Integrated genomic characterization of oesophageal carcinoma

Authors

The Cancer Genome Atlas Research Network

Adam J. Bass

Vésteinn Thorsson

Ilya Shmulevich

Peter W. Laird

Francisco Sánchez-Vega

Toshinori Hinoue

Andrew D. Cherniack

Juliann Shih

Reanne Bowlby

Andrew J. Mungall

Christina Curtis

Jose Antonio Seoane

Ronglai Shen

Hsin-Ta Wu

Benjamin J. Raphael

Nikolaus Schultz

Barbara G. Schneider

Jaffer A. Ajani

Doi

10.1038/nature20805

PMID: 28052061 · DOI: 10.1038/nature20805 · Journal: Nature (2017)

TL;DR

The TCGA Research Network performed integrated multi-platform molecular profiling (whole-exome and low-pass whole-genome sequencing, SNP6 copy number, HM450 methylation, mRNA-seq, miRNA-seq, RPPA) on 164 oesophageal carcinomas and compared them to 359 gastric adenocarcinomas and 36 additional gastroesophageal-junction (GEJ) adenocarcinomas. Oesophageal squamous cell carcinoma (ESCC) and oesophageal adenocarcinoma (ESCA, abbreviated EAC in the paper) are molecularly distinct: ESCC resembles squamous tumours of the head/neck and lung, whereas EAC is essentially indistinguishable from chromosomally-unstable (CIN) gastric adenocarcinoma (STAD). The authors define three ESCC molecular subtypes (ESCC1, ESCC2, ESCC3), find no evidence for an aetiologic role of HPV in ESCC, and argue that EAC and CIN gastric cancer should be considered a single disease entity — strongly cautioning against pooling EAC and ESCC in clinical trials PMID:28052061.

Cohort & data

  • Total profiled: 164 oesophageal carcinomas — 90 ESCC, 72 EAC (ESCA; 61 definite + 11 probable oesophageal), and 2 undifferentiated carcinomas — accompanied by 359 gastric adenocarcinomas and 36 GEJ adenocarcinomas of indeterminate origin (15 definite gastric + 48 probable gastric for the gastric-GEJ slice; 140 fundus/body, 143 antral/pyloric, 13 unlocalized for the bulk gastric set) PMID:28052061.
  • Sample qualification: 322 oesophageal cases submitted to the BCR; 185 qualified by pathology and molecular QC; 171 entered genomic analysis after expert pathology re-review. 388 similarly characterized gastric adenocarcinomas were used as the comparator (incorporating an additional 77 STADs not in the original TCGA STAD analysis) PMID:28052061.
  • Geographic distribution of ESCC: Vietnamese patients (n=41, only Asian population studied), Eastern European, South American, and US/Canadian patients PMID:28052061.
  • Assays: whole-exome sequencing (Agilent SureSelect Human All Exon V5 at the Broad Institute for STAD-labelled samples; Nimblegen SeqCap EZ Human Exome Library v3.0 + custom IDT viral probes at Washington University for ESCA-labelled samples); low-pass (6–8×) whole-genome sequencing on 51 oesophageal cancers; SNP-array copy-number profiling on Affymetrix SNP 6.0; DNA methylation on Illumina HumanMethylation450 (HM450); mRNA-seq and miRNA-seq; and reverse-phase protein array (RPPA) on 113 tumours PMID:28052061.
  • Analysis tools: iCluster integrative clustering across SCNA + methylation + mRNA + miRNA; mutation calling pooled from four centres (Broad, Wash U, UCSC, BCCA) and annotated with Oncotator; significantly mutated genes by MutSigCV2.0; recurrent SCNAs by GISTIC 2.0; allelic copy number / purity / ploidy by ABSOLUTE; structural-variant detection by BreakDancer and Meerkat; mRNA quantification with MapSplice/RSEM; mutation-signature discovery by Bayesian NMF; read alignment with BWA; inter-cohort cluster integration with COCA; patient-identity QC by custom Sequenom SNP genotyping panel and AmpFISTR Identifiler PMID:28052061.
  • Dataset / cBioPortal: stes_tcga_pub (esophageal/stomach TCGA publication cohort); historical comparator stad_tcga_pub.

