Exome and whole genome sequencing of esophageal adenocarcinoma identifies recurrent driver events and mutational complexity

Authors

Austin M. Dulak

Petar Stojanov

Shouyong Peng

Michael S. Lawrence

Cameron Fox

Chip Stewart

Santhoshi Bandla

Yu Imamura

Steven E. Schumacher

Erica Shefler

Aaron McKenna

Kristian Cibulskis

Andrey Sivachenko

Scott L. Carter

Gordon Saksena

Douglas Voet

Alex H. Ramos

Daniel Auclair

Kristin Thompson

Carrie Sougnez

Robert C. Onofrio

Candace Guiducci

Rameen Beroukhim

David Zhou

Lin Lin

Jules Lin

Rishindra Reddy

Andrew Chang

James D. Luketich

Arjun Pennathur

Shuji Ogino

Todd R. Golub

Stacey B. Gabriel

Eric S. Lander

David G. Beer

Tony E. Godfrey

Gad Getz

Adam J. Bass

Doi

10.1038/ng.2591

PMID: 23525077 · DOI: 10.1038/ng.2591 · Journal: Nature Genetics (2013)

TL;DR

Dulak and colleagues performed whole-exome sequencing on 149 esophageal adenocarcinoma (EAC) tumor/normal pairs and whole-genome sequencing on 16 pairs from surgically-resected, treatment-naive cases. They identify a previously unappreciated EAC-specific mutational signature dominated by A>C transversions at AA dinucleotides (29% of all mutations), and statistically nominate 26 significantly mutated genes — confirming TP53, CDKN2A, SMAD4, PIK3CA and chromatin remodelers (ARID1A, SMARCA4, ARID2), plus new candidates including SPG20 (SPART), TLR4, and the RAC1-pathway regulators ELMO1 and DOCK2 (mutated in 17% of cases). Functional assays show that EAC-derived ELMO1 mutants augment NIH/3T3 invasion 2- to 7-fold over wild-type, implicating aberrant RAC1 signaling in EAC tumorigenesis.

Cohort & data

  • 149 fresh-frozen, surgically-resected, treatment-naïve EAC tumor/normal pairs subjected to whole-exome sequencing; 16 pairs also subjected to whole-genome sequencing (15 with matched WES); 14 WGS samples additionally profiled on mRNA expression arrays. Includes adenocarcinomas of the tubular esophagus and gastric-esophageal junction.
  • Cancer types: ESCA (esophageal adenocarcinoma) and GEJ (gastric-esophageal junction adenocarcinoma).
  • Dataset: esca_broad (Nat Genet 2013).
  • WES capture: SureSelect v2 Exome bait (Agilent), Illumina HiSeq; mean coverage 83.3× tumor / 85.9× normal. WGS: ~49× tumor / 30× germline coverage, 101 bp paired reads. Mutation calling via MuTect and Indelocator; rearrangements via dRanger.
  • Assays/methods: whole-exome-seq, whole-genome-seq, mutsig (MutSig significance algorithm), sequenom-genotyping (mass-spectrometry mutation validation), Affymetrix expression arrays (GEO GSE42363). BAM files deposited under dbGaP phs000598.v1.p1.
  • Four MSI-positive hypermutated tumors (mutation frequencies 14.6–50.9/Mb; highest with MSH6 + MSH3 mutations) were excluded from significance analysis, leaving 145 tumors. MSI assayed with a 10-marker panel (D2S123, D5S346, D17S250, BAT25, BAT26, BAT40, D18S55, D18S56, D18S67, D18S487).

