ATP promotes protein coacervation through conformational compaction

Abstract

Adenosine triphosphate (ATP) has been recognized as a hydrotrope in the phase separation process of intrinsically disordered proteins (IDPs). Surprisingly, when using the disordered Arg–Gly/Arg–Gly–Gly (RG/RGG) rich motif from the HNRNPG protein as a model system, we discover a biphasic relationship between the ATP concentration and IDP phase separation. We show that, at a relatively low ATP concentration, ATP dynamically interacts with the IDP, which neutralizes protein surface charges, promotes intermolecular interactions, and consequently promotes phase separation. We further demonstrate that ATP induces a compact conformation of the IDP, accounting for the reduced solvent exchange rate and lower compression ratio during phase separation. As ATP concentration increases, its hydrotropic properties emerge, leading to the dissolution of the phase-separated droplets. Our finding uncovers a complex mechanism by which ATP molecules modulate the structure, interaction, and phase separation of IDPs and accounts for the distinct phase separation behaviors of the charge-rich RGG motif and other low-complexity IDPs.

intrinsically disordered protein, phase separation, ATP, conformational compaction, HNRNPG

Introduction

Intrinsically disordered proteins (IDPs) and intrinsically disordered regions make up ∼40% of human proteome. Unlike folded proteins, IDPs are generally characterized with low-complexity sequences and often comprise a large proportion of charged residues, preventing them from adopting compact conformations (Habchi et al., 2014). These proteins do not naturally collapse or fold efficiently on their own; instead, they remain highly dynamic and conformationally diverse. However, they can be stabilized upon binding to the folded proteins (Wright and Dyson, 2015). Over the past decade, it has been shown that IDPs are also important participants of macromolecular condensates through the processes of liquid–liquid phase separation (LLPS) (Dignon et al., 2020; Sanders et al., 2020; Wiedner and Giudice, 2021). IDPs involved in the LLPS process typically exhibit low-complexity sequence characteristics, including sequences like poly-P/poly-Q, prion-like domain (PLD) rich in Gln–Gly–Ser–Tyr (QGSY), and charged motifs rich in Arg–Gly/Arg–Gly–Gly (RG/RGG) (Chong et al., 2018; Ryan et al., 2018; Alshareedah et al., 2019).

Many physicochemical factors modulate the phase separation of biomolecules. For example, temperature influences molecular thermal motion, which can regulate phase separation (Dignon et al., 2019). On the other hand, ionic strength affects LLPS by modulating electrostatic interactions (Krainer et al., 2021), while macromolecular crowding adjusts intermolecular interactions through spatial steric hindrance, thereby often promoting phase separation (Ghosh et al., 2019).

Adenosine triphosphate (ATP) serves as an important energy currency in cells, typically in millimolar concentrations. Within this range, ATP not only provides a direct energy source for biochemical reactions but also regulates signal transduction and metabolic processes. High concentrations of ATP can also effectively maintain intracellular ionic balance and membrane potential. Microscopically, ATP can interact with protein residues through a range of interactions (Nishizawa et al., 2021; Ou et al., 2021; Ren et al., 2022; Toyama et al., 2022). The nucleobases of ATP form hydrogen bonds, π–π stacking, and NH–π interactions with proteins (Shetty et al., 1996; Meyer et al., 2003; Mao et al., 2004), whereas its phosphate groups bind to charged or polar residues via hydrogen bonding and salt bridges, facilitating complex biochemical processes (Heo et al., 2020; Toyama et al., 2022).

Recent studies have shown that ATP plays a significant role as a biological hydrotrope, which enhances protein solubility, reverses protein phase transition, and prevents the formation of protein fibrils and aggregates (Patel et al., 2017; Mehringer et al., 2021). However, the role of ATP in regulating protein phase separation is likely more nuanced. At relatively low concentration, ATP has been found to promote the phase separation of the RG/RGG-rich IDPs derived from fused in sarcoma (FUS) and cell cycle-associated protein 1 (CAPRIN1) proteins (Kang et al., 2018; Toyama et al., 2022). A recent nuclear magnetic resonance (NMR) study revealed that ATP binding neutralizes the positive charges of CAPRIN1, enabling attractive intermolecular interactions (Toyama et al., 2022). Furthermore, ATP can mediate phase separation of IDP molecules by mediating intermolecular interactions (Kota et al., 2024). Nevertheless, at higher concentrations, ATP has been shown to reverse the electrostatic potential of CAPRIN1 and cause the dissipation of the protein droplets (Toyama et al., 2022).

