A Molecular Grammar Governing the Driving Forces for Phase Separation of Prion-like RNA Binding Proteins

Abstract

Proteins such as FUS phase separate to form liquid-like condensates that can harden into less dynamic structures. However, how these properties emerge from the collective interactions of many amino acids remains largely unknown. Here, we use extensive mutagenesis to identify a sequence-encoded molecular grammar underlying the driving forces of phase separation of proteins in the FUS family and test aspects of this grammar in cells. Phase separation is primarily governed by multivalent interactions among tyrosine residues from prion-like domains and arginine residues from RNA-binding domains, which are modulated by negatively charged residues. Glycine residues enhance the fluidity, whereas glutamine and serine residues promote hardening. We develop a model to show that the measured saturation concentrations of phase separation are inversely proportional to the product of the numbers of arginine and tyrosine residues. These results suggest it is possible to predict phase-separation properties based on amino acid sequences.

Keywords: FUS; PLD; cation-π; intrinsically disordered; low complexity; membraneless compartments; phase separation; prion-like; prion-like RNA binding proteins; saturation concentration.

Figure 1. The saturation concentration of FUS family proteins can span orders of magnitude

(A) Domain structures of FUS family proteins. PLD: prion-like domain; RRM: RNA recognition motif; ZnF: zinc finger; CC: coiled coil; NTD: N-terminal domain. The intrinsically disordered regions are colored in grey. (B) SDS-PAGE of the 22 FUS family proteins tagged with MBP and GFP. (C) Phase separation of FUS family proteins at the physiological salt concentration. Phase separation was induced by reducing the salt concentration from 500 mM to 150 mM through dilution. Bar, 5 μm. (D) Phase separation quantified by the relative amount of condensed protein versus the protein concentration. For the definition of the relative amount of condensed protein, see STAR Methods. Images from (C) were used for the quantification. The saturation concentration is indicated by a red arrow. All data are expressed as the mean ± the standard deviation (SD). See also Figure S1, Tables S1, S2 and S4.

Figure 2. Saturation concentration of FUS is governed by the collective interactions amongst tyrosine residues in the PLD and arginine residues in the RBD

(A) Domain structures of FUS and EWSR1. RBD: RNA binding domain. (B) RBD depletion significantly increases the saturation concentration. Unless specifically mentioned, all the phase separation assays in Figure 2 were conducted at 75 mM KCl. In the mixing assays, the PLD and RBD were mixed at a ratio of 1:1. All data are expressed as the mean ± the standard deviation (SD). (C) Whole genome analysis for tyrosine and arginine number within the disordered regions of human proteins. Shown is a two-dimensional histogram quantifying the frequencies of the disordered regions with specific number of tyrosine and arginine residues. (D) Tyrosine to serine substitution in the PLD or arginine to glycine substitution in the disordered regions of RBD increases the saturation concentration of the full-length FUS. (E) RBDs with distinct arginine number modulate the saturation concentration differently. (F) A cartoon summarizing the phase behavior of the prion-like domain and the full-length FUS. (G and H) Recruitment of clients by scaffold droplets appears to scale with the tyrosine and arginine number. Bar, 5 μm. To show the weak recruitment of some clients, we increased the signal from the clients by the same degree, and this caused the apparent layering of FUS and the clients in the merged channel. The clients are actually well mixed with FUS in the droplets. See also Figures S2 and S3.

Figure 3. Manipulating arginine number bypasses the need for G3BP1 in formation of FUS granules in vivo

(A) Distribution of the newly added arginine residues in FUS(27R) mutant. NLS, nuclear localization signal. The mutant was designed according to the distribution of tyrosine and arginine residues in TAF15 RBD (Figure S3A), which has a relatively low saturation concentration. (B) Addition of arginine residues in FUS lowers the saturation concentration of FUS in vitro. Phase separation assays were conducted at 150 mM KCl. All data are expressed as the mean ± the standard deviation (SD). (C and D) Granule formation by FUS variants in the absence of stress. Shown are the images for HeLa cells transfected with GFP tagged FUS WT (∆NLS) or 27R (∆NLS), in which the NLS was deleted. We used the ∆NLS variants because unlike the full-length WT which dominantly localizes in the nuclear, full-length 27R localizes mainly in the cytoplasm. Bar, 5 m; (D), quantification of the granule number versus cyto mean intensity (mean intensity of the cytoplasm) of FUS variant. 189 and 134 cells were used to quantify the number of the preformed granules in the cells transfected with WT (∆NLS) and 27R (∆NLS), respectively. (E) The stress granule nucleator G3BP1 is absent from the 27R (∆NLS) granules in the absence of stress. Stable HeLa cell line expressing C-terminal mCherry tagged human G3BP1 were transfected with the plasmid expressing the GFP tagged 27R (∆NLS). Bar, 5 μm. (F) Addition of arginine residues and substitution of phenylalanine with tyrosine residues in hnRNPA1a PLD lower the saturation concentration. hnRNPA1a is the short isoform of hnRNPA1. Phase separation assays were conducted at 150 mM KCl. The positions of tyrosine and arginine residues for each construct are indicated by green and red lines, respectively. All data are expressed as the mean ± the standard deviation (SD). See also Figure S4.

