FUS Phase Separation Is Modulated by a Molecular Chaperone and Methylation of Arginine Cation-π Interactions

1Fused in sarcoma (FUS) is an RNA-binding protein involved in RNA transcription, splicing, transport, and translation. FUS undergoes rapid, physiologically reversible phase separation between dispersed, liquid droplet, and hydrogel states. The droplet and hydrogel states are stabilized by hydrogen bonding between antiparallel β sheet motifs formed by core residues 39–95 in the low-complexity (LC) domain (Figure 1A). The ability of FUS and other proteins with intrinsically disordered domains to undergo phase separation likely contributes to the role of FUS in forming transient membrane-free organelles, such as ribonucleoprotein (RNP) granules. These dynamic structures take up, sequester, transport, and then release key RNA and protein cargos that regulate local RNA and protein metabolism in subcellular niches, such as axon terminals and dendrites. When these processes go awry (e.g., because of pathogenic missense mutations), they trigger disease, such as familial amyotrophic lateral sclerosis (fALS) and frontotemporal lobar degeneration (FTLD).

2Given the crucial role of intrinsically disordered proteins like FUS in multiple fundamental biological processes, understanding the molecular and cellular factors that control their reversible condensation would be invaluable. This knowledge could also yield avenues for therapeutic intervention in diseases associated with aberrant assembly of these proteins, such as FUS-associated ALS (fALS-FUS) and frontotemporal lobar degeneration (FTLD-FUS).

3A potentially powerful clue to the identity of such factors is the observation that arginines in FUS, which are predominantly located in the structured C-terminal domain (sCTD), are normally heavily methylated as mono- or dimethylated forms (Figure 1A). However, in FTLD-FUS, FUS is hypomethylated and accumulates in neurons as nuclear and cytoplasmic aggregates that frequently also contain EWS, TAF15, and transportin 1 (TNPO1). These observations suggest that physiological fluctuations in its arginine methylation and/or its interactome might physiologically control FUS phase behavior.

4To assess this possibility, we investigated the effect of manipulating (1) the number and post-translational methylation state of arginines (Figure 1A) and (2) the interactions with known FUS-binding proteins. These experiments confirm that FUS phase separation is exquisitely modulated by (1) arginine methylation state (which tunes the strength of cation-π interactions between the structured C-terminal and the disordered N-terminal domains) and (2) binding of TNPO1, which acts as a molecular chaperone in peripheral compartments of neurons. Crucially, hypomethylation of FUS promotes formation of stable condensates comprising intermolecular β-sheet-rich FUS assemblies that disrupt RNP granule function in neuronal terminals and cause FTLD-FUS.

5Prior work has established that the LC domain of FUS can form β-sheet-rich condensates when cooled at high protein concentrations (50–133 μM) in the presence of crowding agents (e.g., polyethylene glycol and dextran). However, to gain a quantitative understanding of how regions outside the LC domain might influence the phase behavior of full-length FUS, we expressed and purified wild-type, full-length human FUS from eukaryotic Sf9 cells with or without an Emerald GFP (EmGFP) tag. We chose this system over bacterial production systems because it allows analysis of FUS that has undergone physiological eukaryotic post-translational modification. We then used this material to explore the impact of variations in temperature (4°C–37°C) and salt (50–150 mM NaCl or KCl) at physiological FUS protein concentrations (≤5 μM).

6These experiments revealed that variations in temperature (4°C–37°C) had little effect on the phase state of full-length FUS at physiological concentrations (∼1 μM). In contrast, FUS phase behavior was profoundly affected by variations in the concentration of salts, such as NaCl and KCl. Specifically, at 1 μM FUS, decreasing concentrations of NaCl or KCl from 150 to 50 mM caused rapid phase separation of FUS into hundreds of small droplets (2.29 ± 0.15 μm diameter; n = 128 assemblies) (Figure 1B). These assemblies, which appeared within seconds, were approximately spherical (Figures 1C–1E), and underwent fusion events that could be monitored using structured illumination microscopy (Figure 1E; Video S1). These effects were not influenced by the presence or the absence of an EmGFP tag.

7To explore the effects of changes in FUS methylation, we isolated full-length, wild-type human FUS, with or without an EmGFP tag, from eukaryotic Sf9 cells grown in the presence of 25 μM adenosine-2,3-dialdehyde (AdOx), a widely used inhibitor of arginine methyltransferase activity. Arginine methyltransferases are components of RNP granules and therefore likely relevant to the biology of FUS phase separation. Western blots of FUS protein from AdOx-treated cells showed a significant reduction in asymmetrically dimethylated FUS (ADMA FUS) (Figures 1A and S1A).

8To identify which arginines were methylated and to quantitatively assess the reduction in arginine methylation induced by AdOx, we used both isobaric tags for relative and absolute quantitation (iTRAQ) and spectral counting mass spectrometry methods. In full-length FUS from untreated Sf9 cells, methylation was not homogeneously distributed across all 37 arginines (Figure S1B). At least 9 arginines were dimethylated, and these were predominantly located in glycine-rich clusters. However, several arginines were predominantly unmethylated, even when neighboring arginines were dimethylated. The effect of AdOx was also nonuniform. Thus, arginines 216, 259, 407, 473, and 476 were converted from a significantly dimethylated state to a predominantly mono- or unmethylated state. In contrast, arginines 394 and 481 remained predominantly dimethylated (Figure S1B). These differences were robust, being replicated in both the iTRAQ and the spectral counting analyses and were not altered by the presence or the absence of the EmGFP tag.

9To assess the effects of reduced methylation on FUS phase behavior, we repeated the phase transition experiments on hypomethylated FUS (HYPO FUS) purified from AdOx-treated Sf9 cells. HYPO FUS condensed into many small assemblies (1.46 ± 0.11 μm), often with non-spherical shapes, fewer fusion events, and right-shifted the phase diagram in a manner similar to the effects of fALS- FUS mutations (Figures 1B, 1D, 1E, and 2B; Video S2; ADMA FUS, black; FUS P525L, green; HYPO FUS, lower red line; p < 0.0006). These differences were not influenced by the EmGFP tag.

