Conformational Dynamics of Intrinsically Disordered Proteins Regulate Biomolecular Condensate Chemistry

2.1.2.1. Single-Molecule Confocal Fluorescence Microscopy

Single-molecule confocal fluorescence microscopy can be used to probe translational diffusion in liquid droplets. Schuler and co-workers probed IDP conformational dynamics using single-molecule fluorescence microscopy in a different but related context of molecular crowding. In the work by König et al., the IDP prothymosin α (ProTα) was fluorescently labeled at residues 1 and 56 mutated to cysteines and injected into living HeLa cells.¹¹² The measured intracellular diffusion time (τ_Diff = 1.8 ± 0.7 ms in cytosol and τ_Diff = 1.6 ± 0.6 ms in nucleus) was more than 2 times larger than in buffer (τ_Diff = 0.7 ± 0.1 ms) and corresponds to an effective intracellular viscosity of 2.8 ± 1.1 mPa s. In a later study,¹¹³ König et al. repeated experiments in HeLa cells in a medium that contained 20% (w/v) PEG 400.¹¹³ PEG 400 does not cross the cell membrane¹¹⁴ but creates an osmotic efflux of water. This reduced the cell volume by approximately 2-fold. In this study, the diffusion time of ProTα in noncrowded cells was 1.5 ± 0.2 ms and increased to 7 ± 2 ms in crowded cells.

According to the study by König et al., the protein and nucleic acid concentrations in HeLa cells before hyperosmotic stress are 108 ± 25 mg/mL and 23 ± 5 mg/mL, respectively. Notably, HEK 293-F eukaryotic cells had similar total protein concentrations, but only 25 mg/mL were soluble cytoplasmic proteins. Taking into account protein and nucleic acid crowding, we can expect macromolecule concentrations in the cytosol of crowded HeLa cells to be on the order of 100 mg/mL or more, which is similar to values observed in IDP condensates: concentration values of 300 mg/mL and higher were reported for the intrinsically disordered hnRNPA2 low-complexity (LC) domain (determined by NMR spectroscopy)⁸⁴ as well as for the Ddx4 LC domain⁸¹ and about 120 mg/mL for the intrinsically disordered FUS LC domain⁸³ (determined by spectrophotometry).

2.1.2.2. Single Particle Tracking Methods

Single particle tracking (SPT) methods offer an independent probe of droplet microrheology.¹¹⁵ Micro- or nanosized fluorescent beads or other observable particles (such as metal nanoparticles) are incorporated into in vitro formed droplets, and the trajectories of their Brownian motion-caused displacement are analyzed to probe the viscoelastic properties of the droplets. SPT was used to study droplets formed by LAF-1,⁹⁹ yielding a viscosity value of 34 ± 5 Pa s at physiological salt conditions (125 mM NaCl). Upon addition of 5 μM RNA, the droplet viscosity decreased 3-fold to 12.8 ± 0.8 Pa s. Murakami et al. employed SPT to probe viscosities of FUS LC reversible and irreversible gels formed by cyclic cooling and rewarming.¹⁰⁹ Liquid state, reversible and irreversible gel states of the FUS LC were found to have viscosities of 0.4 ± 0.03 Pa s, 3.8 ± 0.4 kPa s, and 15 ± 3 kPa s, respectively.

Liquidlike droplets and cellular condensates are not always homogeneous but can have multiple compartments^101,116,117 (Figure 3b). Detailed insights into the microenvironment of droplets formed by the ubiquitin-binding protein p62 during their maturation were obtained by Pan et al. using SPT in combination with dark-field microscopy.⁸⁰ Membraneless microdroplets formed immediately upon mixing p62 with GFP-labeled polyubiquitin (Ubx8). Before LLPS, gold nanorods (AuNR) were mixed with p62, and some of them were found to be embedded into p62 droplets. FRAP analysis of GFP-Ubx8 diffusion inside droplets revealed an apparent viscosity of 165.3 mPa s. The droplets fused and precipitated on the glass substrate, where they underwent a liquid-to-solid transition. During this transition, the droplets became highly heterogeneous. This included the formation of quasi-solid compartments and multiple vacuoles with external dimensions up to 18–120 μm and vacuole sizes between 0.7 and 25.6 μm.

In droplets that were undergoing liquid-to-solid transitions, the apparent diffusion rate was calculated from the Brownian motion trajectories of AuNR probes in the laboratory frame and was close to 0.3 μm²/s, yielding an apparent viscosity of ≈241.7 ± 157.4 mPa s in the maturing droplets. However, statistical analysis of the displacement of multiple AuNRs embedded in the same droplet revealed much lower diffusion rates relative to the droplet, ≈0.04 ± 0.018 μm²/s, indicating that nanorods trapped inside droplets were quasi stationary. AuNRs trapped in different regions had different diffusion rates ranging from 0.011 to 0.034 μm²/s, further underscoring the high heterogeneity of the droplets. A further rigidification of p62/Ubx8 droplets was observed using PEG-modified AuNRs that did not interact with p62 or Ubx8 and were mostly not trapped inside gel compartments, localizing instead in the liquid vacuoles or in the surrounding disperse phase. About 30 min after LLPS onset, PEG-AuNRs trapped inside vacuoles were repeatedly captured by the vacuole surface, suggesting that the gel phase in droplets was still fluid enough to allow for the formation of nanoscale pores or defects that captured diffusing nanorods. However, 1.5 h after LLPS, nanorods trapped inside vacuoles were experiencing elastic-like collision with their walls without sticking to them. This change in behavior suggests that at later stages of droplet maturation, the gel that surrounds the vacuoles becomes completely solid.

