Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disorder characterized by proteinaceous intracellular inclusions and progressive motor neuron loss. Fused in sarcoma (FUS) is an ALS-associated protein, which carries an aggregation-prone low-complexity N-terminal domain involved in transcriptional activation. While certain mutations in this domain have been shown to accelerate the kinetics of aggregation of FUS, the corresponding impact of other common mutations is still unclear. For example, the P525L mutation affects the proline-tyrosine nuclear localization signal (PY-NLS) at the FUS C-terminal domain and disrupts its interaction with the nuclear import receptor transportin-1, leading to the formation of aberrant FUS deposits in the cytoplasm. The aggregation of mutant FUS may derive from its excessive accumulation within the cell, an increased aggregation propensity induced by the P525L mutation, or both. Here, we investigate whether the P525L mutation modifies the biochemical and biophysical properties of purified FUS by analyzing its phase separation kinetics. We demonstrate that this single amino acid change is sufficient to reduce the solubility of the FUS PY-NLS, leading to a stronger aggregation tendency. These observations indicate that the remarkable aggregation propensity of P525L FUS is linked to both a supersaturation caused by an increased concentration following mislocalization and a change of biochemical properties that reduce its intrinsic solubility.

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disorder of upper and lower motor neurons^[1]. While 90% of ALS cases are idiopathic, the remaining 10% have been associated with inherited mutations that typically segregate in an autosomal dominant fashion^[2].

Mutations in fused in sarcoma (FUS) are the third most common genetic cause of ALS and are associated with the presence of FUS-immunoreactive cytoplasmic inclusions in both cortical and spinal motor neurons of FUS-ALS patients^[3][4][5]. FUS is a multifunctional DNA/RNA-binding protein of 526 amino acids, which shuttles between nucleus and cytoplasm to carry out a variety of tasks linked to nucleic acid metabolism^[6][7]. FUS mutations have been described to span its entire length^[8]. However, the C-terminal proline-tyrosine nuclear localization signal (R/H/KX2−5 PY-NLS) domain distinguishes itself for being a hot spot for pathogenic sequence variations, with about 25 recognized ALS-causing mutations clustering between residues 510–526^[9]. Previous reports have demonstrated that the non-classical PY-NLS of FUS is critical for nuclear localization mediated by the nuclear import receptor transportin-1 and that mutations that disrupt this interaction induce the mislocalization of FUS to the cytoplasm^[10]. Analysis of the crystal structure of the FUS-NLS/transportin-1 complex has shed some light on the structural aspects of the interaction between the two proteins, highlighting the crucial role of a unique α-helical structure in the FUS-NLS, and identifying the hydrophobic residues P525 and Y526 as a key to forming the specific conformation required for binding transportin-1^[11][12]. Additionally, binding affinity studies have revealed that FUS-NLS ALS-associated mutations cause a decrease in affinity ranging from 1.4–fold to over 700–fold, correlating with the extent of protein mislocalization, and more notably, with disease onset and progression in patients^[11].

Of the numerous mutations affecting the FUS C-terminus, P525L has been consistently associated with severe juvenile forms of ALS^{[13][14][15][16][17]}. This severity has been attributed to the strong cytoplasmic accumulation and aggregation of FUS resulting from the disruption of the critical proline in position 525 of the FUS NLS^[11][12]. We wondered whether P525L would drive aggregation by solely increasing the FUS cytoplasmic concentrations above critical levels, thus making it supersaturated^[18], or whether there could be any additional mechanisms underlying the aggressive nature of this mutation. For instance, while the FUS N-terminal domain carries an aggregation-prone low-complexity region whose mutations have been shown to accelerate the kinetics of FUS aggregation^[19], it remains unclear whether C-terminal mutations, such as P525L, may have a similar impact.

In recent years, the field of protein aggregation has rapidly evolved from being a largely unexplored topic of protein chemistry to become a central subject in the biotechnology and medical sciences. This rising interest has been linked to increasing evidence connecting protein misfolding and aggregation to a number of human disorders, including neurodegenerative diseases^[20]. The identification of aggregation-prone toxic species has paved the way for new targets in the arena of drug discovery. For example, a number of strategies to either prevent, delay, or eliminate aggregates are actively being investigated^[21]. Therefore, understanding the aggregation dynamics of FUS in FUS-ALS would represent a distinct advantage for the development of novel therapeutic approaches.

