Use of Machine Learning Reveals Multiple Sub‐Populations in Epithelial Cell Signaling After Injury

Abstract

The purinergic receptors P2X7 and P2Y2, while ion channel and GPCR respectively, play essential roles in coordinating the migration of the corneal epithelium during wound healing. These receptors play related yet distinct roles, with P2X7-induced calcium signaling having an increased probability of propagation between adjacent cells relative to P2Y2-induced signaling events. These indicate that the epithelial cells signal to one another (cell-cell communication) during sheet movement after injury. To delineate the injury response we used specific agonists to and/or inhibitors against the receptors. When the P2X7 receptor is activated with the agonist BzATP we detected many more signaling evens than average in a subset of the cells. We hypothesize that there are distinct sub-populations within the BzATP-induced cells based on the percent of time they are actively signaling.These sub-populations may have different roles in instigating versus propagating P2X7-induced cell-cell communication events. We have approached this question using machine learning to dissect the communication. The analyses used hierarchical clustering to group the cells into sub-populations based on the number of frames in which each cell is actively signaling. Cells induced with BzATP form distinct “high-signaling” and “low signaling” clusters which are 2.5-fold less similar to each other than the clusters observed in cells with P2X7 knocked down using siRNA. Cells stimulated with UTP do not display this distinct clustering pattern regardless of the siRNA knockdown of P2X7. Moreover, knockdown of P2X7 via siRNA or by the competitive inhibitor, A438079, significantly reduces the probability of cell-cell signaling events and cellular migration upon wounding when compared to controls. These results demonstrate that P2X7 has a critical role in cell-cell communication and indicate that there are distinct sub-populations within the P2X7 signaling profile based on the amount of time each cell spends signaling. Future studies have the potentialto dissect communication patterns between cells in pathological tissue.