Key findings

  • EAC and ESCC are molecularly distinct. Independent and integrated clustering across SCNA, methylation, mRNA and miRNA platforms cleanly separates squamous from adenocarcinoma cases. Pathway analysis showed EACs upregulated CDH1 (E-cadherin) signalling and ARF6/FOXA pathways, while ESCCs upregulated Wnt, syndecan and p63 pathways PMID:28052061.
  • Three ESCC molecular subtypes. Integrative clustering yielded ESCC1 (n=50), ESCC2 (n=36), and ESCC3 (n=4 — defined by SMARCA4 mutations, increased DNA methylation and relatively unaltered SCNA profiles). All four ESCC3 tumours came from US/Canadian patients (P=0.001, Fisher’s exact test) and had no analogue in the TCGA HNSCC dataset PMID:28052061.
  • ESCC1 = NRF2-pathway / classical squamous. Characterised by alterations in NFE2L2, KEAP1, CUL3 and ATG7, higher SOX2/TP63 amplification frequency, YAP1 (11q22.1) amplification and VGLL4/ATG7 deletion (suggesting Hippo activation). ESCC1 gene expression resembled the classical subtype of TCGA lung squamous and head and neck squamous carcinomas PMID:28052061.
  • ESCC2 = NOTCH/PI3K with immune infiltration. Higher rates of NOTCH1 and ZNF750 mutation, more frequent inactivating alterations of KDM6A and KMT2D, CDK6 amplification, and inactivation of PTEN or PIK3R1. ESCC2 had greater leukocyte infiltration, higher cleaved Caspase-7, and lower DNA methylation / higher expression of BST2 (P=3×10⁻⁴, Fisher) PMID:28052061.
  • ESCC3 = PI3K-activated, SMARCA4/KMT2D mutant. Four samples; no genetic cell-cycle deregulation; TP53 mutated in only 1/4; all had PI3K-pathway activating alterations and 3/4 had KMT2D alterations on top of SMARCA4. Unique to ESCC — no analogue in TCGA HNSCC PMID:28052061.
  • Significantly mutated genes — ESCC: TP53, NFE2L2, KMT2D (MLL2), ZNF750, NOTCH1, TGFBR2. Significantly mutated genes — EAC: TP53, CDKN2A, ARID1A, SMAD4, ERBB2 PMID:28052061.
  • Recurrent SCNAs — EAC: amplifications of VEGFA (6p21.1), ERBB2 (17p12), GATA6 (18q11.2) and CCNE1 (19q12), and deletion of SMAD4 (18q21.2). Recurrent SCNAs — ESCC: amplifications of SOX2 (3q26.33), TERT (5p15.33), FGFR1 (8p11.23), MDM2 (12q14.3), NKX2-1 (14q13.2), and deletion of RB1 (13q14.2). Novel focal deletions at 3p25.2 in ESCC encompass VGLL4 (Hippo negative regulator) and ATG7 (autophagy) PMID:28052061.
  • Cell-cycle dysregulation differs between subtypes. CDKN2A inactivation in 76% of ESCC (and 76% of EAC, by mutation/deletion/epigenetic silencing); CCND1 amplified in 57% of ESCC vs only 15% of EAC; EAC instead amplifies CCNE1. Additional ESCC tumours had CDK6 amplification or RB1 loss PMID:28052061.
  • RTK / signalling alterations. EAC shows ERBB2 alteration in 32% (vs 3% of ESCC), and additional EACs amplify KRAS, EGFR, IGF1R or VEGFA. ESCC shows amplification or mutation of EGFR in 19%, and PI3K-activating alterations of PIK3CA, PTEN or PIK3R1 in 24%. Four EACs had ERBB2 mutations without amplification, and six ERBB2-amplified cases expressed an ERBB2 exon-12–JUP 3′-UTR fusion transcript that omits the transmembrane and tyrosine-kinase domains PMID:28052061.
