FUS nanoclusters are a distinct state within the dilute phase

Ge, Yingda; Paul, Tapas; Gordiychuk, Margarita; Das, Nilimesh; Zhang, Yaojun; Myong, Sua

doi:10.1038/s41467-025-64909-7

Download PDF

Article
Open access
Published: 12 November 2025

FUS nanoclusters are a distinct state within the dilute phase

Yingda Ge^1,2^na1,
Tapas Paul^2,3^na1,
Margarita Gordiychuk⁴,
Nilimesh Das ORCID: orcid.org/0000-0003-2081-0177^2,3,
Yaojun Zhang^1,4 &
…
Sua Myong ORCID: orcid.org/0000-0001-9098-3423^1,2,3

Nature Communications volume 16, Article number: 9956 (2025) Cite this article

4865 Accesses
1 Citations
16 Altmetric
Metrics details

Subjects

Abstract

Liquid-liquid phase separation of Fused in sarcoma (FUS) has been extensively studied due to its critical role in various physiological and pathological processes. However, the mechanism of this process remains poorly understood. Here we show that by preparing pre-cleaved FUS, we enable kinetic measurement of phase separation from the onset. We identify a distinct state within the dilute phase using a combination of biophysical measurements including dynamic light scattering, fluorescence correlation spectroscopy, and an optimized single-cluster fluorescence imaging assay. This state is termed nanoclusters. They are hundreds of nanometers in diameter and exhibit instantaneous formation far below the saturation concentration, sharply arrested coarsening, rapid mixing, high exchange rate, and resistance to hexanediol-induced dissolution. This state exhibits properties that distinguish it from condensates. Our findings define the nanocluster as a prominent intermediate in the FUS phase separation pathway. This understanding paves the way for further experiments and theories to explore the molecular mechanisms underlying FUS and other phase-separating proteins.

Nanoscale profiling of evolving intermolecular interactions in ageing FUS condensates

Article Open access 29 September 2025

Fibril formation and ordering of disordered FUS LC driven by hydrophobic interactions

Article Open access 25 May 2023

Single-droplet surface-enhanced Raman scattering decodes the molecular determinants of liquid-liquid phase separation

Article Open access 28 July 2022

Introduction

Fused in sarcoma (FUS) is a highly disordered RNA-binding protein that plays important roles in multiple physiological and pathological processes by liquid-liquid phase separation (LLPS). FUS is enriched in multiple nuclear membrane-less organelles^{1,2,3,4,5,6,7}. In DNA damage and transcriptional stress, FUS is enriched in nucleoli^8,9 and other membrane-less organelles, such as splicing condensates and paraspeckles^3,10, where FUS plays critical roles in multiple stages of DNA damage repair^4,5,11 and RNA regulation^3,12. FUS is required for paraspeckle formation in amyotrophic lateral sclerosis (ALS)^10,13. In response to cellular stress, FUS can be recruited to the stress granules^14,15 to protect critical cellular components¹⁶ and post-transcriptional regulation¹⁷ and protein synthesis¹⁸. The failure to regulate the dynamics of stress granules is linked to specific FUS mutations implicated in neurodegenerative diseases¹⁵.

Over the past decade, numerous studies reported the basic parameters of FUS condensates, such as the saturation concentration, C_sat^19,20, viscoelasticity^21,22, and partition coefficient²³. Based on these properties and the structural information^24,25, theory and simulation^26,27,28 of FUS LLPS have also been proposed.

Compared to the study on FUS condensates, much less attention has been given to the behaviors of FUS in the dilute phase or sub-saturated conditions. However, this condition is commonly found in the cell without stress^4,29. Recent studies revealed that even in subsaturated conditions, FUS exists not as homogeneous monomers and small oligomers but as varying sizes of nanoscale clusters (nanoclusters)^30,31,32. These nanoclusters are observed not only in carefully controlled experiments in vitro, but also inside living cells, where broad distributions of cluster sizes have been reported^33,34. Traditional nucleation theory, however, does not explain the nanoclusters^30,35, suggesting that more complex pathways may underlie the formation of phase-separated condensates^36,37. While evidence supports the existence of FUS nanoclusters, a detailed quantitative study on the physical and chemical properties is still lacking.

To investigate the development of FUS nanoclusters and their material properties, we prepared tag-free FUS based on the affinity protein liquid chromatography^19,20,22,38, enabling precise quantification of the protein concentration and tracking the early stages of nanocluster development. Using biophysical methods, including Dynamic Light Scattering (DLS), Fluorescence Correlation Spectroscopy (FCS), and our optimized “Single Cluster Pull-down” (SiCluP) method, we tracked and quantified multiple properties of the nanoclusters, including their critical concentrations, sizes, rate of assembly, diffusion coefficient, exchange rate of the components, etc. We present FUS nanoclusters as a distinct liquid-like intermediate with ~300–500 nm diameter that forms above 200 nM FUS concentration, primarily driven by electrostatic interactions. As a fast-assembling constituent of a dilute phase, FUS nanoclusters could be considered a different phase distinguished from the condensates.

Results

Pre-cleavage enables probing the early development of FUS LLPS

Due to the aggregation-prone nature of FUS-related proteins, they are often expressed with a solubility tag, such as GST or MBP. Therefore, in vitro condensate assays are initiated by applying a protease, such as Tobacco Etch Virus (TEV) protein, to cleave the solubility tag^{19,20,22,38,39}. The protease reaction, however, is slow and inefficient, interfering with detecting early assembly intermediates. To test the kinetics and efficiency of TEV-cleavage on FUS, we terminated the TEV reaction at different time points after initiation. We visualized the cleaved vs. uncleaved FUS by running a gel. The result displays that one hour of TEV incubation cleaves only about half of the substrates (Fig. 1A–C), making it challenging to measure concentration and detect intermediates during LLPS (Fig. 1D).

**Fig. 1: Optimized pre-cleavage FUS (PC-FUS) protocol.**

To overcome the shortcomings, we have modified our protocol for protein preparation. After eluting the MBP-tagged FUS from the Ni-NTA column, we optimized the TEV cleavage buffer by adding KCl and urea to minimize the FUS aggregation while keeping sufficient TEV protease activity. After 8 hours of incubation with TEV, we applied the cleavage reaction mixture to the Ni-NTA column, which captures all 6XHis-tagged species, including the 6XHis-MBP, 6XHis-MBP-FUS (uncleaved) and 6XHis-TEV. Therefore, the desired product, untagged FUS, is expected to flow through the column (Fig. 1E). We demonstrate by SDS-PAGE that this method successfully generated the untagged FUS with high purity (Fig. 1F).

We measured the size of untagged FUS by Dynamic Light Scattering (DLS), which revealed ~10 nm diameter in 0.5–5 μM concentration range, indicating primarily monomer status of FUS when kept in the storage buffer with high salt (Fig. 2A). Next, we compared the conventional TEV-Cleaved “TC” FUS and the Pre-Cleavage “PC” FUS by titrating both versions of FUS and measuring by DLS. PC-FUS was diluted to a desired concentration in a low-salt buffer. The PC-FUS led to a significantly lower C_sat than the TC-FUS (Fig. 2B). For simplicity, we defined any cluster size of 1 μm diameter as the cut-off for the C_sat. The C_sat for TC-FUS was 2-5 μM, in agreement with previous studies^20,30, while the PC-FUS was 0.5-1 μM. Moreover, the PC-FUS immediately formed a ~ 600 nm peak since the lag-time required for TEV cleavage is no longer rate limiting (Fig. 2C). The nanoscale (<1 μm) species, termed nanoclusters^30,35, could not be detected by a wide-field microscope but tracked by DLS (Fig. 2D). As a comparison to this species, the smaller populations that cannot be distinguished from monomers in DLS are categorized as “monomers” while the micron-sized condensates were categorized as “condensates” in our study. We occasionally observed an additional peak at 500 nM, near the C_sat, which we attribute to transient fluctuations or fitting variability rather than a distinct population (Supplementary Fig. S1). Together, the tag-free FUS and the precise measurement tool that enables sub-micron-sized nanocluster detection unveil the early stages of FUS LLPS.