In this study, we use a combination of protein solubility prediction tools and experimental measurements to assess the effects of the P525L mutation on the aggregation of FUS irrespective of its subcellular localization. We predict this mutation to cause a drop in solubility, and, comparing the phase separation profiles of purified wild-type (WT) and P525L FUS, we confirm this prediction experimentally.

The objective of this study is to assess whether the P525L mutation in FUS increases the aggregation propensity of FUS.

To gain information on the solubility profile of FUS, we initially relied on bioinformatics tools. In particular, we used the CamSol software^[22], which allows for rapid in silico computation of the intrinsic solubility scores of each protein residue, and highlights their impact on the overall solubility of the protein under investigation. WT FUS recorded an overall intrinsic solubility value of 3.17, with the lowest solubility scores clustering at its N-terminal low-complexity domain (Fig. 1A). The C-terminal domain of WT FUS showed high solubility (R522 = 2.89, E523 = 2.43, R524 = 2.26, P525 = 1.91, Y526 = 1.45) (Fig. 1B). Introducing the P525L mutation strongly affected the PY-NLS, causing a drastic drop in its solubility profile (R522 = 2.34, E523 = 1.88, R524 = 1.62, L525 = 1.15, Y526 = 0.50) (Fig. 1C). We further confirmed these results using Aggrescan3D^[23], which revealed that P525L induces an increase in the NLS aggregation propensity (Suppl. Fig. 1B). Next, we investigated the impact of P525L on the conformational propensity of FUS using the s2D software^[24]. We observed that both WT and P525L FUS are extensively disordered across their entire length, with the exception of two α-helices in the region of residues 295–305 and 342–352 (Fig. 1D). In particular, we found that the FUS NLS is primarily made up of random coils, as opposed to the α-helical conformation determined for this region by X-ray crystallography using the FUS NLS/transportin-1 complex as a template (PDB ID: 4FQ3 or 4FDD)^[11][12]. These observations suggest that the α-helical structure of this domain (Suppl. Fig. 1A) may derive from a coil-to-helix transition resulting from the specific interaction between these two proteins^[25][26]. Interestingly, calculations using AGADIR (http://agadir.crg.es)^[27][28] on the NLS fragment indicated that the P525L mutation increases the α-helical propensity of the FUS NLS (from 0.32 to 0.74), suggesting that this is how the mutation may alter the stability of the FUS/transportin-1 complex. Additionally, the proline to leucine change in position 525 was flagged as destabilizing for the protein structure based on calculations using FoldX^[29], where mutations resulting in a predicted reduction in protein stability ≥1 kcal/mol are typically considered as disruptive. The calculated difference in free energy between WT and mutated FUS was 3.87 kcal/mol. We speculate that P525L may affect the availability of the PY-NLS at the interface with transportin-1 and become refractory to crucial post-translational modifications. For instance, hypomethylation of arginine residues within the C-terminal domain of FUS has been reported to favor cooperative cation-π interactions with tyrosine residues in the N-terminal low complexity domain inducing phase separation^[30].

In order to substantiate this in silico prediction with experimental measurements, we analyzed the behavior of FUS in vitro using purified proteins. Previous reports described that full-length WT FUS rapidly phase separates at physiological levels (∼1–2 μM) in the presence of low salt concentrations^[30]. This phenomenon is due to the promotion of the interaction between oppositely charged groups on the protein. Based on this principle, we induced phase separation by diluting purified WT FUS protein without further addition of any molecular crowder. This simple step caused the formation of large numbers of small spherical droplets, whose behavior could be monitored in real-time using fluorescence microscopy. We observed that FUS droplets were initially extremely dynamic and characterized by frequent fusion events, which gradually led to the formation of larger assemblies of reduced motility. Over a time-course of 24 h, hundreds of droplets of ∼0.5 µm in diameter had coalesced into a reduced number of much larger droplets (>10 µm) (Suppl. Video). This time-course suggested that the kinetics of droplet evolution towards larger-sized assemblies would be accelerated in the presence of an aggregation-prone mutation. Thus, to evaluate whether P525L affects the phase separation profile of FUS, we aimed at measuring droplet evolution.

