Sequence-specific interactions determine viscoelasticity and aging dynamics of protein condensates

Abstract

Biomolecular condensates are viscoelastic materials. Here, we investigate the determinants of sequence-encoded and age-dependent viscoelasticity of condensates formed by the prion-like low-complexity domain of the protein hnRNP A1 and its designed variants. We find that the dominantly viscous forms of the condensates are metastable Maxwell fluids. A Rouse-Zimm model that accounts for the network-like organization of proteins within condensates reproduces the measured viscoelastic moduli. We show that the strengths of aromatic inter-sticker interactions determine sequence-specific amplitudes of elastic and viscous moduli, and the timescales over which elastic properties dominate. These condensates undergo physical ageing on sequence-specific timescales. This is driven by mutations to spacer residues that weaken the metastability of dominantly viscous phases. The ageing of condensates is accompanied by disorder-to-order transitions, leading to the formation of non-fibrillar, beta-sheet-containing, semi-crystalline, elastic, Kelvin-Voigt solids. Our results suggest that sequence grammars, which refer to amino acid identities of stickers versus spacers in prion-like low-complexity domains, have evolved to afford control over metastabilities of dominantly viscous fluid phases of condensates. This selection is likely to render barriers for conversion from metastable fluids to globally stable solids insurmountable on functionally relevant timescales.

Extended Data Fig. 1.. Protein network structure governs condensate viscoelasticity.

(a) Representative positional trajectory, tracked along the horizontal axis, of a bead thermally fluctuating within the optical trap inside the condensate. (b) The normalized autocorrelation function ($A\left(\tau \right)$), where $\tau$ is the lag time, of positional fluctuations of a bead, confined by the optical trap, inside a condensate. The correlation functions measured along the horizontal and vertical axes are shown. (c) Ensemble-averaged mean square displacement (MSD) of 200 nm beads in condensates formed by WT^+NLS. The solid line is a fit to the equation $MSD\left(\tau \right)=4D{\tau }^{\alpha }+N$ and reveals that at longer timescales (>10 ms) the beads are diffusive. However, at shorter timescales (<10 ms), the beads are sub-diffusive, indicating that the condensates are dominated by an elastic response. The MSD shown in black represents the average noise floor of the measurements. (d) Residuals between computed and measured storage moduli ($G^{\prime }$) for WT^+NLS condensates. These residuals are shown for the single-chain model (red) and collective model (magenta). The residuals were computed as the absolute values of differences between the logarithms of the computed and measured moduli. Values of zero imply exact matches, values of one imply a discrepancy of an order of magnitude, and so on. At low frequencies, the sharp increase in residuals for the collective model is due to increased noise in the experimentally derived storage moduli. (e) Residuals between computed and measured loss moduli ($G^{″}$) using the single chain (blue) or collective chain models (black). The residuals were computed as in panel (d). (f) Comparison of computed and measured dynamical moduli for WT^+NLS condensates. The computations use the single-chain model with weighted Zimm matrices. The computed crossover frequency was rescaled to match the experimental value. No other adjustments or fitting of the data was performed. The inadequacy of the single-chain model is made clear by the fact that this model overestimates the storage modulus by almost two orders of magnitude. Similar overestimates were found for other systems, and they stand in contrast to the results obtained using the collective model shown in Fig. 1b. The error band for the computed data represents the standard deviation from 3 replicates. (g) Comparison of computed and measured dynamical moduli for WT^+NLS condensates with computations that use the single-chain model with unweighted Zimm matrices. The computed dynamical moduli were rescaled as in panel (f). In (f)-(g), the standard deviation from 3 replicates is smaller than the line width used to plot the computed moduli. The error bars for measured data in (f)-(g) represent standard deviations about the mean (± S.D.M.) that were computed using data for 20 separate beads over 3 independent experiments. (h) A schematic diagram of the constructs deployed in the various measurements in this work is summarized. Grey represents non-aromatic spacer residues, whereas colored stripes represent the locations of the indicated aromatic residues.

Extended Data Fig. 2.. Viscoelastic moduli of condensates formed by A1-LCD mutants.

(a) Computed dynamical moduli (elastic ($G^{\prime }$) and viscous ($G^{″}$)), calculated using the collective model, for condensates formed by allF. The error band for the computed data represents the standard deviation from 3 replicates. (b) Measured $G^{\prime }$ and $G^{″}$ for condensates formed by the YtoW variant. The error bars for measured data represent standard deviations about the mean (± S.D.M.) that were computed using data for 20 separate beads over 3 independent experiments. (c, d) Measured $G^{\prime }$ and $G^{″}$ for condensates formed by W- and FtoW variants, respectively. The error bars for measured data in (f)-(g) represent standard deviations about the mean (± S.D.M.) that were computed using data for 20 separate beads over 3 independent experiments.

