Comparative interactomics analysis of different ALS-associated proteins identifies converging molecular pathways

For the co-immunoprecipitation of endogenous protein complexes, N2A or NSC-34 cells from one 10-cm dish were lysed in 150 μl lysis buffer [20 mM Tris–HCl pH 7.5, 150 mM NaCl, 1 % NP-40, 10 % glycerol, complete protease inhibitor cocktail (Roche)], incubated on ice for 10 min and centrifuged at 13,200 rpm for 10 min at 4 °C. Lysates were incubated for 1 h at 4 °C using 1 μg antibodies. Detailed information on antibodies and plasmids used can be found in the Supplementary Material. Samples were incubated with protein A or protein G magnetic beads for 30 min followed by 4 washes in lysis buffer. Precipitated proteins were eluted by boiling in NuPage LDS sample buffer containing 10 mM DTT for 5 min at 90 °C. For RNA dependency experiments, lysates were first treated with RNAse A (Roche) for 30 min at 4 °C. Recombinant FUS (Origene) and FMRP (Abnova) proteins were mixed in buffer (10 % glycerol, 0.5 % NP40, 100 mM glycine, 25 mM Tris–HCl, pH 7.4) and incubated for 1 h with protein A agarose beads pre-incubated with rabbit anti-FUS antibody. Immunoprecipitation was performed as described above. For RNA dependency experiments, total RNA was isolated from HEK293 cells using the RNAeasy mini kit (Qiagen) according to the manufacturer’s instructions and added to the protein mixture at concentrations ranging from 0.1 to 62.5 ng/μl. For the generation of crude mitochondrial fractions, N2A cells were homogenized in homogenization buffer [0.25 M sucrose, 0.2 mM EDTA, 20 mM HEPES pH 7.4, complete protease inhibitor cocktail (Roche)] and a pellet containing crude mitochondria was obtained after differential centrifugation of the lysate.

Proteins were separated in a SDS-PAGE gel and transferred onto nitrocellulose membrane (Hybond-C Extra; Amersham). Membranes were incubated in blocking buffer (TBS, 0.05 % Tween and 5 % milk powder) for 30 min at room temperature (RT), followed by incubation with the appropriate antibody overnight at 4 °C. After several washes with TBS-T, membranes were incubated with appropriate peroxidase conjugated secondary antibodies in blocking buffer for one hour at RT followed by incubation with Super Signal West Dura Extended Duration Substrate (Pierce) and exposed to ECL films (Pierce). For quantification of Western blots, intensity measurements of the protein bands were performed using ImageJ.

Immunoprecipitation of biotin-tagged proteins was performed as described previously [16, 28]. It is important to note that cell lysis could allow bait proteins to interact with proteins that they normally would not be able to bind (e.g., because of their localization in distinct cellular compartments). While this is an important point to consider, so far we have performed bioIPs for >40 different bait proteins (including cytoplasmic, nuclear or membrane-associated proteins) and never found strong evidence for such post-lysis artifacts.

Samples were separated in a NuPAGE Novex 4–12 % Bis/Tris gradient gel following the manufacturer’s instructions. For silver staining, gels were fixed in 50 % methanol for 30 min, followed by incubation in 10 μM DTT for 15 min and incubation in 0.1 % (w/v) AgNO₃ for 20 min. Subsequent gels were developed using 0.25 M anhydrous sodium carbonate containing 0.02 % (w/v) formaldehyde. For mass spectrometry analysis, gels were stained using GelCode blue stain reagent (Pierce). Mass spectrometry analysis of interacting proteins was done as described previously [96]. Raw data were analyzed by MaxQuant (version 1.4.1.2) [12] as described in the Supplementary Material. GO analysis was performed using PANTHER [57].

All animal care and use was in line with institutional, national and European legislation. Mice (C57BL/6) were purchased from Charles River. Primary motor neurons were isolated from the ventral part of spinal cords of E13.5 mouse embryos and cultured as described in the Supplementary Material. After 48 h, cells were transfected using magnetic beads (Oz Biosciences) as described by Fallini et al. [20].

