Glycine-rich peptides from FUS have an intrinsic ability to self-assemble into fibers and networked fibrils

Abstract

Glycine-rich regions feature prominently within intrinsically disordered regions (IDRs) of proteins that drive phase separation and the regulated formation of membraneless biomolecular condensates. Interestingly, the Gly-rich IDRs seldom feature poly-Gly tracts. The protein Fused in Sarcoma (FUS) is an exception. This protein includes two ten-residue poly-Gly tracts within the prion-like domain (PLD) and at the interface between the PLD and the RNA binding domain (RBD). Poly-Gly tracts are known to be highly insoluble, being potent drivers of self-assembly into solid-like, fibrils. Given that the internal concentrations of FUS and FUS-like molecules cross the high micromolar and even millimolar range within condensates, we reasoned that the intrinsic insolubility of poly-Gly tracts might be germane to emergent fluid-to-solid transitions within condensates. To assess this possibility, we characterized the concentration-dependent self-assembly for three non-overlapping 25-residue-long Gly-rich peptides derived from FUS. Two of the three peptides feature 10-residue long poly-Gly tracts. These peptides either form long fibrils based on twisted ribbon-like structures or self-supporting gels based on physical crosslinks of fibrils. Conversely, the peptide with similar Gly contents but lacking in a poly-Gly tract does not form fibrils or gels. Instead, it remains soluble across a wide range of concentrations. Our findings highlight the ability of poly-Gly tracts within IDRs that drive phase separation to undergo self-assembly. We propose that these tracts are likely to contribute to nucleation of fibrillar solids within dense condensates formed by FUS.

Graphical Abstract

INTRODUCTION

Membraneless biomolecular condensates enable spatial and temporal regulation of intracellular biochemical reactions^{1, 2}. Many of these condensates are thought to form and dissolve under the influence of reversible phase transitions driven by multivalent protein and nucleic acid molecules known as scaffolds^{1, 3, 4}. These molecules are biological instantiations of associative polymers⁵ that feature multiple stickers or cohesive motifs that form reversible physical crosslinks^6–8. The valence (number) and strengths of stickers determine the cohesive interactions that drive reversible phase transitions^6–10. Stickers are interspersed by spacers, and these sequence regions contribute to the cooperativity of phase transitions¹¹ that are driven by reversible physical crosslinks among stickers⁵. Phase transitions that give rise to condensates combine density transitions known as phase separation, and networking transitions known as gelation or percolation¹¹. The combination of phase separation and percolation gives rise to condensates that are best described as network fluids with an underlying condensate-spanning network structure^{6–8, 11, 12}. Depending on the timescales associated with the making and breaking of physical crosslinks, the condensates are concordant with being viscous or viscoelastic network fluids^{6, 13}.

Associative biomacromolecules can drive phase transitions through homotypic interactions¹⁴, a blend of homotypic and heterotypic interactions^{6, 12, 15, 16}, or purely heterotypic interactions^{4, 17}. Among the most well-studied systems in vitro are RNA binding proteins with intrinsically disordered regions (IDRs) that can drive phase transitions through homotypic interactions. Examples of such systems include full-length versions of proteins such as fused in sarcoma (FUS), other proteins from the FET family, and disordered prion-like domains (PLDs) excised from these and other systems^{8, 18–23}. Condensates can undergo hardening behavior over time through non-equilibrium transitions such as dynamical arrest and physical aging^24–27. Changes to material properties in condensates can also be realized through equilibrium solidification transitions and / or fibril formation^{28, 29}, wherein condensates serve as crucibles for driving equilibrium transitions^30–32. Mutations within PLDs of FUS and related proteins are associated with the formation of protein-rich deposits in the context of Amyotrophic Lateral Sclerosis (ALS)^{26, 30, 31, 33}. McKnight and coworkers showed that archetypal PLDs from FUS and related proteins can form self-supporting hydrogels characterized by physically crosslinked fibrils^{21–23, 34}. Structural studies using peptide-based microcrystals led Hughes et al., to propose that specific motifs known as low-complexity, aromatic-rich, kinked segments or LARKS provide the cohesive interactions to drive the formation of networks that scaffold peptide or protein-based hydrogels³⁵. These hydrogels are biochemical facsimiles of self-assembled fibrillar networks or SAFINS that are formed by synthetic associative polymers and proteins / peptides with or without low molecular weight organic gelators^36–40. It is noteworthy that hydrogels and fibrils form above concentrations that are in the high micromolar / millimolar range. Patel et al.,³⁰ as well as Molliex et al.,³¹ showed that systems such as the PLDs of FUS and hnRNPA1 and full-length hnRNPA1 can also undergo phase separation and percolation or the less nuanced liquid-liquid phase separation (LLPS) to form spherical, fluid-like condensates at bulk concentrations that are at least one- to two orders of magnitude lower than the threshold concentrations for fibril formation. Dense liquid-like condensates that form via phase separation and percolation or LLPS can become crucibles for forming solid-like fibrils that form networked, self-supporting gels^{30, 41}.

