FUS-ALS mutants alter FMRP phase separation equilibrium and impair protein translation

Cytoplasmic mislocalization of FUS-ALS mutants determines aberrant FMRP condensates and protein synthesis repression.


INTRODUCTION
The fate of mRNAs in the cytoplasm, including their localization, stability and translation, is regulated by RNA binding proteins (RBPs). These RBPs often contain low-complexity regions (LCRs) that promote their phase separation into biomolecular condensates. RNA transport granules and stress granules are two examples of biomolecular condensates and their dynamics, such as trafficking, assembly and disassembly, impact on the availability, localization, and translation of bound RNAs.
FUsed in Sarcoma (FUS) is an LCR-containing RBP involved in various aspects of RNA metabolism. Under physiological conditions, FUS is predominantly localized in the nucleus where it is involved in transcription, mRNA processing, and miRNA biogenesis (1); however, low levels of the protein are also present in the cytoplasm (2). FUS mutations are causative of amyotrophic lateral sclerosis (ALS), a relentless neurodegenerative disorder in which the loss of motor neurons (MNs) leads to a progressive impairment of the neuromuscular system. Most FUS mutations disrupt the C-terminal nuclear localization signal (NLS), leading to a nuclear depletion and a cytoplasmic mislocalization of FUS (3). The loss of FUS nuclear RNA processing functions, however, is not sufficient to induce neurodegeneration on its own, drawing attention to the cytoplasmic roles of FUS and how these are affected by its increased cytoplasmic levels, which ultimately result in cytoplasmic neuronal aggregates in patient post mortem tissue (4,5).
The role of FUS in the cytoplasm and the functional consequences of ALS-linked mutations are poorly understood; however, an increasing number of studies suggest that FUS liquid-liquid phase separation (LLPS) properties play a crucial role in its cytoplasmic gain of function (6)(7)(8). Under physiological conditions, the propensity of FUS to undergo LLPS in the cytoplasm is limited by its low concentration; in contrast, the increased cytoplasmic localization of FUS mutants favors the transition to a phase-separated state. Moreover, decreased interaction with the nuclear import factor and chaperone TNPO1, posttranslational modifications and intrinsic properties given by ALS mutations all contribute to the formation of cytoplasmic FUS condensates (6,7,9,10), the biology and composition of which are, to date, poorly understood. Potential disease mechanisms have been identified in various overexpression systems (7,11), although it remains unclear how well these model the physiological setting, where the expression of FUS, as well as most RBPs, is finely tuned both by auto-and cross-regulatory mechanisms (12).
Here, we use mouse and human models with endogenous expression of FUS-ALS mutations and show that mutant FUS forms cytoplasmic condensates containing the RBP fragile X mental retardation protein (FMRP). FUS and FMRP repress de novo protein synthesis in vitro and in MNs. We find that this is not caused by direct interaction with the translational machinery; instead, our data support a model whereby FUS and FMRP condensates have an antagonistic role to translation, resulting in decreased ribosome occupancy of FMRP mRNA targets. Our findings highlight how mutant FUS can posttranscriptionally alter RBP dynamics and function.
Fus −/− axons = 85.9 ± 5.7), supporting that the increase in axonal FMRP puncta is caused by a gain of function of mutant FUS.

FMRP partitions into FUS condensates
To further explore how mutant FUS induces FMRP granules, we asked whether the two proteins could cophase separate in vitro. We performed an in vitro LLPS assay in which the disordered N-terminal region of FUS conjugated with fluorescein isothiocyanate (FITC) ( FITC FUS LCR ) was coincubated with the disordered C-terminal region of FMRP conjugated with Alexa Fluor 647 ( Alexa647 FMRP LCR ), in the presence of sc1, a G-quadruplex-forming RNA known to bind FMRP (22,23). The observed droplets were positive for both FITC FUS LCR and Alexa647 FMRP LCR ( Fig. 2A), evidence that the FUS disordered N-terminal region phase separates with sc1 RNA and FMRP.
To analyze the partitioning of FMRP into droplets of full-length FUS, we generated wild-type, P525L, and ∆14 recombinant FUS, fused to maltose binding protein (MBP) to enhance solubility and circumvent the strong propensity of recombinant FUS to aggregate ( fig. S4A). Upon cleavage of the MBP tag by Tobacco Etch Virus (TEV) protease, all FUS proteins phase-separated in vitro as detected by turbidity measurements (Fig. 2B). FUS 14 displayed a lower LLPS propensity compared to FUS WT and FUS P525L , likely due to the loss of the C-terminal RG/RGG region ( fig. S4C). To investigate FMRP incorporation into FUS condensates, we added Alexa647 FMRP LCR to preformed FUS droplets in the presence or absence of sc1 RNA. FMRP was equally enriched in FUS WT , FUS P525L , and FUS 14 condensates in the absence of sc1 RNA, and sc1 RNA increased the enrichment of Alexa647 FMRP LCR in these condensates for all FUS proteins ( Alexa647 FMRP LCR enrichment ratio in FUS WT = 1.5 ± 0.04, FUS WT + sc1 = 4.0 ± 0.05, FUS P525L = 1.3 ± 0.004, FUS P525L + sc1 = 3.5 ± 0.17, FUS 14 = 1.4 ± 0.008, and FUS 14 + sc1 = 2.7 ± 0.04; ***P < 0.001 and ****P < 0.0001; Fig. 2 significantly differ is not unexpected as wild-type FUS is physiologically prevalently nuclear, while these assays recapitulate the behavior of the proteins when present in the same compartment. To further analyze FUS-induced FMRP sequestration in a cellular environment, we took advantage of the ability of mutant FUS to generate intracellular condensates upon overexpression (7). Overexpression of either NLS-lacking FUS ( mCherry FUS 513x ) or mutant FUS ( Flag FUS P525L ) in HeLa cells induced the formation of large, cytoplasmic FUS-positive condensates ( fig. S5A, left), which led to the sequestration of endogenous FMRP ( fig. S5A). We found that brief (~18-hour) overexpression of mCherry FUS 513x led to the presence of small FMRP puncta decorating larger FUS condensates (fig. S5A, top), as well as condensates with a more homogeneous distribution of FMRP and FUS ( fig. S5A, middle), possibly reflecting different phases of incorporation or a multiphasic behavior reminiscent of FMRP-Caprin1 condensates (24). Overexpression of Flag FUS P525L in primary MNs also led to the formation of distinct FUS condensates that were positive for endogenous FMRP ( fig. S5; arrowheads indicate FMRP-positive FUS puncta).
Because FUS overexpression may induce the formation of stress granules, the increase in FMRP condensates could be due to cellular stress rather than a direct consequence of cytoplasmic mutant FUS. To address this, we generated mouse embryonic fibroblasts (MEFs) from our ∆14 FUS mouse model, which endogenously expresses mutant FUS. When we looked at endogenous FMRP, we found that the presence of mutant FUS led to a dose-dependent increase in the number of cells with spontaneous FMRP condensates (>0.5 m 2 ) compared to controls (fig. S5, C to F; percentage of MEFs with FMRP puncta: Fus +/+ = 46.2 ± 2.1%, Fus ∆14/+ = 55.8 ± 3.6%, and Fus ∆14/∆14 = 63.8 ± 5.4%; **P < 0.05). The large majority of these puncta were negative for stress granule markers, such as Ras-GAP SH3 domain binding protein 1 (G3BP1) (fig. S5, C and D; percentage of MEFs with G3BP1 puncta Fus +/+ = 1.5 ± 0.6%, Fus ∆14/+ = 0.8 ± 0.4%, Fus ∆14/∆14 = 1.7 ± 0.7%). Together, these findings further indicate that cytoplasmic mutant FUS increases localization of FMRP in cytoplasmic FUS condensates in neuronal and non-neuronal cells and provides evidence that the presence of FMRP in these structures is not a secondary effect triggered by cellular stress.