Heterogeneous nuclear ribonucleoprotein G (HNRNPG), also known as RBMX, is associated with nascent mRNA transcription, involved in alternative splicing of pre-mRNAs, and thus implicated in ultraviolet damage response (Xiang et al., 2017). It is noteworthy that HNRNPG is localized in nuclear speckles (Heinrich et al., 2009; Adamson et al., 2012), which are membraneless organelles formed through phase separation (Galganski et al., 2017). The HNRNPG protein contains an RG/RGG-rich motif near its C-terminus (Figure 1A), which undergoes phase separation. In the present study, using an integrative approach that includes NMR, small-angle X-ray scattering (SAXS), and molecular dynamics (MD) simulations, we show that ATP binding not only neutralizes the charge but also modulates the conformation of the RGG domain derived from HNRNPG, thereby promoting protein coacervation at a relatively low ATP concentration.

Figure 1

ATP modulates the phase separation of the RGG domain derived from HNRNPG. (A) The construct of HNRNPG used in the current study, with the sequence of the C-terminal RGG domain. (B) Differential interference contrast (DIC) micrographs showing that the IDP forms liquid droplets only when the ATP concentration is within the range of 5–20 mM. (C) ATP co-localizes with the IDP in the droplets. Here, Cy3-labeled protein (red) and ATP analog TNP-ATP (blue) were doped into the unlabeled protein at 2% and 5% proportions, respectively. (D) FRAP experiments showing that the IDP is highly mobile in the liquid droplet. GFP was fused to the N-terminus of the RGG domain.

Results

ATP modulates the phase separation behavior of an RGG-rich IDP

We found that the RGG domain (residues 334–391) derived from the HNRNPG protein did not undergo phase separation on its own at protein concentrations up to 300 μM. However, the addition of 5 mM ATP, either alone or with equimolar Mg²⁺, lowered the threshold for phase separation, with protein demixing taking place at concentrations of 200 μM and 250 μM, respectively (Figure 1; Supplementary Figure S1). Since ATP in cells is usually associated with Mg²⁺ ions, in subsequent experiments and simulations, we focused on ATP complexed with Mg²⁺, hereafter simply referred to as ATP. Increasing the concentration of ATP eventually dissolved the protein droplets (Figure 1B), which aligns with its well-established role as a hydrotrope.

We doped ATP with fluorescent analog 2,4,6-trinitrophenol (TNP) and found the dye co-localized with Cy3-labeled RGG domain in the droplets (Figure 1C), similar to previous findings for the FUS protein (Patel et al., 2017). Thus, ATP co-phase-separates with the protein likely through direct interaction and exerts a modulatory effect. Additionally, the liquid droplets exhibited high fluidity, as demonstrated by nearly 100% fluorescence recovery of labeled proteins in the droplets in the fluorescence recovery after photobleaching (FRAP) experiment (Figure 1D).

ATP makes the IDP more conformationally compact

We then assessed the effect of ATP on the conformation of this RGG-rich IDP. We collected SAXS data at three different concentrations (50, 75, and 100 μM) just below the phase separation threshold with or without ATP (Supplementary Figure S2). The SAXS data indicated a more compact protein conformation upon the addition of ATP, as the radius of gyration (R_g) deceased from ∼20 Å to ∼17 Å and the end-to-end distance D_max decreased from ∼70 Å to ∼60 Å (Figure 2A and B). The pair distance distribution function (PDDF) curves also indicated a narrower distance distribution of the protein conformation upon the addition of ATP (Figure 2B). For a rough comparison, a protein of similar size like GB1 (PDB ID: 1GB1, 56 residues) has an R_g of ∼11 Å in its folded state, but the addition of Gnd·HCl unfolded the protein and yielded an R_g value of 23 ± 1 Å (Kohn et al., 2004).

