Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Biology of Human Tumors Clinical Cancer Research Prognostic Impact of Novel Molecular Subtypes of Small Intestinal Neuroendocrine Tumor Anna Karpathakis1,2, Harpreet Dibra1, Chistodoulos Pipinikas1, Andrew Feber1, Tiffany Morris1, Joshua Francis3,4, Dahmane Oukrif1, Dalvinder Mandair1,2, Marinos Pericleous2, Mullan Mohmaduvesh2, Stefano Serra5, Olagunju Ogunbiyi2, Marco Novelli1, TuVinh Luong2, Sylvia L. Asa5, Matthew Kulke4, Christos Toumpanakis2, Tim Meyer1,2, Martyn Caplin2, Matthew Meyerson3,4, Stephan Beck1, and Christina Thirlwell1,2

Abstract

Purpose: Small intestinal neuroendocrine tumors (SINET) are BeadChip (Illumina) array profiling (n ¼ 69), methylated DNA the commonest malignancy of the small intestine; however, immunoprecipitation sequencing (n ¼ 16), copy-number var- underlying pathogenic mechanisms remain poorly characterized. iance analysis (n ¼ 47), and Whole-Genome DASL (Illumina) Whole-genome and -exome sequencing has demonstrated that expression array profiling (n ¼ 43). SINETs are mutationally quiet, with the most frequent known Results: Based on molecular profiling, SINETs can be classified mutation in the cyclin-dependent kinase inhibitor 1B into three groups, which demonstrate significantly different pro- (CDKN1B) occurring in only 8% of tumors, suggesting that gression-free survival after resection of primary tumor (not alternative mechanisms may drive tumorigenesis. The aim of this reached at 10 years vs. 56 months vs. 21 months, P ¼ 0.04). study is to perform genome-wide molecular profiling of SINETs in Epimutations were found at a recurrence rate of up to 85%, and 21 order to identify pathogenic drivers based on molecular profiling. epigenetically dysregulated were identified, including This study represents the largest unbiased integrated genomic, CDX1 (86%), CELSR3 (84%), FBP1 (84%), and GIPR (74%). epigenomic, and transcriptomic analysis undertaken in this Conclusions: This is the first comprehensive integrated molec- tumor type. ular analysis of SINETs. We have demonstrated that these tumors Experimental Design: Here, we present data from integrated are highly epigenetically dysregulated. Furthermore, we have molecular analysis of SINETs (n ¼ 97), including whole-exome or identified novel molecular subtypes with significant impact on targeted CDKN1B sequencing (n ¼ 29), HumanMethylation450 progression-free survival. Clin Cancer Res; 1–9. 2015 AACR.

Introduction inhibitor 1B) have been identified in approximately 8% of tumors in large-scale sequencing studies (3, 4). Mutations occurring in The incidence of small intestinal neuroendocrine tumors CDKN1B are loss-of-function truncating mutations that occur (SINET) is increasing (1) and accounts for 25% of all NETs. To throughout the gene. Mutational status does not appear to cor- date, the underlying pathogenic mechanisms have been poorly relate with expression of p27 (the product of CDKN1B) characterized, and due to relatively rarity, these tumors have not and has no detectable association with clinical characteristics or been subject to large-scale integrated genomic analyses, such as survival (4). Given the low frequency of mutations in this putative The Cancer Genome Atlas (TCGA) or International Cancer tumor suppressor gene, and the absence of other obvious path- Genome Consortium (ICGC) projects. ogenetic genome alterations, we hypothesized that alternative The most frequent genomic event identified in SINETs is loss mechanisms such as epigenetic dysregulation are likely to play a of heterozygosity (LOH) at 18, occurring in 60% role in these tumors (5). to 78% of tumors (2, 3); more recently, recurrent mutations Epigenetic alterations lead to changes in gene expression in the cell cycle regulator CDKN1B (cyclin-dependent kinase without modification of the underlying DNA sequence. DNA methylation and chromatin modifications are two of the mechanisms which regulate gene expression and are frequently 1University College London, London, United Kingdom. 2The Royal Free Hospital, London, United Kingdom. 3The Broad Institute of Harvard disrupted in cancer. It has long been recognized that hyper- and MIT, Cambridge, Massachusetts. 4Dana-Farber Cancer Institute, methylation at the promoter region of a gene can lead to 5 Harvard Medical School, Boston, Massachusetts. UHN Princess reduced gene expression (6). More recently, it has been dem- Margaret Cancer Centre, Toronto, Canada. onstrated that hypermethylation over the gene body leads Note: Supplementary data for this article are available at Clinical Cancer to increased gene expression, and demethylating the gene Research Online (http://clincancerres.aacrjournals.org/). body by application of 5-aza-2-deoxycytidine leads to down- Corresponding Author: Christina Thirlwell, University College London, 72 Hunt- regulation (7). ley St, London WC1E6BT, UK. Phone: 020-76796882; Fax: 020-76796884; Epigenetic dysregulation is known to play a key role in other E-mail: [email protected] NET development, including tumors of pancreatic and bronchial doi: 10.1158/1078-0432.CCR-15-0373 origin (8, 9). Integrated molecular profiling based on methylation 2015 American Association for Cancer Research. analysis has recently been shown to identify clinically distinct

www.aacrjournals.org OF1

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research. Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Karpathakis et al.