2.2.1.1. Local and Segmental Dynamics in IDPs Probed by NMR Relaxation

For a rigid protein, reorientational dynamics can be interpreted in terms of the global rotational tumbling and separate internal motions.^122,123 This distinction starts to break with increasing flexibility in the protein chain. Already in proteins containing multiple domains connected by flexible linkers, the global tumbling rate depends on the mutual conformational arrangement of the domains.^124−126 In intrinsically disordered proteins, there is no stable three-dimensional structure and the global tumbling appears to play little role in the description of their dynamics.^68,127−129 Instead, their conformational dynamics can be best described in terms of motions of individual segments of the IDP chain.^68,127 Some reference in the description of these segmental motions can be taken from models developed to study dynamics of homopolymers in the dilute limit, where intermolecular interactions play a negligible role,^130,131 such as the Rouse model.¹³² An important definition used by these models is the Kuhn segment, a distance along the polymer chain above which the relative dynamics of two monomers is not restricted by monomers between them. According to these models, locally restricted conformational fluctuations inside each Kuhn segment form a component contributing to chain dynamics that is distinct from slower modes describing chain hydrodynamics. These slower modes describe motions in a polymer chain ranging from reorientations of individual Kuhn segments to long-range chain reconfigurations. A similar distinction between local chain dynamics and slower motions of chain segments is present in the analysis of dynamics of proteins with flexible chains, including partially folded¹³³ and intrinsically disordered proteins.^{67,69,134,135}

One of the methods of choice to study IDP dynamics is NMR spin relaxation, which is sensitive to protein motions on time scales between tens of picoseconds and tens of nanoseconds, and provides a residue-specific description thereof.^{67,68,122,123,126,136−13915}N spin relaxation rates depend on the autocorrelation function describing the stochastic reorientation of ¹⁵N–¹H bond vectors and of the closely related ¹⁵N chemical shift anisotropy (CSA) tensors. The “model-free” analysis, typically used in folded proteins to approximate this autocorrelation function and to obtain time scales of global tumbling and internal motions,^122,123 is not applicable to IDPs and unfolded proteins, because these motions cannot be separated.¹⁴⁰ The most successful analyses of NMR relaxation rates in IDPs were performed with an approximation of the autocorrelation function as a sum of three exponentially decaying components (Figure 4)^69,131,135 that was first proposed by Brutscher et al.¹³³ Each of the three components has a characteristic time τ_k, which represents the time scale of a distinct motional mode, and the amplitude A_k representing the contribution of the motional mode to the relaxation rate:^131,135

When the amplitude of the slowest component (A₃) is small, the chain is very flexible. In this case, the more local motions have large amplitudes (A₁ + A₂), which efficiently quench the autocorrelation function. For large A₃ values, on the other hand, local motions are more restricted and chain segmental motions significantly contribute to the measured relaxation rates. Therefore, it is related to the local backbone conformational entropy.¹⁴¹ This relationship is however not straightforward, in particular because of the collectivity of motions of different residues and because not all chain motions are probed by NMR relaxation.¹⁴² Importantly, a faster mode decays fully before a slower mode starts, if the three characteristic times τ_k are sufficiently separated^126,143 (Figure 4). In such a situation, the three components describe statistically independent motions, validating the three-component approach. Otherwise there is no guarantee that the fitted time scales represent distinct motional modes.

The three-component analysis of conformational dynamics is also applicable to unfolded proteins, as shown earlier by Skrynnikov and co-workers.¹⁴⁴ They used ¹⁵N NMR spin relaxation measurements to rescale MD simulations and thus extract from the MD trajectories correlation functions, which describe the reorientation of the backbone ¹⁵N–¹H bond vectors of urea-unfolded ubiquitin. Three exponential components, with average (over the protein sequence) correlation times of 44 ps, 1.4 ns, and 9.4 ns and weights of 30%, 42%, and 28%, respectively, were identified. The shortest 44 ps component was assigned to fast peptide plane librations. The ∼1 ns component was attributed to local changes in the peptide plane dihedral angles. These might be compensated by motions in adjacent residues and therefore do not impact the global shape of the chain. The more long-range motions, which involve the conformational rearrangement of chain segments, were best described by the ∼10 ns component. Skrynnikov and co-workers also pointed out the role of the resistance of the solvent to the long-range chain conformational rearrangements (“hydrodynamic drag”) in the separation between local and segmental motions. Notably, correlation functions obtained from MD simulations performed in vacuum (i.e., without solvent) could be described with only one exponential component, and local changes in individual dihedral angles were resulting in global conformational rearrangements, highlighting the importance of the solvent for IDP dynamics.¹⁴⁴

To gain further insight into the physical origin of fast, intermediate, and slow components of the autocorrelation function of IDPs, Blackledge and co-workers analyzed characteristic times and amplitudes of three components in the C-terminal domain of the nucleoprotein of Sendai virus (N_TAIL).¹³⁵ Dynamic parameters were fitted to 58 relaxation rates measured at four different magnetic fields (14.1–22.3 T) and four temperatures (274–298 K). The observed separation of the extracted time scales supported the independence of the associated motional modes. In addition, the temperature and sequence dependence of their time scales pointed to distinct physical origins, as previously suggested by polymer models. The fastest of the three modes, which had a time scale of tens of picoseconds and exhibited a negligible temperature dependence, was assigned to librational motions of the ¹⁵N–¹H bond vector and the CSA tensor (Figure 5). The dominant contribution to the autocorrelation function came from the intermediate mode with a time scale of about 1 ns. Its flat sequence profiles of both time scale and amplitude (except in the helical region) suggested that it is independent of the chainlike nature of the protein. Supported by its temperature dependence, it was assigned to local backbone sampling (Figure 5). The characteristic times of the slow mode varied between 5 and 25 ns along the N_TAIL sequence. Both the sequence profiles of its amplitude and time scale were bell-shaped (decreasing toward the chain termini), implying that it is related to chainlike or segmental motions. Its temperature dependence correlates with the temperature dependence of the solvent viscosity, indicating that its dynamics are coupled to the solvent, as expected for chain motions. Therefore, it was assigned to the motions of the N_TAIL chain (Figure 5).

