Characterization of design grammar of peptides for regulating liquid droplets and aggregates of FUS

Liquid droplets of aggregation-prone proteins, which become hydrogels or form amyloid fibrils, are a potential target for drug discovery. In this study, we proposed an experiment-guided protocol for characterizing the design grammar of peptides that can regulate droplet formation and aggregation. The protocol essentially involves investigation of 19 amino acid additives and polymerization of the identified amino acids. As a proof of concept, we applied this protocol to fused in sarcoma (FUS). First, we evaluated 19 amino acid additives for an FUS solution and identified Arg and Tyr as suppressors of droplet formation. Molecular dynamics simulations suggested that the Arg additive interacts with specific residues of FUS, thereby inhibiting the cation–π and electrostatic interactions between the FUS molecules. Second, we observed that Arg polymers promote FUS droplet formation, unlike Arg monomers, by bridging the FUS molecules. Third, we found that the Arg additive suppressed solid aggregate formation of FUS, while Arg polymer enhanced it. Finally, we observed that amyloid-forming peptides induced the conversion of FUS droplets to solid aggregates of FUS. The developed protocol could be used for the primary design of peptides controlling liquid droplets and aggregates of proteins.

www.nature.com/scientificreports/ Arg and Tyr additives suppressed FUS droplet formation. We examined the effect of 19 amino acid additives on the formation of MBP-FUS droplets. The amino acids were used at 40 mM except for Tyr, which used at 2 mM, because of the low solubility of Tyr. Cys was not tested, because Cys can form disulfide bonds to the Cys residues of proteins in general. The addition of Gly did not alter the scattering intensity from the MBP-FUS solution (Fig. 1C). Similarly, the other amino acids, except for Arg, Lys, Tyr, and Phe, did not affect droplet formation (Fig. 1C). In contrast, Arg and Tyr significantly decreased the scattering intensity (p = 0.044 for Tyr and Gly and 0.039 for Arg and Gly, two-tailed t-test). Arg suppressed droplet formation at concentrations greater than 5 mM, and the scattering intensity from the droplet was saturated at 40 mM Arg ( Supplementary Fig. S2A). For Lys or Phe, the scattering was reduced to a less extent (p = 0.063 for Lys and 0.074 for Phe, two-tailed t-test with Gly). Accordingly, the four amino acids suppress the formation of MBP-FUS droplets. Next, we examined the effect of the amino acids on the size and shape of the droplets using DIC microscopy. Spherical droplets of MBP-FUS were observed in the presence of any of the 19 amino acids ( Fig. 1D and supplementary Fig. S3). Arg, Lys, Tyr, and Phe did not affect the relative distribution of the cross-section area and circularity of droplets compared to Gly as the control (Fig. 1E).
We examined how such effective additives suppress the formation of MBP-FUS droplets using MD simulations. Molecular dynamics of MBP-FUS was simulated in the presence of Arg, Lys, Phe, or Gly at 100 mM or Tyr at 10 mM. In the simulations, MBP-FUS exhibited a rod-like shape, consistent with that observed through small-angle X-ray scattering analysis 20 ( Fig. 1F and Supplementary Fig. S4). We analyzed the number of contacts between the additives and residues of MBP-FUS, where the contact was defined as a distance of less than 6.5 Å between the centers of two side chains, except for Gly, for which Cα was used (Fig. 1G). We focused on five residues that participate in droplet formation (Arg, Lys, Phe, Tyr, and Asp) 9 and Glu with a negative charge. In particular, the Arg additive exhibited more contacts with Tyr residues of the LC domain and Asp/Glu residues of the RNA-binding domain (Asp residues of the RGG1 domain and Asp/Glu residues of the RRM domain, Supplementary Fig. S5A) than the Gly additive. The cation-π, π-π, and electrostatic interactions between the Arg additives and FUS molecules inhibit the interactions among the FUS molecules. In contrast, the Tyr additive formed contacts with Arg residues of the RNA-binding domain of FUS via cation-π interaction. Accordingly, we propose that Arg and Tyr interact with the droplet-forming residues of FUS, thereby competitively inhibiting the interactions among the FUS molecules and suppressing droplet formation (Fig. 1H).

