(B-79) Computer Vision Pipeline for Automated Analysis of Poloxamer 188 Treatment of bTBI

Introduction:: Triblock amphiphilic copolymer P188 was originally approved by FDA to reduce viscosity in blood transfusion. Using tissue engineering approaches, our laboratory developed in vitro models to establish the efficacy of P188 treatment for blast-induced traumatic brain injury (bTBI) [1]. While P188 can reseal the compromised cell membrane and restore its structure and function, this surfactant is capable of entering the cell and accumulates in the cytoplasm. Potential transport mechanisms may include GTP-protein mediated endocytosis. Moreover, we developed a machine learning-based computer vision pipeline to automate cell segmentation, improve image analysis, and significantly reduce the time typically required of laborious hand-drawn identification and analysis of individual cells from hours to < 15 min. The pipeline also provided spatially and temporally resolved information of every individual cell that was monitored and tracked. Such a pipeline is leading us to formulate novel hypotheses that are tested to determine the effectiveness of P188 to reseal the cell membrane and enhance intrinsic subcellular repair machinery.


Materials and Methods:: Mouse brain endothelial cells (BECs) were cultured to mimic the brain endothelium. P188 was conjugated with Rhodamine 110 (Rh-110) or FITC fluorophore. Cells were incubated with fluorophore-conjugated P188 (10 μM), and fluorescent images were acquired at 1 min intervals for 30 mins for transport mechanism studies. Fluorescent images were acquired for 48 hrs to determine long-term effects of P188. Mechanical damage (shockwave) and saponin-induced cell permeabilization were applied to the BECs to elucidate the ability of P188 to repair the injured brain endothelium. A deep learning-based generalist cellular segmentation algorithm, Cellpose [2], was adopted and modified to automatically outline the boundary of each cell and create regions of interest (ROIs). ‘Mask’ images were generated that delineate individual cells by assigning every ROI a different grayscale color. The unique color was used as a label during cell registration and tracking to facilitate further data analysis.


Results, Conclusions, and Discussions:: Results and Discussion: In control BECs, P188 was quickly transported into cells in < 30 min, and its concentration inside the cell increased by an order of magnitude. However, excess P188 in control cells was found to be exocytosed in < 1 hr (Fig. 1). Transport across the cell membrane was not passive, required intact cytoskeletal organization, and inhibited by dynasore that blocks GTP-protein mediated endocytosis. In saponin-damaged cells, P188 accumulation was even greater and exceeded 3-folds in comparison to that observed in control cells. Cell membrane permeabilization likely created membrane defects (e.g., pores) through which rapid diffusion of P188 was allowed. Transport characteristics and exocytosis were calculated using the computer vision pipeline. The off-the-shelf segmentation AI, Cellpose [2], had a segmentation accuracy ranging from 60-77% in fluorescent images of P188-Rh110 in BECs and greatly reduced the time required for cell segmentation. However, a specialized deep-learning based model for clustering cells (e.g., monolayer of endothelial cells) is deemed necessary to increase accuracy of segmentation and essentially eliminate the need for human-in-the-loop adjustments of incorrect ROIs.

Conclusion: Our machine learning pipeline can be applied to examine in detail the previously unknown therapeutic effects of P188 with time efficiency and data accuracy. We are now re-analyzing the dynamics of P188 uptake in blast-injured cells and under other traumatic conditions (e.g., chemical or thermal). This combinatory strategy of utilizing machine learning to better design the future experiments in a predictive manner will significantly facilitate the development of optimizing this FDA-approved non-biologic treatment for repairing damaged tissues. Machine learning is also enabling the investigator to gain molecular insights with an unprecedented level of details. Additionally, developing a segmentation model trained on clustered cells is expected to lead to an automatic analysis platform for TBI research and drug repurposing through identification of transport mechanisms and cell-drug interactions.


Acknowledgements (Optional): : Authors are grateful for financial support from the University of Texas System Rising STARs Award (J.L), the CPRIT First Time Faculty Award (J.L), and from the Office of Naval Research (N00014-21-1-2039, M.C) and the Alfred R. and Janet H. Potvin Endowment (M.C).

References (Optional): : 1. Inyang, E., Abhyankar, V., Chen, B. et al. Modulation of in vitro Brain Endothelium by Mechanical Trauma: Structural and Functional Restoration by Poloxamer 188. Sci Rep 10, 3054 (2020). https://doi.org/10.1038/s41598-020-59888-2

2. Stringer, C., Wang, T., Michaelos, M. et al. Cellpose: a generalist algorithm for cellular segmentation. Nat Methods 18, 100–106 (2021). https://doi.org/10.1038/s41592-020-01018-x