FUS and FMRP repress translation in vitro
Biomolecular condensates are thought to antagonize translation (25,26), and FMRP phase separation correlates with translation inhibition (23). To test whether FUS condensates can directly affect protein synthesis, we took advantage of an in vitro translation assay in which recombinant MBP-FUS fusion proteins were directly added to a commercial rabbit reticulocyte cytoplasmic extract and its phase separation was induced by TEV protease cleavage. Quantification of the bioluminescence of the firefly luciferase, as a reporter for the translation of its mRNA, showed that all FUS proteins significantly suppressed translation, and addition of FMRP LCR induced a further decrease in luciferase translation (normalized bioluminescence per minute in buffer control = 1 ± 0.12, buffer + FMRP = 0.2 ± 0.02, FUS WT = 0.27 ± 0.05, FUS WT + FMRP = 0.17 ± 0.02, FUS P525L = 0.27 ± 0.03, FUS P525L + FMRP = 0.08 ± 0.01, FUS 14 = 0.15 ± 0.01, FUS 14 + FMRP = 0.08 ± 0.01; ****P < 0.0001; Fig. 3A). These results support FUS having a repressive role on protein synthesis and that its mislocalization to the cytoplasm and not the mutations per se is key to driving this gain-of-function mechanism. FMRP has an additive, although not significant, effect on translation inhibition.
To test whether this deficit was conserved in human models of FUS-ALS, we performed the same assay in iPSC-derived MNs carrying the common ALS-associated NLS mutation P525L and compared them to isogenic controls. Similarly to mouse MNs, AHA labeling was decreased by 20% in mutant FUS human MNs (isogenic control 100 ± 3.1 and FUS P525L/P525L 79.4 ± 2.2; ****P < 0.0001; Fig. 3, E and F), underlining that translation inhibition stems from FUS cytoplasmic mislocalization and is not mutation specific.
In summary, these results show that FUS mutations impair its association with polysomes and that wild-type FUS promotes the association of FMRP with the translation machinery. Given that the global level of protein synthesis is not affected in FUS knockout MNs and that polysomal localization of FUS and FMRP is impaired in both knockout and mutant FUS neurons, we conclude that these alterations cannot be the main driver for the translation phenotype observed in Fig. 3, B to F.

