 7(6); 1–22. ©2017 AACR.

1Cancer Therapeutics and Stratified Oncology, Genome Institute of Singapore. 16Department of Upper Gastrointestinal & Bariatric Surgery, Singapore, Singapore. 2Department of Physiology, Yong Loo Lin School of Singapore General Hospital, Singapore. 17Department of Pathology & Trans- Medicine, National University of Singapore, Singapore. 3Cancer and Stem Cell lational Genomics, Samsung Medical Center, Sungkyunkwan University Biology Program, Duke-NUS Medical School, Singapore. 4NUS Graduate School of Medicine, Seoul, Korea. 18SingHealth/Duke-NUS Institute of Pre- School for Integrative Sciences and Engineering, National University of cision Medicine, National Heart Centre Singapore, Singapore. 19Department Singapore, Singapore. 5Institute of Molecular and Cell Biology, Agency for of Medicine, Division of Hematology-Oncology, Samsung Medical Center, Science, Technology and Research, Singapore. 6Cancer Science Institute Sungkyunkwan University School of Medicine, Seoul, Korea. 20Institute of of Singapore, National University of Singapore, Singapore. 7Laboratory Biomedical Studies, Baylor University, Waco, Texas. 21Cellular and Molecular of Cancer Epigenome, Department of Medical Sciences, National Cancer Research, National Cancer Centre, Singapore. 8 Centre, Singapore. Department of Internal Medicine, Meharry Medical Col- Note: Supplementary data for this article are available at Cancer Discovery 9 lege, Nashville, Tennessee. Department of Medicine, Howard University, Online (http://cancerdiscovery.aacrjournals.org/). Washington, DC. 10Department of Internal Medicine, Yonsei University College of Medicine, Seoul, South Korea. 11Department of Medicine, Yong A. Qamra and M. Xing contributed equally to this article. Loo Lin School of Medicine, National University of Singapore and National Corresponding Author: Patrick Tan, Duke-NUS Medical School, 8 College University Health System, Singapore. 12Department of Gastroenterology Road, Singapore 169857, Singapore. Phone: 65-6516-1783; Fax: 65-6221- & Hepatology, National University Hospital, Singapore. 13Department of 2402; E-mail: [email protected] Haematology-Oncology, National University Hospital of Singapore, Singa- doi: 10.1158/2159-8290.CD-16-1022 pore. 14Department of General Surgery, Singapore General Hospital, Singa- pore. 15Division of Surgical Oncology, National Cancer Centre Singapore, ©2017 American Association for Cancer Research.

OF1 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

INTRODUCTION of biological, functional, and regulatory diversity, as current estimates suggest that 30% to 50% of genes in the human Gastric cancer is the third leading cause of global cancer genome are associated with multiple promoters (15), which mortality with high prevalence in many East Asian countries can be selectively activated as a function of developmental lin- (1). Patients with gastric cancer often present with late-stage eage and cellular state (16). Differential usage of alternative disease (2, 3), and clinical management remains challenging, promoters causes the generation of distinct 5′ untranslated as exemplified by several recent negative phase II and phase regions (5′ UTR) and first exons in transcripts, which in turn III clinical trials (4–7). At the molecular level, studies have can influence mRNA expression levels (17), translational identified characteristic gene mutations (8, 9), copy-number efficiencies (18, 19), and the generation of different protein alterations, gene fusions (10), and transcriptional patterns in isoforms through gain and loss of 5′ coding domains (15, 20). gastric cancer (11, 12). However, few of these have been clini- In cancer, alternative promoters in genes such as ALK (21), cally translated into targeted therapies, with the exception of TP53 (22), LEF1 (23), and CYP19A1 (24) have been reported, HER2-positive gastric cancer and trastuzumab (13). There is producing cancer-specific isoform variants with oncogenic thus a strong need for additional and more comprehensive properties. To date, promoter alterations in cancer have been explorations of gastric cancer, as these may highlight new bio- largely studied on a gene-by-gene basis, and very little is markers for disease detection, predicting patient prognosis or known about the global extent of promoter-level diversity in responses to therapy, as well as new therapeutic modalities. gastric cancer and other solid malignancies. Promoter elements are cis-regulatory elements that func- Promoters in the genome can be experimentally identified tion to link gene transcription initiation to upstream reg- by various methods. Broadly divided into RNA-based or epi- ulatory stimuli, integrating inputs from diverse signaling genomic approaches, the former involves technologies such pathways (14). Promoters represent an important reservoir as RNA sequencing (RNA-seq), CAP analysis gene expression

JUNE 2017 CANCER DISCOVERY | OF2

RESEARCH ARTICLE Qamra et al.

(CAGE), and global run-on sequencing (25–27). For the latter, To enable accurate promoter identification, we inte- active promoters have been shown to exhibit characteristic grated data from multiple histone modifications, selecting chromatin modifications, specifically H3K4me3 positivity, H3K4me3 regions simultaneously codepleted for H3K4me1 H3K27ac positivity, and H3K4me1 depletion (28–31). Com- (“H3K4me3hi/H3K4me1lo regions”; Supplementary Fig. S1; pared with transcriptome sequencing (25), using histone Methods; ref. 42). Comparisons against data from exter- modifications to identify promoters carries certain advan- nal sources, including GENCODE reference transcripts, tages. First, epigenome-guided promoter identification allows ENCODE chromatin state models, and CAGE databases, genomic localization of the promoter element itself, rather validated the vast majority of H3K4me3hi/H3K4me1lo regions than the ensuing transcript product. Second, particularly for as true promoter elements (Supplementary Text; Supplemen- clinical samples, epigenome-guided promoter identification tary Fig. S1). Because primary gastric tissues comprise several is less prone to transcript degradation artifacts caused by 5′ different tissue types, including epithelial cells, immune cells, RNA exonucleases (32). Epigenome-marked promoters may and stroma, we further confirmed that our promoter profiles also highlight transcript classes not easily detectible by other were reflective of bona fide gastric epithelia by comparison means, such as promoters originating via recapping events, against Epigenome Roadmap data for gastric and nongastric short-lived RNAs (27), or unstable RNAs with greater sensi- tissues. Gastric tumor and matched normal promoter pro- tivity to exosome-mediated decay, such as promoter upstream files exhibited the highest correlations to Roadmap gastric transcripts (33) and/or enhancer RNAs (34–36). mucosae and were distinct from other gastrointestinal tissues In this work, we analyzed a gastric cancer cohort of pri- (small intestine, colon mucosa, colon sigmoid), stomach- mary samples and cell lines to survey the landscape of altered associated muscle, skin, and blood (CD14; Supplementary promoter elements in gastric cancer. Our study applied Fig. S2). Primary tissue promoter profiles also showed a sig- microscale histone modification profiling [Nano–chroma- nificant overlap with promoter profiles of gastric cancer cell tin immunoprecipitation sequencing (Nano-ChIP-seq)] to lines (87%), which are purely epithelial in origin, compared profile primary cancers (37, 38), which allows the measure- with gastrointestinal fibroblast lines (58%–69%), and colon ment of epigenomic modifications in vivo, compared with carcinoma lines (59%–74%; Supplementary Fig. S2). laboratory-cultured cell lines that may harbor epigenomic In total, we mapped approximately 23,000 promoter ele- artifacts due to in vitro culture (39, 40). By comparing the ments in the Nano-ChIP-seq cohort. Visual exploration of epigenomic promoter profiles of primary gastric cancers and these promoter elements identified three main promoter cat- matched normal tissues, we observed pervasive alterations in egories: unaltered promoters, promoters gained in tumors promoter usage in gastric cancer, and an important role for (gained somatic or tumor-specific promoters), and promot- somatic promoters in enhancing gastric cancer transcript, ers present in normal gastric tissues but lost or decreased in protein isoform, and immunogenic diversity. We also found gastric cancer (lost somatic or normal-specific promoters; that many somatic promoters observed in gastric cancer are Fig. 1A–C). Representative examples of unaltered promoters deregulated in other cancer types, supporting a generalized included RHOA (Fig. 1A), whereas CEACAM6, an intracel- role for somatic promoters in solid malignancies. To our lular adhesion gene, exhibited somatic promoter gain at the knowledge, our study represents one of the largest and most CEACAM6 transcription start site (TSS) in tumor samples comprehensive surveys of somatic promoters in any single and cell lines (Fig. 1B). Conversely, ATP4A, a parietal cell– tumor type. associated H+/K+ ATPase with decreased expression in gastric cancer (43), exhibited somatic promoter loss (Fig. 1C). RESULTS Both CEACAM6 and ATP4A promoter alterations were correlated with increased and decreased CEACAM6 and ATP4A Identifying Epigenomic Promoter Alterations in gene expression in the same samples, respectively (Fig. 1B Gastric Cancer and C). Using Nano-ChIP-seq (37), we profiled three histone modi- Previous studies have established distinct molecular sub- fication marks (H3K4me3, H3K27ac, and H3K4me1) across types of gastric cancer (11, 12, 44). Because of limited sample 17 gastric cancers, matched normal gastric mucosae (34 sizes, however, we elected in the current study to identify pro- samples), and 13 gastric cancer cell lines, generating 110 epi- moter alterations (“somatic promoters”; ref. 45–47) present genomic profiles (Supplementary Tables S1 and S2 provide in multiple gastric cancer tissues relative to control tissues clinical and sequencing metrics; Fig. 1A). Quality control of irrespective of subtype. Focusing on recurrent alterations the Nano-ChIP-seq data was performed using two independ- also has the benefit of reducing potential artifacts due to ent methods: ChIP enrichment at known promoters and “private” epigenomic variation or individual sample-specific employing the ChIP-seq quality control and validation tool technical errors. Using two complementary read count– ChIP-seq analytics and confidence estimation (CHANCE; ref. based algorithms commonly used for analysis of ChIP-seq 41). Comparisons of Nano-ChIP-seq read densities at 1,000 data (48–50), we identified approximately 2,000 highly recur- promoters associated with highly expressed protein-coding rent somatic promoters, of which 75% were gained in gastric genes confirmed successful enrichment in all H3K27ac and cancers [fold change (FC) ≥ 1.5, q < 0.1]. Two-dimensional H3K4me3 libraries. CHANCE analysis also revealed that the heat map clustering and principal component analysis (PCA) large majority (81%) of samples exhibited successful enrich- plots based on somatic promoters confirmed a separation ment (Supplementary Table S2). We have previously also of gastric cancer samples from normal samples (Fig. 1D; shown that Nano-ChIP signals exhibit a good concordance Supplementary Fig. S3). Somatic promoter H3K4me3 levels with orthogonal ChIP-qPCR results (38). were also highly correlated with H3K27ac signals, commonly

OF3 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE

A Unaltered promoter (RHOA) D Somatic promoter alterations USP4 NICN1 70 GC

70 Normal

H3K4me3 175 GC line Gain (75%)

2,198 GC 4 2,198 RNA Normal 0 Loss chr3:49,343,632-49,502,474 50 Kb 4 Normal GC BE Somatic gain (CEACAM6) 0 CEACAM5 Somatic promoters Unaltered promoters

77 GC 51

77 Normal

231 H3K4me3 GC line − 50 5,000 GC

Activity mark [Log2(H3K27ac)] r = 0.91, P < 0.001

RNA 5,000 Normal − 10 10 50 510 chr19:42,197,907-42,284,660 25 Kb − − Promoter mark [Log2 (H3K4me3)] C F Somatic loss (ATP4A) H3K27me3 HCP in brain TMEM147 H3K27me3 targets in ESCs 68 GC Up in early GC Gain SUZ12 targets in ESCs 68 Normal Up in advanced GC

H3K4me3 282 GC line Down in early GC Up in nasopharyngeal carcinoma 6,309 GC Up in prostate development Loss RNA 6,309 Normal Down in advanced GC SUZ12 targets in ESCs chr19:36,023,273-36,060,416 10 Kb 040 80 Negative log 10 P value (FDR < 0.05)

Figure 1. Somatic promoters in primary gastric adenocarcinoma. A, Example of an unaltered gastric cancer (GC) promoter. The UCSC genome track of the RHOA TSS (shaded box) highlights similar H3K4me3 signals in gastric cancer and matched normal samples. Similar signals are seen in gastric cancer lines. The bottom two tracks display similar levels of RNA expression in the same gastric cancer and matched normal samples (RNA-seq). B, Example of a gained somatic promoter. The UCSC genome track of the CEACAM6 TSS (shaded box) highlights gain of H3K4me3 signals in gastric cancer samples and gastric cancer lines, compared with matched normal samples. In contrast, no changes are observed at the TSS of CEACAM5, an adjacent gene. Concordant tumor-specific gain of RNA expression is shown in the bottom two tracks displaying RNA-seq profiles of the same gastric cancer and matched normal samples. C, Example of a lost somatic promoter. The UCSC genome track of the ATP4A TSS (shaded box) highlights loss of H3K4me3 signals in gastric cancer samples and gastric cancer lines compared with matched normal samples. Concordant tumor-specific loss of RNA expression is shown in the bottom two tracks displaying RNA-seq profiles of the same gastric cancer and matched normal samples. D, Heat map of H3K4me3 read densities (row scaled) of somatic promoters (rows) in primary gastric cancer and matched normal samples. E, Correlation between H3K4me3 promoter signals and H3K27ac activity signals in primary gastric samples (r = 0.91, P < 0.001). Each data point corresponds to a single H3K4me3hi/H3K4me1lo region. Gold points, all promoters; blue points, somatic promoters. Analysis was performed using data from 16 N/T pairs (Supplementary Table S2). F, Top five gene sets associated with canonical gained and lost somatic promoters. Gene sets associated with genes upregulated and downregulated in gastric cancer are rediscovered. Also note that gene sets related to H3K27me3 and SUZ12, a PRC2 component, are enriched. ESC, embryonic stem cell; HCP, high-CpG promoter.

