Whole-exome sequencing of pancreatic cancer defines genetic diversity and therapeutic targets

Authors

Agnieszka K. Witkiewicz

Elizabeth A. McMillan

Uthra Balaji

GuemHee Baek

Wan-Chi Lin

John Mansour

Mehri Mollaee

Kay-Uwe Wagner

Prasad Koduru

Adam Yopp

Michael A. Choti

Charles J. Yeo

Peter McCue

Michael A. White

Erik S. Knudsen

Doi

10.1038/ncomms7744

PMID: 25855536 · DOI: 10.1038/ncomms7744 · Journal: Nature Communications (2015)

TL;DR

Witkiewicz et al. performed whole-exome sequencing on 109 needle-microdissected pancreatic ductal adenocarcinomas (PAAD) paired with germline normals, with the goal of overcoming the low-tumour-cellularity limitation of prior pancreatic-cancer sequencing studies. Microdissection raised mean tumour cellularity above 50% and lifted the average non-synonymous mutation burden to 67 per case, comparable to other solid tumours. They confirmed the canonical KRAS / TP53 / SMAD4 / CDKN2A backbone and nominated 24 significantly mutated genes (MutSigCV, q<0.1), including several with prognostic associations: RBM10 mutations (4%) tracked with longer survival despite aggressive histology, while ARID1A loss (confirmed by IHC in an independent 296-case cohort) tracked with poor outcome. MYC amplification at 8q24.13 was uniquely associated with poor outcome and the adenosquamous (PAASC) subtype. Within the KRAS signalling axis, codon-12 mutations dominated but codon-Q61 alleles showed a remarkably favourable prognosis with reduced pERK staining; BRAFV600E mutations (3%) were mutually exclusive with KRAS, and a patient-derived BRAFV600E PDA cell line was as sensitive to vemurafenib (PLX-4032) as a melanoma control. High-frequency pathway lesions (WNT/β-catenin, chromatin-remodelling SWI/SNF, Hedgehog, DNA repair, RB/cell-cycle) define multiple candidate targeted therapies in PDA.

Cohort & data

  • 109 surgically resected pancreatic ductal adenocarcinoma cases with matched germline (normal tissue n=105 or peripheral blood n=4), all annotated with outcome and etiological features.
  • Histological subtypes: ductal carcinoma NOS / PAAD (n=94, 86%), adenosquamous / PAASC (n=11, 10%), colloid / PAAC (n=4, 4%)). Mostly stage I (n=5) or II (n=97); stage III (n=6) or IV (n=1). Median age 66 (range 29–86); 50% male.
  • Median overall survival: 21 months. Grade 3 (n=27) was associated with poor outcome; nodal status and adenosquamous subtype trended toward poor outcome.
  • Dataset: paad_utsw_2015 — cBioPortal study id; raw reads at NCBI SRA BioProject PRJNA278883 (“Whole Exome Sequencing of Microdissected Pancreatic Cancer”).
  • Sample prep: ~7-μm frozen-tissue slides; tumour epithelium manually needle-dissected (14-gauge needle, dissecting microscope); DNA extracted with QIAamp DNA Mini Kit. Tumour cellularity >50% (vs >60% recommended threshold for unselected resections).
  • Assays: whole-exome-seq using TruSeq Exome Enrichment (FC-121-1048) or Nextera Exome Enrichment (FC-140-1003) on Illumina HiSeq 2500. Mean depth 51.28× (95% CI 49.5–53.1) covering 99.26% of targets at ≥1×; 91.51% of targets at ≥15× coverage; 21 cases re-sequenced to ~123× depth.
  • Variant calling pipeline: reads aligned with bwa to UCSC hg19; SNVs by mutect (≥14 tumour reads, ≥8 normal reads); INDELs by varscan (≥8 reference, ≥3 variant reads, indels manually inspected in IGV). Annotation by Broad Institute oncotator; segmentation with ExomeCNV + DNAcopy (circular binary segmentation); copy-number calls by gistic 2.0 (join_segment_size=4, conf=0.95). Significantly mutated genes by mutsig (MutSigCV, default settings).
  • Validation: 248 non-silent mutations across 132 genes validated by sanger-sequencing in 84 cases — 92% confirmation rate (95% CI 87.6–94.6%). fish with the Vysis LSI MYC dual-colour break-apart probe used to confirm 8q24 MYC amplification. immunohistochemistry for ARID1A (Santa Cruz sc-32761), p53, pERK (Cell Signaling 4370) and p63 (Biolegend 619002) on autostainers; ARID1A IHC scored in an independent cohort of 296 PDA cases.
  • Pathway/subtype clustering: APC affinity-propagation clustering for CNV and pathway analysis; random-forest-classifier-based unsupervised clustering of pathway alteration profiles. Combined re-analysis with the 99-case Biankin et al. ICGC cohort (PMID:23103869)) yielded a 208-case meta-cohort for additional MutSigCV power.

