A C. elegans model of C9orf72-associated ALS/FTD uncovers a conserved role for eIF2D in RAN translation

A hexanucleotide repeat expansion GGGGCC in the non-coding region of C9orf72 is the most common cause of inherited amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). Toxic dipeptide repeats (DPRs) are synthesized from GGGGCC via repeat-associated non-AUG (RAN) translation. Here, we develop C. elegans models that express, either ubiquitously or exclusively in neurons, 75 GGGGCC repeats flanked by intronic C9orf72 sequence. The worms generate DPRs (poly-glycine-alanine [poly-GA], poly-glycine-proline [poly-GP]) and poly-glycine-arginine [poly-GR]), display neurodegeneration, and exhibit locomotor and lifespan defects. Mutation of a non-canonical translation-initiating codon (CUG) upstream of the repeats selectively reduces poly-GA steady-state levels and ameliorates disease, suggesting poly-GA is pathogenic. Importantly, loss-of-function mutations in the eukaryotic translation initiation factor 2D (eif-2D/eIF2D) reduce poly-GA and poly-GP levels, and increase lifespan in both C. elegans models. Our in vitro studies in mammalian cells yield similar results. Here, we show a conserved role for eif-2D/eIF2D in DPR expression.

T he GGGGCC (G4C2) hexanucleotide repeat expansion in the non-coding region of C9orf72 is the most common monogenic cause of inherited amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) 1,2 , and also causes up to 10% of what appears to be sporadic ALS. The G4C2 repeat expands in patients to hundreds or thousands of copies that vary in number in different cells of the same individual. This genetic insult is thought to cause ALS/FTD via three non-mutually exclusive mechanisms: (1) loss of function due to decreased expression of C9ORF72 protein, (2) RNA toxicity from bidirectionally transcribed sense (GGGGCC) and antisense (CCCCGG) transcripts, and (3) proteotoxicity from dipeptide repeat (DPR) proteins produced from the expanded nucleotide repeats 3 .
This study focuses on the molecular mechanisms underlying DPR toxicity. Strong evidence suggests that DPRs are toxic in both cell culture and animal models of disease 4,5 . Despite the presence of the expanded G4C2 repeat in the non-coding region of the RNA and the absence of an AUG initiating codon, DPRs are translated in all three reading frames from both sense and antisense transcripts through a process called repeat-associated non-AUG (RAN) translation 4 . Polyglycine-alanine (poly-GA), poly-glycine-proline (poly-GP), and poly-glycine-arginine (poly-GR) are produced from sense transcripts, whereas poly-proline-arginine (poly-PR), poly-prolinealanine (poly-PA), and poly-GP are generated from antisense transcripts [6][7][8] . DPRs are present in neural cells of patients with C9orf72-associated ALS and FTD, indicating that RAN translation occurs in vivo 6,7,9,10 . A number of potential causes for DPR toxicity have been proposed, including protein aggregation, impaired nucleocytoplasmic transport, and proteasome dysfunction 5 ; however, the relative contribution of each DPR to disease pathogenesis remains unclear.
Recent in vitro and in vivo studies have identified noncanonical translation initiation factors as critical mediators of DPR production. For example, a recent study showed that the ribosomal protein eS25 (RPS25), which is required for efficient translation initiation at the internal ribosome entry site (IRES) of a number of viral and cellular RNAs 11,12 , is also required for RAN translation of poly-GP from G4C2 repeats in yeast, fly, and induced pluripotent human stem cell models 13 . In addition, the eukaryotic translation initiation factor eIF3F, which is thought to control translation of hepatitis C viral (HCV) RNA by binding to the viral IRES 14 , regulates poly-GP production from G4C2 repeats in HEK293 cells 15 . Moreover, a fly study found that eIF4B and eIF4H are important for RAN translation of poly-GR from G4C2 transcripts 16 . Lastly, knockdown of eIF2A, an unconventional translation initiation factor used by cells under stress conditions [17][18][19] , partially affected poly-GA production in HEK293 cells and neural cells of the chick embryo 20 . Collectively, these and other studies 21,22 suggest that multiple factors are involved in RAN translation of DPRs from G4C2 transcripts; yet, their identity and molecular mechanisms of action remain elusive.