Figure 2

The addition of ATP collapses the ensemble conformation of HNRNPG-RGG. (A) The scattering curve of HNRNPG-RGG at 75 μM in the absence (black) and presence of 3.75 mM ATP (red). (B) The PDDF in the absence (black) and presence of 3.75 mM ATP (red), experimentally derived from SAXS data. (C) NMR DOSY measurement affords the translational diffusion coefficients. (D) The distribution of calculated R_g of the protein, obtained from MD simulations.

We further employed NMR experiments to assess the change in protein size. The translational diffusion coefficient obtained from NMR diffusion-ordered spectroscopy (DOSY) measurements showed a large increase in diffusion rate upon the addition of 5 mM ATP (Figure 2C), indicating a compaction of protein conformation. As a control, 3-(trimethylsilyl)-1-propanesulfonic acid sodium salt (DSS) exhibited no change in diffusion rate without or with ATP (Supplementary Figure S3).

With the resonance assignment obtained for protein backbone using the standard triple-resonance NMR spectroscopy (Supplementary Figure S4), we further measured the relaxation rates and heteronuclear NOE (XNOE) of the backbone amide nitrogen atoms (Figure 3). The transverse relaxation rate R₂ decreased upon the addition of 10 mM ATP, whereas the longitudinal relaxation rate R₁ increased. The XNOE values also decreased, indicating that the addition of ATP affects the local dynamic properties of the RGG. Thus, without considering the variations in local dynamics, a decreased R₂/R₁ ratio is resulted from a decreased rotational correlation time τ_c (Kay et al., 1989), thus corroborating the translational diffusion measurement by NMR DOSY.

Figure 3

NMR relaxation analysis of HNRNPG-RGG without or with 10 mM ATP. The residue-specific R₁ (A), R₂ (B), and XNOE (C) were collected at 25°C on a 600 MHz NMR instrument.

The compaction of the protein conformation was further supported by MD simulations. In the absence of ATP, the distribution of the R_g values exhibited a distinct ‘long tail’, indicating the presence of highly extended conformations. Addition of 5-fold excess of ATP resulted in a narrower distribution of the R_g values for the IDP (Figure 2D; Supplementary Figure S5), indicating both conformational compaction and homogenization.

Dynamic association of ATP with HNRNPG-RGG

Our all-atom MD simulations showed that the phosphate groups of ATP molecules are preferentially associated with Arg residues (Figure 4A and B). Interestingly, ATP binding mainly occurs near the N-terminal and C-terminal regions of the protein (Supplementary Figure S6). Besides electrostatic interactions, MD simulations showed that the nucleobase of ATP forms hydrogen bonds with the backbone and sidechain at Arg, Ser, and Gly residues (Figure 4B; Supplementary Figure S7).

Figure 4

Structural models for the ATP-associated HNRNPG-RGG. (A) Representative snapshots from MD simulations indicate that the ATP phosphate group interacts with Arg sidechain, and ATP nucleobase forms hydrogen bonds with adjacent Gly and Ser residues. (B) Statistics of the number of hydrogen bonds formed between ATP and different protein residues, with the phosphate and nucleobase groups tabulated separately. (C) The probability of the indicated number of associated ATP molecules, with the theoretical protein and ATP concentrations at 1.6 mM and 8 mM, respectively. The probability was calculated as the number of the RGG domain associated with the indicated number of ATP molecules divided by the total number of MD snapshots.

MD simulations also showed that the number of the protein-associated ATP molecules (i.e. within 6 Å vicinity of the protein) fluctuates over time, with 2–4 most likely. In fact, in a simulated system containing five ATP molecules, the proportion that all five ATP molecules simultaneously associated with the protein was only 10% (Figure 4C; Supplementary Figure S8), suggesting that the interactions between ATP and the RGG domain are weak and dynamic. Thus, to effectively neutralize protein charges and promote protein phase separation, a 20-fold molar excess of ATP (Figure 1A) is required, which is approximately twice as much as the total number of Arg residues.

ATP slows water exchange for labile protons in the IDP

The addition of 5 or 10 mM ATP to the protein caused little chemical shift perturbations (CSPs) of the backbone amide protons (Supplementary Figure S9). However, the intensities of many peaks increased in the presence of ATP (Figure 5A; Supplementary Figure S10), which can be attributed to two factors. First, ATP binding makes the protein more compact, which would retard the penetration of water molecules into the ‘interior’ of the protein (Figures 2 and 3). Second, the increased peak intensity arises from the direct binding of ATP, thus reducing the exchange of labile protons with water.

