Generating Ensembles of Dynamic Misfolding Proteins

The early stages of protein misfolding and aggregation involve disordered and partially folded protein conformers that contain a high degree of dynamic disorder. These dynamic species may undergo large-scale intra-molecular motions of intrinsically disordered protein (IDP) precursors, or flexible, low affinity inter-molecular binding in oligomeric assemblies. In both cases, generating atomic level visualization of the interconverting species that captures the conformations explored and their physico-chemical properties remains hugely challenging. How specific sub-ensembles of conformers that are on-pathway to aggregation into amyloid can be identified from their aggregation-resilient counterparts within these large heterogenous pools of rapidly moving molecules represents an additional level of complexity. Here, we describe current experimental and computational approaches designed to capture the dynamic nature of the early stages of protein misfolding and aggregation, and discuss potential challenges in describing these species because of the ensemble averaging of experimental restraints that arise from motions on the millisecond timescale. We give a perspective of how machine learning methods can be used to extract aggregation-relevant sub-ensembles and provide two examples of such an approach in which specific interactions of defined species within the dynamic ensembles of alpha-synuclein (alpha Syn) and beta(2)-microgloblulin (beta(2)m) can be captured and investigated.

Automated summary

The early stages of protein misfolding and aggregation involve dynamic disorder and present challenges in capturing atomic level visualization. Identifying specific conformers on-pathway to aggregation within large pools of rapidly moving molecules is complex. This article describes experimental and computational approaches for studying the dynamic nature of protein misfolding and aggregation, and discusses challenges in describing these species due to ensemble averaging of experimental restraints. The potential of machine learning methods in extracting aggregation-relevant sub-ensembles is also explored.