Multiscale coarse-grained models for biological phase separation: Development and applications

Abstract

Many proteins undergo liquid–liquid phase separation (LLPS) in order to generate membraneless organelles, by which the cell can organise its biomolecules and bioprocesses dynamically in space and time. Many of the experimental and theoretical methodologies used to study the formation of these condensates still struggles to capture the relation between microscopic, residue–level details of a protein with its phase behaviour in the bulk. In this thesis, we present three multiscale coarse-grained models with amino acid resolution aimed at studying phase separation of proteins. The Mpipi model balances the dominant role of π–π and hybrid cation–π/π–π interactions with the rest of the interactions, while keeping R–based interactions stronger than K–based ones. The parameterisation is based on atomistic PMF calculations and bioinformatics data on π-based contacts. The Mpipi Recharged model is a finer and more optimised version of Mpipi, that appropriately balances the electrostatic interactions on a pair-by-pair basis, since which all-atom simulations prove the asymmetry between samely-charged and oppositely-charged residues. Both Mpipi and Mpipi recharged are capable of reproducing ensemble averaged experimental observables with high accuracy, from single-molecule properties to phase diagrams of an extensive set of proteins (i.e. hnRNPA1, FUS, Laf 1, DDX4) and their corresponding mutations. Lastly, we also investigated the role of Mg2+ions in regulating LLPS of intranuclear proteins. Atomistic-resolution simulations proved that minimal quantities of Mg2+ions present in the media can significantly alter the phase behaviour of proteins, especially ones with a high number of charged residues. The MagPi model arises from this observations and bioinformatics analysis of the proteome, and can reproduce the phase behaviour of a set of intranuclear proteins (i.e. MED1 IDR, BRD4 IDR, Nanog CTD, and DDX4 and DDX3 variants) qualitatively. Overall, our multiscale modelling approach shows great potential at bridging the gap between atom-level observables, to single-molecule behaviour, to macroscopic phase transitions, as well as its ability to extend the range of the simulations to different solvent conditions or surroundings. Therefore, the work presented in this thesis poses a significant step towards the unification of experiments, computer simulations and real biological LLPS phenomena.