Protein structural transitions critically transform the network connectivity and viscoelasticity of RNA-binding protein condensates but RNA can prevent it

Abstract

Biomolecular condensates, some of which are liquid-like during health, can age over time becoming gel-like pathological systems. One potential source of loss of liquid-like properties during ageing of RNA-binding protein condensates is the progressive formation of inter-protein β-sheets. To bridge microscopic understanding between accumulation of inter-protein β-sheets over time and the modulation of FUS and hnRNPA1 condensate viscoelasticity, we develop a multiscale simulation approach. Our method integrates atomistic simulations with sequence-dependent coarse-grained modelling of condensates that exhibit accumulation of inter-protein β-sheets over time. We reveal that inter-protein β-sheets notably increase condensate viscosity but does not transform the phase diagrams. Strikingly, the network of molecular connections within condensates is drastically altered, culminating in gelation when the network of strong β-sheets fully percolates. However, high concentrations of RNA decelerate the emergence of inter-protein β-sheets. Our study uncovers molecular and kinetic factors explaining how the accumulation of inter-protein β-sheets can trigger liquid-to-solid transitions in condensates, and suggests a potential mechanism to slow such transitions down.

Fig. 1. Structural transitions leading to inter-peptide β-sheet motifs dramatically increase protein binding and promotes droplet densification over time.

a Atomistic Potential of Mean Force (PMF) dissociation curve of a 6-amino acid segment (PDB code: 6BXX) found in the A-LCD-hnRNPA1 sequence from a β-sheet structure formed by 4 peptides (of the same sequence) as a function of the centre-of-mass distance (COM) using the a99SB-disp force field. PMF simulations are conducted at room conditions and physiological salt concentration. Yellow curve represents the interaction strength among peptides with a well-defined folded structure, kinked β-sheet structure, while the purple curve depicts the interaction strength among the same segments but when they are fully disordered. Statistical uncertainty is depicted by colour bands. A representation of the four peptides, both ordered (yellow) and disordered (purple) including the dissociating peptide in blue, is also included in the inset. b Snapshot of an all-atom PMF simulation in which the structured peptide is pulled from the inter-protein β-sheet motif. The distinct residues within the peptides are highlighted by different colours while water is depicted in blue (O) and red (H) and NaCl ions by green (Na⁺) and yellow (Cl⁻) spheres. c Representation of the dynamical algorithm coupled to the residue-resolution model to introduce disorder-to-order transitions according to the protein local environment. When four LARKS segments meet within a given cut-off distance, LARKS binding is strengthened according to the PMF binding free energy difference computed in Panel a. The bending penalty between residues composing LARKS motifs is also enhanced to account for the higher rigidity of structured β-sheet aggregates^,. d Phase diagram of A-LCD-hnRNPA1 in the T-ρ plane for protein condensates with disorder-to-order transitions and subsequent strengthening of inter-molecular protein binding (black symbols; dynamical ageing model), and for the reference model (HPS-Cation-π, refs. 107, 108) where the interaction strength among LARKS is always considered fully disordered (green symbols). Statistical errors are obtained by bootstrapping results from n = 3 independent simulations. A Direct Coexistence simulation snapshot is included in the inset where different protein replicas are depicted by different colours. e Number of inter-peptide (cross) β-sheet transitions as a function of time found in phase-separated condensates at different temperatures (see legend; temperatures are normalized by the critical temperature of A-LCD-hnRNPA1; T_c). f Time-evolution of the condensate density for different temperatures as indicated in the legend of panel e.

Fig. 2. Protein structural transitions severely impact the viscoelastic behaviour of FUS and A-LCD-hnRNPA1 condensates.

a Phase diagram in the T-ρ plane for full-FUS (circles) and FUS-PLD (squares) sequences before disorder-to-β-sheet transitions take place (green symbols), and after condensates become kinetically arrested (black symbols; dynamical algorithm). Filled symbols represent the coexistence densities obtained via Direct Coexistence simulations, while empty symbols depict the estimated critical points by means of the law of rectilinear diameters and critical exponents. Purple and brown crosses depict coexistence densities for the full-FUS reference model and ageing model respectively using system sizes two times larger (i.e., 96 protein replicas). Temperature has been normalized by the critical temperature of full-FUS (T_c,FUS) for the reference model^,. Statistical errors are obtained by bootstrapping results from n = 3 independent simulations. b Shear stress relaxation modulus G(t) of the A-LCD-hnRNPA1 bulk condensed phase at T = 0.92T_c for the reference model (green curve; HPS-Cation-π model^,), and for protein condensates in which local strengthening of LARKS protein binding due to structural transitions is accounted (black curve; dynamical ageing model). Snapshots illustrating a shear stress relaxation computational experiment over A-LCD-hnRNPA1 condensates are included: Bottom, for a liquid-like condensate (reference model), and Top, for an aged condensate. Structured inter-peptide β-sheet motifs are depicted in red, and intrinsically disordered regions in green. c Shear stress relaxation modulus G(t) of FUS-PLD bulk condensates at two different temperatures as indicated in the legend. Light-coloured circles account for condensates before exhibiting binding strengthening due to structural transitions (reference model), while dark-coloured circles represent G(t) for condensates upon ageing (i.e., when the rate of structural transitions has reached a plateau). d Top: Viscosity (η) as a function of temperature (renormalized by the critical temperature T_c) for A-LCD-hnRNPA1 condensates before the emergence of enhanced binding due to local structural transitions (green symbols; reference model), and after the formation of inter-protein β-sheet fibrils within the condensates (black symbols; dynamical model). Bottom: Protein diffusion coefficient within the bulk condensed phase before (green symbols) and after the formation of inter-peptide β-sheet motifs within the condensates (black symbols).