Extended Data Fig. 3.. Temperature-dependent viscosities report on the flow activation energies of the condensates.

(a) Schematic for the temperature-dependent video particle tracking (VPT) measurements. The 200 nm beads (yellow circles) are embedded inside the condensate (blue region) sitting on a substrate (black rectangle). Diffusion of multiple 200 nm polystyrene beads is recorded using 100x objective (grey box) at distinct temperatures (depicted with blue and red symbols for cold and hot). (b) Room-temperature bulk dynamical viscosity of condensates formed by different sequence variants as measured by VPT compared to passive microrheology with optical tweezers (pMOT). Data for viscosities from VPT were collected from at least six trials (n=7 for allF, n=9 for allY, n=6 for WT^+NLS, n=7 for W-, n=13 for YtoW, n=6 for FtoW, and n=6 for allW) featuring at least ~50 tracked particles per trial. Error bars represent standard deviations about the mean (± S.D.M.). Although the viscosity values obtained from these two independent experimental measurements are in good agreement with one another, we note that the terminal zero-shear viscosity, obtained from the slope of the frequency-dependent loss modulus in pMOT measurements, is an asymptotic property, which means that this quantity is difficult to estimate for viscoelastic materials when the elastic and viscous moduli become similar, even below the crossover frequency. The low-frequency limit of the measurement (about 0.1 Hz which corresponds to ~10 s) may not be low enough to estimate the zero-shear viscosity, especially for condensates with high viscoelasticity, such as for variants with Trp residues as stickers. This explains the minor discrepancy in the viscosity values obtained from VPT and pMOT measurements. (c) Representative plots of the ensemble-averaged mean squared displacement (MSD) of 200 nm polystyrene beads within condensates formed by allF. Measurements were performed using VPT-based microrheology at different temperatures as set by a custom-made thermal stage attached to the microscope. (d) Measured temperature-dependent viscosities for the condensates formed by different sequence variants of A1-LCD. The solid lines connect the data points but do not represent fits. Temperature-dependent viscosities were collected from at least five trials (n≥5) for each temperature, featuring at least ~50 tracked particles per trial. Error bars represent standard deviations about the mean (± S.D.M.). (e) Arrhenius plots of measured viscosities for condensates formed by allW, WT^+NLS, allY, and allF. (f) Arrhenius plots using temperature-dependent viscosities from simulations analyzed using the collective model. Data are shown for condensates of allF, WT^+NLS, and allY. (g) Flow activation energies are estimated from computations. These energies are shown in units of RT, where temperature, T = 25 °C, and R is the ideal gas constant in units of kcal / mol-K. The variant names were taken from the work of Bremer et al., (ref. 32) and Farag et al., (ref. 8). The diamonds represent the activation energy, and the whiskers represent the standard deviation in the flow activation energy calculated from the fit, pink diamonds are the hnRNP A1 variants, black diamonds highlight the homopolymer and FUS-LCD values.

Extended Data Fig. 4.. Interplay between sticker strength and spacer solvation determines the timescale of the dynamical arrest of A1-LCD condensates.

(a) Schematic denoting the position of glycine to serine substitution in the WT^+NLS and allY sequence backgrounds. Grey shading represents unmutated or tyrosine residues that were not mutated to serine (b) Trajectories of a 200 nm bead within condensates of WT^30GtoS_. Two different trajectories are shown, one at the time of observation $(t_{\mathrm{obs}})\approx 5\text{min}$ (blue), and the other for $t_{\mathrm{obs}}\approx 7\text{min}$ (red). These data show that WT^30GtoS condensates undergo rapid dynamical arrest within a short time. (c) Ensemble-averaged mean square displacements (MSD) of 200 nm beads diffusing within allY^20GtoS condensates tracked at a high acquisition rate of 500 fps. The solid blue line is a fit to the equation $MSD\left(\tau \right)=4D{\tau }^{\alpha }+N$. Also plotted is the average noise floor (black). These data show that allY^20GtoS condensates are viscoelastic Maxwell fluids at small $t_{\mathrm{obs}}$, with a terminal behavior of a viscous fluid (α = 1). (d) Ensemble-averaged MSD of 200 nm beads (tracked at an acquisition rate of 10 fps) in condensates formed by allY^20GtoS. The MSDs were measured at different values of $t_{\mathrm{obs}}$. Solid lines are the fits to the equation $MSD\left(\tau \right)=4D{\tau }^{\alpha }+N$. Also indicated are the values of the exponents obtained from the fits, which show a clear deviation from the linear behavior for the aged condensates, and the onset of sub-diffusive motions.

