FUS regulates RAN translation through modulating the G-quadruplex structure of GGGGCC repeat RNA in C9orf72-linked ALS/FTD

Abnormal expansions of GGGGCC repeat sequence in the noncoding region of the C9orf72 gene is the most common cause of familial amyotrophic lateral sclerosis and frontotemporal dementia (C9-ALS/FTD). The expanded repeat sequence is translated into dipeptide repeat proteins (DPRs) by noncanonical repeat-associated non-AUG (RAN) translation. Since DPRs play central roles in the pathogenesis of C9-ALS/FTD, we here investigate the regulatory mechanisms of RAN translation, focusing on the effects of RNA-binding proteins (RBPs) targeting GGGGCC repeat RNAs. Using C9-ALS/FTD model flies, we demonstrated that the ALS/FTD-linked RBP FUS suppresses RAN translation and neurodegeneration in an RNA-binding activity-dependent manner. Moreover, we found that FUS directly binds to and modulates the G-quadruplex structure of GGGGCC repeat RNA as an RNA chaperone, resulting in the suppression of RAN translation in vitro. These results reveal a previously unrecognized regulatory mechanism of RAN translation by G-quadruplex-targeting RBPs, providing therapeutic insights for C9-ALS/FTD and other repeat expansion diseases.

DPRs have been shown to exert toxic effects in multiple C9-ALS/FTD models, such as cultured cells, flies, and mice (Choi et al., 2019;Jovičić et al., 2015;May et al., 2014;Mizielinska et al., 2014;Rudich et al., 2017;Wen et al., 2014;Zhang et al., 2016;Zhang et al., 2018). Importantly, the toxicity of DPRs was confirmed in DPR-only flies, which express DPRs translated from non-G 4 C 2 repeat RNAs with alternative codons and show neurodegeneration, whereas RNA-only flies expressing G 4 C 2 repeat RNAs with stop codon interruptions, which eliminate DPRs production, did not show any obvious degenerative phenotypes (Mizielinska et al., 2014). In addition, increased DPR production, but not RNA foci, was reported to correlate with G 4 C 2 repeat-induced toxicity in a C9-ALS/FTD Drosophila model (Tran et al., 2015). Taken together, these studies have strongly suggested that DPRs play a central role in the pathogenesis of C9-ALS/FTD. Indeed, DPRs have been reported to disrupt various biological pathways, such as nucleocytoplasmic transport (Hutten et al., 2020;Jovičić et al., 2015;Zhang et al., 2016) and membraneless organelle dynamics (Kwon et al., 2014;Lee et al., 2016;Lin et al., 2016). Therefore, elucidating the regulatory mechanism of RAN translation is a significant challenge toward developing potential therapies for C9-ALS/FTD.
Since the discovery of RAN translation in 2011 (Zu et al., 2011), many studies to date have focused on its molecular mechanisms, that is, whether it has functional overlap with canonical AUG-dependent translation. Previous studies on C9-ALS/FTD using monocistronic reporters containing a G 4 C 2 repeat sequence revealed cap-dependent translation initiation from the upstream near-cognate CUG initiation codon, requiring the cap-binding eukaryotic translation factor 4F complex (Green et al., 2017;Tabet et al., 2018). On the other hand, studies using bicistronic reporters with a G 4 C 2 repeat sequence in the second cistron also produced DPRs by RAN translation in all reading frames, suggesting capindependent translation initiation within the G 4 C 2 repeat sequence (Cheng et al., 2018;Sonobe et al., 2018). This is reminiscent of internal ribosomal entry site translation initiation, which is another type of noncanonical cap-independent translation in which specific factors are directly recruited to the highly structured mRNA for initiation (Kwan and Thompson, 2019). While such initiation mechanisms of RAN translation have been explored to date, specific roles of the repeat sequence on RAN translation remain enigmatic. Considering a repeat length dependency of RAN translation (Mori et al., 2013c;Zu et al., 2011;Zu et al., 2013), the repeat sequence itself would also be essential for the initiation or elongation steps of RAN translation. Based on our previous findings of the protective role of TDP-43 on UGGAA repeat-induced toxicity in spinocerebellar ataxia type 31 (SCA31) models (Ishiguro et al., 2017), we hypothesized that RBPs specifically binding to repeat sequences of template RNA play a role in RAN translation.
Using Drosophila models of C9-ALS/FTD, we here demonstrate the regulatory roles of the ALS/ FTD-linked RBP FUS on RAN translation from G 4 C 2 repeat RNA, which lead to the significant modulation of neurodegeneration. We found that FUS suppresses RNA foci formation and DPR production, resulting in the suppression of repeat-induced degeneration. This suppressive effect on degeneration was abolished by mutations in the RNA-recognition motif (RRM) of FUS. In contrast, knockdown of endogenous caz, a Drosophila homologue of FUS, enhanced DPR aggregation and RNA foci formation, resulting in the enhancement of repeat-induced degeneration. Moreover, FUS was found to directly bind to G 4 C 2 repeat RNA and modify its G-quadruplex structure as an RNA chaperone, resulting in the suppression of RAN translation in vitro. In addition, other G-quadruplex-targeting RBPs also suppressed RAN translation and G 4 C 2 repeat-induced toxicity in our C9-ALS/FTD flies. These results strongly indicate that FUS regulates RAN translation and suppresses DPR toxicity through modulating the G-quadruplex structure of G 4 C 2 repeat RNA. Our findings shed light on the regulatory mechanisms of RAN translation by G-quadruplex-targeting RBPs and propose novel therapeutic strategies for repeat expansion diseases by regulating RAN translation.

Results
Screening for RBPs that suppress G 4 C 2 repeat-induced toxicity in C9-ALS/FTD flies We established Drosophila models of C9-ALS/FTD that express pathogenic length 42 or 89 G 4 C 2 repeats [(G 4 C 2 ) 42 , or (G 4 C 2 ) 89 flies, respectively] and confirmed that expanded G 4 C 2 repeat sequences induce eye degeneration and motor dysfunction accompanied with the formation of RNA foci and the production of three types of DPRs ( Figure 1-figure supplement 1), consistent with previous studies (Freibaum et al., 2015;Goodman et al., 2019;Mizielinska et al., 2014;Xu et al., 2013). We also established Drosophila expressing normal length 9 G 4 C 2 repeats as a control [(G 4 C 2 ) 9 flies] and found that they did not show eye degeneration, motor dysfunction, RNA foci formation, or DPR aggregation ( Figure 1-figure supplement 1). We selected 18 RBPs that have been reported to bind to G 4 C 2 repeat RNA (Mori et al., 2013a), as well as TDP-43, an ALS/FTD-linked RBP that does not bind to G 4 C 2 repeat RNA (Xu et al., 2013; Figure 1-source data 1), and examined their roles in neurodegeneration in our C9-ALS/FTD fly models. We found that coexpression of FUS, IGF2BP1, or hnRNPA2B1 strongly suppressed the eye degeneration in both flies expressing (G 4 C 2 ) 42 or 89 , which show decreased eye size and loss of pigmentation ( Figure 1A-D and Figure 1-source data 2). Coexpression of five RBPs, namely, hnRNPR, SAFB2, SF3B3, hnRNPA1, and hnRNPL, also partially suppressed the eye degeneration, whereas coexpression of the other six RBPs had no effect, and two RBPs enhanced the phenotypes ( Figure 1A-D and Figure 1-source data 2). In addition, coexpression of TDP-43 had no effect on the eye degeneration in (G 4 C 2 ) 42 flies and resulted in lethality in (G 4 C 2 ) 89 flies, likely due to the toxicity of TDP-43 expression itself ( Figure 1A and D and Figure 1source data 2). The variation in the effects of these G 4 C 2 repeat-binding RBPs on G 4 C 2 repeat-induced toxicity may be due to their different binding affinities to G 4 C 2 repeat RNA and the different toxicity of overexpressed RBPs themselves. We then analyzed the expression levels of G 4 C 2 repeat RNA in flies coexpressing (G 4 C 2 ) 89 and three RBPs that strongly suppressed eye degeneration. We found that coexpression of IGF2BP1 or hnRNPA2B1 significantly decreased G 4 C 2 repeat RNA levels, whereas they were not altered upon coexpression of FUS ( Figure 1E). Although the suppressive effects of Figure 1. Screening for RNA-binding proteins (RBPs) that suppress G 4 C 2 repeat-induced toxicity in C9-ALS/FTD flies. (A) Light microscopic images of the eyes in flies expressing both (G 4 C 2 ) 42 or 89 and the indicated RBPs using the GMR-Gal4 driver. Coexpression of FUS, IGF2BP1, or hnRNPA2B suppressed eye degeneration in both (G 4 C 2 ) 42 and (G 4 C 2 ) 89 flies, indicated by 'Suppression (strong).' Coexpression of hnRNPR, SAFB2, SF3B3, hnRNPA1, or hnRNPL suppressed eye degeneration in either (G 4 C 2 ) 42 or (G 4 C 2 ) 89 flies, indicated by 'Suppression (weak)' (see also Figure 1-source data 2).