**Fig. 2: Characterization of re-cleaved FUS (PC-FUS).**

FUS nanoclusters are distinct from condensates in size and kinetics

The C_sat for FUS is in the 2–5 μM range^19,20. To examine the status of FUS below the C_sat, we titrated both TC- and PC-FUS concentrations 0.2–2 μM and performed DLS measurements at 2, 5, 10, 30, 60, 120, and 180 minutes. For TC-FUS, the minutes indicate the time after TEV protease addition, whereas for PC, they represent the time after the PC-FUS dilution in 100 mM NaCl, which is comparable to the cytosolic monovalent salt condition (40, 41) and the physiological pH of 7.4 (42–43). PC-FUS exhibits nanocluster when FUS is as low as 200 nM. In contrast, TC-FUS assembles into nanocluster only above 1 µM FUS concentration (Fig. 3A, B). Above the C_sat, i.e., at 2 µM, both TC- and PC-FUS assembled into nanoclusters and slowly transitioned into micron-sized condensates (Fig. 3A, B).

**Fig. 3: TEV-cleaved FUS (TC-FUS) vs. pre-cleaved (PC-FUS) titration.**

Besides the C_sat and size, nanoclusters form significantly earlier than the condensates for all conditions (Fig. 3B). For PC-FUS, the transition from monomers to nanoclusters was too fast to be detected by DLS. Moreover, when the FUS concentration increased from 200 nM to 500 nM, the size of nanoclusters also increased. Above 1 μM, nanoclusters persisted while the micron-sized condensates gradually emerged.

Due to the protein concentration (>200 nM) required for DLS and the detection bias of DLS toward large particles, 44 we employed an orthogonal approach: Fluorescence Correlation Spectroscopy (FCS). For all FCS measurements, we doped in 20 nM Cy3-labeled FUS to unlabeled FUS and incubated for one hour before taking the measurement. The Cy3 alone displayed an extremely fast diffusion (gray, Fig. 3C), while 10-100 nM PC-FUS generated a clear right shift (yellow, Fig. 3C), representing the FUS monomers. At and above 200 nM, a slower diffusing species by order of magnitude emerged and dominated the signal (orange, Fig. 3C). Above 1 μM, the second critical concentration, the diffusion slowed down by another order of magnitude (blue, Fig. 3C, D). The diameter (~10 nm) obtained for the 10-100 nM PC-FUS is expected for a monomer FUS while the two regions of larger diameters (100 ~ 1000 nm and >1000 nm) likely corresponded to nanoclusters and condensates, respectively (Fig. 3D, Supplementary Fig. S2). Overall, the FCS and DLS results agree with each other in the scale of magnitude, suggesting that nanoclusters that form above the first critical concentration of ~200 nM FUS exist as a stable state, distinguishable from the micron-sized condensate that forms above the second critical concentration of ~ 1 μM.

SiCluP (Single Cluster Pull-down) of single FUS nanoclusters

Next, we sought to capture individual nanoclusters onto a microscope slide to analyze their biophysical property. We used the Total Internal Reflection Fluorescence (TIRF) microscope to visualize FUS nanoclusters. To assemble nanoclusters that can be tethered to our single-molecule surface, we mixed in MBP-FUS to PC-FUS with 1% Cy3-PC FUS so that the nanoclusters that contain MBP-FUS can be pulled down to an anti-MBP-antibody-coated surface for fluorescence imaging (Fig. 4A, see Methods for details).

**Fig. 4: Single Cluster Pull-down (SiCluP) assay.**

We applied increasing concentrations of the PC-FUS mixture (Fig. 4A–E). Below 100 nM, the fluorescent signal was weak, and most molecules remained monomeric, as confirmed by single-step photobleaching at 10 nM PC-FUS (Supplementary Fig. S3). Once the concentration reached 200 nM, the nanocluster threshold, clearly visible bright fluorescence signals appeared on the imaging surface. Bleaching of these signals became continuous, and individual steps were unresolvable due to fluorophore overlap, consistent with the formation of multimeric nanoclusters (see Supplementary Fig. S3). The signals became markedly brighter (50-350 nM) and their size increased progressively (350 nM–1 μM) as a function of concentration (Fig. 4B). Under condensate-forming conditions (500 nM and 1 μM), the size of the fluorescent spots continued to increase throughout the three hours of measurement (Fig. 4C–E, Supplementary Figs. S4 and S5). Quantitative analysis of the mean diameters of condensates and the early growth of nanoclusters as a function of time reveals that the coarsening dynamics follow a 1/3 scaling law, consistent with diffusion-limited coarsening mechanisms, i.e., Ostwald ripening^40,41 and/or Brownian motion-induced^42,43 coalescence (see Supplementary Information for details).

In contrast to the continuous growth of condensates, under nanocluster-forming conditions (200 nM and 350 nM), the nanoclusters initially coarsened over time but plateaued at an equilibrium diameter of around 500 nm within 30 minutes (Fig. 4C, E), suggesting that the growth of nanoclusters is arrested. Thus, despite the limitation of SiCluP for accurate size determination, this method reliably distinguished the kinetic differences between nanoclusters and condensates. Although the absolute values for the critical concentrations varied slightly from the diffusion-based measurements, nanoclusters and condensates are consistently discernible by their coarsening patterns. The direct visualization of single nanoclusters further validates the distinct population of nanoclusters as major species that form within a narrow concentration range (~250–500 nM).

Next, we sought to measure the nanocluster assembly in real-time using the SiCluP. We first immobilized a few Cy3-FUS to find the focal plane and applied varying concentrations of the FUS mixture while measuring the fluorescence signals. This enables the tracking of the real-time assembly of FUS nanoclusters or condensates from the beginning (Fig. 4F). The assembly rate for the condensate-forming concentrations (500 nM and 1 μM) was higher than that for the nanocluster-forming conditions (200 nM and 350 nM) (Fig. 4G, H). This higher rate of assembly is consistent with the larger nanoclusters forming at 500 nM and 1 μM in the initial 2-5 minutes, revealed by DLS measurement (Fig. 3A). Our observed assembly dynamics can be explained by the initial stage of droplet formation driven by diffusion-limited growth, which depends on the initial protein concentration and C_sat⁴⁴. Based on the DLS and SiCluP results, we propose that nanoclusters inherently differ from condensates in that they undergo another phase that limits assembly to the phase of condensates.

Single FUS nanoclusters exhibit liquid-like properties

To test whether FUS nanoclusters exhibit liquid-like behavior, we first examined their ability to exchange components with the surrounding solution using the SiCluP assay. We prepared two batches of nanoclusters labeled with either Cy3 or Cy5 and monitored their mixing behavior (Fig. 5A–D). Cy3-labeled nanoclusters were first immobilized, followed by the flow-in of Cy5-labeled nanoclusters. Within 1-2 minutes, Cy5 fluorescence colocalized with Cy3 clusters. This mixing was also observed when the labeling order was reversed. Given the rapid exchange kinetics⁴³, we infer that the observed mixing likely results from the exchange of molecules between labeled nanoclusters through the dilute phase (Supplementary Fig. S6), rather than from fusion of the nanoclusters (see Supplementary Information for details).

**Fig. 5: Liquidity and molecular exchange dynamics of FUS nanoclusters revealed by SiCluP and FRAP.**

To test if such mixing arises from simple partitioning rather than true molecular exchange, we performed two additional experiments (Fig. 5E–K). First, we monitored the Cy5 fluorescence intensity over time following the addition of Cy5-labeled nanoclusters to surfaces that already contained Cy3-labeled nanoclusters. The signal increased rapidly and plateaued, conforming to a biphasic exponential profile (Fig. 5E–G and Supplementary Fig. S7A). The fast component (τ₁ ~ 4–7 s) likely reflects the initial partitioning of Cy5-labeled FUS to the surfaces of Cy3-labeled nanoclusters, caused by a temporary disruption of nanocluster equilibrium. The slower component (τ₂ ~ 70–100 s) corresponds to molecular exchange between nanoclusters and the surrounding solution until equilibrium is re-established (see Methods for details). In the second experiment, we introduced unlabeled nanoclusters into Cy3-labeled nanoclusters to avoid dual-color overlap and FRET-related complications. Here, we observed a gradual decrease in Cy3 fluorescence over time, consistent with the replacement of labeled FUS molecules by unlabeled ones from the solution (Fig. 5H–K and Supplementary Fig. S7B). This decay followed a single-exponential curve with a similar time constant (~70–90 s), further supporting the conclusion that nanocluster components dynamically exchange with the surrounding phase. Although initial partitioning may also occur in this case, it is not directly observable due to the monitoring of Cy3 signal loss from pre-immobilized clusters. Together, these experiments demonstrate that FUS nanoclusters continuously exchange molecular components with the surrounding solution and exhibit liquid-like properties on both short and long timescales.