  • Hedgehog / β-catenin / chromatin modifiers. 6% of ESCCs (no EACs) carried inactivating alterations in PTCH1, suggesting active hedgehog signalling. EACs more frequently activated CTNNB1 (β-catenin) and dysregulated TGF-β. SWI/SNF-encoding ARID1A, SMARCA4 and PBRM1 alterations were more common in adenocarcinomas, while ESCCs more frequently altered histone modifiers KDM6A, KMT2D and KMT2C PMID:28052061.
  • GATA4/6 amplification distinguishes EAC. EACs frequently amplified GATA4 and GATA6 developmental factors, paralleling gastric adenocarcinoma biology PMID:28052061.
  • No aetiological role for HPV in ESCC. ESCC HPV-transcript levels by mRNA-seq resembled HPV-negative HNSCC PMID:28052061.
  • Mutation signatures: ESCCs were enriched (vs EAC) in C>A substitutions (associated with smoking/tobacco chewing; P=7×10⁻⁷, Wilcoxon) and APOBEC signatures (P=5×10⁻⁵). Within ESCC the APOBEC signature was overrepresented in the ESCC2 subtype (P=0.03, Kruskal–Wallis) and in patients from Ukraine and Russia (P=0.01, Wilcoxon). Restricting to lifelong nonsmokers, the C>A signature was significantly higher in Vietnamese patients (P=0.013, Wilcoxon), consistent with tobacco chewing. EAC showed a predilection for A>C transversions at AA dinucleotides absent from ESCC PMID:28052061.
  • Vietnamese ESCC patients are enriched in NFE2L2 mutations (24% [10/41] vs 6% [3/49] in other patients; P=0.017, Fisher’s exact test), suggesting common oxidative-stressor exposure or genetic predisposition. All Vietnamese patients carried East-Asian alcohol-metabolism variants in ALDH2/ADH1B PMID:28052061.
  • EAC = CIN gastric cancer. 71/72 EACs were classified as chromosomal-instability (CIN) per the prior TCGA gastric subtyping (EBV / MSI / CIN / GS); none were MSI- or EBV-positive. Among 288 CIN gastroesophageal adenocarcinomas (GEA-CIN), integrative clustering showed no consistent separation between EAC and CIN gastric tumours, even after excluding equivocal GEJ cases. The CIN proportion increases proximally — analogous to the distal-to-rectal CIN gradient in colorectal cancer PMID:28052061.
  • Methylation gradient along the gastroesophageal axis. DNA methylation clustering of GEA-CIN tumours revealed a proximal-to-distal gradient: hypermethylated cluster 1 was enriched in oesophagus/proximal stomach/GEJ, while hypomethylated cluster 4 was enriched in distal stomach. Hypermethylation was significantly more common in EACs than in gastric CIN cancers (70% vs 30%; P=1.0×10⁻⁸, Fisher’s exact test). No GEA-CIN tumour showed MGMT/MLH1 silencing of the kind seen in gastric CIMP, but CDKN2A epigenetic silencing was more common; MGMT and CHFR silencing were also enriched in cluster 1 PMID:28052061.
  • Differentiating EAC from CIN gastric cancer. Within GEA-CIN, EACs had higher rates of SMARCA4 mutation, RUNX1 deletion, VEGFA and MYC amplification; lower APC mutation rates (suggesting less Wnt/β-catenin involvement); and higher rates of FHIT/WWOX fragile-site deletions PMID:28052061.

