Biomolecular condensates form spatially inhomogeneous network fluids

Abstract

The functions of biomolecular condensates are thought to be influenced by their material properties, and these are in turn determined by the multiscale structural features within condensates. However, structural characterizations of condensates are challenging, and hence rarely reported. Here, we deploy a combination of small angle neutron scattering, fluorescence recovery after photobleaching, and bespoke coarse-grained molecular dynamics simulations to provide structural descriptions of model condensates that mimic nucleolar granular components (GCs). We show that facsimiles of GCs are network fluids featuring spatial inhomogeneities across hierarchies of length scales that reflect the contributions of distinct protein and peptide domains. The network-like inhomogeneous organization is characterized by a coexistence of liquid- and gas-like macromolecular densities that engenders bimodality of internal molecular dynamics. These insights, extracted from a combination of approaches, suggest that condensates formed by multivalent proteins share features with network fluids formed by associative systems such as patchy or hairy colloids.

Fig. 1:. Complexation between acidic regions within N130 and R-motifs of rpL5 is required for condensation.

(a) Schematic representation of N130 including the different acidic regions. The amino acid sequence of N130 is also shown. The three acidic regions A0 (see text), A1, and A2 span residues 4–18, 35–44, and 120–133, respectively. The underlined N-terminal residues GSH are cloning artefacts. On the right, we show the overall structure of the hairy colloid generated by superposition of 50 distinct conformations from all-atom simulations. The OD (PDB ID 4N8M), in gray, was modeled as a rigid molecule in the atomistic simulations. (b) The sequence of rpL5. The panel on the right shows a superposition of 50 different conformations extracted from atomistic simulations based on the ABSINTH model. (c) Confocal microscopy images of phase separation of 100 μM N130 upon titrating the concentration of rpL5 in buffer and 150 mM NaCl. N130 is labeled with AlexaFluor488. (d) Two-component phase boundary for N130+rpL5, showing the result of concentration titrations. (e) SANS curve showing the intensity I(q) plotted against q, the scattering vector, for condensates formed by a N130 (200 μM): rpL5 solution at 1:3 stoichiometry. A fitting analysis, described in the Methods, gives peaks at scattering vectors corresponding to 55 Å (left arrow), 77 Å (middle arrow), and 119 Å (right arrow). The SANS curve for N130 alone is shown for comparison. In the interest of clarity, this curve is shifted upwards by 0.03 cm⁻¹.

Fig. 2:. Coarse-grained simulations of N130+rpL5 condensates highlight the importance of an N-terminal acidic region (A0) within N130.

(a) Coarse-graining procedure for N130 and rpL5 systems. To generate sequence- and system-specific CG models for N130 and rpL5 peptides, we start with atomistic simulations of the individual molecules, using CAMPARI (http://campari.sourceforge.net) and ABSINTH . We then prescribe a CG model for the system. Here, the residues of the OD are collectively modeled as one large bead depicted here in gray. Next, for regions outside the OD, the residues are single beads, and the bead types are organized into three groups: 1: E,K,R,D; 2: V,F,M,I,L,Y,Q,N,W,H; and 3: A,P,G,S,T,C, respectively. To determine the optimal interaction parameters for the CG model, we use a machine-learning-based approach using CAMELOT. (b) Bead-to-bead contact maps from coarse-grained simulations of dense-phases comprising a 1:3 ratio of N130 and rpL5. The A1 and A2 regions interact favorably with rpL5 peptides. The contact maps reveal new, hitherto unappreciated interactions involving a region we refer to as A0. (c) Confocal microscopy images of phase separation at a fixed concentration of N130^+A2 upon titrating with rpL5. The A0 tract in the wild type is replaced with the reversed sequence of the A2 tract. (d) Two-component phase boundary for N130^+A2+rpL5, showing the result of concentration titrations. (e) SANS curves show that a broad peak forms at lower q ranges, with the shift implying a shift toward extended conformations in rpL5-induced droplets of the +A2 mutant compared to wild type. SANS data were collected with 200 μM N130 and 600 μM rpL5. For clarity, the curve for N130^+A2+rpL5 is shifted upwards by 0.03 cm^-1. (f) FRAP curves for the condensates containing N130 and N130^+A2, along with the measured recovery times. FRAP curves were collected at 100 μM N130 and 400 μM rpL5 with twelve replicates. Error bars represent the standard deviation of the mean.

Fig. 3:. Radial distribution functions affirm the network fluid structure of N130+rpL5 condensates.

(a) Radial distribution function showing the correlations between particles in a Lennard-Jones fluid along with the volume integral n(r) quantifying the mean coordination number as a function of interparticle separation r. On average, approximately twelve Lennard-Jones particles are found within the first coordination shell. (b) $g(r)$ in black and n(r) in red quantifying the correlations between the ODs of N130 in the simulated N130+rpL5 system. There are local maxima at 53 Å, 95 Å, and 140 Å. These are in accord with the peaks in SANS profiles. On average, approximately four oligomerization domains are found within the first coordination shell. This number is consistent with a network fluid, such as water. (c) $g(r)$ quantifying the correlations between negative charges in the acidic regions of N130 and the positive charges in rpL5. $W(r)=-RT\text{ln}g(r)$ quantifies the free energy change associated with bringing a pair of sites to a distance $r$ from one another. From the site-site $g(r)$, it follows that the free energy change is between −0.5 and 0 kcal/mol over a distance range of 50 Å. These interactions are strongest between A2 and rpL5 and weakest between the A1 tract and rpL5. Averages were calculated over 3 replicates for the Lennard-Jones systems and over 5 replicates for the N130.