While basic relaxation properties of IDPs can be related to those identified in homopolymers (i.e., the presence of local and segmental motions), multiple features of IDP dynamics originate from the specific properties of residues in their chains. Residual secondary structure elements can result in a significant deviation of the dynamic parameters from values observed in the highly flexible parts of the chain. In addition, sequence hydrophobicity^67,145 and the presence of charged residues¹⁴⁶ influence chain conformational dynamics.¹³⁵ Differences in amino acid bulkiness can further influence IDP dynamics, resulting in deviations from random coil-like chain behavior.¹⁴⁷ Importantly, the presence of intermolecular contacts, such as in the condensates formed by LLPS, will likely influence the description of motional modes developed for IDPs in the dilute limit.

2.2.1.2. Segmental Dynamics in IDPs Probed by NMR Relaxometry

Parigi et al. used field-cycling ¹H NMR relaxometry to study long-range segmental dynamics in IDPs. Dispersion profiles of collective, nonresidue specific proton relaxation rates were acquired at 298 K at fields ranging from 0.02 to 45 MHz for four IDPs: the 140-residue protein α-synuclein, the 134-residue β- synuclein, a 99-residue fragment of the protein Tau termed K19, and the 129-residue protein lysozyme in a denatured state. The dispersion profiles were fitted with a collective correlation time τ_R describing correlated motions of IDP segments, associated squared collective order parameter S_C², representing the amplitude of contribution of these motions to the relaxation, and an additive parameter α representing the residual contribution of faster motions. τ_R values of 7.9 ± 0.5 ns, 6.5 ± 0.3 ns, 6.1 ± 0.9 ns, and 5.3 ± 0.4 ns were obtained for α-synuclein, β-synuclein, denatured lysozyme, and Tau K19, respectively, with associated order parameters being close to 0.1. This range of τ_R values (6–9 ns) is similar to the time scale of long-range segmental motions determined in unfolded ubiquitin by Skrynnikov and co-workers (vide supra).

To evaluate the impact of the chain length on segmental dynamics in IDPs, Parigi et al. also measured proton longitudinal relaxation dispersion in a fragment comprising the N-terminal 30 residues of α-synuclein (named N30 α-synuclein) and in htau40 the longest isoform of Tau with 441 residues¹²⁸ (Figure 6). In the case of N30 α-synuclein, it was possible to estimate only the upper (slower) limit of τ_R, which was around 2 ns. For htau40, the dispersion profiles were initially fitted with a single τ_R, resulting in τ_R = 10 ns and S_C² = 0.12. However, this fit was not optimal, and it was repeated with two correlation times, τ_R1 and τ_R2, resulting in τ_R1 = 27 ± 0.5 ns, S_C1² = 0.02 and τ_R1 = 4 ns, S_C2² = 0.17. Based on the differences in the τ_R values of N30 α-synuclein, α-synuclein, and htau40, it was concluded that the correlation times of segmental motions in IDPs increase with the length (or molecular weight) of the protein, and that in IDPs with long chains (such as htau40), additional modes with slower correlation times may be necessary to describe their dynamics. Notably, the chain length dependence of segmental motions is not usually detectable by high-field NMR relaxation experiments in unfolded proteins/protein regions, as the autocorrelation function is already quenched by faster local motions. Field-cycling ¹H NMR relaxometry, however, probes lower frequencies and can therefore be more sensitive to slower chain motions that may appear in longer chains. Indeed, the combination of lower field measurements using NMR relaxometry and residue-specific high-field NMR relaxation experiments are likely to be particularly powerful to dissect IDP motions on multiple time scales.^148−151

The intrinsic dynamics of IDP chains can also be predicted from the amino acid sequence of an IDP taking into account hydrodynamic coupling.¹²⁸ This approach is based on the HYCUD (hydrodynamic coupling of domains) algorithm.^124,152 In HYCUD, an IDP chain is split into multiple rigid nonoverlapping fragments of the size of the Kuhn segment, 14 residues.^70,127 As part of the IDP, each fragment is experiencing an increase in the effective viscosity relative to the viscosity of the solvent, due to the presence of other fragments. This effective viscosity is calculated from the intrinsic viscosity of all other fragments, calculated by HYDROPRO,¹⁵³ and their effective concentration calculated based on the distance to the fragment in question. The rotational correlation time of the fragment in question is then slowed according to the ratio between effective and solvent viscosities. HYCUD was shown to accurately predict τ_R values in different IDPs. In addition, the experimentally observed dependence of τ_R on IDP chain length was well reproduced. Future work has to show to which degree the residual structure, which might change the persistence length,¹⁵⁴ influences the HYCUD-based chain dynamics description of IDPs.

2.2.1.3. Slow and Long-Range Chain Conformational and Reorientation Dynamics Probed by Fluorescence Spectroscopy Methods

Slower and longer-range chain conformational dynamics in IDPs, and their changes upon LLPS, can be accessed using fluorescence spectroscopy methods (Figure 7), such as fluorescence correlation spectroscopy (FCS), which is sensitive to chain motions on timescales slower than the fluorophore lifetime (several nanoseconds). Originally FCS was designed to probe translational diffusion of the fluorescent molecule, as it moves in and out of the measurement volume.^76,155 If the IDP chain contains an attached fluorophore and quenchers (such as aromatic residues), contacts between the quenchers and the fluorophore result in fluorescence quenching through photoinduced electron transfer (PET), and the time scale of the quenching process is reported by the fluorescence intensity correlation function. PET-FCS uses this effect to probe chain reconfiguration events resulting in these contacts on time scales between nanoseconds and milliseconds.⁹⁰ Alternatively, FCS can be combined with Förster resonance energy transfer (FRET). If the protein is labeled with a donor and an acceptor fluorophore, the variations of distance between the two probes would result in variations of the efficiency of an energy transfer between them, which are reported by the donor and acceptor intensity autocorrelation functions and the cross-correlation between donor and acceptor intensities.⁹¹

PET-FCS was used to probe chain dynamics of the disordered N-terminal TAD domain of the tumor suppressor protein p53.¹⁵⁶ p53-TAD natively contains three tryptophan residues, and they were mutated to phenylalanine residues, leaving alternatively tryptophan at position 23 or 53. Residues 13, 31, and 60 were alternatively mutated to cysteines to attach a fluorescent oxazine dye. In this way, it was possible to study the time scales of loop closure for the chain loops 13–23, 23–31, 31–53, and 53–60 of p35-TAD. For all four chain loops, FCS had a fast component, assigned to loop closure kinetics (corresponding to the time scale of ∼100 ns), and a slower microsecond component of minor amplitude, assigned to larger-scale chain conformational reorganizations having an impact on loop closure kinetics (Figure 8). Upon binding of MDM2, the fast ∼100 ns loop-closing component slowed down to the microsecond time scale in the loop 13–23 involved in the binding interface.

