nERdy: network analysis of endoplasmic reticulum dynamics

Abstract

The endoplasmic reticulum (ER) comprises smooth tubules, ribosome-studded sheets, and peripheral sheets that can present as tubular matrices. ER shaping proteins determine ER morphology, however, understanding their role in tubular matrix formation requires reconstructing the dynamic, convoluted ER network. Existing reconstruction methods are sensitive to parameters or require extensive annotation and training for deep learning. We introduce nERdy, an image processing based approach, and nERdy+, a D4-equivariant neural network, for accurate extraction and representation of ER networks and junction dynamics, outperforming previous methods. Comparison of stable and dynamic representations of the extracted ER structure reports on tripartite junction movement and distinguishes tubular matrices from peripheral ER networks. Analysis of live cell confocal and STED time series data shows that Atlastin and Reticulon 4 promote dynamic tubular matrix formation and enhance junction dynamics, identifying novel roles for these ER shaping proteins in regulating ER structure and dynamics.