Figure 1 continued on next page
IGF2BP1 and hnRNPA2B1 could simply be explained by the decreased levels of G 4 C 2 repeat RNA, the molecular mechanisms by which FUS suppresses G 4 C 2 repeat-induced toxicity remain to be clarified. The suppressive effects of FUS on G 4 C 2 repeat-induced toxicity were confirmed using multiple FUS fly lines, showing the significant suppression of decreased eye size and loss of pigmentation in (G 4 C 2 ) 42 or 89 flies coexpressing FUS (Figure 1-figure supplement 2). Therefore, we decided to further focus on FUS, which is another ALS/FTD-linked RBP, and investigated its mechanism of the suppression of G 4 C 2 repeat-induced toxicity.

FUS suppresses G 4 C 2 repeat-induced toxicity via its RNA-binding activity
We next investigated whether the suppressive effects of FUS on G 4 C 2 repeat-induced toxicity are mediated by its binding to G 4 C 2 repeat RNA, using flies expressing FUS with mutations in the RRM (FUS-RRMmut), which have been reported to eliminate its RNA-binding activity (Daigle et al., 2013). Western blot analysis confirmed that the FUS-RRMmut fly line expresses almost an equivalent level of the FUS proteins to the FUS fly line (Figure 2-figure supplement 1). We found that coexpression of FUS-RRMmut did not restore the eye degeneration in flies expressing (G 4 C 2 ) 89 , suggesting that the RNA-binding activity of FUS is essential for its suppressive effects on G 4 C 2 repeat-induced toxicity (Figure 2A -C). We also evaluated the reduced egg-to-adult viability of (G 4 C 2 ) 42 flies and confirmed that this phenotype was rescued by coexpression of FUS, but not by coexpression of FUS-RRMmut ( Figure 2D). Expression of G 4 C 2 repeat RNA in the nervous system of flies after eclosion using the elav-GeneSwitch driver induces motor dysfunction, and coexpression of FUS significantly alleviated this motor dysfunction ( Figure 2E), indicating that FUS suppresses the neuronal phenotypes of flies expressing G 4 C 2 repeat RNA. It is notable that the motor dysfunction caused by the expression of FUS alone was also alleviated by coexpression of (G 4 C 2 ) 42 ( Figure 2E), indicating that the G 4 C 2 repeat RNA conversely suppresses FUS toxicity. This result is consistent with our previous observations in SCA31 flies that UGGAA repeat RNA reduced the aggregation and toxicity of TDP-43 (Ishiguro et al., 2017). Moreover, recent studies demonstrated that RNA buffers the phase separation of TDP-43 and FUS, resulting in the suppression of their aggregation (Maharana et al., 2018;Mann et al., 2019). These findings hence suggest that balancing the crosstalk between repeat RNAs and RBPs neutralizes the toxicities of each other.
FUS suppresses RNA foci formation and RAN translation from G 4 C 2 repeat RNA We next analyzed the effects of FUS expression on RNA foci and DPR production in flies expressing G 4 C 2 repeat RNA. We performed RNA fluorescence in situ hybridization (FISH) of the salivary glands of fly larvae expressing (G 4 C 2 ) 89 and found that coexpression of FUS significantly decreased the number Scale bar: 100 μm. (B) Quantification of eye size in (G 4 C 2 ) 89 flies coexpressing the indicated RBPs (n = 5). (C, D) Quantification of eye pigmentation in (G 4 C 2 ) 89 flies (C) or (G 4 C 2 ) 42 flies (D) coexpressing the indicated RBPs (n = 5). (E) Expression levels of (G 4 C 2 ) 89 RNA in flies expressing both (G 4 C 2 ) 89 and the indicated RBPs using the GMR-Gal4 driver (five independent experiments, n = 25 flies per genotype). The (G 4 C 2 ) 89 (H) fly line expresses (G 4 C 2 ) 89 RNA at a high level (see also Figure 1-figure supplement 1). In (B-E), data are presented as the mean ± SEM; p<0.0001, as assessed by one-way ANOVA; n.s., not significant, * p<0.05, ** p<0.01, and *** p<0.001, as assessed by Tukey's post hoc analysis. The detailed statistical information is summarized in Figure  1-source data 3.
The online version of this article includes the following source data and figure supplement(s) for figure 1: Source data 1. RNA-binding proteins and their cDNA accession numbers screened in the genetic analyses in Figure 1.
Source data 2. Summary of the genetic analyses in Figure 1.
Source data 3. Statistical data related to Figure Figure 3A and B). We confirmed that the expression levels of G 4 C 2 repeat RNA in (G 4 C 2 ) 89 flies were not altered by coexpression of FUS or FUS-RRMmut ( Figure 3C). These results were in good agreement with our previous study on SCA31 showing the suppressive effects of FUS and other RBPs on RNA foci formation of UGGAA repeat RNA through altering RNA structures and preventing (D) Egg-to-adult viability in flies expressing both (G 4 C 2 ) 42 and either FUS or FUS-RRMmut using the GMR-Gal4 driver (>500 flies per genotype). (E) Climbing ability in flies expressing both (G 4 C 2 ) 42 and FUS using the elav-GeneSwitch driver (five independent experiments, n = 100 flies per each genotype). In (B-E), data are presented as the mean ± SEM. In (B, C), p<0.0001, as assessed by one-way ANOVA; n.s., not significant, and *** p<0.001, as assessed by Tukey's post hoc analysis. In (D), n.s., not significant and *** p<0.001, as assessed by Tukey's multiple-comparison test using wholly significant difference. In (E), n.s., not significant, * p<0.05, ** p<0.01, and *** p<0.001, as assessed by two-way repeated-measures ANOVA with Tukey's post hoc analysis. The detailed statistical information is summarized in Figure 2-source data 1.