Figure 5

Solvent exchange and intermolecular interactions of the IDP are reduced in the presence of ATP. (A) The ratios of peak intensities were calculated for the backbone amide protons in the presence and absence of ATP. (B) The solvent exchange rates k_ex were measured by CLEANEX-PM scheme, in the presence and absence of ATP. (C) Intermolecular PRE was measured in the absence and presence of 10 mM ATP. The labeling site of the paramagnetic probe is indicated with an asterisk.

Interestingly, the most notable peak intensity increases were observed for Gly and Ser residues located adjacent to Arg residues. Although Gly and Ser are uncharged and largely hydrophilic, they are structurally more flexible and can readily undergo structural rearrangement (Li et al., 2024) upon the compaction of the protein and ATP binding to neighboring Arg. Moreover, MD simulations indicated that ATP molecules can approach closely to the NH atoms and potentially form a hydrogen bond with neighboring Ser and Gly residues (Figure 4B; Supplementary Figure S11).

We then mutated all the Arg residues in the protein to Ala (RGG_RtoA) and compared the NMR peak intensities in the absence and presence of ATP. We did not observe significant increases in NMR peak intensities for Gly and Ser residues in RGG_RtoA (Supplementary Figure S12). Accordingly, RGG_RtoA did not undergo phase separation in the presence of ATP (Supplementary Figure S13).

We also measured solvent exchange rates of amide protons using the phase-modulated clean chemical exchange (CLEANEX-PM) NMR pulse sequence (Hwanga et al., 1998). Notably, we could not measure the exchange rate for many residues in the free proteins, particularly Ser, due to fast solvent exchange. The solvent exchange rates decreased with the addition of 5 or 10 mM ATP (Figure 5B; Supplementary Figure S14). As the pH of protein solution was carefully controlled, the decreased exchange rates account for the increased peak intensity of the amide protons (Figure 5A).

ATP alters intermolecular interactions of the IDP

As ATP can lead to the demixing of the IDP at a relatively low concentration, it is likely that ATP also promotes protein–protein interactions. We thus measured the intermolecular paramagnetic relaxation enhancement (PRE), by comparing the peak intensity of paramagnetic probe-labeled protein to that of diamagnetic protein. Upon the addition of 10 mM ATP, the relative intensities of most peaks decreased, indicating enhanced associations between the proteins (Figure 5C). However, with the addition of 30 mM ATP, the relative peak intensities increased compared to those with 10 mM ATP, suggesting that high concentration of ATP somehow weakens the interactions between RGG proteins (Supplementary Figure S15).

In addition, MD simulations of two RGG molecules demonstrated that, in the absence of ATP, only a few contacts can be formed, while the addition of 8 mM ATP significantly enhances protein–protein interactions but more ATP (24 mM) significantly reduces the intermolecular contacts between RGG domains (Supplementary Figure S16).

The RGG motif is different from other low-complexity IDPs that phase-separate

Next, we employed various methods to determine the protein concentration in the condensed phase of HNRNPG-RGG. Based on the relative peak intensities in ¹H NMR spectra, we estimated that the concentration was 18.6 mM (117 mg/ml) (Figure 6A). Using sodium dodecyl sulphate–polyacrylamide gel electrophoresis (SDS–PAGE) with serial dilution, we estimated that the protein concentration was 26–28 mM (164–176 mg/ml) (Figure 6B and Table 1; Supplemental Figure S17). Upon doping the protein with a green fluorescent protein (GFP)-tagged version, we estimated the protein concentration at ∼25 mM (158 mg/ml) (Figure 6C). The concentration estimated for HNRNPG-RGG in the condensed phase is several folds lower than that for entangled random-coil polymers in the condensed phase and also lower than the concentrations reported for other phase-separating proteins. For example, the low-complexity domains from FUS, DDX4, hnRNPA2, and ELP3 were estimated at 477 mg/ml, 380 mg/ml, 440 mg/ml, and 500 mg/ml, in their respective condensed phases (Brady et al., 2017; Reichheld et al., 2017; Ryan et al., 2018; Murthy et al., 2019). Thus, the RGG protein remains hydrated (the water molecules were estimated to make up 88% of the volume) in the condensed phase, which accounts for high mobility in the droplet (Figure 1D; Supplemental Figure S1).