10The dramatic effect of FUS arginine methylation on phase separation raises the question of what proportion of FUS needs to be hypomethylated before significant changes in phase behavior could occur. If this proportion was small, then modulating FUS methylation might provide a physiological mechanism to dynamically change FUS assembly. Furthermore, if this process became uncontrolled, the accumulation of excessive quantities of hypomethylated FUS might then cause disease by promoting the formation of stable and biologically irreversible fibrillar hydrogel assemblies in a manner analogous to fALS-FUS mutations. To address this question, we repeated the phase separation experiments at the boundary conditions of 2 μM full-length FUS and 40 mM NaCl at 23°C, but included varying quantities of fully unmethylated FUS (UM FUS) purified from E. coli (0.5%–5% of total FUS). These studies revealed that when UM FUS comprised more than 1% of total FUS there was (1) increased formation of small non-spherical, non-fusing assemblies and (2) the appearance of larger diffuse assemblies that are likely fibrillary hydrogel condensates (Figure 1F, middle and lower panels, p < 0.01). These results indicate that even small quantities of unmethylated FUS (< 5%) could induce transition of dispersed FUS into liquid droplets and its gelation into irreversible fibrillar condensates.

11Our observation that differential methylation of arginines in the C-terminal structured domain of FUS can modulate phase separation in a salt-dependent manner raises the question of what these arginines interact with during condensation. FUS contains only 5 acidic residues in the LC domain, making these unlikely to be the principal drivers of the implied electrostatic interaction. However, FUS has 27 tyrosines (but no tryptophan or phenylalanine) in the LC domain, which might allow protons in the guanidino moiety on the arginine side-chains to form cation-π interactions with electrons in the benzene ring of tyrosines.

12To explore this idea, we investigated phase separation in purified full-length FUS proteins in which (1) multiple arginines in the sCTD were mutated to alanine or lysine or (2) multiple tyrosines in the LC domain were mutated to alanine or phenylalanine. In the arginine mutagenesis studies, we focused on the six arginines (216, 259, 407, 472, 473, and 476) that showed the greatest variability in methylation state after AdOx treatment. In the tyrosine mutagenesis studies, we focused on seven “near core” tyrosines (hereafter “ncYs”: 113, 122, 130, 136, 143, 149, and 161) adjacent to the β-sheet-forming core of the LC domain (amino acids [aa] 39–95). We chose not to investigate tyrosine replacement in the core LC domain to avoid confounding the experiment by disrupting the ability of the core domain to form anti-parallel β sheet assemblies.

13FUS phase separation was abrogated when cation-π interactions were disrupted by (1) replacement of arginines with alanine (FUS 6R→A) (Figure 2A), (2) enzymatic conversion of arginine to citrulline (Cit-FUS) by protein arginine deiminase (PAD), which replaces the positively charged ketimine group ( = NH) with an uncharged ketone group ( = O) (Figures 2A, 2E, and S2), and (3) conversion of the ncYs to alanine (FUS ncY→A) (Figures 2A and 2D). However, phase separation was maintained when cation-π interactions were preserved by (1) substitution of the arginines with lysine (FUS 6R→K), which has a cationic side chain, or (2) substitution of ncYs by phenylalanine (FUS ncY→F), which contains an aromatic ring in its side-chain (Figures 2A, 2C, and 2D).

14Next, we investigated the effect of increasing the cation-π drive by strategic substitution of 9, 16 or 21 additional arginines to create more RGG and GRG motifs in the sCTD (FUS +9R, FUS +16R, and FUS +21R, respectively). The circular dichroism (CD) spectra of these proteins were indistinguishable from either ADMA FUS or HYPO FUS, implying that they were properly folded (Figure S2A). However, these constructs had significantly increased propensity to phase separate (as measured by turbidity; Figure 2B) and form gel-like structures (as measured by increasing numbers of nonspherical condensates; Figures 2C and 2D). This behavior was strongly dependent on the number of extra arginines (Figures 2A and 2B). To confirm that this effect was due to enhanced cation-π interactions, we mutated the ncYs to alanine in the construct with 16 extra arginines (FUS +16R ncY→A), thereby reducing the number of tyrosines available to form cation-π interactions with the extra arginines in the parental FUS +16R construct. This FUS +16R ncY→A construct rescued liquid droplet formation (Figures 2A, 2C, and 2D).

15To examine the implied cooperativity between the tyrosine-rich disordered LC domain and the arginine-rich sCTD we separately purified these two domains and investigated their phase behavior alone or mixed together. The EmGFP-tagged LC fragment formed droplets and gels when cooled at high concentrations (>50 μM FUS), but did not phase separate at 1 μM at 23°C, even when mixed with dextran (Figure 2F). The mCherry-tagged arginine-rich sCTD also showed minimal phase separation under these conditions (Figure 2F). In contrast, robust phase separation rapidly occurred when the LC and sCTD fragments were mixed at 1:1 molar ratios, even in the absence of crowding agents (Figure 2F). However, unlike condensates from full-length FUS, these condensates were unstable, and dissolved within minutes, implying that full stabilization of condensed polymers requires tethering of the LC domain to the sCTD.

16Next, we investigated FUS phase separation in SH-SY5Y cells transiently expressing either YFP-tagged full-length FUS or YFP-tagged versions of the arginine or the tyrosine-modified FUS constructs described above. These cells were then treated for 24 hr with either DMSO control or varying doses (0–20 μM) of AdOx. Hypomethylation of FUS was confirmed by western blotting (Figure S3A).

17The results of these cellular studies were in good agreement with our initial biochemical studies. Thus, SH-SY5Y cells expressing FUS ncY→F, which have intact cation-π drive, were indistinguishable from cells expressing ADMA FUS (Figure 3C). Similarly, AdOx treatment and expression of FUS constructs with additional arginines caused a significant increase in the number of cells displaying intracellular FUS granules (Figures 3A and 3B; p < 0.01). The visible FUS aggregates were accompanied by increased abundance of FUS in RIPA-insoluble fractions of cell lysates (Figures 3A, 3D, 3E, and S3B). Crucially, the magnitude of these increases were dependent on the AdOx dose or number of extra arginines respectively. By contrast, FUS granule formation was significantly reduced in cells expressing FUS variants that diminish cation-π drive (either FUS ncY→A or FUS 6R→A) (Figures 3C and S3B–S3D).

18These experiments support the notion that differential methylation of arginines in the sCTD of FUS can regulate FUS assembly through modification of cation-π interactions with tyrosines in the N-terminal LC domain.

19Because hypomethylated FUS deposits in FTLD-FUS brain tissue contain EWS, TAF15 and TNPO1, we wondered whether these proteins might modulate FUS phase behavior. We therefore performed protein mixing experiments in which purified ADMA FUS or HYPO FUS were mixed with equimolar concentrations of TNPO1, EWS, TAF15, BSA, or buffer alone, and the phase separation behavior was investigated as above. When ADMA FUS or HYPO FUS were mixed with TAF15, EWS, or BSA, their phase behavior was essentially indistinguishable from ADMA FUS or HYPO FUS alone (Figures 4A and 4B). However, TNPO1 strongly suppressed phase separation of both ADMA FUS and HYPO FUS (Figures 4A, fourth column, and 4B, p < 0.001, n = 6 replications).