RNA Paraffin kit). Where nucleic acids were extracted from for- Translational Relevance malin-fixed paraffin-embedded (FFPE) tissue blocks, 6 to 12 6 To date, limited understanding of the underlying biology of mm serial sections were cut and macrodissected, ensuring that SINETs has hindered personalized management. The most >80% of the sample was histologically confirmed as tumor. Fresh frequent genetic mutation (CDKN1B, affecting 8% of tissue samples were frozen in liquid nitrogen within 30 minutes of tumors) has no immediate detectable impact on clinical resection. Serial hematoxylin and eosin–stained sections were phenotype or outcome. Due to their relative rarity, SINETs evaluated to ensure >80% purity of tumor and normal specimens. have not been included in any of the international collabo- rative cancer molecular profiling efforts. We have therefore DNA mutation analysis: whole-exome and targeted CDKN1B performed a Cancer Genome Atlas style–integrated analysis of sequencing a large cohort of SINETs in order to comprehensively charac- Exome sequencing was performed by Joshua Francis on 1 mg terize their molecular profile. We identified three molecular DNA as previously described (3). In brief, exonic region capture subtypes of SINET that are associated with significant differ- was performed utilizing the Agilent V2 capture probe set, and ence in progression-free survival after surgical resection. We sequencing of 76-bp paired end reads was performed utilizing also demonstrated that SINETs are highly epigenetically dys- Illumina HiSeq2000 instruments. A median of 9.15 Gb of unique regulated and characterized a panel of 21 genes that are sequence data was generated per sample, which was aligned to the epigenetically altered in 70% to 80% of cases. These findings target exome using the Burrows–Wheeler Aligner (BWA), resulting highlight the molecular diversity of SINETs and highlight in a median 140 coverage of each base. Targeted sequencing was epigenetic mechanisms as potential drivers of tumorigenesis. performed by multiplex PCR reaction tiling of CDKN1B using 100 Novel molecular profiling could be implemented in the to 200 ng of DNA on the Fluidigm Access Array microfluidic clinical setting to facilitate personalized management and device. PCR products underwent Illumina sequencing on the improve prognosis. MiSeq instrument to a mean coverage of 800.

DNA methylation analysis: Illumina HumanMethylation450 BeadChip subgroups of lung carcinomas, including a neuroendocrine-spe- DNA from SINET samples was analyzed on the HumMeth450 cific "epitype" (10). BeadChip (HM450). The integrity and viability of FFPE samples fi Here, we present the first comprehensive integrated genome- were rst determined using the Illumina FFPE QC Kit. Ligation of fi wide molecular analysis of a large cohort of SINETs. FFPE DNA (Qiagen REPLI-g FFPE Kit) and bisul te conversion of FF and FFPE DNA (Zymo EZ DNA Methylation kit) were per- fi Materials and Methods formed as described previously (12). Quality control of bisul te conversion was performed by quantitative PCR to confirm >98% Patient recruitment conversion. Data analysis was performed using the ChAMP pipe- All patients provided written informed consent for their tissue line (13). Samples where 5% of probes failed detection P value < to be analyzed in this study, which was approved by NHS Camden 0.01 filtering were excluded from further analysis. Normalization and Islington Community Research Ethics Committee (Ref: 09/ with BMIQ and batch correction with ComBat were performed H0722/27). A total of 97 tumors and 25 normal tissue samples according to pipeline defaults. The comparability of array-based were analyzed from 85 individuals. Inclusion criteria were histo- methylation analysis of FFPE and FF tissues has previously been pathologic diagnosis of SINET, and surgical resection of either demonstrated (12). primary tumor or liver metastasis. Exclusion criteria were biopsy CpG island methylator phenotype (CIMP) status was defined sampling of tumor only (due to inadequate amounts of DNA by calculating the median methylation score across all CpG island which was obtained from biopsy samples). Subjects were includ- probes on the HM450: CIMP positive: sample demonstrates >5% ed from three international sites (UK, the United States, and greater median methylation than median score across all tumors; Canada), and recruitment was performed during the period of CIMP negative: >5% less than overall median score. Samples 2011 to 2013. Anonymized clinical information was retrospec- where total CpG methylation score is within 5% of cohort median tively collected from patient records where available including are unclassified. histopathologic details (tumor grade and stage) and progression- free survival (PFS) following resection. PFS was used as the DNA methylation analysis: methylation DNA primary endpoint for prognosis. immunoprecipitation sequencing DNA from 16 tumor and three normal samples underwent Pathology review whole-genome methylation profiling using methylation DNA All cases were reviewed by two expert NET pathologists (T.V. immunoprecipitation sequencing (MeDIP-Seq) as previously Luong, M. Novelli, S.L. Asa, or S. Serra). The European NET Society described (14). Briefly, 1 mg of DNA was sonicated to achieve an grading system was used, which utilizes the Ki-67 proliferation average fragment length of 300 bp. Fragmented DNA underwent index. Low-grade tumors have a Ki-67 <2%, intermediate-grade end repair (NEBNext End Repair Module), dA-tailing (NEBNext tumors have a Ki-67 of 2% to 20%, and high-grade tumors have a dA-Tailing Module), and adapter ligation. DNA purification fol- Ki-67 >20% (11). lowing each step was performed using Ampure XP magnetic beads (Agencourt). Automated MeDIP was performed according to the Nucleic acid extraction Diagenode protocol and utilizing Diagenode-positive and -neg- DNA and RNA were extracted using standard methods (Qiagen: ative DNA controls. Amplification was performed using NebNext QIAamp DNA Mini kit and RNeasy Mini kit; Roche: High Pure High Fidelity 2 PCR Master Mix (NEB); size selection was

OF2 Clin Cancer Res; 2015 Clinical Cancer Research

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research. Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Novel Molecular Subtypes of Intestinal NET

performed by gel excision of 300 to 350 fragments. noma, pancreatic adenocarcinoma; healthy: colon, stomach, Assessment of fold enrichment of regions of known methylation rectum, pancreas) were downloaded and analyzed using the status was performed. Sequencing was performed on the Illumina Marmal-aid webtool (19). Normalization was performed using Hi Seq 2500 in High Output mode (UCL Genomics). MeDIP-seq the ChAMP pipeline. data were analyzed using the custom pipeline MeDUSA (v2.0; ref. 15). BWA was used to align paired end sequence data to the Analysis of publicly available data: expression fi reference , and ltering was performed to remove Affymetrix Human Genome U133 Plus 2.0 Array data were reads that were unable to be aligned as a viable pair and also those downloaded from the Gene Expression Omnibus (GEO) data- pairs in which neither read scored an alignment score of 10, base, accession code GSE9576 (20). Data processing was per- resulting in a mean of 59 million unique reads per sample. formed using GEO2R (21).