The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1. Statistical data related to Figure      aggregation of misfolded repeat RNA as RNA chaperones (Ishiguro et al., 2017), raising the possibility that FUS has RNA-chaperoning activity also for G 4 C 2 repeat RNA. Immunohistochemistry of the eye imaginal discs of fly larvae expressing (G 4 C 2 ) 89 revealed that coexpression of FUS significantly decreased the number of DPR aggregates in (G 4 C 2 ) 89 flies, whereas coexpression of FUS-RRMmut did not ( Figure 3D and E). Quantitative analyses of poly(GP) by immunoassay also demonstrated that poly(GP) levels were greatly decreased in (G 4 C 2 ) 89 flies upon coexpression of FUS, but not FUS-RRMmut ( Figure 3F), indicating that FUS suppresses RAN translation from the G 4 C 2 repeat RNA to reduce DPR production. Considering that the 5′ upstream sequence of the G 4 C 2 repeat in the C9orf72 gene is reported to affect RAN translation activity (Green et al., 2017;Tabet et al., 2018), we used flies expressing the G 4 C 2 repeat sequence with the upstream intronic sequence of the C9orf72 gene, namely, LDS-(G 4 C 2 ) 44 GR-GFP (Goodman et al., 2019). Since this construct has a 3′-green fluorescent protein (GFP) tag in the GR reading frame downstream of the G 4 C 2 repeat sequence, the GR-GFP fusion protein is produced by RAN translation (Figure 3-figure supplement 1). We confirmed that coexpression of FUS significantly decreased the expression level of GR-GFP, whereas coexpression of FUS-RRMmut had no effect ( Figure 3G-I).
We further excluded the possibility that FUS directly interacts with DPRs, rather than with G 4 C 2 repeat RNA, to decrease DPR levels and exert its suppressive effects. Using DPR-only flies expressing DPRs translated from non-G 4 C 2 RNAs with alternative codons (Mizielinska et al., 2014), we confirmed that FUS did not suppress the eye degeneration in DPR-only flies expressing poly(GR), but rather enhanced their phenotypes, likely due to the additive effects of FUS toxicity (Figure 3-figure supplement 2). Together with the finding that FUS decreases not only DPR expression but also RNA foci formation ( Figure 3A and B), these results collectively indicate that FUS indeed interacts with G 4 C 2 repeat RNA and regulates RAN translation from G 4 C 2 repeat RNA in Drosophila models of C9-ALS/ FTD.
Reduction of endogenous caz expression enhances G 4 C 2 repeat-induced toxicity, RNA foci formation, and DPR aggregation To elucidate the physiological role of FUS on RAN translation, we also investigated the role of endogenous caz, a Drosophila homologue of FUS, on G 4 C 2 repeat-induced toxicity in flies expressing G 4 C 2 repeat RNAs. Coexpression of caz as well as FUS suppressed eye degeneration in flies expressing (G 4 C 2 ) 42 or 89 (Figure 4-figure supplement 1). These data suggest that caz is a functional homologue of FUS. In contrast, knockdown of caz by RNA interference or its hemizygous deletion modestly but significantly enhanced the eye degeneration in (G 4 C 2 ) 89 flies ( Figure 4A-D), indicating that reduced caz expression enhances G 4 C 2 repeat-induced toxicity. We next analyzed the effects of caz knockdown the GMR-Gal4 driver (10 independent experiments, n = 50 flies per each genotype). (D) Immunohistochemical analyses of dipeptide repeat proteins (DPRs) stained with anti-DPR antibodies in the eye imaginal discs of fly larvae expressing both (G 4 C 2 ) 89 and either FUS or FUS-RRMmut using two copies of the GMR-Gal4 driver (magenta: poly(GR); orange: poly(GA); green: poly(GP)). Arrowheads indicate cytoplasmic aggregates. Scale bars: 20 μm (low magnification) or 5 μm (high magnification). (E) Quantification of the number of DPR aggregates from the immunohistochemical analyses in (D) (n = 14 or 15 [GR], or 10 [GA or GP]). (F) Immunoassay to determine poly(GP) levels in flies expressing both (G 4 C 2 ) 89 and either FUS or FUS-RRMmut using the GMR-Gal4 driver (three independent experiments, n = 30 flies per each genotype). (G) Western blot analysis of the heads of adult flies expressing both LDS-(G 4 C 2 ) 44 GR-GFP and any of DsRed, FUS or FUS-RRMmut using the GMR-Gal4 driver, using either an anti-GFP (upper panel) or anti-GR antibody (middle panel). (H, I) Quantification of GR-GFP protein levels from the western blot analysis in (G) (nine independent experiments, n = 90 flies per each genotype). In (B, C, E, F, H, I), data are presented as the mean ± SEM. In (B, E, F), p<0.0001, as assessed by one-way ANOVA; n.s., not significant, * p<0.05, ** p<0.01, and *** p<0.001, as assessed by Tukey's post hoc analysis. In (C), p=0.452, as assessed by one-way ANOVA; n.s., not significant, as assessed by Tukey's post hoc analysis. In (H), p=0.0148, as assessed by one-way ANOVA; n.s., not significant and * p<0.05, as assessed by Tukey's post hoc analysis. In (I), p=0.0072, as assessed by one-way ANOVA; n.s., not significant and * p<0.05, as assessed by Tukey's post hoc analysis. The detailed statistical information is summarized in Figure 3-source data 1.
The online version of this article includes the following source data and figure supplement(s) for figure 3: Source data 1. Statistical data related to Figure 3B, C, E, F, H and I.
Source data 2. Source data related to Figure 3G.   on RNA foci formation and DPR production in flies expressing (G 4 C 2 ) 89 . FISH analysis of the salivary glands revealed that knockdown of caz significantly increased the number of nuclei containing RNA foci in (G 4 C 2 ) 89 flies ( Figure 4E and F). We also confirmed that the expression levels of G 4 C 2 repeat RNA in (G 4 C 2 ) 89 flies were not altered by the knockdown of caz ( Figure 4G). Immunohistochemical analysis showed that knockdown of caz significantly increased the number of DPR aggregates in (G 4 C 2 ) 89 flies ( Figure 4H and I). These results indicate that the reduction of caz expression enhances RNA foci formation and DPR aggregation, compatible with the results of FUS coexpression in flies expressing (G 4 C 2 ) 89 (Figure 3), and that FUS functions as an endogenous regulator of RAN translation.