Extended Data Fig. 5.. A1-LCD condensates undergo an age-dependent transition from Maxwell fluid to Kelvin-Voigt solid.

Schematic representation of the linear viscoelastic response of (a) an elastic solid, (b) a viscous liquid, and (c) a Kelvin-Voigt viscoelastic solid. (d-f) Storage ($G^{\prime }$) and loss ($G^{″}$) moduli of allY^30GtoS condensates at the time of observation $(t_{\mathrm{obs}})\approx 5\text{min}$ as calculated from the measured mean squared displacements (MSDs) of embedded 200 nm probe particles (Fig. 3f in the main text). The calculations were performed following the Evans method (d) and the Mason method (e, see Supplementary Methods). A comparison between the results obtained from the two methods is shown in (f). (g-i) Same as (d-f) but for allY^30GtoS condensates at $t_{\mathrm{obs}}\approx 3\text{hrs}$.

Extended Data Fig. 6.. Dynamically arrested A1-LCD condensates are dominantly elastic.

(a) Schematic showing the expected outcomes of a particle embedded within a viscoelastic Kelvin-Voigt solid (top), viscoelastic Maxwell fluid (middle), and viscous liquid (bottom) upon creep deformation by an optical trap. $k_{Condensate}$ and $k_{Trap}$ represent the spring constants for the condensate and the trap, respectively, and ${\eta }_{Condensate}$ is the viscosity of the condensate. As the optical trap exerts a force to drag a particle embedded within the material, the dashpot element applies a dissipative drag force while the spring element applies a restoring force. Except for the viscoelastic solid, the creep deformation is irreversible (the final bead position is distinct from the initial bead position X₀ indicated by the red vertical line). In the case of a viscoelastic solid, the bead recoils back to its initial position due to the stored energy within the elastic element. In both viscoelastic fluids and viscous liquids, the deformation energy is eventually dissipated through the dashpot or the viscous element, giving rise to a terminally viscous behavior. (b) Bead position, with respect to the initial position, as a function of time in response to the programmed trap movement in allY condensates at the time of observation $(t_{\mathrm{obs}})\approx 5\text{min}$ at increasing power levels of the trapping laser. (c) Same as (b) but for the allY condensates at $t_{\mathrm{obs}}\approx 24\text{hrs}$. (d) Bead position, with respect to the initial position, as a function of time in response to the programmed trap movement in allY^20GtoS condensates $t_{\mathrm{obs}}\approx 5\text{min}$ at increasing power levels of the trapping laser. (e) Same as (d) but for the allY^20GtoS condensates at $t_{\mathrm{obs}}\approx 24\text{hrs}$. (f) The force exerted on the bead by the allY condensate network in (b). (g) The force exerted on the bead by the allY condensate network in (c). (h) The force exerted on the bead by the allY^20GtoS condensate network in (d). (i) The force exerted on the bead by the allY^20GtoS condensate network in (e). The percentage values in (b-i) refer to the absolute laser powers used for each measurement.

Extended Data Fig. 7.. Disorder-to-order transitions accompany the dynamical arrest of A-LCD condensates.

(a) Peaks in FTIR spectra correspond to the presence of specific types of secondary structures. (b) FTIR spectra for condensates formed by WT^+NLS. The spectra are shown for time of observation $(t_{\mathrm{obs}})\approx 5\text{min}$ and $t_{\mathrm{obs}}\approx 3\text{hrs}$. (c-g and i) Inverse second derivatives (-d²A/dv²) (black line) of the measured FTIR spectra and contributions from different members of the basis set seen in (a) (light blue lines). A = absorbance; v = wave number. The dashed red line is the sum of these underlying contributions. The dark blue shading quantifies the area under the peak that corresponds to aggregated beta sheets and the estimated contribution is indicated in %. (c) Data are shown here for WT^+NLS condensates for $t_{\mathrm{obs}}\approx 5\text{min}$. (d) The inverse second derivative of FTIR spectra and contributions from different members of the basis set to data obtained for allY condensates at $t_{\mathrm{obs}}\approx 5\text{min}$. (e) – (g) Second derivative of FTIR spectra obtained at different values of $t_{\mathrm{obs}}$ for condensates formed by allY^30GtoS. (h) FTIR spectra for amyloid fibers of insulin collected at $t_{\mathrm{obs}}\approx 24\text{hrs}$. (i) Second derivative of the measured FTIR spectra for insulin fibers and contributions from different members of the basis set. (j) Stacked plots of secondary structure contents for allY^30GtoS condensates plotted as a function of time, where the abscissa refers to different values of $t_{\mathrm{obs}}$.