Genes & alterations

  • TP53 — Significantly mutated in both EAC and ESCC. Notably mutated in only 1/4 ESCC3 tumours PMID:28052061.
  • NFE2L2 — Significantly mutated in ESCC (more so in ESCC1); 24% in Vietnamese cohort vs 6% in others (P=0.017). Activates NRF2 pathway, associated with chemoradiotherapy resistance PMID:28052061.
  • KEAP1, CUL3, ATG7 — NRF2-degradation and autophagy components altered in ESCC1 alongside NFE2L2 PMID:28052061.
  • KMT2D (MLL2), KMT2C (MLL3), KDM6A (UTX) — Histone-modifier alterations frequent in ESCC (especially ESCC2 for KMT2D/KDM6A); KMT2D mutated in 3/4 ESCC3 PMID:28052061.
  • ZNF750, NOTCH1 — Significantly mutated in ESCC (enriched in ESCC2); modulators of squamous cell maturation PMID:28052061.
  • TGFBR2 — Significantly mutated in ESCC PMID:28052061.
  • CDKN2A — Inactivated in 76% of EAC (mutation, deletion or epigenetic silencing) and 76% of ESCC; significantly mutated in EAC; epigenetic silencing more common in proximal GEA-CIN cluster 1 PMID:28052061.
  • ARID1A, SMARCA4, PBRM1 — SWI/SNF complex; more frequently altered in EAC. SMARCA4 mutation defines ESCC3 and is enriched in EAC vs CIN gastric PMID:28052061.
  • SMAD4 — Significantly mutated in EAC; deleted at 18q21.2 in EAC PMID:28052061.
  • ERBB2 — Altered in 32% of EACs vs 3% of ESCCs (amplification or mutation). Four EACs had ERBB2 mutations without amplification. Six amplified EACs expressed an ERBB2 exon-12–JUP 3′-UTR fusion transcript that lacks the transmembrane and tyrosine-kinase domains PMID:28052061.
  • CCND1 — Amplified in 57% of ESCC, 15% of EAC PMID:28052061.
  • CCNE1 — Amplified at 19q12 more often in EAC PMID:28052061.
  • CDK6 — Amplified in additional ESCCs (especially ESCC2) lacking CCND1 amplification PMID:28052061.
  • RB1 — Recurrently deleted (13q14.2) in ESCC PMID:28052061.
  • SOX2, TP63 — Amplified at 3q in 48% of ESCCs (combined); higher in ESCC1 PMID:28052061.
  • TERT, FGFR1, MDM2, NKX2-1 — Recurrent focal amplifications in ESCC PMID:28052061.
  • VGLL4 — Novel 3p25.2 deletion in ESCC, removing this Hippo-pathway negative regulator (and adjacent ATG7); associated with ESCC1 PMID:28052061.
  • YAP1 — Amplified at 11q22.1 in ESCC1, suggesting Hippo activation PMID:28052061.
  • VEGFA, GATA6, GATA4 — Amplified in EAC; GATA6 experimentally validated as an EAC driver in prior work cited by the authors. VEGFA and MYC amplifications more common in EAC than CIN gastric PMID:28052061.
  • EGFR — Amplified or mutated in 19% of ESCCs; also amplified in some EACs PMID:28052061.
  • PIK3CA, PTEN, PIK3R1 — PI3K-pathway activating alterations in 24% of ESCCs (PTEN/PIK3R1 enriched in ESCC2); all 4 ESCC3 tumours had PI3K-activating alterations PMID:28052061.
  • KRAS, IGF1R — Amplified in subsets of EACs alongside ERBB2/EGFR/VEGFA PMID:28052061.
  • PTCH1 — Inactivating alterations in 6% of ESCCs (no EACs), suggesting active hedgehog signalling PMID:28052061.
  • CDH1 — Upregulated E-cadherin signalling pathway in EAC vs ESCC (gene-expression analysis) PMID:28052061.
  • CTNNB1 — β-catenin activation more common in EAC than ESCC PMID:28052061.
  • BST2 — Lower DNA methylation and higher expression in ESCC2 (P=3×10⁻⁴, Fisher); flagged as potential immunomodulatory target PMID:28052061.
  • RUNX1, APC, FHIT, WWOX — RUNX1 deletion higher in EAC vs CIN gastric; APC mutation lower in EAC; fragile-site deletions at FHIT/WWOX higher in EAC PMID:28052061.
  • MGMT, CHFR — Epigenetic silencing enriched in proximal hypermethylated GEA-CIN cluster 1 (relevant to alkylating-agent and microtubule-inhibitor response in prior literature) PMID:28052061.