FRET-FCS was used to probe chain conformational dynamics in the 283-residue disordered cyclin-dependent kinase inhibitor Sic1 (Figure 9).¹⁵⁷ Measurements were performed in Tris buffer (150 mM NaCl, 50 mM Tris at pH7.4) and in pure Milli-Q water to evaluate the impact of charge screening on chain dynamics. In Tris buffer, a nanosecond-time scale component and two millisecond-time scale components (17.3 ± 0.7 ms and 83.7 ± 6.3 ms) to FCS were observed. In pure water, however, only millisecond-time scale components (13.1 ± 0.9 ms and 58.6 ± 6.1 ms) were detected, and the nanosecond-time scale component disappeared. From these data it was concluded that the Sic1 chain was stiffer in pure water compared to Tris buffer due to electrostatic repulsion between positively charged residues in the absence of charge screening, and this led to the disappearance of fast nanosecond-time scale dynamics.

However, in both PET-FCS and FRET-FCS, probe chain conformational dynamics are limited to the relative translational diffusion of two positions in the IDP chain. Additional insight can be gained when measuring FCS in polarized light (polarization-resolved FCS; Pol-FCS),¹⁵⁸ because the efficiency of dye excitation by polarized light depends on its constant fluctuating orientation in space, and the correlation curve thus captures these fluctuations. Indeed, Pol-FCS was used to understand whether heterochromatin protein 1 (HP1) undergoes LLPS in cells.¹⁵⁹ To this end, green fluorescent protein (GFP) was covalently attached to HP1. The hypothesis underlying the experiments was that the presence of LLPS in chromocenters would result in increased local viscosity of the medium and decreased rotational diffusion through protein–protein interactions. Translational diffusion would be, on the other hand, impacted by factors such as binding interactions with chromatin and collisions with diffusion barriers and would be a less precise probe of LLPS. However, the GFP-HP1 rotational correlation time observed in chromocenters (111 ± 8 ns) was similar to that in the surrounding nucleoplasm (117 ± 9 ns), and only in the cytoplasm it was smaller (74 ± 7 ns). This suggested that there might not be a specific viscous microenvironment surrounding chromocenters. The combination of polarization-resolved FCS with fluorescence anisotropy measurements (discussed in section 2.2.4.1) could potentially probe chain reorientation dynamics on timescales from 0.1 ns to milliseconds.¹⁶⁰

In filtered FCS, signals emanating from different fluorescent species present in solution can be detected and processed separately, based on their properties such as fluorescence lifetime and polarization-resolved spectral information.¹⁶¹ These species could also represent the same molecule but in different conformations modulating fluorescence parameters,¹⁶¹ allowing a more precise description of conformational and reconfiguration events governing chain dynamics. Tsytlonok et al. used filtered FCS to study conformational dynamics of the intrinsically disordered cyclin dependent kinase inhibitor protein p27. Fluorescence signals were filtered based on observed FRET efficiencies and fluorescence autocorrelation curves. In addition, cross-correlation curves were analyzed with four exponential decay components describing chain dynamics. The three slowest components (2–2.6 μs, 23.1–25.7 μs, and 297–363 μs) had similar time scales among different constructs. In contrast, the shortest component was significantly different with relaxation times of 22 ± 30 ns, 70 ± 40 ns, and 130 ± 30 ns for the donor/acceptor pairs 29/54, 54/93, and 75/110, respectively. Tsytlonok et al. concluded that the faster 22 ns-time scale dynamics occurring in the p27 region between residues 29–54 is important for p27’s ability to undergo chain expansion in the interaction with Cdk2/Cyclin A.

Chain distance dynamics that occur on time scales slower than milliseconds can be probed by analyzing single-molecule Förster resonance energy transfer (smFRET) efficiency histograms (time scales up to 0.1 s) and total-internal reflection fluorescence (TIRF) intensity trajectories of immobilized molecules (time scales up to 1000 s).⁸⁸ On the opposite side, a more detailed description of chain reorientation dynamics on the nanosecond timescale could be obtained using nanosecond FCS-FRET (nsFCS). nsFCS is based on the detection and analysis of donor and acceptor intensity fluctuations on the timescale of their lifetimes and is sensitive to chain motions in the 10–1000 ns range.¹⁶² Donor and/or acceptor intensity correlation functions can be fitted with a model of diffusion in a potential of mean force describing the random displacement of one probe relative to another with an equilibrium distance between them.⁹²⁻⁹⁴ The relaxation time, τ_r, obtained from this model, corresponds to the reconfiguration time of the chain between the two fluorophores. In addition, the distance distribution function can be extracted from the smFRET transfer efficiency histograms, and the radius of gyration (R_g) of the chain can be estimated. nsFCS is complementary to the “classic” FRET-FCS, and the two methods can be combined, potentially allowing one to observe motions on time scales between nanoseconds and seconds.⁹⁴ Rezaei-Ghaleh et al. used nsFSC to probe long-range chain dynamics in α-synuclein, fluorescently labeled at residues 42 and 92 mutated to cysteines. When using a single-exponent model for fitting the nsFCS data, a τ_r of 58 ± 13 ns was obtained. However, additional MD simulations suggested that a double-exponent model explains better the distance correlation functions obtained from the trajectory. Fitting also the nsFCS experimental data with a double-exponent model yielded a shorter correlation time τ_r1 = 23 ± 4 ns (contributing 66 ± 2%) and a longer correlation time τ_r2 = 136 ± 33 ns (with a contribution of 34 ± 2%), potentially probing two modes in the spectrum of chain relaxation modes predicted by the Rouse model. In addition, a combination between nsFCS and PET-FCS was proposed by Schuler and co-workers.¹⁶³