Key findings

  • Mutation burden. Median genome-wide mutation frequency was 9.9/Mb (range 7.1–25.2/Mb), exceeded only by lung cancer and melanoma — versus 5.6/Mb in CRC genomes. Median 26,161 genome-wide mutations per tumor; median 104 non-silent coding mutations per tumor; median non-silent coding rate 3.51/Mb (range 0.97–10.8/Mb).
  • EAC-specific mutational signature. A>C base changes comprised 34% of all WGS mutations. A>C transversions at AA dinucleotides accounted for 29% of total mutations and showed strong context preference for AAG trinucleotides (49.3/Mb) over AAA/AAC/AAT. 84% of all A>C events were flanked by a 5′ adenine. Sequenom validation of 25 randomly selected AA>C calls confirmed 100% concordance. The signature is not seen in other cancer types.
  • Expression-coupled repair. AA transversions were attenuated in coding versus intergenic regions (16% vs 29%; AAG P=0.001, AAT P=0.0006, AAC P=0.0007, AAA P=0.0006, two-tailed Student’s t-test) and were strongly attenuated by gene expression (three-fold greater attenuation than at other contexts). AA>C mutations were more common on the non-transcribed strand than the transcribed strand (12.4/Mb vs 11.2/Mb, P=0.0016 paired t-test), consistent with transcription-coupled repair acting on the lesion.
  • Rearrangements. 2,952 candidate rearrangements (median 172/tumor, range 77–402); 20% interchromosomal, 55% of intrachromosomal events within 1 Mb. 38 predicted in-frame fusions, but no recurrent fusions. Mutation count and rearrangement count uncorrelated (R²=0.0046).
  • Significantly mutated genes (MutSig FDR q<0.1, n=26). TP53 and CDKN2A were most significant. Previously implicated in EAC: TP53, CDKN2A, ARID1A, PIK3CA, SMAD4. Novel candidates included ELMO1, DOCK2, SPG20 (SPART), TLR4, SMARCA4, ARID2, PBRM1, JARID2, AKAP6, HECW1, AJAP1, NUAK1, and KAT6A (MYST3).
  • RAC1-pathway axis. ELMO1 or DOCK2 mutations occurred in 25 (17%) EAC samples; two tumors harbored mutations in both, and two tumors had two independent DOCK2 mutations. ELMO1 p.K312 was mutated in three tumors (suggestive of gain-of-function). Mutations were also observed in additional RAC1 GEFs (TRIO, TIAM1, VAV2, ECT2). PAK1, a downstream RAC1 effector at 11q13, is recurrently amplified in EAC (per the authors’ prior copy-number work).
  • Chromatin remodelers. SWI/SNF family members ARID1A, SMARCA4, and ARID2 together mutated in 20% of tumors; with PBRM1 and JARID2, 24% (35/145) of EACs harbored chromatin-modifying-factor mutations. A predicted SMARCA4–DNM2 fusion (SMARCA4 exon 11 ↔︎ DNM2 exon 14) was identified by WGS.
  • Other candidates. SPG20 (SPART) mutated in 7% of EACs, with 5 of the events driven by AA transversions. TLR4 mutated in 6% (between p.D379 and p.F487 in the MD-2-interaction region, including p.E439). AKAP6 (8%), HECW1 (8%), AJAP1 (6%), NUAK1/ARK5 (3%), and KAT6A/MYST3 (5%).
  • Pathway disruption. Cell-cycle (CDKN2A): 14% by point mutation, with additional CCND1/CCNE1/CDK6 amplifications. β-catenin pathway mutated in only 9% (two tumors with co-occurring APC + CDH1 or APC + AXIN1; one AXIN1–GALNT7 fusion in ESO-1060). TGFβ/SMAD: 18% (SMAD4 most recurrent, mutated in 10 samples and subject to frequent copy loss). MAPK: no BRAF mutations, NF1 in 3 tumors (2%), KRAS in 5 (3%; three at p.G12, one p.K117N produced by an AA transversion — a known transforming mutation previously reported in CRC). PI3K pathway: most-frequently-mutated oncogenic pathway (13%) — PIK3CA (7), PIK3R1 (5), PTEN (4).
  • ERBB family. Three EGFR mutations (p.S447Y and p.S1153I predicted benign by Polyphen-2). Five ERBB2 mutations including kinase-domain p.D769Y (×2) and p.G776V, previously seen in other cancers.
  • Actionability. 23% of tumors harbored mutations in genes with approved or preclinical targeted agents; combined with amplification, 48% of tumors had a targetable alteration — pointing to the importance of focally amplified RTKs in EAC.
  • AA-transversion biology. The AA signature is unable to create the most common KRAS p.G12 or PIK3CA p.E545 hotspots, and cannot generate a stop codon from an AAG context — explaining why driver hotspots in EAC are largely not AA-derived (0 of 2,570 AA-driven coding mutations were nonsense).