Mutant FUS impairs translation of FMRP target RNAs in vivo
We next investigated whether mutant FUS expression could affect the translation of RNAs specifically bound to either FUS or FMRP in vivo. We crossed ∆14 FUS mutant and FUS knockout lines with mice expressing both the Cre-dependent hemagglutinin (HA)-tagged RPL22 ribosomal subunit (Rpl22 HA , RiboTag) and a MN-specific Cre-recombinase (Chat-Cre) (31). We obtained two triple transgenic lines, Chat-Cre/Fus ∆14/∆14 /Rpl22 HA and Chat-Cre/Fus −/− /Rpl22 HA , allowing us to immunopurify ribosome-bound transcripts from MNs within the mouse spinal cord. We refer to ribosome-associated mRNAs (Ribo) as the "translatome" throughout the manuscript ( Fig. 4A and fig. S8, A to C). To validate MN specificity of purified RNAs, we performed quantitative polymerase chain reaction (qPCR) experiments that confirmed enrichment for the motor neuronal genes Chat and Rpl22 HA and depletion of glial gene Pmp22 (Fig. 4B).
We next asked whether RNAs bound by FUS or by FMRP were affected at the translatome level. We used published and widely used CLIP (individual-nucleotide resolution Cross-Linking and Immuno-Precipitation) data to select RBP-bound RNAs (30,32). Because FUS, unlike FMRP, mostly binds pre-mRNA intronic sequences, we selected only transcripts where FUS binds within the mature RNA sequence [5′ untranslated region (5′UTR), CDS (coding sequence), and 3′UTR]. We plotted the distribution of target RNAs (bound by a defined RBP) and non-target RNAs (not bound) ( fig. S9B) as a cumulative frequency of their z scores and used non-target RNAs with similar expression levels to target RNAs to avoid a bias deriving from gene expression levels ( fig. S9 for FUS and fig. S10 for FMRP). In this analysis, a shift of the cumulative frequency curve toward the right indicates an up-regulation of targets in the condition of interest, and a leftward shift indicates a down-regulation (33,34). The comparison between FUS targets versus non-target controls shows a minor, albeit significant, right shift (up-regulation) of the cumulative distribution in Ribo-Fus ∆14/∆14 (distance D = 0.05; P = 1.20 × 10 −5 ), while no change was detected in the Ribo-Fus −/− dataset (D = 0.02; P = 0.26; fig. S9C). Conversely, when investigating FMRP targets, we found more widespread changes (Fig. 4, E and F, and fig. S10B). We found that FMRP target distribution was down-regulated selectively in Ribo-Fus ∆14/∆14 (left shift, D = 0.16, P = 3.10 × 10 −15 ; Fig. 4G), while no distribution change was present in Ribo-Fus −/− (D = 0.04, P = 0.13; Fig. 4H). This indicates that the presence of mutant FUS in the cytoplasm leads to impaired ribosome association of FMRP targets, supporting the translation inhibition by FUS and FMRP condensates shown in previous assays (Figs. 1 to 3).