Fig. 4:. Interactions between acidic tracts and rpL5 are modular and independent of one another.

(a) g_OD-OD(r) computed after neutralizing the charges in each of the acidic tracts indicated in the legend. (b) g_A0-rpL5(r) quantifies the pair correlations between acidic residues in A0 and basic residues in rpL5. Results are shown for the WT (black), when acidic residues are neutralized in A0 (blue), A1 (green), and A2 (magenta). Neutralizing charges within A0 weakens the interactions between A0 and rpL5. However, neutralizing the charges within A1 and A2 does not significantly affect the interactions between A0 and rpL5. (c) g_A0-rpL5(r) quantifies the pair correlations between acidic residues in A1 and basic residues in rpL5. Results are shown for the WT (black), when acidic residues are neutralized in A0 (blue), A1 (green), and A2 (magenta). Neutralizing charges within A1 weakens the interactions between A1 and rpL5 (green curve). However, neutralizing the charges within A0 and A2 does not significantly affect the interactions between A1 and rpL5. (d) g_A2-rpL5(r) quantifies the pair correlations between acidic residues in A2 and basic residues in rpL5. Results are shown for the WT (black), when acidic residues are neutralized in A0 (blue), A1 (green), and A2 (magenta). Neutralizing charges within A2 weakens the interactions between A2 and rpL5 (magenta curve). However, neutralizing the charges within A0 and A1 does not significantly affect the interactions between A2 and rpL5. In all cases, averages were calculated over 5 replicates.

Fig. 5:. Linkers between R-motifs in rpL5 and changes to the charge of A2 influence the local structure of the network fluid.

(a) SANS curves for N130 with each of the synthetic rpL5 peptide variants. The data for the wild-type N130+rpL5 system are replotted from Fig. 1e. (b) g_OD-OD(r) between the ODs for dense phases of each synthetic peptide + N130 mixture. The disordered linker weakens nearest neighbor interactions whilst preserving the strengths of and shifting the preferred positions of next-nearest-neighbor correlations. Averages were calculated over five replicates. The sequences of the linkers are as follows: 10L: GSRRRRGSGSYYGSGSRKRLV; 16L: GSRRRRGSGSGSGYYSGSGSGSRKRLV; and 20L: GSRRRRGSGSGSGSGYYSGSGSGSGSRKRLV.

Fig. 6:. Flowchart describing the graph-theoretic analysis of coarse-grained simulations of dense-phases of N130 and rpL5.

We start by selecting a set of residues of interest. As an illustrative example, we pick the acidic residues in N130 and the basic residues in rpL5. To determine an edge, we compute the $g(r)$ between sets of beads where the first minimum serves as the distance cutoff for bead adjacency. Given this cutoff, we construct the bead-to-bead adjacency matrices. The system includes the multidomain N130, and performing appropriate block-summations of the bead-to-bead adjacency matrix generates the molecular adjacency matrix. The latter matrix is analyzed using standard graph-theoretic analyses. As an example, a random embedding is shown where the size of a node corresponds to the degree centrality of that node.

Fig. 7:. Each acidic region of N130 in the N130+rpL5 dense phase imparts a different network structure onto the system.

(a) Degree distributions, $P(k)$, for pure Lennard-Jones (LJ) systems. Because the density of the vapor ($\mathrm{L}\mathrm{J}$) is low, the most likely degree is zero. For the liquid $\rho =0.01,T=1.000)$, the degree distributions shift towards higher values with a mean around twelve, where the widths of the distributions correspond to the inherent variation in the number of bonds particles can make in the locally spatially inhomogeneous environment of a liquid. For the solid $(\rho =1.5,T=0.758)$, the degree distribution peaks very sharply at twelve. (b) Degree distributions, $P(k)$, for the complementary charge interactions between the different acidic tracts of N130 and the rpL5 peptides. Unlike the graphs in panel (a), the distributions display bimodality, which is an indication of a bipartite graph. Consistent with the radial distribution functions in Fig 3, we see that A2 has the largest degrees, followed by A0 and A1. In all cases, averages were calculated across five replicates.

Fig. 8:. Motions within dense phases of N130+rpL5 show bimodality.

(a) MSD of the OD plotted against the lag time shows that N130 has super-diffusive and sub-diffusive regimes. Here, the abscissa is a unitless parameter $t^{\mathrm{\prime }}=t/t_D$ where $t_D$ is the timescale over which the motion of the OD is purely diffusive. The red region in the panel indicates the timescales that fit best to purely diffusive motion. There is a timescale below $t_D$ where the motion is super-diffusive and a timescale above $t_D$ where the motion is sub-diffusive, with the regimes and corresponding exponents indicated in the panel. (b) Histograms of the exponents calculated for the mean square displacements of individual ODs. The bimodal distribution reflects the presence of super-diffusive and sub-diffusive regimes. (c) MSDs of the OD and charged residues in each of the acidic regions and the rpL5 peptides. In contrast to the OD, the acidic regions and the peptides show only sub-diffusive motions on all timescales. In all cases, averages were calculated over five replicates.