Poly-Arg additive enhanced FUS droplet formation by bridging FUS molecules.
We hypothesized that, compared with Arg monomers, Arg polymers interact more effectively with FUS, because Arg polymers have the potential to form multivalent interactions. To test this hypothesis, we investigated the effect of Arg polymers on the formation of MBP-FUS droplets. Poly-Arg (15-70 kDa; median, 200-mer; R200) significantly enhanced the scattering intensity from MBP-FUS droplets, at a concentration of 40 mM with respect to Arg units (p < 0.0001, two-tailed t-test, Fig. 2A); in contrast, Arg monomer suppressed droplet formation. Poly-Arg effectively promoted droplet formation at concentrations greater than 0.5 mM with respect to Arg units (Supplementary Fig. S2B). Additionally, poly-Lys (30-70 kDa; median, 220-mer) increased the scattering intensity compared to Lys monomer (p = 0.045, two-tailed t-test, Fig. 2A). The effect of poly-Lys was lesser than that of poly-Arg likely due to the weaker cation-π and electrostatic interactions with FUS (Fig. 1F). The DIC images displayed the MBP-FUS droplets in the presence of poly-Arg (Fig. 2B). The occurrence percentage of large droplets with cross-section area greater than 3 µm 2 was increased, coincident with the decrease in small droplets with 1-2 µm 2 (Fig. 2C). The circular shape of the droplets was not considerably changed (Fig. 2C). In contrast, poly-Lys did not affect the shape and size of the droplets, supporting the observation that the poly-Lys was less effective in regulating droplet formation than poly-Arg ( Supplementary Fig. S6). We additionally confirmed the uptake of poly-Arg labeled with a fluorophore, Alexa488, into the MBP-FUS droplets using DIC and fluorescence microscopy (Fig. 2D). Furthermore, we confirmed the effect of poly-Arg on the formation of FUS droplets following the cleavage of the MBP tag by TEV protease in the absence of dextran (p = 0.01, two-tailed t-test, Fig. 2E). Overall, the results demonstrated that the polymer of Arg significantly promoted the FUS droplet formation, while Arg monomers exhibited the opposite effect.
To identify the effective peptide length for promoting FUS droplet formation, we examined the effect of Arg peptide length on the formation of MBP-FUS droplets. The peptide concentrations were adjusted to 40 mM in the Arg monomer unit. The scattering intensities of Arg dimer (R2) and trimer (R3) were higher than that of the Arg monomer (p < 0.02, two-tailed t-test, Fig. 2F), comparable to that in the absence of additives (Fig. 1C). In 5-mer peptide (R5), the intensity was further enhanced (p = 0.008, two-tailed t-test, Fig. 2F). The MD simulations of MBP-FUS in the presence of R, R2, R5 or R10 showed similar patterns, that is, an increase in the absolute number of contacts in the LC-and RNA-binding domains with increase in the Arg peptide length (Fig. 2G). Additionally, the polymerization of Arg did not affect the binding regions over the FUS sequence (Supplementary Figs. S5B and S7). Collectively, the polymer length of Arg affects the droplet formation tendency, likely because the more Arg residues the polymers have, the more strongly they interact with FUS molecules via cation-π and electrostatic interactions, leading to FUS bridging and hence FUS droplet formation (Fig. 2H). www.nature.com/scientificreports/ assumed the action by R10 inside the cell, since R9 and other arginine-rich peptides are known as a cell-penetrating peptide 24 . In contrast, the addition of Arg decreased the area of FUS clusters significantly (Fig. 3A,B). Therefore, these results demonstrated that Arg and poly-Arg additives can work for regulating FUS cluster formation in cells in a way similar to in vitro conditions.  Fig. S8). In the presence of the Arg additive, the size of the FUS aggregates decreased (Fig. 4A). Thin filament structures in the aggregates were observed. In contrast, the poly-Arg (R200) additive increased the aggregate size (Fig. 4A). Similar result was obtained after incubation of MBP-FUS without MBP-tag cleavage for a week at 4 °C (Supplementary Fig. S9). In both cases, the fluorescence images suggested the presence of amyloids in the aggregates (Fig. 4A).