The valence and patterning of aromatic residues as well as the cationic residues such as Arg have been shown to be the main drivers of phase separation and percolation of PLDs and full-length FUS-like molecules^{8–10, 42–45}. Aromatic residues (Tyr / Phe) are the main stickers within PLDs and Arg residues are the main stickers within RNA binding domains (RBDs)⁸. In both domains, stickers are interspersed by spacers, and these are typically made up of Gly / Ser residues⁸. Theory and simulation predict that, to zeroth order, the effective solvation volumes of spacers determine the synergies between density and networking transitions¹¹, the cooperativity of phase transitions, the width of the two-phase regime, the interfacial tension between dilute and dense phases⁴⁶, and internal dynamics of molecules within condensed phases⁴⁷. These predictions have been tested in experiments directed at full-length FUS and FUS-like molecules as well as PLDs studied as autonomous units^{8, 10}. The key finding is that Gly, when compared to Ser, enhances the driving forces for phase transitions and gives rise to more rapid internal dynamics for molecules within condensates^{8, 10}.

Interestingly, one seldom finds more than 4–5 Gly residues in a row within various FET family proteins (Figure 1). An exception to this is the PLD and the region that straddles the PLD and RBD in FUS. Specifically, there are two sequence stretches viz., FUS^156−180 and FUS^211−235 that feature ten-residue-long poly-Gly tracts. The longest poly-Gly tract in the intervening stretch, FUS^181−210, is only four residues long. Poly-Gly tracts are of interest because these systems become highly insoluble above a length threshold of 6–7 residues^48–51. Further, investigations in ultra-dilute aqueous solutions⁵² have revealed that poly-Gly molecules in aqueous solvents behave like polymers in poor solvents providing the tract length is larger than 7–10 residues^{53, 54}. Long poly-Gly tracts are known to drive the formation of fibrillar solids with polyglycine II-like or other extended structures at the molecular level⁵⁵.

Figure 1: Analysis of Gly content and the presence of poly-Gly tracts of different lengths within FUS / FET family proteins.

(a) Compositional analysis showing the Gly content within eight FUS / FET family proteins from the human proteome. (b) Numbers of poly-Gly tracts of length n = 3, 4, 5, 6, 7, and 10 within each of the eight human FUS / FET family proteins. The sequence of FUS is unique in having two poly-Gly tracts of length 10.

Recent studies have shown that concentrations of FUS, FUS-like molecules, and PLDs can cross the millimolar thresholds within dense phases of condensates^{9, 56, 57}. Motivated by these observations we ask if sequence stretches with poly-Gly tracts self-assemble into long fibers or fibrillar networks at concentrations that are greater than 10² – 10³ μM? To answer this question, we performed comparative biochemical studies of the self-assembling properties of three 25-residue peptides from FUS: P1: FUS^156−180, P2: FUS^181−210, and P3: FUS^211–235. We find that P1 and P3 form, supramolecular fibers, and self-supporting fibrillar networks, respectively, whereas P2 is soluble throughout the concentration ranges we investigated. These findings suggest that sequence regions with poly-Gly tracts are likely to contribute as drivers of fiber formation or the formation of fibrillar networks within disordered dense phases formed by molecules like FUS.

MATERIALS AND METHODS

Materials:

Three Gly-rich peptides, each 25 residues long, were synthesized using solid-phase peptide synthesis. The uncapped peptides, designated as P1, P2, and P3, were excised from the full-length FUS protein. The sequences of each of the peptides are shown in Table 1. The peptides, purchased from peptides&elephants (https://www.peptides.de/), Germany, were 96% pure and were formulated as trifluoracetic acid (TFA) salts. The identities of the peptides were confirmed using mass spectrometry, and the peptides were used in the experiments without further purification. Thioflavin T (ThT), a fluorescent dye used for assays of fibril formation, was purchased from Sigma-Aldrich, Germany.