To identify factors involved in RAN translation of DPRs in vivo, we developed two C. elegans models for C9orf72-associated ALS/FTD. C. elegans provides a powerful system for the study of molecular and cellular mechanisms underlying neurodegenerative disorders due to: (1) its short lifespan (~3-4 weeks), (2) its compact nervous system (302 neurons in total), (3) the ability to study neuronal morphology and function with singlecell resolution, (4) the fact that most C. elegans genes have human orthologs, and (5) the ease of genomic engineering and transgenesis, which enables the rapid generation of worms that harbor human gene mutations, permitting in vivo modeling of neurodegenerative disorders.
In this study, we initially developed C. elegans transgenic animals that carry, under the control of a ubiquitous promoter, 75 copies of the G4C2 repeat flanked by intronic C9orf72 sequences. Compared to controls, these animals produce DPRs (poly-GA, poly-GP, poly-GR), and display evidence of neurodegeneration, as well as locomotor and lifespan defects. A second set of transgenic worms expressing the 75 G4C2 repeats and the flanking intronic sequences exclusively in neurons displayed similar phenotypes, suggesting that DPR production in this cell type is sufficient to cause disease phenotypes. These C. elegans models provide an opportunity to study in vivo the molecular mechanisms underlying DPR production. Through a candidate approach, we identified a role for the non-canonical translation initiation factor eif-2D (ortholog of human eIF2D) in RAN translation of G4C2 repeats. Genetic removal of eif-2D did not affect the formation of G4C2 RNA foci, but prominently decreased poly-GA steady-state levels, mildly affected poly-GP levels, and improved locomotor activity and lifespan in both C. elegans models. Supporting the phylogenetic conservation of our C. elegans findings, in vitro knockdown of eIF2D in mouse and human cells led to a similar decrease in poly-GA steady-state levels. Together, these observations suggest eIF2D may play an important role in RAN translation from G4C2 repeats, a finding with biomedical implications for C9orf72-associated ALS and FTD.

Results
Generation of a C. elegans model for C9orf72-mediated ALS/ FTD. The C. elegans genome contains a C9orf72 ortholog (alfa-1), which is involved in lysosomal homeostasis, cellular metabolism, and stress-induced neurodegeneration [23][24][25] , however, alfa-1 does not contain G4C2 repeats. Therefore, to study the molecular mechanisms underlying DPR production from G4C2 repeats, we generated worms that carry a transgene encoding 75 copies of the G4C2 sequence under the control of a ubiquitous (snb-1) promoter 26 (Fig. 1a). The widespread expression of this promoter was independently validated ( Supplementary Fig. 1). These 75 G4C2 copies are flanked with intronic sequences normally found upstream (113 nucleotides) and downstream (99 nucleotides) of the G4C2 repeats in the human C9orf72 gene (Fig. 1a). To monitor the expression of poly-GA, the most amyloidogenic DPR 27,28 , a nanoluciferase (nLuc) reporter was placed in the poly-GA reading frame. Hereafter, we will refer to these transgenic animals as C9 ubi (Fig. 1a). In parallel, we generated four control C. elegans strains: (a) UAG ubi worms carry an identical sequence to the C9 ubi animals, but the upstream translation initiation codon CUG, which is required for translation of poly-GA in vitro 20,[29][30][31] , is mutated to UAG (Fig. 1a), (b) ΔC9 ubi worms lack the G4C2 repeats and the intronic sequences flanking the G4C2 repeats ( Fig. 1a), (c) ΔG4C2 repeats ubi worms lack the G4C2 repeats, but maintain the intronic sequences ( Supplementary Fig. 2a), and (d) the C9 ΔnLuc ubi worms carry an identical sequence to the C9 ubi animals, but lack nLuc ( Supplementary Fig. 2a).