Arg and poly-Arg additives regulated FUS cluster formation similarly in cells. To test if
To determine the percentage of FUS molecules incorporated into the aggregate, we determined the concentration of soluble molecules in the supernatant by measuring the OD at 280 nm after centrifuging the samples (Fig. 4B). The OD at 280 nm (absorbance by FUS plus scattering) was corrected by subtracting the OD at 350 nm (scattering). In the absence of additives, 7% of the FUS molecules (corresponding to 2.1 µM) were in solution, indicating that 93% of the FUS molecules was included into the aggregate (Fig. 4C). When poly-Arg was added, the percentage of FUS in the aggregate slightly increased to 95%. In contrast, the Arg additive significantly reduced the percentage in the aggregate to 58%. Overall, the results demonstrated that the Arg additive suppresses the FUS aggregate formation, while poly-Arg promotes it. The similarity in the effect of the additives for the droplets and aggregates implies that aggregate regulation is achieved through regulation of the intermediate state (liquid droplets) (Fig. 4D).

QQQQ and NNNN additives at high concentrations drove liquid droplets of FUS to solid aggregations.
Since the cross-beta structures formed by intermolecular H-bonds participate in the formation of liquid droplets, as well as gel-like droplets and amyloid fibrils 7,10 , we evaluated the effect of QQQQ (Q4) and NNNN (N4), amyloid-related peptides, on the droplet formation of MBP-FUS at 5 µM. Q4 and N4 www.nature.com/scientificreports/ at 10 mM, corresponding to 40 mM of monomer, increased the scattering intensity to some extent, suggesting the enhancement of droplet formation (Fig. 5A), where the OD 350 of Q4 or N4 was subtracted from that of the sample containing MBP-FUS plus Q4 or N4, because Q4 and N4 produced a small scattering effect. In contrast, YGQS, composed of four residues frequently observed in LLPS proteins, did not affect droplet formation. The DIC images demonstrated spherical droplets in the presence of the three different peptides (Fig. 5B). The crosssection area and circularity analysis of the MBP-FUS droplets indicated no significant effect upon the addition of Q4 and N4 (Supplementary Fig. S10). Q4 and N4 additives at low concentrations slightly enhanced FUS droplet formation without modulating the shape and size distribution. We examined the effect of Q4 and N4 additives in MBP-FUS droplets at higher concentrations. The solution contained 15 µM MBP-FUS and 30 mM Q4 or N4. Non-spherical aggregates of MBP-FUS were generated in the presence of Q4 and N4 additives (Fig. 5C). The aggregates exhibited a high fluorescence intensity from PicoGreen, which suggested amyloid formation in the aggregates (Fig. 5C). As a control, spherical droplets were detected in the absence of Q4 or N4 additives (Fig. 5C). Additionally, the droplets exhibited high fluorescence intensity from PicoGreen, implying amyloid formation in the FUS droplets, consistent with the results of a previous study 26 . The circularity distribution of each droplet or aggregate became broader in the presence of Q4 or N4 additives, supporting non-spherical aggregates induced by the additives (Fig. 5D). In addition, the additives increased the frequency of droplets or aggregates with a cross-section area greater than 10 µm 2 (Fig. 5D). Additionally, www.nature.com/scientificreports/ MBP-FUS labeled with Alexa488 was incorporated into the aggregates, confirming the aggregation of MBP-FUS (Fig. 5E). Thin filament structures in the aggregates supported amyloid formation, especially in the presence of Q4 (Fig. 5E). The time-lapse DIC images demonstrated that MBP-FUS formed non-spherical aggregates; subsequently, the aggregates assembled and became larger (Fig. 5E). Long storage with Q4 or N4 solution promoted the aggregation of MBP-FUS. In addition, Q4 and N4 themselves formed small non-spherical aggregates containing amyloids under this condition ( Supplementary Fig. S11). Therefore, we propose that the small amyloid core formed by Q4 and N4 induced the conversion of the FUS droplets to FUS aggregates.