DISCUSSION
Cytoplasmic mislocalization and nuclear depletion of FUS are hallmarks of FUS-ALS, and the degree of cytoplasmic misplacement induced by different disease-causing mutations correlates with disease severity (3). The molecular mechanisms underlying FUSmediated neuronal toxicity are unclear; however, alterations in phase separation propensities and protein translation have been proposed to have a relevant role (7,27,37,38). Moreover, FUS expression levels, similarly to many RBPs, are physiologically highly regulated, and both exogenous overexpression and complete depletion were found to affect numerous cellular processes, causing neurodegeneration and lethality (4,(39)(40)(41). We investigated the consequences of mutant FUS mislocalization using both mouse and human models, where FUS mutants are expressed at endogenous levels. We analyzed neurons that express mutant FUS both in heterozygosity, as in patients with ALS, and in homozygosity, to better uncover phenotypes. We found that FUS forms condensates where FMRP, another RBP strongly linked to translation regulation and neurodegeneration (42)(43)(44)(45), is sequestered. We show that endogenous expression of FUS mutants is sufficient to impair translation both in mouse and human MNs. Last, we demonstrate that translation of FMRP target RNAs is impaired in vivo in MNs, establishing a pathogenesis paradigm by which mutant FUS impairs translation of specific sets of transcripts by altering the phase behavior of another RBP (Fig. 5).
Recently, numerous studies have demonstrated that FUS can undergo LLPS and that cytoplasmic FUS granules are biomolecular condensates (7,11). Cytoplasmic condensates have been shown to be involved in various aspects of RNA metabolism. Not only phase separation of specific RBPs is required for the formation of transport granules and stress granules but also the fine tuning of condensate assembly and disassembly is key to controlling RNA availability and, ultimately, translation. In agreement with this, activity-dependent protein translation has been associated with the disassembly of RNA granules and the consequent release of RBP-bound transcripts. Neuronal activity triggers -actin mRNA granule disassembly (46,47), associated with activity-dependent protein synthesis; similarly, injury-induced disassembly of G3BP1 granules results in the translation of G3BP1-bound RNAs (26).
Mutant ∆14 FUS condensates are present throughout the MN cell body and neurites (Fig. 1, A and B) and are likely to have an impact on RBP dynamics and RNA metabolism before the formation of pathological aggregates seen in post mortem tissue (5). The composition of biomolecular condensates is heterogeneous, and the proteins that partition in these structures dictate their biological function. While LCRs of proteins are required for LLPS, RBPs containing these regions have different propensities to phase-separate, and in vitro studies have demonstrated that FUS is particularly prone to undergo this process. Moreover, FUS phase separation can favor the copartitioning of so-called "client proteins" that contain LCRs but would not normally form condensates at physiological concentrations (48). FUS condensates could, therefore, act as a scaffolding in which other proteins partition, and the resulting aberrant compartmentalization of these factors could determine alterations in their biological activity. With this in mind, we questioned whether the presence of cytoplasmic FUS condensates could alter the distribution and partitioning of a wider RBP network. Because some FUS-interacting RBPs are also found in FUS inclusions in post mortem tissue or are disrupted in models of FUS-ALS (4,15,49), we decided to analyze the distribution of the two most well-characterized RBPs: FMRP and SMN. When we analyzed the distribution of axonal FMRP, we found a dose-dependent increase in FMRP condensates in mutant FUS primary MNs (Fig. 1, C, D, H, and I) and iPSCderived MNs, which is not due to alterations in FMRP expression levels but likely reflects a shift in the equilibrium between diffuse FMRP and FMRP granules. We investigated the axonal distribution of these RBPs because they have key roles in RNA transport and axonal function; however, it is likely that the effect is not restricted to this compartment, and we found alterations in the frequency of FMRP condensates also in non-neuronal MEF cells ( fig. S5, C and D). Although FMRP granules can be induced by cellular stress, FUSinduced FMRP condensates are negative for stress granule markers ( fig. S5, C and D). We therefore propose that the increased FMRP condensate formation is due to a cytoplasmic aberrant function of mutant FUS. Although other cytoplasmic RBPs containing LCRs are likely to copartition within FUS condensates, we did not find any alteration in the distribution of SMN ( fig. S2, K and L). Although the expression of mutant FUS may still affect SMN functionality [for example at the nuclear level, as previously reported (4)], this indicates that, in addition to protein-protein interaction, other factors are required to promote FUS-induced condensate formation.
We report that expression of FUS mutants at endogenous levels impairs translation, both in physiological models of disease (heterozygous FUS mutant mice) and in MNs carrying homozygous mutations. A recent study, where FUS was overexpressed in human embryonic kidney (HEK) 293 cells, has shown that FUS-mediated translation inhibition was associated with increased cosedimentation of FUS mutants with polysomes (38). In our model, however, we find that endogenously expressed mutant FUS does not associate with the translational machinery using biochemical and imaging approaches, a difference that could be due to the overexpression of the protein or to an alternative regulation in HEK293 cells. While FUS-mediated translation regulation has not yet been studied in detail, the role of FMRP as a translation repressor is well established. FMRP inhibits translation through several mechanisms, including polysome binding, miRNA and RNA-induced silencing complexdependent repression, or binding to translation initiation factors (43). More recently, FMRP LLPS has been shown to correlate with translation inhibition in vitro (23), and in vivo FMRP granule formation decreases the translation of FMRP-bound synaptic targets (50). This adds complexity to the FMRP-dependent regulation of translation and supports a model in which an increase in FMRP condensates could be associated with decreased translation of FMRP targets. To explore how the translation landscape is affected by mutant FUS expression in vivo, we sequenced ribosome-engaged transcripts from MNs. FUS mutations induce transcript expression changes, and we have previously shown that these are comparable to but weaker than FUS knockout (4,51). However, when we compared the translatome datasets, we found specific changes in the ∆14 FUS compared to the knockout translatome (Fig. 4D), demonstrating that mutant FUS affects the translatome through a gain-offunction mechanism. When we analyzed FMRP target RNAs, we found a selective depletion of these transcripts from the purified ribosomal fractions of our mutant ∆14 FUS model, with no changes in FUS knockout (Fig. 4G). The depletion induced by mutant FUS is notably similar to that described in an FMRP knockout model In agreement with FMRP having a concurrent role in FUSmediated toxicity, FMRP coexpression can rescue FUS-dependent denervation in zebrafish (15). However, because FMRP LLPS behavior, rather than expression levels, is affected by mutant FUS expression, and given that overexpression of FMRP itself promotes its LLPS, we believe that it is unlikely that FMRP overexpression could rescue the translation deficit in our model, although it may have some localized effect on specific targets (52).
Cytoplasmic mislocalization of FUS also occurs in the absence of disease-associated mutations, both in ALS cases caused by other genetic determinants (53) and in cases of frontotemporal lobar degeneration (54,55). In our in vitro translation assays, we show that wild-type FUS impairs translation in a similar manner to ALS mutants (Fig. 3A). Although FUS-ALS mutations may also affect the biophysical properties of the condensates, such as preventing their dissolution (7), and therefore cause a worsening of the phenotype, FUS mislocalization may be sufficient to drive translation repression in these pathologies. Moreover, in our in vitro experiments, FUS WT and FUS P525L have a higher propensity to undergo LLPS compared to FUS ∆14 (Fig. 2B), which is likely due to the lack of the C-terminal RGG/RG region ( fig. S4C). Because both ALS-mutants cause a very aggressive disease in patients and NLS-lacking FUS mutants, such as FUS ∆14 , have a stronger mislocalization compared to NLS missense mutants, such as FUS P525L , this highlights how a combination of mislocalization and LLPS behavior can determine FUS cytoplasmic aberrant function and result in comparable cellular toxicity.
In addition to the aforementioned mechanisms, mislocalization of FUS can result in alterations of the protein cellular concentration by affecting its autoregulation and therefore leading to the worsening of its aberrant cytoplasmic functions. Previous studies have shown that FUS-ALS mutations result in increased Fus RNA levels (51,56); we here find Fus to be increased also in the mutant translatome dataset. This further supports that the altered transcript expression indeed results in increased FUS protein translation, which, in turn, could be crucial in increasing its cytoplasmic partitioning and create a vicious cycle leading to increased FUS and FMRP condensate formation.
Our results support a model whereby the presence of cytoplasmic mutant FUS-positive condensates alters the partitioning of a wider network of RBPs, which may affect, in multiple ways, protein translation. While under physiological conditions FMRP is in an equilibrium between a diffuse and a condensed state, increased cytoplasmic levels of FUS due to ALS mutations promote the phase separation of both proteins (Fig. 5, top). We find that wild-type FUS has a critical role in FMRP association with the translational machinery, but because this effect is comparable in both mutant FUS expression and in loss of FUS, it cannot be accounted for the translation repression detected only in mutant FUS-expressing MNs. Instead, our data support a model where increased FMRP condensate formation, induced by mutant FUS, reduces the availability of FMRP-bound RNAs, ultimately altering their translation (Fig. 5). It was recently shown that an altered cross-regulation between FUS, FMRP, and the RBP HuD results in an aberrant axonal phenotype in FUS-ALS models (52). It is therefore likely that FUS and FMRP and possibly other LCR-containing RBPs with a similar biophysical behavior form a wider network and alterations in their cytoplasmic localization can influence their LLPS behavior. This would result in the generation of heterogeneous condensates in which the RNAs that are bound by RBPs present in the condensates are sequestered and their translation inhibited, with a major impact on overall neuronal functionality. FUS mislocalization and the aberrant phase separation of the RBP network can therefore alter neuronal function via a misregulation of finely tuned translational requirements, possibly contributing to ALS pathogenesis and ultimately affecting MN survival. All mouse lines were backcrossed onto C57BL/6 J animals for more than five generations. Both ∆14 Fus knockin and Fus knockout animals were maintained in heterozygosity, because homozygous mice die perinatally. Both mouse lines were crossed with heterozygous HB9::GFP mice when required.