Table 1:

Details of the peptide sequences used in this work^†

Peptide	Amino acid sequence	% Gly	FCR	NCPR
P1: FUS156−180	-GQQNQYNSSSGGGGGGGGGGNYGQD-	44	0.04	−0.04
P2: FUS186−210	-SGGGSGGGYGNQDQSGGGGSGGYGQ-	56	0.04	−0.04
P3: FUS211−235	-QDRGGRGRGGSGGGGGGGGGGYNRS-	60	0.20	+0.12

^†The longest poly-Gly stretches within each sequence are bold-faced. FCR and NCPR refer to the fraction of charged residues and the net charge per residue, respectively. These parameters were computed using CIDER⁵⁸.

Ultraviolet circular dichroism (UV-CD) spectroscopy:

UV-CD spectra between 200 nm and 280 nm were recorded on an Applied Photophysics Chirascan plus spectrophotometer. Different concentrations of peptide samples were prepared by adding peptide stock into 50 mM phosphate buffer (PB), pH 7.4 at 25°C. The stock solutions (5 mM) were prepared by adding deionized water to the peptide and mixing thoroughly at 25°C. The pH of the stock solutions was highly acidic due to TFA. The pH was brought to neutral (~7.0) by titrating with 1 M KOH. We used quartz cuvettes (Hellma, Jena, Germany) of path length 1 mm. Each spectrum represents the average of six measurements, three measurements each from two independent sample preparations, where each measurement was the average of five successive scans. CD spectra were converted from machine units (mdeg) to mean residue ellipticity ([θ], deg cm² dmol⁻¹ residue⁻¹) using the equation [θ] = θ/(N−1)Lc, where θ is in machine units, N is the number of amino acids in the peptide, L is path length and c is molar concentration.

Table-top assays of forming self-supporting networks:

Each peptide was thoroughly mixed in Eppendorf tubes with deionized water at 25°C to generate 5 mM stock solutions. The pH was brought to ~7 using 1 M KOH. After mixing, the solutions were left undisturbed for six hours. Following this, the Eppendorf tubes were inverted, placed on a benchtop, and pictures were taken to obtain qualitative, macroscopic assessments of whether the solutions were clear, cloudy, or formed self-supporting gels.

ThT binding assay:

Fluorescence emission spectra of ThT, excited at 450 nm, were recorded between 470 nm and 580 nm on a TECAN 20M plate reader using excitation and emission bandwidths of 5 nm. Samples with different concentrations of peptides were prepared by diluting the peptide stocks (5 mM) with 50 mM Tris buffer pH 7.4 at 25°C. For each measurement, 10 μM of ThT was added. As a control, 50 mM Tris buffer pH 7.4 was used with 10 μM of ThT. Fluorescence spectra of peptide samples incubated at various concentrations were collected in the emission range between 470 nm to 580 nm. For each peptide, ThT binding assays were performed at three different concentrations viz., 100 μM, 200 μM, and 400 μM. Five second shaking was used prior to the measurements. Each spectrum represents the average of three successive scans and three independent measurements.

Transmission electron microscopy (TEM):

Each peptide stock solution (5 mM) was diluted 50-fold to 100 μM using a 50 mM Tris, pH 7.4 buffer and 50 mM Tris pH 7.4 buffer with 200 mM KCl; 2 μL of sample was loaded on carbon-coated copper grids (400 mesh), incubated for 2 mins, and drawn off using a filter paper wick. The samples were stained with 2 μL of 1.5% phosphotungstic acid (PTA), pH 7.5 for 30 seconds for better contrast. For each condition, we prepared three grids and scanned to acquire approximately 45 separate images. Visualization was performed using Morgagni TEM (ThermoFisher) operated at 80 kV. Micrographs were acquired with a Morada camera (EMSIS, Muenster, Germany).

Fourier Transform Infrared (FTIR) spectroscopy:

