Association of Distinct Mutational Signatures With Correlates of Increased Immune Activity in Pancreatic Ductal Adenocarcinoma

Abstract

Importance: Outcomes for patients with pancreatic ductal adenocarcinoma (PDAC) remain poor. Advances in next-generation sequencing provide a route to therapeutic approaches, and integrating DNA and RNA analysis with clinicopathologic data may be a crucial step toward personalized treatment strategies for this disease.

Objective: To classify PDAC according to distinct mutational processes, and explore their clinical significance.

Design, setting, and participants: We performed a retrospective cohort study of resected PDAC, using cases collected between 2008 and 2015 as part of the International Cancer Genome Consortium. The discovery cohort comprised 160 PDAC cases from 154 patients (148 primary; 12 metastases) that underwent tumor enrichment prior to whole-genome and RNA sequencing. The replication cohort comprised 95 primary PDAC cases that underwent whole-genome sequencing and expression microarray on bulk biospecimens.

Main outcomes and measures: Somatic mutations accumulate from sequence-specific processes creating signatures detectable by DNA sequencing. Using nonnegative matrix factorization, we measured the contribution of each signature to carcinogenesis, and used hierarchical clustering to subtype each cohort. We examined expression of antitumor immunity genes across subtypes to uncover biomarkers predictive of response to systemic therapies.

Results: The discovery cohort was 53% male (n = 79) and had a median age of 67 (interquartile range, 58-74) years. The replication cohort was 50% male (n = 48) and had a median age of 68 (interquartile range, 60-75) years. Five predominant mutational subtypes were identified that clustered PDAC into 4 major subtypes: age related, double-strand break repair, mismatch repair, and 1 with unknown etiology (signature 8). These were replicated and validated. Signatures were faithfully propagated from primaries to matched metastases, implying their stability during carcinogenesis. Twelve of 27 (45%) double-strand break repair cases lacked germline or somatic events in canonical homologous recombination genes-BRCA1, BRCA2, or PALB2. Double-strand break repair and mismatch repair subtypes were associated with increased expression of antitumor immunity, including activation of CD8-positive T lymphocytes (GZMA and PRF1) and overexpression of regulatory molecules (cytotoxic T-lymphocyte antigen 4, programmed cell death 1, and indolamine 2,3-dioxygenase 1), corresponding to higher frequency of somatic mutations and tumor-specific neoantigens.

Conclusions and relevance: Signature-based subtyping may guide personalized therapy of PDAC in the context of biomarker-driven prospective trials.

Figure 1.. Mutational Signatures in Primary and Metastatic Pancreatic Ductal Adenocarcinoma

A, Bar plot of proportion of 7 merged signatures in each of the 160 discovery tumors, sorted by hierarchical clustering (dendogram at bottom), showing germline (dark blue), somatic (mauve), and occult (white) double-strand break repair (DSBR) etiologies and heat maps for total number of single-nucleotide variants (SNVs), total number of neoantigens, total number of indels, total number of short deletions (dels) greater than 3 base pairs (bp), total number of structural deletions, and transcriptional subtypes (Moffitt tumor class, Collisson class, and Bailey class) in cases for which RNA sequencing is available for the tumor. B, Bar plots of proportion of 7 merged signatures in paired primary tumors and metastases from 4 cases. ADEX indicates aberrantly differentiated endocrine exocrine.

Figure 2.. Etiologic Stratification of Double-Strand Break Repair Genomes

Boxplots of proportion of single-nucleotide variants (SNVs) attributed to signature 3, number of short deletions greater than 3 base pairs (bp) in length, number of structural variants, and number of large (structural) deletions in the double-strand break repair subtype divided by etiology—germline, BRCAness/occult, or somatic —and in the age-related subtype, for amalgamated discovery and replication cohorts. All values are significantly greater in both double-strand break repair germline and BRCAness/occult groups relative to the age-related subtype (P < .001 for each, Wilcoxon test). The horizontal line in the middle of each box indicates the median, while the top and bottom borders of the box mark the 75th and 25th percentiles, respectively. The top whisker marks the 75th percentile plus 1.5 times the interquartile range. The bottom whisker marks the 25th percentile minus 1.5 times the interquartile range. Each point indicates a tumor genome.

Figure 3.. Association of Genetic Inactivations With Double-Strand Break Repair (DSBR) Signature

Scatterplot of proportions of cases with biallelic inactivation of every gene in the DSBR subtype primary tumors (n = 27) vs those in the age-related subtype primary tumors (n = 169) for the amalgamated discovery and replication cohorts. Driver genes include CDKN2A, SMAD4, and TP53. FDR indicates false discovery rate.

Figure 4.. Integrated Genomic and Transcriptomic Features of Antitumor Immunity in Pancreatic Ductal Adenocarcinoma

A, Heat map of median expression of gene sets representative of categories of immune function by signature group for discovery cohort cases with tumor cellularities between 20% and 80%. B, Scatterplot of number of neoantigens vs number of somatic single-nucleotide variants (SNVs) per tumor, colored by signature-based subtype, for 137 discovery cohort cases to which we confidently assigned HLA class 1 genotypes. The regression line from the linear model (y ~ x) is shown in black with areas between confidence bands shaded in gray. APC indicates antigen-presenting cell; DSBR, double-strand break repair; MHC, major histocompatibility complex; MMR, mismatch repair; pDC, plasmacytoid dendritic cell.

Figure 5.. Correlation of Immunohistochemistry With RNA Sequencing

A, Programmed cell death 1 ligand (PD-L1) and CD8 immunohistochemical (IHC) expression in representative cancer tissue microarray spots showing high and low expression of PD-L1 and CD8 counts. B, Median (dotted lines) and interquartile ranges (shaded regions) of expression of PD-L1, CD8A, and cytolytic activity (left-hand y axis) and absolute counts of cells with IHC staining for CD8 (right-hand y axis) at each level of PD-L1 IHC staining (0-3) (see Methods). CD8 staining cell counts and CD8A expression were strongly correlated (P < .001, r = 0.744, Pearson correlation). Programmed cell death 1 ligand and cytolytic activity expression were significantly higher across PD-L1 staining levels (P for PD-L1 = .006, P for cytolytic activity = .01, PD-L1 0-1 vs 2-3 staining, Wilcoxon test). FPKM indicates fragments per kilobase of exon per million fragments mapped.