Several independent transgenic lines for C9 ubi , UAG ubi , ΔC9 ubi , ΔG4C2 repeats ubi , and C9 ΔnLuc ubi worms were generated and tested as follows. First, we performed quantitative PCR with reverse transcription (RT-qPCR) and found that mRNA corresponding to intronic sequences that flank the G4C2 repeats is only detected in C9 ubi , UAG ubi , ΔG4C2 repeats ubi , and C9 ΔnLuc ubi animals, whereas nLuc mRNA is only detected in C9 ubi , UAG ubi , and ΔG4C2 repeats ubi animals ( Fig. 1b and Supplementary Fig. 2b, c). Second, we sought to evaluate whether G4C2 transcripts form RNA foci in our transgenic animals, similar to observations made in other C9orf72 models 32,33 . RNA fluorescent in situ hybridization (FISH) revealed G4C2 RNA foci in head and tail neurons, ventral nerve cord motor neurons, as well as non-neuronal cells in C9 ubi and UAG ubi , but not in control (ΔC9 ubi ) animals ( Fig. 1c and Supplementary Fig. 3a).
Next, we tested whether DPRs (poly-GA, poly-GP, poly-GR) from the G4C2 sense transcript are produced in the C9 ubi animals. We focused on these because the 75 G4C2 repeats are downstream of the snb-1 promoter, suggesting the sense transcript may be the one primarily transcribed and translated. Western blotting and luciferase assays showed that poly-GA was robustly expressed in C9 ubi worms when compared to ΔC9 ubi , UAG ubi , and ΔG4C2 repeats ubi control animals ( Fig. 1d-f and Supplementary Fig. 3b). Corroborating these observations, immunofluorescence staining revealed the presence of poly-GA aggregates in head neurons, ventral nerve cord motor neurons, and non-neuronal cells of C9 ubi animals (Fig. 1i). We made similar observations in C9 ΔnLuc ubi worms ( Supplementary  Fig. 3c).
To evaluate potential age-dependent aggregation of poly-GA, we took advantage of the optical transparency of C. elegans and generated a GFP version of the C9 ubi animals (termed C9-GFP ubi ) by replacing nLuc with GFP. Similar to mouse C9orf72 models 33,34 , we indeed observed an age-dependent poly-GA::GFP aggregation in neurons and non-neuronal cells of C9-GFP ubi animals (Supplementary Fig. 4).
Although we did not detect poly-GP or poly-GR with western blotting ( Supplementary Fig. 5a), more sensitive ELISA assays detected robust expression of poly-GP and modest production of poly-GR in C9 ubi worms, above the background level, observed in ΔC9 ubi controls (Fig. 1g, h). The background levels of poly-GP and poly-GR likely reflect the abundance of GP and GR motifs in the C. elegans proteome. Collectively, the above results demonstrate that RAN translation occurs in C. elegans.
Lastly, our findings in UAG ubi animals show that mutation of the non-canonical initiation codon CUG did not affect C9orf72 mRNA levels nor the formation of G4C2 RNA foci (Fig. 1c), but dramatically reduced poly-GA production ( Fig. 1d-f). In contrast, the levels of poly-GP and poly-GR were not reduced in UAG ubi animals (Fig. 1g, h). We also confirmed this finding in HEK293 cells in vitro ( Supplementary Fig. 5b, c), consistent with a previous report in human induced pluripotent stem cell (iPSC)-derived motor neurons 35 . Interestingly, the nLuc reporter gene is translated in ΔG4C2 repeats ubi animals, which lack G4C2 repeats but carry the intronic C9orf72 sequences including the upstream CUG (Supplementary Fig. 6). However, it remains unclear whether this CUG can initiate translation in the absence of G4C2 repeats. Importantly, we found that the median survival of C9 ubi worms is significantly shorter when compared to ΔC9 ubi and UAG ubi controls (P < 0.0001) ( Fig. 2a and Supplementary  Fig. 7a-c). The lack of an effect on median survival of UAG ubi animals suggests that the G4C2 RNA foci present in these animals do not affect organismal survival. Together, these findings suggest that one or more of the three DPRs (poly-GA, poly-GP, poly-GR) generated from the 75 G4C2 repeats in the C9 ubi animals likely contribute to the reduced survival of these animals, thereby establishing a C. elegans model for C9orf72-mediated ALS/FTD. C9 ubi animals display progressive locomotion defects and motor neuron degeneration. At the level of animal behavior, we conducted swimming and crawling assays on C9 ubi worms. The swimming assay was performed in a longitudinal fashion at juvenile (larval stage 4, L4) and adult (days 2 and 5) stages. By quantifying the rate of body flexion while animals are swimming, we found an age-dependent defect in C9 ubi animals compared to three control strains (ΔC9 ubi , UAG ubi , ΔG4C2 repeats ubi ) ( Fig. 2b and Supplementary Fig. 7d). Next, we performed a highresolution behavioral analysis of freely crawling adult (day 2) C. elegans using automated worm tracking technology 36,37 .
