Clinical genomic implications of transcriptional subtypes in pancreatic cancer.

Abstract

4145 Background: Transcriptional profiling of pancreatic cancers (PC) has defined classical and basal subtypes; basal tumors have worse outcomes. Mesenchymal (MES) and neural-like progenitor (NRP) subtypes are increasingly recognized and enriched post-therapy. Initial data suggests worse outcomes to FOLFIRINOX (FFX) compared with gemcitabine nab-Paclitaxel (GnP) in basal tumors. Several clinical trials are ongoing to investigate this. Here, we examined the clinical implications of transcriptional subtypes in a large, real-world dataset. Methods: Retrospective IRB exempt, deidentified data was examined from NextGen DNA and RNA sequencing performed on PCs at Caris Life Sciences (Phoenix, AZ). Classical and basal cell states were identified using RNA-seq and the PurIST algorithm in a genomic cohort or GATA6 and KRT5 expression levels in a clinical cohort. Tumor microenvironment immune cell composition on RNA seq was performed using QuantiSeq. Survival was obtained from insurance claims data and calculated from first treatment date to last known contact. Kaplan-Meier estimates were calculated for patient cohorts. P values were adjusted using Benjamini-Hochberg correction. Results: A total of 7,250 PCs were profiled in the genomic cohort. 3,063 tumors (42.2%) were strongly classical (SC), 2,015 tumors (27.8%) were strongly basal (SB) and the remaining had mixed phenotypes. MES and NRP marker genes were significantly co-expressed with each other, with basal genes, and anti-correlated with classical genes. When compared to SC, SB had significantly higher mutation rates in KRAS (93% vs. 88%), TP53 (83% vs. 72%) and ARID1A (12% vs. 8%), whereas SMAD4 (23% vs. 17%) mutations were more common in SC (all q < 0.05). There were no differences in mutation rates in homologous recombination or mismatch repair genes. SB had a significantly higher fraction of M1 macrophages (fold change [FC]: 1.14) and neutrophils (FC 1.16), whereas SC tumors had higher M2 macrophages (FC 1.18), NK (FC 1.2), and dendritic cells. Overall proportions of CD4/8 T cells were low and not different. Interestingly, SB had higher levels of PD-L1 by IHC (4.8% vs. 35%) and higher expression of immune exhaustion genes including CTLA4 (FC 1.19), TIM3 (FC 1.22) and PD-1 (FC 1.43) (all q < 0.05). The clinical cohort had 1,623 patients. Basal tumors had an inferior survival (median survival: 8.2 months (mo) vs 13.3 mo (Hazard Ratio (HR) 0.67, p < 0.00001)) and showed a significant improvement in outcomes when treated with upfront FFX vs GnP (n = 80 vs 90, Median: 15.8 vs 7.4 mos., HR 0.68, p = 0.021). This difference between FFX vs GnP was less pronounced in classical tumors (n = 70 vs 89, Median: 17.3 vs 15.4 mos, HR 0.70, p = 0.049). Conclusions: Our work represents the largest known real world molecular comparison of transcriptional subtypes of PC. Differential outcomes for patients with basal tumors treated with FFX versus GnP warrants further investigation in prospective studies.