Key findings

  • Microdissection raised mutation discovery: average 67 non-synonymous mutations per case — a substantial increase over the 26/case reported in the prior 99-case bulk-tissue PDA exome series (PMID:23103869)) — consistent with stromal contamination masking somatic events in unselected resections.
  • Mutator-phenotype subset: cases in the top mutation-burden quartile harboured lesions in mismatch-repair genes (MSH2, MSH6, MLH1, MLH3, PMS2, POLE, EXO1) and displayed a T→C-at-CTG mutation signature consistent with mismatch-repair deficiency; overall PDA spectra were dominated by C→T (age-associated) with minimal APOBEC TCW signal.
  • Smoking signature: smokers showed an enrichment of G→T transversions matching the canonical smoking signature, and had ~half the median survival of non-smokers (HR=2.8734, P=0.069).
  • MYC amplification (8q24.13): the only CNA among GISTIC peaks uniquely associated with poor overall survival (P=0.0013). MYC amplification was significantly over-represented in adenosquamous histology and was confirmed by break-apart FISH (no translocation). MYC-amplified cases did not show higher mutation burden or co-occurrence with other PDA-hallmark mutations. MYC amplification was also detectable in PanIN precursor lesions adjacent to invasive disease.
  • APC clustering of CNV partitioned cases into 6 clusters; clusters 5 and 6 (high overall CNV) were enriched for mutations or homozygous deletions in DNA double-strand-break-repair genes (ATM, CHEK2, BRCA1, BRCA2, FANCA, FANCD2, FANCM, NBN, RAD51AP2, RAD54B, RAD9A, XRCC4) — but not TP53. Cluster 6 trended toward worst survival (HR=3.9115, P=0.034 vs cluster 1).
  • 24 significantly mutated genes (MutSigCV) occurring in ≥3.5% of cases: KRAS 92%, TP53 50%, SMAD4 19%, FLG 10%, ATXN1 7%, ARID1A 6%, COL14A1 6%, CDKN2A 6%, RNF43 6%, RP1L1 6%, ITGAE 6%, GLI3 6%, GNAS 6%, SPTA1 6%, BCLAF1 5%, AXIN1 5%, NIN 5%, RBM10 4%, IRF6 4%, HDAC2 4%, PABPC1 4%, PIK3CA 4%, USP10 4%, CDH13 4%. Combined 208-case meta-MutSigCV with Biankin et al. added lower-frequency cancer genes: ATM, ARID2, TGFBR2, ACVR1B.
  • GNAS hotspots in IPMN-derived PDA: GNAS mutations (n=6) were in hotspot codon 201 (R201C, R201H); all four colloid carcinomas plus two conventional PDAs carried GNAS lesions, and all GNAS-mutant invasive cases were derived from intraductal papillary mucinous neoplasm (IPMN precursors. Four of six GNAS-mutant cases also carried KRAS mutations.
  • RBM10 favourable prognosis: RBM10 mutations (4 cases) were associated with longer overall survival (log-rank P=0.0345) despite all RBM10-mutant cases being high grade, pT3, with lymph-node metastasis in 3 of 4 cases. RBM10-mutant cases did not arise from IPMN and were not associated with GNAS mutations.
  • ARID1A loss confirmed prognostic in independent cohort: ARID1A protein deficiency by IHC in an additional 296 PDA cases was associated with poor overall survival (Kaplan–Meier P=0.0202). ARID1A-deficient PDA cell lines showed vulnerability to siRNA depletion of ARID1B, recapitulating the ARID1A-loss synthetic-lethal interaction reported in other cancers.
  • KRAS allele heterogeneity drives prognosis: 92% KRAS mutation rate, dominated by codon 12 (G12D/G12V/G12C/G12R/G12S) with codon 13 and codon 61 (Q61H/Q61K/Q61R) at lower frequency (Fig. 4b–c). Codon-12 mutations had uniformly poor prognosis, but codon-61 cases had a remarkably favourable prognosis (log-rank P=0.01999), despite histological/clinical features otherwise indicative of aggressive disease. By IHC, codon-61 cases showed lower pERK staining than other KRAS-mutant cases, suggesting allele-specific differences in MAPK pathway output.
  • KRAS-wildtype PDA carries RAS-effector lesions: in the 8% KRAS-wild-type subset, the authors identified PIK3CA activating mutations (4% overall) and BRAFV600E (3% overall) — all mutually exclusive with KRAS — plus additional cancer-gene lesions (STK11, GNAS, CHEK2, RB1).
  • BRAFV600E PDA is vemurafenib-sensitive in vitro: a patient-derived cell line (PDA_014, BRAFV600E) established from one of the BRAF-mutant cases was as sensitive to vemurafenib (PLX-4032) as the MNT1 BRAFV600E melanoma control across ERK phosphorylation, cell-cycle (BrdU) and viability assays; KRAS-mutant PL45 PDA cells were resistant.
  • High-frequency pathway disruption (>20% of cases) beyond KRAS/TP53: TGF-β (43% SMAD4, 9% NOTCH4, 7% TGFBR2, 6% ACVR1B, SMAD3 3%); NOTCH (10% NOTCH1, 6% NOTCH2, 1% NOTCH3, MAML1/MAML2/MAML3/JAG1/JAG2 at low frequency); Hedgehog (8% GLI3, 8% SMO, 6% LRP2, 3% GLI2, 2% PTCH1); WNT/β-catenin (9% AXIN1, 7% RNF43, 3% AXIN2, 2% APC, 2% CTNNB1, 2% TCF4); RB pathway (36% CDKN2A, 36% CDKN2B, 9% CDK4, 6% CCND1, 3% RB1, 2% CDKN2D); SWI/SNF (7% SMARCA2, 6% ARID1A, 5% PBRM1, 4% SMARCA4, 4% ARID1B, SMARCC1/SMARCC2 at low frequency); DNA repair (9% ATM, 6% CHEK2, 6% BCLAF1, 6% RAD51AP2, 5% BRCA1, 4% FANCF, BRCA2 1%, etc.). Pearson correlations between pathways were weak, suggesting largely independent acquisition.
  • Pathway-based subtypes correlate with prognosis: random-forest + APC clustering of pathway alterations identified subtypes; cases with isolated KRAS-pathway alterations (alone or with TP53) — clusters 1 & 2 — had poor prognosis, and cases with more complex multi-pathway deregulation trended toward even poorer outcome (HR=0.4369, P=0.1009 for clusters 1&2 vs others).