This analysis can quantitate several features related to locomotion (e.g., velocity, crawling amplitude, body curvature). We found multiple locomotor defects in C9 ubi animals, such as reduced velocity and impaired body curvature . These defects were not present in ΔC9 ubi and UAG ubi , and ΔG4C2 repeats ubi controls, indicating that a hallmark of ALS, namely motor deficits, is recapitulated in the C9 ubi animal model. Crawling defects were also present in older (adult day 6) C9 ubi animals, but not in ΔC9 ubi and UAG ubi controls ( Supplementary Videos 1-3). The presence of locomotion and survival defects in C9 ubi , but not UAG ubi animals is unlikely to be a result of different G4C2 mRNA amounts since both showed comparable mRNA levels (Fig. 1b).
The locomotion defects could be attributed, at least partially, to the degeneration of ventral nerve cord motor neurons, which are essential for C. elegans locomotion. To test this possibility, we fluorescently labeled the cholinergic motor neurons present in the nerve cord of C9 ubi and ΔC9 ubi animals with the unc-17/ VAChT::gfp transgenic reporter, which allows for single-cell resolution in our analysis (Fig. 2c, d). Next, we quantified the number of cholinergic motor neurons (unc-17/VAChT::gfp positive) in C9 ubi and ΔC9 ubi animals at juvenile (L4) and adult (days 2 and 5) stages. Although no differences were observed at L4 and day 2, we found a significant reduction in the number of motor neurons that populate the nerve cord of C9 ubi animals at day 5 ( Fig. 2e), arguing for progressive motor neuron degeneration. Because the snb-1 promoter is broadly active, we questioned whether other neuron types also degenerate. Although a similar longitudinal analysis did not reveal dopaminergic neuron degeneration ( Supplementary Fig. 11), we cannot exclude the possibility that other neuron types in addition to cholinergic neurons also degenerate in C9 ubi animals.
DPR production selectively in neurons is sufficient to cause disease phenotypes in vivo. We found that poly-GA is present both in neuronal and non-neuronal cells of C9 ubi animals because the snb-1 promoter is active in multiple cell types ( Fig. 1i and Supplementary Fig. 4a, b). To test whether DPR production from the G4C2 repeats exclusively in neurons is sufficient to cause disease-associated phenotypes, we generated a second set of transgenic animals by replacing the snb-1 promoter with a fragment of the unc-11 promoter (unc-11 prom8) known to be exclusively active in all C. elegans neurons 26 . We refer to these animals as C9 neuro , ΔC9 neuro , and UAG neuro (Fig. 3a). First, we confirmed by RT-qPCR that the G4C2 repeat RNA (with C9orf72 intronic mRNA) and the nLuc mRNA are transcribed in C9 neuro and UAG neuro animals at comparable levels (Fig. 3b, c). Fig. 1 A C. elegans model for C9orf72-mediated ALS/FTD. a Schematic diagram showing ΔC9 ubi , C9 ubi , and UAG ubi constructs with the snb-1 promoter. The orange and green arrows show primer location to detect C9orf72 and nLuc mRNAs, respectively. nLuc: nanoluciferase; UTR: untranslated region. b The G4C2 repeat mRNA (C9orf72 intronic mRNA) and act-1 mRNA were assessed by RT-PCR on three different transgenic lines per genotype: ΔC9 ubi (kasEx154, kasEx259, kasEx260), C9 ubi (kasEx153, kasEx261, kasEx262), or UAG ubi (kasEx155, kasEx263, kasEx264) (mean ± s.e.m.). One-way ANOVA with Dunnett's multiple comparisons test was performed. c Representative fluorescent images of the head of adult (day 1) ΔC9 ubi (kasIs9), C9 ubi (kasIs7) and UAG ubi (kasIs10) animals. N = 25. G4C2 RNA foci are visualized with a Quasar 670 probe. Nuclei are demarcated with DAPI (blue). Scale bar = 10 μm. G4C2 RNA foci are detected in head neurons and motor neurons of the retrovesicular ganglion (circled) both in C9 ubi and UAG ubi animals. The signal detected on the left of the region indicated "Head neurons" belongs to non-neuronal cells (no neurons reside that far anterior in the C. elegans head). d, e Western blotting with poly-GA and α-tubulin antibodies and quantification on three different transgenic lines per genotype: ΔC9 ubi (kasEx154, kasEx259, kasEx260), C9 ubi (kasEx153, kasEx261, kasEx262), or UAG ubi (kasEx155, kasEx263, kasEx264) (mean ± s.e.m.). One-way ANOVA with