Figure 6

Determination of the protein concentration in the condensed phase. (A) Overlay of the 1D NMR spectra of the RGG domain in dispersed and phase-separated states. Based on the peak intensities, the protein in the condensed phase is ∼186 times more concentrated than in the diluted state. The original spectra without scaling are shown on the left. (B) Protein concentration determined using SDS–PAGE. The electrophoretic bands of HNRNPG-RGG are displayed, and the complete gel image is shown in Supplemental Figure S16. Lane numbers and corresponding sample types are indicated. Lanes marked with an asterisk (*) were not utilized in the analysis. The band in lane 6 is overexposed, and the band intensity data for lane 9 fall outside the range of the standard curve. The linear fitting of protein concentration and band intensity, the linear relationship function, and R² value are indicated. The condensed phase sample data are marked in red on the curve. Detailed information about the samples can be found in Table 1. (C) Protein concentration determined using fluorescence intensity. The fluorescence intensity of GFP-labeled RGG proteins at different concentrations was measured. The linear fitting of protein concentration and fluorescence intensity (fluorescence images on the top) is shown. The linear relationship function and R² value are indicated. The fluorescence intensity of the condensed phase of RGG protein doped with 2% GFP-labeled RGG proteins was measured and marked in red.

Table 1

The concentration of proteins in the condensed phase determined using SDS–PAGE.

Lane	Sample	Concentration (μM)a	Intensity	Dilute factorb
1	Dilute	25	8.10E + 07	1
2	Dilute	50	8.68E + 07	1
3	Dilute	100	9.47E + 07	1
4	Dilute	150	1.04E + 08	1
5	Dilute	200	1.13E + 08	1
6c	Dilute	250	1.36E + 08	1
7	Condensed	28384	1.02E + 08	1/200
8	Condensed	26068	8.87E + 07	1/400
9	Condensed	NAd	7.85E + 07	1/800

a

For dilute samples, the concentration is measured directly, while for samples in the condensed phase, the concentration is obtained by fitting based on the linear relationship between concentration and band intensity.

b

To avoid signal overexposure when measuring samples in the condensed phase, the sample is diluted by the corresponding factor before measuring the band intensity. This ensures that the measured intensity falls within the linear range of the detection system, allowing for accurate quantification.

c

Due to the overexposure of the stripe brightness, it is not used for the linear fitting of the standard curve.

d

Because the band intensity of the strip is lower than the range of the standard curve, it is not used for measuring the concentration of the condensed protein.

Furthermore, we prepared a protein construct from the N-terminal PLD of FUS (Figure 7A). The FUS-PLD is rich in QGSY residues but has no charged residue. The FUS protein readily phase-separates and can eventually assembles into fibrillar structure. It has been reported that ATP functions as a hydrotrope to dissipate FUS condensate (Patel et al., 2017). We found that the addition of ATP caused little NMR CSP or enhancement (Figure 7B and C; Supplementary Figure S18) and the slower translational diffusion (Figure 7D; Supplementary Figure S19), suggesting that the transient association between ATP molecules and the protein observed in MD simulations (Supplementary Figures S20 and S21) does not lead to a conformational compaction.

Figure 7

NMR titration of ATP to FUS-PLD. (A) The construction of FUS, with the PLD at its N-terminus. (B) CSPs for the ¹H–¹⁵ N of FUS-PLD in the presence of 10 mM ATP. (C) Relative peak intensities with the addition of 5 mM (red) and 10 mM (blue) ATP to those without ATP. (D) NMR DOSY measurement affords the translational diffusion coefficients. The NMR experiments were performed with FUS-PLD at 100 μM in the presence of 5 mM ATP.

Discussion

ATP and ATP complexed with Mg²⁺ have been shown as hydrotropes that help to stabilize aggregation-prone proteins in the diluted phase (Patel et al., 2017). In this study, we uncovered a novel mechanism by which ATP is involved in the regulation of protein structure and interactions. We showed that ATP regulates the phase separation of an RG/RGG-rich protein through charge neutralization and conformational compaction, which has not been reported for other IDPs undergoing phase separation.