Of note, the purification of P525L FUS was particularly challenging due to its strong aggregation in standard buffer conditions, which made the protein rapidly unavailable for column purification. This phenomenon could be readily monitored by measuring turbidity over time, with which we observed that lysates containing P525L FUS quickly progressed towards the aggregated state (Fig. 1E). This process was prevented by the addition of urea (Fig. 1E), a known protein denaturing agent. Indeed, FUS patient inclusions have been reported to be insoluble in RIPA buffer but soluble in highly concentrated urea^[31]. Following this observation, we included urea in each purification buffer, which enabled us to isolate P525L FUS. To favor droplet formation in these new conditions, we tested the addition of dextran as a molecular crowding agent. Including 5% dextran induced phase separation of WT FUS into protein droplets (Fig. 1F). In contrast, P525L FUS rapidly formed aggregates without even transitioning through phase separation (Fig. 1F). Interestingly, these aggregates stained positive for the amyloidophilic dye AmytrackerTM (Suppl. Fig. 1C), which we previously tested for its ability to recognize amyloidogenic proteins by administering it to cells transfected with α-synuclein fibril seeds (Suppl. Fig. 1D). We further confirmed aggregation via turbidity measurements (Fig. 1G). Based on these data, we concluded that the P525L mutation greatly increases the propensity of FUS for aggregation.

Since P525L FUS aberrantly accumulates in the cytoplasm, its concentration may increase above its solubility limits. To test this hypothesis, we monitored the phase behavior of FUS at increasing protein concentrations. In order to decouple the consequences of the point mutation on phase transition, we analyzed the protein in its WT form (Fig. 1H). As it was previously described that cytotoxicity correlates with increasing cytoplasmic FUS concentrations in iPSC-derived neurons^[32], we tried to recapitulate these conditions in our in vitro phase separation assay. To this end, we monitored FUS droplet formation over time using 1X, 2X, 4X, and 8X FUS (Fig. 1I). We confirmed that FUS-eGFP droplets increase in size upon longer incubation periods (Fig. 1J), with 8X FUS droplets reaching a size plateau after 2 h. We also found that increasing concentrations of the protein promoted earlier nucleation events and faster progression towards the formation of larger assemblies (Fig. 1J). Importantly, droplet circularity was significantly reduced at the latest time point and concentration considered (Fig. 1K), suggesting the beginning of a liquid-to-solid phase transition. In conclusion, these results are in line with the notion that the increased cytoplasmic concentration of FUS caused by its mislocalization promotes its aggregation.

Healthy cells attempt to refold misfolded proteins using molecular chaperons, or degrade them via the proteasome and/or autophagy^[33]. However, these mechanisms are not infallible, and their efficiency declines with age^[34][35]. Amyloid-like aggregates are typically resistant to degradation^[36], and progressive overload of the cellular systems in charge of protein homeostasis will lead to the gradual accumulation of aggregated species, eventually perturbing cellular function. In human neurons, increased death rates have been observed following a cytoplasmic accumulation of mutant FUS^[32], although the exact molecular mechanisms causing decreased viability have not been fully elucidated. Mutant aggregation-prone FUS may exert cytotoxicity via toxic gain of function by physically disrupting organelle trafficking along neuronal axons^[37], or by sequestering cytosolic proteins and impairing global as well as local protein synthesis^[38]. In this context, reducing protein accumulation and aggregation by targeting the production and processing of mutant FUS, could open new therapeutic avenues.

Our results indicate that the severity of P525L FUS ALS is not only linked to a disruption in the interaction between FUS and transportin-1, which induces the accumulation of mutant FUS in the cytoplasm, but also to changes in the biochemical and biophysical properties of the FUS PY-NLS, which reduce the intrinsic solubility of FUS.

The physiological concentration of WT FUS in the nucleus was previously reported (∼1–2 μM)^[30]. Unfortunately, biochemical measurements of cytoplasmic FUS levels in healthy cells have not been reported to date. Studies performed on human iPSC-derived neurons to characterize the relative abundance of cytoplasmic FUS in cells exhibiting different levels of pathology indicating that neurons with mislocalized FUS displayed 2X, 4X, and 8X more cytoplasmic protein than healthy cells^[32]. These fold change values were obtained by analyzing the FUS fluorescence signal in confocal micrographs. In the absence of a reference concentration value for cytoplasmic FUS, we used the known nuclear concentration of FUS to build up a concentration range that reflected the fold change observed in human neurons. To more effectively recapitulate physiological conditions, the present study would require repetition following assessment of the actual biological concentration of WT FUS in the cellular cytoplasm.

The purified FUS protein used in this study contains an MBP-eGFP-His tag, which according to CamSol analysis reduced the overall protein solubility from an intrinsic solubility value of 3.17 to a value of 1.64. While this reduction in solubility facilitated phase-separation in vitro, future experiments would benefit from the analysis of an untagged FUS control to corroborate the aggregation behavior observed in this study. This could be confirmed including additional biophysical techniques to the ones explored in this work, such as sedimentation analysis or SDS resistance.