FUS directly binds to and modulates the G-quadruplex structure of G 4 C 2 repeat RNA, resulting in the suppression of RAN translation in vitro
We next confirmed the direct interaction of FUS with G 4 C 2 repeat RNA by the filter binding assay. We found that His-tagged FUS binds to the (G 4 C 2 ) 4 RNA in a dose-dependent manner, but not to the control (AAAAAA) 4 RNA ( Figure 5A), and His-tagged FUS-RRMmut had almost no binding affinity to the (G 4 C 2 ) 4 RNA, consistent with a previous study (Mori et al., 2013a). We also confirmed the interaction of FUS with the G 4 C 2 repeat RNA in our C9-ALS/FTD flies by showing the colocalization of FUS with the RNA foci ( Figure 5-figure supplement 1), consistent with a recent study using C9-ALS/ FTD patient fibroblasts (Bajc Česnik et al., 2019). Since G 4 C 2 repeat RNA was reported to form both G-quadruplex and hairpin structures (Fratta et al., 2012;Haeusler et al., 2014;Reddy et al., 2013;Su et al., 2014), we next characterized the interactions of FUS with G 4 C 2 repeat RNA having different structures. G 4 C 2 repeat RNA is known to form G-quadruplex structures in the presence of K + , whereas they form hairpin structures in the presence of Na + (Su et al., 2014). Surface plasmon resonance (SPR) analyses demonstrated that FUS preferentially binds to (G 4 C 2 ) 4 RNA with the G-quadruplex structure in KCl buffer ( Table 1, dissociation constant (K D ) = 1.5 × 10 -8 M) and weakly to (G 4 C 2 ) 4 RNA with the hairpin structure in NaCl buffer ( Table 1, K D = 1.3 × 10 -7 μM). We also confirmed that FUS has poor binding affinity to (G 4 C 2 ) 4 RNA in LiCl buffer ( Table 1, K D = 1.4 × 10 -5 μM), which destabilizes the G-quadruplex structure (Hardin et al., 1992), and was an almost similar level to its binding affinity to the negative control (A 4 C 2 ) 4 RNA (not shown). These results suggest the preferential binding of FUS to G 4 C 2 repeat RNA with the G-quadruplex structure, which is consistent with a previous report showing preferential binding of FUS to G-quadruplex structured Sc1 and DNMT RNAs (Ozdilek et al., 2017).
Considering that higher-order structures, including G-quadruplex and hairpin structures, are reported to be involved in RAN translation (Mori et al., 2021;Simone et al., 2018;Wang et al., 2019;Zu et al., 2011), we next investigated the effects of FUS on the structure of G 4 C 2 repeat RNA. The circular dichroism (CD) spectrum of (G 4 C 2 ) 4 RNA in KCl buffer was found to exhibit a positive peak at approximately 260 nm and a negative peak at 240 nm ( Figure 5B, black line), consistent with previous reports (Fratta et al., 2012;Haeusler et al., 2014;Reddy et al., 2013;Su et al., 2014). Interestingly, upon the addition of FUS, these two peaks were notably shifted to longer wavelengths with substantial CD shown in (C) (n = 10). (E) Fluorescence in situ hybridization (FISH) analyses of G 4 C 2 repeat RNA in the salivary glands of fly larvae expressing (G 4 C 2 ) 89 using the GMR-Gal4 driver, with knockdown of caz (red: G 4 C 2 RNA; blue [DAPI]: nuclei). Arrowheads indicate RNA foci. Scale bar: 20 μm. (F) Quantification of the number of nuclei containing RNA foci from the FISH analyses in (E) (n = 10). (G) Expression levels of (G 4 C 2 ) 89 RNA in fly larvae expressing (G 4 C 2 ) 89 using the GMR-Gal4 driver, with knockdown of caz (four independent experiments, n = 20 flies per each genotype). (H) Immunohistochemical analyses of DPRs stained with anti-DPR antibodies in the eye imaginal discs of fly larvae expressing (G 4 C 2 ) 89 using two copies of the GMR-Gal4 driver, with the knockdown of caz. (magenta: poly(GR); orange: poly(GA); green: poly(GP)). Arrowheads indicate cytoplasmic aggregates. Scale bars: 20 μm (low magnification) or 5 μm (high magnification). (I) Quantification of the number of DPR aggregates from the immunohistochemical analyses in (H) (n = 10). In (B, D, F, G, I), data are presented as the mean ± SEM. In (B, D), p<0.0001, as assessed by one-way ANOVA; *** p<0.001, as assessed by Tukey's post hoc analysis. In (F, G, I), n.s., not significant, * p<0.05, ** p< 0.01, and *** p<0.001, as assessed by the unpaired t-test. The detailed statistical information is summarized in Figure 4-source data 1.
The online version of this article includes the following source data and figure supplement(s) for figure 4: Source data 1. Statistical data related to Figure 4B, D, F, G and I.   , indicating that this change in CD spectrum is attributed to structural changes in the (G 4 C 2 ) 4 RNA. We also observed CD spectrum changes to some extent in the (G 4 C 2 ) 4 RNA upon the addition of FUS in NaCl buffer, but not in LiCl buffer, confirming an interaction between FUS and hairpin-structured (G 4 C 2 ) 4 RNA as well ( Figure 5C and D). We further analyzed the interaction between FUS and G 4 C 2 repeat RNA by imino proton nuclear magnetic resonance (NMR). In KCl buffer, the NMR signals of the imino proton for the G-quadruplex structure of (G 4 C 2 ) 4 RNA were detected in the region around 10-12 ppm ( Figure 5-figure supplement 2D), consistent with previous studies (Fratta et al., 2012;Su et al., 2014). Upon the addition of FUS, the NMR intensities of (G 4 C 2 ) 4 RNA were decreased in an FUS concentration-dependent manner ( Figure 5-figure supplement 2D), further supporting the possibility that that FUS interacts with and modulates the G-quadruplex structure of (G 4 C 2 ) 4 RNA. Collectively, these results indicate that FUS directly binds to G 4 C 2 repeat RNA, preferentially to its G-quadruplex form and modulates its higher-order structures. These structural alterations of G 4 C 2 repeat RNAs by FUS did not require ATP or interactions with other proteins, suggesting its role as an RNA chaperone for G 4 C 2 repeat RNA (Rajkowitsch et al., 2007).
To further clarify the direct link between the binding of FUS to G 4 C 2 repeat RNA and its effects on RAN translation, we employed a cell-free in vitro translation assay using rabbit reticulocyte lysate. We designed a reporter construct containing the 80 G 4 C 2 repeat sequence with the 5′ upstream intronic sequence of the C9orf72 gene and the Myc tag sequence in the GA reading frame at the 3′ downstream ( Figure 5E). This upstream sequence contained multiple stop codons in each reading frame and lacked AUG initiation codons. We confirmed by western blotting that this reporter system indeed produces GA-Myc by RAN translation, consistent with previous studies (Green et al., 2017;Tabet et al., 2018). Notably, upon the addition to this translation system, FUS suppressed RAN translation efficiently, whereas FUS-RRMmut did not. FUS decreased the expression levels of GA-Myc at as low as 10 nM and nearly eliminated RAN translation activity at 100 nM. At 400 nM, FUS-RRMmut weakly suppressed the GA-Myc expression levels probably because of the residual RNA-binding activity ( Figure 5F and G). Taken together, these results indicate that FUS suppresses RAN translation in vitro through direct interactions with G 4 C 2 repeat RNA as an RNA chaperone. controls, respectively. (B-D) CD spectra of (G 4 C 2 ) 4 RNA incubated with or without FUS in the presence of 150 mM KCl (B), NaCl (C), or LiCl (D). The CD spectrum of FUS alone was subtracted from that of (G 4 C 2 ) 4 RNA incubated with FUS. The original data are shown in Figure 5-figure supplement 2B-2D. (E) Schema of the template RNA containing the (G 4 C 2 ) 80 sequence and 113 nucleotides of the 5′-flanking region of intron 1 of the human C9orf72 G 4 C 2 repeat sequence. A Myc tag in the GA frame was introduced downstream of the (G 4 C 2 ) 80 repeat sequence. (F) Western blot analysis of samples from in vitro translation using rabbit reticulocyte lysate in the presence or absence of increasing concentrations of FUS or FUS-RRMmut. The GA-Myc fusion protein was detected by western blotting using the anti-Myc antibody. (G) Quantification of the GA-Myc fusion protein in (F) (n = 3). In (G), data are presented as the mean ± SEM; * p<0.05, ** p<0.01, and *** p<0.001, as assessed by the unpaired t-test. The detailed statistical information is summarized in Figure 5-source data 1.
The online version of this article includes the following source data and figure supplement(s) for figure 5: Source data 1. Statistical data related to Figure 5G.
Source data 2. Source data related to Figure 5F.    Table 1. Association (k a ) and dissociation (k d ) rate and dissociation constants (K D ) between FUS and (G 4 C 2 ) 4 RNA in different buffers as assessed by surface plasmon resonance (SPR) analysis. Identification of G-quadruplex-targeting RBPs that suppress G 4 C 2 repeat-induced toxicity in C9-ALS/FTD flies Considering that FUS suppresses G 4 C 2 repeat-induced toxicity as an RNA chaperone through its preferential binding to the G-quadruplex structure of G 4 C 2 repeat RNA ( Figure 5 and Table 1), we hypothesized that other G-quadruplex-targeting RBPs might have similar suppressive effects on G 4 C 2 repeat-induced toxicity. To investigate this possibility, we selected six representative G-quadruplextargeting RBPs, all of which are known to bind to G 4 C 2 RNA as well (Cooper-Knock et al., 2014;Haeusler et al., 2014;Mori et al., 2013a;Xu et al., 2013; Figure 6-source data 1). Intriguingly, coexpression of EWSR1, DDX3X, DDX5, or DDX17 significantly suppressed eye degeneration in (G 4 C 2 ) 89 flies without altering G 4 C 2 RNA expression ( Figure 6A-D). As expected, these RBPs also decreased the number of poly(GA) aggregates in the eye imaginal discs ( Figure 6E and F). Their effects on G 4 C 2 repeat-induced toxicity, repeat RNA expression, and RAN translation were consistent with those of FUS. In support of our results, DDX3X was previously reported to suppress RAN translation and G 4 C 2 repeat-induced toxicity in cell culture in a helicase-activity-dependent manner (Cheng et al., 2019). On the other hand, coexpression of DHX9 or DHX36 suppressed eye degeneration by reducing G 4 C 2 repeat RNA levels ( Figure 6A-D). Since G-quadruplex-targeting RBPs have diverse biological functions, including transcription, RNA processing, translation, and RNA stabilization (Dumas et al., 2021), these different effects among G-quadruplex-targeting RBPs on G 4 C 2 repeat RNA expression might be attributed to their different roles in RNA metabolism. Thus, some G-quadruplex-targeting RBPs regulate RAN translation and G 4 C 2 repeat-induced toxicity by binding to and possibly by modulating the G-quadruplex structure of G 4 C 2 repeat RNA.