Clinical implications

  • Do not pool EAC and ESCC in clinical trials. Molecular characteristics differ across all platforms tested; the authors explicitly caution against combining histologic subtypes for neoadjuvant, adjuvant or systemic-therapy trials PMID:28052061.
  • EAC may be treated as CIN gastric cancer. The molecular indistinguishability of EAC from CIN gastric adenocarcinoma supports unifying these as a single disease entity for therapeutic and trial purposes (analogous to colorectal adenocarcinoma along the colon) PMID:28052061.
  • ERBB2-directed therapy may extend beyond approved indication. Trastuzumab is FDA-approved for gastric / GEJ adenocarcinomas; the authors note ERBB2-positive EACs are routinely treated off-label with trastuzumab and that ERBB2-mutant (non-amplified) EAC patients may also benefit. The functional consequence of the EAC-specific ERBB2–JUP fusion transcript (which lacks transmembrane / kinase domains) is unclear PMID:28052061.
  • Cell-cycle inhibitors are rationally indicated, especially in ESCC. Combined CDKN2A loss and CCND1 / CDK6 amplification in ESCC supports CDK4/6 inhibitor evaluation PMID:28052061.
  • Subtype-directed strategies for ESCC:
    • ESCC1 (NRF2-pathway / classical squamous, SOX2/TP63-amplified, Hippo-activated): consider Hippo / YAP1-targeted approaches.
    • ESCC2 (NOTCH/PI3K/immune-infiltrated): higher cleaved Caspase-7 implies XIAP-directed agents could enhance apoptosis; lower BST2 methylation suggests BST2-inhibition potential.
    • ESCC3 (PI3K-activated, SMARCA4/KMT2D mutant): all PI3K-pathway-activated, supporting PI3K-pathway inhibition PMID:28052061.
  • Methylation-driven biomarker hypotheses for proximal GEA-CIN. Enrichment of MGMT and CHFR silencing in cluster 1 (oesophagus/proximal-stomach) raises the prospect of differential alkylating-agent and microtubule-inhibitor responses (citing prior literature) PMID:28052061.
  • HPV status is not a useful biomarker in ESCC. Unlike cervical SCC and HNSCC, HPV does not appear to play an aetiologic role in ESCC PMID:28052061.
  • Geographic / population stratification matters. NFE2L2-enriched ESCC1 in Vietnamese patients, APOBEC-enriched ESCC2 in Eastern European patients, and ESCC3 confined to US/Canada highlight the need for geographically diverse trial recruitment PMID:28052061.

Limitations & open questions

  • Small ESCC3 sample (n=4). All four ESCC3 tumours derive from US/Canadian patients; the authors note that additional samples are needed to confirm the subtype is genuinely confined to ESCC and to refine its molecular definition PMID:28052061.
  • EAC vs CIN gastric distinction is gradual, not categorical. Although the integrated data support a single GEA-CIN entity, additional samples are required to refine the progressive proximal-to-distal gradient of features and to determine whether subtle EAC-enriched features (SMARCA4, RUNX1 deletion, VEGFA/MYC amplification, FHIT/WWOX fragile-site loss) define a clinically meaningful subgroup of CIN GEA PMID:28052061.
  • GEJ misannotation could inflate EAC–gastric similarity. The authors repeated their analysis after excluding equivocal GEJ cases and saw no separation, but residual misclassification cannot be ruled out PMID:28052061.
  • Functional consequence of the ERBB2–JUP fusion transcript is unknown. Six ERBB2-amplified EACs expressed a fusion that omits transmembrane and tyrosine-kinase domains; its biology and impact on trastuzumab response are unresolved PMID:28052061.
  • Tobacco chewing as the source of the ESCC C>A signature in Vietnamese nonsmokers is inferential; direct exposure data are unavailable. The relationship between East-Asian alcohol-metabolism variants (ALDH2 / ADH1B — entire Vietnamese cohort carrier) and NFE2L2 mutation could not be statistically tested PMID:28052061.
  • No prospective biomarker validation. Subtype-directed therapeutic suggestions (XIAP, BST2, Hippo/YAP, PI3K) are hypothesis-generating, not tested in this study.
  • No statistical power calculation. The authors note that sample sizes were not predetermined and investigators were not blinded — standard for TCGA but limits prespecified inference.
  • ESCC3 has no HNSCC analogue — unclear whether this reflects a true ESCC-restricted biology or undersampling in TCGA HNSCC PMID:28052061.