Time scales probed by fluorescence correlation spectroscopy methods provide information about longer-range chain reconfiguration modes than those probed by NMR relaxation and relaxometry. Because the current data suggest that the conformational dynamics of IDPs in condensates are slowed down (reviewed in section 2.2.3.1), FCS experiments could be particularly useful to interrogate the slowed down motions and thus shed light on the changes in the dynamic properties of IDPs upon phase separation. However, quantitative analysis of time scales and rates extracted from FCS analysis should be done with care. First, the time scales obtained in different methods are not equivalent. PET-FCS reports only on the formation of contacts shorter than 10 Å, whereas FRET-FCS reports on events where dyes are in a distance range between 10 and 100 Å, and a conversion is required¹⁶⁴ to compare values from two methods. Polarized FCS probes rotational reconfiguration of chain segments and not their relative translational diffusion, and though the two are related to the chain reconfiguration dynamics, a chain model would be required to reconcile the two values. Importantly, as in fluorescence anisotropy measurements described later in this review, the fitted time scales are likely representing averaged values over the timescale range to which the method was sensitive in a given system.

2.2.3.1. Fast IDP Motions Inside Condensates

While still in its infancy, first studies have started to investigate residue-specific dynamics of IDPs inside condensates. In the LC domain of FUS, the ¹⁵N transverse relaxation rate R₂ strongly increased upon LLPS from <5 s^–1 to 15–35 s^–1 (Figure 12).⁸³ Notably, ¹⁵N R₂ relaxation in unfolded proteins and IDPs has been linked to segmental chain dynamics.¹²⁷ It is therefore likely that the LLPS-induced increase in the ¹⁵N transverse relaxation rate R₂ is caused by the slowing down of segmental dynamics. In addition, the ¹⁵N–¹H heteronuclear NOEs slightly increased in the FUS LC domain upon LLPS (Figure 12). Because ¹⁵N–¹H heteronuclear NOEs are more sensitive to high frequency motions, the data indicate that concentrating the FUS LC domain into condensates only slightly restricts or slows high-frequency motions but predominantly slows down segmental chain dynamics.

In the LC domain of the hnRNPA2 protein, a component of RNA-processing membraneless organelles,^174,17515N R₁ rates as well as ¹⁵N–¹H heteronuclear NOEs increase upon LLPS (Figure 13).⁸⁴ The increased ¹⁵N–¹H heteronuclear NOEs pointed to slowed or restricted local motions. However, ¹⁵N R₂ rates were quite similar before and after LLPS (Figure 13). Because the NMR relaxation measurements were performed at 65 °C, the ¹⁵N R₂ rates might be influenced or even be dominated by fast amide hydrogen exchange with water.⁸⁴ Further measurements at physiological temperatures are thus required to dissect the influence of LLPS on the conformational dynamics of the LC domain of hnRNPA2.

In the phase separation of ELP₃, an elastin-like polypeptide, the translational diffusion rate decreased by 2 orders of magnitude, from ≈100 μm²/s to ≈1 μm²/s.⁸⁵ LLPS of ELP₃ also resulted in an increase in ¹⁵N–¹H heteronuclear NOEs and ¹⁵N R₂ relaxation rates (from ≤5 s^–1 to >15 s^–1 at 14.1 T, 37 °C), consistent with a slow down or restriction of both local motions and chain reconfigurations, respectively.⁸⁵

In these pioneering NMR relaxation studies of IDP dynamics under LLPS conditions, only a reduced set of spin relaxation rates was acquired (¹⁵N R₁, ¹⁵N R₂, and heteronuclear NOEs), which is not enough for their analysis in terms of timescales and amplitudes of motional components. Such an analysis would, however, be necessary to quantify the exact impact of LLPS on both local and segmental chain motions and delineate the contributions to the measured rates from either slower timescales of motions or reduced local chain backbone conformational entropy resulting in decreased amplitudes of local motions.

2.2.3.2. Chain Flexibility and R₁ Relaxation

The ¹⁵N spin relaxation studies available so far demonstrate that ¹⁵N R₂ relaxation rates and ¹⁵N–¹H heteronuclear NOE values increase when IDPs are concentrated inside condensates. The picture is however less clear in the case of residue-specific R₁ relaxation rates. In the LC domain of hnRNPA2, phase separation was accompanied by a strong increase in ¹⁵N R₁ rates from 0.5 to 1.25 to 1.25–1.75 s^–1 (Figure 13). In contrast, the ¹⁵N R₁ rates of the FUS LC domain decreased upon LLPS by approximately a factor of 1.5 from 1.3 to 1.7 s^–1 to 0.75–1.2 s^–1 (Figure 12). Thus, ¹⁵N R₁ relaxation rates respond in an opposite manner to phase separation in the two systems.

Residue-specific differences in the sensitivity of ¹⁵N R₁ relaxation rates to increased molecular crowding and viscosity were previously reported for the highly flexible N-terminal region of mitogen-activated kinase kinase 4 (MKK4).¹⁷² Addition of increasing concentrations of the molecular crowding agent dextran increased the ¹⁵N R₁ rates for the N-terminal ∼25 residues of MKK4 (Figure 14).¹⁷² In contrast, lower ¹⁵N R₁ rates were observed for residues 55–80 in the presence of high concentrations of dextran (Figure 14). Notably, neither the 25 N-terminal residues nor residues 55–80 of MKK4 contain significant amounts of regular secondary structure.¹⁷⁶ Thus, even in a single disordered polypeptide chain, the ¹⁵N R₁ relaxation rates can respond in distinct ways to increased viscosity.