Discussion
In this study, we revealed a novel regulatory mechanism of RAN translation from expanded G 4 C 2 repeat RNA by the ALS/FTD-linked RBP FUS, which suppresses DPR production and neurodegeneration in C9-ALS/FTD Drosophila models (Figures 1-4). FUS directly binds to G 4 C 2 repeat RNA and modulates its G-quadruplex structure as evident by CD and NMR analyses ( Figure 5, Figure 5-figure supplement 2), and suppresses RNA foci formation in vivo ( Figure 3A and B), suggesting its functional role as an RNA chaperone. This is reminiscent of our recent study on SCA31, in which we demonstrated a novel role of the ALS/FTD-linked RBPs TDP-43, FUS, and hnRNPA2B1 as RNA chaperones binding to UGGAA repeat RNA and altering its structure, resulting in the suppression of its neurotoxicity through reducing RNA foci formation and repeat polypeptide translation (Ishiguro et al., 2017). Considering the similarities of the effects of FUS on G 4 C 2 repeat RNA and UGGAA repeat RNA, we conclude that FUS functions as an RNA chaperone also for G 4 C 2 repeat RNA to regulate its RAN translation. The suppressive effects of RBPs in several noncoding repeat expansion diseases by the amelioration of their sequestration into RNA foci have been reported. For example, in myotonic dystrophy type 1, MBNL1 was shown to be sequestered into CUG repeat RNA foci, and overexpression of MBNL1 in a mouse model was found to compensate its functional loss, resulting in the reversal of myotonia (Kanadia et al., 2006). Similarly, previous studies reported the suppressive effects of other RBPs on neurodegeneration, such as hnRNPA2B1 in fragile X ataxia/tremor syndrome (Sofola et al., 2007), and Pur-α and Zfp106 in C9-ALS/FTD (Xu et al., 2013;Celona et al., 2017). The suppressive effects of these RBPs have been thought to result from the supplementation against their loss-of-function due to their sequestration into RNA foci, although their effects on gain-of-toxic disease pathomechanisms, that is, RAN translation and repeat RNA expression, remain to be elucidated. In contrast, in this study we demonstrated that FUS suppresses neurodegeneration in C9-ALS/FTD by directly targeting G 4 C 2 repeat RNA and inhibiting RAN translation as an RNA chaperone. Similar suppressive effects of RBPs by targeting UGGAA repeat RNA in SCA31 as RNA chaperones have also been reported (Ishiguro et al., 2017). In addition, we also showed that the expression of IGF2BP1, hnRNPA2B1, DHX9, and DHX36 decreased G 4 C 2 repeat RNA expression and suppressed eye degeneration in our C9-ALS/ FTD Drosophila model (Figures 1 and 6), likely via the reduction of DPR levels. Similarly, we recently reported that hnRNPA3 reduces G 4 C 2 repeat RNA expression levels, leading to the suppression of neurodegeneration in C9-ALS/FTD fly models (Taminato et al., 2023). Interestingly, these RBPs have been reported to be involved in RNA decay pathways as components of the P-body or interactors with the RNA deadenylation machinery (Tran et al., 2004;Katahira et al., 2008;Geissler et al., 2016; Hubstenberger et al., 2017), possibly contributing to the reduced expression levels of G 4 C 2 repeat RNA. In myotonic dystrophy type 2 models, MBNL1 was also reported to retain CCUG repeat RNA in the nucleus, resulting in the suppression of RAN translation (Zu et al., 2017), implying various mechanisms of the effects of RBPs depending on the combination of RBPs and repeat RNA. Nevertheless, our findings highlighted the previously unrecognized roles of RBPs directly interacting with repeat RNA and modulating gain-of-toxic pathomechanisms, including RAN translation in noncoding repeat expansion diseases.
Several studies have indicated the importance of higher-order structures of repeat RNA in RAN translation. In SCA8 models, hairpin-forming CAG repeat RNA was shown to be RAN-translated to produce polyglutamine proteins, but switching the CAG repeats to non-hairpin-forming CAA repeats abolished the RAN translation (Zu et al., 2011). In C9-ALS/FTD, G 4 C 2 repeat RNA has been reported to form both hairpin and G-quadruplex structures (Fratta et al., 2012;Haeusler et al., 2014;Reddy et al., 2013;Su et al., 2014). Although the effect of each structure on RAN translation remains largely unknown, small molecules binding to the hairpin structure or the G-quadruplex structure were both reported to inhibit RAN translation from the G 4 C 2 repeat RNA, resulting in reduced DPR levels Mori et al., 2021). These findings are in accordance with our results showing that FUS modifies the G-quadruplex structure as well as the hairpin structure of G 4 C 2 repeat RNA as an RNA chaperone and reduces DPR production. We further found that G-quadruplex-targeting RNA helicases, including DDX3X, DDX5, and DDX17, which are known to bind to G 4 C 2 repeat RNA (Cooper-Knock et al., 2014;Haeusler et al., 2014;Mori et al., 2013a;Xu et al., 2013), also suppress RAN translation and G 4 C 2 repeat-induced toxicity without altering the expression levels of G 4 C 2 repeat RNA in our Drosophila models. These results suggest that not only ATP-independent RNA chaperones, but also ATP-dependent RNA helicases may regulate RAN translation through modifying the higher-order structures of template repeat RNA. Consistently, a previous study also reported that DDX3X inhibits RAN translation from G 4 C 2 repeat RNA in a helicase activity-dependent manner (Cheng et al., 2019). Knockdown of another RNA helicase, DHX36, has been reported to both promote (Cheng et al., 2019) and inhibit (Liu et al., 2021;Tseng et al., 2021) RAN translation, possibly due to the different effects on repeat RNA structures depending on the experimental conditions. Unfortunately, most of these studies reporting the effects of RBPs on RAN translation have limitations of the detailed structural analyses of repeat RNA. In this study, focusing on FUS, we performed a series of molecular structural analyses, in vitro translation assays, and in vivo genetic analyses to clarify the structure-function relationship of G 4 C 2 repeat RNA and provide compelling evidence for the modifying effects of FUS on repeat RNA structures leading to the suppression of RAN translation and repeat-induced toxicity in vivo.
FUS has an RRM domain for RNA binding and a low complexity (LC) domain involved in protein interactions, and exerts multifaceted functions, such as RNA transcription, RNA splicing, RNA transport, and formation of membraneless organelles, such as stress granules and nuclear paraspeckles via liquid-liquid phase separation (Lagier-Tourenne et al., 2010). Recent studies reported that argininerich DPRs, such as poly(GR) and poly(PR), interact with LC domain-containing RBPs, including FUS, and alter their liquid-liquid phase separation, resulting in the disruption of the dynamics and functions of membraneless organelles (Kwon et al., 2014;Lee et al., 2016;Lin et al., 2016). These findings raise the possibility that FUS may exert its suppressive effects by directly interacting with DPRs. However, we showed that FUS does not suppress eye degeneration in DPR-only flies (Figure 3-figure supplement 2), indicating that a direct interaction between FUS and DPRs is unlikely to be the mechanism of the suppression of DPR toxicity in our C9-ALS/FTD flies. This result supports our conclusion that FUS suppresses G 4 C 2 repeat-induced toxicity through direct binding to G 4 C 2 repeat RNA. GMR-Gal4 driver (orange: poly(GA)). Arrowheads indicate cytoplasmic aggregates. Scale bars: 20 μm (low magnification) or 5 μm (high magnification). (F) Quantification of the number of poly(GA) aggregates from the immunohistochemical analyses in (E) (n = 10). In (B, C, D, F), data are presented as the mean ± SEM; p<0.0001, as assessed by one-way ANOVA; n.s., not significant, * p<0.05 and *** p<0.001, as assessed by Tukey's post hoc analysis. The detailed statistical information is summarized in Figure 6-source data 2.
The online version of this article includes the following source data for figure 6: Source data 1. RNA-binding proteins and their cDNA accession numbers screened in the genetic analyses in Figure 6.
Source data 2. Statistical data related to Figure 6B-D, F.