20Similar results were obtained in cell-based experiments in SH-SY5Y cells expressing YFP-FUS together with mCherry vector alone, mCherry-EWS, mCherry-TAF15, or mCherry-TNPO1. Thus, although both EWS and TAF15 colocalized with HYPO FUS granules induced by AdOx, their co-expression had no impact on granule formation (Figure 4C, top-right and bottom-left panels; Figure 4D, p < 0.001, n = 8 replications). By contrast, TNPO1, suppressed FUS granule formation, and TNPO1 was largely absent from HYPO FUS granules that did form (Figure 4C, lower-right panel; Figure 4D; p < 0.001, Figure 4E, p < 0.05, n = 8 replications). These inhibitory effects of TNPO1 were not attributable to (1) TNPO1-induced changes in the abundance or methylation state of FUS (Figure S4A) or (2) AdOx-induced changes in the abundance of EWS, TAF15 or TNPO1 (Figure S4B).

21Taken together, these experiments lead to the intriguing conclusion that TNPO1 may act as a cellular molecular chaperone for both ADMA FUS and HYPO FUS.

22To gain insight into the secondary and quaternary structures of ADMA FUS and HYPO FUS assemblies during phase separation, we applied two complementary approaches. The first approach employed the amyloidophylic fluorescent dyes—thioflavin T (ThT) and pentameric formyl thiophene acetic acid (pFTAA)—as chemical probes that could be applied to both protein and cellular preparations. The second approach employed atomic force microscopy-basedinfrared nanospectroscopy (AFM-IR; see the next section).

23ThT showed only minimal binding and fluorescence enhancement upon addition to either purified ADMA FUS protein condensates (Figure S5A, black line) or purified HYPO FUS protein condensates (Figure S5A, red line). By comparison, equimolar concentrations of α-synuclein (a conventional amyloid-forming protein) displayed robust ThT binding and fluorescence (Figure S5A, purple line). In cell-based experiments, ThT displayed minimal binding and fluorescence to HYPO FUS in AdOx-treated SH-SY5Y cells (data not shown). This result is in good agreement with prior studies showing poor binding of ThT to pathological fibrillar FUS assemblies in human FTLD-FUS and fALS-FUS tissues and in C. elegans models. Further work with ThT was abandoned. However, pFTAA showed more promising results.

24pFTAA is a high-affinity, cell permeant, luminescent oligothiophene dye that discriminates different conformers of β-sheet-containing aggregates of tau and PrP. pFTAA showed modest binding and fluorescence with ADMA FUS condensates (Figure 5A, first column of images), but significantly greater binding and fluorescence with condensates composed of HYPO FUS, FUS +9R, or FUS +16R (Figure 5A, p < 0.001). Crucially, pFTAA was also able to detect the chaperone-like activity of TNPO1. Thus, premixing TNPO1 with ADMA FUS or HYPO FUS dramatically reduced the number of assemblies that bound pFTAA (Figure S5B, p < 0.001). In cell-based experiments, pFTAA displayed significant binding and fluorescence with HYPO FUS condensates in AdOx-treated SH-SY5Y cells (Figure 5B).

25These experiments suggest that liquid droplet condensates of ADMA FUS contain a small proportion of FUS in an antiparallel β sheet conformation, and this antiparallel β sheet content is then significantly increased upon conversion to hydrogel-like HYPO FUS condensates.

26To further explore the relationship between the three-dimensional morphology and the secondary and quaternary structures of individual ADMA FUS and HYPO FUS assemblies, we next applied a recently developed single-molecule technique that combines atomic force microscopy (AFM) with infrared nanospectroscopy (IR). In contrast to conventional bulk approaches, AFM-IR provides a unique tool to probe, at nanoscale resolution, the morphological, nanomechanical, chemical, and secondary and quaternary structural properties of individual protein assemblies, a feature crucial for characterizing heterogeneous molecular systems.

27We applied this approach to investigate ADMA FUS, HYPO FUS and FUS +16R condensates generated using the same conditions used in earlier experiments. The condensates were placed on zinc selenide (ZnSe) windows. AFM-IR was then used to acquire nanoscale resolved maps on the 3D morphological (Figure 6A) and nanomechanical properties of the assemblies (Figure 6B). The nanomechanical state was assessed by measuring the shifts in the tip-sample contact resonance (Figure 6C), which varies monotonically as a function of the intrinsic Young’s modulus of the sample. Because of the complexity of defining the absolute nanomechanical properties of soft biological samples, we measured the stiffness of each assembly relative to the stiffness of the underlying ZnSe window. We then acquired nanoscale resolved infrared (IR) spectra from several locations within each assembly (denoted by “+” in Figure 6A). Because each nanospectroscopy measurement has a lateral resolution down to 20 nm, it allows exquisite characterization of the chemical properties and secondary and the quaternary chemical structures across multiple locations in individual condensates. The average spectrum from all assemblies for each group was then calculated (Figure 6F), and the corresponding average second derivatives in amide band I (Figure 6G) were evaluated to extract the principal structural components of the condensates. Cumulatively, we acquired 216 spectra with corresponding detailed nanomechanical data from ADMA FUS, HYPO FUS, and FUS +16R assemblies.

28ADMA FUS condensates had relatively homogeneous morphological and mechanical properties, with high sphericity and low relative intrinsic stiffness akin to that of a liquid. In agreement with the nanomechanical data, the chemical responses of different ADMA FUS liquid droplets were also relatively homogeneous, all being composed of α-helical, native β sheet, random coil, β-turn, and residual antiparallel intermolecular/cross-β sheet structures (Figures 6D, top panel, and S6A–S6D).

29By contrast, HYPO FUS condensates were both morphologically and mechanically heterogeneous. Crucially, this heterogeneity existed both within individual HYPO FUS assemblies and between different HYPO FUS condensates (Figures 6A–6D, second row, and S6A). This intra-sample heterogeneity is quantitatively demonstrated in Figure 6 for a representative HYPO FUS condensate. This particle has a spherical component (red circle, middle panel, Figure 6B) fused to a non-spherical component (orange ellipse, middle panel, Figure 6B). The spherical component showed softer mechanical properties, like those of ADMA FUS condensates. The nonspherical component showed stiffer nanomechanical features more suggestive of a gel. This regional heterogeneity within a single condensate is of note because it suggests conversion between liquid droplet and hydrogel conformations.