The FTIR data were obtained using a Nicolet iS5 MIR FTIR (ThermoFischer Scientific) instrument with Specac ATR scanning head from L.O.T.-Oriel. Two different approaches were used to collect the FTIR data. In the first approach, 5 mM stock solutions of peptides were directly used to collect the data with water as a background. In the second approach, the peptide samples were dialyzed against DI water using 0.5–1 kDa molecular weight cut off Float-A-Lyser (Spectra) for 24 hours and then lyophilized. These samples were used to collect data with air as background. Each FTIR spectrum represents the average of hundred consecutive scans. FTIR spectra were checked for an appropriate ratio of amide I and II band intensity, low noise, and the absence of other artifacts to confirm proper data acquisition and background subtraction. Data were analyzed using the GRAMS/AI software (Thermo Scientific). Second derivative spectra were calculated from the absorbance spectra in the amide I region using a Savitzky-Golay filter⁵⁹, third order, with a 79- to 159-point window, depending on the dataset. The second derivative spectra were then inverted and baseline-corrected assuming flat baseline segments connecting minima between 1710 and 1600 cm⁻¹. Peaks between 1600 and 1710 cm⁻¹ from the inverted second derivative spectra were then fit with the minimum number of Gaussian curves that still provided a clean fit, as determined by the Akaike Information Criterion⁶⁰. To confirm that bands identified by derivative analysis were not artifactual, we also performed a Fourier self-deconvolution (FSD) analysis on the IR absorbance spectra in the amide I region⁶¹. Only bands that appeared in both analyses were considered trustworthy. For FSD, we used gamma values of 6.7, 8.3 and 5.4 for peptides 1–3 respectively, with a smoothing factor of 95%. The resulting spectra were baseline-corrected assuming a flat baseline between 1720 and 1580 cm⁻¹. In both derivative and FSD analyses, the region from 1700 to 1800 cm⁻¹ remained flat and smooth in dry samples. This indicates the absence of confounding contributions from water vapor. The solvated P1 sample resulted in a weaker signal than the others and showed some signs of water vapor contribution. This is likely because P1 forms fibrils that precipitate from solution, making it difficult to apply to the sample stage as a solution, resulting in a low concentration sample. However, spectra from dry and solvated samples for a given peptide showed similar dominant features. The positions of the corroborated peaks were compared to values reported in the literature to identify dominant structural elements within lyophilized samples.

RESULTS

Gly-rich peptides excised from FUS are disordered:

Peptides P1, P2, and P3 are of identical lengths. We computed the predicted disorder contents across each of the sequences using IUPRED2a⁶² (Figure 2a). Scores greater than 0.5 indicate a preference for disorder, which refers to a preference for heterogeneous conformational ensembles. The predictions suggest that all three peptides are highly disordered.

Figure 2: Peptides P1, P2, and P3 are predicted to be intrinsically disordered.

(a) Predictions of disorder propensities using IUPRED2a. (b) Measured UV-CD spectra, plotted as mean residue ellipticity as a function of wavelength, for each of the three peptides. The peptide concentrations were 200 μM for each of the peptides.

We characterized the ensembles of conformations in solution using UV-CD spectroscopy (Figure 2b). Differences / similarities among the three spectra are attributable, at least in part, to the lengths of poly-Gly tracts (Table 1). The measured spectra have mean residue ellipticities that are near zero across all wavelengths. This derives in part from of the high Gly contents, which lead to a deficiency of chiral centers. Additionally, the spectral flattening is also indicative of highly assembled or aggregated material^{63, 64}. However, we were unable to detect concentration-dependent changes to the CD spectra for the range of concentrations that were tested (50 μM-200 μM) (SI Figure S1).

The CD spectra of P1 and P3 both show clear signatures of antiparallel β- sheet and β- turns. In the spectrum for the P1 peptide, a minimum near 232 nm and a maximum near 210 nm point to contributions from type-I β-turns. The minimum in the spectrum for the P3 peptide extends from 220 nm to 235 nm and is preceded by a maximum at or below 200 nm. This is suggestive of contributions from β-sheeted conformations⁶⁵. Gly-rich regions in polyglycine II conformations have similar backbone ψ-angles as extended β structures. These systems form fibers through intermolecular hydrogen bonding. Polyglycine II structures have been observed in crystallographic studies of the snow flea antifreeze protein⁶⁶, the core of acetophenone carboxylase⁶⁷, and in original fiber diffraction studies that dissected the molecular structure of polyglycine⁶⁸. Indeed, Ramachandran et al.,⁶⁸ noted that the numbers of fibers with right- or left-handed twists will likely be equal in solution, thus giving rise to solutions that lack optical activity. These observations, which hold for long polyglycine chains, could also be an explanation for the flattened CD spectra of peptides P1 and P3, although standard deconvolutions of these spectra might suggest beta-sheeted conformations⁶⁹. Overall, the weak CD signals are likely to be due to lower concentrations of peptide in solution due to poor solubility. The CD spectrum of the P2 peptide shows features that are inverted compared to P1 and P3, with a positive band at 230 nm followed by a negative band approaching 200 nm. This is a signature of the presence of some degree of unbundled left-handed helical structures, such as segments of polyproline II or polyglycine II helices^{65, 70, 71}.