Genes & alterations

  • TP53 — most significantly mutated tumor suppressor in EAC; loss-of-function pattern (mutations were not manually re-reviewed in BAM for TP53).
  • CDKN2A — second most significant gene; recurrent point mutations and the dominant lesion in the 14% of EACs with point-mutation-driven cell-cycle disruption.
  • SMAD4 — mutated in 10/145 tumors; most recurrently altered gene in the TGFβ/SMAD pathway (18% of EAC), with frequent additional copy-number loss.
  • PIK3CA — most frequently mutated of the actionable genes; 7 tumors. Author note: hotspot p.E545 cannot be generated by an AA transversion.
  • ARID1A, ARID2, SMARCA4, PBRM1, JARID2 — SWI/SNF and broader chromatin-remodeling mutations in 24% of EACs (SMARCA4 also part of a putative SMARCA4–DNM2 fusion).
  • ELMO1 — significantly mutated; recurrent p.K312 mutation in 3 tumors. EAC-derived mutants (p.F59L, p.K312E, p.K312T, p.K349R, p.T421N) increase NIH/3T3 invasion 2–7-fold over wild-type (matrigel transwell assay; P values reported in figure 3C). Wild-type ELMO1 itself increased invasion 7-fold over GFP control (P=0.0040).
  • DOCK2 — significantly mutated; RAC1 GEF dimerization partner of ELMO1. Two tumors carried two independent DOCK2 mutations.
  • RAC1 — not directly mutated, but upstream regulators ELMO1, DOCK2, TRIO, TIAM1, VAV2, ECT2 are recurrently mutated; downstream effector PAK1 is recurrently amplified at 11q13.
  • SPART (formerly SPG20) — mutated in 7% of EACs; 5 mutations generated by AA transversions.
  • TLR4 — mutated in 6% of EACs, with mutations clustered in the MD-2 interaction region (p.D379–p.F487), including p.E439.
  • KAT6A (formerly MYST3) — lysine acetyltransferase mutated in 7 specimens (5%).
  • AKAP6 (8%), HECW1 (8%), AJAP1 (6%), NUAK1 (3%) — additional significant candidates.
  • KRAS — mutated in 5 tumors (3%); 3 at p.G12, plus p.K117N (a known transforming allele) generated by AA transversion.
  • ERBB2 — mutated in 5 tumors; kinase-domain p.D769Y (×2) and p.G776V previously reported in other cancers.
  • EGFR — 3 mutations (p.S447Y, p.S1153I), predicted non-deleterious by Polyphen-2.
  • PIK3R1 (5 tumors), PTEN (4 tumors) — additional PI3K-pathway mutations.
  • NF1 (2%), BRAF (none) — MAPK-pathway mutations are uncommon in EAC, contrasting with CRC.
  • APC, CDH1, AXIN1, CTNNB1 — β-catenin-pathway mutations in only 9% of EACs; two tumors with co-occurring APC + CDH1 or APC + AXIN1; an AXIN1–GALNT7 fusion identified by WGS in ESO-1060.
  • MSH6, MSH3 — co-mutated in the most hypermutated MSI-positive EAC (one of four MSI-positive cases excluded from significance analysis).
  • CCND1, CCNE1, CDK6 — frequent focal amplifications contributing to cell-cycle deregulation.