To further quantify the exchange dynamics and molecular mobility within nanoclusters, we developed a nanocluster-suited FRAP (Fluorescence Recovery After Photobleaching) assay, which is conventionally used to determine the liquidity of condensates^45,46,47. We adapted it to our surface-tethered SiCluP system. Nanoclusters were periodically photobleached, and fluorescence recovery was monitored by varying the recovery interval between each event (Fig. 5L and Supplementary Fig. S7C). This modified FRAP approach utilizes intermittent excitation to minimize photodamage, allowing for repeated recovery measurements from the same cluster. FRAP recovery curves were generated for nanoclusters and condensates at various concentrations and fitted with exponential functions to extract exchange rates and apparent diffusion coefficients (Fig. 5M–T). The recovery times (~40–90 s) are slightly shorter than the exchange timescales observed in perturbation-based assays above, likely reflecting re-equilibration after component addition in these assays, whereas FRAP was performed under steady-state conditions.

When normalized by surface area²⁸, nanoclusters’ exchange rates exceeded 3000 molecules/(µm²·s), comparable to condensates (see Supplementary Table S2), indicating that nanoclusters and condensates exhibit similarly fast molecular exchange despite differences in size and assembly mechanism. We also quantified the apparent diffusion coefficients (D_app) for nanoclusters and condensates using the standard FRAP relation D_app = R² / (π²·τ), where R is the radius of the bleached nanoclusters/condensates and τ the recovery time. The calculated D_app values ranged from 1.6 × 10⁻⁴ to 2.2 × 10⁻⁴ µm²/s across FUS concentrations from 200 nM to 1000 nM (see Supplementary Table S2). We note that the recovery of a fully bleached droplet involves both internal mixing of molecules and exchange with the surroundings; whichever process is slower limits the recovery time. If internal mixing is the rate-limiting factor here, Dapp would reflect the molecular mobility within the nanoclusters/condensates. The similar values of exchange rates and D_app between nanoclusters and condensates, as indicated by the FRAP measurements, thus suggest that they exhibit comparable molecular mobility and liquid-like dynamics.

Nanoclusters’ tolerance to mechanical and chemical disruption

What factors regulate the formation and internal molecular interaction network of nanoclusters vs. condensates? We introduced various physical or chemical disturbances to nanoclusters and condensates to elucidate the molecular network of their constituents. For consistency, we used 1 µM of PC-FUS as this condition has both nanoclusters and condensates (Fig. 3B).

We perturbed the protein mixture mechanically by applying pipetting and centrifugation after incubating 1 µM of PC-FUS for three hours, prior to mechanical disruption. Firstly, the mechanical shear force applied by 30 seconds of pipetting did not disrupt the micron-size condensates and nanoclusters (Fig. 6A, top).

**Fig. 6: Perturbation assays of FUS nanoclusters and droplets.**

Next, we applied mechanical disruption by centrifuging 21,000 x g for 20 minutes. The DLS measurements taken before and after centrifugation revealed the disappearance of the condensates while leaving the nanoclusters in solution (Fig. 6A, bottom). As previous studies have used this level of centrifugation to separate the dense and dilute phases, 52 our result reveals that nanoclusters remain in the supernatant, i.e., the dilute phase, after spinning down the dense phase, i.e., the condensates. This result demonstrates that nanoclusters are the constituent of the dilute phase. In addition, the absorbance at 280 nm decreased (Supplementary Fig. S8), confirming that centrifugation separated the condensates from the solution but was unable to either sediment or disperse the nanoclusters.

Next, we challenged both species to several chemicals commonly used to disturb LLPS. First, we tested the effect of ionic perturbation by adding varying concentrations of NaCl, which can prohibit LLPS by screening the Coulomb interactions using ionic screening^47,48. The titration started from 100 mM NaCl, and our buffer conditions were used for in vitro LLPS assays^19,22. The nanocluster and condensate remained unchanged up to 260 mM, but above 300 mM, both condensates and nanoclusters dissolved to become monomers (Fig. 6B). Even with finer titrations around 300 mM NaCl (Supplementary Fig. S9), no clear boundaries were found to distinguish the dispersion of nanoclusters and condensates. These results indicate that nanoclusters and condensates are destabilized at the same level of ionic screening. Thus, the Coulomb interactions shielded by ionic screening contributed similarly to stabilizing nanoclusters and micron-sized condensates. However, high NaCl concentrations, instead of delaying the formation of droplets, accelerate the transition from the nanocluster state to the condensate state. This finding implies that another type of interaction may also be involved in governing the kinetics during this transition.

So, we applied 1,6-Hexanediol, which disrupts hydrophobic interactions to perturb condensates in vivo and in vitro. While 0.1% of 1,6-hexanediol did not change the formation of both condensates and nanoclusters, 0.2% of 1,6-hexanediol selectively disrupted the condensates (Fig. 6C). Nanoclusters persisted even up to 2% of 1,6-hexanediol. Above 2%, the DLS measurement was not feasible due to the severe scattering signal from the high level of 1,6-hexanediol. The selective dissolution of condensates reveals that condensates, but not nanoclusters, are stabilized by hydrophobic interactions. Furthermore, condensate dissolution at high 1,6-hexanediol concentration leads to the formation of nanoclusters rather than monomers, emphasizing the stability of nanoclusters in these conditions. As a comparison to the effect of NaCl, which disrupts hydrophobic effects and decouples the stable states of nanoclusters and droplets, also delays the development of droplets, indicating that the loss of hydrophobic effects slows down the kinetics of droplet formation.

As a positive control, we tested sodium dodecyl sulfate (SDS), a potent protein denaturant. SDS started disrupting nanoclusters and micron-sized condensates at the same concentration, 0.005% (w/v) (Fig. 6D). Therefore, by disrupting both Coulomb interactions and hydrophobic interactions, both nanoclusters and micron-sized condensates are dissolved with the same strength of disruption.

Here, we demonstrate that, unlike the micron-sized condensates that rely on both Coulomb and hydrophobic interactions to stabilize, nanoclusters are as sensitive to the disruption of Coulomb interactions but significantly less impacted by the disruption of hydrophobic interactions (Fig. 6E). These results imply that these two species’ driving forces and inner interaction networks could be inherently different. This finding may help explain why the size of nanoclusters is more confined, and they are more resistant to mechanical and chemical disruptions.

RNA regulates the size and kinetics of nanoclusters

The LLPS of FUS is regulated by RNA^19,22,39, which has been demonstrated as an RNA-dependent reentrant LLPS behavior^49,50,51. PC-FUS co-condenses with unstructured 50-nucleotide poly-Uracil RNA. (Fig. 7A). Based on these findings, we inquired about the potential impact of RNA on nanoclusters. We applied 0 to 10 µM of U50 (50-nt of poly-Uracil), a standard unstructured RNA to 1 µM of PC-FUS. The result shows that even applying low concentrations of U50 RNA (0.25 and 0.5 µM) can accelerate condensate formation from nanoclusters (Fig. 7B, C). The size of the nanoclusters and condensates increased upon the addition of 0.25 µM U50. However, upon further increasing the U50 to 0.5 µM-10 µM, the sizes of both species decreased as a function of increasing U50 concentration, displaying the reentrant impact of RNA (Fig. 7C, D).

**Fig. 7: Dual effects of 50-nucleotide poly-Uracil RNA (U50) on nanoclusters and droplets.**

To monitor the RNA dependence on FUS nanoclusters without the ratio metric variation, also based on the previous work on a one-to-one ratio of FUS binding to U50 RNA¹⁹, we co-varied both PC-FUS and RNA concentrations while maintaining the FUS: RNA ratio at 1:1. When the concentrations varied 200-500 nM, the size of the nanoclusters were consistently larger in the presence of equimolar RNA (Fig. 7E).