The RGG-rich IDP is positively charged, making the protein to adopt an ensemble of extended and dynamic conformations. Electrostatic repulsion also inhibits intermolecular interactions (stage I in Figure 8). However, ATP preferentially binds near Arg residues, neutralizing positive charges. Although charge neutralization has recently been reported for other RGG-rich proteins (Toyama et al., 2022), we show here that ATP causes the conformational compaction of the protein. Subsequently, multiple protein molecules can coalesce in a highly hydrated form (stages II and III in Figure 8).

Figure 8

A scheme illustrating how ATP binding modulates the phase separation behavior of an RGG-rich IDP. The RGG domain contains a high density of positive charges that causes both intramolecular expansion and intermolecular repulsion (I). The addition of ATP neutralizes the positive charges, allowing intermolecular interactions to take place and causing the protein to adopt a more compact conformation (II). When a critical ATP concentration is reached, multiple copies of proteins come together and demix (III). Nevertheless, excessive ATP inhibits intermolecular interactions and dissipates liquid droplets (IV).

The role of ATP in inducing conformational compaction was corroborated through SAXS, NMR, and MD analyses. The transient association of ATP and resulting compaction of the protein protect backbone amides from solvent exchange. Initially, ATP's phosphate moiety facilitates this protection, allowing its hydrophobic regions to engage in intermolecular interactions. However, as ATP concentration increases (stage IV in Figure 8), the compacted and possibly negatively charged proteins are prevented from further interactions. In this stage, ATP acts as a hydrotrope, dissolving the phase-separated droplets.

The phase separation process of IDPs is driven by complex multivalent interactions. For instance, FUS-PLD, containing multiple aromatic amino acids, can form π–π stacking and cation–π interactions when phase-separating. In contrast, the phase separation of disordered proteins containing RGG motifs appears mainly driven by electrostatic interactions. ATP regulates electrostatic interactions while compacting protein structure. Significantly, this compacted conformation of the RGG domain prevents further compression and inhibits solidification and aggregation, as for proteins like FUS-PLD. The findings thus uncover a novel regulatory mechanism of ATP on protein phase separation.

Materials and methods

Sample preparation

The wild-type HNRNPG-RGG domain (L334–Y391) and RGG_RtoA mutant (all Arg to Ala) were cloned into the pET11a vector (GenBank: CAG33028.1). The N-terminus of the RGG domain connects the GB1 protein (His-tagged GB1 for RGG_RtoA) through the cleavage site of tobacco etch virus (TEV) protease. The fusion protein was expressed in BL21 Star (DE3) cells. Escherichia coli bacteria were cultured in minimal M9 medium or LB medium to prepare isotope-enriched or unlabeled proteins. For preparing the isotope-labeled protein, 1 g/L U-¹⁵N-labeled NH₄Cl (Cambridge Isotope Laboratories) and/or 2 g/L U-¹³C-labeled glucose (Sigma-Aldrich) were added to the minimal M9 medium. The protein was induced at an OD₆₀₀ of 0.8 with isopropyl-β-D-thiogalactopyranoside (IPTG) at a final concentration of 0.25 mM for 16 h at 23°C. The wild-type protein was purified with Sepharose SP, Sephacryl S100, and Source-S columns (GE Healthcare). The RGG_RtoA mutant protein was first purified with HisTrap^TM FF (GE Healthcare) and then buffer-exchanged to TEV protease restriction buffer containing 50 mM Tris, 1 mM ethylenediaminetetraacetic acid (EDTA), and 1 mM dithiothreitol (DTT) at pH 7.6. The GB1 protein was removed with TEV protease at 25°C for 1 h. The RGG domain was purified with the second round of Source-S column for wild-type and Source-Q column for RGG_RtoA mutant. Final samples were buffer-exchanged to 20 mM NaH₂PO₄ and 20 mM NaCl at pH 6.8 using Amicon Ultra (Millipore).

For the ATP–Mg solution preparation, ATP (ATP·Na₂, BBI, 987-65-5) was dissolved in water at a final concentration of 500 mM and NaOH was added to adjust pH to 6.8. MgCl₂ (magnesium chloride hexahydrate, Diamond, 7792-28-6) was dissolved in water at a final concentration of 500 mM. The two solutions were mixed in the ratio of 1:1.