30IR spectra of HYPO FUS assemblies were also highly heterogeneous, and this heterogeneity correlated with the nanomechanical heterogeneity. The regions within a single droplet possessing higher stiffness were also the ones possessing higher absorption at 1,695 cm−1, corresponding to antiparallel cross-β sheets. More importantly, on average, deconvolution of amide band I of the HYPO FUS IR spectra revealed increased antiparallel cross-β sheet, random coil and β-turn content compared to ADMA FUS assemblies (Figures 6G, S6B, S6C, and S6E). Furthermore, amide band II of HYPO FUS assemblies was shifted toward lower wave-numbers, confirming independently increased hydrogen bonding (Figures 6G, S6B, S6C, and S6E). In addition, on average, HYPO FUS assemblies had reduced and shifted signals originating from the methyl group absorption (δas(CH3), methyl asymmetric stretching at 1,445 cm−1), confirming their lower methylation state (Figure S6F).

31Remarkably, at the level of individual HYPO FUS assemblies there was considerable heterogeneity in the spectral data. Thus, HYPO FUS condensates with liquid-like nanomechanical properties (e.g., within the red circle in Figure 6B) had IR spectra like those of ADMA FUS condensates. But the HYPO FUS condensates with stiffer nanomechanical properties (e.g., within the orange ellipse in Figure 6B) had higher antiparallel cross-β sheet content and increased intermolecular hydrogen bonding, suggesting the presence of a hydrogel-like structure.

32The FUS +16R condensates displayed a predominantly non-spherical 3D morphology and stiffer nanomechanical properties, like those of the gelled HYPO FUS condensates (Figures 6A–6D, lower panels). Quantitative analysis of the IR spectra for FUS +16R assemblies was complicated by partial overlap in the absorption spectra of the extra arginine side-chains, which absorb at 1,635–1,675 cm−1. As a result, it was not possible to quantitatively compare the secondary and quaternary structural composition of FUS +16R with HYPO FUS or ADMA FUS assemblies. Nevertheless, the spectra for FUS +16R reveal significant intermolecular amyloidogenic antiparallel (1,695 cm−1) and parallel β sheet (1,625 cm−1) content, related to the dense network of intermolecular hydrogen bonding, and lead to the observed stiffer nanomechanical properties of FUS +16R condensates.

33These results support the notions that (1) both liquid droplet and hydrogel phase transitions are associated with increasing inter-molecular hydrogen bonding and increasing antiparallel cross-β sheet, random coil and β-turn structures and (2) these structural shifts are associated with selective binding of amyloidophyllic dyes, such as pFTAA. Finally, the AFM-IR analyses of FUS +16R condensates demonstrate that enhancing the arginine: tyrosine cation-π interaction by increasing the number of arginines in the C-terminal regions of FUS outside the core LC domain, promotes formation of stabilized hydrogels. This observation suggests that FUS constructs with additional arginines (“cation-π enhanced” constructs) can at least partially replicate the propensity of HYPO FUS to form pathologically stable FUS granules, and might therefore be useful as a molecular model of FTLD-FUS (see below).

34The experiments described above support our hypothesis that FUS phase transition can be regulated by (1) methylation of arginines in the sCTD of FUS and (2) interactions with TNPO1. We were therefore curious to determine whether manipulation of the arginine methylation status of FUS and its interaction with TNPO1 might alter FUS RNP granule function in distal neuron terminals.

35To address this question, we examined FUS assembly, FUS conformational state, and FUS RNP granule function in ex vivo Xenopus retinal neuron cultures prepared as previously described. The distribution of FUS assemblies was assessed in mock-treated or AdOx-treated axons (20 μM for 30 min) either by anti-FUS immunofluorescence (for endogenous FUS) or by GFP fluorescence (for live imaging of axonal FUS granules in neurons expressing GFP-tagged FUS) (Figures 7A–7F). A caveat to the use of AdOx to induce hypomethylation of FUS is the potential for AdOx to alter the methylation state of numerous other neuronal proteins. To circumvent this caveat, in parallel experiments, we also expressed the “cation-π enhanced” constructs in axon terminals (FUS +5R, FUS +7R, FUS +9R, FUS +16R, and FUS +21R).

36In good agreement with our earlier biochemical and cellular experiments, AdOx treatment of isolated axons and axon terminals induced the formation of bright FUS granules that showed increased pFTAA fluorescence (Figures 7A–7F; p < 0.05, p < 0.01), Crucially, the parallel experiments in neurons expressing “cation-π enhanced” constructs revealed an arginine dose-dependent increase in similar axonal FUS granules (Figure 7D, p < 0.001). pFTAA labeling was not investigated in neurons expressing “cation-π enhanced” constructs because the excitation/emission spectra of their GFP tags overlap those of pFTAA.

37Next, we used puromycin labeling of nascent proteins in isolated axon terminals to assess the impact of AdOx (HYPO FUS) and of “cation-π enhanced” constructs on FUS RNP granule function (Figures 7G–7I). These experiments demonstrated that new protein synthesis was significantly attenuated in both AdOx-treated axon terminals (HYPO FUS), and in the axon terminals expressing “cation-π enhanced” constructs (Figures 7G–7I; p < 0.001). Crucially, the magnitude of this effect (∼0.60–0.80 of control) approximated that of fALS-FUS mutations (∼0.80 of control).

38Outside of the nucleus, TNPO1 colocalizes with some cytoplasmic RNA granules, where it coexists with FUS, purine-rich element binding protein A (Pur-α, which modulates toxicity of ALS-associated FUS mutants), and Staufen-1 (a marker of neuronal transport granules). In this non-nuclear role, TNPO1 facilitates the import of protein components into RNA granules in a Ras-related nuclear protein-GTP-independent fashion. In agreement with this prior work, we found that mCherry-TNPO1 is also present within motile granules in axon terminals (Video S3).

39Given that TNPO1 is expressed in axon terminals, and may function as a molecular chaperone, we next investigated whether modest overexpression of TNPO1 might restore FUS RNP granule function in axon terminals treated with AdOx or expressing “cation-π enhanced” FUS constructs. As predicted, mCherry-TNPO1 fully rescued new protein synthesis in both AdOx-treated neurons and in neurons expressing “cation-π enhanced” constructs (Figures 7E and 7I, p < 0.001). However, mCherry-TNPO1 had no effect on new protein synthesis in mock-treated (ADMA FUS) axon terminals.

40The experimental data described here reveal that the phase behavior of FUS is modulated by (1) the post-translational methylation state of arginines in the structured C-terminal domain and (2) by molecular chaperones such as TNPO1. This conclusion provokes several broad lines of thought.

41Previous reports have shown that the N-terminal LC domain of FUS (residues 1–214) is necessary and sufficient for phase separation and gelation of FUS, and does so by forming intermolecular β-sheet-rich fibrils. The experiments described here support and extend this view. Thus, our AFM-IR experiments clearly reveal that intermolecular hydrogen bonding between β sheet regions contributes to both liquid droplet and hydrogel formation.

