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

Abstract

Pancreatic ductal adenocarcinoma (PDA) has a dismal prognosis and insights into both disease etiology and targeted intervention are needed. A total of 109 micro-dissected PDA cases were subjected to whole-exome sequencing. Microdissection enriches tumour cellularity and enhances mutation calling. Here we show that environmental stress and alterations in DNA repair genes associate with distinct mutation spectra. Copy number alterations target multiple tumour suppressive/oncogenic loci; however, amplification of MYC is uniquely associated with poor outcome and adenosquamous subtype. We identify multiple novel mutated genes in PDA, with select genes harbouring prognostic significance. RBM10 mutations associate with longer survival in spite of histological features of aggressive disease. KRAS mutations are observed in >90% of cases, but codon Q61 alleles are selectively associated with improved survival. Oncogenic BRAF mutations are mutually exclusive with KRAS and define sensitivity to vemurafenib in PDA models. High-frequency alterations in Wnt signalling, chromatin remodelling, Hedgehog signalling, DNA repair and cell cycle processes are observed. Together, these data delineate new genetic diversity of PDA and provide insights into prognostic determinants and therapeutic targets.

Figure 1. Overall mutational landscape of PDA and association with etiological features of disease.

(a) Representative image of the needle dissection approach used to enrich tumour cells for whole-exome sequencing (scale bar, 600 μM) (b) Mutational burden across the sequenced cohort. The presence of genetic alterations in genes associated with mutator phenotypes are shown for the cases with the top 25% of mutational burden. Association with mutational burden was determined by a hypergeometric test. (c) Mutation spectra of the hypermutated cases, top quartile cases with mutator mutations and others (** denotes P<0.05, Student's t-test). The boxes show the distance between the first and third quartile with the whiskers extending up to 1.5 times the interquartile range. (d) Unsupervised hierarchical clustering of cases based on mutation spectra. (e) Mutation spectra based on smoking status, the increase in G>T transversion is significant (P<0.05, Student's t-test). The boxes show the distance between the first and third quartile with the whiskers extending up to 1.5 times the interquartile range. The unfilled circles represent possible outliers. (f) Association of PDA smoking status with overall survival. Hazard ratio and P value were obtained from Cox proportional hazard test.

Figure 2. Copy number alterations in PDA.

(a) APC clustering of PDA cases by CNV revealed several distinct clades and associated clusters. Clusters 5 and 6 exhibit higher overall CNV relative to the other clusters. (b) Clusters 5 and 6 are enriched for mutations or homozygous deletion of genes involved in double-strand break repair, but not in TP53. Significance was determined by hypergeometric test. (c) Cluster 5 and 6 trend towards poor survival relative to clusters with less CNV. Hazard ratio and P values were obtained from Cox proportional hazard test. (d) GISTIC analysis of cases reveals chromosomal regions that are significantly deleted/amplified. (e) Kaplan–Meier analysis of the association of the 8q24.13 locus amplification with overall survival. P value was obtained from Cox proportional hazard test. (f) Fluorescence in situ hybridization with MYC break apart probes demonstrates amplification in the absence of translocation at the MYC locus (Scale bar, 5 μM).

Figure 3. Significantly mutated genes in PDA.

(a) Mutational significance was determined for SNV and INDELS from the 109 sequenced cases using MutsigCV. Genes were subjected to unsupervised clustering, and the frequency and P value as determined by MutsigCV are shown. (b) Kaplan–Meier analysis of select significantly mutated genes. P value was obtained from Cox proportional hazard test. (c) IHC was used to confirm loss of ARID1A protein in a large cohort of PDA cases (scale bar, 150 μM). Diminished ARID1A protein level was associated with overall survival. P value was obtained from Cox proportional hazard test.

Figure 4. KRAS/BRAF-Pathway in PDA.

(a) Oncomap showing the distribution of KRAS, BRAF and PIK3CA mutation in the cohort. (b) KRAS mutations in the cohort exhibit a diverse spectrum concentrated at known oncogenic codons 12, 13 and 61 (c) Analysis of KRAS mutations in the PDA cohort in comparison with all cancer cases in cbioportal (d) Kaplan–Meier analysis of codon 12 versus codon 61 mutated cases. Kaplan–Meier analysis of codon 12 versus codon 61 mutated cases. P value was obtained from Cox proportional hazard test. (e) Representative staining of pERK in pancreatic cancer cases (scale bar, 150 μM), and the association of pERK staining in PDA cases with KRAS codon 12 versus codon 61 mutations. The boxes show the distance between the first and third quartile with the whiskers extending up to 1.5 times the interquartile range. (f) Analysis of BRAF mutations identified in the PDA cohort in comparison with the analysis of all cancer cases in bioportal. (g) Impact of PLX-4032 on the phosphorylation of ERK in the indicated cell lines. PL45 is KRAS mutated, PDA_014 is a patient-derived cell line harbouring BRAF V600E and MNT1 is a BRAF V600E melanoma cell line (h) Impact of PLX-4032 on cell cycle progression in the indicated cell lines (data analysis was performed in triplicate and error bars denote s.d.). (i) Impact of PLX-4032 on viability in the indicated cell lines (data points are from at least six values and error bars denote s.d.).

Figure 5. Pathways in PDA.

(a) Oncomaps and schematics of pathways with combined genetic lesions occurring at a frequency >15% in the PDA cohort. (b) Correlation between pathways in the PDA cohort reveals relatively weak interactions between pathways. (c) Random forest-based clustering of pathways. (d) Overall survival related to genetic alterations targeting KRAS and TP53 in clusters 1 and 2 versus all others. P value was obtained from Cox proportional hazard test.