Primary motor neurons were fixed with 4 % PFA for 15 min at RT, permeabilized with 0.1 % Triton X-100 in PBS for 5 min at RT, blocked in PBS containing 2.5 % BSA, and incubated with primary antibodies in BSA supplemented with 5 % normal goat serum for 1 h at RT. After several washes in PBS, cells were incubated with a mixture of the appropriate Alexa Fluor^®-labeled secondary antibodies (Life Technologies) for 1 h at RT. Then, cells were washed, counterstained with 4′,6′-diamidino-2-phenylindole (DAPI) (Sigma), washed extensively with PBS and mounted in DABCO.

Consent for autopsy was obtained in concordance with institutional regulations. Patient and control details are included in Supplementary Table S3. Double immunohistochemistry and double fluorescent-labeling were performed as described in the Supplementary Material. Due to antibody incompatibilities we were unable to perform double fluorescent-labeling for all interactors. Sections were analyzed using a laser scanning confocal microscope (Leica TCS Sp2, Wetzlar, Germany).

Zebrafish were kept and maintained under standard conditions. Injections were performed at the one-cell stage. FUS WT, FUS R521C and mCherry-FMRP were subcloned into a pCS2+ vector and mRNA was transcribed using the mMESSAGE Machine kit (Ambion) followed by phenol–chloroform purification and ethanol precipitation. mRNAs were diluted in RNAse free water (with 1 % phenol-red dye) at a concentration of 75–200 ng/μl and were pulse-injected using a pressure ejector. Touch-evoked escape response (TEER) was measured in developmentally normal zebrafish, as described previously [41].

For immunohistochemical analysis, zebrafish larvae were fixed in 4 % PFA overnight at 4 °C. Larvae were rinsed in PBS with 0.1 % tween (PBS-T) and then permeabilized with collagenase A (1 mg/ml) for 45 min at RT and stained consecutively with Alexa Fluor^® 488-conjugated alpha-bungarotoxin and anti-SV2 antibody as described in the Supplementary Material.

For FMRP granule analysis, primary motor neurons were imaged using a Zeiss Axioskop2 microscope in combination with Axiovision SE64 software. For quantification of FMRP granules, an automated method in ImageJ was designed and statistical significance was tested using an ANOVA. For the zebrafish NMJ analysis, images were acquired using an Olympus Fluoview FV1000 confocal microscope and overlap between pre- and postsynapse was determined using the JACoP plugin for ImageJ [6] with Costes’ automatic thresholding.

Zebrafish embryos were collected at 72 hpf and yolk sacs were removed by triturating in ice-cold Ringer’s salt solution (116 mM NaCl, 2.9 mM KCl, 1.8 mM CaCl₂, 5 mM HEPES) with 50 mM EDTA. Then, 200–500 embryos were homogenized per condition in homogenization buffer (320 mM sucrose, 4 mM HEPES supplemented with protease inhibitors) and synaptosomes were obtained after differential centrifugation of the lysate.

Polysomal fractionations from N2A cells were performed using a protocol adapted from Gismondi et al. [27].

Identification of protein binding partners of proteins associated with ALS can aid the characterization of the pathogenic mechanisms underlying this disease [21, 23, 28, 52, 76, 77, 90]. To further dissect ALS disease mechanisms and to possibly unveil common disease pathways downstream of multiple ALS-associated proteins, we used a biotin-streptavidin pull down system [16, 28] in combination with mass spectrometry to determine the interactomes of six ALS-associated proteins. ATXN2, C9orf72, FUS, OPTN, TDP-43 and UBQLN2 were selected on basis of genetic and pathological data. These proteins are the focus of many current studies on ALS pathogenesis, are localized to ALS-associated inclusions in a large proportion of patients [5, 43], and/or are poorly characterized at the functional level (C9orf72).