Fig. 3. Imbalanced protein binding can drive condensate ageing but not droplet reshaping.

Landscape of the protein contact free energy variation upon condensate ageing for bulk FUS-PLD condensates at T/T_c,FUS = 0.785 (a) and A-LCD-hnRNPA1 condensates T = 0.97T_c (b). ΔG/k_BT is obtained from the residue contact probability ratio between aged condensates and liquid-like condensates before ageing. Colour map projections of the free energy landscape in 2-dimensions are also included. c Network connectivity of aged FUS-PLD condensates at T/T_c,FUS = 0.785 (Left) and T/T_c,FUS = 0.861 (Right) computed using a primitive path analysis. d Evolution of droplet sphericity during condensate ageing at T/T_c,FUS = 0.81. The percentage of transitioned fully disorder LARKS into structured inter-peptide β-sheet motifs was ~75% along the trajectory. The horizontal red dashed line represents the average sphericity for a FUS-PLD condensate in a liquid-like state (reference model). Two representative configurations of the condensate surrounded by the surface map (light green) are provided with LARKS residues belonging to inter-peptide β-sheet clusters depicted in dark green and fully disorder residues belonging to distinct protein replicas depicted by a blue-to-grey colour range. Details on the sphericity parameter evaluated through a Solvent Available Surface Area (SASA) analysis as well as of the primitive path analysis and the protein contact free energy calculations are provided in the Supplementary Methods.

Fig. 4. FUS disorder-to-order transitions are hindered by RNA which collectively contributes to blocking protein high-density fluctuations.

a Temperature–density phase diagram for pure FUS (red symbols) and two polyU/FUS mixtures with different RNA concentrations as indicated in the legend. Filled circles represent the coexistence densities before structural transitions take place (reference model), and square symbols depict densities after ageing occurs (ageing model, i.e., >70% of LARKS within the condensates are engaged in inter-protein β-sheet motifs). Empty symbols indicate the estimated critical points obtained through the law of rectilinear diameters and critical exponents. Please note that temperatures are normalized by the critical T of pure FUS (T_c,FUS). Statistical errors are obtained by bootstrapping results from n = 3 independent simulations. b Critical temperature of FUS/polyU mixtures as a function of the polyU/FUS mass ratio evaluated for the reference model (green circles) and for the dynamical ageing model (black circles). Symbols above the horizontal dotted line imply LLPS enhancement while those below indicate phase-separation hindrance. The statistical uncertainty shown in Panel a also applies for b. c Time-evolution of inter-protein β-sheet transitions (in percentage) within bulk condensates at different polyU/FUS mass ratios and at T/T_c,FUS = 0.96. Dashed lines account for second-order reaction fits to our data employed to estimate the kinetic constant of inter-protein β-sheet formation at distinct RNA concentrations. Inset: Structural transition kinetic constants as a function of polyU/FUS mass ratio within bulk phase-separated droplets. The brown cross depicts the computed kinetic constant in presence of inert polymers of the same length and concentration than polyU strands (black symbols). d Landscape of the average protein contact free energy variation upon condensate ageing of FUS droplets in absence of RNA at T/T_c,FUS = 0.96. ΔG/k_BT is computed from the residue contact probability ratio between aged droplets (dynamical model) and liquid-like droplets (reference model). e Free energy inter-molecular variation computed from the molecular contact probability of pure FUS aged condensates and polyU/FUS (at 0.24 polyU/FUS mass ratio) aged condensates at T = 0.96T_c,FUS and after an ageing time interval of ~1 μs for both systems.

Fig. 5. A-LCD-hnRNPA1 droplet ageing driven by disorder-to-order transitions is decelerated by inclusion of high polyU RNA concentration.

a Time-evolution of droplet density (Top) and percentage of inter-protein β-sheet transitions within the condensates (Bottom) measured at different polyU/A-LCD-hnRNPA1 mass ratios and at T = 0.97T_c (where T_c refers to the critical temperature of the pure protein condensate). Dashed lines in the bottom panel depict the second-order reaction fits employed to evaluate the kinetic constant (k_dir) of inter-protein β-sheet formation (see details on the Supplementary Methods). b Estimated kinetic constants from a second-order reaction analysis to the number of inter-protein β-sheet transitions over time for different polyU/A-LCD-hnRNPA1 mass ratios at T = 0.97T_c using two different residue-resolution models: HPS-Cation-π (black circles,^–) and Mpipi (red circles,). Black empty diamonds account for test control simulations in presence of inert polymers of the same length and concentration than polyU strands in the protein mixtures using the HPS-Cation-π model. Symbol sizes account for the estimated uncertainty while dotted and continuous lines are included as a guide for the eye. c Landscape of the protein and polyU (wide band) contacts free energy variation upon condensate ageing measured in polyU/A-LCD-hnRNPA1 phase-separated bulk droplets at T = 0.97T_c and 0.18 polyU/A-LCD-hnRNPA1 mass ratio. The same timescale for observing condensate ageing in pure A-LCD-hnRNPA1 droplets shown in Fig. 3b was explored here. d Shear stress relaxation modulus of polyU/A-LCD-hnRNPA1 aged condensates at T = 0.97T_c for different polyU/protein mass ratios. The time interval to observe structural transitions before quenching them through our dynamical algorithm and compute G(t) was the same for all concentrations, ~1 μs.