JUNE 2017 CANCER DISCOVERY | OF4

RESEARCH ARTICLE Qamra et al. regarded as a marker of active regulatory activity (42, 51). Atlas (TCGA) consortium (n = 321 gastric cancer samples, This correlation was observed across all somatic promoters n = 33 matched normals). To perform this analysis, RNA- (r = 0.84, P < 0.001, Fig. 1E), and also when gained somatic seq reads from TCGA samples were mapped against the promoters and lost somatic promoters were analyzed sepa- epigenome-guided somatic promoter regions defined by the rately (r = 0.78, P < 0.001 for gained somatic; r = 0.82, P < discovery samples and normalized to calculate fold-change 0.001 for lost somatic, Supplementary Fig. S3). Pathway differences in expression between gastric cancer samples analysis revealed that both gained somatic promoters and versus normal samples (see Methods). Similar to the discov- lost somatic promoters were significantly associated with ery series, we observed that TCGA gastric cancer samples expression gene sets previously reported to be upregulated also exhibited significantly increased expression at gained and downregulated in gastric cancer, respectively (Fig. 1F). somatic promoters, whereas lost somatic promoters exhib- These included upregulated oncogenes (MET, ABL2), cell ited decreased expression, relative to either all promoters adhesion genes (CEACAM6), and claudin family members (P < 0.001; Fig. 2C) or unaltered promoters (P < 0.001; Sup- (CLDN7, CLDN3). Fifteen percent to 18% of somatic promot- plementary Fig. S4). We further tested the tissue specificity ers mapped to noncoding RNAs, including HOTAIR and of the gastric cancer somatic promoters by querying RNA- PVT1, previously associated with gastric cancer (Supplemen- seq data from other tumor types, including colon cancer, tary Table S3; refs. 52, 53). Additional analyses at increasing kidney renal clear cell carcinoma (ccRCC), and lung adeno- thresholds of stringency (FC from 1.5 to 2 and FDR from 0.1 carcinoma (LUAD; Fig. 2D). Almost two thirds (n = 1,231, to 0.001) yielded similar results, supporting the robustness 63%, FC ≥ 1.5) of gastric cancer somatic promoters were also of this analysis (Supplementary Fig. S3). These results dem- differentially regulated in TCGA colon cancer samples, and onstrate that normal gastric epithelia and gastric cancers similarly, a significant proportion of gastric cancer somatic can be distinguished on the basis of epigenomic promoter promoters were also associated with differential RNA-seq profiles. expression in TCGA ccRCC (n = 939, 48%, FC ≥ 1.5) and LUAD samples (n = 1,059, 54%, FC ≥ 1.5; Fig. 2D). This Somatic Promoters in Gastric Cancer Exhibit result suggests that many gastric cancer somatic promoters Deregulation in Diverse Cancer Types are also likely associated with deregulated promoter activity To explore relationships between epigenomic promoter in other solid epithelial malignancies. alterations and gene expression, we analyzed RNA-seq data from the same discovery cohort (∼106 million reads/sample), Role of Alternative Promoters quantifying RNA-seq transcript reads mapping to the epi- By comparing the somatic promoters against the reference genome-guided promoter regions or directly downstream. GENCODE database (V19; ref. 57), we discovered extensive Examining somatic promoter regions (Fig. 2A provides use of alternative promoters (18%) in gastric cancers, defined an illustrative example of a gained somatic promoter), we as situations where a common unaltered promoter is present observed significantly increased expression at gained somatic in both normal tissues and tumors (canonical promoter) promoters in gastric cancers and significantly decreased but a secondary tumor-specific promoter is engaged in the expression at lost somatic promoters, compared with either latter (alternative promoter). The remaining 82% of somatic all promoters (P < 0.001; Fig. 2B), or unaltered promoters promoters corresponded to single major isoforms or unan- (P < 0.001; Supplementary Fig. S4; refs. 27, 51, 54). Among notated transcripts (see below). Fifty-seven percent of the other types of epigenetic modifications, previous studies have alternative promoters occurred downstream of the canonical also reported a reciprocal relationship between active regula- promoter. Using multiple RNA-seq analysis methods, we tory regions and DNA methylation (54, 55). Using Infinium confirmed that transcript isoforms driven by alternative pro- 450K DNA methylation arrays, we identified 7,505 CpG sites moters are overexpressed in gastric cancers to a significantly overlapping the somatic promoter regions (5,213 sites for greater degree than canonical promoters in the same gene gained somatic promoters; 2,292 sites for lost somatic pro- (Methods; Supplementary Fig. S6). For example, HNF4A, a moters). Promoters gained in gastric cancer were significantly transcription factor overexpressed in gastric cancer (58–61), is hypomethylated compared with all promoters (P < 0.001, driven by two promoters (P1 and P2). At the HNF4A canoni- Wilcoxon test), whereas promoters lost in gastric cancer were cal promoter (“P2”), we observed equal promoter signals in hypermethylated (P < 0.001, Wilcoxon test; Fig. 2B, bottom). gastric cancer tissues and normal tissues; however, we also As DNA methylation typically occurs in CpG-rich regions further observed gain of an additional promoter in gastric (56), we then repeated the analysis focusing only on CpG cancers at a TSS 45 kb downstream (“P1”). Similarly, HNF4A island–bearing promoters (Methods). Similar to the orig- P1 promoter gains were also observed in gastric cancer cell inal results, CpG island–bearing promoters gained in gastric lines (Fig. 3A), with RNA-seq analysis supporting expression cancer were significantly hypomethylated compared with all of the HNF4A P1 isoform in gastric cancers. Alternative pro- CpG island–bearing promoters (P < 0.001, Wilcoxon test), moter usage was also observed at the EPCAM gene, frequently whereas CpG island–bearing promoters lost in gastric cancer used to identify circulating tumor cells, causing expression of were hypermethylated (P < 0.001, Wilcoxon test; Supplemen- EPCAM transcript ENST00000263735.4 (Fig. 3B). Notably, tary Fig. S5). both the HNF4A and EPCAM alternative isoforms exhibited To validate the somatic promoters in a larger independ- significantly greater cancer overexpression compared with ent gastric cancer cohort and also to examine their behavior their canonical isoforms (Supplementary Fig. S6). Other in other cancer types, we proceeded to query RNA-seq data genes associated with tumor-specific alternative promoters, of 354 gastric cancer samples from the The Cancer Genome many reported for the first time, includeNKX6-3 (FC = 1.83,

OF5 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE

Somatic TSS Gastric cancer Gastric normal (For illustrative purposes) signal H3K4me3

Discovery samples Validation cohorts (n = 967, RNA-seq)

BCRNA-seq GC - TCGA D Colon cancer (n = 9 pairs) (n = 354) (n = 326) *** *** *** *** *** *** *** *** *** 8

6 6 6

4 4 4

2 2 2

0 0 RPKM log2 (T/N) RPKM log2 (T/N) 0 RPKM log2 (T/N) −2 −2 −2 −4 −4 −4

All promoters All promoters All promoters

Somatic gainSomatic in GC loss in GC Somatic gainSomatic in GC loss in GC Somatic gainSomatic in GC loss in GC

DNA methylation Kidney cancer Lung cancer (n = 20 pairs) (n = 170) (n = 115)

0.4 *** *** NS *** *** *** *** *** *** 4 0.2 6 2 4 0.0 0 2

T/N β -value −0.2

RPKM log2 (T/N) RPKM log2 (T/N) 0 −2 −2 −0.4 −4 −4

All promoters All promoters All promoters

Somatic gainSomatic in GC loss in GC Somatic gainSomatic in GC loss in GC Somatic gainSomatic in GC loss in GC

*** P < 0.001

Figure 2. Association of somatic promoters with gene expression in gastric cancer (GC) and other tumor types. A, Example of a gastric cancer somatic promoter (red, H3K4me3 signal in gastric cancer; blue, H3K4me3 signal in gastric normal). Example is for illustrative purposes only. B, Changes in RNA- seq expression (top) and DNA methylation (bottom) in discovery samples between somatic promoters and all promoters. Top, box plot depicting changes in RNA-seq expression between 9 paired primary gastric cancer and gastric normal samples at genomic regions exhibiting somatic promoters (gained and lost; ***, P < 0.001, Wilcoxon test). Bottom, box plot depicting changes in DNA methylation (β-values) at regions exhibiting somatic promoters between 20 paired gastric cancer and gastric normal samples, compared with all promoters (***, P < 0.001, Wilcoxon test). C, Independent validation cohorts. Box plot depicting changes in RNA-seq expression at genomic regions exhibiting somatic promoters across 354 (321 gastric cancer, 33 normal) TCGA stomach adenocarcinoma (STAD) samples, compared with all promoters (***, P < 0.001, Wilcoxon test). D, Somatic promoters in other cancer types. Box plot depicting changes in RNA-seq expression at genomic regions exhibiting gastric cancer somatic promoters compared against all promoters, across 326 TCGA colon adenocarcinoma (COAD) samples (286 COAD, 40 normal; ***, P < 0.001, Wilcoxon test), 170 TCGA kidney ccRCC samples (98 ccRCC and 72 normal; ***, P < 0.001, Wilcoxon test), and 115 TCGA LUAD samples (58 LUAD, 57 normal; ***, P < 0.001 somatic gain vs. all promoters and somatic gain vs. somatic loss, Wilcoxon test). T/N, tumor/normal.

JUNE 2017 CANCER DISCOVERY | OF6

RESEARCH ARTICLE Qamra et al.

HNF4A ADCanonical TSS Somatic TSS

50 GC RASA3 (translated) 50 Normal C2 C2 RASGAP BHK 254 Canonical TSS H3K4me3 GC line N C

850 GC Somatic TSS N C

850 Normal PH BHK 530

RNA TCGA GC 530 TCGA GC normal chr20:42,977,507-43,066,933 20 Kb

B E EPCAM SNU1967 cells (GC) Canonical TSS Somatic TSS

** RASA3-Ctl NS 29 GC ** 6 29 Normal 110 H3K4me3 GC line RASA3-CanT 4

2,366 GC 2,366 2 Normal RASA3-SomT

RNA 2,409 TCGA GC membrane transwell

% Area of migrated cells/ 0 2,409 TCGA GC normal

chr2:47,557,600-47,627,802 20 Kb RASA3-Ctl RASA3-CanTRASA3-SomT

RASA3 C Somatic TSS GES1 cells (normal)

** 18 70 GC 16

/ RASA3-Ctl 70 Normal 14 NS

H3K4me3 270 GC line 12 10 420 GC RASA3-CanT 8 420 Normal 6 RNA 180 TCGA GC

transwell membrane transwell 4

180 TCGA GC normal % Area of migrated cells RASA3-SomT 2 chr13:114,764,810-114,771,899 2 Kb 0 ** P < 0.01 Primer 5’ RACE TSS GG GG TTAAGGTTCC T CC RASA3-Ctl RASA3-CanTRASA3-SomT

Figure 3. Alternative promoters in gastric cancer (GC). A, UCSC browser track of the HNF4A gene. Gastric cancer and matched gastric normal samples have equal H3K4me3 signals at the canonical HNF4A promoter. However, an alternative promoter, seen by H3K4me3 gain, can be observed at a downstream TSS in gastric cancers compared with matched normals. At the RNA level, both in-house and TCGA STAD samples also show gain of gene expression at the alternate promoter TSS compared with normal samples. B, UCSC browser track of the EPCAM gene. Another example of alternative promoter usage at a downstream TSS. Gain of H3K4me3 is observed at a TSS downstream of the canonical promoter, while the canonical promoter exhibits equal H3K4me3 signals in gastric cancer and gastric normal. Gain of RNA-seq expression can also be observed in gastric cancer at the alternative promoter– driven transcript in both in-house and TCGA STAD samples. C, UCSC browser track of the RASA3 gene, demonstrating H3K4me3 and RNA-seq signals highlighting gain of promoter activity at an unannotated TSS (dark gray box) corresponding to a novel N-terminal truncated RASA3 transcript. Expression of this variant transcript was validated through 5′ Rapid Amplification of cDNA Ends (RACE) in gastric cancer lines (bottom). D, Functional domains of the translated RASA3 canonical and alternate isoform. The alternate transcript is predicted to encode a RASA3 protein missing the RASGAP domain. E, Effect of overexpression of RASA3 canonical (CanT) and alternate (SomT) isoforms on the migration capability of SNU1967 (top) and GES1 (bottom) cells. Representative images of RASA3-Ctl (empty vector), RASA3-CanT, and RASA3-SomT in migration assays (n = 3). Bar plots show the percentage area of migrated cells versus the area of Transwell membrane. Data are shown as mean ± SD; n = 3 (**, P < 0.01; Student one sided t test).

OF7 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE q < 0.05) and GRIN2D (FC = 1.9, q < 0.001). A complete list lower levels of active GTP-bound RAS compared with either of gastric cancer tumor–specific promoters is provided in empty vector or RASA3 SomT transfected cells, indicating that Supplementary Table S4. RASA3 CanT has higher RASGAP activity (Supplementary To explore the influence of alternative promoters on protein Fig. S7). diversity, we identified 714 tumor-specific promoter alter- To address functions of RASA3 SomT, we transfected ations predicted to change N-terminal protein composition the RASA3 CanT and SomT isoforms into SNU1967 gastric and also supported by both H3K4me3 and RNA-seq data. cancer cells. Compared with untransfected cells, transfection The vast majority of these alterations (>95%) were in-frame of RASA3 SomT into SNU1967 cells significantly stimulated to that of the canonical protein. Of these, 47% (n = 338) were migration (P < 0.01) and invasion (P < 0.01), whereas RASA3 predicted to cause gains of new N-terminal peptides in tumors CanT significantly suppressed invasion P( < 0.001; Fig. 3E, (see Methods). To confirm protein-level expression of these Supplementary Fig. S7). Similarly, transfection of RASA3 N-terminal peptides in gastrointestinal cancer, we queried SomT into GES1 cells significantly stimulated migration P( < publicly available peptide spectral data of 90 TCGA colorectal 0.01, Fig. 3E) and invasion (P < 0.01, Supplementary Fig. S7), cancer and 60 normal colon samples (62, 63). Colorectal can- whereas RASA3 CanT did not. When tested on KRAS-mutated cer data were used for this analysis, as large-scale proteomic AGS gastric cancer cells that are innately highly migratory, data of primary gastric cancers are not currently available, expression of RASA3 CanT potently suppressed migration, and because many gastric cancer somatic promoters are also whereas RASA3 SomT exhibited significantly less attenua- observed in colorectal cancer (Fig. 2D). Across all proteins tion (P = 0.03, Supplementary Fig. S7). These results suggest detected, proteins with gained promoters were overexpressed that tumor-specific use ofRASA3 SomT is likely to increase in cancer to a significantly greater extent than proteins with- gastric cancer cell migration and invasion. Notably, RASA3 out gained promoters (P < 0.001; 63% vs. 54%; Fisher test). CanT and SomT transfections did not alter SNU1967, GES1, Then, examining specific N-terminal peptides predicted to or AGS cellular proliferation rates (Supplementary Fig. S7). be gained in tumors, we confirmed protein expression of 33% To confirm that these observations are not due to nonphysi- (112/338) in the colorectal cancer data (Supplementary Table ologic in vitro expression levels, we then examined NCC24 S5), of which 51.8% were overexpressed in colorectal cancer gastric cancer cells, which normally express high endogenous samples relative to normal colon samples (FDR, 10%). In a levels of RASA3 SomT and minimal RASA3 CanT (Supple- separate experiment, we further investigated whether these mentary Fig. S7). Silencing of endogenous RASA3 SomT N-terminal peptides also exhibit tumor overexpression in using two independent siRNA constructs significantly inhib- proteomic data from 3 gastric cancer cell lines and 1 normal ited NCC24 migration and invasion (P < 0.01–0.001; Supple- gastric epithelial line (GES1; see Methods). Similar to the colo- mentary Fig. S7), consistent with RASA3 SomT playing a role rectal cancer data, 48% of the N-terminal peptides were over- in promoting cancer migration and invasion. expressed in the gastric cancer lines relative to normal GES1 In an earlier study (38), we reported a transcript isoform of gastric cells, and again, a significantly greater proportion of the MET receptor tyrosine kinase driven by an internal alter- proteins with gained promoters were overexpressed in cancer native promoter, which has been independently confirmed compared with proteins without gained promoters (P < 0.001; in other cancer types (65). However, functional implications Fisher test). Taken collectively, these analyses suggest that of this MET variant (Var) remain unclear. RNA-seq and 5′ alternative promoters may contribute significantly toward RACE analysis confirmed transcript expression of this shorter proteomic diversity in gastrointestinal cancer. isoform, predicted to harbor a truncated SEMA domain (Sup- To examine possible functions of somatic promoters in can- plementary Fig. S8). To assess functional differences between cer development, we focused on RASA3, which encodes a RAS wild-type (WT) and Var MET, we performed transient trans-

GTPase-activating protein required for Gαi-induced inhibition fections of MET-WT and MET-Var into HEK293 cells. In both of MAPK (64). In both gastric cancers (50%) and gastric cancer untreated and HGF-treated conditions, MET-Var transfected lines, we observed gain of promoter activity at an intronic cells exhibited significantly higher levels of pGAB1 (Y627), region 127 kb downstream apart from the canonical RASA3 a key mediator of MET signaling [e.g., 2.48- to 3.95-fold TSS (Fig. 3C, top; Supplementary Fig. S7). RNA-seq and 5′ comparing MET-Var vs. MET-WT, P = 0.003 (untreated), P < Rapid Amplification of cDNA Ends (RACE) analysis con- 0.05 (T15 and T30); ref. 66]. In addition, in HGF-untreated firmed expression of this shorterRASA3 isoform (Fig. 3C, bot- samples, cells transfected with MET-Var also exhibited higher tom), and expression of this shorter RASA3 isoform was also pERK1/2 levels (2.74-fold) and higher pSTAT3 (Y705; refs. observed in TCGA RNA-seq data (Fig. 3C). Using isoform- 67–70) levels (1.80-fold) compared with MET-WT [P = 0.023 specific quantitative PCR, we confirmed that although both and P = 0.026 for pERK and pSTAT3 (Y705), respectively]. the canonical full-length RASA3 and shorter RASA3 isoform These results suggest that expression of the MET-Var isoform are overexpressed in gastric cancer tissues relative to matched may promote MET downstream signaling kinetics in a man- normal tissues (P < 0.01 Student one sided t test), the shorter ner important for gastric cancer tumorigenesis. RASA3 isoform is overexpressed to a significantly greater extent (FC = 2.64, P = 0.01, Student one sided t test; Supplementary Somatic Promoters Correlate with Fig. S7). Compared with the canonical full-length RASA3 Tumor Immunity protein (CanT), the shorter 31-kDa RASA3 somatic isoform Cancer immunoediting is a process where developing (SomT) is predicted to lack the N-terminal RASGAP domain tumors sculpt their immunogenic and antigenic profile to (Fig. 3D). Consistent with these predictions, transfection of evade host immune surveillance (71, 72). Mechanisms of RASA3 CanT into GES1 normal gastric epithelial cells induced cancer immunoediting are diverse, including upregulation