We also tested two ALS-linked FUS mutants that displayed aberrant condensation in previous studies^19,22. Under the conditions containing a one-to-one molar ratio of U50 RNA, while R244C still showed similar nanocluster sizes to wild-type, G156E mutant formed significantly smaller nanoclusters (Supplementary Fig. S10).

Discussion

In this study, we unveiled distinct nanocluster species that exhibit unique physical and kinetic properties with the optimized protein purification methods and the optimized TIRFM-based Single Cluster Pull-down (SiCluP). The optimized pre-cleavage protocol enabled the examination of FUS in its pure form, i.e., without the interference of the MBP tag. Based on the results obtained by DLS and FCS, we found that PC-FUS’s condensate (diameter> 1 μm) formation occurs at a significantly lower FUS concentration and at a higher rate than MBP-FUS. This difference indicates that the protease digestion required for the MBP cleavage limits the accuracy of measuring the kinetics of FUS LLPS. Therefore, the pre-cleavage protocol was critical in enabling the kinetic measurements.

Additionally, the PC-FUS enabled us to distinguish and examine the intermediate-state nanoclusters systematically. Although previous research reported the nanocluster formation in varying contexts^30,31,32,35, the nanoclusters were rarely compared with the condensates regarding kinetics and physical properties. In this work, we determined that the critical concentration for nanocluster formation is approximately 200 nM under physiological conditions of salt and pH. The transition from homogeneous monomers to nanoclusters occurs abruptly upon reaching the concentration threshold (200 nM). In contrast, the C_sat of FUS LLPS is above 1 µM, and the development of condensates continued even after 2 h after initiation. Even under condensate-forming conditions, nanoclusters are detected from the beginning, preceding the appearance of the condensates (>1 µm), and persist even after three hours. These results indicate that nanoclusters are inherently distinct from condensates, characterized by their size, critical concentration, and formation kinetics. Thus, nanoclusters are a unique “phase” well distinguished from condensates.

Enabled by SiCluP, we can track the growth of nanoclusters and test their material properties while visualizing them under optical microscopes. We validated the two-step transition of FUS from nanoclusters to condensates using TIRFM by titrating the FUS concentration. Taking advantage of the single-molecule resolution of the TIRFM, we measured the assembly of individual FUS nanoclusters under different FUS concentrations. The initial assembly rate differed significantly between the nanoclusters and condensates, i.e., it increased substantially when the FUS concentration exceeded the critical saturation concentration (C_sat) at ~500 nM. Using the same platform, we performed FRAP for nanoclusters and condensates. The exchange rates between the nanocluster/condensate and monomer/oligomer phases suggested a monomer concentration of 74 nM in equilibrium with nanoclusters and condensates, consistent with the threshold concentration for nanocluster formation (see Methods and Supplementary Information for details). Using SiCluP, we measured the assembly, size, coarsening, and fluidity of nanoclusters, which other methods cannot directly capture. Based on our SiCluP results, we observed that nanoclusters rapidly exchange their components with the surrounding dilute phase. How can nanoclusters persist without coarsening below Csat and coexist with condensates above C_sat?

To explain this behavior, we propose a conceptual model that incorporates both free energy and kinetic considerations (see Supplementary Fig. S11 and Supplementary Information for details), building on previous models that consider multiple states on the free-energy landscape and multiple steps along the pathway of LLPS^34,36,37,52. At low concentrations, nanocluster and condensate states are thermodynamically inaccessible, and only monomers and small oligomers are stable. As the concentration increases, nanoclusters become metastable intermediates, while condensates remain disfavored. Only above C_sat does the condensate state become energetically favored, enabling condensate formation and their co-existence with monomers and nanoclusters. Kinetically, the cooperative nature of these transitions means that flux toward condensates is negligible at intermediate concentrations below the critical saturation concentration (C_sat), thereby reinforcing the stability of nanoclusters in this regime. This would explain how nanoclusters can exhibit internal fluidity and dynamic exchange without merging into condensates until phase-separating conditions are met. This would also explain the co-existence of nanoclusters with condensates above C_sat^31,33,34, as nanoclusters represent a metastable state in equilibrium with both monomers and condensates. Together, our observations and interpretations, in addition to prior studies on the pathway of FUS assembly^36,37, contribute to an evolving understanding of LLPS, where nanoclusters represent a distinct and potentially stable intermediate state along the phase transition pathway.

The distinct physical and kinetic parameters we observed in nanoclusters led us to hypothesize that the intrinsic inner molecular interaction network of nanoclusters may differ from the condensates. Both nanoclusters and condensates dissolved into monomers at similar salt concentrations (>300 mM NaCl), suggesting the importance of electrostatic interactions for both species. In contrast, the hydrophobic disturbance by 1,6-hexanediol resulted in the disruption of condensates at 0.2%, while nanoclusters persisted even when 1,6-hexanediol reached 2%. These results indicate that electrostatic interactions primarily hold together nanocluster but not hydrophobic interactions, while both are critical for condensation assembly^53,54. Considering the difference in strength and effective range of the two types of interactions, excluding hydrophobic interactions might be one reason for the nanoclusters’ limited size. Compared to the condensates, the nanoclusters have fewer non-specific hydrophobic effects and more electrostatic interactions, such as the pi-cation interactions between the tyrosine-rich, low-complexity domains and Arginine-rich RNA binding domains⁵⁵. One plausible scenario is that the nanoclusters fuse via hydrophobic interactions to assemble into condensate. Our proposal aligns with previous work³⁰ which suggests that different types of interactions govern the formation of clusters and condensates, respectively. Extending these understandings, the lower C_sat in the presence of RNA could be rationalized as the RNA expands the nanoclusters’ size by partly shielding the electrostatic interactions while strengthening the hydrophobic interactions. Further test on RNA targets and FUS mutants can be conducted in upcoming studies. Based on the findings in this work, more physiology-related features of nanoclusters can be possibly unveiled.

The “two-step transition” behavior observed in the FUS liquid-liquid phase separation process adds to the complexity of condensate formation. In this work, we confirm the existence of FUS nanoclusters and provide the detailed physical and chemical properties that differentiate FUS nanoclusters from condensates, which was greatly facilitated by the SiCluP platform. Although we do not present a precise molecular mechanism for nanocluster formation and their transition to condensates, we are paving the way toward a deeper understanding of liquid-liquid phase separation. Our findings may help reshape the theory of LLPS and decipher in vivo nanoclusters that may operate under similar principles.

Methods

Plasmid preparation

The his-tagged MBP-FUS is encoded by the E.coli expression vector generated by Genscript, which was also used in many of our previous work^{11,19,22,38,39,56}. The design of the target protein, starting from the N-terminus, includes a 6XHis-tag, MBP, TEV-protease site, and then FUS.

Protein purification

NEB BL21(DE3) cells containing the expression vectors were cultured in 500 mL of LB with 50 mg/L of Kanamycin until the OD 600 reached 0.4. Then, the expression was induced by 0.25 mM of IPTG at 30 °C for 2 h. The cell pellets were resuspended and lysed by sonication in the FUS Lysis Buffer (1 M KCl, 1 M Urea, 50 mM Tris-HCl, pH 7.4, 10 mM imidazole, 1.5 mM β-mercaptoethanol, 1% NP-40, 5% glycerol, 5 mg/L of culture RNase A, and one protease inhibitor cocktail tablet). After removal of the insoluble portion of the lysate by centrifuging under 14,000 × g for 30 minutes and filtering with the 0.22 µm filter, the sample was applied to a Histrap FF Crude 5 mL column using the AKTA Pure 25 system and eluted with an imidazole gradient. (Elution Buffer: 1 M KCl, 1 M Urea, 50 mM Tris-HCl, pH 7.4, 1.5 mM β-mercaptoethanol, 5% glycerol, 500 mM imidazole).