Genes & alterations

  • **KRAS — mutated in 92% of cases. Codon-12 alleles (G12D/G12V/G12C/G12R/G12S) dominate and confer poor prognosis; codon-13 and codon-61 (Q61H/Q61K/Q61R) alleles are minority. Codon-61 cases have favourable survival (P=0.01999) and lower pERK staining despite aggressive pathology, suggesting allele-specific MAPK output. Mutually exclusive with BRAF and PIK3CA lesions.
  • **TP53 — mutated in 50%, with strong p53-IHC stabilization concordance.
  • **SMAD4 — mutated/deleted in 19–43% across analyses; backbone of the TGF-β pathway lesions in PDA.
  • **CDKN2A / CDKN2B — frequently deleted (36% each in pathway analysis); the dominant RB-pathway lesion. Co-occurs with CDK4 (9%) and CCND1 (6%) amplification and RB1 loss (3%) in a smaller subset.
  • **MYC — focal amplification at 8q24.13 in a subset of cases; uniquely associated with poor overall survival (P=0.0013) and adenosquamous histology (PAASC). Confirmed by FISH; no MYC translocations. Detectable already in PanIN precursor lesions.
  • **BRAF — V600E mutations in 3% of cases; mutually exclusive with KRAS. Patient-derived PDA_014 cell line confirmed vemurafenib-sensitive.
  • **PIK3CA — activating mutations in 4% of cases; mutually exclusive with KRAS; nominate PI3K inhibitor candidates in PDA.
  • **RBM10 — 4% of cases; loss-of-function mutations associated with prolonged overall survival despite aggressive histology, mirroring the favourable association seen in lung adenocarcinoma. Independent of IPMN origin and GNAS status.
  • **ARID1A — 6% mutated by sequencing; protein loss by IHC in independent 296-case cohort is significantly associated with poor outcome (P=0.0202). ARID1A-deficient PDA cell lines are vulnerable to ARID1B depletion (synthetic lethal).
  • **RNF43 — 6% mutated; WNT-pathway tumour suppressor; predicted sensitivity to porcupine inhibitors (LGK974).
  • **GNAS — 6% mutated, hotspot codon 201 (R201C/R201H); all invasive cases derived from IPMN precursors; present in all four colloid (PAAC) and two conventional PDAs; frequent co-mutation with KRAS.
  • **AXIN1 / AXIN2 / APC — recurrently mutated WNT-pathway components nominating tankyrase inhibitors (XAV939) as therapeutic candidates.
  • **ATM, CHEK2, BRCA1, BRCA2, FANCF, FANCA, FANCD2, FANCM, BCLAF1 — DNA double-strand-break and Fanconi-anaemia pathway lesions enriched in high-CNV clusters; nominate olaparib (PARP inhibitor) and cross-linking agents like mitomycin-c as therapeutic candidates.
  • NOTCH1–4, JAG1/JAG2, MAML1–3 — NOTCH-pathway alterations totalling 31% of cases (amplification/mutation), nominating γ-secretase inhibitor (MK0752) candidates.
  • **SMO, GLI2, GLI3, PTCH1, LRP2 — Hedgehog-pathway alterations.
  • **TGFBR2, TGFBR1, TGFB1, ACVR1B, ACVR1C, SMAD3, SMAD6 — TGF-β-axis lesions augmenting SMAD4 loss.