42However, our data also lead to the conclusion that FUS phase separation is regulated by additional factors beyond just the intermolecular β-sheet-forming LC domain. The experiments described here reveal that phase separation is also driven by multivalent cation-π interactions, which occur physiologically between multiple arginines in the sCTD and multiple tyrosines in the LC domain. In support of this, we have shown that cation-π pairing and FUS phase separation are impeded by (1) replacement of C-terminal domain arginines by alanines, (2) conversion of these arginines to citrullines, or (3) replacement of N-terminal tyrosines with alanines. Conversely, we have shown that phase separation is maintained by amino acid replacements that preserve the cation-aromatic ring pairing (i.e., arginine to lysine; tyrosine to phenylalanine), rather than planar-π or planar-planar interactions. Finally, we have shown that phase separation is augmented in an arginine dose-dependent manner by increasing the number of arginines in the C-terminal domain, presumably by enhancing the number of cation-π interactions. Two further observations underscore the importance of domains outside the LC domain in promoting liquid phase separation and gelation of FUS. First, as shown here, FUS phase separation is also modulated by TNPO1, which binds to FUS via its structured C-terminal domain. Second, fALS-FUS mutations map predominantly to the C-terminal domain.

43Additional work will be required to fully understand the mechanics of the co-operative interaction between the C-terminal domain and the intermolecular β-sheet-forming LC domain. Our mixing experiments suggest a model in which multivalent cation-π interactions initiate phase separation, thereby bringing LC domains close together in restricted volumes and at higher local concentration. We propose that this close apposition permits the formation of more stable, intermolecular hydrogen-bonded β-sheet-rich condensates driven by the LC domain. The transient nature of the condensation events that occur when physically separate LC domain and C-terminal domain proteins are mixed could arise through several mechanisms. For instance, the untethered C-terminal proteins might be less efficient in restraining the egress of LC domain peptides from nascent condensates during initial sol→droplet phase separation. Alternatively, the untethered arginine-rich C-terminal domain proteins might bind to tyrosines within the intermolecular β-sheet-forming parts of the LC domain, and interfere with further condensation.

44Our observation that some HYPO FUS assemblies contain regions that are liquid droplet-like while other regions are gel-like, suggests liquid droplet and hydrogel states are alternate but mechanistically related, and potentially inter-convertible states within a single FUS assembly. This interpretation is supported by theoretical work indicating that in systems close to a critical point, variations in interaction strength or solvation volume can allow coexistence of phase separation and gelation.

45There are likely to be active processes to maintain FUS and other phase-separating proteins in a dispersed state, and to reverse gelled forms. As is the case for TDP43, autophagy and proteasome pathways are likely important components of this quality control system. However, molecular chaperones are also likely to be involved. Indeed, our experiments show that TNPO1 is a molecular chaperone for ADMA FUS and HYPO FUS that acts in distal axonal compartments of neurons, as well as at the nuclear pore.

46Our observations that arginine methylation status profoundly influences FUS phase separation, and that adding very small amounts of unmethylated FUS (≤5%) to ADMA FUS results in rapid phase separation and gelation raises the possibility that arginine methylation state might be a physiological method to regulate FUS phase behavior. This conclusion raises several critical questions. Which differentially methylated arginines are essential for a change in FUS phase behavior? How do unmethylated arginines increase the cation-π interaction strength? Are specific tyrosines more important than others? In some Tudor domain proteins, protein:protein binding occurs because of interactions between asymmetrically dimethylated arginines on one protein and specific clusters of aromatic amino acids on the other protein that are arranged into three-dimensional “aromatic cages”. Does a similar process occur during formation of the cation-π interactions in FUS? If so, do unmethylated arginines alter the thermodynamics of interactions with such “cages”? Or do these unmethylated arginines simply form promiscuous interactions with tyrosines that are not in specific clusters or cages, as proposed for other Tudor domain proteins? Further work will be required to address these questions. However, because the tyrosine-rich LC domain of FUS is natively disordered, if they exist, such tyrosine clusters or cages are unlikely to have the same highly ordered three dimensional structure of the tyrosine-based “aromatic cages” in Tudor domain proteins.

47Finally, how is methylation and demethylation of individual arginines regulated, both physiologically and pathologically? Multiple PRMT enzymes are known, at least some of which are components of RNP granules. However, to date, only a single arginine demethylase has been identified (Jumonji domain-containing 6-JMJD6).

48While not directed at generating a model of FTLD-FUS, the experiments reported here have obvious implications for understanding this disorder. Our observation that HYPO FUS and fALS mutant FUS assemblies have similar biochemical and biophysical properties, and similar effects on FUS RNP granule function, suggests that gel-like assemblies of hypomethylated wild-type FUS and of fALS mutant FUS represent a common final mechanism for FUS-related neurodegeneration.

49We speculate that neuron subtype-specific differences in molecular chaperones, methylases, and demethylases might account for why hypomethylated non-mutant FUS accumulates in fronto-temporal neurons in FTLD-FUS, while normally methylated fALS- mutant FUS accumulates in corticospinal and spinal neurons in fALS-FUS. Why FUS becomes hypomethylated and inadequately chaperoned in FTLD-FUS is unknown. Mutations have not been detected in the PRMT genes tested to date. However, our results offer other candidate genes that are worth investigating for disease-causing mutations, including TNPO1, PADs, and JMJD6.

50One important difference between our model and FTLD concerns the fact that nuclear assemblies of FUS in FTLD contain EWS, TAF15, and TNPO1. In our model TNPO1 is largely absent from FUS aggregates. The explanation for this difference is not immediately apparent. It might arise from the fact that in our model the level of TNPO1 expression is not a limiting factor because it is overexpressed. In contrast, in neurons TNPO1 might be titrated out by an excess of hypomethylated FUS. Regardless, the work reported here provides a starting point to investigate how pathological methylation and how pathological phase separation of FUS that escapes from its normal molecular chaperones might be targeted therapeutically.

51Clearly, several reagents and methods developed here including pFTAA, AFM-IR, and the cation-π-enhanced constructs will be useful tools to delve further into the biophysics of FUS phase separation, and to create molecular models of increased FUS phase separation propensity in FTLD-FUS.

52Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Peter St George-Hyslop (phs22@cam.ac.uk and p.hyslop@utoronto.ca).

53SHSY-5Y cells were cultured in DMEM high glucose medium (Sigma) supplemented with 10% FCS and 100 units/mL of penicillin and 100 μg/mL of streptomycin in a humidified incubator at 37°C and 5% CO2. SH-SY5Y cells stably expressing EYFP-FUS were generated by electroporation, followed by selection with geneticin. Cells were transiently transfected with plasmids of EYFP-FUS, FUS mutants, mCherry, mCherry-TAF, mCherry-EWS, and mCherry-TNPO1 using lipofectamine 3000 (Thermo Fisher Scientific) according to the manufacturer’s instructions. AdOx (Adenosine-2′,3′-dialdehyde, Sigma), or equal volume of DMSO vehicle, was added to cells for 24 hours at a final concentration of 20μM, unless otherwise stated in the figure legends.