Gene expression profiling: Illumina Whole-Genome DASL Integrated pathway analyses assay Pathway analyses were performed on WebGestalt (22). RNA was assessed using the Agilent Bioanalyzer; samples with RNA concentrations of >40 ng/mL and RNA Integrity Number (RIN) > 1.5 were taken forward for analysis. The whole-genome Availability of data cDNA-mediated annealing, selection and ligation (WG-DASL; Raw data from this study will be deposited in GEO (accession Illumina) assay was performed following the manufacturer's number: GSE73832). instructions (UCL Genomics). Quantile normalization and back- ground correction were performed in an Illumina GenomeStudio Results software. Samples where 70% probes failed detection P value < Genome-wide DNA methylation, gene expression, and copy- 0.05 filtering were excluded from further analysis. The limma number variance (CNV) profiling were performed on 97 SINET package was used to identify differentially expressed probes (16). tumors (primaries and liver metastases) collected from 85 individuals. A flowchart demonstrating the analyses the sam- Copy-number variance analysis ples underwent is shown in Fig. 1. The majority of analyses were Genome-wide copy-number variance was inferred from performed on 72 primary SINETs with 25 liver metastases HM450 signal intensity data within the ChAMP pipeline (17). utilized as a validation cohort. Of the 72 SINET primary tumor Whole-chromosome arm copy number was determined using an cases, there was a preponderance of females (34 vs. 25 males, 80% length threshold and a segment mean threshold of 0.2. gender unknown in 13 cases). Median age at diagnosis was 62 Analysis of focal copy-number alterations was performed using years. Where site of primary within the small intestine was the GISTIC 2.0 algorithm (18). documented, almost all were of ileal origin (40 vs. 4 duodenal/ jejunal, 28 not specified). Clinical information regarding Ki-67 Analysis of publicly available data: methylation and staging was incomplete in some cases, but where available Raw methylation data from TCGA cohorts (cancer: colon the majority had metastatic disease (28 vs. 9 localized disease), adenocarcinoma, rectum adenocarcinoma, stomach adenocarci- and the majority were low-grade tumors (grade 1: n ¼ 34, grade

Total cohort Cases = 85 SINET primary tumors n = 72 Liver metastases n = 25

Analysis performed Figure 1. Flowchart describing number of specimens included in each stage of Methylaon analysis HM450 primary n = 49 Expression analysis molecular analysis. HM450, SCNA analysis: DASL primary n = 31 Mutaon analysis HM450 liver mets n = 20 HM450 primary n = 47 Primary n = 29 DASL liver mets n = 12 HumanMethylation450 BeadChip assay MeDIP primary n = 16 (2 samples failed and were removed from downstream analysis); SCNA, somatic copy-number variance analysis; DASL, whole-genome cDNA- mediated annealing, selection, and Detailed analyses ligation assay (1 sample failed and was removed from downstream analysis). Molecular subgroupings Epigenec drivers Meth & SCNA: primary n = 47 Meth primary n = 49 Mutaon: primary n = 29 Expn primary n = 31 Expn: primary n = 17 Survival: primary n = 32 Panel validaon: GIPR validaon: Methylaon MeDIP n = 16 HM450 liver mets n = 20 Expn (external) n = 6 DASL liver mets n = 11

www.aacrjournals.org Clin Cancer Res; 2015 OF3

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research. Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Karpathakis et al.

2: n ¼ 14, grade 3: n ¼ 2; Supplementary Table S2). Twenty- CDKN1B mutations were located within subgroup A (chr18 LOH) nine cases from the cohort had previously been sequenced for tumors. CDKN1B mutations; three of 29 (10%) of these primary SINETs Unsupervised analysis determined that the epigenetic profile of harbored detectable mutations (3). SINETs also clearly differentiates the three groups identified by CNV analysis (Fig. 2B). Supervised analysis of 275 genes differ- Novel molecular subgrouping of SINETs entially methylated [>30% differentially methylated (Db), P < CNV analysis identified recurrent whole-chromosome/whole 0.001] between the three subgroups (Fig. 2A; Supplementary arm alterations in SINET primary tumors [n ¼ 47; chr4 (gain, 19% Table S1) found that significant differential methylation between of cases), chr5 (gain, 11%), chr18 (loss, 64%), and chr20 (gain, subgroups was observed in EGFR, mTOR, and VEGF signaling 17%)], consistent with previous reports (23). Unsupervised hier- pathways. archical clustering of whole-chromosome/arm-level CNVs iden- Expression analysis was performed to identify differentially tified three distinct molecular subgroups of SINET primary expressed genes between the three subgroups where matched data tumors. Group A harbored chr18 LOH only (18LOH group, were available (n ¼ 17), a panel of 41 probes (unadj P < 0.001) 55% of tumors), group B no large copy-number alterations was identified which distinguished these groups with 88% accu- (NoCNV group, 19%), and group C harbored multiple copy- racy (15/17 cases; Supplementary Fig. S3). Finally, a CIMP status number alterations (including gain of 4, 5, and 20; MultiCNV was defined in SINETs based on overall methylation across all group, 26%; Fig. 2A). Of note, all three tumors that harbored analyzed CpG sites (CIMP positive n ¼ 18, CIMP negative n ¼ 13).

