AI-based AlphaFold2 significantly expands the structural space of the autophagy pathway

A structural understanding of entire cellular processes has been an uncharted realm, until now. The artificial intelligence-based tool AlphaFold2 (AF2) has substantially changed the prediction accuracy, and predicted models of entire proteomes are now available. Here, we have examined AF2’s prediction of 38 core macroautophagic/autophagic proteins and 378 interacting partners representing the human autophagic interactome. Prior to AF2, ~50% of the proteins lacked atomistic level resolution and we found significant improvement in structural coverage by AF2, with an addition of ~ 47% of the residues modeled with reasonable confidence. We also augmented this structural information with μs timescale molecular dynamics simulations, in particular, ATG2, ATG10, and ATG14. ATG2A, a bipartite membrane protein with rodlike architecture was predicted with high accuracy and our simulations revealed dynamic transitions of cavity-lining residues that might play a critical role in regulating lipid transfer. In addition, a promising approach of multimeric prediction by AF2 revealed the architecture of ATG7-ATG10, a tetrameric complex that participates in conjugation machinery in autophagy. By combining computational and experimental approaches, we demonstrated that three salt bridges were crucial to ATG7-ATG10 complex formation and mutating these residues abrogated the binding. We have also generated a web resource with curated AF2 structural models, simulated conformational ensemble, and structural analysis that will be highly pertinent to the autophagy community. Altogether, our work presents a robust pipeline to utilize AF2 as a tool for a starting point to provide the dynamic behavior of molecules in a given biological pathway.