Animals
For RiboTag experiments, both RiboTag and ChAT-Cre homozygous mice were crossed with either Fus ∆14/+ or Fus +/− animals. Double transgenic mice were subsequently crossed to obtain experimental progeny.
All experiments were carried out following the guidelines of the UCL (University College London) Queen Square Institute of Neurology Genetic Manipulation and Ethics Committees and in accordance with the European Community Council Directive of 24 November 1986 (86/609/EEC). All procedures for the care and treatment of animals were carried out under license from the U.K. Home Office in accordance with the Animals (Scientific Procedures) Act 1986 Amendment Regulations 2012 and were approved by the UCL Queen Square Institute of Neurology Ethical Review Committee.

MEFs and cell lines
MEFs were isolated from the embryonic tissue discarded from the primary MN preparation. Viscera were removed, and the remaining tissue was triturated with a blade and incubated in 0.25% trypsin for 20 min at 37°C. Trypsin was quenched, and cells were plated in DMEM (Thermo Fisher Scientific) containing 10% fetal bovine serum (FBS) and 1× penicillin/streptomycin. After 3 days in culture, cells were immortalized by transfection with Lipofectamine 3000 (Invitrogen) with the simian virus 40 (SV40) T antigen. After 5 to 7 passages at low density, the cultures presented a homogeneous cell population and started to grow steadily.
HeLa and MEF cells were maintained in DMEM containing 10% FBS and 1× penicillin/streptomycin and were maintained in a humidified incubator at 37°C with 5% CO 2 . Transfection was performed with Lipofectamine 3000 (Invitrogen) according to the manufacturer's instructions, 0.3 to 0.6 g of DNA per coverslip was used, and cells were analyzed 18 to 48 hours after transfection.

Immunofluorescence
Cells were fixed in paraformaldehyde (PFA) solution [4% PFA and 4% sucrose in phosphate-buffered saline (PBS)] for 15 min at room temperature (RT). Samples were then permeabilized and blocked in a solution containing 10% HRS (Horse serum), 0.5% BSA, and 0.2% Triton X-100 in PBS for 15 min. Primary antibodies were diluted in blocking solution (10% HRS and 0.5% BSA in PBS) and incubated for 1 hour at RT. Secondary antibodies were diluted in blocking solution (10% HRS and 0.5% BSA in PBS) and incubated for 1 hour at RT.
Coverslips were mounted using Mowiol or Fluoromount-G (Thermo Fisher Scientific); MFCs were mounted with Ibidi mounting media. Imaging was carried out using a Zeiss LSM 780 inverted confocal microscope with a 40× oil immersion lens with 1.3 numerical aperture or with a Zeiss LSM 710 inverted confocal microscope with a 63× oil immersion lens with 1.4 numerical aperture. Images were digitally captured using ZEN 2010 software and analyzed using Fiji (ImageJ).

Immunohistochemistry
For immunohistochemical analysis, 3-month-old mice were perfused with saline, followed by 4% PFA solution. Spinal cords were dissected, incubated in 20% (w/v) sucrose, embedded in Tissue-Tek optimum cutting temperature compound (OCT) (Sakura Finetek, 4583), and sectioned with an OTF cryostat (Bright Instruments). Slices were mounted on microscope slides, and sections were encircled with a hydrophobic barrier pen (Dako, S2002), permeabilized by three 10-min washes with 0.3% Triton X-100 in PBS, and blocked for 1 hour in 10% BSA and 0.3% Triton X-100 in PBS. Samples were then probed with primary antibodies overnight; the samples were then washed three times before incubation with secondary antibodies for 1 hour. Slides were then washed, mounting media were added, and samples were covered with 22 mm by 50 mm cover glass.

Cellular translation assay
AHA labeling assays were carried out as previously described (61). Briefly, neurons were incubated in a neuronal methionine-free media consisting of methionine and cysteine-free DMEM supplemented with 0.26 mM l-cysteine, 0.23 mM sodium pyruvate, 10 mM HEPES (pH 7.4), 0.067 mM l-proline, 0.674 M zinc sulfate, 5 nM B12, 1× GlutaMAX, 1× B27, 1× penicillin/streptomycin, BDNF (1 ng/ml), GDNF (0.1 ng/ml), and CNTF (10 ng/ml) for 30 min before the addition of 2 mM AHA or vehicle control for 30 min. Anisomycin (40 M) was preincubated for 20 min and coincubated with AHA. Neurons were then fixed and permeabilized, and AHA was labeled by click chemistry using the Click-iT Cell Reaction Buffer Kit with an Alexa Fluor 555 alkyne (1 M) following the manufacturer's instructions.

Proximity ligation assays
PLA was performed with Duolink In Situ Orange PLA reagents according to the manufacturer's protocol (Sigma-Aldrich).