Citations from this paper used in the wiki

  • “we performed a comprehensive molecular analysis of 164 carcinomas of the oesophagus derived from Western and Eastern populations” (Abstract).
  • “Oesophageal squamous cell carcinomas resembled squamous carcinomas of other organs more than they did oesophageal adenocarcinomas” (Abstract).
  • “Squamous cell carcinomas showed frequent genomic amplifications of CCND1 and SOX2 and/or TP63, whereas ERBB2, VEGFA and GATA4 and GATA6 were more commonly amplified in adenocarcinomas” (Abstract).
  • “Oesophageal adenocarcinomas strongly resembled the chromosomally unstable variant of gastric adenocarcinoma, suggesting that these cancers could be considered a single disease entity” (Abstract).
  • “we identified 90 ESCCs, 72 EACs (61 definite oesophageal and 11 probable oesophageal), 36 GEJ carcinomas of indeterminate origin” (p. 2).
  • “In ESCC, we identified significantly mutated genes, TP53, NFE2L2, MLL2, ZNF750, NOTCH1 and TGFBR2” (p. 3).
  • “In EAC, we identified significant mutations in TP53, CDKN2A, ARID1A, SMAD4 and ERBB2” (p. 3).
  • “Inactivation of CDKN2A and amplification of CCND1 were present in 76% and 57% of squamous tumours, respectively” (p. 3).
  • “EACs had a wider range of potentially oncogenic amplifications, most commonly of ERBB2, which was altered in 32% of EACs, but in only 3% of ESCCs” (pp. 3–4).
  • “Transcriptome data identified six cases with ERBB2 amplification that expressed a fusion transcript in which exon 12 of ERBB2 was fused to the 3′ untranslated region of neighbouring gene JUP” (p. 4).
  • “Integrative clustering of ESCC data using iCluster revealed two classes … We designated the distinct set of tumours with these features as subtype ESCC3, thus dividing ESCCs into three molecular subtypes: ESCC1 (n = 50), ESCC2 (n = 36) and ESCC3 (n = 4)” (p. 4).
  • “All four ESCC3 tumours were derived from patients from the USA and Canada (P = 0.001, Fisher’s exact test)” (p. 5).
  • “24% in the Vietnamese cohort (10 out of 41) versus 6% in other patients (3 out of 49; P = 0.017, Fisher’s exact test)” (p. 5).
  • “ESCC HPV transcript levels resembled HPV-negative HNSCC tumours … These data do not support an aetiologic role for HPV in ESCC” (p. 6).
  • “71 of 72 EACs classified as CIN … No EACs were positive for MSI or EBV” (p. 6).
  • “The proportion of cancers showing more frequent DNA hypermethylation (that is, clusters 1 or 2) was significantly higher among EACs than among gastric CIN cancers (70% versus 30%, respectively; P = 1.0 × 10−8, Fisher’s exact test)” (pp. 6–7).
  • “Our analyses therefore argue against approaches that combine EAC and ESCC for clinical trials of neoadjuvant, adjuvant or systemic therapies” (p. 7).
  • “we may view GEA as a singular entity, analogously to colorectal adenocarcinoma” (p. 7).

This page was processed by crosslinker on 2026-05-14.