Lipid droplet size profiling in yeast

Lipid droplets are central hubs of cellular lipid metabolism and play crucial roles in health and disease. The yeast Saccharomyces cerevisiae serves as a powerful model for studying lipid droplet biology at the cellular level. Estimation of the number and size distribution of lipid droplets in a cell population is critical for functional studies. However, accurate segmentation and quantification of lipid droplets through image-based methods present significant challenges. Organelle motion, point-spread function (PSF) overlap of closely associated organelles and heterogeneous fluorescence labeling often compromise histogram-, shape- or machine-learning-based methods. Here, we present an alternative, seeding-based radial raytracing approach that is independent of image histograms and considers the PSF-blurred nature of imaged organelles. To improve lipid droplet resolution and contrast, we applied custom 3D deconvolution and integrated a supervised machine-learning step that corrected typical deconvolution artifacts. An integrated, extensible toolset including deep-learning-based cell registration enables detailed lipid droplet statistics, including user-defined size classes. As a proof of principle, we applied our approach to analyze lipid droplets in wild-type and mutant cells lacking the Sei1-Ldb16 seipin complex required for regular lipid droplet biogenesis.