Proteins that readily undergo phase separation are characterized by architectural features that facilitate self-interaction and aggregation, thus hampering purification^[39]. Therefore, standard purification protocols often do not apply. To isolate P525L FUS we used 2 M urea. This procedure prevented aggregation while preserving relatively mild conditions, and enabled us to obtain enough protein for testing. However, purified P525L FUS was comparatively much less abundant than WT, and it did not reach the highest purity after extraction. Increasing urea concentration to 6–8 M would likely boost protein yields and quality. For these reasons, we analyzed the concentration dependence of the phase behavior of WT FUS, while future studies will be required to extend these results to P525L FUS.

We showed that purified P525L FUS rapidly forms aggregates and that these aggregates stained positively for the amyloidophilic dye AmytrackerTM. This suggested that FUS may form ordered fibers rather than amorphous aggregates at the tested conditions. However, further experiments will be required to thoroughly assess the morphology of in vitro FUS aggregates, for instance by using electron microscopy.

Although this study was performed in vitro to eliminate a variety of sources of variability, we are aware that the primary limitation of any in vitro study is that it may be reductive. For instance, cell-based studies have recently shown the existence of a number of other mechanisms affecting the phase behavior and aggregation propensity of FUS, including low cytoplasmic RNA levels^[40], differential interactions with other RNA-binding proteins^[32] and active recruitment to RNP granules^[41][42]. In this broader context, our study contributes to highlighting a complex scenario orchestrated by multiple players and events.

Bioinformatic analysis

Protein solubility was predicted using CamSol intrinsic^[22]. The CamSol method is based on a combination of algorithms that assign a score to the intrinsic solubility of protein residues based on linear protein sequences. This score is tested against solubility changes that have been experimentally determined and reported in the literature. Aggregation propensity was validated using Aggrescan3D^[23]. Aggrescan 3D does not simply rely on linear protein sequence but takes into consideration protein structure as well as empirical data on aggregation derived from the cellular setting. In particular, the analysis of the aggregation profiles of a number of mutant Aβ peptides in E. Coli has allowed deriving a scale of intrinsic aggregation values for each natural amino acid, which represents the basis of the algorithm. Contrary to standard sequence-based algorithms, Aggrescan 3D uses 3-dimensional structures as inputs, i.e. PDB files. This enables accurate predictions especially in the case of structured proteins, where aggregation-prone regions may be naturally masked inside the hydrophobic core of the protein. To analyze the aggregation propensity of the FUS C-terminus, we used the PDB entry "4FDD", with a distance of aggregation of 10 Å and static mode. Results referred to chain B, corresponding to the 498–526 fragment of FUS (GPGKMDSRGEHRQDRRERPY). FoldX measurements relative to the free energy of the FUS C-terminus were obtained in conjunction with Aggrescan 3D results. Intrinsic disorder and secondary structure were predicted with s2D^[24], which exploits information from NMR measurements to predict the probability distributions of secondary-structures in disordered proteins using their linear sequence as an input. AGDIR was used to predict the helical content of the GPGKMDSRGEHRQDRRERPY and GPGKMDSRGEHRQDRRERLY peptides, respectively.

Purification of recombinant proteins

Full-length recombinant human MBP-FUS-eGFP-His protein was isolated from eukaryotic Sf9 cells cultured in suspension in Insect Cell Medium (ESF921, Expression Systems). For protein over-expression, Sf9 cells were infected with recombinant baculoviruses at a density of 1×106 cells/mL. After 4 days of incubation at 27°C in agitation, cells were harvested for protein purification. To this end, infected Sf9 cells were centrifuged at 4000 × g for 30 min. Pellets were resuspended in lysis buffer (50 mM Tris-HCl pH 7.4, 1 M KCl, 0.1% CHAPS, 5% glycerol, 1 mM DTT, and protease inhibitors) and disrupted mechanically. The lysate was ultra-centrifuged at 40,000 rpm for 20 min at room temperature (RT). Following resin equilibration in 50 mM Tris-HCl pH 7.4, 1 M KCl, 5% glycerol, and 1 mM DTT, the clear supernatant was poured through a nickel (Ni)-Excel column to allow for FUS-eGFP binding via its His-tag. The protein was then washed with the same buffer containing 20 mM imidazole and eventually eluted with 250 mM imidazole. Next, the eluate was loaded onto an amylose column (E8021L, NEB) for further protein selection. The bound protein was eluted with a buffer containing 10 mM maltose. Following protein concentration, the eluate was eventually loaded onto an AKTA system for size-exclusion chromatography and the fractions containing FUS-eGFP were pooled. To allow for P525L FUS protein purification, all buffers were supplemented with 2 M urea. Purified proteins were concentrated using 3–50 kDa Amicon Ultra 0.5–15 Centrifugal filters (Sigma). Concentrations were determined to measure their absorbance at 280 nm using theoretical extinction coefficients calculated with Expasy ProtParam. Proteins were aliquoted, flash-frozen in liquid nitrogen, and stored at -80°C.