Fig. 1:. A1-LCD condensates are viscoelastic Maxwell fluids with sequence-specific storage and loss moduli.

(a) Setup for passive microrheology with optical tweezers (pMOT) measurements. The microscopy image shows an optically trapped 1 μm polystyrene bead within a condensate formed by WT^+NLS A1-LCD. In the schematic, the condensate is shown in blue, the optical trap as a red, hourglass-shaped object, the bead in white, and the slide surface in black. (b) Dynamical moduli for WT^+NLS condensates. The figure shows comparisons between measured and computed moduli, where computations use the collective model. $G’$ and $G”$ are the frequency-dependent storage and loss moduli, respectively. (c) Dynamical moduli computed from simulations using the collective model for WT^+NLS condensates. This highlights the frequency-dependent scaling of storage and loss moduli above ($G^{\prime }\to G_{\mathrm{\infty }}$ and $G^{\prime \prime }\sim {\omega }^{-1}$) and below ($G^{\prime }\sim {\omega }^2$ and $G^{\prime \prime }\sim \omega$) the crossover frequency, which is annotated by the blue dashed line. $G^{\prime }=G^{\prime \prime }$ at the crossover frequency. (d-f) Measured dynamical moduli of condensates formed by three sequence variants: (d) allF, (e) allY, and (f) allW. The computed moduli for allF are shown in Extended Data Fig. 2a. In panel (e) we see that the collective model provides an accurate accounting of the measured moduli. We lack a coherent LaSSI model for coarse-grained lattice simulations of Trp-containing sequences (see Supplementary Note 3). For panels (b), and (d)-(f), the error bars for measured data represent standard deviations about the mean that was computed using data for 20 condensates. For panels (b) and (e), the error bands for computed data are standard deviations across 3 replicates. (g) Loss tangent plotted against the frequency for condensates formed by seven variants. For (g-i), the yellow and grey regions in the graph represent the dominantly viscous or elastic regimes, respectively. (h) A diagram-of-states based on measured moduli at 10 Hz. A reference line for $G^{\prime }=G^{\prime \prime }$ is shown. The range of the color bar is $G^{\prime }=0.02\mathrm{P}\mathrm{a}$, $G^{\prime \prime }=50\mathrm{P}\mathrm{a}$ (viscous) to $G^{\prime }=50\mathrm{P}\mathrm{a}$, $G^{\prime \prime }=0.8\mathrm{P}\mathrm{a}$ (elastic). (i) Delineation of the dominantly viscous versus elastic regimes in terms of relaxation times, computed as the inverse of the crossover frequency, for condensates formed by each of the seven variants. For panels (b, d, e, f, h, i), the experimentally measured data represent mean values from data for 20 separate beads over 3 independent experiments; the error bars represent ± S.D.M.

Fig. 2:. Internal viscosity of condensates and the driving forces for phase separation are inversely correlated.

(a) Plot of computed, temperature- and variant-dependent viscosity and saturation concentration $\left(c_{\mathrm{s}\mathrm{a}\mathrm{t}}\right)$ values. The data for different temperatures and variants collapse, without any adjustments, onto a single master curve. The red line is a linear fit to the plot of $\mathrm{l}\mathrm{n}\left(\mathrm{v}\mathrm{i}\mathrm{s}\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{y}\right)$ versus $\mathrm{l}\mathrm{n}\left(c_{\mathrm{s}\mathrm{a}\mathrm{t}}\right)$. (b) Plots of the measured temperature-dependent viscosities versus $c_{\text{sat}}$ for seven different systems (see legend). However, while there is a strong negative correlation between $\mathrm{l}\mathrm{n}\left(\mathrm{v}\mathrm{i}\mathrm{s}\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{y}\right)$ versus $\mathrm{l}\mathrm{n}\left(c_{\mathrm{s}\mathrm{a}\mathrm{t}}\right)$, the slopes are variant-specific. Viscosity measurements are extracted from videos taken from three individually prepared samples. (c) Computed flow activation energies for allY, allF and WT^+NLS the whiskers represent the standard deviation in the flow activation energy calculated from the fit. (d) Measured flow activation energies for allY, allF, WT^+NLS, and allW. Diamonds in (c, d) indicate the flow activation energy (in $RT$) at 25°C, error bars represent the error obtained from the linear fit. The r values in (a, b) are Pearson’s correlation coefficients.

