Accurate model of liquid-liquid phase behaviour of intrinsically-disordered proteins from optimization of single-chain properties

Abstract

Many intrinsically disordered proteins (IDPs) may undergo liquid-liquid phase separation (LLPS) and participate in the formation of membraneless organelles in the cell, thereby contributing to the regulation and compartmentalisation of intracellular biochemical reactions. The phase behaviour of IDPs is sequence-dependent, and its investigation through molecular simulations requires protein models that combine computational efficiency with an accurate description of intra- and intermolecular interactions. We developed a general coarse-grained model of IDPs, with residue-level detail, based on an extensive set of experimental data on single-chain properties. Ensemble-averaged experimental observables are predicted from molecular simulations, and a data-driven parameter-learning procedure is used to identify the residue-specific model parameters that minimize the discrepancy between predictions and experiments. The model accurately reproduces the experimentally observed conformational propensities of a set of IDPs. Through two-body as well as large-scale molecular simulations, we show that the optimization of the intramolecular interactions results in improved predictions of protein self-association and LLPS.

Disease‐linked TDP‐43 hyperphosphorylation suppresses TDP‐43 condensation and aggregation

“…Two amino acids were deemed in contact in the simulations with the HPS model when their inter‐bead distance was < 2 1/6 σ ij where σ ij = ½(σ i + σ j ) is the average of bead diameter of the respective amino acids i and j . The concentrations c dilute and c dense of dilute and dense phases, respectively, were determined by adapting the workflow of Tesei et al ( https://github.com/KULL‐Centre/papers/tree/main/2021/CG‐IDPs‐Tesei‐et‐al ) from the simulations of the HPS model (preprint: Tesei et al , 2021 ). The excess free energy of transfer from the dilute to the dense phase was then computed as Δ G trans = − RT ln ( c dense / c dilute ), where R is the gas constant and T is the absolute temperature.…”

Physics-driven coarse-grained model for biomolecular phase separation with near-quantitative accuracy

“…To validate our model parameters, we first compare the Mpipi model with seven other residue-level coarse-grained models for LLPS, namely the KH (Kim-Hummer) [28], HPS-KR (hydrophobicity scale) [28], FB-HPS [26] and HPS-Urry [29] models, as well as the HPS model with augmented cationπ interactions [schemes (i) and (ii)] [40]. We also include analyses for TSCL-M2, the M2 parameter set of Tesei et al [27] proposed while our work was under review. Das et al [40] recently provided a thorough comparison of the KH, HPS-KR, HPS+cation-π(i) and HPS+cation-π(ii) models.…”

“…Recently, Tesei and coworkers [27] used experimental data to reparameterise the hydrophobicity scale of HPS-KR via a Bayesian parameter-learning procedure. The M2 parameter set predicted well both single-molecule and collective behaviours of the tested IDPs; we have therefore included benchmarks for this parameter set in our work.…”

“…Because LLPS is a collective phenomenon, coarse-graining is essential to reduce the system dimensionality while retaining essential physicochemical information and allowing sufficient sampling of phase space in computationally tractable time scales. There are numerous possible approaches for parameterising biomolecular coarse-grained models [21] , from 'bottom-up' strategies that rely on higher-resolution models [9,10,[22][23][24], to 'knowledge-driven' approaches that aim to reproduce experimental properties using a data-based parameterisation [25][26][27], to 'top-down' strategies that account for emergent behaviour by approximating fundamental physical forces [28,29], to combinations of these [30,31]. Coarsegrained models can also be broadly classed as 'system-specific', bottom-up parameterisations focussing on finding an optimum representation for a particular system using fine-grained simulations as a reference, often derived in a systematic way using for instance iterative Boltzmann inversion [32,33] or force matching [34,35], and 'transferable', either bottom-up or top-down parameterisations, aiming to achieve a generally applicable potential.…”

Improving Martini 3 for disordered and multidomain proteins

“…Proteins were coarse-grained using the Martinize2 python script, placed in a dodecahedral box using Gromacs and solvated using the Insane python script ( Wassenaar et al, 2015 ). Initial box size was chosen by using starting structures from simulations in Tesei et al (2021b) corresponding to the 95th percentile of R g -distributions and using Gromacs editconf with the flag -d 4 . 0 .…”