JUNE 2017 CANCER DISCOVERY | OF8

RESEARCH ARTICLE Qamra et al. of immune checkpoint inhibitors, such as PD-L1 (72). To alterations and intratumor T-cell activity in various primary explore potential contributions of somatic promoters to gastric cancer cohorts. First, to detect promoter alterations tumor immunity, we identified somatic promoter–associated in a cohort of 95 gastric cancer–normal pairs [Singapore (SG) N-terminal peptides with high predicted affinity binding to cohort], we generated a customized NanoString panel target- various MHC class I HLA alleles (Supplementary Tables S6 ing the top 95 recurrent gastric cancer somatic promoters, and S7), which are required for antigen presentation to CD8+ measuring transcripts associated with either the canonical cytotoxic T cells (IC50 ≤ 50 nmol/L, Fig. 4A; ref. 73). Analy- promoter or the alternative promoter. There was a significant sis of recurrent somatic promoter–associated peptides using correlation between the NanoString data and RNA-seq (Sup- the NetMHCpan-2.8 (74) algorithm against patient-specific plementary Fig. S10, r = 0.65, P < 0.001), with approximately MHC class I alleles revealed a significant enrichment in high- 35% of transcripts driven by alternate promoters upregulated affinity MHC I binding compared with multiple control in more than half of the gastric cancers (Fig. 4D). Second, to peptide populations, including canonical gastric cancer pep- examine markers of T-cell activity in these same gastric cancer tides (average 36% vs. 24%; P < 0.01), randomly selected pep- samples, we analyzed previously published microarray data tides (P < 0.001), and C-terminal peptides (P < 0.01; Fig. 4B (75) to measure CD8A (a measure of CD8+ tumor-infiltrating shows HLA-A, B, and C combined; Supplementary Fig. S9A lymphocytes), and granzyme A (GZMA) and perforin (PRF1; depicts data for HLA-A only). The majority of these high- refs. 76–78), which are both T-cell effectors and validated affinity somatic promoter–associated peptides corresponded markers of T-cell cytolytic activity. We confirmed that these to situations where the somatic transcript lacking the N- three genes (CD8A, GZMA, and PRF1) were not themselves terminal peptide is overexpressed in tumors relative to nor- associated with somatic promoters. Comparing the top and mal tissues (78% lost; 76/97 high-affinity peptides, Fig. 4C), bottom quartiles, gastric cancers with high somatic promoter and the proportion of high-affinity MHC-binding peptides usage exhibited significantly lowerGZMA and PRF1 levels among lost peptides was significantly greater than among (P < 0.001 and P = 0.01, Wilcoxon test) indicating lower T-cell gained peptides (37% vs. 21%, P < 0.05, Fisher test). Notably, cytolytic activity (Fig. 4E, top left), and also a trend toward because transcripts driven by the N-terminal lacking somatic lower CD8A levels (P = 0.14, Wilcoxon one-sided test). These TSSs are also overexpressed in tumors to a significantly findings support a lower level of tumor antigenicity in gas- greater degree than transcripts driven by the canonical TSS tric cancers with high somatic promoter usage, as recurrent (P < 0.05, Wilcoxon one-sided test; Supplementary Fig. S6), N-terminal peptides lost through somatic promoters are such a scenario would be expected to result in relative depletion predicted to be collectively more immunogenic than peptides of N-terminal immunogenic peptides in tumors. Interestingly, gained through somatic promoters (Fig. 4C). Using two dif- an analogous N-terminal analysis using RNA-seq data alone ferent algorithms (ASCAT, ref. 79, and ESTIMATE, ref. 80), (in the absence of epigenomic data) revealed that epigenome- we further confirmed that the decreasedGZMA and PRF1 guided N-terminal peptides exhibited significantly higher pre- levels are independent of tumor purity differences between dicted immunogenicity scores compared with RNA-seq–only gastric cancers (Supplementary Fig. S10). Similar results were identified peptides (36.10% vs. 27% for MHC presentation, obtained upon splitting the gastric cancer samples based on P = 0.02, Fisher test), suggesting that epigenome-guided pro- median promoter usage score (GZMA, P < 0.001; and PRF1, moter identification can provide complementary value to P = 0.03). Patients with gastric cancers exhibiting high RNA-seq–only guided analyses (Supplementary Fig. S9). somatic promoter usage (top 25%) also showed poor survival To explore whether somatic promoters might contribute compared with patients with gastric cancers with low somatic to reducing tumor antigen burden and immunoreactivity in promoter usage (bottom 25%; Fig. 4E, top right, HR = 2.56, vivo, we proceeded to examine correlations between promoter P = 0.02). Again, dividing patients by their median somatic

Figure 4. Somatic promoters correlate with immunoediting signatures. A, Schematic outlining alternative promoter usage [H3K4me3 box, overlapping gastric cancer (GC) in red and normal gastric tissue in blue] leading to alternative transcript usage (transcript box) and N-terminally truncated protein isoforms (protein box). B, Bar plot showing the average percentage of peptides with predicted high-affinity binding to MHC class I (HLA-A, B, and C, IC50 ≤ 50 nmol/L). N-terminal peptides associated with recurrent somatic promoters (alternative promoters) show significantly enriched predicted MHC I binding compared with canonical gastric cancer peptides (P < 0.01, Fisher test), random peptides from the human proteome (P < 0.001), and C-terminal peptides (P < 0.01) derived from the same genes exhibiting the N-terminal alterations. Canonical peptides refer to peptides derived from protein- coding genes overexpressed in gastric cancer through nonalternative promoters. C, Percentage (%) of high-affinity peptides predicted to bind different patient-specific HLA alleles categorized by somatic gain or loss. Most alleles have a greater number of N-terminal lost peptides predicted to have high binding affinity. The percentage of patients bearing specific HLA alleles is denoted inside the brackets. D, Quantification of somatic promoter expression using NanoString profiling. Top, distinct NanoString probes were designed to measure the expression of alternate and canonical promoter–driven transcripts. Two probes were designed for each gene, a canonical probe at the 5′ transcript marked by unaltered H3K4me3, and an alternate probe at the 5′ transcript of the somatic promoter. Bottom, heat map of alternative promoter expression from 95 gastric cancer and matched normal samples. Gastric cancer samples have been ordered left to right by their levels of somatic promoter usage. E, Association between somatic promoters and T-cell immune correlates. NS, not significant. Samples with high somatic promoter usage are in red, whereas those with low usage are in blue. Top left, expression of T-cell markers CD8A (P = 0.1443) and the T-cell cytolytic markers GZMA (P = 0.0001) and PRF1 (P = 0. 00806) in gastric cancer samples with either high or low somatic promoter usage (SG cohort). Samples with high alternative promoter usage show lower expression of immune markers. All P values are from Wilcoxon one-sided test. Top right, Kaplan–Meier analysis comparing overall survival curves between validation samples with high somatic promoter usage (top 25%) and low somatic promoter usage (bottom 25%; HR = 2.56, P = 0.02). Bottom left, expression of T-cell markers CD8A (P = 0.02), GZMA (P = 0.01), and PRF1 (P = 0.03) in TCGA STAD with either high or low somatic promoter usage. T-cell markers were evaluated by RNA-seq [transcripts per million (TPM)]. Bottom right, expression of T-cell markers CD8A (P = 0.035), GZMA (P = 0.001), and PRF1 (P = 0.025) in Asian Cancer Research Group (ACRG) gastric cancer samples with either high or low somatic promoter usage. All P values are from Wilcoxon one-sided test. F, EPIMAX heat map of total cytokine responses (fold change relative to actin) for 15 peptide pools against 9 donors. G, Individual cytokine responses against 15 peptides for two individual donors (donor 2 and donor 3) showing complex cytokine responses (FC ≥ 2). *, P < 0.05; **, P < 0.01; ***, P < 0.001.

OF9 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE

High-affinity MHC-I binding (average %) A Canonical TSS Somatic TSS B (IC50 < 50 nmol/L) 0 10 20 30 40 H3K4me3 Recurrent somatic Canonical Transcript * Random peptidome * * C-terminal Protein *

C D Gene X A*02:03A*02:07 (15%)A*11:01 (23%)A*11:02 (46%)A*24:02 (8%)A*24:07 (31%)A*24:10 (8%)A*33:03 (8%)A*34:01 (38%)B*15:01 (8%)B*15:21 (8%)B*15:27 (8%)B*27:04 (8%)B*38:02 (15%)B*39:01 (15%)B*40:01 (8%)B*40:06 B*46:01(38%) (8%)B*51:01 (31%)B*55:02 (8%)B*58:01 (8%)C*01:02 (31%)C*03:02 (31%) (31%) H3K4me3 100 C*03:04C*03:67 (8%)C*04:01 (8%)C*04:03 (23%)C*07:02 (8%)C*08:01 (31%)C*12:02 (15%)C*14:02 (15%)C*15:02 (8%) (8%) NanoString Canonical Alternate 75 probe

% 50

25 ONECUT2 IGF2BP3 0 Gain Loss LIF High-affinity binders MET HNF4A EPCAM E RASA3 Low alternate promoter usage High alternate promoter usage CEACAM6 JAK3 SG series 1.0 SG series NS *** 9 ** 0.8 HR = 2.56 (1.11–5.89) Log-rank P = 0.023 8 0.6 PSCA Low (n = 22) ATP4A 7 ESRRG 0.4 Loss Gain GIF 6 Survival rate NormalGC (n = 95) High (n = 20) 0.2 −40 4 Gene expression 5

(log2 normalized counts) 0.0 F Total cytokine CD8A GZMA PRF1 050 100 150 Donor 1 Time (months) Donor 2 Donor 3 TCGA ACRG Donor 4 ** * 40 *** Donor 5 10 Donor 6 30 *

pression * 8 Donor 7 pression

(TPM) 20 Donor 8

* 6 to actin) Log FC (relative Donor 9 10 Gene ex 4 Gene ex DST IKZF3 METMIB2MRC2NOS2PLEC −6 +6 0 DNAH3 LAMA3 PTGDSRASA3TRPM2 2 EPS8L1FRMD4B (log2-normalized counts) PLEKHG5 CD8A GZMA PRF1 CD8A GZMA PRF1

G GM-CSFIFNγIL2 IL3 IL4 IL7 IL9 IL10 IL13 IL15IL17AsCD40LTNFα

1,500 250 Donor 2Donor 3 200 1,000 to actin) to actin)

100 500 FC (relati ve FC (relati ve

0 0 5 DST MET DST MET IKZF3 MIB2MRC2NOS2 PLEC IKZF3 MIB2MRC2NOS2 PLEC DNAH3EPS8L1 LAMA3 PTGDSRASA3TRPM2 DNAH3EPS8L1 LAMA3 PTGDSRASA3TRPM2 FRMD4B PLEKHG FRMD4B PLEKHG5

JUNE 2017 CANCER DISCOVERY | OF10

RESEARCH ARTICLE Qamra et al. promoter usage score also showed similar survival differences testing for both T-cell proliferation and cytokine production. (Supplementary Fig. S10, HR = 1.81, P = 0.04). First, we identified N-terminal peptides predicted to exhibit high To validate these findings, we then analyzed two other HLA-binding affinities across a pool of healthy peripheral blood prominent gastric cancer cohorts: one from TCGA and mononuclear cell (PBMC) donors. Second, selecting 15 alter- another from the Asian Cancer Research Group (ACRG). In native promoter–associated peptides for testing, we generated the TCGA cohort, availability of RNA-seq data allowed us to peptide pools for each peptide (Supplementary Tables S9 and infer somatic promoter usage directly from next-generation S10; Methods), which were then used to stimulate PBMCs from sequencing data (Fig. 2C). Similar to the Singapore cohort, 9 healthy donors. T-cell proliferation and cytokine production TCGA gastric cancers with high somatic promoter usage levels were measured and benchmarked against control peptides (top 25%) exhibited decreased CD8A (P = 0.002, Wilcoxon (Supplementary Table S11). Across all 135 exposures (15 pep- one-sided test), GZMA (P = 0.001, Wilcoxon one-sided test), tides across 9 donors), we observed strong cytokine responses for and PRF1 levels (P = 0.005, Wilcoxon one-sided test, Fig. 4E, 79 peptide pools (58%; FC ≥ 2 relative to actin peptides; Fig. 4F bottom left) compared with gastric cancers with low somatic and G) inducing complex Th1, Th2, and Th17 polarizations in a promoter usage (bottom 25%) in a manner independent of donor-dependent fashion (Supplementary Fig. S11). tumor purity (Supplementary Fig. S10). Notably, as pre- To test the immunogenic capacity of specific N-terminal vious studies have suggested that somatic mutation burden peptides in a more cellular setting, we then assessed responses may also correlate with intratumor T-cell cytolytic response of T cells previously primed to recognize either altered or (81), we further repeated the analysis after adjusting for the WT peptides, when cocultured with HLA-matched isogenic total number of missense mutations in each sample using a gastric cancer cells expressing either altered or WT peptides, regression-based approach. Even after correcting for somatic respectively (Supplementary Fig. S11). Similar approaches mutation burden, we still observed decreased CD8A (P = have been used by previous investigators to investigate pro- 0.02, Wilcoxon one-sided test), GZMA (P = 0.01, Wilcoxon tein and peptide immunogenicity (84–86). By MHC-I affinity one-sided test), and PRF1 expression (P = 0.03, Wilcoxon one- screening, a VMCDIFFSL nonamer in the WT RASA3 N-ter- sided test) in samples with high somatic promoter usage (top minus was predicted to exhibit high MHC-I affinity binding

25% against bottom 25%; Supplementary Fig. S10). for both the HLA-A02:01 (IC50 = 6.93 nm) and HLA-A02:06 We leveraged a third independent cohort of gastric cancer (IC50 = 9.74 nm) alleles. Using HLA-A*02:06 T cells that are samples from ACRG (11). Using NanoString to target cross-reactive to HLA-A*02:01–positive AGS cells (87, 88), 89 canonical and alternative promoters along with various we tested the release of IFNγ from primed T cells after expo- immune markers, we profiled 264 primary gastric cancer sure to AGS lysates expressing either RASA3 CanT or SomT samples from the ACRG cohort (11). Forty percent of alter- isoforms. ELISA assays demonstrated that T cells primed to native promoter transcripts showed tumor-specific expression recognize RASA3 CanT released significantly more IFNγ when in more than half of the samples (Supplementary Fig. S10). cocultured with RASA3 CanT–expressing AGS cells than Once again, samples with high somatic promoter usage (top when cocultured with RASA3 SomT–expressing AGS cells. 25%) showed significantly lower expression of T-cell cytolytic In contrast, T cells primed with RASA3 SomT did not exhibit activity markers, including CD8A (P = 0.035, Wilcoxon one- appreciable IFNγ release when cocultured with RASA3 SomT– sided test), CD4A (P = 0.005, Wilcoxon one-sided test), GZMA expressing AGS cells (Supplementary Fig. S11). Thus, under (P = 0.001, Wilcoxon one-sided test), and PRF1 (P = 0.025, similar in vitro conditions, RASA3 CanT is capable of eliciting Wilcoxon one-sided test; Fig. 4E, bottom right; Supplemen- a stronger immune response than RASA3 SomT, consistent tary Fig. S10). Similar results were obtained upon splitting the with the RASA3 CanT N-terminus being more immunogenic. gastric cancer samples based on median promoter usage score Taken collectively, these in vitro results demonstrate that pep- (Supplementary Table S8). Also, after adjusting for mutational tides predicted to exhibit relative depletion in gastric can- burden (for cases where information is available), samples cers through somatic promoters can produce immunogenic with high somatic promoter usage still showed decreased CD8A responses, with the magnitude of immune responses depend- (P = 0.167, Wilcoxon one-sided test), GZMA (P = 0.009, Wil- ing on both peptide sequence and host immune background. coxon one-sided test), and PRF1 (P = 0.03, Wilcoxon one-sided test) expression (Supplementary Fig. S10). Taken collectively, Somatic Promoters Are Associated these results, observed across multiple gastric cancer cohorts with EZH2 Occupancy and assessed using diverse technologies (microarray, RNA-seq, To identify potential oncogenic mechanisms driving the NanoString), all support a significant association between somatic promoters, we intersected the genomic locations somatic promoter usage and reduced tumor immunity levels. of the somatic promoters with transcription factor binding Importantly, the decreased levels of T-cell cytolytic activity sites of 237 transcription factors from 83 different tissues associated with somatic promoter usage are likely independent (89). Regions exhibiting somatic promoters were significantly of tumor purity and mutational load. enriched in regions associated with EZH2 (P < 0.01) and SUZ12 (P < 0.01) binding (Fig. 5A; Supplementary Table S12), Somatic Promoter–Associated Peptides confirming earlier findings on a smaller cohort (38). Both Are Immunogenic In Vitro EZH2 and SUZ12 are components of the Polycomb repres- To functionally test the ability of N-terminal peptides sor complex 2 (PRC2) epigenetic regulator complex, which is depleted in gastric cancer to elicit immune responses, we con- upregulated in many cancer types, including gastric cancer ducted in vitro assays using the high-throughput Epitope Maxi- (90–95). To validate these findings, we then performed EZH2 mum (EPIMAX) platform (82, 83), which allows multiepitope ChIP-seq on HFE-145 normal gastric epithelial cells (Methods).