Figure 6 continued
In summary, we here provided evidence that FUS modulates the structure of G 4 C 2 repeat RNA as an RNA chaperone and regulates RAN translation, resulting in the suppression of neurodegeneration in C9-ALS/FTD fly models. Recent advances in genome sequencing technology unveiled that such expansions of repeat sequences cause more than 50 monogenic human diseases (Malik et al., 2021) and are also associated with psychiatric diseases such as autism (Mitra et al., 2021;Trost et al., 2020). Thus, our findings contribute to the elucidation of the repeat-associated pathogenic mechanisms underlying not only C9-ALS/FTD, but also a broader range of neuromuscular and neuropsychiatric diseases than previously thought, and will advance the development of potential therapies for these diseases.

Flies
All fly stocks were cultured and crossed at 23°C or 25°C in standard cornmeal-yeast-glucose medium. Male adult flies were used for the climbing assay and GeneSwitch experiments. 3-to 5-day-old female adult flies were used for the evaluation of eye phenotype using a stereoscopic microscope model SZX10 (Olympus). Female third-instar larvae were used for quantitative real-time polymerase chain reaction (PCR), RNA FISH, and immunohistochemistry experiments. The transgenic fly line bearing the GMR-Gal4 transgene has been described previously (Yamaguchi et al., 1999).  (Frickenhaus et al., 2015), Dr. Nancy Bonini (Goodman et al., 2019), and Dr. Brian McCabe (Wang et al., 2011), respectively. Other transgenic fly lines were generated in this study. Full genotypes of the fly lines used in all figures and their cultured temperatures are described in Supplementary file 1.

Generation of constructs and transgenic flies
Artificially synthesized (G 4 C 2 ) 50 sequences flanked at the 5′ end with an EagI recognition site and at the 3′ end with a PspOMI recognition site were subcloned into T-vector pMD20 (Takara Bio). To generate a longer repeat size, the pMD20-(G 4 C 2 ) 50 vector was digested with EagI and PspOMI, followed by ligation into the pMD20-(G 4 C 2 ) 50 vector linearized by digestion with EagI. This vector was digested with EcoRI and HindIII, and subcloned into the pcDNA3.1/myc-His(−)A vector (Thermo Fisher Scientific). We accidentally obtained the pcDNA3.1/myc-His(−)A-(G 4 C 2 ) 9 vector at this step. The pcDNA3.1/myc-His(−)A-(G 4 C 2 ) n vector was digested with EcoRI and XbaI, and subcloned into the Drosophila pUAST vector. These constructs have no start codon sequence (ATG) upstream of the G 4 C 2 repeat sequence (Figure 1-figure supplement 1A). These pUAST-(G 4 C 2 ) n vectors were amplified with a recombinase-mutated SURE2 Escherichia coli strain (Agilent Technologies) at 28°C for 72 hr to prevent repeat length contraction. The number of G 4 C 2 repeats in the pUAST-(G 4 C 2 ) 9 or 50 vectors was determined by sequencing. To determine the number of G 4 C 2 repeats in the pUAST-(G 4 C 2 ) 89 vector, transposable element insertional mutagenesis using EZ-Tn5<KAN-2>Insertion Kit (Epicentre) and sequencing were performed. The entire sequence of the insert in the pUAST vector is shown in Figure 1-figure supplement 1-source data 1.
To generate pUAST-FUS or pUAST-TDP-43 vectors, the Gateway Vector Conversion System (Thermo Fisher Scientific) was used. The human FUS or human TARDBP cDNA was subcloned into the pENTR/D-TOPO vector (Thermo Fisher Scientific). To generate the Gateway destination vector pUAST-DEST, we inserted the Gateway cassette A sequence (Thermo Fisher Scientific) into the pUAST vector. The pUAST-FUS or pUAST-TDP-43 vectors were generated using Gateway recombination reactions (Thermo Fisher Scientific). The FUS RRM mutant construct (pUAST-FUS-RRMmut), in which leucine residues at positions 305, 341, 359, and 368 in the FUS protein were substituted to phenylalanine, was generated by PCR and the In-Fusion Cloning system (Takara Bio). To generate the other pUASTattB-RBP vectors, each cDNA encoding the RBP shown in Figure 1-source data 1 and Figure 6-source data 1 was subcloned into the pUASTattB vector (VectorBuilder). To establish transgenic flies harboring UAS-(G 4 C 2 ) n , UAS-FUS, UAS-FUS line 2, UAS-FUS-RRMmut, and UAS-TDP-43, the pUAST-(G 4 C 2 ) n , pUAST-FUS, pUAST-FUS line 2, pUAST-FUS-RRMmut, and pUAST-TDP-43 vectors, respectively, were injected into fly embryos of the w 1118 strain. To establish transgenic flies harboring the other UAS-RBP constructs including UAS-FUS line 3, pUASTattB-RBP vectors were injected into fly embryos of the attP40 strain. By employing site-specific transgenesis using the pUASTattB vector, each transgene was inserted into the same locus of the genome and was expected to be expressed at the equivalent levels. These transgenic flies were established using standard methods at BestGene Inc.