Cleavage of the MBP-tag

The eluted MBP-FUS first underwent a buffer exchange process using a 4 mL AmiconUltra 30 K Centrifuge filter to decrease the imidazole concentration to less than 50 mM so that the TEV protease could function more efficiently. (FUS Binding Buffer used for buffer exchange: 1 M KCl, 1 M Urea, 50 mM Tris-HCl, pH 7.4, 1.5 mM β-mercaptoethanol, 5% glycerol, 10 mM imidazole). After filtering the aggregates with a 0.22 µm filter, TEV protease (NEB P8112S, 2 µL/mL of sample) was added. After being incubated at room temperature for 6 hours and going through a 0.22 µm filter, the sample was applied to another HisTrap column, and the flow-through was collected as the final cleaved product. After adding Glycerol to make the final glycerol v/v concentration 25%, the Pre-Cleaved FUS was aliquoted and flash frozen by liquid-nitrogen and stored under −80 °C.

Protein labeling

For the FCS measurements and condensate imaging, FUS needs to be labeled with fluorophores. In this study, we used NHS-ester conjugated Cy3-sulfo for FUS labeling so that no further mutations or modifications needed to be included in the protein sequence. FUS was first transferred into the FUS labeling buffer (1 M KCl, 1 M Urea, 100 mM NaHCO₃, 5% Glycerol, and 280 µM β-mercaptoethanol in H₂O) to minimize the remaining Tris, which NHS-ester also targets. After adding dye (5X to 10X of the molar protein concentration), the mixture was incubated in the dark for one hour. Then, the product was transferred back to the FUS binding buffer, and the extra dye was removed. All the buffer exchange processes were performed with the Zeba desalting columns, 7 K MWCO. After adding extra glycerol, making the final concentration 25%, and being flash-frozen with liquid-nitrogen, the labeled protein was stored under −80 °C and ready to use.

Dynamic light scattering measurements

Dynamic Light Scattering (DLS) measurements were conducted with a DynaPro Plate Reader II from Wyatt Technology. To avoid being interfered with by unrelated impurities, all of the buffer used in DLS measurements had to be filtered by 0.22 µm filters right before the experiments, and pipette tips also needed to be individually packed and newly opened before use. If not explicitly mentioned, the buffer conditions for DLS assays were 100 mM NaCl, 50 mM Tris-HCl pH 7.5, 1 mM EDTA, and 1 mM DTT, which in this work was also named condensate buffer.

To initiate measurements, FUS was directly added from the highly concentrated stock to the final condition (at least 50-fold dilution was required). This approach avoided potential condensation during the dilution process and minimized the effects of salt and urea in the stock. After preparation in low-binding tubes, the sample was transferred into a Corning Low-binding 96-well plate for the measurements. Measurements were performed in the presence and absence of unstructured 50-nucleotide poly-Uracil RNA. The results acquired by the DynaPro Plate Reader II were further analyzed using Dynamic7 software provided by Wyatt Technology.

For mechanical disturbance introduced by pipetting, the procedure is as follows: Pipetting 100 µL from the total of 135 µL samples, using the large orifice P200 pipette tip, completing 60 times within 30 seconds.

Fluorescence correlation spectroscopy measurements

Fluorescence Correlation Spectroscopy (FCS) measurements were performed using a MicroTime 200 system (PicoQuant) equipped with a 532 nm continuous-wave laser for excitation. The laser beam was focused into the sample using a 60 x water-immersion objective (NA 1.2), with the sample positioned on plastic sample chambers (μ-Slide, ibidi) over the objective mounted on a piezo stage. For all of the measurements, the light was focused at 30 μm from the slide surface. The average power at the back aperture of the objective was maintained around 5 μW. Fluorescence photons were separated from scattered laser light with a triple-band filter, further filtered with a 582/64 nm band-pass filter, and detected with single-photon avalanche diodes. The intensity trace of the fluorescence signal as a function of time is then subject to the correlation analysis to generate the autocorrelation curve. The signal series were extracted and analyzed using a Python script adopted from previously published work^42,43. Measured correlation functions were fitted with a model for translational diffusion through a 3D Gaussian-shaped confocal volume:

$$G\left(\tau \right)=\frac{1}{{\left(\sum {N}_{i}\right)}^{2}}\sum {N}_{i}{D}_{i}(\tau ),\left(i=1,\,2\right)$$

(1)

$${D}_{i}\left(\tau \right)={\left(1+\frac{\tau }{{{\tau }_{D}}_{i}}\right)}^{-1}{\left(1+{w}^{2}\frac{\tau }{{{\tau }_{D}}_{i}}\right)}^{-\frac{1}{2}},\,\left(i=1,\,2\right)$$

(2)

where $\tau$ is the lag time, ${N}_{i}$ is the number of corresponding fluorescently labeled species in the confocal volume, ${{\tau }_{D}}_{i}$ is the translational diffusion time of that species, and $\omega$ is the ratio of the axial and lateral radii of the confocal volume.

For calibration, we used free Cy3-dye and collected a series of autocorrelation curves. Then, by fixing the diffusion coefficient D to the known diffusion coefficients of the probes (241 ${{\mu m}}^{2}/s$), we performed the fittings to determine the structural parameter $({\omega }^{2})$ of the confocal volume. D is inserted in the fitting equation through the relation $\omega=\frac{l}{r}\,$ and $D=\frac{{r}^{2}}{4{\tau }_{D}}$. In the subsequent fittings, ${\omega }^{2}$ is kept fixed, and other parameters are allowed to vary.

We used the following equation to calculate the hydrodynamic radius of any species (${R}_{H}^{i}$).

$${R}_{H}^{i}={R}_{H}^{{Cy}3}\times \frac{{\tau }_{D}^{i}}{{\tau }_{D}^{{Cy}3}}$$

(3)

The hydrodynamic radius of free Cy3 dye in water is taken as 0.81 nm⁵⁷.

Buffer conditions, if not explicitly mentioned, were used the condensate buffer. On top of the regular samples, 10 nM of Cy3-labeled PC-FUS was doped into each sample as a fluorescent reporter. All the measurements are taken for around 600 seconds. Every measurement was repeated three times.

Condensate imaging

Condensate imaging with a Nikon Ti-2 widefield fluorescent microscopy system. If not specifically mentioned, the samples were prepared in the condensate buffer with 10 nM of fluorescent-labeled FUS doped in and the fluorescent reporter. Samples were added to a 96-well plate with a lid and untreated glass bottom for imaging. The images and movies were further analyzed by a set of MATLAB scripts adopted from previously published work¹⁹ and modified to fit our purpose in this work.

Single cluster pull-down (SiCluP) assay

All single-molecule and single-nanocluster capture and visualization were performed using prism-type total internal reflection fluorescence (TIRF) microscopy with standard sample preparation and imaging protocols^58,59. We followed a well-optimized protocol to prepare polyethylene glycol (PEG) and 1-2% biotin-PEG-passivated quartz slides and coverslips^59,60. NeutrAvidin (1 mg/mL) was flowed into the microfluidic sample chamber, incubated for 2 minutes, and then washed out the unbound NeutrAvidin. Biotin-conjugated anti-MBP antibody (400 times dilution) was immobilized on the PEG-passivated slide surface through biotin-NeutrAvidin interactions.

A mixture of MBP-FUS (50 nM), PC-FUS (200 nM), and Cy3-labeled PC-FUS (1 nM) in a buffer containing 50 mM Tris-HCl (pH 7.5), 100 mM NaCl, 1 mM EDTA, and 1 mM DTT was then applied to the anti-MBP antibody immobilized slide surface. The fluorescence signal from Cy3-labeled PC-FUS was visualized due to nanocluster formation on the slide surface. Field-of-view images were recorded as short movies (~2 seconds) at different imaging areas over various intervals.

The exact measurements were performed at different PC-FUS concentrations while keeping all other components constant. The recorded movies were processed using a standard mapping file in Interactive Data Language (IDL), and the nanocluster size was calculated using a custom-written MATLAB code. All SiCluP-related experiments were independently repeated three times. In each round, more than ~2000 nanoclusters were analyzed for size measurements, and over ~35 individual traces were used to quantify nanocluster assembly rates, partitioning behavior, exchange dynamics, and FRAP recovery.

Nanocluster assembly

Different concentrations of PC-FUS (200 nM and above) were added to a mixture containing a fixed concentration of MBP-FUS (50 nM) and Cy3-labeled PC-FUS (1 nM). The mixture was then immediately applied to the anti-MBP antibody-coated slide surface, then recording a long movie (~10 minutes) at room temperature (23 °C ± 2 °C). Single-molecule time traces were selected using a custom MATLAB code and fitted to an exponential function in Origin.