OF11 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE

AB(For illustrative purposes)

40 Gastric cancer EZH2 Somatic Gastric normal SUZ12 Gain signal 30 Loss H3K4me3 All promoters % Altered by EZH2 inhibition (GSK126) 0 2468101214 20 Somatic * 10 * Unaltered Binding enrichment * 0 All genes * 237 Transcription factors

CDSomatic loss (SLC9A9) Somatic loss (PSCA)

62 GC 98 GC 62 Normal 98 Normal H3K4me3 H3K4me3

12 GSK126 D6 30 GSK126 D6 12 DMSO D6 30 DMSO D6 20 39 RNA GSK126 D9 RNA GSK126 D9 20 DMSO D9 39 DMSO D9

chr3:143,561,025-143,571,802 2.5 Kb chr8:143,759,851-143,765,700 1 Kb

Figure 5. Somatic promoters are associated with EZH2 occupancy. A, Binding enrichment of ReMap-defined transcription factor–binding sites at genomic regions exhibiting somatic promoters. Transcription factors were sorted according to their binding frequency at all H3K4me3-defined promoter regions. EZH2 and SUZ12 binding sites significantly overlap regions exhibiting somatic promoters (gained and lost;P < 0.01, empirical distribution test). B, Proportion of RNA transcripts associated with somatic promoters changing upon GSK126 treatment in IM95 cells, compared with RNA transcripts associated with unaltered promoters. The top somatic promoter figure is for illustrative purposes only. Unaltered promoters were defined as all gene promoters except the somatic promoters. The proportion of genes changing upon treatment, as a proportion of all genes, is also shown. Somatic promoters are more likely to change expression after GSK126 treatment relative to unaltered promoters (OR = 1.46, P < 0.001) or all GSK126-regulated genes (OR = 9.21, P < 0.001, Fisher test). ***, P < 0.001. C, UCSC browser track of the SLC9A9 TSS, a gene with loss of promoter activity [overlapping gastric cancer (GC; red) and normal gastric tissue (blue) H3K4me3]. Gain of expression is seen after inhibition of EZH2 using GSK126 in IM95 cells at both day 6 (D6) and day 9 (D9) treatment. D, UCSC browser track of the PSCA TSS, with loss of promoter activity [GC (red) and normal gastric tissue (blue) H3K4me3]. Gain of expression is seen after inhibition of EZH2 using GSK126 in IM95 cells at both day 6 (D6) and day 9 (D9) treatment.

Concordant with the previous findings, we observed signifi- moters exhibiting deregulation after GSK126 challenge (8.8%, cant enrichment of EZH2 binding sites at somatic promoters OR = 1.46, P < 0.001, Fisher test, Fig. 5B), suggesting height- compared with all promoters (enrichment score 27 vs. 13 for ened sensitivity of somatic promoters to EZH2 inhibition. all promoters, P < 0.01), and this EZH2 enrichment remained The proportion of somatic promoters deregulated after EZH2 significant when the gained somatic (enrichment score 28, inhibition was also greater than the total proportion of genes P < 0.01) and lost somatic promoters (enrichment score 24, (as defined by GENCODE) regulated by GSK126 (1.5%, OR= P < 0.01) were analyzed separately (Supplementary Fig. S12). 9.21, P < 0.001, Fig. 5B). Of those promoters exhibiting both To experimentally test whether inhibiting EZH2/PRC2 GSK126 deregulation and also mapping to somatic promoters activity might modulate somatic promoter usage in gastric lost in primary gastric cancer, 89.6% were reactivated follow- cancer, we treated IM95 gastric cancer cells with GSK126, a ing GSK126 administration (78/87, FC ≥ 2, q < 0.1; Methods), highly selective small-molecule inhibitor of EZH2 methyl- consistent with EZH2 functioning to repress these promoters. transferase activity (90, 96). This line was selected because it For example, Fig. 5C and D highlights two lost somatic pro- has previously been shown to be sensitive to EZH2 depletion moters (SLC9A9 and PSCA), exhibiting expression gain after (Supplementary Fig. S12; ref. 97). RNA-seq analysis of GSK126 treatment (Fig. 5). These results thus suggest a role GSK126-treated IM95 cells at two treatment time points for EZH2 in regulating somatic promoters in gastric cancer. (days 6 and 9) confirmed that genes upregulated upon EZH2 inhibition are enriched in previously identified PRC2 target Somatic Promoters Reveal Novel gene sets (Supplementary Fig. S12). GSK126 treatment caused Cancer-Associated Transcripts deregulation of 2,134 promoters in total. Of 1,959 promot- Finally, when analyzing the altered somatic promoters with ers exhibiting somatic alterations in primary gastric cancers respect to proximity to known genes, we found that somatic (Fig. 1D), GSK126 treatment caused deregulation of 251 promoters could be classified into annotated and unanno- somatic promoters in IM95 cells (12.8%). This proportion was tated categories. Annotated promoters were defined as pro- significantly greater than the proportion of unaltered promoters mapping close (<500 bp) to a known GENCODE

JUNE 2017 CANCER DISCOVERY | OF12

RESEARCH ARTICLE Qamra et al.

A BC Tissue-specific enrichment <0.50.5–2 kb 2–10 kb20–50 kb10–20 kb>50 kb kb 1.00 CAGE+ 100 92% 85% Distance from 2,000 *** 80 annotated TSS 0.75 66% 1,500 60 0.50 41% 1,000 40 33% 0.25 20 500 0 0.00 erage RNA-seq reads 0 % of promoter regions Av

Normal GC Somatic Loss Gain Median genoSkyline score Gl ESCFetal Lung All BloodBreast Heart Muscle Epithelium Somatic DEF 44 140 ABCA13 unannotated TSS Shallow 42 135 sequencing Low All 35 GC Total population 40 promoters 130 expn 35 38 125 Normal Random sampling 87 GC line High 36 120 expn by NGS 59 anscripts ( × 100) 34 Somatic 115 Deep seq GC ChIP Tr promoters 59 # 110 H3K4me3 RNA 32 Deep seq normal Deep 30 105 RNA59 H3K4me3 GC sequencing 20 40 60 80 chr7:48, 236, 742-48, 243, 150 1 Kb 106120 139 # Aligned RNA-seq reads (M)

Figure 6. Somatic promoters reveal novel cancer-associated transcripts. A, Distribution of distances for different promoter categories to the nearest annotated TSSs. Left, the first bar plot shows distance distributions for promoters present in gastric normal tissues, the second for promoters present in gastric cancer (GC) samples, and the third for promoters exhibiting somatic alterations (i.e., different in tumor vs. normal). Right, the bar plots present distance distributions associated with either lost or gained somatic promoters. A substantial proportion of gained somatic promoters occupy locations distant from previously annotated TSSs (red, green, purple, blue, orange). B, Median functional scores of unannotated promoters as predicted by GenoSky- line across 7 different tissues. Unannotated promoters exhibited high functional scores for gastrointestinal, fetal, and embryonic stem cell (ESC) tissues. C, Box plot depicting average RNA-seq reads for CAGE-validated promoters, comparing either all promoters or somatic promoters and also supported by CAGE data (***, P < 0.001, Wilcoxon one-sided test). Somatic promoters are observed to have lower levels of RNA-seq expression. D, Cartoon depicting proposed effects of dynamic range on Nano-ChIP-seq and RNA-seq sensitivity in detecting lowly expressed transcripts. NGS, next-generation sequencing. Because of a more restricted dynamic range, epigenomic profiling may detect active promoters missed by RNA-seq, due to the random sampling of abundantly expressed genes by RNA-seq. E, Down-sampling and up-sampling analysis. The y-axis depicts the number of transcripts detected that overlap either all promoters (blue line) or somatic promoters (red line) at varying RNA-seq depths. Original primary sample RNA-seq data were sequenced at approximately 106 M reads, which were down-sampled to 20, 40, and 60 M reads. Deep RNA-seq data were additionally generated at approximately 139 M read depth. F, Cancer-associated transcripts detected at deep but not regular RNA-seq depth. The UCSC genome browser track for ABCA13 shows an example of a novel transcript detected by Nano-ChIP-seq at a read depth of 20 M but detected by RNA-seq only at a read depth of approximately 139 M (deep sequencing GC). This transcript is not detected by regular-depth RNA-seq (GC).

TSS, whereas unannotated promoters were those mapping data (54). We observed that gastrointestinal tissues had the to genomic regions devoid of known GENCODE TSSs. The third highest median score after embryonic stem cell (ESC) majority of promoters present in nonmalignant tissues, and and fetal tissues, consistent with our tumors being gastric in also promoters unchanged between tumors and normal tis- lineage and also dedifferentiated (Fig. 6B). In a separate analy- sues, mapped closely to previously annotated TSSs (72%– sis, recent studies have also suggested that endogenous repeat 92%). In contrast, only 41% of somatic promoters mapped to elements in the human genome may contribute significantly annotated promoter locations, whereas the remaining 59% to regulatory element variation (100), and hypomethylation mapped to “unannotated” locations, distant from GENCODE of repeat elements can induce cancer-associated transcription TSSs and in many cases 2 to 10 kb away (Fig. 6A). (101). We found that unannotated promoters were also sig- To test the functional relevance of these unannotated pro- nificantly enriched for the repeat elements ERV1 (P < 0.0001 moters, we used GenoCanyon, a nucleotide-level quantifica- unannotated vs. all) and L1 (P < 0.0001 unannotated vs. all; tion of genomic functional potential that integrates multiple Supplementary Fig. S13). levels of conservation and epigenomic information (98). We Compared with annotated promoters, unannotated pro- observed that 81% of the unannotated promoter regions exhib- moters exhibited weaker H3K27ac signals, suggesting that ited a maximum genome-wide functional score of greater than the former might have lower activity and decreased gene 0.9 (range 0–1), indicating high functional potential. To ascer- expression levels (Supplementary Fig. S13). Supporting this, tain tissue-type specificities, we then applied tissue-specific somatic promoters, even those supported by CAGE tags (indi- annotations using GenoSkyline (99), an extension of the cating true promoters), exhibited significantly lower RNA- GenoCanyon framework integrating Roadmap Epigenomics seq expression levels compared with all CAGE tag–supported

OF13 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE promoters (Fig. 6C). We thus hypothesized that unannotated membrane glycoprotein that has been proposed as a marker for promoters might be associated with low transcript levels, circulating tumor cells (119), and EPCAM expression levels have thereby rendering them more challenging to detect by con- been correlated with prognosis of patients with gastric cancer ventional depth transcriptome sequencing given the very wide (120). However, little is known about the specific cellular mech- dynamic range of cellular transcriptomes (10–10,000 tran- anisms driving high EPCAM expression in gastric cancer. Our scripts per cell for different genes; Fig. 6D; ref. 102). To test this finding thatEPCAM is regulated in gastric cancer not through possibility, we employed both down-sampling and up-sampling its canonical promoter, but instead through a cancer-specific analysis. Not surprisingly, decreasing levels of RNA-seq depth alternative promoter may lend credence to recent reports sug- caused a concomitant decrease in detected somatic promoter gesting that in addition to acting as an experimentally con- transcripts. For example, down-sampling to approximately venient surface marker, EPCAM may actually play a more direct 40 M reads caused approximately 250 transcripts (FPKM > 0; pro-oncogenic role in stimulating cellular proliferation (121). Fig. 6E) to be rendered undetectable at somatic promoters. Another novel example of an alternative promoter–associ- More convincingly, in the reciprocal experiment, we experi- ated gene, identified for the first time in our study, isRASA3 . mentally generated deep RNA-seq data for 5 matched gastric Although a functional role for RASA3 in cancer remains to cancer/normal pairs (average read depth 140 M compared with be definitively established, studies from other biological fields standard 100 M) and confirmed the additional detection of have shown that RASA3 can inhibit RAP1 (122), which in 435 new somatic promoter–associated transcripts (FPKM > 0; turn has been implicated in invasion and metastasis in vari- Fig. 6E). We estimate that usage of deep RNA-seq data allowed ous cancers (123, 124). RASA3 depletion can enhance signal- us to discover additional transcripts for 22% of the unanno- ing by integrins (125) and MAPK (64), and the possibility that tated promoters, not previously detectable at regular depth RASA3 can act as a tumor suppressor has also been recently RNA-seq (Fig. 6F). These results demonstrate that despite suggested through independent cross-species cancer studies being associated with bona fide cancer-associated transcripts, (126). Our results suggest that RASA3 may play a more com- many somatic promoters defined by epigenomic profiling may plex role in cancer, as the expression of WT RASA3 inhibited have been missed by conventional-depth RNA-seq. cell migration and invasion in gastric cancer cell lines, whereas N-terminal Var RASA3 enhanced migration and invasion. A third example of an alternative promoter–driven gene is MET, DISCUSSION which has been extensively investigated as a target for cancer Identifying somatically altered cis-regulatory elements and therapy (127–129). Although we and others have previously understanding how these elements direct cancer-associated gene reported (38, 65) the expression of an N-terminal truncated expression (103) represents a critical scientific goal (104, 105). MET-Var in cancer, functional implications of this truncated Here, we defined close to 2,000 promoters exhibiting altered MET-Var have remained unclear. In this study, experimental activity in gastric cancer, indicating that somatic promoters in assessment of MET WT and Var signaling revealed that trun- gastric cancer are pervasive. Promoters are traditionally defined cated MET variants may have different downstream signaling as proximal cis-regulatory elements that recruit general tran- effects compared with full-length MET isoforms. Under the scription factors to initiate transcription (106, 107). However, experimental conditions used, we observed significant dif- selection and activation of TSSs by RNA polymerase at core pro- ferences in phosphorylation patterns of ERK, STAT3, and moters is dependent on multiple factors. Core promoters are dif- GAB1, in a manner consistent with MET-Var being more pro- ferentially distributed between genes of different functions (15, oncogenic compared with MET-WT, as both ERK, STAT3, 106), and chromatin distributions and epigenetic landscapes of and GAB1 have been shown to facilitate MET-induced sig core promoter regions can also differ in a tissue-specific man- naling (130–132). The MET signaling pathway is known to ner (15, 108–110). Presence of multiple transcription initiation be particularly complex, with multiple feedback loops (133), sites within the same gene can generate distinct transcript iso- and understanding how expression of the N-terminal short forms with different 5′ UTRs that can act to regulate gene MET isoform might promote downstream survival signaling expression (111–113), and usage of alternative 5′ UTRs can also will be an important subject of future research, particularly affect both translation and protein stability of cancer-associ- in light of recent clinical trials targeting MET in lung cancer ated genes such as BRCA1, TGFβ, and ERG (18, 114–117). Such using antibodies that have been unsuccessful (5). findings demonstrate that specific promoter element activity Our study also revealed an unexpected relationship between is complex and cell-context dependent, with impact on down- somatic promoters and tumor immunity. Specifically, we dis- stream transcriptional, translational, and functional processes. covered that alternative promoter isoforms overexpressed in A significant proportion ∼( 18%) of somatic promoters cor- gastric cancer were significantly depleted of N-terminal pep- responded to alternative promoters. In cancer, alternative pro- tides predicted to be potentially immunogenic, based on com- moter utilization is of major relevance, as increasing numbers putational predictions of high-affinity MHC class I binding and of genes (e.g., LEF1, TP53, TGFB3) are now being shown to other immunologic assays. We believe that finding is relevant exhibit distinct alternative promoter–associated isoforms that to cancer immunity, as it builds on previous findings from the differentially affect malignant growth (21, 118). In the current literature establishing the existence of self-reactive T cells, the study, we identified alternative promoters in genes both known potential immunogenicity of overexpressed tumor antigens, and novel to gastric cancer biology with significant clinical and and the process of tumor immunoediting. First, although the translational implications. For example, we discovered an alter- majority of self-reactive T cells are clonally deleted during early native promoter at the EPCAM gene locus specifically activated development, numerous groups have also demonstrated the in gastric tumors. In gastric cancer, EPCAM encodes a trans- frequent persistence of self-reactive T cells in the periphery

JUNE 2017 CANCER DISCOVERY | OF14

RESEARCH ARTICLE Qamra et al.

