(A-22) Validation of Photostimulatory Responses in Islet Cells Using Computer Vision Algorithms

Introduction:: Development of potential treatments of Diabetes Mellitus is urgently needed. Because hormone secretion from the islet cells critically depends on the calcium dynamics, modulation of calcium spiking in β- and α- cells to regulate insulin and glucagon, respectively, has been extensively studied. Our laboratory has shown that through activation of the voltage-sensitive calcium channels by non-invasive electric field stimulation (EFS) [1], the calcium dynamics can be modulated in the islet cells. More recently, we developed a computer vision pipeline, BetaBuddy [2] to automatically perform cellular segmentation, registration, and tracking over time. Here, we applied the pipeline to re-analyze the multiple images of mouse βTC6 and αTC1 cells that were stimulated photonically to promote hormone secretion through modulating calcium dynamics. Unlike EFS, photostimulation (referred to as photobiomodulation) is postulated to couple to the electron transfer chain in mitochondria that increases calcium spiking and ATP production.

Materials and Methods:: Mouse-derived βTC6 and αTC1 insulinoma cells were cultured in high glucose DMEM (4.5 g/L glucose, Sigma) and low glucose DMEM (1.0 g/L glucose, Sigma), as needed. The cell culture conditions were described in detail in our previous publication [3]. Briefly, Fluo-8 was used to observe changes in calcium activity with 1 drop of NucBlue to visualize nuclei. For photobiomodulation (PBM) studies, cells were exposed to an 810 nm laser (Cytonsys; Austin, TX) for 60 s at an intensity of 150 mW/cm2 that produced an energy density of 9 J/cm2 (fluence). Cells were imaged for 2 min prior to photostimulation to establish a baseline calcium activity. Following exposure to photostimulation, the cells were then imaged for 2 min at t = 5, 10, and 15 min time points to minimize fluorophore degradation. Fluo-8 images were previously analyzed manually. Using our computer vision pipeline, the same image stacks were automatically segmented, registered, and tracked over time. All registered cells then underwent automated background subtraction, normalization, and calcium spike detection processes. Calcium spikes were defined as 10% increases in fluorescence from a specified cell’s preceding peak in intensity. Fold-rise in calcium activity was calculated at each time point post-stimulation, and these results were compared to earlier findings determined through manual segmentation and analysis.

Results, Conclusions, and Discussions::
Results and Discussion: A flowchart of the machine learning pipeline, referred to as BetaBuddy, is shown in Figure 1. Once cell segmentation is completed and each cell is registered, the pipeline can accurately track the time-dependent changes of intracellular calcium dynamics and display them in a single plot. Statistical analyses indicate significant increases (e.g., calcium spikes/cell) in both β-and α-cells post-photostimulation. This is in general agreement with manual cell segmentation and analysis that we reported previously [3]. However, automated analyses show that the manual cell segmentation may have resulted in an overestimation of calcium spiking, which was found to be more noticeable in α-cells. Inconsistent manual segmentation may be attributed as a potential cause of such discrepancy. The BetaBuddy pipeline can therefore be applied to analyze subcellular responses more accurately and elucidate intricate details of the complex biosystems. Furthermore, photobiomodulation appears to be a promising non-invasive, and non-biologic stimulus to promote hormone secretion from the islet cells.




Conclusion: While manual segmentation methods have proven useful, automated analysis by the BetaBuddy pipeline appears to provide a powerful alternative. Moreover, computer vision pipelines are faster (hours vs. < 15 min) and more efficient for batch analysis of large datasets such as multiple stacks of subcellular images. The automated pipeline essentially eliminates the human-in-the-loop requirement and is capable of segmenting more cells that are often found spatially clustered and thus excluded when performing manual segmentation. Finally, the current pipeline is being modified to probe and determine cell-to-cell communication and its effects in calcium dynamics and, ultimately, the regulation of hormone secretion. Characterization of isolated islets using machine learning pipelines will provide a reliable methodology for transplantation of viable islets in Type I diabetic patients.




Acknowledgements (Optional): : Authors are grateful for the support of the NIH Predoctoral Training Grant (HL134613; PI, Kytai Nguyen), the University of Texas System Rising STARs Award (J.L), the CPRIT First Time Faculty Award (J.L), and the Alfred R. and Janet H. Potvin Endowment (M.C).

We would like to thank Dr. Caleb Liebman for his work culturing insulinoma (βTC-6) mouse cells and providing the fluorescein calcium images.

References (Optional): :
1. Liebman C, Vu TM, Phillips A, Chen B, Cho M. Altered β-cell calcium dynamics via electric field exposure. Ann Biomed Eng. 2021;49(1):106-114. doi:10.1007/s10439-020-02517-w




2. Alsup AM, Fowlds K, Cho M, Luber JM. BetaBuddy: An end-to-end computer vision pipeline for the automated analysis of insulin secreting β-cells. Published online April 6, 2023:2023.04.06.535890. doi:10.1101/2023.04.06.535890




3. Liebman C, Loya S, Lawrence M, Bashoo N, Cho M. Stimulatory responses in α- and β-cells by near-infrared (810 nm) photobiomodulation. J Biophotonics. 2022;15(3):e202100257. doi:10.1002/jbio.202100257