Deep manifold learning reveals hidden dynamics of proteasome autoregulation

The 26S proteasome regulates proteostasis and myriad cellular processes. Here we reconstitute the hidden dynamics of human 26S proteasome during protein degradation using cryo-EM data-driven deep learning, which identified 64 proteasomal conformers. These conformers revealed substrate-dependent conformational entanglement of two 19S regulatory particles in the doubly capped holocomplexes. The rate-limiting step in single-nucleotide exchange dynamics of the AAA-ATPase motor is ATPase re-engagement with the substrate, while the rate-limiting step in proteasomal degradation is the initiation of substrate unfolding by the AAA-ATPase motor following deubiquitylation. Nine rare conformers delineated novel ubiquitin-binding sites on the RPN2 and 5 subunits, along with a ubiquitylation site on RPN10, which were verified using site-directed mutagenesis and liquid chromatography-tandem mass spectrometry. Opening of the 20S core particle gate was found to upregulate proteolytic activity by allosterically promoting nucleophilic attack on the scissile peptide bond. Our systematic analyses illuminate a grand hierarchical allostery underlying ubiquitin-mediated proteasome autoregulation.