54Xenopus laevis embryos were fertilized in vitro and raised in 0.1x Modified Barth’s Saline at 18°C. Capped mRNAs of mCherry, mCherry-TNPO1 or FUS-GFP were synthesized using mMessage mMachine SP6 Transcription Kit (ThermoFisher Scientific), polyadenylated using Poly(A)-tailing kit (ThermoFisher Scientific), and injected into the two dorsal blastomeres at four-cell stage as described. Eye primordia from stage 34 embryos were dissected and cultured in 60% L15 on laminin-coated coverslips at 20°C for 24 hours. This research has been regulated under the Animals (Scientific Procedures) Act 1986 Amendment Regulations 2012 following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body (AWERB).

55Constructs encoding FUS residues 1-526 and its mutants, LC-mEmerald (aa1-214) and CTF-mCherry (aa215-526), were cloned into pACEBac2 vector with a TEV cleavable N-terminal MBP tag and an EmGFP or mCherry-6xHis- C-terminal tag. Proteins were expressed and purified from insect Sf9 cells infected with the baculovirus. After four days of infection cells were harvested by spinning at 4000rpm for 30 minutes. Cell pellets were mixed with the resuspension buffer containing 50 mM Tris, 1 M KCl, 0.1% CHAPS, 1 mM DTT, 5% glycerol at pH 7.4, and proteins purified using three steps purification scheme including, Ni-NTA affinity column, Amylose affinity column followed by size exclusion chromatography in the buffer containing 50 mM Tris, 1 M KCl, 1 mM DTT, 5% glycerol at pH 7.4. For Thioflavin T binding experiment, FUS samples were produced without the C-terminal GFP tag.

56Constructs encoding full length human EWS or human TAF15 were cloned into pBACEBac2 vector with a TEV protease cleavable N-terminal MBP tag. Proteins were expressed in Sf9 cells and purified on an amylose column. Fusion proteins were subjected to TEV protease cleavage and the MBP tag was further removed by size exclusion chromatography.

57Gene encoding TNPO1 protein was cloned into pOPINS vector containing an N-terminal His-Sumo Tag and a ULP protease cleavage site separating the tag from TNPO1. Protein was expressed in E. coli BL21(DE3) in an overnight TB autoinduction media at 37°C for 5 hours followed by an overnight incubation at 25°C. Cells were harvested and subjected to lysis using high pressure cell disruption system. Clarified lysate was loaded on a Ni-NTA column and purified using standard procedure. Protein containing fractions were pooled, and dialysed in 25mM HEPES pH 7.5, 50mM NaCl, 1mM DTT and 5% glycerol buffer after addition of ULP protease to remove the His-Sumo Tag. Protein was further purified on a size-exclusion column and the fractions containing purified protein were pooled for all subsequent experiments.

58For seeding experiments, a construct encoding full length FUS protein was cloned into pOPINS vector with an N-terminal His6-SUMO tag. Protein was expressed in E. coli BL21(DE3) using TB autoinduction medium, with induction at 25°C for 24 hours. Cells were lysed by sonication and proteins purified on Ni-NTA resin, followed by buffer exchange step into 50 mM Tris, 500 mM NaCl, pH 7.4.

59For the preparation of hypomethylated FUS samples, the insect cell cultures were subjected to a repeated dose of 25μM AdOx solubilised in DMSO for four days. Protein was subjected to same methods of purification as unmodified FUS.

60For the preparation of citrullinated FUS sample, 20μg of Protein arginine deiminases (PAD) cocktail comprising five (PAD1-5) full length GST tagged recombinant proteins was mixed with prewashed 50μl of anti-GST agarose bead in PAD cocktail buffer (0.1M Tris, 10mM CaCl2, pH 7.4). After incubating at 4°C for 1 hour, the sample was spun, supernatant was discarded and the beads were washed three times with PAD buffer before being incubated with 100μg of purified Em-GFP- FUS protein in 100μl of PAD buffer plus 200mM NaCl. The mixture was incubated at 37°C shaking for 2 hours. Post incubation the sample was centrifuged and the supernatant containing the citrullinated protein was collected and kept at 4°C until further analysis.

61AdOx treated YFP-FUS expressing SHSY-5Y cells were lysed in RIPA buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 0.25% sodium deoxycholate, 1% NP-40, 1 mM EDTA) with Protease and Phosphatase Inhibitor cocktail (Pierce) and 1mM PMSF for 20 minutes on ice. Supernatants were cleared by centrifugation and equal amounts of protein taken for total lysate input or for immunoprecipitation (IP) with anti-FUS antibody or normal mouse IgG control, followed by incubation with Protein G-Agarose (GE Healthcare) or Dynabeads Protein G (Invitrogen). Beads were washed four times with RIPA buffer before the addition of 1x LDS sample buffer (Thermo) containing 2.5 mM DTT (LSD/DTT), and samples analyzed by standard immunoblotting techniques. Protein expression was quantified using densitometry analysis with ImageJ software, and the amount of ADMA-FUS, normalized to FUS input, was expressed relative to control treated protein levels.

62Equal numbers of AdOx or DMSO treated YFP-FUS expressing cells were lysed directly in 1x LDS/DTT for input control, or in RIPA buffer as performed for IP. Lysates were cleared of insoluble material by centrifugation (16,000 xg) at 4°C, the RIPA soluble supernatant diluted in 1x LDS/DTT, and the RIPA insoluble pellet washed again in RIPA buffer, re-centrifuged and suspended in 1x LDS/DTT. All samples were denatured by heated at 95°C, and analyzed by immunoblotting.

63FUS purified proteins (0.5 μM-2 μM) were subjected to a series of NaCl concentration (40mM-500mM) in a total volume of 20 μL. Samples were deposited on 8-well glass bottom Ibidi slides, incubated at room temperature for 10 minutes before being imaged on a Zeiss Axiovert 200M microscope with Improvision Openlab software using 100X magnification objective. Droplet formation was followed over time by collecting a series of images in both the bright field and FITC channels. ImageJ software was used in all image processing. For all purified proteins n ≥ 3.

64Purified LC-mEmerald and CTF-mCherry were buffer exchanged to a buffer containing 150mM NaCl. 1μM each were mixed and the concentration of the NaCl dropped to 50mM to induce droplet formation.