Fig. 3:. Mutations that alter spacer solvation of A1-LCD accelerate the dynamical arrest of condensates.

(a) Relative changes in viscosities within condensates at time of observation $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 5\mathrm{m}\mathrm{i}\mathrm{n}$, calibrated against the allY system, and quantified as a function of different numbers of Gly to Ser substitutions. Data are shown for relative viscosities from measurements and computations, where the latter uses the collective model for the Zimm matrix. Error bars represent the standard deviation in the estimate of the mean relative viscosity. Experimental viscosities are obtained from at least 6 experiments ($n=9$ for allY, $n=7$ for allY^20GtoS, and $n=6$ for allY^30GtoS) featuring at least 50 tracked particles in each measurement. Saturation concentration $\left(c_{\text{sat}}\right)$ values shown for ≈ 20 °C. (b) Average diffusion coefficient ($D_{avg}$) of 200 nm polystyrene beads within WT^30GtoS condensates measured as a function of time following the start of sample imaging. The black line shows an exponential decay fit with the characteristic decay times ${\mathrm{\tau }}_{\text{decay}}$ as indicated. The time zero in the graph corresponds to time of observation, $t_{\text{obs}}\approx 5\mathrm{m}\mathrm{i}\mathrm{n}$. (c) Same as (b) but for allY^30GtoS condensates. (d) Histograms for the Root Mean Squared Displacement (RMSD) of 200 nm polystyrene beads measured at 10 s lag times within condensates at $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 5\mathrm{m}\mathrm{i}\mathrm{n}$ and $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 24\mathrm{h}\mathrm{r}\mathrm{s}$. The top, middle, and bottom panels show histograms for condensates at different times of $t_{\mathrm{o}\mathrm{b}\mathrm{s}}$ (ages of condensates are in the legends) formed by allY, allY^20GtoS, allY^30GtoS, respectively. (e) Ensemble-averaged Mean Squared Displacement (MSD) of 200 nm beads in allY condensates at $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 5\mathrm{m}\mathrm{i}\mathrm{n}$ (blue) and $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 24\mathrm{h}\mathrm{r}\mathrm{s}$ (red). The solid black line is the fit to the equation $MSD\left(\tau \right)=4D{\tau }^{\alpha }+N$. The value of the exponent obtained from the fit shows diffusive motion. (f) Same as (e) but for allY^30GtoS condensates. The values of the exponents show the onset of sub-diffusive motion. For panels (b)-(f), the experiments were repeated at least three times. (g) Temperature-dependent compliance computed from the simulations for condensates formed by allY and allY^30GtoS. T is temperature in Kelvin (K). We use the collective model for computing the Zimm matrix. Error bars are standard deviations of the mean compliance values computed using 3 replicates. Solid lines connect the data points.

Fig. 4:. Dynamically arrested A1-LCD condensates are Kelvin-Voigt solids with dominantly elastic behavior.

(a) Schematic of the optical tweezer-based creep test microrheology experiment. A bead (white circle) is trapped within a condensate (blue shading) using an optical trap (red hourglass shape). The trap is programmed to move a set distance (here 5 μm) at a constant velocity (0.5 μm/s). The bead position and the force exerted by the condensate on the bead are monitored via a camera and a quadrant photodetector, respectively. The black rectangle represents the slide surface. (b) Expected bead displacement in response to the trap movement during the creep test for a viscous liquid, a viscoelastic Maxwell fluid, and a Kelvin-Voigt solid. The red-shaded regions depict the period over which the stress is applied. (c) Bead displacement as a function of time in response to the programmed trap movement in allY and allY^20GtoS condensates at the time of observation, $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 5\mathrm{m}\mathrm{i}\mathrm{n}$. (d) Same as (c) but for aged allY and aged allY^20GtoS condensates ($t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 24\mathrm{h}\mathrm{r}\mathrm{s}$). Note that the bead in the allY^20GtoS condensate recoils back to its initial position after ~1 μm travel, indicating that the condensate is exerting an elastic force that overpowers the force exerted by the optical trap. As shown in panel (b), this response is characteristic of a viscoelastic solid. (e) Force exerted on the bead by the condensate network during the creep test experiment. Data are shown for allY and allY^20GtoS condensates at $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 5\mathrm{m}\mathrm{i}\mathrm{n}$. The saturation of the force is a characteristic of the viscous drag force, which depends on the velocity of the trap, the viscosity of the condensate, and the size of the bead. (f) Same as (e) but for condensates at $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 24\mathrm{h}\mathrm{r}$. The data in the bottom of panel (f) indicate that aged condensates formed by allY^20GtoS behave like Kelvin-Voigt solids. Comparison with the top panel highlights the fact that these condensates transition from being Maxwell fluids at early time points to dominantly elastic, Kelvin-Voigt solids as they age. The experiments were repeated at least three times.