The flattening of the CD spectra suggests that peptides P1 and P3 undergo self-assembly into larger structures^{63, 64}. Taken together with the classical analysis of Ramachandran et al.,⁶⁸ it would seem that the polyglycine II structure might be of relevance in these systems, as might a competition with sterically allowed extended β-strand conformations within β-sheets. The backbone ϕ and ψ angles for peptides in polyproline II and polyglycine II conformations are very similar to one another (ϕ ≈ −60° and ψ ≈ +140°)^{72, 73}. Our results are noteworthy in light of recent suggestions indicating Gly-rich regions likely adopt polyproline II / polyglycine II conformations within biomolecular condensates⁷⁴.

Macroscopic assessments:

We used a table-top visualization assay to investigate the macroscopic consequences of concentration-dependent changes to peptide solutions. The qualitative results are presented in Figure 3. They show that a 5 mM stock solution of peptide P1 forms cloudy solutions, whereas peptides P2 and P3 form clear solutions and self-supporting gels, respectively.

Figure 3: Peptide P3 (right) forms macroscopically visible self-supporting gels.

In contrast the peptides P1 (left) and P2 (middle) form cloudy and clear solutions, respectively.

Next, we used TEM to investigate the morphologies formed by each of the three peptides. The supramolecular structures formed by the P1 peptide are concordant with being fibers formed by the twisting of ribbons around a long axis (Figure 4a, SI Figures S2a, S2b, S2c). In contrast, the P2 peptide does not form micron-scale assemblies that are either ordered or disordered (SI Figures S2d, S2e, S2f). The peptide P3, which forms macroscopically visible self-supporting gels, is characterized by networks of physically crosslinked fibrils (Figure 4b, SI Figures S2g, S2h, S2i).

Figure 4: TEM images of supramolecular assemblies formed by P1 (a) and P3 (b) at 50 mM tris buffer pH 7.4.

The peptide P1 forms micron-sized fibrils made of ribbons twisting around a long axis. In contrast, the peptide P3 forms a dense network of physically crosslinked fibrils that are shorter than those formed by the P1 peptide.

Concentration-dependence of the formation of fibrils measured using ThT fluorescence:

ThT is routinely used in histological and biochemical studies to stain fibrillar assembles. Binding of ThT to fibrillar structures hinders the rotation of the molecular rotor. This leads to increased quantum yield in the fluorescence, as evident by the increase in the emission spectrum intensity at 482 nm upon excitation at 450 nm. Figure 5 shows the concentration dependence of ThT emission spectra. At equivalent peptide concentrations, the quantum yield at 482 nm is ~5-fold higher for peptide P1 when compared to P3. Unlike peptides P1 and P3, which have 10-residue long poly-Gly tracts within their sequences, the ThT fluorescence does not change as a function of concentration for the peptide P2. This is consistent with macroscopic assays showing clear solutions (Figure 3) and the absence of ordered supramolecular assemblies in TEM imaging (SI Figures S2d, S2e, S2f).

Figure 5: ThT fluorescence intensities measured for each of the three peptides at three different concentrations.

(a) Peptide P1, (b) peptide P2, and (c) peptide P3. The emission spectra for peptides P1 (a), and P3 (c) show a concentration dependent increase in intensity with maxima at 482 nm. This is consistent with the peptides forming fibrillar structures, albeit with different micron-scale organization. P2 (b) shows no increase in intensity at various concentrations.

Given the higher charge content of peptide P3 compared to peptide P1, we hypothesized that the network of shorter fibrils might be driven in part by the electrostatic repulsions among cationic Arg moieties. To test this hypothesis, we measured the dependence of ThT fluorescence in peptide solutions with different concentrations of monovalent salt (KCl). These measurements were performed at peptide concentrations of 100 μM. For peptide P3, the fluorescence intensity at 482 nm, the maximum in the emission spectrum, increases by a factor of 1.6 as the KCl concentration is increased from zero excess salt to 300 mM (Figure 6a). In contrast, for the P1 peptide the fluorescence intensity at 482 nm decreases by a factor of 0.6 as the KCl concentration is increased from zero excess salt to 300 mM (Figure 6b). The maximum fluorescence intensities for peptides P1 and P3 are roughly equivalent to one another at a KCl concentration of 300 mM. This suggests that the higher salt concentrations unmask the intrinsic fibril forming tendencies of poly-Gly tracts, which are shared by the peptides.

Figure 6:

Dependence of ThT fluorescence on salt concentration for P1 (a) and P3 (b).