Image analysis
Images were analyzed using Fiji (ImageJ). Digital deconvolution was performed using the following plug-ins: "Diffraction PSF 3D" was used to generate a theoretical point spread function for each wavelength, and "Parallel spectral deconvolution 2D" was used for the generation of the deconvolved image. Puncta number was quantified manually using the "Cell Counter" plug-in. SynPAnal (62) was used to quantify axonal puncta size. PLAs were analyzed by using Fiji particle analysis and PLA puncta with sizes comprising 0.05 to 3.00 m were quantified. In translation assays, MNs were selected by HB9::GFP expression and imaged. The AHA-Alexa Fluor 555 signal within MN cell bodies was quantified as mean fluorescence intensity. All analyses were performed on blinded samples.

Live imaging and analysis
Live axonal transport assays were performed in motor neuronal cultures grown in MFCs at DIV 7 to 8 as previously described (63). Briefly, for signaling endosomes and mitochondria transport assays, neurons were incubated with either 30 nM Alexa Fluor 555-HcT or 125 nM MitoTracker Deep Red (Molecular Probes, Invitrogen) for 30 min at 37°C. Cells were then washed, and new MN media containing 20 mM HEPES-NaOH (pH 7.4) were added. After 15 min, transport was assessed at 37°C using an inverted Zeiss LSM 780 microscope equipped with a Zeiss 40×, 1.3 numerical aperture differential interference contrast Plan-Apochromat oil immersion objective. Images were taken at 2 Hz for 2 to 4 min. Cargoes were tracked using the Fiji plug-in TrackMate (64), and data analysis was performed in R.
For coimmunoprecipitation (co-IP) assays, E12.5-14.5 brains were used. Samples were homogenized in lysis buffer [20 mM HEPES (pH 7.4), 150 mM NaCl, 10% glycerol, Halt phosphatase, and protease inhibitor cocktail] and incubated on a rotating wheel at 4°C for 1 hour; nuclei and cellular debris were spun down at 20,000g for 20 min. Supernatants were collected, 1 mg of protein lysate was used per co-IP, and protein of interest was immunoprecipitated with 2 g of antibody or appropriate immunoglobulin G control overnight. Protein A Sepharose beads (Sigma-Aldrich) were used to purify the antibody/protein complex; precipitates were washed three times before being eluted in Laemmli buffer.
Samples were separated on precast 4 to 15% Mini-PROTEAN TGX Stain-Free protein gels (Bio-Rad) and transferred onto a polyvinylidene difluoride membrane using a semidry Trans-Blot Turbo system (Bio-Rad), or NuPAGE 4 to 12% bis-tris protein gels were used, and proteins were blotted onto a nitrocellulose membrane using a Novex system (GE Healthcare). Western blots were developed with a Classico substrate (Millipore) and detected with a ChemiDoc imaging system (Bio-Rad). Densitometric quantification of bands was carried out using the software Image Lab (Bio-Rad).

Polysome profiling
Cytoplasmic lysates from frozen E17.5 brains were prepared as described previously (29). Tissue was pulverized in a mortar under liquid nitrogen. The tissue powder was dissolved in 10 mM tris-HCl (pH 7.5), 10 mM NaCl, 10 mM MgCl 2 , 1% Triton X-100, 1% sodium deoxycholate, RiboLock RNase Inhibitor (0.4 U/ml; Thermo Fisher Scientific), 1 mM dithiothreitol, cycloheximide (0.2 mg/ml), and DNase I (5 U/ml; Thermo Fisher Scientific). Following a first centrifugation step for 1 min at 14,000g at 4°C to remove tissue debris, the supernatant was centrifuged for 5 min at 14,000g to pellet nuclei and mitochondria. Cleared supernatants were then loaded on a linear 15 to 50% sucrose gradient in 10 mM tris-HCl (pH 7.5), 100 mM NaCl, and 10 mM MgCl 2 and ultracentrifuged in a SW41Ti rotor (Beckman) for 1 hour and 40 min at 40,000 rpm at 4°C in a Beckman Optima LE-80 K ultracentrifuge. After ultracentrifugation, gradients were fractionated in 1-ml volume fractions with continuous monitoring absorbance at 254 nm using an ISCO UA-6 ultraviolet detector. Proteins were extracted from each sucrose fraction of the profile using the methanol/chloroform protocol and solubilized in a sample buffer.
MBP-FUS and mutants were expressed in Escherichia coli BL21 DE3 cells with rare codons for R, I, P, and L using chloramphenicol and kanamycin for selection. Following cell lysis by sonication, the protein was purified by nickel-affinity chromatography. The lysis buffer used contained 20 mM sodium phosphate, 500 mM NaCl, 5 mM -mercaptoethanol, and 20 mM imidazole (pH 7.4). One Complete Protease Inhibitor tablet (Sigma-Aldrich) was added to the lysate from 2 liters of growth. The column was washed with the same buffer supplemented with 40 mM imidazole. Protein was eluted in lysis buffer with 400 mM imidazole. The protein was then further purified using gel filtration chromatography with a buffer containing 50 mM tris-HCl (pH 7.6), 150 mM NaCl, 5 mM -mercaptoethanol, and 1 mM EDTA.

