Immune Cytolytic Activity Stratifies Molecular Subsets of Human Pancreatic Cancer

Abstract

Purpose: Immunotherapy has the potential to improve the dismal prognosis in pancreatic ductal adenocarcinoma (PDA), but clinical trials, including those with single-agent PD-1 or PD-L1 inhibition, have been disappointing. Our aim was to examine the immune landscape of PDA as it relates to aspects of tumor biology, including neoepitope burden.Experimental Design: We used publicly available expression data from 134 primary resection PDA samples from The Cancer Genome Atlas to stratify patients according to a cytolytic T-cell activity expression index. We correlated cytolytic immune activity with mutational, structural, and neoepitope features of the tumor.Results: Human PDA displays a range of intratumoral cytolytic T-cell activity. PDA tumors with low cytolytic activity exhibited significantly increased copy number alterations, including recurrent amplifications of MYC and NOTCH2 and recurrent deletions and mutations of CDKN2A/B In sharp contrast to other tumor types, high cytolytic activity in PDA did not correlate with increased mutational burden or neoepitope load (MHC class I and class II). Cytolytic-high tumors exhibited increased expression of multiple immune checkpoint genes compared to cytolytic-low tumors, except for PD-L1 expression, which was uniformly low.Conclusions: These data identify a subset of human PDA with high cytolytic T-cell activity. Rather than being linked to mutation burden or neoepitope load, immune activation indices in PDA were inversely linked to genomic alterations, suggesting that intrinsic oncogenic processes drive immune inactivity in human PDA. Furthermore, these data highlight the potential importance of immune checkpoints other than PD-L1/PD-1 as therapeutic targets in this lethal disease. Clin Cancer Res; 23(12); 3129-38. ©2016 AACR.

Figure 1.. Stratification of human PDA based on cytolytic index.

(A) Cytolytic index (geometric mean of expression of GZMA and PRF1 in transcripts per million (tpm)) across TCGA tumor types. Kidney renal clear cell carcinoma (KIRC, n = 606), lung adenocarcinoma (LUAD, n = 116), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC, n = 309), lung squamous cell carcinoma (LUSC, n = 553), pancreatic adenocarcinoma (PAAD, n = 134, red underline), stomach adenocarcinoma (STAD, n = 418), head and neck squamous cell carcinoma (HNSC, n = 566), colon adenocarcinoma (COAD, n = 328), skin cutaneous melanoma (SKCM, n = 105), bladder urothelial carcinoma (BLCA, n = 427), esophageal carcinoma (ESCA, n = 196), liver hepatocellular carcinoma (LIHC, n = 371), thyroid carcinoma (THCA, n = 572), ovarian serous cystadenocarcinoma (OV, n = 309), glioblastoma multiforme (GBM, n = 169), and prostate adenocarcinoma (PRAD, n = 555). Inset, cytolytic index between normal pancreas expression levels obtained from Genotype-Tissue Expression (GTEx) project and TCGA PAAD. (B) Distribution of cytolytic genes within pancreatic adenocarcinoma. Gene set variation analysis (GSVA) signature scores for cytolytic index distinguished top decile (orange) and bottom quartile (green) samples for cytolytic-high (CYT High) and low (CYT Low) tumors, respectively.

Figure 2.. Cytolytic index correlates with classifiers of PDA subtypes.

(A) Hierarchical clustering of GSVA signature scores for gene programs defining PDA subtypes from Collisson et al., 2011, Moffitt et al., 2015, and Bailey et al., 2016. (B) Distribution of GSVA signature scores for each PDA subtype program between cytolytic-high and low tumors. ** = FDR adjusted P-values ≤ 0.05 and N.S. = Not statistically significant. (C) Hierarchical clustering of Moffitt Normal Stroma gene expression between cytolytic-high and low tumors showing enrichment in cytolytic-high tumors. (D) Hierarchical clustering of Bailey GP1 Pancreatic Progenitor gene expression between cytolytic-high and low tumors showing enrichment in cytolytic-low tumors.

Figure 3.. Low cytolytic index is associated with increased copy number alterations in PDA.

(A) Co-mutation plot showing significantly mutated genes (SMGs, FDR < 0.1) in cytolytic subsets in the PAAD dataset. Red boxes indicate mutation. SMGs that correlate with cytolytic subtypes (p < 0.05) are highlighted by green or orange circles in the left column. Genome MuSiC (v0.4) FDR p-values for SMGs are plotted in –log10 on the right. (B) Nonsynonymous mutation spectra across PDA cytolytic subsets. (C) KRAS mutation types across PAAD dataset and association with cytolytic index, showing no statistically significant correlation between KRAS mutations and cytolytic subsets. (D) GISTIC2.0 analysis identified recurrent somatic copy number alterations (SCNA) in cytolytic-low tumors. Recurrent amplifications at 8q24.21 (MYC), 1p12 (NOTCH2), 8p11.22 (FGFR1), and deletions at 9p21.3 (CDKN2A/B), 18q21.2 (SMAD4) in CYT low tumors. (E) Total SCNA were calculated for each TCGA PAAD patient and were significantly increased in cytolytic-low tumors (Mann-Whitney). (F) Co-mutation plot of copy number alterations and non-silent SNVs/INDELs in genes amplified or lost in cytolytic-low PDA tumors.

Figure 4.. Inhibitory checkpoint molecules, but not neoepitope load, are associated with cytolytic index in PDA.

(A) Local regression curves (Spearman rank correlation) between cytolytic index and total mutation count and boxplot distributions between cytolytic subsets (Mann-Whitney). (B) Local regression curves and boxplot distributions between cytolytic subsets for cytolytic index and total MHC class I neoepitopes (50 nM predicted binding affinity) and (C) cytolytic index and number of mutations generating ≥ 1 neoepitopes to MHC class I. (D) Local regression curves and boxplot distributions between cytolytic subsets for cytolytic index and total MHC class II neoepitopes (<1% rank) and (E) cytolytic index and number of mutations generating ≥ 1 neoepitopes to MHC class II. (F) Differentially expressed chemokines, cytokines, and inhibitory checkpoint molecules between cytolytic-high (top decile) and low (bottom quantile) samples across TCGA. Fold change between subtypes indicated by color. Size of circle indicates statistical significance (−log10(Adjusted P value)). Arrow and box indicate no differential expression of PD-L1 (CD274) in between cytolytic subtypes in TCGA PAAD dataset. (G) Distribution of inhibitory immune checkpoint index (geometric mean of TPM values) across PAAD TCGA. Checkpoint molecules: CD274 (PD-L1), IDO2, PDCD1LG2 (PD-L2), CTLA4, IDO1, ADORA2A (A2AR), LAG3, PDCD1 (PD1), TIGIT, HAVCR2 (TIM3), VISTA (C10orf54), VTCN1 (B7-H4). (H) Local regression curve showing statistical significant relationship between cytolytic index and inhibitory immune checkpoint index in PDA (Spearman rank correlation). (I) Expression of differentially expressed Treg markers in PDA subsets. (J) Expression of differentially expressed inhibitory checkpoint molecules. N.S = not statistically significant, ** = FDR adjusted P-values ≤ 0.1.