We used TEM to investigate the morphologies of the P1 and P3 peptides in the presence of 200 mM KCl with 50 mM Tris pH 7.4. Under these conditions, the long fibers were broken up and instead form smaller nanoscale fibers (Figure 7a). In contrast, the P3 peptide forms micron-scale assemblies in presence of salts (Figure 7b). For P1 (NCPR −0.04), salt interferes with fibril formation. However, for P3 (NCPR +0.12), salt enhances fibril formation. The parsimonious explanation is that electrostatic repulsions are screened at higher salt concentrations.

Figure 7: TEM images of supramolecular assemblies formed by P1 (a) and P3 (b) at 50 mM tris buffer pH 7.4 with 200 mM KCl.

The peptide P1 forms nanoscale fibers that are shorter than those observed at low salt – compared to Figure 4a. In contrast, the peptide P3 forms networks of micron-sized fibrils.

Finally, we used FTIR spectroscopy to obtain comparative assessments of the molecular level structures formed by the three peptides at high concentrations (5 mM). The data are shown in Figures 8 and 9. Water interference is a persistent challenge in acquiring FTIR spectra due to the overlap of water bending modes with the Amide-I polypeptide band. We collected FTIR data using peptides dissolved in buffer and subtracted a buffer-only spectrum from each of the peptide spectra. However, due to the challenges of low solubility of some of the peptides and / or difficulty in isolating assembled material in sufficient quantities, the FTIR signals were weak and noisy. However, by analyzing the data with both Fourier self-decomposition (FSD) and 2^nd derivative (SD) band sharpening, we were able to identify bands that were consistent in both FSD and SD analyses and therefore less likely to be due to noise. Furthermore, we also measured the FTIR spectra of lyophilized material and found that all the most prominent and defining bands observed in the solvated peptides were also present in the lyophilized material, allowing us to interpret the spectra of solvated peptides with confidence. The resulting band assignments are summarized in Table 2.

Figure 8:

FTIR spectra of P1, P2, and P3; (a) in buffer (with buffer subtracted) and (b) lyophilized.

Figure 9: Deconvolution of the FTIR spectra in buffer.

The top row shows Fourier self-deconvolutions of the absorbance spectra whereas the bottom row shows second derivatives of these spectra. The raw data, smoothed as described in the methods section, are shown in gray. Fits to the spectra using the component Gaussians (green) are shown in black. See text for band assignment. Panels (a), (d) correspond to peptide P1; panels (b), (e) to peptide P2, and panels (c), (f) to peptide P3.

Table 2:

FTIR band assignment for peptides P1-P3 under both solvated and dry conditions

	Peptide P1				Peptide P2				Peptide P3
	solvated		dry		solvated		dry		solvated		dry
	FSD cm−1	SD cm−1	FSD cm−1	SD cm−1	FSD cm−1	SD cm−1	FSD cm−1	SD cm−1	FSD cm−1	SD cm−1	FSD cm−1	SD cm−1
AP β-sheet	1701	1702	1691	1693	1698	-	1699	1698	-	1702	1695	1698
Agg β-str	-	-	-	-	-	-	-	-	1686	1689	1688	1688
PPII	-	-	-	-	1681	1681	1684	1684	-	-	-	-
β-turn	1678	1674	1674	1674	-	-	-	-	-	-	-	-
PPII or β-sh	-	-	1668	-	1670	1670	1668	1668	1671	1667	1675	1668
RC or β-sh	1658	1656	1657	1657	1660	1658	-	-	1655	1655	-	1652
PPII or RC	1644	1644	-	-	1645	1645	1645	1648	-	1647	1643	1644
β-sheet	-	-	1641	1640	-	1635	-	1638	1640	1641	-	1639
Agg β-str	1631	1631	-	-	1628	1627	-	1629	1626	1626	1625	1625
AP β-sheet	1616	1613	1616	1616	1614	1615	1612	1612	1613	1610	1614	1611

FSD: Fourier self-deconvolution; SD: second derivative; AP: antiparallel; Agg: aggregated; PPI: polyproline II; β-sh: β-sheet; β-str: β-strand

The FTIR spectrum (Figure 9) for peptide P1 confirms the presence of β-turns observed by CD, with a prominent band at 1678 cm⁻¹. Additionally, the P1 peptide shows a band at 1631 cm⁻¹, which is indicative of β-sheets. Also present are the 1701 and 1616 cm⁻¹ bands, which taken together with the band at 1631 cm⁻¹ are indicative of antiparallel β-sheet structures.