Figure 2. SINETs may be subdivided into three groups based on copy-number alterations, epigenetic profile, and CDKN1B status. A, three subgroups of SINET primary tumors are identified based on unsupervised hierarchical clustering of whole-chromosome/arm-level somatic copy-number variance profiles. CDKN1B mutations are found only in subgroup A (chr18 LOH). B, unsupervised hierarchical clustering of the methylation status of the top 500 most variable probes across the SINET primary tumor cohort differentiates three subgroups of SINET distinguished by CNV status. C, PFS analysis following surgical resection of primary tumor demonstrates poorer prognosis in the NoCNV and MultiCNV subgroups compared with chr18 LOH subgroup. D, the methylation profile of 21 epigenetically dysregulated genes distinguishes SINET from other GI malignancies and normal GI tissue. Abbreviations: 18LOH, LOH chromosome 18; MultiCNV, multiple arm-level copy-number alterations; NoCNV, no arm-level copy-number alterations; OS, overall survival; TvN, tumor/normal classification.

OF4 Clin Cancer Res; 2015 Clinical Cancer Research

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research. Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Novel Molecular Subtypes of Intestinal NET

A significant association between CIMP negativity and chr18 LOH was performed by MeDIP-seq of an independent cohort of 16 SI was noted (Fisher exact test, P < 0.032). primary NET samples. Validation of altered expression of the 23 Thirty-two samples classified by molecular subtype were ana- candidates was performed using publicly available data (GEO lyzed for PFS following surgical resection of primary tumor. After accession number: GSE9576; ref. 20). These confirmatory steps a median follow-up of 56 months, average PFS across all 32 endorsed the validity of 91% (21 of 23) of the proposed panel of samples was 40.6 months. Using a Cox proportional hazard candidate epigenetically regulated genes in SINET pathogenesis model, a difference in the PFS was identified between the sub- (Table 2; Supplementary Fig. S3). Altered methylation in at least groups with the 18LOH group having superior PFS (median not 10 of these 21 genes was demonstrated in 65% of tumor speci- reached at 10-year follow-up) when compared with both the mens, demonstrating a significant rate of epigenome alterations. NoCNV group (56 months, P ¼ 0.10) and the MultiCNV group When the top 5 most frequently altered candidates from this panel (21 months, P ¼ 0.02; Fig. 1C). Interestingly, the two high-grade [caudal type homeobox 1 (CDX1), fructose-1,6-bisphosphatase 1 tumors in the cohort were both included in the MultiCNV poor (FBP1), transmembrane protein 171 (TMEM171), ganglioside- prognosis group; however, intermediate- and low-grade tumors induced differentiation associated protein 1 like 1 (GDAP1L1), did not segregate by subgroup. This indicates that this molecular and cadherin, EGF LAG seven-pass G-type receptor (CELSR3)] classification may have utility as an adjunct to Ki-67 in order to were selected, 82% of tumor specimens demonstrate altered refine prognostication. A trend for younger age at diagnosis was methylation in at least 4 of the 5 candidates. identified on comparison of the 18LOH group (median age 67) Utilizing publicly available TCGA methylation data from and the MultiCNV group (median age 54, P ¼ 0.06). The key other gastrointestinal (GI) tumor samples, we evaluated the characteristics of these molecular subgroups of SINET primary methylation status of the panel of 21 candidate genes across tumors are summarized in Table 1. a range of tumors. We determined that the methylation signa- ture of this panel robustly distinguished SINETs from other GI Identification of epigenetic drivers of NET pathogenesis malignancies and normal GI tissue (positive predictive value Methylation profiling of SINET primary tumors (n ¼ 49) 95.8%; Fig. 1D). identified 130,083 methylation variable positions [MVP; Benja- Of particular interest is hypermethylation of the gastric mini–Hochberg (BH) adj P < 0.0005] differentiating tumor and inhibitory polypeptide receptor gene [GIPR;whichwasseen normal tissues. Of these MVPs, 5,514 (4.2%) were hypomethy- in 74% of SINET (median methylation 0.67 vs. normal 0.29, lated in tumor by >30% and 1,841 (1.4%) were hypermethylated P < 2.2e 16)]. To investigate the status of GIPR further, DNA in tumor by >30%. KEGG (Kyoto Encyclopedia of Genes and methylation and gene expression profiling were performed on a Genomes) pathway analysis of 2,749 differentially methylated discretesetof20(methylation)and11(expression)SINET genes identified enrichment of multiple cancer-related pathways, liver metastases. This confirmed that promoter hypomethyla- including MAPK, Wnt, and PI3K–mTOR signaling pathways (BH tion,genebodyhypermethylation(Fig.3A),andincreased adj P < 0.0005; Supplementary Fig. S2). expression (Fig. 3B) of GIPR were positively correlated with Gene expression profiling of SINET primary tumors (n ¼ 31) the development of liver metastases. Finally, through compar- identified 2,016 genes which were differentially expressed ison with TCGA data, it was determined that hypermethylation between tumor and normal (BH adj P < 0.05), of these, 1,300 at GIPR was sensitive and specific for the detection of demonstrated a minimum 2-fold average difference in expression. SINET compared with other GI malignancies with an AUC of Disease association analysis of these genes identified significant 0.991 (95% confidence interval, 0.991–0.999; Fig. 3C). Of enrichment of intestinal and neoplastic disease processes, as well note, in addition to GIPR, two further G protein–coupled as endocrine and neurological system disorders. receptor genes (prolactin releasing hormone receptor (PRLHR) Integration of DNA methylation and gene expression data and CELSR3) and three enzymes [pancreatic lipase related identified 61 genes where altered methylation was associated protein 2 (PNLIPRP2), protein tyrosine phosphatase receptor with a change in expression (Supplementary Table S3). Of these type N (PTPRN), and proprotein convertase subtilisin/kexin genes, 23 were found to be altered in the majority of SINETs. type 1 (PCSK1)] represent potentially druggable targets within Validation of methylation changes in this panel of 23 candidates the 21-gene panel.