Imaging and quantification of fly eyes
Light microscopic images of the eyes of 3-to 5-day-old female flies were taken using a stereoscopic microscope model SZX10 (Olympus) with a CCD camera DP21 (Olympus). Images shown are representative eye phenotypes of the fly crosses. Crosses were performed three times to validate the specific phenotypes. Eye size and pigmentation were quantified as previously reported (Saitoh et al., 2015). 5 or 10 eyes per genotype were analyzed. Data were normalized by setting the values of samples from flies expressing one copy of EGFP ( Figure 4D), those expressing two copies of EGFP ( Figures 1B-D,  2B, C, and 6B and C, Figure 1-figure supplement 2B-D, and Figure 4-figure supplement 1B-D), or those expressing both EGFP and GFP-IR ( Figure 4B), as 100.

Egg-to-adult viability of flies
Mated female flies were placed on grape juice agar with yeast paste for 24 hr. Eggs were collected from the surface of the grape juice agar, and the number of eggs was counted and placed on new standard fly food. After eclosion, the number of adult flies was counted. Egg-to-adult viability was calculated by dividing the number of adult flies by the number of eggs. More than 500 eggs per genotype were used. Data were normalized by setting the values of samples from flies expressing two copies of EGFP as 100 ( Figure 2D).

Climbing assay
Twenty male flies were gently introduced into a glass vial. After a 5 min adaptation period, the bottom of the vial was gently tapped and the height the flies reached in 10 s was recorded using a digital video camera, and scored as follows: 0 (lower than 2 cm), 1 (from 2 to 3.9 cm), 2 (from 4 to 5.9 cm), 3 (from 6 to 7.9 cm), 4 (from 8 to 9.9 cm), and 5 (higher than 10 cm). Five trials were performed in each experiment at intervals of 20 s. The assay was performed between 8:00 and 10:00. Climbing scores were calculated as an average of five trials.

GeneSwitch experiments
Flies were crossed in the absence of RU486 (mifepristone) on standard fly food. 1-day-old adult male flies were transferred to Formula 4-24 Instant Drosophila medium (Wako) with RU486 (100 µg/mL) for the indicated periods. Every 2 or 3 d, flies were transferred to new medium with RU486. Climbing assays were performed at 0, 7, and 14 d after the start of RU486 treatment ( Figure 2E).
After RNA FISH, samples were scanned using a Zeiss LSM710 or LSM880 confocal laser-scanning microscope along the z-axis direction. One z-stack image was taken per salivary gland using ZEN imaging software (Zeiss). RNA foci-positive nuclei in more than 30 cells per salivary gland were counted, and the percentage of nuclei containing RNA foci in the salivary gland was calculated. Ten salivary glands were analyzed for each genotype.
The number of DPR aggregates in the eye discs was quantitatively measured using ZEN imaging software (Zeiss) and ImageJ (Schneider et al., 2012), as follows: (1) selection of photoreceptor neurons within the 13 developing ommatidia in rows 2 and 3 at the posterior end of the eye discs (Saitoh et al., 2015) by DAPI or Hoechst 33342 staining because these ommatidia are at similar stages of development and can be easily identified; and (2) counting of the number of DPR aggregates with a diameter of larger than 2 µm in the cytoplasm. 10-15 eye discs were analyzed for each genotype.

Measurement of poly(GP) protein levels
The heads of 5-day-old female flies expressing both (G 4 C 2 ) 89 (H) and either EGFP, FUS, or FUS-RRMmut using the GMR-Gal4 driver were collected and stored at -80°C. Samples were prepared using a previously reported method (Tran et al., 2015). Poly(GP) levels were measured by a sandwich immunoassay that uses Meso Scale Discovery electrochemiluminescence detection technology, as described previously (Su et al., 2014). Data were normalized by setting the values of samples from flies expressing (G 4 C 2 ) 89 (H) and EGFP ( Figure 3F) as 100.

Western blotting
To assess the expression levels of FUS and FUS-RRMmut (Figure 2-figure supplement 1), or GR-GFP ( Figure 3G-I), 10 heads of 5-day-old female flies expressing FUS or FUS-RRMmut, or both LDS-(G-4 C 2 ) 44 GR-GFP and either FUS or FUS-RRMmut using the GMR-Gal4 driver were homogenized in 100 µL of sodium dodecyl sulfate (SDS) sample buffer using a pestle, boiled for 5 min, and centrifuged at 10,000 × g for 3 min at 25°C. 5 µL of each supernatant were run on a 5-20% gradient polyacrylamide gel (Wako) and then transferred onto an Immun-Blot polyvinylidene fluoride membrane (Bio-Rad). Membranes were blocked with 5% skim milk in PBS containing 0.1% Tween-20 (PBST) or PVDF Blocking Reagent for Can Get Signal (TOYOBO) for 2 hr at RT, and then incubated overnight at 4°C with a rabbit polyclonal anti-FUS antibody (A300-302A, Bethyl Laboratories), a rat monoclonal anti-poly(GR) antibody (clone 5A2, MABN778, Millipore), a Living Colors EGFP mouse monoclonal antibody (632569, Clontech), or a mouse monoclonal anti-actin antibody (clone AC-40, A4700, Sigma-Aldrich) at 1:1000 dilution as the primary antibody. After washing three times with PBST, membranes were incubated for 2 hr at RT with either HRP-conjugated anti-rat, anti-rabbit, or anti-mouse antibody (31470, 31460, or 31430, respectively, Invitrogen) at 1:5000 dilution as the secondary antibody, washed three times with PBST, treated with SuperSignal West Dura chemiluminescent substrate (Thermo Fisher Scientific), and imaged using the LuminoGraphII imaging system (ATTO). Data were normalized by setting the average values of samples from flies expressing FUS (Figure 2-figure supplement 1B) or those expressing both LDS-(G 4 C 2 ) 44 GR-GFP and DsRed as 100 ( Figure 3H and I).