SDS page and Coomassie

Proteins were diluted in 4X Lämmli buffer, boiled for 5 min at 95°C, loaded onto an SDS-page gel, and run for 25 min at 200 V. Gels were incubated in Coomassie blue Fast Stain (GeneCopoeia) for band revelation.

Phase separation experiments

Unless otherwise stated, phase separation was induced by diluting the protein stock (1 M KCl) to 40 mM KCl and a working concentration of 1.5 µM FUS (1X) in the presence of 5% dextran. After vortexing, the mixture was pipetted onto slides (Ibidi), and FUS-eGFP droplet evolution was monitored over time using an Axiovert 200M microscope (Zeiss) at 100X magnification. Droplet size and circularity were calculated using Fiji after removing background noise using the Process -> Subtract Background function. The phase-separation time course reported in "Supplementary Video" displays images acquired at 0 min, 10 min, 30 min, 45 min, 1 h, 1.5 h, 2 h, 3 h, 4 h, 5 h, 6 h, 8 h and 24 h post-induction of phase separation.

Turbidity assay

Following induction of phase-separation as described above, samples were pipetted onto 96-well black/clear bottom plates (6005182, PerkinElmer) and absorbance was measured at 405 nm using a Spectra Max M5 plate reader (Molecular Devices) endowed with a SoftMax Pro 7.1 software. Turbidity measurements of Sf9 cell lysates were performed by sampling a fraction of the lysate prior to column purification and recording absorbance values over time while leaving the plate at RT.

Structural properties of FUS assemblies

AmytrackerTM dyes are molecules that become highly fluorescent when bound to their target, i.e. pre-fibrillary and fibrillary states of amyloids. To stain FUS aggregates, the AmytrackerTM 680 (Ebba Biotech) stock was diluted 1:10,000 in the phase-separation mixture. Images were acquired using the red channel.

NSC34 cell culture and transfection

Cells were cultured in T75 flasks using DMEM medium (D5796, Sigma) supplemented with 10% FBS (A15-751, PAA) and split with trypsin at a 1:3 dilution when 80% confluent. Transfection was performed on 6-well plates. Fully confluent wells were supplemented with 1 µM aggregated α-synuclein protein added directly to the medium. After 24 h, cells were incubated with AmytrackerTM 680 diluted 1:500 in fresh medium for 30 min prior to fixation. Cells were fixed with 4% PFA, stained with Hoechst, and mounted for imaging. Images were acquired using an Axiovert 200M microscope (Zeiss) at 100X magnification.

Generation of α-synuclein seed fibrils

Seed fibrils were produced as previously described by Buell et al.^[43], with minor modifications. Briefly, 500 µL aliquots of α-synuclein at concentrations ranging from 500 to 800 µM were incubated in 20 mM phosphate buffer, pH 6.5 for 48–72 h at 40°C and stirred at 1500 rpm with a Teflon bar on an RCT Basic Heat Plate (IKA). Fibrils were diluted to a monomer equivalent concentration of 200 µM, divided into aliquots, flash-frozen in liquid N2 and stored at -80°C. Prior to use, the 200 µM fibril stock was sonicated for 15 sec using a probe sonicator (Bandelin, Sonopuls HD2070), set at 10% maximum power and 50% cycle.

Lara Marrone was supported by the Boehringer Ingelheim Fonds (BIF) and the Hans und Ilse Breuer Stiftung.

The authors would like to thank Naunehal Matharu and William Meadows for their technical assistance. We additionally thank Dr. Jared Sterneckert for his advice. Finally, we commemorate the late Sir Professor Christopher Dobson and thank him for his support for this project.

The authors declare no conflicts of interest.

Not applicable.

No fraudulence is committed in performing these experiments or during processing of the data. We understand that in the case of fraudulence, the study can be retracted by ScienceMatters.