  • **SMARCA2, SMARCA4, SMARCC1, SMARCC2, ARID1A, ARID1B, PBRM1 — SWI/SNF chromatin-remodelling lesions in >42% of cases.
  • STK11, CHEK2, RB1 — additional cancer-gene lesions identified in KRAS-wildtype cases.
  • FLG, ATXN1, COL14A1, RP1L1, ITGAE, SPTA1, NIN, BCLAF1, IRF6, HDAC2, PABPC1, USP10, CDH13 — additional MutSigCV-significant genes not previously highlighted in PDA; functional roles in PDA biology remain to be validated.
  • TERT (5p15.33), PDGFA (7p22.3) — recurrently amplified loci identified by GISTIC.

Clinical implications

  • BRAFV600E PDA as a vemurafenib candidate: the 3% of cases harbouring BRAF V600E (mutually exclusive with KRAS) define a small but pharmacologically tractable subset. The patient-derived PDA_014 cell line is as sensitive to PLX-4032 as MNT1 melanoma cells, supporting BRAF-mutation testing in KRAS-wildtype PDA.
  • PIK3CA-mutant PDA: 4% of cases carry activating PIK3CA mutations; the authors nominate buparlisib (BKM120) and gdc-0941 as candidate PI3K-pathway therapies.
  • RB-pathway dysregulation as a palbociclib opportunity: combined CDKN2A loss (>36%), CDK4 amplification, and CCND1 amplification (≥9% combined) make PDA a high-prior candidate for CDK4/6 inhibition with palbociclib (PD-0332991) or LEE-11; the authors and others (Franco et al. 2014) had previously shown synergy with pathway-selective agents in PDA models.
  • MYC amplification as a CDK9 / BET-bromodomain target: MYC amplification (~14% of cases) and association with poor outcome and adenosquamous histology nominate CDK9 (PHA767491) and BET-bromodomain (jq1) inhibitors.
  • DNA repair deficiency as olaparib / mitomycin-c opportunity: 14% Fanconi-anaemia-pathway alterations and 3% BRCA1/2 alterations nominate cross-linking agents (mitomycin C) and PARP inhibitors (olaparib). The high-CNV clusters (clusters 5 and 6) enriched for DSB-repair lesions could be a stratification biomarker.
  • WNT/β-catenin therapy in RNF43- or AXIN1/2/APC-loss cases: ~11% AXIN1/2 or APC loss, plus 7% RNF43 loss, nominate porcupine inhibitors (LGK974) and tankyrase inhibitors (XAV939). These remain actionable in the presence of activating KRAS.
  • NOTCH-pathway inhibition with γ-secretase inhibitors (MK0752) — 31% of cases harbour NOTCH-family amplifications/mutations.
  • Allele-specific KRAS prognosis: codon-61 KRAS mutations identify a favourable-survival subset within otherwise aggressive PDA; suggests molecular stratification beyond simple KRAS-mutant vs wild-type and may have implications for emerging allele-selective RAS inhibitors.
  • RBM10 as a candidate favourable prognostic biomarker in resected PDA, independent of histological grade and lymph-node status. ARID1A loss (by IHC) as a candidate unfavourable prognostic biomarker, validated in 296 cases.
  • MYC amplification as a candidate unfavourable prognostic biomarker for adenosquamous and high-CNV PDA, detectable by routine FISH.