652μM FUS protein was mixed with various NaCl concentrations in a 50 μL total volume in a Greiner 96 well half-are clear microplate. Sample were incubated at room temperature for 10 minutes prior to the absorption (turbidity) measurement at 600nm in a SpectraMax microplate reader. Readings were recorded in triplicate for each protein sample. All assays were performed in triplicate (n = 3).

66Thioflavin (ThT) binding was evaluated by monitoring ThT fluorescence. The ThT solution, containing 10μM of ThT in 50mM Tris, 40mM NaCl (pH 7.4) buffer, was mixed with 2μM of control treated FUS, 2μM AdOx treated FUS and 5μM Synuclein and incubated for 15 minutes at room temperature. Fluorescence emission spectra of ThT, excited at 446 nm, were recorded between 455 and 550 nm on a PerkinElmer LS55 luminescence spectrometer using excitation and emission bandwidths of 2.5 nm. All binding experiments were carried out in triplicate (n = 3).

67Circular dichroism (CD) spectrum of ADMA FUS, HYPO FUS, FUS +9R, FUS +16R, FUS +21R, FUS ncY→A, FUS ncY→F were measured on a JASCO-810 Spectropolarimeter at 25°C. 5μM of each purified protein was placed in a 1 mm path length quartz cuvette and the far-UV spectrum recorded in the wavelength range of 195 – 250 nm. Scans were repeated ten times and then averaged to yield a final spectrum for each construct.

68Cells expressing FUS were imaged on μ-slides glass bottomed chambers (ibidi GmbH, Germany). Pentameric formyl thiophene acetic acid (pFTAA) was used at a concentration of 300nM diluted in PBS for cellular assays.

69To image FUS in solution, borosilicate glass coverslips (VWR international, 22 × 22 mm, product number 631-0124) were cleaned using an argon plasma cleaner (PDC-002, Harrick Plasma) for at least 1 h to remove any fluorescent residues. Prior to use, each batch of cover-slides were tested for fluorescent artifacts. FUS was diluted at a concentration of 1 μM in low salt buffer, adding pFTAA at a final concentration of 300nM. A drop of 8 μL of this mixture was placed over a coverslip and another coverslip was placed on top.

70Imaging was performed using a 60x Plan Apo TIRF, NA 1.45 oil objective, (Nikon Corporation) mounted on an Eclipse TE2000-U microscope (Nikon Corporation) fitted with a Perfect Focus unit. Fluorescence was collected by the same objective was separated from the returning beam by a dichroic (Di01-405/488/532/635, Semrock), and passed through an emission filter (FF03-525/50-25, Semrock) for both CFP and pFTAA signals. Cells expressing CFP-FUS were excited with a 405nm laser (LBX-405-50-CIR-PP, Oxxius), while pFTAA was excited with a 488 nm laser (0488-06-01-0060-100, Cobolt MLD). The excitation power was 25 W/cm2 for both lasers measured in epifluorescence mode. Cells were imaged in epifluorescence mode, while protein in solution was imaged in HiLo. The images were recorded on an EMCCD camera (Evolve 512, Photometrics) operating in frame transfer mode (EMGain of 6.8 e−/ADU and 250 ADU/photon). Each pixel was 241 nm in length. The microscope was controlled with Micromanager software, and bursts of images were recorded at 20 frames per second. Each analyzed image corresponds to an average of 50 images.

71In the experiments mixing FUS with TNPO1, the untagged FUS assemblies were mixed with an equimolar concentration of TNPO1, and the same NaCl and pFTAA concentrations and imaging paradigms. The pFTAA-positive assemblies were automatically counted using an ImageJ (NIH) macro with the function Find Maxima, and statistics were computed with MATLAB (Mathworks) and GraphPad Prism.

72Samples of purified Sf9-cell expressed MBP-FUS-mEmerald from AdOx-treated and untreated cultures, totalling 100 μg of protein each, were separately incubated at 60°C for 30 minutes with 20 mM Tris(2-carboxyethyl)phosphine. Subsequently, sample solutions were adjusted to 7.45 M urea and 45 mM 4-vinylpridine before being incubated in the dark for one hour. The denatured and reduced protein samples were concentrated by centrifugation within Microcon YM-30 filter cartridges (EMD Millipore, Billerica, MA, USA). The filters embedded in these cartridges were washed with 300μl of 10 mM triethylammonium bicarbonate (TEAB) then covered in 50 μL of 10 mM TEAB containing 2 μg of bovine chymotrypsin (for methylation analysis) or porcine trypsin (for citrullination analysis) (Promega, Fitchburg, WI, USA) and shaken for five minutes at 600 rpm.Proteolytic digestion was carried out at 37°C for 14 hours.

73Next, peptides were collected by passing the digest and an additional 50 μl volume of 0.5 M NaCl through the cartridges. Chymotryptic digests were divided for unlabeled and iTRAQ analyses. One aliquot of each digest was diluted with 20 μl of 1 M TEAB before addition of iTRAQ 8plex reagent in 210 μL of ethanol (Sciex, Concord, ON, Canada). iTRAQ reaction mixtures were left for 2 hours then concentrated in a centrifugal evaporator at 36°C before being diluted in 100 μL of 0.1% formic acid in water then combined. The samples were acidified with formic acid then desalted using Bond Elut OMIX C18 pipette tips (Agilent, Santa Clara, California USA) according to the manufacturer’s instructions. Desalted samples were concentrated in a centrifugal evaporator at 36°C, diluted in 0.1% formic acid, and analyzed on an Easy-nLC 1000-Orbitrap Fusion system (Thermo Scientific, Waltham, MA, USA). Mass spectra were collected over 60-minute-long HPLC runs, during which the acetonitrile content of the mobile phase was increased from 0 to 30% (v/v) over 40 minutes, then to 99.9% over 10 minutes and finally held at 99.9% acetonitrile for 10 minutes. The two mobile phases used were water and acetonitrile, each with 0.1% formic acid (v/v). The flow rate of the HPLC system was 300 nl/min and the Acclaim PepMap RSLC (Thermo Scientific) analytical column used was 25 cm long with a 75 μm internal diameter, packed with 2 μm C18 particles having 100 Å pores.

74All mass spectra were collected in positive ion mode with a nanoflow electrospray ionization source potential of 2200 V and an ion transfer tube temperature of 275°C. The data acquisition cycle for the iTRAQ analysis was 3 s long, beginning with an orbitrap precursor ion scan from m/z 400 to 2,000 at a resolution of 30,000 followed by MS2 scans and MS3 scans of the most abundant precursor ions and their dissociation products respectively. The MS3 spectra were collected at a resolution of 30,000. Ions detected in any survey scan having charge states less than 2 or greater than 6 or having intensities under 10,000 counts or that had been subjected to MS2 four times in the preceding 20 s of analysis were excluded from MS2 and MS3. Instrument parameters used in the analysis of unlabelled samples were identical to those used in the iTRAQ analysis, except MS3 scans were omitted.