Table 1. Characteristics of three molecular subgroups of SINET primary tumors Subgroup A B C Tumors, n (% of total) 26 (55%) 9 (19%) 12 (26%) CNV profile 18LOH No CNV Multiple CNV (18LOH and amplifications 4, 5, 20) Median arm-level CNVs per tumor (range) 2 (1–6) 0 (0–0) 6 (2–11) CDKN1B mutation rate (% of total) 3 mut (11.5%) 0 mut (0%) 0 mut (0%) 14 wt (53.8%) 7 wt (77.7%) 5 wt (41.7%) CIMP positive 8 (30.8%) 5 (55.6%) 5 (41.7%) CIMP negative 11 (42.3%) 1 (11.1%) 1 (8.3%) Median age at diagnosis, y 67 60 54 Stage, metastatic (%) 8 (30.8%) 3 (33.3%) 5 (41.7%) Stage, localized (%) 3 (11.5%) 2 (22.2%) 2 (16.7%) PFS after resection (months) Not reached 56 21 NOTE: Three novel molecular subgroups of SINET are defined by distinct copy number and epigenetic profiles, with prognostic impact on PFS following resection of primary tumor.

www.aacrjournals.org Clin Cancer Res; 2015 OF5

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research. Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Karpathakis et al.

Table 2. Candidate epigenetically regulated global drivers of NET development: a panel of 23 genes, 21 of which are validated in independent cohorts and by alternate techniques Differential methylation Differential expression Normal % SINETs MeDIP Normal Fold % SINETs GEO Region SI SINET Db P affected validation SI SINET change P affected validation Downregulation CDX1 Promoter 0.55 0.90 0.35 <0.001 85.7 Significant 5,347 1,541 3.5 <0.001 80.6% Significant C20orf54 Promoter 0.34 0.64 0.29 <0.001 53.1 Significant 2,315 663 3.5 0.006 83.9% Significant EFNA2 Promoter 0.36 0.56 0.20 <0.001 34.7 Significant 3,384 661 5.1 <0.001 93.5% Not validated FBP1 Promoter 0.43 0.82 0.38 <0.001 83.7 Significant 10,591 2,195 4.8 <0.001 96.8% Significant GATA5 Body 0.89 0.69 0.20 <0.001 36.7 Trend 735 179 4.1 <0.001 96.8% Significant NGEF Body 0.89 0.72 0.16 <0.001 20.4 Significant 3,753 599 6.3 <0.001 96.8% Significant PNLIPRP2 Body 0.77 0.47 0.30 <0.001 63.3 Significant 1,457 341 4.3 0.007 90.3% Trend TMEM171 Promoter 0.18 0.61 0.43 <0.001 85.7 Significant 3,570 607 5.9 <0.001 90.3% Significant TRIM15 Promoter 0.53 0.87 0.33 <0.001 63.3 Trend 696 206 3.4 0.001 100.0% Significant Upregulation CELSR3 Body 0.10 0.57 0.47 <0.001 83.7 Significant 2,037 7,664 3.8 0.007 87.1% Significant CNTNAP5 Promoter 0.26 0.09 0.17 <0.001 28.5%a Significant 380 4,258 11.2 0.003 93.5% Significant C3orf14 Promoter 0.32 0.12 0.20 <0.001 20.4%a Significant 316 1,074 3.4 0.006 83.9% Significant DSCAM Promoter 0.50 0.15 0.35 <0.001 34.7 Significant 175 1,194 6.8 0.035 87.1% Significant GDAP1L1 Promoter 0.38 0.08 0.30 <0.001 79.6 Trend 188 736 3.9 0.010 93.5% Significant GIPR Body 0.30 0.66 0.36 <0.001 73.5 Significant 349 1,915 5.5 0.012 83.9% Significant KCNH6 Body 0.07 0.36 0.29 <0.001 69.4 Significant 1,058 4,879 4.6 0.007 83.9% Significant LMX1B Body 0.30 0.59 0.28 <0.001 55.1 Significant 386 3,450 8.9 0.003 87.1% Trend PCSK1 Promoter 0.38 0.08 0.29 <0.001 67.3 Trend 322 2,413 7.5 0.033 93.5% Significant PRLHR Promoter 0.44 0.13 0.31 <0.001 55.1 Trend 300 3,843 12.8 0.027 93.5% Trend PTPRN Promoter 0.38 0.11 0.27 <0.001 34.7 Trend 446 4,236 9.5 0.009 90.3% Significant RUNDC3A Body 0.08 0.42 0.34 <0.001 65.3 Significant 544 3,898 7.2 0.016 83.9% Significant SCGN Promoter 0.18 0.10 0.08 <0.001 71.4%b Not validated 981 6,126 6.2 0.021 87.1% Significant SNTG1 Promoter 0.28 0.07 0.21 <0.001 28.5%a Trend 193 2,131 11.1 0.002 93.5% Significant NOTE: A large proportion of SINETs are affected by greater than 25% differential methylation, and greater than 2-fold alteration in expression of 23 genes. Abbreviation: Db, differential methylation. a>20% Db; b>10% Db.