The kinetic rate corresponding to nanocluster formation was extracted from the fitted traces. By combining the assembly rates of nanoclusters at different concentrations, statistical significance was calculated using a two-tailed, two-sample Student’s t-test with p < 0.001.

To understand the coarsening dynamics, we fit the time evolution of nanocluster/condensate size using $\left\langle R\left(t\right)\right\rangle={\left({\left\langle {R}_{0}\right\rangle }^{1/\alpha }+{kt}\right)}^{\alpha }$, where $\left\langle {R}_{0}\right\rangle$ is the mean initial radius, ${k}$ is the growth rate, and α is the dynamic scaling exponent⁴⁰. Since the coarsening of nanoclusters was arrested around 30 minutes, we first applied the fit to data at 500 nM and 1000 nM PC-FUS concentrations. A global fit to these two datasets yielded $\alpha=0.334\pm 0.051$, which is consistent with $\alpha=1/3$ – a scaling in agreement with diffusion-limited coarsening via Ostwald Ripening⁴¹ or Brownian motion-induced coalescence^42,43. We therefore fixed the scaling exponent $\alpha$ at 1/3 and performed individual fits to the coarsening dynamics for each concentration using $\left\langle R\left(t\right)\right\rangle={\left({\left\langle {R}_{0}\right\rangle }^{3}+{kt}\right)}^{1/3}$. All fittings were performed using the Levenberg–Marquardt algorithm⁶¹ for solving non-linear least squares problems, implemented in the Python module SciPy⁶².

TIRFM-based nanocluster size measurement

We measured the early stages of nanocluster growth by monitoring the initial fluorescence intensity of captured clusters. Due to the inherent limitations of TIRFM and potential camera saturation, long-term measurements focused on nanocluster size rather than fluorescence intensity. To ensure consistent nanocluster size measurements, we calibrated the TIRFM setup using commercially available fluorescent beads of known diameters (see Supplementary Information for details). All nanocluster images were acquired using identical optical settings (laser power, exposure time, and gain) as used during bead calibration. In cases where high protein concentrations led to visible streaks or signs of signal saturation typically aligned with the laser path these regions were excluded from analysis. Our custom MATLAB pipeline applies binary thresholding, area-based filtering, and contour detection to identify circular features and exclude non-uniform or saturated signals. This approach improves the quality of size measurements even when signals approached detector saturation and ensured that measured distributions accurately represent individual nanoclusters. The image processing pathway used to determine nanocluster size is illustrated in Supplementary Fig. S5.

Nanocluster mixing

Two batches of nanoclusters were prepared simultaneously: one containing Cy3-labeled PC-FUS (along with MBP-FUS and PC-FUS) and the other containing Cy5-labeled PC-FUS (with PC-FUS only). The Cy3-labeled MBP-FUS nanocluster was applied to the slide surface and immobilized through anti-MBP interactions. Both samples were incubated for one hour.

After incubation, the Cy5-labeled FUS nanocluster (without MBP-FUS) was applied to the sample chamber containing the Cy3-labeled FUS nanocluster. Short movies were recorded, and the field-of-view images were analyzed in ImageJ to check for colocalization. Similarly, the reverse experiment was performed, where the Cy5-labeled FUS nanocluster was immobilized first, followed by the addition of the Cy3-labeled FUS nanocluster.

Nanocluster FRAP

Surprisingly, the Cy3-labeled FUS nanocluster did not exhibit photobleaching under intense laser exposure, whereas the Cy5-labeled FUS nanocluster showed partial photobleaching. Therefore, the Cy5-labeled FUS nanocluster was used for FRAP experiments, maintaining a fixed laser exposure time while varying the recovery intervals without changing the imaging position.

A fixed red laser exposure time of 20 seconds induced photobleaching, followed by recovery intervals ranging from 5 to 180 seconds. The maximum recovered intensity at each time point was extracted from individual nanocluster FRAP measurements and plotted over time.

The fluorescence recovery data at each PC-FUS concentration were shifted and normalized as ${I}_{{{\rm{norm}}}}\left(t^{\prime} \right)={I}_{{{\rm{norm}}}}\left(t-{t}_{\min }\right)=\left(I\left(t\right)-{I}_{\min }\right)/\left({I}_{\max }-{I}_{\min }\right)$, where $I\left(t\right)$ is the fluorescence intensity at time $t$, ${I}_{\min }$ is the minimum value of the intensity curve, ${t}_{\min }$ is the corresponding time at ${I}_{\min }$, and ${I}_{\max }$ is the maximum recovery. ${I}_{\max }$ was taken as the mean of the last two data points under the assumption that the recovery curve has reached a plateau there. Using the mean of the last two points instead of the last one alone helps mitigate variability in the data for the plateau intensity. The normalized fluorescence recovery curve ${I}_{{{\rm{norm}}}}\left(t^{\prime} \right)$ exhibits the properties that ${I}_{{{\rm{norm}}}}\left({t}^{{\prime} }=0\right)=0$ and ${I}_{{{\rm{norm}}}}\left({t}^{{\prime} }\to+\infty \right)=1$. The characteristic timescale of recovery $\tau$ for each PC-FUS concentration was obtained by fitting the normalized intensity data to ${I}_{{{\rm{norm}}}}\left(t^{\prime} \right)=1-{e}^{-{t}^{{\prime} }/\tau }$²⁸ using the Levenberg–Marquardt algorithm⁶¹ implemented in the Python module SciPy⁶². To avoid the inherent heterogeneity of protein, each condition took more than 100 nanometers/condensates into quantification.

Statistics and reproducibility

All experiments were repeated independently multiple times with consistent results, demonstrating reproducibility. Data shown in Figs. 4H and 7D–E were obtained from individual and independent events. All nanocluster size measurements were performed using the SiCluP assay at different time points, analyzing 2000–3000 nanoclusters from 15–20 short movies acquired at different imaging positions. Nanocluster formation rates were calculated by fitting single-cluster time traces. For FRAP measurements at different FUS concentrations, ~100 single-cluster traces of periodic photobleaching at different time intervals were combined. Data shown in Fig. 4H represent counts of individual nanocluster formation rates. All raw data are provided in the Source Data file.

No statistical method was used to predetermine sample size. No data were excluded from the analyses. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.

Data availability

The custom analysis codes are archived in Zenodo and available at https://doi.org/10.5281/zenodo.17162358. Source data are provided with this paper.