To identify interactors, biotin-GFP-tagged (bioGFP) constructs were generated for ATXN2, C9orf72, FUS, OPTN, TDP-43, and UBQLN2 as well as for a common ALS-associated mutant variant of each of these proteins (ATXN2 31Q, 39Q, FUS R521C, OPTN E478G, TDP-43 M337V and UBQLN2 P497H) (Fig. 1a). Hexanucleotide repeat expansions in C9orf72 cause ALS but these expansions are found in the non-coding part of the gene and are therefore not present in C9orf72 protein [17, 68]. Nevertheless, C9orf72 was included to further define its functional role(s). Neuronal cells (N2A) were co-transfected with plasmids encoding the bacterial biotin ligase BirA and one of the bioGFP-tagged constructs. Importantly, the subcellular localization of transfected epitope-tagged proteins was indistinguishable from the reported distribution of their wild-type or ALS-mutated endogenous counterparts (Fig. 1b). Furthermore, immunoblotting showed that expression of epitope-tagged ATXN2, FUS, OPTN, and TDP-43 was equivalent to or lower than endogenous proteins, while exogenous UBQLN2 was expressed at slightly higher levels. Endogenous expression of C9orf72 was low in N2A cells consistent with weak expression of C9orf72 in other cell types [33, 83] (Fig. 1c). Biotinylated bait proteins and their binding partners were captured from N2A cell lysates with magnetic streptavidin beads, separated on gel and subjected to immunoblotting with anti-GFP antibodies or silver staining to confirm pull down efficiency (Figs. 1d, S1, S2). Following this confirmation, samples were subjected to mass spectrometry analysis. Identification and quantification of proteins was carried out using MaxQuant software [11] and statistical analyses were performed using the open PERSEUS environment. For pairwise comparison of GFP and WT interactomes, and WT and mutant interactomes, Welch’s t test statistics were applied with an FDR of 0.01 and S0 of 1.5 (the S0 parameter sets a threshold for minimum fold change). With these criteria, 163 interactors of ATXN2, 53 interactors of C9orf72, 123 interactors of FUS, 33 interactors of OPTN, 140 interactors of TDP-43, and 104 interactors of UBQLN2 were identified. These interactors included both known and novel binding partners of the different ALS-associated protein baits. It is important to note that some of the identified interactors may engage in indirect interactions that require other proteins for binding to the selected bait proteins. Below each of the interactomes identified in this study is discussed in more detail.

Mutations may not only cause protein mislocalization or affect enzymatic activity but could also influence protein–protein interactions. To determine whether the ALS-associated mutations selected here influence protein–protein interactions, we compared the interactomes of wild-type and mutant proteins. No statistically significant differences were detected between the interactomes of wild-type and mutant ATXN2, FUS and TDP-43. However, OPTN and UBQLN2 wild-type and mutant proteins showed altered protein interaction patterns (Fig. 3a, b). This difference was most prominent for mutant OPTN, where interactions with 13 proteins were detected for OPTN WT but not OPTN E478G. These interactors were Casein kinase II subunit alpha (Csnk2a1, Csnk2a2), Pre-mRNA-splicing factor ATP-dependent RNA helicase DHX15 (Dhx15), IgE-binding protein (Iap), cAMP-dependent protein kinase type I-alpha regulatory subunit (Prkar1a), Rnf31, Dolichyl-diphospho-oligosaccharideprotein glycosyltransferase subunit 1 (Rpn1), Sqstm1, Tax1bp1, TBC1 domain family member 15 (Tbc1d15), Ub, Vcp and ATPase WRNIP1 (Wrnip1) (Fig. 3a). Other interactors that were specific for wild-type OPTN but that did not pass our selection criteria were Protein 2410002F23Rik, BAG family molecular chaperone regulator 2 (Bag2), DnaJ homolog subfamily B member 6 (Dnajb6), FAS-associated factor 2 (Faf2), Flotillin-1 (Flot1), Flotillin-2 (Flot2), Pericentriolar material 1 protein (Pcm1), Peptidyl-prolyl cis-trans isomerase A (Ppia), Proteasome subunit alpha type-7 (Psma7), Ras-related protein Rab-1A (Rab1a), RanBP-type and C3HC4-type zinc finger-containing protein 1 (Rbck1), Sphingosine-1-phosphate lyase 1 (Sgpl1), Store-operated calcium entry-associated regulatory factor (Tmem66) and Ufd11. The proteins that failed to bind mutant OPTN function in ubiquitin-dependent protein degradation and ER transport.