The PLD (M1–S165) of FUS, which connects the His-tagged GB1 protein through the cleavage site of TEV protease, was cloned, expressed, and purified following the same procedure as that for the RGG_RtoA protein. The final sample was prepared in 50 mM 2-(N-Morpholino)ethanesulfonic acid and 250 mM NaCl at pH 5.5.

Fluorescent labeling

A cyanine 3 maleimide (Cy3; AAT Bioquest, 142)-labeled protein sample, where Cy3 is connected to S386C on the RGG domain, was prepared by adding 1.5 times amount of Cy3 into the protein in the buffer containing 50 mM 4-(2-hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES) and 150 mM NaCl at pH 7.4. The excess Cy3 probe was removed by a desalting column (GE Healthcare). TNP-ATP triethylammonium salt (APExBIO Technology, B7066) was dissolved in water at a final concentration of 500 mM (NaOH added to adjust pH to 6.8) and then mixed with 500 mM MgCl₂ in the ratio of 1:1.

The GFP-tagged HNRNPG-RGG domain, N-terminus connecting the His tag through the flexible linker GSGSGS, was cloned, expressed, and purified with HisTrap^TM FF (GE Healthcare). The final sample was prepared in 20 mM NaH₂PO₄ and 20 mM NaCl at pH 6.8.

Characterization of phase separation properties

The formation of protein droplets was examined at 25°C with different RGG domain concentrations (100, 200, 250, and 300 μM) in the presence of the ATP–Mg complex (or ATP alone) at 0, 1, 5, 10, 15, 20, or 30 mM. The sample was mixed, kept at room temperature for 30 min, and then transferred to the self-made chamber. The chamber was observed under a Nikon A1 confocal laser-scanning microscope using a 60× oil objective (Nikon).

For the FRAP experiment, droplet samples were prepared using 300 μM RGG protein contaning 5% GFP-tagged RGG domain and 10 mM ATP. A circular region of interest (ROI) was drawn at the position of each droplet and bleached with 100% laser power for 60 sec at 488 nm. The mean fluorescence intensity from the ROI was collected until recovery was complete. The experiment was repeated for three times.

NMR experiments

The protein samples for the NMR experiment were prepared in the respective buffers with a 10% D₂O addition. Phase separation was induced by mixing 300 μM ¹⁵N-labeled RGG domain with 10 mM ATP in a glass centrifuge tube for centrifugation at 7000 rpm for 10 min at 7°C. The phase separation formed tiny droplets that coalesced into large droplets, which were transferred to the NMR sample tube (NORELL^®, NI5CCI-B). The above steps were repeated for several times until enough NMR samples were obtained (40 ml of 300 μM ¹⁵N-labeled RGG domain sample was consumed).

All NMR experiments were performed at 298 K on Bruker 600 MHz and 700 MHz spectrometers equipped with cryogenic probes. The NMR data were further processed using TopSpin 3.5 (Bruker), NMRPipe 2016 (Delaglio et al., 1995), and CCPNmr Analysis V2.4 (Vranken et al., 2005), respectively. A series of NMR experiments were acquired for backbone assignment, including ¹H–¹⁵ N heteronuclear single quantum correlation (HSQC), etc.

NMR titration experiment

The initial ¹⁵N-labeled protein sample was prepared at 200 μM. The ¹H–¹⁵N HSQC spectra were recorded with the addition of ATP at the final concentration of 0, 5, or 10 mM.

Water exchange experiment

Water-selective experiments were performed using the CLEANEX-PM approach and recorded using a standard pulse sequence in Topspin 3.5 (fhsqccxf3gpph). The mixing times (T_m) were set at 0, 10, 15, 20, and 30 ms. ATP was added to the protein sample at the final concentration of 0, 5, or 10 mM. The peak volumes corresponding to different T_m and ATP concentrations were processed, and exchange rates were fitted using the algorithm described by Hwanga et al. (1998).