References

Carey, J. L. & Guo, L. Liquid-liquid phase separation of tdp-43 and fus in physiology and pathology of neurodegenerative diseases. Front Mol. Biosci. 9, 1–20 (2022).
Article ADS Google Scholar
Portz, B., Lee, B. L. & Shorter, J. Trends in Biochemical Sciences Review FUS and TDP-43 Phases in Health and Disease. Trends Biochem Sci 1–14 https://doi.org/10.1016/j.tibs.2020.12.005 (2020).
Guo, Y. E. et al. Pol II phosphorylation regulates a switch between transcriptional and splicing condensates. Nature 572, 543–548 (2019).
Article CAS PubMed PubMed Central ADS Google Scholar
Reber, S. et al. The phase separation-dependent FUS interactome reveals nuclear and cytoplasmic function of liquid–liquid phase separation. Nucleic Acids Res 49, 7713–7731 (2021).
Article CAS PubMed PubMed Central Google Scholar
Altmeyer, M. et al. Liquid demixing of intrinsically disordered proteins is seeded by poly(ADP-ribose). Nat. Commun. 6, 8088 (2015).
Article CAS PubMed ADS Google Scholar
Wang, B. et al. Liquid–liquid phase separation in human health and diseases. Signal Transduct Target Ther 6, 290 (2021).
Patel, A. et al. A Liquid-to-solid phase transition of the als protein fus accelerated by disease mutation. Cell 162, 1066–1077 (2015).
Article CAS PubMed Google Scholar
Lafontaine, D. L. J. J., Riback, J. A., Bascetin, R. & Brangwynne, C. P. The nucleolus as a multiphase liquid condensate. Nat. Rev. Mol. Cell Biol. 22, 165–182 (2021).
Article CAS PubMed Google Scholar
Martinez-Macias, M. I. et al. FUS (fused in sarcoma) is a component of the cellular response to topoisomerase I–induced DNA breakage and transcriptional stress. Life Sci. Alliance 2, e201800222 (2019).
Article PubMed PubMed Central Google Scholar
Tyzack, G. E. et al. Paraspeckle components NONO and PSPC1 are not mislocalized from motor neuron nuclei in sporadic ALS. Brain 143, e66–e66 (2020).
Article PubMed PubMed Central Google Scholar
Rhine, K. et al. Poly(ADP-ribose) drives condensation of FUS via a transient interaction. Mol. Cell 82, 969–985.e11 (2022).
Article CAS PubMed PubMed Central Google Scholar
Kwon, I. et al. Phosphorylation-Regulated Binding of RNA Polymerase II to Fibrous Polymers of Low-Complexity Domains. Cell 155, 1049–1060 (2013).
Article CAS PubMed PubMed Central Google Scholar
Hennig, S. et al. Prion-like domains in RNA binding proteins are essential for building subnuclear paraspeckles. J. Cell Biol. 210, 529–539 (2015).
Article CAS PubMed PubMed Central Google Scholar
Wolozin, B. & Ivanov, P. Stress granules and neurodegeneration. Nat. Rev. Neurosci. 20, 649–666 (2019).
Article CAS PubMed PubMed Central Google Scholar
Marrone, L. et al. Isogenic FUS-eGFP iPSC reporter lines enable quantification of fus stress granule pathology that is rescued by drugs inducing autophagy. Stem Cell Rep. 10, 375–389 (2018).
Article CAS Google Scholar
Lee, A. K. et al. Translational repression of g3bp in cancer and germ cells suppresses stress granules and enhances stress tolerance. Mol. Cell 79, 645–659.e9 (2020).
Article CAS PubMed Google Scholar
Curdy, N. et al. Stress granules in the post-transcriptional regulation of immune cells. Front. Cell Dev Biol. 8, 611185 (2021).
Baymiller, M. & Moon, S. L. Stress Granules as Causes and Consequences of Translation Suppression. Antioxid. Redox Signal 39, 390–409 (2023).
Article CAS PubMed PubMed Central Google Scholar
Niaki, A. G. et al. Loss of Dynamic RNA Interaction and Aberrant Phase Separation Induced by Two Distinct Types of ALS/FTD-Linked FUS Mutations. Mol. Cell 77, 82–94.e4 (2020).
Article CAS PubMed Google Scholar
Koch, L., Pollak, R., Ebbinghaus, S. & Huber, K. Early Stages of FUS Droplet Formation via Liquid-Liquid Phase Separation. Langmuir https://doi.org/10.1021/acs.langmuir.4c01243 (2024).
Jawerth, L. et al. Protein condensates as aging Maxwell fluids. Science (1979) 370, 1317–1323 (2020).
CAS Google Scholar
Rhine, K. et al. ALS/FTLD-Linked Mutations in FUS glycine residues cause accelerated gelation and reduced interactions with wild-type FUS. Mol. Cell 80, 666–681.e8 (2020).
Article CAS PubMed PubMed Central Google Scholar
Davis, R. B., Kaur, T., Moosa, M. M. & Banerjee, P. R. FUS oncofusion protein condensates recruit mSWI/SNF chromatin remodeler via heterotypic interactions between prion-like domains. Protein Sci. 30, 1454–1466 (2021).
Article CAS PubMed PubMed Central Google Scholar
Berkeley, R. F., Kashefi, M. & Debelouchina, G. T. Real-time observation of structure and dynamics during the liquid-to-solid transition of FUS LC. Biophys. J. 120, 1276–1287 (2021).
Article CAS PubMed PubMed Central ADS Google Scholar
Murray, D. T. et al. Structure of FUS protein fibrils and its relevance to self-assembly and phase separation of low-complexity domains. Cell 171, 615–627.e16 (2017).
Article CAS PubMed PubMed Central Google Scholar
Li, M., Chen, G. & Zhang, Z. Modeling the effects of phosphorylation on phase separation of the FUS low-complexity domain. Biophys. J. 122, 2636–2645 (2023).
Article CAS PubMed PubMed Central ADS Google Scholar
He, G. et al. Phase-separating pyrenoid proteins form complexes in the dilute phase. Commun. Biol. 6, 1–10 (2023).
Article Google Scholar
Zhang, Y. et al. The exchange dynamics of biomolecular condensates. Elife 12, 484641 (2024). 2022.03.16.
Article Google Scholar
Shelkovnikova, T. A., Robinson, H. K., Southcombe, J. A., Ninkina, N. & Buchman, V. L. Multistep process of FUS aggregation in the cell cytoplasm involves RNA-dependent and RNA-independent mechanisms. Hum. Mol. Genet 23, 5211–5226 (2014).
Article CAS PubMed PubMed Central Google Scholar
Kar, M. et al. Phase-separating RNA-binding proteins form heterogeneous distributions of clusters in subsaturated solutions. Proc. Natl Acad. Sci. 119, 1–30 (2022).
Article Google Scholar
He, C., Wu, C. Y., Li, W. & Xu, K. Multidimensional Super-Resolution Microscopy Unveils Nanoscale Surface Aggregates in the Aging of FUS Condensates. J. Am. Chem. Soc. 145, 24240–24248 (2023).
Article CAS PubMed PubMed Central ADS Google Scholar
Davis, R. B., Supakar, A., Ranganath, A. K., Moosa, M. M. & Banerjee, P. R. Heterotypic interactions can drive selective co-condensation of prion-like low-complexity domains of FET proteins and mammalian SWI/SNF complex. Nat. Commun. 15, 1–17 (2024).
Article CAS Google Scholar
Narayanan, A. et al. A first order phase transition mechanism underlies protein aggregation in mammalian cells. Elife 8, e39695 (2019).
Lan, C. et al. Quantitative real-time in-cell imaging reveals heterogeneous clusters of proteins prior to condensation. Nat. Commun. 14, 4831 (2023).
Article CAS PubMed PubMed Central ADS Google Scholar
Kusumi, A., Tsunoyama, T. A., Suzuki, K. G. N., Fujiwara, T. K. & Aladag, A. Transient, nano-scale, liquid-like molecular assemblies coming of age. Curr. Opin. Cell Biol. 89, 102394 (2024).
Article CAS PubMed Google Scholar
Ray, S. et al. Mass photometric detection and quantification of nanoscale α-synuclein phase separation. Nat. Chem. 15, 1306–1316 (2023).
Article CAS PubMed Google Scholar
Martin, E. W. et al. A multi-step nucleation process determines the kinetics of prion-like domain phase separation. Nat. Commun. 12, 4513 (2021).
Article CAS PubMed PubMed Central ADS Google Scholar
Ganser, L. R., Ge, Y. & Myong, S. Single-Molecule Fluorescence Methods to Study Protein-RNA Interactions Underlying Biomolecular Condensates. in 149–160 https://doi.org/10.1007/978-1-0716-2663-4_7 (2023).
Rhine, K., Vidaurre, V. & Myong, S. RNA Droplets. Annu Rev. Biophys. 49, 247–265 (2020).
Article CAS PubMed PubMed Central Google Scholar
Berry, J., Brangwynne, C. P. & Haataja, M. Physical principles of intracellular organization via active and passive phase transitions. Reports on Progress in Physics 81, (2018).
Lifshitz, I. M. & Slyozov, V. V. The kinetics of precipitation from supersaturated solid solutions. J. Phys. Chem. Solids 19, 35–50 (1961).
Article ADS Google Scholar
Binder, K. & Stauffer, D. Theory for the slowing down of the relaxation and spinodal decomposition of binary mixtures. Phys. Rev. Lett. 33, 1006–1009 (1974).
Article ADS Google Scholar
Siggia, E. D. Late stages of spinodal decomposition in binary mixtures. Phys. Rev. A (Coll. Park) 20, 595–605 (1979).
Article CAS ADS Google Scholar
Berry, J., Weber, S. C., Vaidya, N., Haataja, M. & Brangwynne, C. P. RNA transcription modulates phase transition-driven nuclear body assembly. Proceedings of the National Academy of Sciences 112, (2015).
Muzzopappa, F. et al. Detecting and quantifying liquid–liquid phase separation in living cells by model-free calibrated half-bleaching. Nat. Commun. 13, 1–15 (2022).
Article ADS Google Scholar
Yokosawa, K. et al. Concentration Quantification of the Low-Complexity Domain of Fused in Sarcoma inside a Single Droplet and Effects of Solution Parameters. J. Phys. Chem. Lett. 13, 5692–5697 (2022).
Article CAS PubMed Google Scholar
Duan, C. & Wang, R. A Unified Description of Salt Effects on the Liquid-Liquid Phase Separation of Proteins. ACS Cent. Sci. 10, 460–468 (2024).
Article CAS PubMed PubMed Central Google Scholar
MacAinsh, M., Dey, S. & Zhou, H.-X. Direct and Indirect Salt Effects on Homotypic Phase Separation. eLife 13, (2024).
Blazquez, S. et al. Location and Concentration of Aromatic-Rich Segments Dictates the Percolating Inter-Molecular Network and Viscoelastic Properties of Ageing Condensates. Adv. Sci. 10, 1–18 (2023).
Article Google Scholar
Tejedor, A. R., Garaizar, A., Ramírez, J. & Espinosa, J. R. RNA modulation of transport properties and stability in phase-separated condensates. Biophys. J. 120, 5169–5186 (2021).
Article CAS PubMed PubMed Central ADS Google Scholar
Banerjee, P. R., Milin, A. N., Moosa, M. M., Onuchic, P. L. & Deniz, A. A. Reentrant Phase Transition Drives Dynamic Substructure Formation in Ribonucleoprotein Droplets. Angew. Chem. 129, 11512–11517 (2017).
Article ADS Google Scholar
Golani, G. et al. Mesoscale properties of protein clusters determine the size and nature of liquid-liquid phase separation (LLPS). Commun. Phys. 8, 226 (2025).
Article CAS Google Scholar
Murthy, A. C. et al. Molecular interactions underlying liquid−liquid phase separation of the FUS low-complexity domain. Nat. Struct. Mol. Biol. 26, 637–648 (2019).
Article CAS PubMed PubMed Central Google Scholar
Ahlers, J. et al. The key role of solvent in condensation: Mapping water in liquid-liquid phase-separated FUS. Biophys. J. 120, 1266–1275 (2021).
Article CAS PubMed PubMed Central ADS Google Scholar
St George-Hyslop, P. et al. The physiological and pathological biophysics of phase separation and gelation of RNA binding proteins in amyotrophic lateral sclerosis and fronto-temporal lobar degeneration. Brain Res 1693, 11–23 (2018).
Article Google Scholar
Ganser, L. R. & Myong, S. Methods to study phase-separated condensates and the underlying molecular interactions. Trends Biochem Sci. 45, 1004–1005 (2020).
Article CAS PubMed PubMed Central Google Scholar
Gebhardt, C. et al. Molecular and spectroscopic characterization of green and red cyanine fluorophores from the alexa Fluor and AF Series*. ChemPhysChem 22, 1566–1583 (2021).
Article CAS PubMed PubMed Central Google Scholar
Paul, T. & Myong, S. Protocol for generation and regeneration of PEG-passivated slides for single-molecule measurements. STAR Protoc. 3, 101152 (2022).
Article CAS PubMed PubMed Central Google Scholar
Roy, R., Hohng, S. & Ha, T. A practical guide to single-molecule FRET. Nat. Methods 5, 507–516 (2008).
Article CAS PubMed PubMed Central Google Scholar
Paul, T., Ha, T. & Myong, S. Regeneration of PEG slide for multiple rounds of single-molecule measurements. Biophys. J. 120, 1788–1799 (2021).
Article CAS PubMed PubMed Central ADS Google Scholar
Levenberg, K. enneth a Method for the Solution of Certain Non-Linear Problems in Least Squares. Q Appl Math. 2, 164–168 (1944).
Article MathSciNet Google Scholar
Virtanen, P. et al. SciPy 1.0: fundamental algorithms for scientific computing in Python. Nat. Methods 17, 261–272 (2020).
Article CAS PubMed PubMed Central Google Scholar