Limitations & open questions

  • Mean exome depth (~51×) was modest; deeper resequencing of 21 cases to ~123× recovered mostly low-AF events, suggesting a residual undercall of subclonal mutations even after microdissection.
  • The 109-case cohort is single-institution (UT Southwestern + Thomas Jefferson) and resectable-disease only (mostly stage I–II); generalizability to unresectable/metastatic PDA is untested.
  • Functional consequences of many novel MutSigCV genes (e.g., FLG, ATXN1, COL14A1, RP1L1, ITGAE, SPTA1, NIN, BCLAF1, IRF6, HDAC2, PABPC1, USP10, CDH13) are not established; their oncogenic vs passenger status awaits orthogonal validation.
  • The authors did not identify UPF1 mutations in their adenosquamous cases despite Liu et al. (Nat. Med. 2014) reporting frequent UPF1 mutation in pancreatic adenosquamous carcinoma — discordance unresolved.
  • KRAS codon-61 favourable-prognosis finding is based on a small subset of the 92% KRAS-mutant cases (relatively rare allele); requires replication in larger cohorts.
  • RBM10 favourable-survival association is based on only 4 RBM10-mutant cases — statistically suggestive (P=0.0345) but underpowered.
  • BRAF V600E sensitivity to vemurafenib was shown in a single patient-derived line; clinical activity in BRAF-mutant PDA patients remains untested.
  • Drivers of chromosomal instability (the high-CNV cluster 5/6 phenotype) beyond DSB-repair lesions are not identified — additional structural-variant or replication-stress mechanisms are not interrogated by exome alone.
  • No matched RNA-seq or proteomic data accompany the exomes, limiting interpretation of expression-level consequences of CNAs (e.g., MYC amplification) and pathway flux for the proposed combination strategies.

Citations from this paper used in the wiki

  • “A total of 109 micro-dissected PDA cases were subjected to whole-exome sequencing… environmental stress and alterations in DNA repair genes associate with distinct mutation spectra.” (Abstract)
  • “Amplification of MYC is uniquely associated with poor outcome and adenosquamous subtype.” (Abstract; confirmed in Fig. 2e, P=0.0013.)
  • “RBM10 mutations associate with longer survival in spite of histological features of aggressive disease.” (Abstract; log-rank P=0.0345 in Fig. 3b.)
  • “KRAS mutations are observed in >90% of cases, but codon Q61 alleles are selectively associated with improved survival.” (Abstract; log-rank P=0.01999 in Fig. 4d.)
  • “Oncogenic BRAF mutations are mutually exclusive with KRAS and define sensitivity to vemurafenib in PDA models.” (Abstract; PDA_014 cell-line data in Fig. 4g–i.)
  • “The average mutation burden observed in our cohort was 67 non-synonymous events per case, equivalent to multiple other solid-tumour types.” (Results — Tissue cell enrichment and mutation burden.)
  • “Immunohistochemical analyses of additional 296 PDA cases demonstrated that ARID1A protein deficiency was significantly associated with poor outcome in this expanded cohort (Fig. 3c, P=0.0202).” (Results — Mutated genes associated with prognosis.)
  • “All GNAS-mutated invasive carcinoma cases were derived from a precursor, intraductal pancreatic neoplasm (IPMN). Four of the six PDAs harbouring GNAS alteration had concomitant mutations in KRAS.” (Results — Significantly mutated genes.)
  • “Table 2 — Tabular summary of potential pharmacological strategies for specific genetic pathway deregulation.” (Table 2, listing BRAF→vemurafenib, PIK3CA→BKM120/GDC-0941, CDKN2A loss / CDK4 amp / CCND1 amp → PD-0332991/LEE-11, MYC amp → PHA767491 / JQ1, FANC family → mitomycin C, BRCA family → olaparib, RNF43 loss → LGK974, AXIN1/2 or APC loss → XAV939, NOTCH family amp → MK0752.)
  • “Tumor sequencing data is deposited at SRA … with the BioProject ID: PRJNA278883 and Title: Whole Exome Sequencing of Microdissected Pancreatic Cancer.” (Additional information.)

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