(134). For example, analysis of transgenic mice has shown that In conclusion, our study indicates an important role for 25% to 40% of autoreactive T cells are likely to escape clonal somatic promoters in gastric cancer. We also note that a sig- deletion even in the presence of the deleting ligand (135), and nificant portion (52%) of the somatic promoters localized to in humans, Yu and colleagues have demonstrated that clonal unannotated TSSs, consistent with recent studies indicating the deletion prunes the T-cell repertoire but does not fully elim- existence of hundreds of transcript loci remaining to be anno- inate self-reactive T-cell clones (136). Importantly, although tated (180). Interestingly, a large portion of the human transcrip- such self-reactive T cells are typically low-avidity and are not tome has been shown to originate from repetitive elements that capable of recognizing self-antigens under normal physiologic can exhibit promoter activity and/or express noncoding RNAs conditions (137–140), they still retain the ability to become (181, 182). Unannotated promoters activated in our gastric can- activated and to produce effector and memory cells under con- cer study were found to be enriched in ERV1 and L1 repeat ele- ditions of appropriate stimulation, such as infection and the ments that have been shown to be associated with stage-specific mounting of antitumor responses (141, 142). transcription in early human embryonic cells (183), suggesting Second, in cancer, several studies have shown that self- a yet-unknown functional role for these promoters. Analysis of reactive T cells can exhibit immunologic activity toward these unannotated promoters is likely to provide fertile ground overexpressed tumor antigens, even if these antigens are also for new and hitherto unanticipated insights into mechanisms of expressed at lower levels in normal tissues. One well-known gastric cancer development and progression. example is the melanocyte differentiation antigen Melan-A/ MART1, which is both expressed by normal melanocytes and overexpressed in malignant melanoma cells (143–147). T-cell METHODS recognition of Melan-A/MART1 has been detected in 50% of Primary Tissue Samples and Cell Lines patients with melanoma (148), and even healthy individuals Primary patient samples were obtained from the SingHealth tissue have been shown to exhibit a disproportionately high fre- repository with approvals from the SingHealth Centralised Institu- quency of Melan-A/MART1–specific T cells in the peripheral tional Review Board and signed patient informed consent. “Normal” blood (148). Besides Melan-A/MART1, other examples of (nonmalignant) samples used in this study refer to samples harvested tumor-associated self-antigens (149–151) inducing immu- from the stomach, from sites distant from the tumor and exhibiting nologic recognition in both healthy individuals and patients no visible evidence of tumor or intestinal metaplasia/dysplasia upon with cancer (152) include tyrosinase-related proteins (TRP1 surgical assessment. Tumor samples were confirmed by cryosection- ing to contain >60% tumor cells. FU97, IM95, MKN7, OCUM1, and and TRP2; refs. 153–159) and glycoprotein (gp) 100 (147, RERF-GC-1B cell lines were obtained from the Japan Health Science 160–163) in melanoma, and P1A in mastocytoma cells (164). Research Resource Bank. AGS, KATOIII and SNU16, Hs 1.Int and Such examples clearly demonstrate that in certain cases, Hs 738.St/Int gastrointestinal fibroblast lines were obtained from normally expressed proteins can still become immunogenic the ATCC. NCC-59, NCC-24, and SNU-1967 and SNU-1750 were when overexpressed in cancer. Third, tumor immunoediting, obtained from the Korean Cell Line Bank. YCC3, YCC7, YCC21, the acquired capacity of developing tumors to escape immune and YCC22 were gifts from Yonsei Cancer Centre (Seoul, South control, is a recognized hallmark of cancer (165–172). Tumor Korea). HFE145 cells were a gift from Dr. Hassan Ashktorab (How- immune escape can occur via different mechanisms, such as ard University, Washington, DC). GES1 cells were a gift from Dr. through upregulation of immune checkpoint inhibitors (e.g., Alfred Cheng, Chinese University of Hong Kong. Cell line identities PD-L1) and altered transcription of antigen-presenting genes were confirmed by short tandem repeat DNA profiling using ANSI/ ATCC ASN-0002-2011 guidelines in mid-late 2015. All cell lines were (173–176) or tumor-specific antigens. For example, decreased negative for Mycoplasma contamination as assessed by the MycoAlert expression of melanoma antigens (e.g., gp100, MART1, and Mycoplasma Detection Kit (Lonza) and the MycoSensor qPCR Assay P1A) has been associated with melanoma progression to Kit (Agilent Technologies). PBMCs from healthy donors were col- later disease stages (177). Besides overt downregulation of lected under protocol CIRB ref no. 2010/720/E. the entire gene, it is thus highly plausible that transcriptional changes affecting splice forms and promoter variants ChIP-seq may also contribute to tumor immunoediting. For example, Nano-ChIP-seq was performed as described previously (38) with very recent work (178) in B-cell acute lymphoblastic leuke- slight modifications (see Supplementary Text). Eight Nano-ChIP-seq mia has described the production of N-terminally truncated libraries were multiplexed (New England Biolabs) and sequenced on CD19 variants in response to CD19 CART (chimeric antigen 2 lanes of a HiSeq2500 sequencer (Illumina) to an average depth of receptor–armed T cells) therapy, clearly showing that pro- 20 to 30 million reads per library. We assessed ChIP library qualities moter transcript variants can indeed arise as a consequence (H3K27ac, H3K4me3, and H3K4me1) using two different methods, of immunologic pressure. Taken collectively, we believe that ChIP enrichment assessment and CHANCE (see Supplementary these previously established findings all point to a plausible Text; ref. 41). For EZH2 ChIP-seq, EZH2 antibodies (catalog #5246, Cell Signaling Technology) were used for ChIP. Thirty nanograms of role for alternative promoters in reducing the immunogenic ChIPed DNA was used for each sequencing library preparation (New potential of tumors. In this regard, our observation that England Biolabs). somatic promoter regions exhibit a significant overlap with binding targets of the PRC2 epigenetic regulator complex, Promoter Analysis and are particularly sensitive to EZH2 inhibition, suggests Promoter (H3K4me3hi/H3K4me1lo) regions were identified by cal- that pharmacologic approaches for reawakening somatic culating the H3K4me3:H3K4me1 ratio for all H3K4me3 regions promoter–associated epitopes might represent an attractive merged across normal and gastric cancer samples. We estimated the strategy for increasing antitumor T-cell immunoreactivity required sample size to achieve 80% power and 10% type I error (http:// and antitumor activity (86, 179). powerandsamplesize.com/) based on the average signals of top 100

OF15 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE differential promoters between tumor and normal samples. This result were extracted using IDPicker’s idQuery tool (189). Differentially yielded a recommended sample size of 11 (average), which is met in our expressed peptides were identified by fitting a linear model (limma study (16 N/T). Regions with H3K4me3:H3K4me1 ratios <1 in both R; ref. 190) on quantile-normalized and log2-transformed spectral normal and gastric cancer samples were excluded from further analysis. counts. For gastric cancer cell line mass spectrometry, AGS, GES1, For all analyses performed in this study, promoter regions were defined SNU1750, and MKN1 proteomic profiles were generated using nano- as genomic locations exhibiting H3K4me3hi/H3K4me1lo signals, and for flow liquid chromatography on an EASY-nLC 1200 system coupled all subsequent analyses, it was only within this predefined H3K4me3hi/ to a Q Exactive HF mass spectrometer (Thermo Fisher Scientific; H3K4me1lo subset that H3K4me3 signals were compared. H3K27ac Supplementary Text). The Q Exactive HF was operated with a TOP20 data were used for correlative analysis. H3K4me3 data (FASTQs) for MS-MS spectra acquisition method per MS full scan. MS scans were colon carcinoma lines were downloaded from public databases: HCT116 conducted with 60,000 resolution and MS-MS scans with 15,000 res- and Caco2 from ENCODE and V503 and V400 from GSE36204. To olution. For data analysis, raw files were processed with MaxQuant compare promoter signals between gastric cancer and normal samples, (191) version 1.5.2.8 against the UniProt annotated human protein we used the DESeq2 (46) and edgeR (47) bioconductor packages using database (192). Carbamidomethylation was set as a fixed modifica- a read count matrix of ChIP-seq signals, adjusting for replicate information, whereas methionine oxidation and protein N-acetylation were tion. Regions with FCs greater than 1.5 (FDR = 0.1) were selected as considered as variable modifications. Search results were processed significantly different. The criteria of FC = 1.5 and q < 0.1 was based with MaxQuant filtered with an FDR of 0.01. The match between run on previous literature comparing ChIP-seq profiles using DESeq2 and option and Label-Free Quantification (LFQ; ref. 193) was activated. edgeR also using similar thresholds (49, 50). Significantly altered pro- LFQ intensities were filtered for potential contaminants and reverse moters identified by DESeq2 overlapped almost completely with altered proteins, and log2 transformed. They were then imputed using the promoters found by edgeR. A regularized log transformation of the open-source software Perseus (0.5 width, 1.8 downshift; ref. 194) and DESeq2 read counts was used to plot PCAs and heat maps. fitted using linear models (limma R; ref. 190).

Transcriptome Analysis Molecular Biology and Biochemistry RNA-seq data were obtained from the European Genome-Phenome Procedures for 5′ rapid amplification of cDNA ends (5′ RACE), gene Archive under accession no. EGAS00001001128. Data were processed cloning, Western blotting (MET variants), RASA3 mRNA measure- by first aligning to GENCODE v19 transcript annotations using ment, and RAS-GTP assays are presented in the Supplementary Text. TopHat v2.0.12 (184). Cufflinks 2.2.0 was used to generate FPKM abundance measures. For identification of novel transcripts, Cufflinks Transfection with RASA3 siRNAs was used without employing a reference transcript annotation. Tran- Two RASA3 siRNAs were used to silence the RASA3 SomT transcripts were then merged across all gastric cancer and normal samples script in NCC24 cells [hs.Ri.RASA3.13.1 TriFECTa Kit DsiRNA and compared against GENCODE annotations to identify novel Duplex (Integrated DNA Technologies), and Silencer Select Pre- transcripts using Cuffmerge 2.2.0. Deep-depth strand-specific RNA- Designed siRNA s355 (Life Technologies)]. NCC24 cells were trans- seq was also performed on 10 additional primary samples (paired- fected either with the above two siRNAs or a nontargeting control end 101 bp). TCGA datasets were downloaded from TCGA Data (ON-TARGETplus non-targeting pool, Dharmacon) at a final con- Portal (https://tcga-data.nci.nih.gov/tcga) in the form of FASTQ files, centration of 100 nmol/L for 48 hours, subsequently followed by which were then aligned to GENCODE v19 transcript annotations qPCR and Western validation and migration/invasion assays. using TopHat v2.0.12. To analyze promoter-associated RNA expression, RNA-seq reads from TCGA samples (tumor and normal) were Cell Proliferation, Migration, and Invasion Assays mapped against the genomic locations of promoter regions originally 3 defined by epigenomic profiling in the discovery samples, including For cell proliferation, 3 × 10 GES1, SNU1967, and AGS cells were all promoters, gained somatic promoters, and lost somatic promot- plated into 96-well plates in media with 10% FBS and left overnight to ers (see Fig. 1). RNA-seq reads mapping to these epigenome-defined attach. The next day (day 0), cells were transiently transfected with WT promoter regions were then quantified and normalized by promoter and Var RASA3 constructs using Lipofectamine 3000 (Thermo Fisher length (kilobases) and by total library size, and FCs in expression were Scientific). The amount of the constructs was 40 ng per well for AGS computed between tumor and normal TCGA sample groups. Length and 100 ng per well for GES1 and SNU1967 cells. Cell proliferation of promoter loci was defined as the number of base pairs (bp) between was measured by the WST-8 assay (Cell Counting Kit-8, Dojindo) the start and stop genomic coordinate of the H3K4me3 region as from 24 to 120 hours posttransfection. WST-8 solution (10 μL) was identified by the peak caller program CCAT v3.0 (185). Isoform level added per well, and the absorbance reading was measured at 450 nm quantification for alternative promoter–driven transcripts was per- after 2 hours of incubation in a humidified incubator. To deter- formed using Cufflinks (FPKM; ref. 186), Kallisto (TPM; ref. 187), and mine cell-migratory capacities, RASA3 WT and Var-transfected GES1, MISO (isoform-centric analysis; ref. 188). Assigned counts for each SNU1967, and AGS cells and siRNA-treated NCC24 cells were tested isoform were normalized by DESeq2. using Corning Costar 6.5-mm Transwell with 8.0-μm Pore Polycar- bonate Membrane Inserts (3422, Corning). AGS cells (2.5 × 104), GES1 4 4 4 Other Analyses cells (2 × 10 ), SNU1967 cells (3 × 10 ), and NCC24 cells (5 × 10 ) were suspended in 0.1 mL serum-free RPMI medium and added to the Other analyses, including DNA methylation analysis, survival top of the Transwell insert. RPMI (0.6 mL) containing 10% FBS was analysis, gene set enrichment analysis, analysis of repetitive elements, added into the bottom well as a chemoattractant. After incubation functional element analysis using GenoCanyon and GenoSkyline, for 24 hours at 37°C in a 5% CO2 incubator, cells were fixed with 3.7% and analysis of transcription factor binding sites, are presented in formaldehyde and permeabilized with 100% methanol. Nonmigrated the Supplementary Text. cells were scraped off with cotton swabs from the upper surface of the membrane. Migrated cells were stained with 0.5% crystal violet. The Mass Spectrometry and Data Analysis number of migrated cells was represented as the total area of migrated Peptide-level spectral data for 90 colon and rectal cancer samples cells versus the area of Transwell membrane calculated using ImageJ (63) were downloaded from the CPTAC portal (https://cptac-data- software. For cell invasion assays, the above Transwell inserts were portal.georgetown.edu/cptac/s/S016) generated by the Clinical Pro- coated with 0.1 mL (300 μg/mL) Corning Matrigel matrix (354234, teomic Tumor Analysis Consortium (NCI/NIH). Spectral counts Corning) for 2 to 4 hours at 37°C before use.

JUNE 2017 CANCER DISCOVERY | OF16

RESEARCH ARTICLE Qamra et al.

Altered Peptide and Antigen Prediction and 10% heat-inactivated FCS (HyClone)] for 5 days. Individual pep- Altered peptides were defined as variant N-terminal protein tide pools of each alternative promoter were added at the start of the sequences arising from somatic alterations in alternative promoter culture at a concentration of 1 μg/mL for each peptide. At the end usage. The following filters were applied to select the pool of altered of day 5, cells were stained with LIVE/DEAD Fixable Near-IR Dead peptides: (i) FC of at least 1.5 for alternate versus canonical RNA-seq Cell Stain Kit (Life Technologies) and labeled with CD4-BUV737 expression; (ii) only one canonical and one alternate isoform per gene (BD Biosciences), CD8-PacificBlue (BD Biosciences), CD3-PE loci; and (iii) annotated transcripts confirmed as protein coding by (BioLegend), CD19-PE/TexasRed (Beckman Coulter), and CD56- GENCODE. Canonical promoters were defined as regions exhibit- APC (BD Biosciences). In addition, magnetic bead–based cytokine ing unaltered H3K4me3 peaks. Random peptides from the human multiplex analysis (human cytokine panel 1, Millipore, Merck) was proteome were generated from amino acid sequences of GENCODE performed on cell culture supernatants to measure secreted cytokine coding transcripts. N-terminal peptide gains were identified as cases levels. in which the alternative transcript was associated with a different 5′ region predicted to result in a different translated protein sequence IFNf Assays compared with the canonical transcript. For each N-terminal altered To test the immunogenicity of the RASA3 WT and Var protein protein, we evaluated binding of 9-mer peptides using the NetMHC- sequences, CD14+ monocytes were isolated from an HLA-A*02:06 pan 2.8 using a strict threshold of IC50 ≤ 50 nmol/L to identify strong donor by positive selection using magnetic beads (Miltenyi Biotec). MHC binders (74, 195). Antigen predictions were performed against Dendritic cells (DC) were generated by GM-CSF (1,000 IU/mL) and patient-specific HLA types of gastric cancer samples predicted using IL4 (400 IU/mL) and further matured by TNF (10 ng/mL), IL1b OptiType (196). OptiType was run using default parameters, except (10 ng/mL), IL6 (10 ng/mL; Miltenyi Biotec), and PGE2 (1 μg/mL; BWA mem was used as an aligner for prefiltering reads aligning to the Stemcell Technologies) for 24 hours. The DCs were then primed with OptiType-provided reference sequences. AGS cell lysates expressing WT RASA3 or Var RASA3 for 24 hours, before being cocultured with T cells from the same donor at the ratio Association of Cytolytic Markers with of 1:5. After 5 days of coculture with DCs, T cells were isolated by Alternative Promoter Usage positive selection using CD3 magnetic beads (Miltenyi Biotec) and cocultured with AGS cells expressing either WT or Var RASA3 at the Local immune cytolytic activity was evaluated using the expression ratio of 20:1 for 2 days. Supernatants were harvested and IFNγ release of GZMA and PRF1 as previously used by Rooney and colleagues (81). was measured by ELISA (R&D Systems). Tumor content was estimated using two algorithms, ASCAT (aberrant cell fraction; ref. 79) and ESTIMATE (tumor purity; ref. 80). Expression data for the SG series were downloaded (GSE15460) and normalized NanoString Analysis using the robust multiarray average algorithm in the “affy” R package NanoString nCounter Reporter CodeSets were designed for 95