To gain insight into the interplay between chain flexibility and ¹⁵N R₁ relaxation rates, we modeled the influence of molecular crowding on R₁ spin relaxation rates. We took into account previous observations that increased molecular crowding predominantly impacting the characteristic times of different motional contributions but not their amplitudes.¹⁷²¹⁵N R₁ relaxation rates were calculated using the three-component description of the autocorrelation function (Figure 4). We used typical values of characteristic times and amplitudes obtained from ¹⁵N relaxation measurements in flexible and rigid parts of MKK4 at different concentrations of the crowding agent.¹⁷² Molecular crowding strongly slows down slower chain motions represented by the slow component with the characteristic time τ_slow (Figure 15a). In addition, the intermediate component with the characteristic time τ_int, which represents local backbone motions, slightly increased with increased viscosity (Figure 15a). The flexible parts of MKK4 had a major contribution of the intermediate component (A_int > 0.5) and a very small contribution of the slow component (A_slow < 0.1). The more rigid parts of MKK4, on the other hand, had a sizable contribution of the slow component (A_slow ≈ 0.3–0.4).

To represent both a more flexible and a more rigid IDP, we compared the two three-component models A_slow = 0.05/A_int = 0.55/A_fast = 0.4 and A_slow = 0.4/A_int = 0.4/A_fast = 0.2, respectively. For the motional model with a smaller amplitude of slow motions (A_slow = 0.05), the ¹⁵N R₁ rate increased with increasing the concentration of the crowding agent (red line in Figure 15b). In contrast, the ¹⁵N R₁ rate decreased from 2 s^–1 to 1.5 s^–1 for the slow motion model (A_slow = 0.4; blue line in Figure 15b). When a single exponentially decaying component defines the autocorrelation function, the ¹⁵N R₁ rate behaves nonlinearly with its correlation time: for correlation times below ∼2.5 ns, R₁ increases, while above ∼2.5 ns it decreases (Figure 15c). Therefore, when faster (τ_fast, τ_int < 2.5 ns) and slower (τ_slow > 2.5 ns) components define chain dynamics simultaneously, the ¹⁵N R₁ rate behavior depends on the contribution of each component. This analysis shows that in a flexible chain, where faster/intermediate backbone motions have a major contribution (A_int) to the correlation function, R₁ values will increase with slowing down of characteristic times of motions in more crowded conditions. In a more rigid chain, where these motions are more restricted and their contribution is low, R₁ values will decrease with crowding. Depending on the relative contribution of different motional modes to the correlation function, R₁ relaxation rates can thus decrease or increase upon LLPS of IDPs.

We also point out that ¹⁵N–¹H heteronuclear NOEs in the monomeric state of FUS LC are positive (Figure 12c), consistent¹⁷⁷ with more restricted backbone dynamics than in hnRNPA2, where negative NOE values are present (Figure 13c). In addition, the LC domains of hnRNPA2 and FUS differ in their compaction. Despite similar lengths of the two proteins (163 residues for FUS; 151 residues for hnRNPA2), the FUS LC domain has a larger hydrodynamic radius (3.32 nm) than the LC domain of hnRNPA2 (2.89 nm).⁸⁴ This difference in hydrodynamic radii can be explained by the collapse of hnRNPA2 LC, which is favored by its high glycine content (47%) (Figure 13b). Both data support a higher flexibility of the hnRNPA2 backbone when compared to the polypeptide chain of the LC domain of FUS. The decrease in the ¹⁵N R₁ relaxation rates in FUS LC upon phase separation (Figure 12c) is therefore likely a consequence of its more rigid (when compared to hnRNPA2) backbone with lower local conformational entropy.

In the analysis of LLPS behavior of ¹⁵N R₁ relaxation rates provided here, the chain flexibility (the amplitudes of chain conformational sampling) was assumed to remain constant upon LLPS. As we will discuss later, this is not necessarily true: IDP chains can become even more rigid upon phase separation. In the case of FUS LC, this effect could decrease ¹⁵N R₁ values even further in the condensed phase.

2.2.3.3. Molecular Crowding and Intermolecular Contacts

LLPS-induced slowing of IDP backbone motions was also reported for the intrinsically disordered N-terminal 236 residues of the germ-granule protein Ddx4.⁸¹ The measured diffusion constant of Ddx4 in the condensed phase was very small, 0.75 ± 0.04 μm²/s. This value corresponds to a protein of an apparent hydrodynamic radius of ∼550 nm in dilute aqueous solution. Despite such slow apparent reorientational dynamics, amide resonances and sharp ¹³C side-chain resonances were present in the NMR spectra, indicating that local motions are still rapid inside the condensate. Residue-specific ¹⁵N R₂ relaxation rates increased from 1 to 6 s^–1 in the dilute phase to 16.4 ± 6.1 s^–1 in the condensed phase (at 18.8 T, 30 °C)⁸¹ (Figure 16). On the other hand, ¹⁵N–¹H heteronuclear NOEs, which are sensitive to local motions on the pico- to nanosecond time scale, did not change significantly upon LLPS, in agreement with the observed line width of the NMR signals. Brady et al. further analyzed the ¹⁵N R₁/R₂ relaxation rates and ¹⁵N–¹H heteronuclear NOEs in terms of the S²τ_c product. S²τ_c is the square of the order parameter S, which describes the amplitude of the amide bond vector motions, multiplied by the residue-specific chain tumbling time τ_c. The analysis yielded S²τ_c = 1.3 ± 0.6 ns for the dilute phase of Ddx4, and 8.5 ± 3.0 ns for the condensed phase. The dynamics of the disordered Ddx4 chain are thus slower inside the condensate.