Epigenetically dysregulated genes on chromosome 18 most favorable PFS (not reached at 10-year follow-up) after The most common subgroup of SINETs was group A chr18 resection and an older age at diagnosis suggesting a less aggressive LOH (26/47, 55% of our cohort). Historically, this has also been phenotype. A second subgroup is characterized by the absence of the most frequently described molecular alteration in these arm-level CNVs and is associated with a high level of CIMP tumors; however, tumorigenic drivers from this region have not positivity and an intermediate PFS (56 months). The final sub- been identified to date. We therefore interrogated chr18 to iden- group comprises 26% of tumors and is characterized by the tify epigenetic changes which may represent the "second hit" presence of multiple CNVs, a significantly poorer PFS (21 following LOH. In total, 1,043 probes on chr18 demonstrated months) and younger age at onset suggesting a more aggressive differential methylation (P < 0.05; Db >10%) between tumors clinical phenotype. The prognostic impact of these newly defined affected by chr18LOH and normal SI (Supplementary Table S4). subgroups requires validation in an independent cohort; howev- Furthermore, 25 genes were found to be differentially expressed er, we suggest that these novel molecular classifications may be (adj P < 0.05) between chr18LOH tumors and normal SI. From utilized as an adjunct to Ki-67 to optimize accurate prognostica- this list, several genes on chr18 where epigenetic changes were tion for patients. Furthermore, SINETs demonstrating the Mul- associated with reduced expression were identified including: tiCNV profile may be considered for trials of more aggressive laminin alpha 3 (LAMA3), serpin peptidase inhibitor clade B management. member 5 (SERPINB5), and tumor necrosis factor receptor super- The association between chr18LOH and CIMP negativity sug- family member 11a NFKB activator (RANK or TNFRSF11A). gests that aberrant methylation plays an even greater role in tumors lacking the characteristic chr18 LOH, and is supported by previously published data in SINET and colorectal adenocar- Discussion cinoma (24, 25). Here, we present the first comprehensive integrated molecular We have demonstrated that SINETs are epigenetically dysregu- analysis of SINETs, characterizing DNA mutation, methylation, lated tumors and are therefore may be candidates for trials of and RNA expression profiles of a large cohort of tumors. Despite novel targeted epigenetic therapies. This is in keeping with prior being a relatively small and heterogeneous cohort when com- molecular analyses of pancreatic and bronchial NETs demon- pared with other TCGA-type studies, it comprises the largest strating epigenetic dysregulation in these populations. A panel of integrated genomic analysis in this tumor type. 21 epigenetically dysregulated genes has been identified and We have identified three subgroups of SINET distinguished by validated. Epigenetic changes affecting this panel appear to be molecular profiling. The largest group is defined by chr18LOH specific to SINETs with respect to gastrointestinal adenocarcino- and is associated with the presence of CDKN1B mutations, and mas. In particular, aberrant methylation of GIPR is proposed as a CIMP negativity. Patients classified within this subgroup have the mechanism of upregulation of this gene. GIPR overexpression in

OF6 Clin Cancer Res; 2015 Clinical Cancer Research

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research. Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Novel Molecular Subtypes of Intestinal NET

Figure 3. GIPR is epigenetically dysregulated in SINETs but not in other gastrointestinal malignancies. A, in SINETs, GIPR is hypomethylated in the promoter region and hypermethylated at the gene body, both changes known to be associated with increased gene expression. Progressive aberrant methylation in liver metastases is identified compared with primary tumors. B, GIPR is significantly overexpressed in SINETs, and further increased expression is noted in liver metastases. C, the methylation profile of GIPR distinguishes SINET (hypermethylated across the gene body) from other gastrointestinal malignancies and normal tissue, which show an absence of methylation at this locus. Abbreviations: 3U, 30UTR; 5U, 50UTR; CpG I, CpG island; LM, liver metastases; NSI, normal small intestine; SI prim, small intestinal neuroendocrine primary tumor; TSS1500, within 1,500 bp from transcription start site; TSS200, within 200 bp of transcription start site.

SINETs has been reported previously, and radiolabeled ligands are previously been described in NETs but is a feature of gastric in development as novel diagnostic agents for NETs particularly adenocarcinoma, prostate cancer, and breast cancer, and in the those not expressing somatostatin receptors (26); however, this is case of breast and prostate cancers, hypermethylation of the gene the first description of aberrant methylation being associated with is associated with poorer clinical outcomes (29–31). SERPINB5 GIPR upregulation in NETs. The success of novel radioligands has been described as a "metastasis suppressor gene," and hyper- suggests that therapeutic agents targeted to the GIPR receptor methylation of this gene is associated with metastatic potential in could be developed and may increase the range of molecularly pancreatic adenocarcinoma and poorer clinical outcomes in targeted therapeutics available for use in NET patients. The CDX1- breast cancer (32, 33). Methylation of SERPINB5 in NETs has encoded transcription factor plays a key role in intestinal differ- previously been demonstrated by Verdugo and colleagues; how- entiation, and CDX1 has been identified in both colorectal and ever, determination of hypermethylation compared with normal gastric cancers as a tumor suppressor gene where downregulation tissue was not performed (34). is secondary to hypermethylation (27). FBP1 undergoes promoter This study has been performed as a comprehensive basic hypermethylation in colorectal cancer and hepatocellular carci- molecular profiling experiment. It is therefore limited by a lack noma but has not previously been associated with NETs (28). of functional validation studies. In SINETs, such studies are These are examples of genes within the 21-gene panel, which difficult to perform due to the lack of validated cell lines or represent epigenetically dysregulated downregulated genes in animal models of these tumors. These data remain exploratory SINETs. at this stage and require further validation in a separate cohort. Finally, we have identified genes found on chromosome 18 This would allow for external validation of the panel of 21 genes which are affected by a "second hit" of epigenetic dysregulation identified in this study as the BH FDR approach used allows for following LOH of this region. LAMA3 hypermethylation has not some false-positive results.

www.aacrjournals.org Clin Cancer Res; 2015 OF7

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research. Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Karpathakis et al.