Figure 4. Distinct saturation concentration is governed by a code of associating aromatic and basic amino acids

(A) Representative interactions amongst aromatic and basic amino acids in FUS family proteins. (B) Tuning phase separation driving forces by different combinations of aromatic and basic amino acids. The in vitro phase separation assays in Figure 4 were conducted at 75 mM KCl. Bar, 5 μm. (C) Phase separation of FUS tyrosine mutants quantified by the relative amount of condensed protein versus the protein concentration. All data are expressed as the mean ± the standard deviation (SD). (D) Foci formation by FUS tyrosine mutants at DNA damage sites. Bar, 10 m. All data are expressed as the mean ± the standard deviation (SD). (E) Phase separation of FUS arginine mutants quantified by the relative amount of condensed protein versus the protein concentration. All data are expressed as the mean ± the standard deviation (SD). (F) Foci formation by FUS arginine mutants at DNA damage sites. Bar, 10 m. All data are expressed as the mean ± the standard deviation (SD). See also Tables S3 and S4.

Figure 5. Tuning protein phase behavior by electrostatic interactions

(A) Phase separation of FUS PLD variants in the absence or presence of crowding agents. Bar, 5 μm. (B) Addition of negative charges promotes phase separation of FUS in vitro. The phase separation assays were conducted at 75 mM KCl. All data are expressed as the mean ± the standard deviation (SD). (C) Foci formation by FUS 6D at DNA damage sites. Bar, 10 m. All data are expressed as the mean ± the standard deviation (SD). The images showing foci formation by WT are the same as those in Figures 4D and 4E, because the assays in Figures 5C, 4D and 4E were done in parallel. (D) Comparison of c_sat values of FUS variants determined by measuring the supernatant concentration at 75 mM KCl (see also Table S4). The inset shows the c_sat values of FUS FL and PLD in the presence of dextran (Figure S2D). (E) A hierarchy of interactions involving aromatic and charged residues governs the saturation concentration. See also Figure S5 and Table S4.

Figure 6. Glycine and serine/glutamine residues act oppositely to modulate the material property of droplets after phase separation

(A) Distribution of glycine, serine and glutamine residues within the PLD region of the full-length protein and the relative mutants used. (B) Glycine, serine and glutamine residues have minor effects on the saturation concentrations for phase separation. Phase separation assays were conducted at 75 mM KCl. All data are expressed as the mean ± the standard deviation (SD). (C) Droplets formed by G→A mutant fuse slower than those formed by WT. Size-normalized median fusion times are 32.9 s/μm for G→A and 0.093 s/μm for WT, respectively. (D) Droplets formed by G→A mutant stop fusing after 40 min while WT droplets do not change dynamics (quantification). Fusion was monitored over time and was scored successful (1) or not (0) according to whether the resulting droplet relaxed to approximately spherical shape within 2 min. (E) Glycine residues maintain droplet liquidity. Bar, 3 μm. (F) Serine and glutamine residues promote hardening. Bar, 5 μm. (G) Dynamic change of the condensates formed by WT, S→A or Q→G mutant over time. All data are expressed as the mean ± the standard deviation (SD). (H) Mobile fraction of FUS variants over time. The mobile fraction of FUS variants was obtained from the FRAP experiments. See also Figure S6 and Movie S1.

Figure 7. Model to explain the phase behavior of FUS family proteins

(A) Hierarchical organization that gives rise to formation of condensates depicted here as spherical droplets. Collective complementary interactions amongst the associative motifs such as tyrosine and arginine residues drive self-association, and electrostatic interactions play a modulatory role. The amino acid compositions of spacers, which are mainly enriched in glycine, glutamine and serine residues, largely determine intra-condensate dynamics of proteins. While the saturation concentration is governed by a code of associating aromatic and charged residues (Figure 5E), the spacers have little effect on the driving forces for droplet formation. (B) Material properties are controlled mainly by glycine, serine and glutamine residues. Glycine-rich spacers yield highly dynamic droplets by imparting flexibility onto the amino acid backbone. This is manifest as droplets that undergo rapid internal re-arrangement and rapid macroscopic relaxation. The presence of serine and glutamine residues in spacers appears to reduce the internal re-arrangements and relaxation, leading to the reduction of droplet liquidity of droplets. (C) Comparison of the inverse of the product of the numbers of tyrosine and arginine residues (ordinate) to the measured saturation concentrations (abscissa). The c_sat values determined directly by measuring the supernatant concentrations were used (see Figure S1 and Table S4). The measured numbers were fit to a model of the form c_sat = k(n_Tyr n_Arg)^–1. Here, n_Tyr and n_Arg denote the number of tyrosine and arginine residues, respectively. The only adjustable parameter in the fitting procedure was k, which places the valence on the same concentration scale as the experimental data. The fit shown here, which was performed on a logarithmic scale, uses a value of k = 6.5 mM. (D) Identifying potential scaffold proteins that could undergo phase separation at physiologically relevant concentrations. This figure includes a two-dimensional histogram quantifying the frequencies of the disordered regions with the specific number of tyrosine and arginine residues. Overlaid atop this histogram are contours showing the predicted values of saturation concentration (c_sat) in μM. The plot shows the locations, in terms of numbers of tyrosine and arginine residues as well as the calculated values of c_sat for proteins with predicted saturation concentrations that are on a par with the FET family of proteins. See also Figure S7, Tables S4 and S5.