BPS2026 – Revealing the conformational dynamics of Hsp104 using Markov state modeling and machine learning

Molecular chaperones play a central role in protein quality control mechanisms, as major components of the cellular proteostasis network. In response to various cellular stresses, proteins can undergo misfolding and aggregations, which are strongly associated with a wide spectrum of human disorders. Chaperones such as heat shock protein 104 (Hsp104), a disaggregase found in yeast, and its bacterial homolog ClpB, members of the AAA+ (ATPases associated with diverse cellular activities) superfamily, protect against such deleterious pathways by mediating protein disaggregation. Hsp104 is a powerful nanomachine which disassembles a wide range of substrates, from amorphous aggregates to highly ordered amyloids, while its bacterial homolog ClpB only processes amorphous aggregates. Hsp104 is a donut-shaped hexameric ring complex, which threads the substrate protein through its central axial channel via conserved pore loops. While recent structural investigations have shed light on functional aspects of the machine, the detailed understanding of its mechanisms is still missing. Our study uses molecular dynamics (MD) simulations and machine learning (ML) approaches to elucidate the allosteric network underlying the machine function at atomic resolution and to identify the structural features that drive the conformational changes of the machine. We performed MD simulations of distinct configurations of Hsp104, in the presence or absence of a substrate protein and/or nucleotide, in both pre- and post-hydrolysis hexamer conformations. Conventional MD simulations are limited in capturing functional conformation changes of complex biomolecules, which can be addressed using a powerful statistical framework called Markov State Models (MSM), for exploring long-timescale dynamics from simulations that probe relatively short biological timescales. MSM was estimated by using PyEMMA using biochemically relevant structural and energetic descriptors calculated at the level of secondary structures and ML was used to extract the key features driving state transitions.