UBQLN2 P497H showed increased binding affinity for clathrin heavy chain 1 (Cltc) compared to UBQLN2 WT (Fig. 3b). Clathrin is a major component of clathrin-coated pits (CCPs) and vesicles involved in intracellular trafficking and receptor endocytosis. In summary, our pull down experiments do not reveal dramatic changes in protein–protein interactions caused by ALS-associated mutations, with the exception of OPTN E478G.

The experimental design of our study, i.e., parallel analysis of six ALS-associated proteins using an identical biochemical workflow, provided a unique opportunity to investigate potential overlap between ALS-associated interactomes. Comparison of the different wild-type interactomes identified high overlap between ATXN2, FUS and TDP-43 interactomes and a considerable overlap between the interactomes of OPTN and UBQLN2 (Fig. 3c, d). In contrast, the C9orf72 interactome did not overlap with any of the other interactomes. ATXN2, FUS and TDP-43 shared 69 interactors (Supplementary Table S2). These included RNA helicases, eIFs, Elavl proteins, hnRNPs, ribosomal subunits, components of the exon junction complex, spliceosome and other RNA-binding proteins. GO analysis of the shared proteins identified processes such as RNA processing, posttranscriptional regulation of gene expression, splicing and gene expression, which is in line with the reported functions of ATXN2, FUS and TDP-43 (Fig. 3c). The overlap between OPTN and UBQLN2 included Sqstm1, Vcp, Ubc, Tbk1 and Tax1bp1. These shared interactors have reported roles in autophagy and NF-κB signaling (Fig. 3d). Thus, ATXN2-FUS-TDP-43 and OPTN-UBQLN2 share interactors that are part of common cellular protein complexes and are likely to affect similar cellular processes.

Molecular pathways shared by different ALS-associated proteins provide an opportunity to understand and treat ALS pathogenesis in larger groups of patients. To exploit this opportunity, we selected six interactors shared by ATXN2, FUS and TDP-43 on the basis of interactome abundance scores, literature and availability of experimental tools (FMRP, Upf1, Caprin1, HuD (Elavl4), Pabpc4, and Dhx9). First, co-immunoprecipitation was used to confirm the binding of these interactors to endogenous ATXN2, FUS and TDP-43 (Fig. 4a–c). One model to explain how ALS mutations cause neuron degeneration is that mutant proteins sequester their interactors into aggregates thereby causing their depletion and defects in specific cellular processes (e.g. [28, 60]). Therefore, we next determined whether Fmrp, Upf1, Caprin1, HuD, Pabpc4 or Dhx9 show an altered distribution in the presence of mutant FUS. Several ALS-associated mutations in FUS cause robust cytosolic mislocalization and formation of inclusions, in experimental settings or human motor neurons, respectively (e.g. [28, 29, 31, 48, 81, 82, 91]). GFP, wild-type FUS or FUS R521C were transfected in primary mouse motor neurons and subjected to immunocytochemistry for FUS and the different interactors at 2 days in vitro (DIV2). Following transfection of FUS R521C condensed accumulations of mutant FUS protein were detected in the cytoplasm of motor neurons that were clearly distinct from the weak, homogeneous cytoplasmic distribution observed for wild-type FUS. In line with their ability to bind FUS, all interactors were detected in mutant FUS cytoplasmic accumulations in primary motor neurons and showed a change in their cytoplasmic distribution compared to wild-type FUS neurons (Figs. 4d, e, 5). To assess whether a similar colocalization occurs in ALS patients, we performed immunohistochemistry for FUS and each of the six interactors on spinal cord sections from two patients harboring a FUS mutation (R521C) (Supplementary Table S3). Expression of FMRP, UPF1, CAPRIN1 and HUD, but not of PABPC4 and DHX9, was observed in mutant FUS inclusions in spinal motor neurons (Fig. S4; Supplementary Table S4) (70–85 % of FUS inclusions displayed immunolabeling for interactors). Because we confirmed FMRP, UPF1, CAPRIN1 and HUD as shared interactors of FUS and TDP-43 (Fig. 4b, c), we next asked whether these proteins also localize to TDP-43-positive inclusions. Indeed, CAPRIN1, FMRP and HUD were detected in phospho-TDP-43 inclusions in spinal motor neurons in ALS patients (Fig. S5, S6; Supplementary Table S3) (80–100 % of pTDP-43 inclusions displayed immunolabeling for interactors). Thus, FMRP, Caprin1 and HuD bind endogenous FUS and TDP-43, and colocalize with mutant FUS and TDP-43 in aggregate-like structures in spinal motor neurons. Of these interactors, FMRP and HuD have been implicated recently in TDP-43-induced motor neuron degeneration [2, 14, 20, 34]. To examine whether these proteins are indeed components of shared pathways downstream of multiple ALS-associated proteins and to further characterize their role in ALS pathogenesis, we selected one of these candidates, FMRP, for further functional studies in relation to FUS mutations.