(197) and log2 transformed. Affymetrix SNP Array 6.0 data for the SG genes (83 upregulated in gastric cancer and 11 downregulated) and 5 series were downloaded from GSE31168 and GSE85466. Mutation housekeeping genes (AGPAT1, CLTC, B2M, POL2RL, and TBP covering frequencies for TCGA stomach adenocarcinoma (STAD) samples were a broad expression range) on the SG series samples. For each gene, we downloaded from the TCGA STAD publication data (https://tcga- designed three probes, targeting (i) the 5′ end of the alternate promoter data.nci.nih.gov/docs/publications/stad_2014/; ref. 198) using level 2 location; (ii) the 5′ end of the canonical promoter (defined by promoter curated MAF files (QCv5_blacklist_Pass.aggregated.capture.tcga. regions of equal enrichment in both GC and normal samples or the uuid.curated.somatic.maf) filtered for “Missense” variant classifi- longest protein-coding transcript); and (iii) a common downstream cation. Expression data for TCGA STAD samples (TPM) were com- probe. A separate NanoString assay was designed for 88 genes on the puted using the Kallisto algorithm (187). Raw SNP Array 6.0.CEL files ACRG cohort, using similar criteria. Vendor-provided nCounter soft- for TCGA gastric cancers (STAD) were downloaded from the GDC ware (nSolver) was used for data analysis. Raw counts were normal- data portal (https://gdc-portal.nci.nih.gov/). Access to this dataset ized using the geometric mean of the internal positive control probes was obtained using Database of Genotypes and Phenotypes (dbGaP) included in each CodeSet. credentials and an ID issued by eRA commons. Precomputed ESTI- MATE scores for TCGA STAD were downloaded from http://bioin- EZH2 Inhibition formatics.mdanderson.org/estimate/ and converted to tumor purity using the formula cos (0.6049872018 + 0.0001467884 × ESTIMATE IM95 cells were treated with GSK126 (Selleck Chemicals), a score; ref. 80). Preprocessed expression data for the ACRG series were selective EZH2 inhibitor (96, 199), at a concentration of 5 μmol/L. downloaded from GSE62254, and precomputed ASCAT scores were Cell proliferation was monitored in 96-well plates posttreatment obtained from collaborators (J. Lee). Expression of cytolytic markers with GSK126 using the CellTiter-Glo Luminescent Cell Viability was adjusted for missense mutation and tumor purity frequencies Assay (Promega) for three independent experiments. For RNA- using a spline regression model. seq analysis, total RNA was extracted using the Qiagen RNeasy Mini Kit according to the manufacturer’s instructions. Cells were EPIMAX Assays treated with GSK126 (Selleck Chemicals; dissolved in DMSO) at a concentration of 5 μmol/L. Control cells were treated with the Peptides for 15 representative alternative promoters were syn- same concentration of DMSO (0.1%). RNA-seq differential analysis thesized by GenScript (Supplementary Table S10). Control peptide for promoter loci was carried out using edgeR (47) on read counts pools for human actin were purchased from JPT [PM-ACTS, PepMix mapping to H3K4me3 regions estimated using featureCounts Human (Actin) JPT]. PBMCs were obtained from 9 healthy volun- (200). RNA-seq gene level differential analysis was performed using teers, of which 8 PBMC samples were HLA typed (Supplementary cuffdiff2.2.1. Table S9). PBMCs were labeled with 1 μmol/L CFSE (Life Technolo- gies, Thermo Fisher Scientific) and cultured at a density of 2 × 105 cells per well in complete culture medium [cRPMI comprising RPMI- Accession Codes 1640 medium (Gibco, Thermo Fisher Scientific), 15 mmol/L HEPES Genomic data for this study have been deposited in the National (Gibco), 1% nonessential amino acid (Gibco), 1 mmol/L sodium Center for Biotechnology GEO database, under accession numbers pyruvate (Gibco), 1% penicillin/streptomycin (Gibco), 2 mmol/L GSE51776 and GSE75898 (https://www.ncbi.nlm.nih.gov/geo/query/ l-glutamine (Gibco), 50 μmol/L β2-mercaptoethanol (Sigma, Merck), acc.cgi?token=kfoxqeamzfetpal&acc=GSE75898).

OF17 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE

Disclosure of Potential Conflicts of Interest TRIO-013/LOGiC–A randomized phase III trial. J Clin Oncol 2016; 34:443–51. P. Tan has ownership interest (including patents) in ASTAR. No 5. Roche. Roche provides update on phase III study of onartuzumab in potential conflicts of interest were disclosed by the other authors. people with specific type of lung cancer. Available from: http://www. roche.com/media/store/releases/med-cor-2014-03-03.htm. Authors’ Contributions 6. Ohtsu A, Shah MA, Van Cutsem E, Rha SY, Sawaki A, Park SR, Conception and design: A. Qamra, M. Xing, N. Padmanabhan, et al. Bevacizumab in combination with chemotherapy as first-line P.K.H. Chow, B.T. Teh, P. Tan therapy in advanced gastric cancer: a randomized, double-blind, Development of methodology: M. Xing, J.J.T. Kwok, J.S. Lin, X. Yao, placebo-controlled phase III study. J Clin Oncol 2011;29:3968–76. B.T. Teh, P. Tan . 7 Dikken JL, van Sandick JW, Maurits Swellengrebel HA, Lind PA, Putter Acquisition of data (provided animals, acquired and managed H, Jansen EP, et al. Neo-adjuvant chemotherapy followed by surgery patients, provided facilities, etc.): M. Xing, N. Padmanabhan, and chemotherapy or by surgery and chemoradiotherapy for patients S. Zhang, C. Xu, Y.S. Leong, A.P.L. Lim, Q. Tang, X. Yao, X. Ong, with resectable gastric cancer (CRITICS). BMC Cancer 2011;11:329. M. Lee, S.T. Tay, E.G. Santoso, C.C.Y. Ng, A. Jusakul, D. Smoot, S.Y. Rha, 8. Wang K, Kan J, Yuen ST, Shi ST, Chu KM, Law S, et al. Exome sequencing identifies frequent mutation of ARID1A in molecular K.G. Yeoh, W.P. Yong, P.K.H. Chow, W.H. Chan, H.S. Ong, K.C. Soo, subtypes of gastric cancer. Nat Genet 2011;43:1219–23. K.-M. Kim, W.K. Wong, B.T. Teh, D. Kappei, J. Lee, P. Tan 9. Zang ZJ, Cutcutache I, Poon SL, Zhang SL, McPherson JR, Tao J, Analysis and interpretation of data (e.g., statistical analysis, bio- et al. Exome sequencing of gastric adenocarcinoma identifies recur- statistics, computational analysis): A. Qamra, M. Xing, J.J.T. Kwok, rent somatic mutations in cell adhesion and chromatin remodeling Q. Tang, W.F. Ooi, J.S. Lin, T. Nandi, X. Yao, A. Ng, S.Y. Rha, S.G. Rozen, genes. Nat Genet 2012;44:570–4. D. Kappei, J. Connolly, P. Tan 10. Yao F, Kausalya JP, Sia YY, Teo AS, Lee WH, Ong AG, et al. Recurrent Writing, review, and/or revision of the manuscript: A. Qamra, fusion genes in gastric cancer: CLDN18-ARHGAP26 induces loss of M. Xing, J.J.T. Kwok, A. Ng, S.Y. Rha, K.G. Yeoh, W.P. Yong, P.K.H. Chow, epithelial integrity. Cell Rep 2015;12:272–85. S.G. Rozen, B.T. Teh, J. Connolly, P. Tan . 11 Cristescu R, Lee J, Nebozhyn M, Kim KM, Ting JC, Wong SS, et al. Administrative, technical, or material support (i.e., reporting or Molecular analysis of gastric cancer identifies subtypes associated organizing data, constructing databases): A.P.L. Lim, A.T.L. Keng, with distinct clinical outcomes. Nat Med 2015;21:449–56. H. Ashktorab, S.G. Rozen, B.T. Teh, P. Tan 12. Tan IB, Ivanova T, Lim KH, Ong CW, Deng N, Lee J, et al. Intrinsic Study supervision: P.K.H. Chow, P. Tan subtypes of gastric cancer, based on gene expression pattern, predict Other (provided the cell lines): H. Ashktorab survival and respond differently to chemotherapy. Gastroenterology 2011;141:476–85. Acknowledgments 13. Bang YJ, Van Cutsem E, Feyereislova A, Chung HC, Shen L, Sawaki A, et al. Trastuzumab in combination with chemotherapy versus We thank the Sequencing and Scientific Computing teams at the chemotherapy alone for treatment of HER2-positive advanced gas- Genome Institute of Singapore for providing sequencing services and tric or gastro-oesophageal junction cancer (ToGA): a phase 3, open- data management capabilities, and the Duke-NUS Genome Biology label, randomised controlled trial. Lancet 2010;376:687–97. Facility for sequencing services. We also thank Dr. Shyam Prabhakar 14. Lenhard B, Sandelin A, Carninci P. Metazoan promoters: emerging for helpful discussions. We thank Dr. Wanjin Hong for the gift of characteristics and insights into transcriptional regulation. Nat Rev HEK293 cells and pCI-Puro-HA vector and Dr. Alfred Cheng for the Genet 2012;13:233–45. gift of GES1 cells. 15. Davuluri RV, Suzuki Y, Sugano S, Plass C, Huang TH. The functional consequences of alternative promoter use in mammalian Grant Support genomes. Trends Genet 2008;24:167–77. This work was supported by a core institutional grant from the 16. D’Alessio JA, Wright KJ, Tjian R. Shifting players and paradigms in Genome Institute of Singapore under the Agency for Science, Technol- cell-specific transcription. Mol Cell 2009;36:924–31. ogy and Research; core funding from Duke-NUS Medical School; and . 17 Bieberstein NI, Carrillo Oesterreich F, Straube K, Neugebauer KM. National Medical Research Council grants TCR/009-NUHS/2013, First exon length controls active chromatin signatures and tran- BnB/0005b/2013 (BnB11dec069), and NMRC/STaR/0026/2015. scription. Cell Rep 2012;2:62–8. 18. Zammarchi F, Boutsalis G, Cartegni L. 5 UTR control of native ERG and of Other sources of support include the Cancer Science Institute of ′ Tmprss2:ERG variants activity in prostate cancer. PLoS One 2013;8:e49721. Singapore, NUS, under the National Research Foundation Singapore 19. Ong CK, Leong C, Tan PH, Van T, Huynh H. The role of 5′ untrans- and the Singapore Ministry of Education under its Research Centres lated region in translational suppression of OKL38 mRNA in hepa- of Excellence initiative. tocellular carcinoma. Oncogene 2007;26:1155–65. 20. Valen E, Pascarella G, Chalk A, Maeda N, Kojima M, Kawazu C, et al. Received September 13, 2016; revised October 27, 2016; accepted Genome-wide detection and analysis of hippocampus core promot- March 16, 2017; published OnlineFirst March 20, 2017. ers using DeepCAGE. Genome Res 2009;19:255–65. . 21 Wiesner T, Lee W, Obenauf AC, Ran L, Murali R, Zhang QF, et al. Alternative transcription initiation leads to expression of a novel REFERENCES ALK isoform in cancer. Nature 2015;526:453–7. . 1 Ferlay J, Soerjomataram I, Dikshit R, Eser S, Mathers C, Rebelo 22. Muller M, Schleithoff ES, Stremmel W, Melino G, Krammer PH, M, et al. Cancer incidence and mortality worldwide: sources, Schilling T. One, two, three–p53, p63, p73 and chemosensitivity. methods and major patterns in GLOBOCAN 2012. Int J Cancer Drug Resist Updat 2006;9:288–306. 2015;136:E359–86. 23. Arce L, Yokoyama NN, Waterman ML. Diversity of LEF/TCF action 2. Layke JC, Lopez PP. Gastric cancer: diagnosis and treatment options. in development and disease. Oncogene 2006;25:7492–504. Am Fam Physician 2004;69:1133–40. 24. Agarwal VR, Bulun SE, Leitch M, Rohrich R, Simpson ER. Use of 3. Schmidt N, Peitz U, Lippert H, Malfertheiner P. Missing gastric alternative promoters to express the aromatase cytochrome P450 cancer in dyspepsia. Aliment Pharmacol Ther 2005;21:813–20. (CYP19) gene in breast adipose tissues of cancer-free and breast 4. Hecht JR, Bang YJ, Qin SK, Chung HC, Xu JM, Park JO, et al. cancer patients. J Clin Endocrinol Metab 1996;81:3843–9. Lapatinib in combination with capecitabine plus oxaliplatin in 25. Carninci P, Kasukawa T, Katayama S, Gough J, Frith MC, Maeda human epidermal growth factor receptor 2-positive advanced or N, et al. The transcriptional landscape of the mammalian genome. metastatic gastric, esophageal, or gastroesophageal adenocarcinoma: Science 2005;309:1559–63.

JUNE 2017 CANCER DISCOVERY | OF18

RESEARCH ARTICLE Qamra et al.

26. Forrest AR, Kawaji H, Rehli M, Baillie JK, de Hoon MJ, Haberle V, et al. A 49. Ross-Innes CS, Stark R, Teschendorff AE, Holmes KA, Ali HR, promoter-level mammalian expression atlas. Nature 2014;507:462–70. Dunning MJ, et al. Differential oestrogen receptor binding is . 27 Core LJ, Martins AL, Danko CG, Waters CT, Siepel A, Lis JT. Analysis of associated with clinical outcome in breast cancer. Nature 2012; nascent RNA identifies a unified architecture of initiation regions at 481:389–93. mammalian promoters and enhancers. Nat Genet 2014;46:1311–20. 50. Decker KF, Zheng D, He Y, Bowman T, Edwards JR, Jia L. Persistent 28. Wang Z, Zang C, Rosenfeld JA, Schones DE, Barski A, Cuddapah S, androgen receptor-mediated transcription in castration-resistant et al. Combinatorial patterns of histone acetylations and methyla- prostate cancer under androgen-deprived conditions. Nucleic Acids tions in the human genome. Nat Genet 2008;40:897–903. Res 2012;40:10765–79. 29. Barski A, Cuddapah S, Cui K, Roh TY, Schones DE, Wang Z, et al. . 51 Okitsu CY, Hsieh JC, Hsieh CL. Transcriptional activity affects the High-resolution profiling of histone methylations in the human H3K4me3 level and distribution in the coding region. Mol Cell Biol genome. Cell 2007;129:823–37. 2010;30:2933–46. 30. Creyghton MP, Cheng AW, Welstead GG, Kooistra T, Carey BW, 52. Zhang ZZ, Shen ZY, Shen YY, Zhao EH, Wang M, Wang CJ, et al. Steine EJ, et al. Histone H3K27ac separates active from poised HOTAIR long noncoding RNA promotes gastric cancer metastasis enhancers and predicts developmental state. Proc Natl Acad Sci U S A through suppression of Poly r(C)-Binding Protein (PCBP) 1. Mol 2010;107:21931–6. Cancer Ther 2015;14:1162–70. . 31 Rada-Iglesias A, Bajpai R, Swigut T, Brugmann SA, Flynn RA, 53. Ding J, Li D, Gong M, Wang J, Huang X, Wu T, et al. Expression and Wysocka J. A unique chromatin signature uncovers early develop- clinical significance of the long non-coding RNA PVT1 in human mental enhancers in humans. Nature 2011;470:279–83. gastric cancer. Onco Targets Ther 2014;7:1625–30. 32. Gallego Romero I, Pai AA, Tung J, Gilad Y. RNA-seq: impact of RNA 54. Kundaje A, Meuleman W, Ernst J, Bilenky M, Yen A, Heravi-Moussavi degradation on transcript quantification. BMC Biol 2014;12:42. A, et al. Integrative analysis of 111 reference human epigenomes. 33. Preker P, Almvig K, Christensen MS, Valen E, Mapendano CK, Nature 2015;518:317–30. Sandelin A, et al. PROMoter uPstream Transcripts share characteris- 55. Smith ZD, Meissner A. DNA methylation: roles in mammalian tics with mRNAs and are produced upstream of all three major types development. Nat Rev Genet 2013;14:204–20. of mammalian promoters. Nucleic Acids Res 2011;39:7179–93. 56. Jones PA. Functions of DNA methylation: islands, start sites, gene 34. Kim TK, Hemberg M, Gray JM, Costa AM, Bear DM, Wu J, et al. bodies and beyond. Nat Rev Genet 2012;13:484–92. Widespread transcription at neuronal activity-regulated enhancers. . 57 Harrow J, Frankish A, Gonzalez JM, Tapanari E, Diekhans M, Nature 2010;465:182–7. Kokocinski F, et al. GENCODE: the reference human genome anno- 35. Andersson R, Refsing Andersen P, Valen E, Core LJ, Bornholdt J, Boyd tation for The ENCODE Project. Genome Res 2012;22:1760–74. M, et al. Nuclear stability and transcriptional directionality separate 58. Chia NY, Deng N, Das K, Huang D, Hu L, Zhu Y, et al. Regulatory functionally distinct RNA species. Nat Commun 2014;5:5336. crosstalk between lineage-survival oncogenes KLF5, GATA4 and 36. Andersson R, Gebhard C, Miguel-Escalada I, Hoof I, Bornholdt J, GATA6 cooperatively promotes gastric cancer development. Gut Boyd M, et al. An atlas of active enhancers across human cell types 2015;64:707–19. and tissues. Nature 2014;507:455–61. 59. Chang HR, Nam S, Kook MC, Kim KT, Liu X, Yao H, et al. HNF4 . 37 Ng JH, Kumar V, Muratani M, Kraus P, Yeo JC, Yaw LP, et al. In vivo alpha is a therapeutic target that links AMPK to WNT signalling in epigenomic profiling of germ cells reveals germ cell molecular sig- early-stage gastric cancer. Gut 2016;65:19–32. natures. Dev Cell 2013;24:324–33. 60. Tanaka T, Jiang S, Hotta H, Takano K, Iwanari H, Sumi K, et al. 38. Muratani M, Deng N, Ooi WF, Lin SJ, Xing M, Xu C, et al. Nanoscale Dysregulated expression of P1 and P2 promoter-driven hepatocyte chromatin profiling of gastric adenocarcinoma reveals cancer- nuclear factor-4alpha in the pathogenesis of human cancer. J Pathol associated cryptic promoters and somatically acquired regulatory 2006;208:662–72. elements. Nat Commun 2014;5:4361. . 61 Takano K, Hasegawa G, Jiang S, Kurosaki I, Hatakeyama K, Iwanari 39. Tanasijevic B, Dai B, Ezashi T, Livingston K, Roberts RM, Rasmussen H, et al. Immunohistochemical staining for P1 and P2 promoter- TP. Progressive accumulation of epigenetic heterogeneity during driven hepatocyte nuclear factor-4alpha may complement mucin human ES cell culture. Epigenetics 2009;4:330–8. phenotype of differentiated-type early gastric carcinoma. Pathol Int 40. Smiraglia DJ, Rush LJ, Fruhwald MC, Dai Z, Held WA, Costello JF, 2009;59:462–70. et al. Excessive CpG island hypermethylation in cancer cell lines ver- 62. Edwards NJ, Oberti M, Thangudu RR, Cai S, McGarvey PB, Jacob sus primary human malignancies. Hum Mol Genet 2001;10:1413–9. S, et al. The CPTAC data portal: a resource for cancer proteomics . 41 Diaz A, Nellore A, Song JS. CHANCE: comprehensive software research. J Proteome Res 2015;14:2707–13. for quality control and validation of ChIP-seq data. Genome Biol 63. Zhang B, Wang J, Wang X, Zhu J, Liu Q, Shi Z, et al. Proteog- 2012;13:R98. enomic characterization of human colon and rectal cancer. Nature 42. Andersson R, Sandelin A, Danko CG. A unified architecture of tran- 2014;513:382–7. scriptional regulatory elements. Trends Genet 2015;31:426–33. 64. Nafisi H, Banihashemi B, Daigle M, Albert PR. GAP1(IP4BP)/ 43. Raja UM, Gopal G, Rajkumar T. Intragenic DNA methylation con- RASA3 mediates Galphai-induced inhibition of mitogen-activated comitant with repression of ATP4B and ATP4A gene expression in protein kinase. J Biol Chem 2008;283:35908–17. gastric cancer is a potential serum biomarker. Asian Pac J Cancer 65. The Cancer Genome Atlas Research Network, Linehan WM, Prev 2012;13:5563–8. Spellman PT, Ricketts CJ, Creighton CJ, Fei SS, et al. Comprehensive 44. The Cancer Genome Atlas Research Network. Comprehensive molec- molecular characterization of papillary renal-cell carcinoma. N Engl ular characterization of gastric adenocarcinoma. Nature 2014;513: J Med 2016;374:135–45. 202–9. 66. Nishida K, Hirano T. The role of Gab family scaffolding adapter 45. Rapaport F, Khanin R, Liang Y, Pirun M, Krek A, Zumbo P, et al. proteins in the signal transduction of cytokine and growth factor Comprehensive evaluation of differential gene expression analysis receptors. Cancer Sci 2003;94:1029–33. methods for RNA-seq data. Genome Biol 2013;14:R95. . 67 Darnell JE Jr, Kerr IM, Stark GR. Jak-STAT pathways and transcrip- 46. Love MI, Huber W, Anders S. Moderated estimation of fold change and tional activation in response to IFNs and other extracellular signal- dispersion for RNA-seq data with DESeq2. Genome Biol 2014;15:550. ing proteins. Science 1994;264:1415–21. 47. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor 68. Ihle JN. Cytokine receptor signalling. Nature 1995;377:591–4. package for differential expression analysis of digital gene expres- 69. Wen Z, Zhong Z, Darnell JE Jr. Maximal activation of transcription sion data. Bioinformatics 2010;26:139–40. by Stat1 and Stat3 requires both tyrosine and serine phosphoryla- 48. Steinhauser S, Kurzawa N, Eils R, Herrmann C. A comprehensive tion. Cell 1995;82:241–50. comparison of tools for differential ChIP-seq analysis. Brief Bioin- 70. Decker T, Kovarik P. Serine phosphorylation of STATs. Oncogene form 2016;17:953–66. 2000;19:2628–37.