External validation of methylation profiling would be pre- M. Pericleous, O. Ogunbiyi, T.V. Luong, S.L. Asa, M. Kulke, C. Toumpanakis, ferred; however, there are no publicly available comparable T. Meyer, M. Caplin, C. Thirlwell datasets, emphasizing the importance of all investigators making Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Karpathakis, T. Morris, M. Novelli, S.L. Asa, available results of molecular studies in rare tumors such as C. Thirlwell SINETs. Writing, review, and/or revision of the manuscript: A. Karpathakis, C. Pipi- In summary, we have performed the largest unbiased integrated nikas, A. Feber, T. Morris, S. Serra, O. Ogunbiyi, T.V. Luong, S.L. Asa, M. Kulke, genomic, epigenomic, and transcriptomic profile of small intes- C. Toumpanakis, T. Meyer, M. Caplin, S. Beck, C. Thirlwell tinal NETs yet reported. The findings presented here form the basis Administrative, technical, or material support (i.e., reporting or organizing for a novel molecular classification of SINET tumors in the clinical data, constructing databases): H. Dibra, C. Pipinikas, J. Francis, D. Oukrif, M. Mohmaduvesh, S. Serra setting and underscore the feasibility of stratifying NET patients by Study supervision: C. Thirlwell molecular features. These mutationally quiet tumors have abun- Other (histopathology/immunohistochemistry analysis and interpretation): dant epigenetic changes, which include putative drivers of M. Novelli tumorigenesis. Grant Support fl Disclosure of Potential Con icts of Interest This study was supported by Cancer Research UK and the National Institute M. Meyerson reports receiving a commercial research grant from Bayer, and for Health Research through the UCL Experimental Cancer Medicine Centre and holds ownership interest (including patents) in and is a consultant/advisory UCL Hospitals Biomedical Research Centre. Specimen biobanking was sup- board member for Foundation Medicine. C. Thirlwell is a consultant/advisory ported by the Raymond and Beverly Sackler Foundation. The Beck lab contri- board member for Ipsen. No potential conflicts of interest were disclosed by the bution was supported by a Royal Society Wolfson Research Merit Award other authors. (WM100023) and EU-FP7 projects EPIGENESYS (257082) and BLUEPRINT (282510). Authors' Contributions The costs of publication of this article were defrayed in part by the payment of advertisement Conception and design: A. Karpathakis, M. Meyerson, S. Beck, C. Thirlwell page charges. This article must therefore be hereby marked in Development of methodology: A. Karpathakis, D. Oukrif, M. Novelli, S. Beck, accordance with 18 U.S.C. Section 1734 solely to indicate this fact. C. Thirlwell Acquisition of data (provided animals, acquired and managed patients, Received February 13, 2015; revised June 25, 2015; accepted June 25, 2015; provided facilities, etc.): A. Karpathakis, H. Dibra, J. Francis, D. Mandair, published OnlineFirst July 13, 2015.

References 1. Yao JC, Hassan M, Phan A, Dagohoy C, Leary C, Mares JE, et al. One affin-embedded tissue using the Illumina Infinium HumanMethylation27 hundred years after "carcinoid": epidemiology of and prognostic factors for BeadChip. Methods 2010;52:248–54. neuroendocrine tumors in 35,825 cases in the United States. J Clin Oncol 13. Morris TJ, Butcher LM, Feber A, Teschendorff AE, Chakravarthy AR, Woj- 2008;26:3063–72. dacz TK, et al. ChAMP: 450k chip analysis methylation pipeline. Bioin- 2. Kulke MH, Freed E, Chiang DY, Philips J, Zahrieh D, Glickman JN, et al. formatics 2014;30:428–30. High-resolution analysis of genetic alterations in small bowel carcinoid 14. Butcher LM, Beck S. AutoMeDIP-seq: a high-throughput, whole genome, tumors reveals areas of recurrent amplification and loss. Genes Chromo- DNA methylation assay. Methods 2010;52:223–31. somes Cancer 2008;47:591–603. 15. Wilson GA, Dhami P, Feber A, Cortazar D, Suzuki Y, Schulz R, et al. 3. Francis JM, Kiezun A, Ramos AH, Serra S, Pedamallu CS, Qian ZR, et al. Resources for methylome analysis suitable for gene knockout studies of Somatic mutation of CDKN1B in small intestine neuroendocrine tumors. potential epigenome modifiers. Gigascience 2012;1:3. Nat Genet 2013;45:1483–6. 16. Smyth GK. Linear models and empirical bayes methods for assessing 4. Crona J, Gustavsson T, Norlen O, Edfeldt K, Akerstrom T, Westin G, et al. differential expression in microarray experiments. Stat Appl Genet Mol Somatic mutations and genetic heterogeneity at the CDKN1B locus in Biol 2004;3:Article3. small intestinal neuroendocrine tumors. Ann Surg Oncol 2015 Jan 14. 17. Feber A, Guilhamon P, Lechner M, Fenton T, Wilson GA, Thirlwell C, et al. [Epub ahead of print]. Using high-density DNA methylation arrays to profile copy number 5. Versteeg R. Cancer: tumours outside the mutation box. Nature 2014; alterations. Genome Biol 2014;15:R30. 506:438–9. 18. Mermel CH, Schumacher SE, Hill B, Meyerson ML, Beroukhim R, Getz G. 6. Esteller M. Epigenetics in cancer. N Engl J Med 2008;358:1148–59. GISTIC2.0 facilitates sensitive and confident localization of the targets of 7. Yang X, Han H, De Carvalho DD, Lay FD, Jones PA, Liang G. Gene body focal somatic copy-number alteration in human cancers. Genome Biol methylation can alter gene expression and is a therapeutic target in cancer. 2011;12:R41. Cancer Cell 2014;26:577–90. 19. Lowe R, Rakyan VK. Marmal-aid–a database for Infinium HumanMethyla- 8. Fernandez-Cuesta L, Peifer M, Lu X, Sun R, Ozretic L, Seidel D, et al. tion450. BMC Bioinformatics 2013;14:359. Frequent mutations in chromatin-remodelling genes in pulmonary carci- 20. Leja J, Essaghir A, Essand M, Wester K, Oberg K, Totterman TH, et al. Novel noids. Nat Commun 2014;5:3518. markers for enterochromaffin cells and gastrointestinal neuroendocrine 9. Jiao Y, Shi C, Edil BH, de Wilde RF, Klimstra DS, Maitra A, et al. DAXX/ carcinomas. Mod Pathol 2009;22:261–72. ATRX, MEN1, and mTOR pathway genes are frequently altered in pancre- 21. Barrett T, Wilhite SE, Ledoux P, Evangelista C, Kim IF, Tomashevsky M, et al. atic neuroendocrine tumors. Science 2011;331:1199–203. NCBI GEO: archive for functional genomics data sets–update. Nucleic 10. Karlsson A, Jonsson M, Lauss M, Brunnstrom H, Jonsson P, Borg A, et al. Acids Res 2013;41:D991–5. Genome-wide DNA methylation analysis of lung carcinoma reveals one 22. Wang J, Duncan D, Shi Z, Zhang B. WEB-based GEne SeT AnaLysis neuroendocrine and four adenocarcinoma epitypes associated with patient Toolkit (WebGestalt): update 2013. Nucleic Acids Res 2013;41: outcome. Clin Cancer Res 2014;20:6127–40. W77–83. 11. Rindi G, Kloppel G, Couvelard A, Komminoth P, Korner M, Lopes JM, 23. Hashemi J, Fotouhi O, Sulaiman L, Kjellman M, Hoog A, Zedenius J, et al. et al. TNM staging of midgut and hindgut (neuro) endocrine tumors: a Copy number alterations in small intestinal neuroendocrine tumors deter- consensus proposal including a grading system. Virchows Arch mined by array comparative genomic hybridization. BMC Cancer 2013; 2007;451:757–62. 13:505. 12. Thirlwell C, Eymard M, Feber A, Teschendorff A, Pearce K, Lechner M, et al. 24. Fotouhi O, Fahmideh MA, Kjellman M, Sulaiman L, Hoog A, Zedenius J, Genome-wide DNA methylation analysis of archival formalin-fixed par- et al. Global hypomethylation and promoter methylation in small