Fig. 5:. Aged A1-LCD condensates show identical mesoscale morphology and absence of amyloid fibrils.

(a) Representative images from no fewer than four DIC microscopy images of different condensates at time of observation $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 5\mathrm{m}\mathrm{i}\mathrm{n}$ and at $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 24\mathrm{h}\mathrm{r}\mathrm{s}$. Insulin fibrils are shown as positive control. (b) ThT fluorescence ratios of the maximal / minimal intensities for freshly prepared condensates and condensates at $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 24\mathrm{h}\mathrm{r}\mathrm{s}$ for various A1-LCD variants. No changes were observed in ThT staining for condensates formed as a function of aging. As a positive control, we show data for amyloid fibrils formed by insulin after 24 hrs from sample preparation. At least four independent samples were prepared for each variant. Each diamond represents an individual data point, and the horizontal line indicates the mean.

Fig. 6:. Aged A1-LCD condensates are beta-sheet-containing non-fibrillar, semi-crystalline solids.

(a) Fourier transform infrared (FTIR) spectra for allY condensates at time of observation $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 5\mathrm{m}\mathrm{i}\mathrm{n}$ and $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 23\mathrm{h}\mathrm{r}\mathrm{s}$. (b) $t_{\mathrm{o}\mathrm{b}\mathrm{s}}$-dependent FTIR spectra for condensates of allY^30GtoS. (c) Difference spectra referenced to $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 5\mathrm{m}\mathrm{i}\mathrm{n}$, shown as change in absorbance units (ΔAU), show a decrease in signal for non-beta-sheeted regions and an increase in signal for beta-sheets, specifically aggregated beta-sheets (see Extended Data Fig. 7a). The difference spectra also reveal the presence of an isosbestic point, which is suggestive of a two-state disorder-to-order transition. (d) Freeze fracture deep-etch EM images of condensates of allY^30GtoS imaged at $t_{\mathrm{o}\mathrm{b}\mathrm{s}}\approx 24\mathrm{h}\mathrm{r}\mathrm{s}$. (e) Zoom-in image of a semi-crystalline material formed by allY^30GtoS. The semi-crystalline morphology is annotated by the presence of three regions labeled I, II, and III, which refer to an amorphous core (I), a lamellar crystalline region (II), and a halo generated by the reorientation of the samples during imaging. This halo is caused by the softness of the materials. Panels (d) and (e) are representative of a single sample.

Fig. 7:. Proposed free energy profile showing bistability and intra-well ruggedness.

The well corresponding to low values of the order parameter is a disordered, terminally viscous Maxwell fluid. The well on the right, corresponding to high values of the order parameter is the ordered, Kelvin-Voigt solid. The barrier for disorder-to-order transitions is expected to be high, $>30RT$, for condensates formed by naturally occurring PLCDs. This barrier is lowered, by destabilizing fluid-like states via mutations to spacers that increase their effective solvation volumes. Within the metastable and globally stable wells, we expect there to be ruggedness due to polymorphisms in the solid phase, and the contributions from conformational heterogeneity in the fluid phase. In a spring-dashpot representation, the spring (elastic element) is in series with the dashpot (viscous element) for a Maxwell fluid and in parallel for a Kelvin-Voigt solid. Conversion between material states can be depicted using a Zener model, which includes springs in series (E2) and in parallel (E1). The relative importance of E2 versus E1, depicted by pale versus bold coloring of the springs, is used to depict the conversion between states where E2 is dominant to states where E1 is dominant. The ruggedness we depict for the metastable fluid phase well is captured by the flow activation energy measurements. $F$ is the free energy of the condensate with fixed density $\left(c_{\text{dense}}\right)$, represented in terms of the order parameter $\mathrm{\Phi }\mid c_{\text{dense}}$.