OF19 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE

. 71 Schumacher TN, Schreiber RD. Neoantigens in cancer immuno- 93. Varambally S, Dhanasekaran SM, Zhou M, Barrette TR, Kumar- therapy. Science 2015;348:69–74. Sinha C, Sanda MG, et al. The polycomb group protein EZH2 is 72. Mittal D, Gubin MM, Schreiber RD, Smyth MJ. New insights into involved in progression of prostate cancer. Nature 2002;419:624–9. cancer immunoediting and its three component phases–elimination, 94. Kleer CG, Cao Q, Varambally S, Shen R, Ota I, Tomlins SA, et al. equilibrium and escape. Curr Opin Immunol 2014;27:16–25. EZH2 is a marker of aggressive breast cancer and promotes neoplas- 73. Sette A, Vitiello A, Reherman B, Fowler P, Nayersina R, Kast tic transformation of breast epithelial cells. Proc Natl Acad Sci U S A WM, et al. The relationship between class I binding affinity and 2003;100:11606–11. immunogenicity of potential cytotoxic T cell epitopes. J Immunol 95. Hock H. A complex Polycomb issue: the two faces of EZH2 in 1994;153:5586–92. cancer. Genes Dev 2012;26:751–5. 74. Hoof I, Peters B, Sidney J, Pedersen LE, Sette A, Lund O, et al. Net- 96. McCabe MT, Ott HM, Ganji G, Korenchuk S, Thompson C, Van MHCpan, a method for MHC class I binding prediction beyond Aller GS, et al. EZH2 inhibition as a therapeutic strategy for lym- humans. Immunogenetics 2009;61:1–13. phoma with EZH2-activating mutations. Nature 2012;492:108–12. 75. Ooi CH, Ivanova T, Wu J, Lee M, Tan IB, Tao J, et al. Oncogenic . 97 Cheng LL, Itahana Y, Lei ZD, Chia NY, Wu Y, Yu Y, et al. TP53 pathway combinations predict clinical prognosis in gastric cancer. genomic status regulates sensitivity of gastric cancer cells to the PLoS Genet 2009;5:e1000676. histone methylation inhibitor 3-deazaneplanocin A (DZNep). Clin 76. Johnson BJ, Costelloe EO, Fitzpatrick DR, Haanen JB, Schumacher Cancer Res 2012;18:4201–12. TN, Brown LE, et al. Single-cell perforin and granzyme expression 98. Lu Q, Hu Y, Sun J, Cheng Y, Cheung KH, Zhao H. A statistical reveals the anatomical localization of effector CD8+ T cells in influ- framework to predict functional non-coding regions in the human enza virus-infected mice. Proc Natl Acad Sci U S A 2003;100:2657–62. genome through integrated analysis of annotation data. Sci Rep . 77 Ji RR, Chasalow SD, Wang L, Hamid O, Schmidt H, Cogswell J, et al. 2015;5:10576. An immune-active tumor microenvironment favors clinical response 99. Lu Q, Powles RL, Wang Q, He BJ, Zhao H. Integrative tissue- to ipilimumab. Cancer Immunol Immunother 2012;61:1019–31. specific functional annotations in the human genome provide novel 78. Herbst RS, Soria JC, Kowanetz M, Fine GD, Hamid O, Gordon MS, insights on many complex traits and improve signal prioritization et al. Predictive correlates of response to the anti-PD-L1 antibody in genome wide association studies. PLoS Genet 2016;12:e1005947. MPDL3280A in cancer patients. Nature 2014;515:563–7. 100. Villar D, Berthelot C, Aldridge S, Rayner TF, Lukk M, Pignatelli 79. Van Loo P, Nordgard SH, Lingjaerde OC, Russnes HG, Rye IH, Sun M, et al. Enhancer evolution across 20 mammalian species. Cell W, et al. Allele-specific copy number analysis of tumors. Proc Natl 2015;160:554–66. Acad Sci U S A 2010;107:16910–5. . 101 Wolff EM, Byun HM, Han HF, Sharma S, Nichols PW, Siegmund 80. Yoshihara K, Shahmoradgoli M, Martinez E, Vegesna R, Kim H, KD, et al. Hypomethylation of a LINE-1 promoter activates an alter- Torres-Garcia W, et al. Inferring tumour purity and stromal and nate transcript of the MET oncogene in bladders with cancer. PLoS immune cell admixture from expression data. Nat Commun 2013;4: Genet 2010;6:e1000917. 2612. 102. Marinov GK, Williams BA, McCue K, Schroth GP, Gertz J, Myers RM, . 81 Rooney MS, Shukla SA, Wu CJ, Getz G, Hacohen N. Molecular and et al. From single-cell to cell-pool transcriptomes: stochasticity in genetic properties of tumors associated with local immune cytolytic gene expression and RNA splicing. Genome Res 2014;24:496–510. activity. Cell 2015;160:48–61. 103. Kouzarides T. Chromatin modifications and their function. Cell 82. Skibinski DA, Hanson BJ, Lin Y, von Messling V, Jegerlehner A, Tee 2007;128:693–705. JB, et al. Enhanced neutralizing antibody titers and Th1 polariza- 104. Rivera CM, Ren B. Mapping human epigenomes. Cell 2013;155:39–55. tion from a novel Escherichia coli derived pandemic influenza vac- 105. Jones PA, Baylin SB. The fundamental role of epigenetic events in cine. PLoS One 2013;8:e76571. cancer. Nat Rev Genet 2002;3:415–28. 83. Palucka AK, Ueno H, Fay J, Banchereau J. Dendritic cells: a critical 106. Kadonaga JT. Perspectives on the RNA polymerase II core promoter. player in cancer therapy? J Immunother 2008;31:793–805. Wiley Interdiscip Rev Dev Biol 2012;1:40–51. 84. Robbins PF, Lu YC, El-Gamil M, Li YF, Gross C, Gartner J, et al. Mining .107 Sandelin A, Carninci P, Lenhard B, Ponjavic J, Hayashizaki Y, Hume exomic sequencing data to identify mutated antigens recognized by DA. Mammalian RNA polymerase II core promoters: insights from adoptively transferred tumor-reactive T cells. Nat Med 2013;19:747–52. genome-wide studies. Nat Rev Genet 2007;8:424–36. 85. Tran E, Turcotte S, Gros A, Robbins PF, Lu YC, Dudley ME, et al. 108. Frith MC, Valen E, Krogh A, Hayashizaki Y, Carninci P, Sandelin A. Cancer immunotherapy based on mutation-specific CD4+ T cells in A code for transcription initiation in mammalian genomes. Genome a patient with epithelial cancer. Science 2014;344:641–5. Res 2008;18:1–12. 86. Matsushita H, Vesely MD, Koboldt DC, Rickert CG, Uppaluri R, 109. Carninci P, Sandelin A, Lenhard B, Katayama S, Shimokawa K, Magrini VJ, et al. Cancer exome analysis reveals a T-cell-dependent Ponjavic J, et al. Genome-wide analysis of mammalian promoter mechanism of cancer immunoediting. Nature 2012;482:400–4. architecture and evolution. Nat Genet 2006;38:626–35. . 87 Sidney J, Southwood S, Mann DL, Fernandez-Vina MA, Newman 110. Rach EA, Winter DR, Benjamin AM, Corcoran DL, Ni T, Zhu J, MJ, Sette A. Majority of peptides binding HLA-A*0201 with high et al. Transcription initiation patterns indicate divergent strate- affinity crossreact with other A2-supertype molecules. Hum Immu- gies for gene regulation at the chromatin level. PLoS Genet 2011;7: nol 2001;62:1200–16. e1001274. 88. Torikai H, Akatsuka Y, Miyauchi H, Terakura S, Onizuka M, . 111 Trinklein ND, Aldred SJ, Saldanha AJ, Myers RM. Identification Tsujimura K, et al. The HLA-A*0201-restricted minor histocompati- and functional analysis of human transcriptional promoters. bility antigen HA-1H peptide can also be presented by another HLA- Genome Res 2003;13:308–12. A2 subtype, A*0206. Bone Marrow Transplant 2007;40:165–74. 112. Zhang T, Haws P, Wu Q. Multiple variable first exons: a mechanism 89. Griffon A, Barbier Q, Dalino J, van Helden J, Spicuglia S, Ballester for cell- and tissue-specific gene regulation. Genome Res 2004;14: B. Integrative analysis of public ChIP-seq experiments reveals a com- 79–89. plex multi-cell regulatory landscape. Nucleic Acids Res 2015;43:e27. 113. Araujo PR, Yoon K, Ko D, Smith AD, Qiao M, Suresh U, et al. Before 90. Simon JA, Lange CA. Roles of the EZH2 histone methyltransferase it gets started: regulating translation at the 5′ UTR. Comp Funct in cancer epigenetics. Mutat Res 2008;647:21–9. Genomics 2012;2012:475731. . 91 Matsukawa Y, Semba S, Kato H, Ito A, Yanagihara K, Yokozaki H. 114. Pal S, Gupta R, Kim H, Wickramasinghe P, Baubet V, Showe Expression of the enhancer of zeste homolog 2 is correlated with LC, et al. Alternative transcription exceeds alternative splicing in poor prognosis in human gastric cancer. Cancer Sci 2006;97:484–91. generating the transcriptome diversity of cerebellar development. 92. Fujii S, Ochiai A. Enhancer of zeste homolog 2 downregulates Genome Res 2011;21:1260–72. E-cadherin by mediating histone H3 methylation in gastric cancer 115. Sobczak K, Krzyzosiak WJ. Structural determinants of BRCA1 cells. Cancer Sci 2008;99:738–46. translational regulation. J Biol Chem 2002;277:17349–58.

JUNE 2017 CANCER DISCOVERY | OF20

RESEARCH ARTICLE Qamra et al.