75Peptide sequencing and quantification from the LC-MS data were performed on Proteome Discoverer version 2.1.0.81 software (Thermo Scientific) with peptide-to-spectrum matches produced by Sequest HT. Cleavage specificity was set to the C-termini of phenylalanine, leucine, tryptophan and tyrosine residues with up to two missed cleavages allowed. Precursor and product ion mass tolerance were 20 ppm and 0.4 Da respectively. The FASTA protein database used for the search included all human canonical and isoform entries (Uniprot version Dec 20, 2015, downloaded Mar 1, 2016), all Uniprot annotated Spodoptera frugiperda entries (downloaded Mar 13, 2017), bovine chymotrypsin (Uniprot accessions P00766, P00767, Q7M3E1), the small ubiquitin-like modifier of Saccharomyces cerevisiae (Uniprot accession Q12306), as well as the sequence of the MBP-FUS-mEmerald construct. False discovery rate estimation was performed using Percolator and quantification was undertaken with Reporter Ion Quantifier algorithm, both under default settings. The mass spectra were also interpreted using PEAKS Studio version 8 (Bioinformatics Solutions Incorporated, Waterloo, ON, Canada) with default settings.

76Analysis by conventional Atomic Force Microscopy and nanoIR2 (Anasys Instrument, USA) was performed on hydrophobic ZnSe windows (Platypus Technologies, USA). An aliquot of 10 μl of each sample was deposited on the surface for 1 minute. Successively, the droplet was rinsed by 1 mL of Milli-Q water and dried by a gentle stream of nitrogen.

77A nanoIR2 platform (Anasys, USA), which combines high resolution and low noise AFM with a tunable OPO laser with top illumination configuration was used. The samples morphology was scanned by the nanoIR microscopy system, with a rate line within 0.05-0.2 Hz and in contact mode. A silicon gold coated PR-EX-nIR2 (Anasys, USA) cantilever with a nominal radius of 30 nm and an elastic constant of about 0.2 N/m was used. In order to avoid and reduce polarization effects, because of the gold coating of the tip, IR light was polarized parallel to the surface of deposition. All images were acquired with a resolution between 800x200 and 1000x500 pixels per line. The AFM images were treated and analyzed using SPIP software. The height images were first order flattened, while IR and stiffness related maps where only flattened by a zero order algorithm (offset). Average frequency shift, related to the intrinsic stiffness of the sample, was calculated on 3 WT, 4 hypo-methylated (HM) and 4 FUS +16R droplets. The last ones could be divided into different regions according to their relative stiffness. Relative stiffness was calculated as the normalized ratio of the average frequency shift of each region. The relative values were measured at both the 1st and 2nd resonance of oscillation of the cantilever with consistent results.

78The spectra were collected by placing the AFM tip on the top of the FUS droplets (Figure S1) with a laser wavelength sampling of 2 cm-1 with a spectral resolution of 4 cm-1 and 256 co-averages, within the range 1400-1800 cm-1.

79Within a droplet, several spectra at different positions were acquired. For each droplet, an average spectrum was obtained as the average of the spectra at different positions within its area and by subtracting the baseline signal of the substrate. Successively, they were smoothed by Savitzky-Golay filter (second order, 9 points) and normalized. Spectra second derivatives were calculated and smoothed by Savitzky-Golay filter (second order, 9 points)). In total, 55 spectra were acquired for three methylated FUS droplets, 73 spectra for four hypo-methylated assemblies and 88 for four FUS +16R droplets. Relative secondary and quaternary organization was evaluating integrating the area of the different secondary structural contribution in the amide band I. Spectra were analyzed using the microscope’s built-in Analysis Studio (Anasys) and OriginPRO. All measurements were performed at room temperature and with laser power between 1%–4% of the maximal one and under controlled Nitrogen atmosphere with residual real humidity below 5%.

80For live cells, images were taken immediately after 24 h mock and AdOx treatment. For fixed cell quantification, cells were initially fixed with 4% PFA in medium at 37°C for 15 min and then with 4% PFA in PBS at room temperature for 15 min. Hoechst or DAPI was used for nuclear counter-staining. Images were taken using a Zeiss 710 confocal microscope. For transient transfected cells, AdOx was added 3 hours before transfection. All the experiments were replicated at least three times. To quantify the percentage of cells with FUS granule clusters, more than 250 cells were counted for each sample.

81For immunocytochemistry detecting endogenous FUS proteins, heat-induced antigen retrieval in sodium citrate buffer (10mM sodium citrate, 0.05% Tween 20, pH 6.0) was performed after 4% paraformaldehyde/15% sucrose fixation. Cultures were subsequently permeabilized with 0.1% Triton (Sigma) for 5 minutes, blocked in 5% heat-inactivated goat serum and incubated at 4°C overnight with anti-FUS antibody (Abcam, ab70381). For puromycin labeling, axons pre-treated with DMSO or 20μM AdOx for 30 minutes were severed from their cell bodies and subsequently incubated with 10μg/ml puromycin (Sigma) for 10 minutes. After fixation, permeablisation and blocking steps, puromycin-incorporated nascent peptides were labeled with Alexa Fluor 647-conjugated anti-puromycin antibody (1:250, Millipore) overnight. 161 Randomly selected noncollapsed growth cones in each condition from 3 independent biological replicates were imaged using a Nikon Optiphot inverted microscope equipped with a 60x oil-immersion objective and a CCD camera (Hamamatsu).

82For quantitation of fluorescence intensity, the growth cone outline was traced on the phase contrast image using Volocity (PerkinElmer), then superimposed on the fluorescent image. The software calculated the fluorescent intensity within the growth cone, giving a measurement of pixel intensity per unit area. The growth cone outline was then placed in an adjacent area clear of cellular material to record the background fluorescent intensity. This reading was subtracted from the growth cone reading, yielding the background-corrected intensity. All quantitative analysis was performed ‘blind’ to experimental condition and normalized to the control.

83FUS-GFP fluorescence in live distal axon segments was imaged under a Perkin Elmer Spinning Disk UltraVIEW ERS, Olympus IX81 inverted microscope with a 60x 1.4NA silicone oil objective, equipped with a Flash4.0 camera (Hamamatsu). Images were acquired at maximum speed for 1 minute with 500ms exposure time.

84In all figures the mean and SEM are described along with the number of biological replications, and the statistical tests applied.

85Raw image data of full western blots are deposited in Mendeley Data at https://doi.org/10.17632/4mjh8y579j.1.