Cryo-EM and artificial intelligence visualize endogenous protein community members

Abstract

Cellular function is underlined by megadalton assemblies organizing in proximity, forming communities. Metabolons are protein communities involving metabolic pathways such as protein, fatty acid, and thioesters of coenzyme-A synthesis. Metabolons are highly heterogeneous due to their function, making their analysis particularly challenging. Here, we simultaneously characterize metabolon-embedded architectures of a 60S pre-ribosome, fatty acid synthase, and pyruvate/oxoglutarate dehydrogenase complex E2 cores de novo. Cryo-electron microscopy (cryo-EM) 3D reconstructions are resolved at 3.84–4.52 Å resolution by collecting <3,000 micrographs of a single cellular fraction. After combining cryo-EM with artificial intelligence-based atomic modeling and de novo sequence identification methods, at this resolution range, polypeptide hydrogen bonding patterns are discernible. Residing molecular components resemble their purified counterparts from other eukaryotes but also exhibit substantial conformational variation with potential functional implications. Our results propose an integrated tool, boosted by machine learning, that opens doors for structural systems biology spearheaded by cryo-EM characterization of native cell extracts.