Mitogen-activated protein kinase signaling is a conserved mechanism explaining phenotypic heterogeneity across diverse drivers of epithelial-mesenchymal transition in pancreas cancer

Introduction: The epithelial-mesenchymal transition (EMT) is a developmental process aberrantly reactivated in cancers, including pancreatic ductal adenocarcinoma (PDAC), that contributes to disease progression and chemoresistance [1-3]. EMT occurs heterogeneously among PDAC ductal cells in vitro and in vivo, which frustrates efforts to identify the underlying signaling regulatory mechanisms [4]. We hypothesize that cell-to-cell variation in the activities of specific signaling pathways that drive EMT explains heterogeneity in EMT induction within cell populations.

Materials and

Methods: We developed a workflow to analyze EMT signaling and phenotypic heterogeneity by: 1) using iterative indirect immunofluorescence imaging [5] (4i) to quantify EMT phenotypes and signaling pathway activity proxies for up to seven purported EMT-regulating pathways with single-cell resolution and 2) implementing a multivariate mutual information computational model based on binned imaging data to identify the pathways that are most informative of EMT phenotype. We used this integrated experimental and computational approach to quantify the extent to which variation in the activation of purported EMT signaling pathway nodes explains the EMT state of the human PDAC cell line HPAF-II. EMT was induced by treating cells with exogenous growth factors found in the tumor microenvironment (50 ng/mL epidermal growth factor, 50 ng/mL hepatocyte growth factor, 10 ng/mL transforming growth factor β1, or two combinations of those growth factors) or with chemotherapeutics (10 ng/mL gemcitabine or 5-fluorouracil). EMT markers and proxies for signaling pathway activities were assessed daily for up to 96 hours. CellProfiler was used to align images from separate rounds of 4i, segment cells, and quantify whole-cell, cell-edge, cytoplasmic, and nuclear fluorescence intensities. In vivo mutual information calculations were conducted using a patient-derived xenograft mouse model of PDAC with the Mesmer algorithm for tissue section cell segmentation. To develop a cell population-level model for EMT regulation by signaling to compare against our single-cell mutual information model results, we used immunoblotting combined with linear regression. Mutual information model results were validated through pharmacological inhibition and siRNA-mediated knockdown studies.

Results, Conclusions, and Discussions: Four rounds of 4i imaging captured a heterogeneous induction of EMT and signaling pathway activation (Figure 1A-C). A mutual information analysis of these data identified ERK and SRC signaling as explaining significantly more EMT heterogeneity than other signaling pathways in both growth factor and chemotherapy-driven EMT (Figure 1D). A linear model of population-level western blot measurements of these signals incorrectly suggested that JNK activity (phosphorylated c-Jun proxy) holds greater importance than ERK activity (phosphorylated ERK proxy). Inhibition or knockdown of each pathway showed that MEK inhibition reduced EMT more than did JNK inhibition but that the combination of the two effected the largest reduction in EMT. In MEK-inhibited cells, mutual information analysis revealed that JNK becomes activated and compensates for lost ERK signaling to drive EMT. This JNK-mediated compensation for EMT was also detected in patient derived xenograft tumors treated with or without the MEK inhibitor selumetinib. Our results demonstrate the pervasive importance of ERK for diverse drivers of EMT and the need to account for the effects of heterogeneous signaling pathway activation in EMT regulation to identify optimal drug targets for therapeutic efficacy.

Acknowledgements (Optional): This work was supported by NCI U01 CA243007 (MJL), NCI F31CA275364 (MCB), NCI P30 CA44579, and NCI 5T32CA009109.

References: [1] Roche, J. Cancers. 2018 [2] Rhim, AD. Cell. 2012 [3] Zheng, X. Nature. 2015 [4] Porter, RL. PNAS. 2019 [5] Gut, G. Science. 2018.