Compared to P1, the P3 peptide shows similar although slightly shifted bands, with a weaker β-turn band at 1671 cm⁻¹, antiparallel β-sheet bands at 1686 and 1613 cm⁻¹, and an extremely strong band at 1626 cm⁻¹. This type of increase in a ~1626 cm⁻¹ band with respect to a ~1613 cm⁻¹ band is indicative of shorter β-sheet fibers⁷⁵. Additionally, the FTIR spectrum of the P3 peptide has a band at 1640 cm⁻¹ that is also attributable to β-sheeted structures⁷⁶. The prominence of β-sheeted features in the FTIR spectrum for the P3 peptide is also consistent with interpretations of CD spectra for this peptide.

The FTIR spectrum of the P2 peptide shows bands at 1681 cm⁻¹, 1670 cm⁻¹, and 1645 cm⁻¹. These bands are all indicative of the presence of polyproline II structure on some length scale^{55, 77, 78}. These results support our interpretations of the CD spectrum for the P2 peptide. While a band at 1645 cm⁻¹ might typically be assigned to a random coil, it has been shown that a band in this vicinity can also be indicative of polyproline II structure, especially given the presence of other bands that are attributable to polyproline II structure (i.e., 1681 cm⁻¹, 1670 cm⁻¹)⁷⁷. The FTIR spectrum of the P2 peptide also shares some less prominent peaks with those of the P1 and P3 peptides. Of interest are the bands at 1698 cm⁻¹, 1628 cm⁻¹ and 1614 cm⁻¹, which point to a weak tendency for antiparallel β-sheet conformations. Furthermore, the FTIR spectra of all three peptides show a significant band in the vicinity of 1655–1660 cm⁻¹ which are attributable to either random coil (P2, 1660 cm⁻¹) or β-turn (P1 and P3, 1658 cm⁻¹ and 1655 cm⁻¹, respectively)^{77, 78}.

DISCUSSION

We studied the conformational properties and morphologies formed by self-assembling peptides derived from the protein FUS. The local concentrations of each of the peptides are likely to be in the high micromolar to low millimolar range within condensates formed by phase separation of FUS. Interest in the three peptides (Table 1) was driven by the unusual presence of 10-residue long poly-Gly tracts within peptides P1 and P3. The peptide P2 serves as a control system with greater than 50% Gly content and the absence of a poly-Gly tract that is longer than four residues.

What types of intrinsic conformational and morphological signatures might Gly-rich regions display at high concentrations? We answered this question by performing comparative assessments of the concentration-dependent conformational and morphological characteristics of assemblies formed by the peptides P1, P2, and P3. The concentrations used in our experiments are likely to be concordant with the local concentrations of peptides within condensates formed by FUS. Peptides P1 and P3 form supramolecular assemblies that are long twisted fibrils in the case of P1 and shorter, networked fibrils that give rise to macroscopic self-supporting gels in the case of P3. In contrast, the solutions of peptide P2 are macroscopically clear. We do not find any evidence for the formation of ordered supramolecular assemblies with this peptide. This is interesting and relevant because most 25-residue Gly-rich stretches within FUS and FET family proteins have compositional biases that are akin to that of the P2 sequence. The FTIR and UV-CD data for the P2 peptide point to the presence of polyproline and / or polyglycine II structures. Given the peptide concentrations used in our experiments, these structures likely form lower molecular weight complexes via homotypic interactions. Notably, it has been demonstrated that the increased exposure of protein backbones, featured in extended left-handed helices can facilitate nucleic acid recognition⁷⁹, which is interesting given the ability of FUS to interact with RNA and single-stranded DNA⁸⁰. Our findings and those of others⁷⁸ might warrant further exploration of the ability of G-rich regions such as the P2 peptide to assist or enhance RNA binding within condensates or serve as enablers of condensate formation through heterotypic interactions with RNA⁸¹.

