Abstract 4750: Time-of-diagnosis prognostication of pancreatic ductal adenocarcinoma based on single cell RNA-seq and digital cytometry

Abstract

Abstract Introduction: Recent work has described tumor heterogeneity in pancreatic ductal adenocarcinoma (PDAC). Here we use scRNA-seq followed by CIBERSORTx and CytoTRACE to profile specific cell states within PDAC tumors at time-of-diagnosis, and correlate these states with clinical outcomes. Methods: We performed diagnostic endoscopic ultrasound (EUS) guided core needle biopsies in ten patients with locally advanced/resectable PDAC, from which we generated scRNA-seq data to identify unique cell types and states. To validate the prognostic utility of these signatures, we utilized CIBERSORTx to deconvolve bulk RNA-seq data from two external datasets - 1) 170 cases from The Cancer Genome Atlas (TCGA) and 2) 51 cases from Kirby et al (Mol Oncol 2016). We also used CytoTRACE to predict tumor cell differentiation status from the scRNA-seq data. Results: From ~9,800 total cells in our single cell cohort, we identified 16 cell types and 25 cell states. Bulk RNA-seq deconvolution was performed using the expression profiles from these 25 cell states. Two groups of malignant ductal cells were found to negatively influence patient survival (p of .008 and .041), and a group CD14+ monocytes and CD4+ T cells were found to positively influence patient survival (p of .011 and .012). We performed further analysis on the two malignant ductal cell populations, which displayed gene expression profiles similar to PDAC subtypes (squamous-like and progenitor) previously described by RNA-seq gene set enrichment analysis (Bailey et al; Moffitt et al). On CytoTRACE analysis of our scRNA-seq data, the squamous-like tumor cell population remained distinct from the larger progenitor population, and was predicted to be significantly more differentiated. We stratified TCGA and Kirby samples based upon cell fraction of this squamous-like population. A Kaplan-Meier estimate was used to quantify survival over time of the two groups (>7% squamous-like vs. <7% squamous-like). The >7% group displayed a marked decrease in survival - log-rank p-values of .025 (TCGA) and .00002 (Kirby). When filtering for only TCGA samples that went on to receive chemotherapy this trend remained highly significant (p of .022), with 46% of the >7% group vs. 82% of the >7% group surviving beyond one year. Conclusion: scRNA-seq from EUS core biopsies identified distinct prognostic tumor cell state fractions that were corroborated by pre-treatment bulk RNA-seq through deconvolution. These novel investigational and digital cytometric approaches not only confirmed prior tumor subtypes and their prognoses, but also enabled the quantification of their fractions for additional prognostic value. Patients with certain cell state fraction profiles may benefit from combinatorial therapeutic interventions, especially in the context of clinical trials. Citation Format: Erik Storrs, Abul Usmani, Bradley Krasnick, Chris Wetzel, Thomas Hollander, Prathamesh Chati, Faridi Qaium, Hephzibah Anthony, Ian Sloan, Natalie Cosgrove, Gabriel Lang, Vladimir Kushnir, Daniel Mullady, Dayna Early, William Hawkins, Koushik Das, Ryan Fields, Aadel Chaudhuri. Time-of-diagnosis prognostication of pancreatic ductal adenocarcinoma based on single cell RNA-seq and digital cytometry [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 4750.