Biophysical Society Thematic Meeting| Padova 2019

Quantitative Aspects of Membrane Fusion and Fission

Poster Abstracts

47-POS Board 47 AUTOMATED IDENTIFICATION OF SECRETORY GRANULES IN TIRF IMAGES OF PANCREATIC BETA CELLS Fabio Scarpa 1 ; Elisa Candeo 1 ; Morten Gram Pedersen 1 ; 1 University of Padova, Department of Information Engineering, Padova, Italy Introduction: The loss of first-phase insulin secretion is an early sign of developing type 2 diabetes (T2D). Using a high-resolution TIRF microscopy, we can visualize a subset of docket insulin granules in human beta cells for which localized Ca2+ influx triggers exocytosis with high probability and minimal latency. This pool of granules is absent in beta cells from human T2D donors, and also their insulin release mechanism is slow and not synchronized. To evaluate granule dynamics underlying secretory activity in beta cells, Syntaxin and the insulin granules were marked with fluorescent proteins EGFP and NPY-Cherry. This activity was recorded in a sequence of images. Along the sequence, one can observe a luminous pulse followed by the disappearance of light, which likely represents an exocytosis event, through which the insulin is secreted. Until now, the location of the granules and their disappearance were identified manually. Methods: We propose an algorithm able to detect automatically the peaks of fluorescence and to recognize light disappearance, corresponding to granule location and hormone secretion respectively. The proposed algorithm is mainly based on a Laplacian of Gaussian filter and local thresholding. The former increases sharpness and emphasizes small circular objects, while the latter recognizes the objects with maximum brightness, i.e. granules. This analysis on single image is followed by an analysis on the stack of images, able to identify where and when some granules eventually disappear. Results and conclusions: The algorithm was evaluated on a data-set composed by 9 sequences of images, provided by the University of Uppsala , Sweden . Compared to manual analysis, the proposed algorithm identified granules with 90% accuracy, 79% sensibility and 100% specificity. The automated process is very fast (few seconds per image), objective and reproducible.

89