Clinical implications

  • HER2 (ERBB2)-targeting agents. The authors note that trastuzumab is currently the only approved targeted agent for EAC/GEJ adenocarcinoma, with use guided by ERBB2 overexpression/amplification. They argue that recurrent kinase-domain ERBB2 mutations (5/145 tumors; p.D769Y, p.G776V) and the co-occurrence of ERBB2 amplification + mutation in 3% of samples justify exploring ERBB2 mutation as an additional biomarker for HER2-directed therapy in EAC.
  • Actionability landscape. Mutations in genes with clinically approved or preclinical-stage inhibitors were present in 23% of tumors (PIK3CA most frequent); when amplifications were also counted, 48% of tumors had a targetable alteration. The authors highlight that amplified RTKs are more common than RTK mutations in EAC and call for strategies tailored to amplified targets.
  • Mutational signature as etiologic clue. The AA-transversion signature is unique to EAC and biased toward intergenic / low-expression regions; the authors hypothesize a link to gastric refluxate (e.g. oxidative damage from 8-oxo-dGTP-like precursors), opening a path toward identifying specific environmental carcinogens behind EAC’s rising incidence.
  • No clinical-outcome (survival) endpoints are analyzed in this paper.

Limitations & open questions

  • The AA-transversion mutational signature is described but the specific mutagen(s) remain unidentified; the authors propose oxidative damage (e.g. 8-oxo-dGTP analog) as a candidate based on E. coli evidence but acknowledge experimental confirmation is needed.
  • AA transversions accumulate preferentially in low-expression and non-coding regions and cannot generate the canonical KRAS p.G12 or PIK3CA p.E545 hotspots or stop codons in AAG contexts, leading to a potential under-representation of this signature in classical drivers; the broader functional contribution of AA-derived mutations to oncogenesis is therefore harder to quantify.
  • Functional validation is limited to ELMO1 (NIH/3T3 matrigel invasion); DOCK2, SPG20/SPART, TLR4, AKAP6, HECW1, AJAP1, NUAK1, and KAT6A/MYST3 are nominated statistically but not experimentally validated here.
  • MSI-positive tumors (4/149, ~3%) were excluded from significance analysis to avoid confounding; the genomic landscape of MSI-positive EAC is therefore not characterized in this cohort.
  • All samples are treatment-naïve surgical resections — the mutational landscape of post-chemotherapy/post-radiation or metastatic EAC is not addressed.
  • No survival or treatment-response analyses are performed; the clinical relevance of the newly nominated drivers awaits validation in outcome-annotated cohorts.
  • Comparison to other gastrointestinal adenocarcinomas: contrasts with CRC (PMID:22810696) — β-catenin and KRAS/BRAF mutations are much less common in EAC than CRC, and the AA signature is EAC-specific.

Citations from this paper used in the wiki

  • “We analyze the mutation spectra from whole exome sequencing of 149 EAC tumors/normal pairs, 15 of which have also been subjected to whole genome sequencing.” (Abstract)
  • “Statistical analysis of exome data identified significantly mutated 26 genes. Of these genes, four (TP53, CDKN2A, SMAD4, and PIK3CA) have been previously implicated in EAC. The novel significantly mutated genes include chromatin modifying factors and candidate contributors: SPG20, TLR4, ELMO1, and DOCK2.” (Abstract)
  • “ELMO1 or DOCK2 are mutated in 25 (17%) EAC samples with two samples having mutations in both factors and two samples have two independent mutations in DOCK2.” (Results)
  • “Compared to GFP control, wild-type ELMO1 increased invasion by 7-fold (P = 0.0040, Student’s T-test, unpaired)… ELMO1 mutations (p.F59L, p.K312E, p.K312T, p.K349R, p.T421N) further resulted in a significant increase (2 to 7-fold) in invasion compared to wild-type ELMO1.” (Results)
  • “A to C transversions at AA sites accounted for 29% of the total mutations.” (Results)
  • “AA>C mutations are more common when the AA sites are located on the non-transcribed strand (12.4/Mb vs. 11.2/Mb; P = 0.0016, Student’s T-test, paired).” (Results)
  • “Although we identified potentially actionable genomic alterations in 48% of samples, the ERBB2 (HER2)-targeted antibody, trastuzumab, is the only targeted agent used in the treatment of EAC/GEJ adenocarcinomas with its use guided by overexpression and genomic amplification of ERBB2.” (Discussion)

This page was processed by crosslinker on 2026-05-09.