OF8 Clin Cancer Res; 2015 Clinical Cancer Research

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research. Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Novel Molecular Subtypes of Intestinal NET

intestinal neuroendocrine tumors: an in vivo and in vitro study. Epigenetics inactivation of laminin alpha3 chain in gastric cancer. Int J Oncol 2014;9:987–97. 2011;39:593–9. 25. Ogino S, Kawasaki T, Kirkner GJ, Ohnishi M, Fuchs CS. 18q loss of 30. Sathyanarayana UG, Padar A, Suzuki M, Maruyama R, Shigematsu H, heterozygosity in microsatellite stable colorectal cancer is correlated with Hsieh JT, et al. Aberrant promoter methylation of laminin-5-encoding CpG island methylator phenotype-negative (CIMP-0) and inversely with genes in prostate cancers and its relationship to clinicopathological fea- CIMP-low and CIMP-high. BMC Cancer 2007;7:72. tures. Clin Cancer Res 2003;9:6395–400. 26. Gourni E, Waser B, Clerc P, Fourmy D, Reubi JC, Maecke HR. The glucose- 31. Sathyanarayana UG, Padar A, Huang CX, Suzuki M, Shigematsu H, Bekele dependent insulinotropic polypeptide receptor: a novel target for neuro- BN, et al. Aberrant promoter methylation and silencing of laminin-5- endocrine tumor imaging–first preclinical studies. J Nucl Med 2014; encoding genes in breast carcinoma. Clin Cancer Res 2003;9:6389–94. 55:976–82. 32. Mardin WA, Petrov KO, Enns A, Senninger N, Haier J, Mees ST. SERPINB5 27. Rau TT, Rogler A, Frischauf M, Jung A, Konturek PC, Dimmler A, et al. and AKAP12 - expression and promoter methylation of metastasis suppres- Methylation-dependent activation of CDX1 through NF-kappaB: a link sor genes in pancreatic ductal adenocarcinoma. BMC Cancer 2010;10:549. from inflammation to intestinal metaplasia in the human stomach. Am J 33. Cheol Kim D, Thorat MA, Lee MR, Cho SH, Vasiljevic N, Scibior-Bent- Pathol 2012;181:487–98. kowska D, et al. Quantitative DNA methylation and recurrence of breast 28. Chen M, Zhang J, Li N, Qian Z, Zhu M, Li Q, et al. Promoter hypermethyla- cancer: a study of 30 candidate genes. Cancer Biomark 2012;11:75–88. tion mediated downregulation of FBP1 in human hepatocellular carcino- 34. Verdugo AD, Crona J, Starker L, Stalberg P, Akerstrom G, Westin G, et al. ma and colon cancer. PLoS One 2011;6:e25564. Global DNA methylation patterns through an array-based approach in 29. Ii M, Yamamoto H, Taniguchi H, Adachi Y, Nakazawa M, Ohashi H, et al. small intestinal neuroendocrine tumors. Endocr Relat Cancer 2014;21: Co-expression of laminin beta3 and gamma2 chains and epigenetic L5–7.

www.aacrjournals.org Clin Cancer Res; 2015 OF9

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research. Published OnlineFirst July 13, 2015; DOI: 10.1158/1078-0432.CCR-15-0373

Prognostic Impact of Novel Molecular Subtypes of Small Intestinal Neuroendocrine Tumor

Anna Karpathakis, Harpreet Dibra, Chistodoulos Pipinikas, et al.

Clin Cancer Res Published OnlineFirst July 13, 2015.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-15-0373

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2015/07/14/1078-0432.CCR-15-0373.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://clincancerres.aacrjournals.org/content/early/2015/10/23/1078-0432.CCR-15-0373. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.

Downloaded from clincancerres.aacrjournals.org on September 28, 2021. © 2015 American Association for Cancer Research.