Dunnett's multiple comparisons test was performed. f Luciferase assay was performed in worm lysates from the same three transgenic lines per genotype shown in panel e (mean ± s.e.m.). g, h ELISA for poly-GP and poly-GR in ΔC9 ubi (kasIs8), C9 ubi (kasIs7), and UAG ubi (kasIs10) animals (n = 4, mean ± s.e.m.). One-way ANOVA with Dunnett's multiple comparisons test was performed. i Immunofluorescent staining for poly-GA (red) in ΔC9 ubi (kasIs9) and C9 ubi (kasIs10) adult (day 1) worms. Representative images of the head and ventral nerve cord region taken at ×40 magnification. N = 20. Anterior to the left. Motor neuron nuclei in blue (DAPI) shown with arrowheads. Scale bar = 20 μm. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26303-x Similar to C9 ubi animals, we found through western blotting that poly-GA is robustly produced in C9 neuro animals, but not in control animals lacking G4C2 repeats and intronic sequences (ΔC9 neuro ) or in animals having the non-canonical translation initiation codon CUG mutated to UAG (UAG neuro ) (Fig. 3d, e). Importantly, we found a dramatic decrease in the median survival of C9 neuro animals when compared to ΔC9 neuro and UAG neuro controls (P-value < 0.0001) ( Fig. 3f and Supplementary Fig. 12). Our findings establish C9 neuro animals as a model for C9orf72mediated ALS/FTD, and further suggest that decreased levels of poly-GA proteins in neurons are sufficient to improve ALS/FTDassociated phenotypes.
Genetic removal of eif-2D/eIF2D reduces DPR production and ameliorates lifespan and locomotion defects in C9 ubi animals. The C9 ubi and C9 neuro worm models for C9orf72-mediated ALS/ FTD offer a unique opportunity to uncover the molecular mechanisms responsible for RAN-translated DPRs from G4C2 repeats. Our results suggest that RAN translation of one DPR, poly-GA, in both in vivo models (C9 ubi and C9 neuro worms) requires the non-canonical initiation codon CUG (Figs. 1d-h and 3d, e). We, therefore, surveyed the literature for factors known to initiate translation via this codon. Two proteins, the eukaryotic translation initiation factors eIF2A and eIF2D, are known to deliver tRNAs to the CUG codon and initiate translation  38,39 . A single ortholog for each factor is present in the C. elegans genome, that is, eif-2A for human eIF2A and eif-2D for human eIF2D. We, therefore, questioned whether eif-2A and/or eif-2D are necessary for DPR expression from the G4C2 repeats. To test this, we obtained mutant animals that carry strong lossof-function (putative null) alleles for eif-2A (gk358198) or eif-2D (gk904876). Both alleles carry missense mutations resulting in a premature STOP codon: W136Amber for eif-2A (gk358198) and R25Opal for eif-2D (gk904876) (see "Methods" section). Since these mutant animals are viable and fertile, we crossed them to the C9 ubi worms and generated eif-2A (gk358198); C9 ubi and eif-2D (gk904876); C9 ubi animals. Next, we found via RT-qPCR that the G4C2 repeat RNA with flanking C9orf72 intronic sequence, as well as the nLuc mRNA are normally transcribed in both eif-2A ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26303-x (gk358198); C9 ubi and eif-2D (gk904876); C9 ubi animals ( Fig. 4a, b), indicating that loss of eif-2A or eif-2D does not affect transcription or stability of the repeat RNA. Consistent with these observations, G4C2 RNA foci can be detected in eif-2D (gk904876); UAG ubi , and eif-2D (gk904876); C9 neuro animals ( Supplementary Fig. 13). Of note, we found a prominent decrease in poly-GA steady-state levels in eif-2D (gk904876); C9 ubi compared to C9 ubi animals, but no effect in eif-2A (gk358198); C9 ubi animals ( Fig. 4c, d), implicating eif-2D in poly-GA expression. A rescue assay further established a causal role for eif-2D in poly-GA expression ( Supplementary Fig. 14). The observed rescue was partial likely due to the mosaic nature of the eif-2D rescuing transgene. Interestingly, we observed a modest but statistically significant decrease in poly-GP production in eif-2D (gk904876); C9 ubi animals ( Supplementary Fig. 15), suggesting other factors, in addition to eif-2D, are also involved in poly-GP translation. We did not observe any effects on poly-GR expression in eif-2D (gk904876); C9 ubi animals ( Supplementary Fig. 15).