Studies of phase separation and percolation, or less nuanced phenomena such as LLPS, typically treat Gly residues as low effective solvation volume spacers^{8, 10}. Such spacers are likely to enhance the: (i) synergy between phase separation and percolation¹¹; (ii) the driving forces for phase separation; and (iii) rearrangements of condensate spanning networks whereby physical crosslinks are made and broken on timescales that are likely to be commensurate with the timescales for diffusion of proteins into and out of condensates⁸. This view of Gly-rich regions is also bolstered by the fact that Gly is a disruptor of ordered structures, typically featuring in loop and turn regions of autonomously foldable proteins⁸². As a result, the combination of Gly and Pro residues are viewed as drivers of conformational heterogeneity and inhibitors of forming fibrillar morphologies⁸³. These expectations stand in contrast to the under-appreciated fact that poly-Gly tracts are highly insoluble molecules in aqueous solvents^{48, 84}. They can only be solubilized in harsh conditions such as 100% formic acid or in the presence of volatile salts such as trifluoroacetic acid or very high concentrations of LiCl⁵³. Further, polyglycine, which is insoluble in aqueous solvents can form fibrous structures with nematic ordering^{50, 55} as has been reported in diffraction studies performed in non-aqueous organic solvents. Within these assemblies, poly-Gly molecules are thought to adopt polyglycine II conformations and strand-like extended conformations that promote intermolecular hydrogen bonding⁵⁵. A series of studies have also investigated the conformational properties of poly-Gly tracts and Gly-rich peptides in aqueous solvents at ultra-low concentrations^{53, 85, 86}. These investigations, which used a combination of molecular simulations⁵² and fluorescence correlation spectroscopy measurements^{53, 85}, concluded that water is a poor solvent for poly-Gly molecules. Accordingly, they form heterogeneous ensembles of collapsed conformations as individual molecules in ultra-dilute solutions. These compact structures are thought to be stabilized by the higher favorability of intramolecular amide-amide interactions when compared to amide-water interactions in aqueous solutions^{87, 88}. As peptide concentrations increase, the interface between poly-Gly molecules and the aqueous solvent, which is a poor solvent for poly-Gly, is better minimized by replacing intramolecular interactions with intermolecular ones, thereby driving the molecules out of solution⁸⁹. This is the conceptual basis for the ultra-low solubility of poly-Gly molecules. Overall, the findings of prior studies, specifically the finding that water is a poor solvent for poly-Gly, which in turn promotes peptide self-association at high peptide concentrations, are consistent with our current findings that peptides P1 and P3 that feature 10-residue-long poly-Gly tracts, self-assemble into fibrils and self-assembled fibrillar networks.

CONCLUSION

Both peptide solubility studies and assessments of solvent quality based on single molecule or small ensemble experiments suggest that the transition from a soluble system to a strongly self-associating system occurs between poly-Gly lengths of 6–10 residues⁵³. Since most low complexity domains that drive phase separation and biomolecular condensate formation do not feature long poly-Gly tracts (Figure 10), little attention has been paid to the poly-Gly effect as a contributor to transitions driven by self-assembly of Gly-rich regions within condensates. However, the human version of FUS features two well-defined poly-Gly tracts (Figure 1), one within the PLD and the other straddling the interface between the PLD and RBD. Uninterrupted poly-Gly tracts seem to be a shared feature of FUS sequences across orthologs (Figure 10). Interestingly, these features are absent in other FET family proteins (Figure 10, SI Figures S4, S5, and S6).

Figure 10: Violin plots summarizing the statistics associated with finding uninterrupted poly-Gly (left) and poly-Ser tracts in FUS and orthologs of FUS as well as members of FET family proteins.

A key finding is that human FUS and orthologs of FUS consistently include long poly-Gly tracts in their sequences.

We find that the peptides with 10-residue-long poly-Gly tracts, designated as P1 and P3, enable concentration dependent self-assembly leading to fibrillar structures. Specifically, P1 forms long, twisted fibrillar structures whereas P3 enables the formation of fibril-based crosslinked structures that, macroscopically, are self-supporting gels. Our findings suggest that Gly-rich regions, specifically tracts that are long enough to be influenced by the intrinsic self-associativity of poly-Gly molecules, will likely transition from being spacers that regulate the driving forces for phase separation and percolation to nucleators of liquid-to-solid transitions within condensates formed by molecules such as FUS. It is noteworthy that the fibril-forming attributes of the Gly-rich peptides excised from FUS are not anticipated by algorithms that are used to predict amyloid forming / aggregation / solubility propensities (SI, Figures S7–S9).

Our findings point to a strategy whereby the self-assembly characteristics of 20–25 residue long peptides, excised from full-length proteins, can be investigated to identify regions that are likely to be intrinsic drivers of fibril formation within condensates. Our results also highlight the importance of and rationale for negative selection, whereby long poly-Gly tracts tend to be avoided, at least in the subset of RNA binding proteins we have analyzed thus far (Figure 10). Whether this also implicates polyproline II / polyglycine II structural motifs of regions with high Gly content albeit short poly-Gly stretches as modulators of driving forces for condensate formation or interactions within condensates⁷⁴, be they homotypic or heterotypic in nature, will require closer scrutiny, preferably in the context of the full-length proteins and multicomponent mixtures.