Diffusion coefficient measurement

Pulsed field gradient NMR diffusion experiments on RGG and PLD, with or without ATP, were performed as pseudo-2D experiments using the standard Bruker pulse sequence with a gradient strength from 5% to 95%. To evaluate the solution viscosity in the presence or absence of ATP, 0.02% DSS (Sigma-Aldrich, 178837) was added to the sample. The data were analyzed by integrating resonance corresponding to the sidechains in RGG and PLD. The diffusion coefficient was fitted using the following equation:

$$\begin{equation}{{I}_G} = {{I}_0}\exp \left[ { - {{{\left( {2\pi \gamma \sigma G} \right)}}^2}D\left( {\Delta - \frac{\sigma }{3}} \right) \times {{{10}}^4}} \right].\end{equation}$$

Due to the more complex dynamic variations of FUS-PLD, a single exponential function cannot accurately fit the diffusion coefficient. Therefore, we employed a dual exponential function to fit the diffusion coefficient using the following equation:

$$\begin{eqnarray} {{I}_G} &=& {{I}_0}\exp \left[ { - {{{\left( {2\pi \gamma \sigma G} \right)}}^2}D1\left( {\Delta - \frac{\sigma }{3}} \right) \times {{{10}}^4}} \right] \nonumber\\ && +\, {{I}_0}\exp \left[ { - {{{\left( {2\pi \gamma \sigma G} \right)}}^2}D2\left( {\Delta - \frac{\sigma }{3}} \right) \times {{{10}}^4}} \right].\end{eqnarray}$$

I₀ is the signal intensity at a gradient strength of zero, G is the gradient strength, D is the diffusion coefficient, σ is the gradient pulse duration, and △ is the diffusion time.

NMR relaxation experiment

The ¹⁵N-labeled protein was prepared at 200 μM and 10 mM ATP was added. Motions of the backbone of RGG were measured at 600 MHz using standard pulse sequences (hsqct1etf3gpsi.2, hsqct2etf3gpsi, and hsqcnoef3gpsi). Delay for gradient recovery was set to 16.96, 169.6, 339.2, and 508.8 ms for R₂ experiments and 20, 200, 400, 600, and 700 ms for R₁ experiments, respectively. The recovery time for the XNOE experiment was set to 5 sec.

PRE experiment

The paramagnetic probe-labeled protein, where S-(1-oxyl-2,2,5,5-tetramethyl-2,5-dihydro-¹H-pyrrol-3-yl) methyl methanesulfonothioate (MTSL; Toronto Research Chemicals, O875000) was connected to S363C on the RGG domain, was prepared by adding two times amount of MTSL into the protein in the buffer containing 50 mM HEPES and 150 mM NaCl at pH 7.4. The excess MTSL probe was removed by a desalting column (GE Healthcare). The ¹H–¹⁵N HSQC spectra were recorded with or without MTSL labeling. For intermolecular PRE experiments, 100 μM MTSL-labeled ¹⁴N protein was mixed with 100 μM ¹⁵N-labeled protein. ATP was added at the final concentration of 10 or 30 mM. The diamagnetic sample was prepared by mixing 100 μM ¹⁴N protein and 100 μM ¹⁵N-labeled protein without MTSL labeling. The relaxation delay (d1) was set to 5 sec.

SAXS experiment

Protein samples for the SAXS experiments were prepared at 50, 75, and 100 μM. ATP was added to the samples at a 50-fold concentration (2.5, 3.75, and 5 mM, respectively). All SAXS experiments for the RGG domain and ATP were collected at the BL19U2 beamline at the National Facility for Protein Science Shanghai. Each time, 1-sec exposure time was used, and 20 frames were recorded and averaged for further analysis. The scattering data for the corresponding buffer and the substrate from the sample data were also recorded. The theoretical scattering curve was calculated from the related PDB file using CRYSOL modules in the ATSAS 2.8 software package (Franke et al., 2017). The PDDF was calculated from the scattering curve using the PRIMSQT module in ATSAS 2.8.

Acknowledgements

We thank the staff of the BL19U2 beamline at the National Facility for Protein Science Shanghai and Shanghai Synchrotron Radiation Facility for assistance during SAXS data collection.

Funding

The work was supported by grants from the National Natural Science Foundation of China (92353304, 31971155, and 21991081) and the Youth Innovation Promotion Association of the Chinese Academy of Sciences (2020329).

Conflict of interest: none declared.

Author notes

These authors contributed equally to this work.