FMRP is encoded by the FMR1 gene. Expansions of more than 200 CGG repeats in the 5′UTR of FMR1 cause FXS and an intermediate repeat length of 50–200 repeats causes FXTAS, a late-onset neurodegenerative disorder. FMRP is a translational repressor with many different mRNA targets in neurons and plays an important role in synapse formation, including NMJ development and plasticity [95].

First, we used co-immunoprecipitation to confirm that Fus and Fmrp interact in motor neuron-like NSC-34 cells and primary cortical neurons (Fig. 6a, b). FUS and FMRP associate with many RNAs and other RNA-binding proteins. Thus, their interaction may be indirect and mediated by other proteins or RNAs. However, immunoprecipitation of recombinant FUS and FMRP proteins revealed a direct interaction in the absence of any other proteins, RNA molecules or cellular components (Fig. 6c). This indicates that FUS and FMRP can interact directly. To determine which region(s) of FUS mediate the interaction with FMRP, a series of FUS deletion mutants was generated and used in co-immunoprecipitation experiments. Full-length FUS and truncation mutants lacking the N-terminal QGSY region (Δ1–165) or the glycine-rich region (Δ165–276) bound FMRP. In contrast, FUS mutants lacking the C-terminal RGG-rich domain (Δ285–372) or the RNA recognition motifs (Δ360–501) did not interact with FMRP (Fig. 6d, e). Both FUS and FMRP are RNA-binding proteins and bind to and regulate many different RNAs [15, 32, 49]. We therefore investigated whether RNA modulates binding of FMRP to FUS. Interestingly, RNase treatment induced an increased interaction between Fus and Fmrp (441 % of RNase negative condition, p = 0.039) (Fig. 6f). To examine this effect of RNA in more detail, purified FUS and FMRP proteins were co-immunoprecipitated in the presence of an increasing amount of RNA. In line with the results from the RNase treatment, RNA application inhibited FUS-FMRP binding in a concentration-dependent manner (0.1 ng/μl = 70.4 %, p = 0.27; 0.5 ng/μl = 56.5 %, p = 0.08; 2.5 ng/μl = 23.3 %, p = 0.007; 12.5 ng/μl = 11.6 %, p = 0.003 and 62.5 ng/μl = 4.6 %, p = 0.002 of FUS-FMRP binding at 0 ng/μl RNA) (Fig. 6g). Together, these findings identify endogenous and direct binding of FUS and FMRP. Further, they suggest that this interaction can be modulated by RNA (Fig. 6h).