Protein expression and purification of FMRP-C-terminal
The low-complexity disordered region of human FMRP 445-632 (referred to as FMRP LCR ) was expressed and purified as previously described (23). Briefly, His-Small Ubiquitin-like Modifier (SUMO)-FMRP was transformed into E. coli BL21-CodonPlus(DE3) RIL cells and grown at 37°C in LB. Protein expression was induced with 0.5 mM isopropyl--d-thiogalactopyranoside (IPTG) at an OD 600 (optical density at 600 nm) ~ 0.6 and grown overnight at 25°C. Cells were harvested and lysed in lysis buffer containing 6 M guanidinium chloride (GdnHCl), 50 mM tris-HCl (pH 8.0), 500 mM NaCl, 20 mM imidazole, and 2 mM -mercaptoethanol. Harvested cells were sonicated for 4.5 min (2 s on, 1 s off) and centrifuged. The supernatant of the lysate was then purified by nickel-affinity chromatography equilibrated with the lysis buffer. The column was washed in lysis buffer without 6 M GdnHCl and then eluted in buffer containing 50 mM tris-HCl (pH 8.0), 500 mM NaCl, 300 mM imidazole, and 2 mM -mercaptoethanol. The His-SUMO tag was cleaved with ubiquitinlike protease, while dialyzed against cleavage buffer (50 mM tris-HCl, 150 mM NaCl, 20 mM imidazole, and 2 mM -mercaptoethanol at pH 7.4) overnight at 4°C. FMRP was separated from the His-SUMO tag by nickel-affinity chromatography following the same steps described above. The fractions containing FMRP were collected, and successful separation of FMRP from the His-SUMO tag was verified with SDS-polyacrylamide gel electrophoresis (SDS-PAGE) gel. FMRP was concentrated and further purified using gel filtration chromatography with a buffer containing 4 M GdnHCl, 50 mM tris-HCl (pH 8.0), 500 mM NaCl, and 2 mM -mercaptoethanol.

Fluorescence protein labeling
An Alexa Fluor 647 fluorescent dye was added to the only cysteine (C584) in FMRP 445-632 via a maleimide linkage following the manufacturer's instruction with slight modifications. First, FMRP was dialyzed into a buffer containing 50 mM tris-HCl (pH 7.5), 100 mM NaCl, and 4 M GdnHCl. To ensure that any residual reducing agents were removed, the protein was desalted using a Hi-Trap desalting column (GE Healthcare). After desalting, the protein sample was immediately reacted with 5× Alexa Fluor 647 (Thermo Fisher Scientific) maleimide dye. The reaction was incubated overnight at 4°C and quenched with an excess of reducing agent [dithiothreitol (DTT)] the following day. To remove any unreacted dye, the protein was passed through a Hi-Trap desalting column (GE Healthcare) and an S75 gel filtration column equilibrated in buffer containing 50 mM tris-HCl (pH 7.5), 100 mM NaCl, 4 M GdnHCl, and 2 mM DTT. Successful dye separation was confirmed by running the protein sample on an SDS-PAGE gel and then visualizing any remaining free dye with a fluorescence reader, ChemiDoc MP System (Bio-Rad). Labeled proteins were either frozen or dialyzed into specific assay buffers.

RNA preparation
Sc1 RNA (GCUGCGGUGUGGAAGGAGUGGUCGGGUUGCG-CAGCG) was purchased from Sigma-Aldrich as lyophilized samples. Stocks (100 M) were reconstituted in water and stored at −20°C. Working stocks were diluted into specific assay buffers.

Turbidity measurements
For turbidity measurements, OD 600 of MBP-FUS was obtained using a SpectraMax i3× multimode plate reader (Molecular Devices) at 25°C. The samples were prepared by mixing varying concentrations of MBP-FUS with 0.5 M TEV protease in a buffer containing 25 mM tris-HCl (pH 7.4), 150 mM KCl, and 2 mM DTT. Samples were equilibrated for 5 min before reading the turbidity. Turbidity was measured at intervals of 35 s for a total of 20 min. The change in turbidity was calculated from the slope (∆ absorbance/min) from 0 to 5 min. Apparent C sats are calculated as previously described (48).

In vitro co-LLPS assays
In vitro phase separation assays of LCRs of FUS and FMRP were performed using FITC FUS LCR (50 M) and Alexa647 FMRP LCR (50 M) in the presence of 1 M sc1 RNA in a buffer containing 25 mM sodium phosphate (pH 7.4), 50 mM KCl, and 2 mM DTT. For partitioning assays using full-length FUS proteins (FUS WT , FUS P525L , and FUS 14 ), MBP-FUS (10 M) phase separation was induced by TEV protease (0.5 M) cleavage before the addition of Alexa647 FMRP LCR (1 M) and sc1 RNA (0.5 M) in a buffer containing 25 mM sodium phosphate (pH 7.4), 150 mM KCl, and 2 mM DTT.

Fluorescence microscopy of phase-separated samples
Fluorescence images of phase-separated droplets were imaged on a confocal Leica DMi8 microscope equipped with a Hamamatsu C9100-13 electron multiplying charge-coupled device (EM-CCD) camera with a 63× objective. Alexa Fluor 647 fluorescence was detected using a 637-nm laser, and FITC fluorescence was detected using a 491-nm laser. In experiments with MBP-FUS and MBP-FUS mutants, samples were incubated with TEV protease for 10 min before imaging. All phase-separated droplets were imaged on a 96-well glass plate (Eppendorf). Two-or threefold concentrated protein or RNA samples were prepared to account for the dilution in mixing with other components to achieve desired final concentrations. Note that no molecular crowding reagents were used. Images represent droplets settled to the bottom of the plate. The images were processed using Volocity (PerkinElmer) and ImageJ.

In vitro partitioning assay
To determine the partitioning of FMRP, images of droplets with the addition of 5% Alexa647 FMRP LCR were acquired as described above and analyzed with ImageJ. An image of the buffer in the absence of any protein was used to subtract any background artifacts. In ImageJ, masks were defined using the Otsu threshold method while applying several criteria to the particle picking algorithm: droplets are required to have a radius greater than 1 m and with the circularity of 0.5 to 1.0. The intensity of the bulk background solution is defined as the mean intensity within a circular region of interest with a diameter of 5 m that does not contain any phase-separated droplets. Fluorescence enrichment ratios were calculated from the ratio of the mean fluorescence intensity (inside droplet)/mean fluorescence intensity (background outside of droplet). Droplets were randomly imaged, and measurement represents three independent experiments.