Surface plasmon resonance analyses
The binding of FUS to (G 4 C 2 ) 4 RNA was analyzed using a Biacore T200 instrument (GE Healthcare).
(G 4 C 2 ) 4 RNAs biotinylated at the 5′ end in 10 mM HEPES pH 6.8 and 500 mM MCl (M=K, Na, or Li) was injected over the streptavidin-coated surface of a sensor chip SA (GE Healthcare). The amount of immobilized RNA was as follows: 240 resonance unit (RU) in KCl, 363 RU in NaCl, or 319 RU in LiCl buffer condition. Binding experiments were performed using the single-cycle kinetics method. The running buffer used was 20 mM HEPES (pH 6.8), 1 mM MgCl 2 , 0.05% Tween-20, and 150 mM KCl, NaCl, or LiCl. FUS was diluted in the running buffer and injected sequentially over the RNA-immobilized sensor surface in increasing concentrations (0.016, 0.031, 0.063, 0.13, or 0.25 µM). Sensorgrams were obtained at 25°C, 30 µL/min flow rate, 60 s of contact time, and 120 s of dissociation time.

Circular dichroism spectroscopy
CD spectra were measured at 25°C using a spectropolarimeter model J-820 (JASCO). (G 4 C 2 ) 4 RNA was synthesized by GeneDesign Inc and dissolved in 20 mM HEPES (pH 6.8), 18.75 mM NaCl, 10 mM MgCl 2 , 0.625% glycerol, 0.625 mM β-cyclodextrin, and 0.0625 mM EDTA with 150 mM KCl, NaCl, or LiCl. RNA samples containing 150 mM KCl were first heated at 95°C for 5 min and then cooled to RT to form the G-quadruplex structure. The other samples were not heated. FUS (1 μM) was added to the RNA sample (4 μM) and mixed before recording the spectrum. CD spectra were recorded at a speed of 50 nm min -1 and a resolution of 1 nm, and 10 scans were averaged.

Nuclear magnetic resonance spectroscopy
All one-dimensional 1 H NMR spectral data were recorded using AVANCE III 800 MHz NMR spectrometers equipped with a TXI cryogenic probe (Bruker BioSpin) at 25°C. (G 4 C 2 ) 4 RNA dissolved in 20 mM HEPES (pH 6.8), 150 mM KCl, 18.75 mM NaCl, 10 mM MgCl 2 , 0.625% glycerol, 0.625 mM β-cyclodextrin, and 0.0625 mM EDTA was first heated at 95°C for 5 min and then cooled to room temperature to form the G-quadruplex structure. The RNA (10 μM) was mixed with FUS at molar ratios (RNA:FUS) of 1:0, 1:0.2, 1:0.4, and 0:1. The samples were then prepared at a final concentration of 10% D 2 O before recording their spectra. 1 H NMR data were acquired using simple single 90° hard-pulse excitations following solvent signal suppression with a jump-and-return pulse scheme. Free induction decay data (1600 points in total) were collected by repeating the scans (2600 times) with an interscan delay of 2.5 s. All NMR data were processed using Topspin 3.6 software (Bruker BioSpin).

RNA synthesis for in vitro translation
For preparation of the C9-RAN reporter plasmid, the pEF6-C9orf72 intron1-(G 4 C 2 ) 80 vector was digested with HindIII and NotI to obtain the fragment C9orf72 intron1-(G 4 C 2 ) 80 and subcloned into the pcDNA5/FRT vector (Thermo Fisher Scientific). To add the T7 promoter upstream of the C9orf72 intron 1 sequence in this pcDNA5/FRT-C9orf72 intron1-(G 4 C 2 ) 80 vector, a forward primer including T7 promoter sequences with the 5′-terminal region of C9orf72 intron 1 flanked at the 5′ end with an HindIII recognition site, and a reverse primer recognizing the 3′-terminal region of C9orf72 intron 1 sequences including a BssHII recognition site were designed, and used to amplify a fragment containing C9orf72 intron 1 with a T7 promoter by PCR. Then, this fragment was subcloned into the pcDNA5/FRT-C9orf72 intron1-(G 4 C 2 ) 80 vector digested by HindIII and BssHII. In addition to the T7 promoter, the Myc tag in the GA frame downstream of (G 4 C 2 ) 80 was introduced into this vector.
The reporter plasmids were linearized with XbaI. Linearized DNA was in vitro transcribed using mMESSAGE mMACHINE T7 Transcription Kit (Invitrogen) according to the manufacturer's instructions. T7 reactions were carried out at 37°C for 2 hr, treated with TURBO DNaseI for 15 min at 37°C to remove the DNA template, and then polyadenylated with E. coli Poly-A Polymerase (NEB) for 1 hr at 37°C. Synthesized mRNAs were purified by LiCl precipitation. The size and quality of the synthesized mRNAs were verified on a denaturing RNA gel.
In vitro translation assay mRNAs of C9orf72 intron1-(G 4 C 2 ) 80 with a Myc tag in the GA frame were in vitro translated with Flexi Rabbit Reticulocyte Lysate System (Promega) according to the manufacturer's instructions. Translation reactions were performed with 10 ng/μL mRNA and contained 30% rabbit reticulocyte lysate, 10 μM amino-acid mix minus methionine, 10 μM amino-acid mix minus leucine, 0.5 mM MgOAc, 100 mM KCl, and 0.8 U/μL Murine RNAse Inhibitor (NEB). FUS or FUS-RRMmut at each concentration (10, 100, 200, 400, and 1000 nM) was preincubated with mRNA for 10 min to facilitate the interaction between FUS protein and G 4 C 2 repeat RNA, and added for translation in the lysate. Samples were incubated at 30°C for 90 min before termination by incubation on ice. 10 μL of samples were analyzed by 13% SDS-polyacrylamide gel electrophoresis and western blotting using a mouse monoclonal anti-c-Myc antibody (clone 9E10, Wako) as the primary antibody.  Figure 2E, and the unpaired t-test was used for Figures 4F, G, I and 5G, and Figure 2-figure supplement 1B. Differences in means were considered statistically significant at p<0.05. All statistical analyses were performed using GraphPad Prism version 8.3.4 (GraphPad Software, LLC).

Quantification and statistical analysis
As the sample sizes used in this study were similar to previous publications (Freibaum et al., 2015;Goodman et al., 2019;Mizielinska et al., 2014;Xu et al., 2013), statistical analyses were performed afterward without interim data analysis. Data were not excluded and were collected and processed randomly. Sample collection and analyses for the measurement of poly(GP) protein levels were performed in a double-blind manner. Data collection and analyses for other experiments were not performed in a blind manner regarding the conditions of the experiments.
, and UAS-FLAG-caz and UAS-FLAG-FUS flies, respectively. We acknowledge Bloomington Drosophila Stock Center and Vienna Drosophila Resource Center for providing various fly stocks. We thank Drs H Imataka and K Machida (University of Hyogo) for kindly providing human tRNAs for in vitro translation. We also thank the members of our laboratory for helpful discussions, K Yamamoto, E Wakisaka, T Yamashita, and A Sugai for their technical assistance, Dr. H Akiko Popiel for critical reading and English editing of the manuscript, and the Center for Medical Research and Education, Graduate School of Medicine, Osaka University, for supplying experimental equipment.

Additional files
Supplementary files • Supplementary file 1. Full genotypes of the fly lines and their cultured temperatures.

• MDAR checklist
Data availability All data generated or analysed during this study are included in the manuscript and supporting files.