Since eif-2D is necessary for steady-state levels of at least two of the DPRs, loss of eif-2D gene activity may improve survival in C9 ubi animals. Indeed, we found that the eif-2D (gk904876); C9 ubi animals display a normal lifespan, similar to ΔC9 ubi and UAG ubi animals ( Fig. 4e and Supplementary Figs. 16,17). Interestingly, the eif-2D (gk904876); UAG ubi animals also display a normal median survival despite producing G4C2 RNA foci (Supplementary Figs. 13a, 17 and 18), suggesting that the improved survival of eif-2D (gk904876); C9 ubi animals are due to decreased DPR (poly-GA and/or poly-GP) expression.
We next asked whether the locomotor defects of C9 ubi animals (detected by automated worm tracking) can be ameliorated upon genetic removal of eif-2D, and found this to be the case (Supplementary Figs. 8 and 9). Of note, eif-2D (gk904876) single mutants or eif-2D (gk904876); UAG ubi animals did not display any prominent locomotor defects compared to controls (ΔC9 ubi and UAG ubi ) ( Supplementary Fig. 19). In summary, this behavioral analysis on C9 ubi animals complements our molecular and survival assays (Fig. 4), providing compelling evidence that loss of eif-2D ameliorates the pathogenic phenotype associated with RAN translation of DPRs (poly-GA, poly-GP) from C9orf72 G4C2 nucleotide repeats.
Lastly, we crossed the eif-2D (gk904876) mutants to the C9 neuro animals to test whether neuronal eif-2D gene activity is critical for poly-GA production and animal survival. Although transcription of the G4C2 repeats (C9orf72 intronic mRNA) increased in eif-2D (gk904876); C9 neuro for unclear reasons (Fig. 4f, g), we found a prominent decrease in poly-GA production in these animals (Fig. 4h, i). Importantly, loss of eif-2D restored survival of C9 neuro animals to levels similar to ΔC9 neuro and eif-2D (gk904876) control animals ( Fig. 4j and Supplementary Fig. 20). Similar to the case of eif-2D (gk904876);UAG ubi , the eif-2D (gk904876); UAG neuro animals also display a normal median survival ( Supplementary  Fig. 21).
eif-2D/eIF2D is broadly expressed and controls poly-GA steady-state levels independently of G4C2 repeat length. Eukaryotic translation initiation factors are often expressed broadly in many tissues, but the expression pattern of C. elegans eif-2D and its vertebrate orthologs remains unknown. Through CRISPR/Cas9 genome editing (SunyBiotech), we engineered an endogenous reporter allele (syb3364) by introducing the gfp-3xFlag sequence before the stop codon of eif-2d ( Fig. 5a and see "Methods" section). This reporter showed cytoplasmic expression (consistent with the role of eif-2D in translation) in most, if not all, C. elegans tissues (Fig. 5a, b).