Mutant FUS expression causes defects in synaptic transmission at the NMJ in zebrafish [1]. Similarly, loss of the Drosophila homolog of FMRP, dFXR, leads to NMJ defects in Drosophila, possibly through upregulation of the MAP1B homolog Futsch [95]. Interestingly, recent work identifies both Futsch and FMRP as modifiers of TDP-43-dependent toxicity in Drosophila [13, 14]. Here, we investigated whether FMRP is also able to modulate mutant FUS-induced ALS phenotypes in vivo, using zebrafish embryos as a model. First, we further characterized NMJ defects induced by mutant FUS to extend previous electrophysiological studies revealing abnormal synaptic transmission. Injected embryos were immunostained at 72 hpf with anti-synaptic vesicles 2 (SV2) antibodies, to stain the presynaptic compartment of NMJs, and alpha-bungarotoxin (BTX), to detect the postsynaptic compartment (Fig. 7a, b). As expected, in NIC and FUS WT injected embryos, considerable overlap was observed between pre- and postsynaptic staining (NIC: R ² = 0.46, FUS WT: R ² = 0.46, p = 0.91) (Fig. 7b, c). However, injection of FUS R521C caused a significant decrease in the overlap between pre- and postsynaptic markers (FUS R521C: R ² = 0.39, p < 0.01), hinting at a loss of NMJ integrity. We then asked whether co-expression of FMRP and FUS R521C would restore abnormalities in NMJ morphology. Indeed, co-expression of FMRP with FUS R521C restored NMJ defects induced by FUS R521C (Fig. 7a–c). Overlap of pre- and postsynaptic markers was similar to NIC and FUS WT, indicating normal NMJ integrity (FUS R521C + FMR1: R ² = 0.46, p = 0.86) (Fig. 7c).

To investigate the functional consequence of diminished NMJ integrity, we assessed zebrafish locomotion at 72 hpf (Fig. 7d). Developing zebrafish normally respond to touch by swimming away from the touch stimulus (known as the touch-evoked escape response, TEER). A low number of NIC or FUS WT larvae showed abnormal TEER (NIC: 0.37 %, FUS WT: 0.55 %, p = 0.99). In contrast, FUS R521C injection led to a significantly higher number of larvae with abnormal TEER (FUS R521C: 3.68 %, p < 0.05). In line with the ability of exogenous FMRP to restore NMJ integrity, zebrafish larvae co-injected with FUS R521C and FMRP displayed normal TEER. These data show that FMRP overexpression rescues mutant FUS-induced ALS phenotypes.

Our previous work demonstrated that sequestration of SMN in mutant FUS aggregates causes an axonal depletion of SMN and defects in axonal morphology [28]. This study indicates that FMRP is also captured in mutant FUS aggregates. Therefore, we assessed whether axonal FMRP levels were decreased in primary motor neurons transfected with FUS R521C (as shown for Smn). However, quantification of the number of endogenous FMRP granules did not reveal differences between neurons transfected with GFP, FUS WT or FUS R521C (Fig. S7). As the subcellular distribution of FMRP appeared unaltered by mutant FUS, we next examined whether FMRP’s function as a regulator of local translation at synaptic sites was affected. Disruption of the repressive effect of FMRP on local translation could lead to increased translation of FMRP target mRNAs. To study this model, we used zebrafish embryos and focused on a well-characterized FMRP target, MAP1B. Consistent with our observation in transfected primary motor neurons, total FMRP protein levels were unchanged in zebrafish embryos expressing FUS R521C (Fig. 8a, b). Similarly, total expression of Map1b was unaffected (Fig. 8c). However, because FMRP regulates local translation at synaptic sites, we next examined Map1b protein levels in the synaptosomal compartment. Synaptosomes were isolated from injected zebrafish and analyzed by Western blot (Fig. S8). This analysis revealed a marked increase in synaptic Map1b levels following mutant FUS expression (Fig. 8d–f). Importantly, co-injection of FMRP RNA normalized enhanced Map1b levels induced by mutant FUS (Fig. 8d–f). Q-PCR analysis of total and synaptosomal fractions did not reveal changes in the mRNA expression of Map1b (Fig. 8g, h). Together, these data suggest that mutant FUS may impair translational repression of FMRP at the synapse, thereby increasing protein levels of FMRP targets such as MAP1B.

