A fully automated morphological analysis of yeast mitochondria from wide-field fluorescence images.

Mitochondrial morphology is an important parameter of cellular fitness. Although many approaches are available for assessing mitochondrial morphology in mammalian cells, only a few technically demanding and laborious methods are available for yeast cells. A robust, fully automated and user-friendly approach that would allow (1) segmentation of tubular and spherical mitochondria in the yeast Saccharomyces cerevisiae from conventional wide-field fluorescence images and (2) quantitative assessment of mitochondrial morphology is lacking. To address this, we compared Global thresholding segmentation with deep learning MitoSegNet segmentation, which we retrained on yeast cells. The deep learning model outperformed the Global thresholding segmentation. We applied it to segment mitochondria in strain lacking the MMI1/TMA19 gene encoding an ortholog of the human TCTP protein. Next, we performed a quantitative evaluation of segmented mitochondria by analyses available in ImageJ/Fiji and by MitoA analysis available in the MitoSegNet toolbox. By monitoring a wide range of morphological parameters, we described a novel mitochondrial phenotype of the mmi1Δ strain after its exposure to oxidative stress compared to that of the wild-type strain. The retrained deep learning model, all macros applied to run the analyses, as well as the detailed procedure are now available at https://github.com/LMCF-IMG/Morphology_Yeast_Mitochondria .