Slowed conformational dynamics of IDPs can arise from both increased viscosity and intermolecular contacts inside condensates. To evaluate to which extent higher R₂ relaxation rates and S²τ_c values in the condensate can exclusively be attributed to increased molecular crowding, a mutant of Ddx4 was constructed in which all 14 phenylalanines were mutated to alanine (termed Ddx4_14FtoA; Figure 16). Ddx4_14FtoA can be concentrated to 370 mg/mL without undergoing phase separation. The concentration of 370 mg/mL of Ddx4_14FtoA in the nonphase-separated state is similar to the concentration of the wild-type protein in the condensed phase (380 mg/mL). ¹⁵N R₂ relaxation rates measured in Ddx4_14FtoA were 9.8 ± 3.9 s^–1, and the S²τ_c product was ∼5.3 ns. Despite similar levels of molecular crowding, both the ¹⁵N R₂ relaxation rates and the S²τ_c product was lower for the Ddx4_14FtoA mutant protein when compared to wild-type Ddx4 inside the condensate (¹⁵N R₂ = 16.4 ± 6.1 s^–1; S²τ_c = 8.5 ± 3.0 ns). The additional slowing of chain dynamics in the condensed phase compared to the Ddx4_14FtoA mutant was attributed to intermolecular contacts. At the same time, the removal of 14 aromatic phenylalanine side chains in Ddx4_14FtoA might have had itself an impact on the chain flexibility and thus on the ¹⁵N R₂ relaxation rates, for example, via the reduction of the number of intramolecular hydrophobic interactions.

Biomolecules that phase separate often contain multiple sites that engage in intra- and intermolecular interactions, a feature that is called multivalence. Multivalent molecules have a tendency to assemble into larger complexes, which increases with the number (valence) and affinity of the interacting sites. This assembly decreases their solubility and promotes phase separation, which in turn facilitates intermolecular contacts and promotes the formation of even larger complexes.¹⁰ The two processes are coupled, resulting in the formation of interconnected networks of macromolecules and a density increase in the condensed phase.¹⁷⁸ Mittag and co-workers adapted a stickers-and-spacers model, developed for associative polymers,^179,180 to simulate the phase separation behavior of intrinsically disordered prion-like domains (PLDs) and LC domains.¹⁷⁸ They then applied the analysis to the phase separation of the LC domain of heterogeneous nuclear ribonucleoprotein A1 (hnRNPA1 LC). Residues or groups of residues that are involved in intermolecular interactions promoting phase separation were called “stickers”, and other residues that are interspersed between stickers were called “spacers” (Figure 17). “Stickers” were associated with aromatic residues, which are often present in PLDs and contribute to their phase separation.^41,98,181 This assignment was substantiated by experimentally measured ¹⁵N R₂ NMR spin relaxation rates, featuring elevated R₂ rates around aromatic residues, reporting on restricted chain motions. A lattice-based coarse grained model with a single bead (sticker or spacer) representing each residue was used to simulate phase separation binodals. In the model, sticker–sticker interactions were set to be more energetically favorable than sticker–spacer and spacer–spacer interactions. Simulated binodals for wild-type hnRNPA1 LC and several mutants (having more or less aromatic residues than the wild-type protein) matched experimentally determined phase transition points. The critical temperature and the width of the two-phase regime increased with an increasing number of aromatic residues in hnRNPA1 LC variants, emphasizing the role of valence in phase separation.

Notably, intermolecular interactions that promote LLPS are not only restricted to interactions between aromatic residues.^10,182 Electrostatic interactions between segments or residues of opposite charge^{98,99,183−185} were also found to drive LLPS. In addition, noncharged residues can participate in LLPS. Polyglutamine polymers were found to phase separate, possibly through the formation of hydrogen bonds.¹⁸⁶ In elastin-like polypeptides, contacts between Ala, Val, and Pro residues were formed in the condensate, suggesting the role of intermolecular hydrophobic interactions in LLPS.⁸⁵ Kim et al. combined a specific labeling scheme with edited-filtered nuclear Overhauser effect spectroscopy (NOESY) to probe intermolecular contacts relevant for the LLPS of the C-terminal intrinsically disordered region of RNA-binding CAPRIN1 protein at a residue-specific level.¹⁸² Whereas the importance of the interaction between aromatic side chains for CAPRIN1 LLPS was confirmed, a previously not fully appreciated role of backbone interactions, in particular H^α–H^N contacts involving aromatic and nonaromatic residues, was established. These contacts were suggested to be stabilized through amide hydrogen bond formation and/or π–π interactions. Indeed, three short regions with significantly increased aliphatic and aromatic to amide proton NOEs in CAPRIN1 condensate, ⁶²⁴GYR⁶²⁶, ⁶³⁸GYR⁶⁴⁰, and ⁶⁶⁰RDYSGYQ⁶⁶⁶, were identified. Mutating the first three residues in these regions to ASA attenuated LLPS. NMR ¹⁵N R₂ spin relaxation rates were also particularly elevated in the ⁶³⁸GYR⁶⁴⁰ and ⁶⁶⁰RDYSGYQ⁶⁶⁶ regions both in monomeric CAPRIN1 and in the CAPRIN1 condensate, indicating the presence of intramolecular interactions restricting its chain dynamics. Notably, R₂ relaxation rates decreased significantly in these regions upon the LLPS-disrupting mutation to ASA, suggesting a link between intramolecular interactions in the monomeric CAPRIN1 and the intermolecular interactions stabilizing LLPS.

2.2.4.1. Fluorescence Anisotropy

Fluorescence anisotropy was used to probe LLPS-induced changes in the conformational dynamics of the intrinsically disordered protein α-synuclein.⁵⁹ Experimental fluorescence anisotropy decay curves were fitted in the dispersed monomeric state and the condensed phase with a single exponential decay. From this analysis, rotational correlation times of 1.0 and 1.6 ns were derived for α-synuclein outside and inside the droplets, respectively. In addition, Ray et al. noted that the decay curve had a positive y-intercept (∼0.1), consistent with the appearance of the contribution of much slower (>10 ns) motions inside droplets.⁵⁹ The amplitude of faster motions is therefore restricted inside droplets, consistent with reduced chain flexibility and decreased local conformational entropy. This explanation was also provided by the authors,⁵⁹ who related the chain rigidification to the specific intermolecular interactions that drive α-synuclein droplet formation.