116. Arrick BA, Lee AL, Grendell RL, Derynck R. Inhibition of translation 138. Morgan DJ, Kreuwel HT, Fleck S, Levitsky HI, Pardoll DM, Sherman of transforming growth factor-beta 3 mRNA by its 5′ untranslated LA. Activation of low avidity CTL specific for a self epitope results in region. Mol Cell Biol 1991;11:4306–13. tumor rejection but not autoimmunity. J Immunol 1998;160:643–51. .117 Han B, Dong Z, Liu Y, Chen Q, Hashimoto K, Zhang JT. Regulation 139. Sandberg JK, Franksson L, Sundback J, Michaelsson J, Petersson M, of constitutive expression of mouse PTEN by the 5′-untranslated Achour A, et al. T cell tolerance based on avidity thresholds rather region. Oncogene 2003;22:5325–37. than complete deletion allows maintenance of maximal repertoire 118. Vizoso M, Ferreira HJ, Lopez-Serra P, Carmona FJ, Martinez-Cardus diversity. J Immunol 2000;165:25–33. A, Girotti MR, et al. Epigenetic activation of a cryptic TBC1D16 140. Poplonski L, Vukusic B, Pawling J, Clapoff S, Roder J, Hozumi transcript enhances melanoma progression by targeting EGFR. Nat N, et al. Tolerance is overcome in beef insulin-transgenic mice Med 2015;21:741–50. by activation of low-affinity autoreactive T cells. Eur J Immunol 119. Toyoshima K, Hayashi A, Kashiwagi M, Hayashi N, Iwatsuki M, 1996;26:601–9. Ishimoto T, et al. Analysis of circulating tumor cells derived from .141 McMahan RH, Slansky JE. Mobilizing the low-avidity T cell reper- advanced gastric cancer. Int J Cancer 2015;137:991–8. toire to kill tumors. Semin Cancer Biol 2007;17:317–29. 120. Warneke VS, Behrens HM, Haag J, Kruger S, Simon E, Mathiak M, 142. Enouz S, Carrie L, Merkler D, Bevan MJ, Zehn D. Autoreactive et al. Members of the EpCAM signalling pathway are expressed in T cells bypass negative selection and respond to self-antigen gastric cancer tissue and are correlated with patient prognosis. Br J stimulation during infection. J Exp Med 2012;209:1769–79. Cancer 2013;109:2217–27. 143. Coulie PG, Brichard V, Van Pel A, Wolfel T, Schneider J, Traversari C, . 121 Chaves-Perez A, Mack B, Maetzel D, Kremling H, Eggert C, Harreus et al. A new gene coding for a differentiation antigen recognized by U, et al. EpCAM regulates cell cycle progression via control of cyclin autologous cytolytic T lymphocytes on HLA-A2 melanomas. J Exp D1 expression. Oncogene 2013;32:641–50. Med 1994;180:35–42. 122. Stefanini L, Paul DS, Robledo RF, Chan ER, Getz TM, Campbell 144. Sensi M, Traversari C, Radrizzani M, Salvi S, Maccalli C, Mortarini RA, et al. RASA3 is a critical inhibitor of RAP1-dependent platelet R, et al. Cytotoxic T-lymphocyte clones from different patients activation. J Clin Invest 2015;125:1419–32. display limited T-cell-receptor variable-region gene usage in HLA- 123. Che YL, Luo SJ, Li G, Cheng M, Gao YM, Li XM, et al. The C3G/Rap1 A2-restricted recognition of the melanoma antigen Melan-A/ pathway promotes secretion of MMP-2 and MMP-9 and is involved MART-1. Proc Natl Acad Sci U S A 1995;92:5674–8. in serous ovarian cancer metastasis. Cancer Lett 2015;359:241–9. 145. Kawakami Y, Eliyahu S, Delgado CH, Robbins PF, Rivoltini L, 124. House CD, Wang BD, Ceniccola K, Williams R, Simaan M, Olender Topalian SL, et al. Cloning of the gene coding for a shared human J, et al. Voltage-gated Na+ channel activity increases colon cancer melanoma antigen recognized by autologous T cells infiltrating transcriptional activity and invasion via persistent MAPK signaling. into tumor. Proc Natl Acad Sci U S A 1994;91:3515–9. Sci Rep 2015;5:11541. 146. Kawakami Y, Eliyahu S, Sakaguchi K, Robbins PF, Rivoltini L, 125. Molina-Ortiz P, Polizzi S, Ramery E, Gayral S, Delierneux C, Oury C, Yannelli JR, et al. Identification of the immunodominant peptides et al. Rasa3 controls megakaryocyte Rap1 activation, integrin signaling of the MART-1 human melanoma antigen recognized by the major- and differentiation into proplatelet. PLoS Genet 2014;10:e1004420. ity of HLA-A2-restricted tumor infiltrating lymphocytes. J Exp Med 126. Tang J, Li Y, Lyon K, Camps J, Dalton S, Ried T, et al. Cancer driver- 1994;180:347–52. passenger distinction via sporadic human and dog cancer compari- . 147 Cox AL, Skipper J, Chen Y, Henderson RA, Darrow TL, Shabanowitz son: a proof-of-principle study with colorectal cancer. Oncogene J, et al. Identification of a peptide recognized by five melanoma- 2014;33:814–22. specific human cytotoxic T cell lines. Science 1994;264:716–9. .127 Gherardi E, Birchmeier W, Birchmeier C, Vande Woude G. Targeting 148. Zippelius A, Pittet MJ, Batard P, Rufer N, de Smedt M, Guillaume MET in cancer: rationale and progress. Nat Rev Cancer 2012;12:89–103. P, et al. Thymic selection generates a large T cell pool recognizing a 128. Peters S, Adjei AA. MET: a promising anticancer therapeutic target. self-peptide in humans. J Exp Med 2002;195:485–94. Nat Rev Clin Oncol 2012;9:314–26. 149. Houghton AN, Taormina MC, Ikeda H, Watanabe T, Oettgen HF, 129. Kawakami H, Okamoto I, Arao T, Okamoto W, Matsumoto K, Old LJ. Serological survey of normal humans for natural antibody Taniguchi H, et al. MET amplification as a potential therapeutic to cell surface antigens of melanoma. Proc Natl Acad Sci U S A target in gastric cancer. Oncotarget 2013;4:9–17. 1980;77:4260–4. 130. Tanimura S, Chatani Y, Hoshino R, Sato M, Watanabe S, Kataoka T, 150. Livingston PO, Natoli EJ, Calves MJ, Stockert E, Oettgen HF, et al. Activation of the 41/43 kDa mitogen-activated protein kinase Old LJ. Vaccines containing purified GM2 ganglioside elicit GM2 signaling pathway is required for hepatocyte growth factor-induced antibodies in melanoma patients. Proc Natl Acad Sci U S A cell scattering. Oncogene 1998;17:57–65. 1987;84:2911–5. .131 Zhang YW, Wang LM, Jove R, Vande Woude GF. Requirement of .151 Houghton AN, Eisinger M, Albino AP, Cairncross JG, Old LJ. Surface Stat3 signaling for HGF/SF-Met mediated tumorigenesis. Onco- antigens of melanocytes and melanomas. Markers of melanocyte gene 2002;21:217–26. differentiation and melanoma subsets. J Exp Med 1982;156:1755–66. 132. Weidner KM, Di Cesare S, Sachs M, Brinkmann V, Behrens J, Birchmeier 152. Lewis JJ, Janetzki S, Schaed S, Panageas KS, Wang S, Williams L, W. Interaction between Gab1 and the c-Met receptor tyrosine kinase is et al. Evaluation of CD8(+) T-cell frequencies by the Elispot assay responsible for epithelial morphogenesis. Nature 1996;384:173–6. in healthy individuals and in patients with metastatic melanoma 133. Trusolino L, Bertotti A, Comoglio PM. MET signalling: principles immunized with tyrosinase peptide. Int J Cancer 2000;87:391–8. and functions in development, organ regeneration and cancer. Nat 153. Brichard V, Van Pel A, Wolfel T, Wolfel C, De Plaen E, Lethe B, et al. Rev Mol Cell Biol 2010;11:834–48. The tyrosinase gene codes for an antigen recognized by autolo- 134. Ota K, Matsui M, Milford EL, Mackin GA, Weiner HL, Hafler DA. gous cytolytic T lymphocytes on HLA-A2 melanomas. J Exp Med T-cell recognition of an immunodominant myelin basic protein 1993;178:489–95. epitope in multiple sclerosis. Nature 1990;346:183–7. 154. Brichard VG, Herman J, Van Pel A, Wildmann C, Gaugler B, Wolfel 135. Bouneaud C, Kourilsky P, Bousso P. Impact of negative selection T, et al. A tyrosinase nonapeptide presented by HLA-B44 is recog- on the T cell repertoire reactive to a self-peptide: a large fraction of nized on a human melanoma by autologous cytolytic T lympho- T cell clones escapes clonal deletion. Immunity 2000;13:829–40. cytes. Eur J Immunol 1996;26:224–30. 136. Yu W, Jiang N, Ebert PJ, Kidd BA, Muller S, Lund PJ, et al. Clonal 155. Robbins PF, el-Gamil M, Kawakami Y, Stevens E, Yannelli JR, deletion prunes but does not eliminate self-specific alphabeta Rosenberg SA. Recognition of tyrosinase by tumor-infiltrating lym- CD8(+) T lymphocytes. Immunity 2015;42:929–41. phocytes from a patient responding to immunotherapy. Cancer Res .137 Schild H, Rotzschke O, Kalbacher H, Rammensee HG. Limit of 1994;54:3124–6. T cell tolerance to self proteins by peptide presentation. Science 156. Wolfel T, Van Pel A, Brichard V, Schneider J, Seliger B, Meyer zum 1990;247:1587–9. Buschenfelde KH, et al. Two tyrosinase nonapeptides recognized on

OF21 | CANCER DISCOVERY JUNE 2017 www.aacrjournals.org

Epigenomic Promoter Alterations in Gastric Cancer RESEARCH ARTICLE

HLA-A2 melanomas by autologous cytolytic T lymphocytes. Eur J 179. Wolfel T, Hauer M, Schneider J, Serrano M, Wolfel C, Klehmann-Hieb Immunol 1994;24:759–64. E, et al. A p16INK4a-insensitive CDK4 mutant targeted by cytolytic . 157 Topalian SL, Rivoltini L, Mancini M, Markus NR, Robbins PF, T lymphocytes in a human melanoma. Science 1995;269:1281–4. Kawakami Y, et al. Human CD4+ T cells specifically recognize a 180. Djebali S, Davis CA, Merkel A, Dobin A, Lassmann T, Mortazavi shared melanoma-associated antigen encoded by the tyrosinase A, et al. Landscape of transcription in human cells. Nature 2012; gene. Proc Natl Acad Sci U S A 1994;91:9461–5. 489:101–8. 158. Wang RF, Appella E, Kawakami Y, Kang X, Rosenberg SA. Identi- . 181 Faulkner GJ, Kimura Y, Daub CO, Wani S, Plessy C, Irvine KM, et al. fication of TRP-2 as a human tumor antigen recognized by cytotoxic The regulated retrotransposon transcriptome of mammalian cells. T lymphocytes. J Exp Med 1996;184:2207–16. Nat Genet 2009;41:563–71. 159. Mattes MJ, Thomson TM, Old LJ, Lloyd KO. A pigmentation-associ- 182. Speek M. Antisense promoter of human L1 retrotransposon drives ated, differentiation antigen of human melanoma defined by a pre- transcription of adjacent cellular genes. Mol Cell Biol 2001;21:1973–85. cipitating antibody in human serum. Int J Cancer 1983;32:717–21. 183. Goke J, Lu X, Chan YS, Ng HH, Ly LH, Sachs F, et al. Dynamic 160. Trager U, Sierro S, Djordjevic G, Bouzo B, Khandwala S, Meloni A, transcription of distinct classes of endogenous retroviral elements et al. The immune response to melanoma is limited by thymic selec- marks specific populations of early human embryonic cells. Cell tion of self-antigens. PLoS One 2012;7:e35005. Stem Cell 2015;16:135–41. .161 Castelli C, Rivoltini L, Andreola G, Carrabba M, Renkvist N, 184. Kim D, Pertea G, Trapnell C, Pimentel H, Kelley R, Salzberg SL. Parmiani G. T-cell recognition of melanoma-associated antigens. TopHat2: accurate alignment of transcriptomes in the presence J Cell Physiol 2000;182:323–31. of insertions, deletions and gene fusions. Genome Biol 2013; 162. Vijayasaradhi S, Bouchard B, Houghton AN. The melanoma antigen 14:R36. gp75 is the human homologue of the mouse b (brown) locus gene 185. Xu H, Handoko L, Wei X, Ye C, Sheng J, Wei CL, et al. A signal-noise product. J Exp Med 1990;171:1375–80. model for significance analysis of ChIP-seq with negative control. 163. Kawakami Y, Robbins PF, Wang X, Tupesis JP, Parkhurst MR, Kang Bioinformatics 2010;26:1199–204. X, et al. Identification of new melanoma epitopes on melanosomal 186. Trapnell C, Williams BA, Pertea G, Mortazavi A, Kwan G, van Baren proteins recognized by tumor infiltrating T lymphocytes restricted MJ, et al. Transcript assembly and quantification by RNA-Seq by HLA-A1, -A2, and -A3 alleles. J Immunol 1998;161:6985–92. reveals unannotated transcripts and isoform switching during cell 164. Brandle D, Bilsborough J, Rulicke T, Uyttenhove C, Boon T, Van differentiation. Nat Biotechnol 2010;28:511–5. den Eynde BJ. The shared tumor-specific antigen encoded by mouse . 187 Bray NL, Pimentel H, Melsted P, Pachter L. Near-optimal prob- gene P1A is a target not only for cytolytic T lymphocytes but also for abilistic RNA-seq quantification. Nat Biotechnol 2016;34:525–7. tumor rejection. Eur J Immunol 1998;28:4010–9. 188. Katz Y, Wang ET, Airoldi EM, Burge CB. Analysis and design of RNA 165. Dunn GP, Koebel CM, Schreiber RD. Interferons, immunity and sequencing experiments for identifying isoform regulation. Nat cancer immunoediting. Nat Rev Immunol 2006;6:836–48. Methods 2010;7:1009–15. 166. Dunn GP, Old LJ, Schreiber RD. The three Es of cancer immunoed- 189. Ma ZQ, Dasari S, Chambers MC, Litton MD, Sobecki SM, iting. Annu Rev Immunol 2004;22:329–60. Zimmerman LJ, et al. IDPicker 2.0: Improved protein assembly .167 Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. with high discrimination peptide identification filtering. J Proteome Cell 2011;144:646–74. Res 2009;8:3872–81. 168. Dunn GP, Bruce AT, Ikeda H, Old LJ, Schreiber RD. Cancer immu- 190. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. limma noediting: from immunosurveillance to tumor escape. Nat Immunol powers differential expression analyses for RNA-sequencing and 2002;3:991–8. microarray studies. Nucleic Acids Res 2015;43:e47. 169. Shankaran V, Ikeda H, Bruce AT, White JM, Swanson PE, Old LJ, et al. . 191 Cox J, Mann M. MaxQuant enables high peptide identification rates, IFNgamma and lymphocytes prevent primary tumour development individualized p.p.b.-range mass accuracies and proteome-wide and shape tumour immunogenicity. Nature 2001;410:1107–11. protein quantification. Nat Biotechnol 2008;26:1367–72. 170. Schreiber RD, Old LJ, Smyth MJ. Cancer immunoediting: integrat- 192. UniProt Consortium. UniProt: a hub for protein information. ing immunity’s roles in cancer suppression and promotion. Science Nucleic Acids Res 2015;43:D204–12. 2011;331:1565–70. 193. Cox J, Hein MY, Luber CA, Paron I, Nagaraj N, Mann M. Accurate . 171 Koebel CM, Vermi W, Swann JB, Zerafa N, Rodig SJ, Old LJ, et al. proteome-wide label-free quantification by delayed normalization Adaptive immunity maintains occult cancer in an equilibrium state. and maximal peptide ratio extraction, termed MaxLFQ. Mol Cell Nature 2007;450:903–7. Proteomics 2014;13:2513–26. 172. Vesely MD, Kershaw MH, Schreiber RD, Smyth MJ. Natural innate and 194. Tyanova S, Temu T, Sinitcyn P, Carlson A, Hein MY, Geiger T, et al. adaptive immunity to cancer. Annu Rev Immunol 2011;29:235–71. The Perseus computational platform for comprehensive analysis of 173. Hicklin DJ, Marincola FM, Ferrone S. HLA class I antigen down- (prote)omics data. Nat Methods 2016;13:731–40. regulation in human cancers: T-cell immunotherapy revives an old 195. Nielsen M, Lundegaard C, Blicher T, Lamberth K, Harndahl M, story. Mol Med Today 1999;5:178–86. Justesen S, et al. NetMHCpan, a method for quantitative predictions 174. Nie Y, Yang G, Song Y, Zhao X, So C, Liao J, et al. DNA hypermethyl- of peptide binding to any HLA-A and -B locus protein of known ation is a mechanism for loss of expression of the HLA class I genes sequence. PLoS One 2007;2:e796. in human esophageal squamous cell carcinomas. Carcinogenesis 196. Szolek A, Schubert B, Mohr C, Sturm M, Feldhahn M, Kohlbacher 2001;22:1615–23. O. OptiType: precision HLA typing from next-generation sequenc- 175. Fonsatti E, Sigalotti L, Coral S, Colizzi F, Altomonte M, Maio M. ing data. Bioinformatics 2014;30:3310–6. Methylation-regulated expression of HLA class I antigens in mela- .197 Gautier L, Cope L, Bolstad BM, Irizarry RA. affy–analysis of Affyme- noma. Int J Cancer 2003;105:430–1. trix GeneChip data at the probe level. Bioinformatics 2004;20:307–15. 176. Soong TW, Hui KM. Locus-specific transcriptional control of HLA 198. The Cancer Genome Atlas Research Network. Comprehensive mol- genes. J Immunol 1992;149:2008–20. ecular characterization of gastric adenocarcinoma. Nature 2014; . 177 de Vries TJ, Fourkour A, Wobbes T, Verkroost G, Ruiter DJ, van 513:202–9. Muijen GN. Heterogeneous expression of immunotherapy candidate 199. Diaz E, Machutta CA, Chen S, Jiang Y, Nixon C, Hofmann G, et al. proteins gp100, MART-1, and tyrosinase in human melanoma cell Development and validation of reagents and assays for EZH2 pep- lines and in human melanocytic lesions. Cancer Res 1997;57:3223–9. tide and nucleosome high-throughput screens. J Biomol Screen 178. Sotillo E, Barrett DM, Black KL, Bagashev A, Oldridge D, Wu G, 2012;17:1279–92. et al. Convergence of Acquired Mutations and Alternative Splicing 200. Liao Y, Smyth GK, Shi W. featureCounts: an efficient general pur- of CD19 Enables Resistance to CART-19 Immunotherapy. Cancer pose program for assigning sequence reads to genomic features. Discov 2015;5:1282–95. Bioinformatics 2014;30:923–30.

JUNE 2017 CANCER DISCOVERY | OF22

Epigenomic Promoter Alterations Amplify Gene Isoform and Immunogenic Diversity in Gastric Adenocarcinoma

Aditi Qamra, Manjie Xing, Nisha Padmanabhan, et al.

Cancer Discov Published OnlineFirst March 20, 2017.

Updated version Access the most recent version of this article at: doi:10.1158/2159-8290.CD-16-1022

Supplementary Access the most recent supplemental material at: Material http://cancerdiscovery.aacrjournals.org/content/suppl/2017/03/18/2159-8290.CD-16-1022.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerdiscovery.aacrjournals.org/content/early/2017/05/05/2159-8290.CD-16-1022. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.