Discussion
In this study, we propose an experiment-guided protocol for designing peptides that can regulate the formation of liquid droplets and solid aggregates. Owing to the tremendous number of peptide candidates, experimental screening may not be successful. In contrast, the developed protocol can efficiently explore droplet-regulating peptides. The protocol is essentially composed of two simple steps, and it can be widely applied to various proteins of interest. The first step is evaluation of the effect of 19 amino acid additives on the droplets of a protein of interest and identification of effective amino acid that can suppress droplets, such as Arg or Tyr for FUS. The second step is synthesis of the polymer of the identified amino acid. The synthesized polymer would bridge at least two protein molecules and promote droplet formation, such as poly-Arg for FUS. The use of the effective amino acids in designing the peptide can significantly reduce the pool of potential peptide candidates. Using this protocol, we succeeded in identifying additives that suppressed and enhanced the formation of FUS droplets and aggregates.
To alter the state of proteins from liquid droplets to aggregates or amyloids, we need to consider intermolecular β-sheet formation between the backbones. The above protocol, focusing on the side chain effect, does not consider this aspect. Nevertheless, the β-sheet-forming peptide serves for this purpose. The Q-rich region of amyloid-forming proteins, such as the LC domain of FUS forms intermolecular β-sheets, which leads to aggregation. The β-sheet-forming peptide supports this intermolecular β-sheet formation for the protein of interest. N4 or Q4 can form amyloids by itself [27][28][29] (Supplementary Fig. S11) and induce FUS to form liquid droplets at low concentrations and to convert from the droplet state to the aggregate state, including amyloids, at high concentrations (Fig. 5). This phenomenon may be similar to cross-amyloid aggregation and cross-seeding: coaggregation occurs between different proteins or peptides 30 . For example, poly-Glu enhances Tau aggregation 31 . We propose that the cross-β structure of FUS is induced near the small amyloid core of Q4 and N4 and that it propagates over other FUS molecules, thus promoting the aggregation of FUS.
The experiment-guided protocol and MD simulations provide information on the design grammar for LLPSregulating peptides. As an additive, poly-Arg, in this study, accelerated FUS droplet formation more effectively than poly-Lys. Arg and Lys in the polymer possess a positive charge, exhibiting cation-π and electrostatic interactions with FUS. Arg may exhibit a stronger cation-π interaction than Lys 32 . In addition, Arg enables π-π interactions via its π-bonded guanidium group 33,34 . Such different interactions between Arg and Lys resulted in the Arg additives forming a greater number of contacts with Tyr and Phe residues of FUS compared to Lys additives, according to MD simulation (Fig. 1G). Consistent with this result, delocalization of positive charge within the guanidium group of Arg can result in hydrogen bond formation with the backbone of FUS 35 (Fig. 1G).
Poly-Arg additive promotes the droplet formation of FUS by contacting to Tyr residues of the LC domain, Asp residues of the RGG1 domain, and Asp/Glu residues of the RRM domain. These interactions are similar with those formed in FUS droplets, such as Arg residues of the RNA-binding domain and Tyr residues of the LC domain 9,10 and electrostatic interactions with Asp 9 . Accordingly, poly-Arg bridges the key residues of FUS which participate in the droplet formation.
In conclusion, we propose an experiment-guided protocol for characterizing the design grammar of peptides that can regulate the formation of liquid droplets and solid aggregates. The design grammar would depend on the proteins of interest; however, Arg and poly-Arg, which were identified as effective additives in this study, might serve for other proteins, because the LLPS proteins provide many donors for cation-π, π-π, and electrostatic interactions with these additives. In drug design, this protocol could be used for the primary design of peptides targeting disease-related proteins formed via liquid droplets. The designed peptide might be further optimized toward achieving aggregate suppression with high affinity using other established methods such as phage display. www.nature.com/scientificreports/ ther purification. R2, R3, and R10 were purchased from GenScript. R10, R5, Q4, N4, and YGQS were synthesized using a standard Fmoc-based solid-phase peptide synthesis. These peptides were cleaved from the resin using trifluoroacetic acid at 95%, water at 2.5%, and triisopropylsilane at 2.5%. The cleaved peptides were precipitated in cold diethyl ether. The resultant peptides except for R10 were dissolved in 0.1% TFA (HCl for R10) and lyophilized. The purity and identity of each peptide were verified using HPLC and mass spectrometry (LTQ Orbitrap XL ETD, Thermo Fisher Scientific Inc.).