Increased rotational correlation times were also observed upon phase separation of the full-length human tau protein htau40.¹⁸⁷ To this end, four double-cysteine mutants, Tau_17/244, Tau_149/244, Tau_244/354, and Tau_354/433 (position of first cysteine mutation/position of second cysteine mutation), were prepared and labeled them the fluorescent dye AF488. In the monomeric form of htau40, the fluorescence anisotropy curves were best fitted with correlation times of 1.9 ± 0.2 ns, 1.5 ± 0.1 ns, 1.5 ± 0.1 ns, and 1.2 ± 0.1 ns for Tau_17/244, Tau_149/244, Tau_244/354, and Tau_354/433, respectively. In droplets, these correlation times increased to 4.7 ± 0.7 ns, 2.6 ± 0.2 ns, 3.2 ± 0.2 ns, and 2.7 ± 0.2 ns, respectively. However, in contrast to the α-synuclein study, no contribution of slower motions appeared in fluorescence anisotropy decay curves in droplets.

Fluorescence anisotropy decay in IDPs is also often analyzed with multiple exponential decay components.^188,189 To probe the conformational dynamics of the repeat domain of the intrinsically disordered protein tau in liquidlike droplets, the fragment tau-K18 (residues 244–372 of htau40) was labeled at the two native cysteine residues Cys291 and Cys322. To describe the rotational tumbling of the fluorescent probe, the observed time-resolved fluorescence anisotropy decay was fitted with a sum of two exponential components.⁸⁷ The correlation time of the slow component decreased from ∼3 ns to ∼1.1 ns upon LLPS. In parallel, the correlation time of the fast component decreased from 440 to 240 ps, whereas its contribution increased from 52% to 73% (Figure 18b). This would point to faster dynamics in tau-K18 inside droplets, whereas the inverse effect, i.e., deceleration of the conformational dynamics in IDPs upon LLPS, is generally observed. According to this model, the tau chain in the monomeric state is partially collapsed and the slow component represents its global tumbling. Upon LLPS, the tau chain would undergo a conformational expansion, and the slow component now represents the backbone torsional mobility in the dihedral angle space of expanded disordered conformations, potentially explaining its faster correlation time.⁸⁷ A similar behavior was observed by Dogra et al. using fluorescence anisotropy in the LLPS of the disordered oligopeptide repeat domain of the melanosomal protein Pmel17.¹⁹⁰ The fluorescence anisotropy decay curves were fitted with two components. The correlation time of the slow component decreased from 1.76 to 1.1 ns upon LLPS, and the correlation time of the fast component decreased from 390 to 230 ps.

The observed changes in the fluorescent anisotropy decay of tau-K18 and Pmel17 point to different conformational dynamics of the proteins before and after LLPS. However, it is less clear if these changes can be connected to changes in the overall compaction of these IDPs upon phase separation. One limitation for example is related to the description of the dynamics of IDPs in terms of global tumbling. Excluding the cases where the IDP chain is largely collapsed, NMR spin relaxation^68,127,128 and fluorescence anisotropy¹²⁹ studies questioned the appropriateness of describing the dynamics of IDPs in terms of overall or global tumbling. Instead, a notion of motions of chain segments or segmental motions was proposed. The ∼3 ns correlation time of the slow component, which was observed for nonphase separated tau-K18 at 25 °C, is in the 2–6 ns range of the slow component characteristic time as detected by NMR spin relaxation in an IDP chain without secondary structure elements at 25 °C.¹³⁵ On the other hand, the value of ∼1.1 ns obtained upon LLPS is more close to the range of 0.5 to 1.3 ns, which was assigned to local backbone sampling motions.¹³⁵ It is therefore possible that in tau-K18, the long-range chain motions actually become much slower upon LLPS, as reported in other systems, but its contribution to the decay curve is small. In addition, the apparent decrease in both the slow and fast motional components upon LLPS might result from an averaging between the time scales of chain dynamics, local backbone sampling, and rotational mobility of the fluorescent probe (Figure 18a). In NMR spin relaxation studies, when two components were used instead of three to describe IDP dynamics, their characteristic times were situated in between the characteristic times obtained by a three-component analysis.¹³⁴

2.2.4.2. EPR Spectroscopy

LLPS-induced changes in the conformational dynamics of a fragment of tau were probed by continuous-wave EPR spectroscopy.⁴⁵ To this end, the tau fragment Δtau187 (residues 225–441 of htau40) was spin-labeled at cysteine 322. Subsequently, Δtau187 was concentrated into droplets through complex coacervation with RNA. Continuous-wave EPR spectra were recorded at 25 °C. No change in the EPR profile was observed upon complex coacervation (Figure 19). The overlapping EPR profiles were fitted with a single rotational correlation time of 425 ± 16 ps. Lin et al. therefore suggested that Δtau187 inside of RNA-induced droplets has the same dynamical properties as in the monomeric state.⁴⁵

Because different IDPs and IDP fragments might behave differently upon LLPS, in particular when LLPS is induced in different ways (for example, self-coacervation versus complex coacervation), Δtau187 in droplets might indeed retain the same dynamical properties as in the monomeric state. At the same time, it is important to note that EPR spin labels and fluorescence tags attached to an IDP chain may have intrinsically larger amplitudes of reorientation relative to the IDP backbone. The time scale of 425 ps obtained by Lin et al. in Δtau187 thus might result from the averaging of time scales of Δtau187 local backbone dynamics and the rotational mobility of the spin label (Figure 18a). Fast motions of the probe (spin label or fluorescent tag) may dominate the autocorrelation function that describes the reorientation rendering EPR spectroscopy and fluorescence anisotropy experiments less sensitive to the slower motions of the IDP backbone chain.