Sampling Fibril-like States from an Amyloid-β Monomer Using a Data-Driven Collective Variable

Abstract

Intrinsically disordered proteins (IDPs) and proteins with intrinsically disordered regions (IDRs) make up ∼30% of the human genome, and are implicated in a number of human diseases including cancer and neurodegenerative diseases. The aggregation of IDPs into fibril structures has been linked to cytotoxicity in diseases like Parkinson's and Alzheimer's diseases, the two most common neurodegenerative diseases worldwide. The mechanism of aggregation is still poorly understood, and represents one of the key challenges in deciphering the progression of these diseases. Molecular dynamics (MD) simulations serve as a powerful tool for understanding the structure and dynamics of IDPs, but conventional MD often struggles to sample biologically relevant states of IDPs, due to the wide range of states they can adopt. One common solution is metadynamics, which offers the ability to reconstruct the free energy surface after a simulation is complete. To run metadynamics, a collective variable (CV) must be chosen. A CV is a function of atomic coordinates that can adopt a range of values for different states of a system. Choosing a CV for IDP systems can be very challenging, but recently data-driven approaches allow for automatically creating a CV from a non-linear transformation of atomic coordinates. Here we present simulations of residues 10-35 of the amyloid β-peptide – a key IDP involved in Alzheimer's – which employ metadynamics with a data-driven CV. We obtained a range of fibril-like states, some of which included short helical turns and all of which showed an elongation of the N-terminal segment of the peptide. Additionally, simulations in the presence of a fibril-derived dimer were caught in kinetic traps that were not reversible even under metadynamics sampling. This work contributes to a better understanding of the structures the amyloid-β monomer can adopt and to the aggregation process as a whole.