Download references

Acknowledgements

We thank Dr. Ibraheem Alshareedah for sharing the Python scripts of FCS data analysis. We thank Sophie Skanchy for the MATLAB scripts of image masking and droplet quantification. This project was supported by RF1 NS113636-01 and R01 AG071326-01.

Author information

These authors contributed equally: Yingda Ge, Tapas Paul.

Authors and Affiliations

Department of Biophysics, Johns Hopkins University, Baltimore, MD, USA
Yingda Ge, Yaojun Zhang & Sua Myong
Program in Cellular and Molecular Medicine, Boston Children’s Hospital, Boston, MA, USA
Yingda Ge, Tapas Paul, Nilimesh Das & Sua Myong
Department of Pediatrics, Harvard Medical School, Boston, MA, USA
Tapas Paul, Nilimesh Das & Sua Myong
Department of Physics and Astronomy, Johns Hopkins University, Baltimore, MD, USA
Margarita Gordiychuk & Yaojun Zhang

Authors

Yingda Ge
View author publications
Search author on:PubMed Google Scholar
Tapas Paul
View author publications
Search author on:PubMed Google Scholar
Margarita Gordiychuk
View author publications
Search author on:PubMed Google Scholar
Nilimesh Das
View author publications
Search author on:PubMed Google Scholar
Yaojun Zhang
View author publications
Search author on:PubMed Google Scholar
Sua Myong
View author publications
Search author on:PubMed Google Scholar

Contributions

Y.G. and T.P. designed the study, performed experiments, and analyzed the raw data. Y.G. was responsible for sample preparation, wide-field imaging, DLS, and FCS measurements. T.P. conducted the SiCluP experiments and analyzed the corresponding data. M.G. contributed to theoretical modeling and fitting of the SiCluP results. N.D. collaborated with Y.G. on sample preparation and FCS measurements. S.M. and Y.G. wrote the original manuscript, with T.P., Y.Z., Y.Z., and S.M. contributing to its editing and revision. Y.Z. and M.G. co-wrote the computational section.

Corresponding author

Correspondence to Sua Myong.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information

Transparent Peer Review file

Source data

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Cite this article

Ge, Y., Paul, T., Gordiychuk, M. et al. FUS nanoclusters are a distinct state within the dilute phase. Nat Commun 16, 9956 (2025). https://doi.org/10.1038/s41467-025-64909-7

Download citation

Received: 20 January 2025
Accepted: 01 October 2025
Published: 12 November 2025
Version of record: 12 November 2025
DOI: https://doi.org/10.1038/s41467-025-64909-7

Subjects

Abstract

Similar content being viewed by others

Nanoscale profiling of evolving intermolecular interactions in ageing FUS condensates

Fibril formation and ordering of disordered FUS LC driven by hydrophobic interactions

Single-droplet surface-enhanced Raman scattering decodes the molecular determinants of liquid-liquid phase separation

Introduction

Results

Pre-cleavage enables probing the early development of FUS LLPS

FUS nanoclusters are distinct from condensates in size and kinetics

SiCluP (Single Cluster Pull-down) of single FUS nanoclusters

Single FUS nanoclusters exhibit liquid-like properties

Nanoclusters’ tolerance to mechanical and chemical disruption

RNA regulates the size and kinetics of nanoclusters

Discussion

Methods

Plasmid preparation

Protein purification

Cleavage of the MBP-tag

Protein labeling

Dynamic light scattering measurements

Fluorescence correlation spectroscopy measurements

Condensate imaging

Single cluster pull-down (SiCluP) assay

Nanocluster assembly

TIRFM-based nanocluster size measurement

Nanocluster mixing

Nanocluster FRAP

Statistics and reproducibility

Data availability

References

Acknowledgements

Author information

Authors and Affiliations

Contributions

Corresponding author

Ethics declarations

Competing interests

Peer review

Peer review information

Additional information

Supplementary information

Supplementary Information

Transparent Peer Review file

Source data

Source data

Rights and permissions

About this article

Cite this article

Share this article

Search

Quick links