Automated detection of polysomes in cryoelectron tomography

Abstract

Ribosomes and messenger RNA assemble to polysomes during protein synthesis. Cryoelectron tomography enables detection and identification of large macromolecular complexes under physiological conditions making the method uniquely suitable to study the supercomplexes that govern translation of mRNA into proteins. Here, we describe a method for automated assignment of polysomes in cryoelectron tomograms using the positions and orientations of ribosomes, as localized by template matching on tomographic data, as input. On the basis of a training dataset of expert-curated polysomes in cryoelectron tomograms, we define the relative 3D arrangements of neighboring ribosomes in polysomes. This prior distribution is used in a probabilistic framework for polysome assignment: the localized ribosomes from a tomogram are represented as a graph of which the edge weights are defined by the prior distribution. A Markov Random Field is embedded on the graph structure, and a message-passing algorithm is used to infer a polysome-label for each ribosome, i.e., to cluster ribosomes into polysomes. The performance of the method is assessed based on simulated tomograms and experimental tomograms indicating that polysome detection is reliable for typical signal-to-noise ratios of cryoelectron tomograms.