In vitro translation assay
In vitro translation rates represent the increase in luminescence as a function of time using a standard rabbit reticulocyte lysate system (Promega) with luciferase mRNA (Promega). Manufacturer's instructions were followed with a few modifications. Briefly, each reaction (30 l) contains 12.6 l of rabbit reticulocyte lysate, 0.5 l of luciferase mRNA (1 mg/ml), 0.3 l of amino acid mixture minus leucine (1 mM), 0.3 l of amino acid mixture minus methionine (1 mM), 2 l of TEV protease (15 M), and 14.3 l of 5 M protein (MBP-FUS/FMRP) or buffer [25 mM sodium phosphate (pH 7.4), 50 mM KCl, and 2 mM DTT). First, the reaction was incubated for 10 min, and then, end-point luminescence measurements were carried out in intervals of 10 min up to 50 min. Each end-point luminescence measurement contained 75 l of luciferase substrate mixed with 2.5 l of unpurified translation mixture measured in a white opaque 96-well plate (Corning 3990). A SpectraMax i3× multimode plate reader (Molecular Devices) at 25°C was used to detect the luminescence. The translation rates represent the line of best fit from the end-point luminescence readings as a function of time.

RiboTag
The RiboTag method was performed as described previously (65) with modifications. Briefly, E17.5 spinal cords were homogenized using TissueRuptor (Qiagen) in a buffer [50 mM tris-HCl (pH 7.4), 100 mM KCl, 12 mM MgCl 2, and 1% NP-40] supplemented with cycloheximide (0.1 mg/ml), heparin (1 mg/ml), SuperaseIn RNase (Thermo Fisher Scientific). Lysates were cleared by centrifugation at 10,000g for 10 min, and 5% of the lysate was saved as input. To reduce nonspecific-binding protein G magnetic beads (Dynabeads, Thermo Fisher Scientific) were added to the lysate and incubated for 2 hours at 4°C. Next, 5 l of anti-HA antibody (Sigma-Aldrich) was added to the precleared lysate and incubated for 2 hours at 4°C. Later, 100 l of beads slurry with 2 l of SuperaseIn were added to the lysate followed by 2-hour incubation in the cold room. After precipitation, beads were washed five times in the wash buffer [300 mM KCl, 1% NP-40, 50 mM tris-HCl (pH 7.4), 12 mM MgCl 2 , and cycloheximide (0.1 mg/ml)]. Beads were eluted in Qiazol (Qiagen), and RNAs were isolated using the RNeasy Micro Kit (Qiagen). Ten percent of the beads were used for WB and eluted in Laemmli sample buffer (Thermo Fisher Scientific) supplemented with 100 mM DTT. The quality control of RNA was performed using TapeStation (Agilent).

RNA sequencing and analysis of RiboTag samples
Libraries were prepared using NEBNext mRNA Ultra II in the UCL Genomics facility and sequenced (75 base pairs single end) to an average depth of 18 million reads. Each sample was aligned to the Mus musculus (house mouse) genome assembly GRCm38 (mm10) with STAR (v2.4.2a) (66). Reads were coordinate-sorted and marked for PCR duplicates using Novosort (1.03.09). Gene expression was quantified using HTSeq using the Ensembl mm10 (v82) mouse transcript reference (67). Differential gene expression was calculated using DESeq2 (68) comparing the IP samples between the Fus −/− (n = 4) and Fus ∆14/∆14 (n = 5) with the same number of their respective littermate controls, in two separate analyses. The significance level was set at a false discovery rate-adjusted P value of 10%. To compare the two analyses, each nominal P value was converted into a z score and given the sign of the log 2 fold change. We defined condition-specific genes as having an adjusted P value of <0.1 in one condition and >0.1 in the other. A list of FMRP target genes was obtained from a published HITS (high-throughput sequencing)-CLIP experiment (30). Entrez IDs were converted to Ensembl IDs using g:Convert from the g:Profiler suite of tools (69). A list of FUS target peaks was obtained from a published iCLIP experiment (32). Peaks were annotated using annotatr (70) using gene models included in the TxDb.Mmusculus.UCSC.mm9.knownGene and org.Mm.eg.db R packages (71,72). FUS targets were filtered to only use peaks overlapping coding and UTR regions. GO terms analysis was performed in R using the enrichGO function from the clusterProfiler package (73).

Statistical analysis
Unless otherwise stated, data were obtained using cells from at least three independent preparations, which are visualized in different shades of gray in the graphs. The number of cells studied is given in the figure legends. GraphPad Prism or R was used for statistical analysis. Normality of data distribution was tested using D'Agostino and Pearson normality test. One-way analysis of variance (ANOVA), followed by Dunnet's post hoc test, was used for normally distributed data and multiple comparisons, while Sidak's post hoc test was used for pairwise comparisons. Kruskal-Wallis, followed by Dunn's post hoc test, was used for not normally distributed data and multiple comparisons. Friedman's test, followed by Dunn's post hoc test, was used to compare normally distributed paired samples. Individual differences were assessed using individual Student's t tests. Data are shown as means ± SEM. Kolmogorov-Smirnov test was used in cumulative frequency analysis to test differences between targets and nontargets of FUS and FMRP.

SUPPLEMENTARY MATERIALS
Supplementary material for this article is available at http://advances.sciencemag.org/cgi/ content/full/7/30/eabf8660/DC1 View/request a protocol for this paper from Bio-protocol.