Materials and methods
Labeling with a fluorophore. The cysteine mutant of MBP-FUS was labeled with Alexa488 (Thermo Fisher) in a solution containing 100 mM Tris, 500 mM KCl, and 1 mM TCEP using maleimide chemistry and purified using gel filtration (PD MiniTrap G-25; GE Healthcare). Because MBP-FUS, bound nonspecifically to unreacted dyes, was eluted, the bound dyes were further removed using a centrifugal filter (Amicon Ultra-4, Millipore) by increasing the concentration of KCl to 1.5 M. The N-terminus of poly-R was labeled with Alexa488 in a solution containing 100 mM NaHCO 3 and 500 mM NaCl (pH 8.3) using succinimidyl ester chemistry and purified using gel filtration. The labeling ratios were determined to be 1. Scattering measurements. We analyzed light scattering from the droplets of MBP-FUS or FUS by measuring the OD 350 using an absorbance spectrometer (NanoDrop One; Thermo Fisher).
DIC microscopy. We used the DIC mode of the inverted microscope (IX-73; Olympus, Tokyo, Japan), as described previously 36 . The sample solution was casted on the coverslip and covered by the slide glass. The DIC images were captured at 40 × or 60 × magnification at 21 °C. The cross-section area and circularity of the droplets were calculated using the Image J software.

Fluorescence microscopy for in vitro measurements.
To evaluate the uptake of poly-R-Alexa488 or MBP-FUS-Alexa488 into droplets and aggregates of non-labeled MBP-FUS, we used the fluorescence mode of the inverted microscope (IX-73; Olympus) 36 . The excitation and emission wavelengths were 470-490 nm and 515-550 nm, respectively. Fluorescence images were captured at 21 °C.

MD simulation. MD simulations were conducted to investigate interactions between MBP-FUS and several
amino acid additives. Because the MBP-FUS structure was not solved, it was generated from the crystal structures of MBP (PDB ID: 1Y4C), the RRM domain (PDB ID: 2LA6), and the ZnF domain (PDB ID: 6G99) 37 of FUS. The disordered regions of FUS were modeled as extended structures. Through connecting these structured and disordered domains, the overall structure of MBP-FUS was constructed. To equilibrate the structure, 300 ns MD simulation was performed with implicit solvent. The protein was described using the AMBER14SB force field 38 and Generalized Born energy 39 . Next, the protein was simulated in a water/amino acid additive box. These molecules were placed in a dodecahedron box with 24.1 nm sides. The system contained one molecule of MBP-FUS, 311,439 water molecules, 589 amino-acid additive molecules corresponding to 100 mM, and 895 sodium and chloride ion molecules corresponding to 150 mM. When the total charge of the system was not neutral, additional sodium or chloride ions were also included. The protein and amino acid additives were described using the AMBER99SB force field 40 . Since the zwitterion form of amino acids is not prepared in this force field, we built their molecular model with restrained electrostatic potential (RESP) charges 41 (Supplementary text). Water was described by the TIP3P model 42 . Sodium and chloride ion models were obtained from the literature 43