To show that interactomics analyses can lead to the identification of shared interactors of different ALS-associated proteins that are relevant for ALS pathogenesis, we focused on FMRP. FMRP is a translational repressor at synaptic sites [15] and FMRP overexpression rescues TDP-43 toxicity in Drosophila [14]. Our study identified FMRP as an interactor of TDP-43, ATXN2 and FUS. Therefore, to explore a more general role for FMRP in ALS pathogenesis we studied the potential link between FMRP and mutant FUS. We show endogenous, direct and RNA-sensitive interactions between FMRP and FUS, and found that FMRP localizes to mutant FUS-positive cytoplasmic aggregates in spinal motor neurons. FMRP has been previously shown to localize to mutant FUS granules in mouse fibroblast cells [91], but not in SH-SY5Y cells [74]. Further, exogenous expression of FMRP rescued defects in NMJ integrity and motor behavior in zebrafish embryos caused by mutant FUS.

How does FMRP modulate cellular defects caused by mutant FUS or TDP-43? Previous studies show that sequestration of SMN in cytosolic aggregates by mutant FUS causes its axonal depletion [28]. Other ALS-associated mutations have also been proposed to act by depleting proteins from their site of action in the cell [30, 58]. However, mutant FUS did not cause reduced FMRP levels in mouse motor neuron axons or synaptosomes derived from zebrafish embryos. Nevertheless, protein but not mRNA expression of a well-characterized FMRP target, Map1b, was increased at the synapse. MAP1B is a microtubule stabilizing protein regulating NMJ morphology [71]. Overexpression of the Drosophila MAP1B homolog Futsch induces NMJ abnormalities and impaired neurotransmission [95]. Intriguingly, overexpression of wild-type or mutant TDP-43 causes a small synaptic downregulation of Futsch in Drosophila, which can be reversed by FMRP overexpression. These data together with our results suggests that strict control of synaptic MAP1B expression is essential for maintaining motor neuron morphology and survival. Although it is unclear why FUS and TDP-43 have opposing effects on local translation of MAP1B, it is possible that both proteins affect MAP1B via distinct mechanisms. For example, although TDP-43 is present on polysomes [13], we only detected weak to no FUS expression in polysomal fractions (Fig. S10). But how do FUS mutations affect FMRP and its downstream mRNA targets? One possibility is that the increase in cytosolic FUS caused by ALS mutations leads to enhanced binding between FUS and FMRP, thereby preventing FMRP from binding to its target mRNAs on stalled polyribosomes leading to increased translation of MAP1B [15]. In line with this model, preliminary studies reveal enhanced binding between mutant FUS and FMRP and show increased expression of mutant as compared to wild-type FUS in FMRP granules in primary motor neurons (Fig. S11; A. M. B., M. K. and R. J. P., unpublished observations). Alternatively, FUS may alter the localization of FMRP at the synapse or by itself bind and control translation of MAP1B mRNA. Future studies are required to further examine the role and mechanism-of-action of FMRP in ALS pathogenesis.

In conclusion, we report a novel dataset comprised interactomes of six ALS-associated proteins that will provide a framework for future studies into the pathogenic mechanisms underlying ALS or other diseases linked to the proteins studied here [e.g., frontotemporal dementia (FTD; C9orf72, FUS, TDP-43) or SCA2 (ATXN2)]). We show that the interactors of multiple different ALS-associated proteins converge on a limited number of molecular complexes and related processes, namely RNA metabolism (ATXN2, FUS, hnRNPA1, hnRNPA2B1, MATR3 and TDP-43), autophagy and NF-κB signaling (OPTN, SQSTM1, TBK1, UBQLN2 and VCP) and mitochondrial functioning (C9orf72). Finally, the observation that FMRP, a shared interactor of ATXN2, FUS, and TDP-43, rescues FUS toxicity in an in vivo ALS model demonstrates that comparative interactomics analyses can aid in the identification and characterization of disease relevant proteins. This work showing that FMRP can reverse pathogenic effects caused by mutant FUS and TDP-43 identifies FMRP as an interesting therapeutic target in ALS.