Despite its broad expression, eif-2D does not appear to play a role in canonical translation because the loss of eif-2D gene activity in C. elegans did not affect the expression of an AUGinitiated green fluorescent protein (GFP) transgene or endogenous reporter alleles for two AUG-initiated transcription factors known to be expressed in C. elegans ventral nerve cord motor neurons ( Supplementary Fig. 22). These findings together with the normal lifespan and locomotion pattern of eif-2d mutants (described above) suggest eif-2D does not broadly affect canonical translation in C. elegans, albeit we cannot exclude the possibility of eif-2D being required for translation of other AUG-initiated transcripts.
Next, we sought to determine whether eif-2D controls poly-GA steady-state levels independently of G4C2 repeat length. To test this, we generated transgenic animals that carry 5 G4C2 repeats, referred to as C9 5xG4C2 ubi , in which nLuc is in the poly-GA frame ( Supplementary Fig. 23). Although the RNA levels of intronic C9orf72 and nLuc were similar between C9 5xG4C2 ubi and eif-2d (-); C9 5xG4C2 ubi animals, a luciferase assay showed that the level of nLuc (indicative of poly-GA expression) was significantly decreased in the eif-2d (-); C9 5xG4C2 ubi animals ( Supplementary Fig. 23). These results together with the data on C9 ubi animals that carry 75 G4C2 repeats (Figs. 4 and 5) suggest that eif-2D controls poly-GA steady-state levels independently of G4C2 repeat length.
The SUI1 domain of eif-2D/eIF2D is required for poly-GA expression in C. elegans. To obtain mechanistic insights, we interrogated the protein domains of C. elegans eIF-2D. Similar to human eIF2D, we found a PUA domain at the N terminus and a SUI1 domain at the C-terminus of C. elegans eIF-2D (Fig. 5a). In eukaryotes, the SUI1 domain is found in translation initiation factors eIF1 and DENR 38 . Because SUI1 is important for accurate translation initiation codon recognition by eIF1 40,41 , we hypothesized that the SUI1 domain of C. elegans eIF-2D is required for DPR translation. To test this, we first employed CRISPR/Cas9 genome editing and selectively replaced the SUI1 domain with a 3xFLAG epitope (Fig. 5a). Homozygous animals lacking the SUI1 domain (termed eif-2D ΔSU1I ) expressed a truncated eIF-2D protein ( Supplementary Fig. 24) and were viable. When crossed to C9 ubi , we observed a prominent reduction in poly-GA expression with no change in the levels of C9orf72 intronic a Schematic of constructs used in ΔC9 neuro (kasEx159), C9 neuro (kasEx157), and UAG neuro (kasEx160) animals. The orange and green arrows show location of primers used to detect C9orf72 and nLuc mRNAs, respectively. nLuc: nanoluciferase; UTR: untranslated region. b The G4C2 repeat RNA (with C9orf72 intronic RNA) and act-1 mRNAs were assessed by RT-PCR. c The nLuc and act-1 mRNAs were assessed by RT-PCR. Experiments were repeated four times using 1 transgenic line per genotype, mean ± s.e.m.). One-way ANOVA with Dunnett's multiple comparisons test was performed. d Worm lysates were processed for western blotting and immunostained with poly-GA and αtubulin antibodies. e Quantification of poly-GA expression on western blots. The experiments were repeated three times using 1 transgenic line per genotype, mean ± s.e.m.). One-way ANOVA with Dunnett's multiple comparisons test was performed. f Lifespan of ΔC9 neuro (kasEx159), C9 neuro (kasEx157), and UAG neuro (kasEx160) animals. The experiments were repeated three times using 1 transgenic line per genotype, mean ± s.e.m. ΔC9 neuro (kasEx159), n = 75; C9 neuro (kasEx157), n = 75; UAG neuro (kasEx160), n = 75). See Supplementary Fig. 12 (Fig. 5c-e). Second, we tested animals carrying the gk561128 allele that generates a nonsense mutation (Q512Ochre) within the SUI1 domain (Fig. 5a). Again, we observed a strong reduction in poly-GA expression, but no effect on C9orf72 intronic mRNA (Fig. 5a, f-h). These observations strongly suggest the SUI1 domain of eIF-2D has a key role in poly-GA translation.
Knockdown of eIF2D suppresses poly-GA production in mammalian cells. The prominent reduction of poly-GA in eif-2D mutants suggests that C. elegans eIF-2D plays a crucial role in poly-GA translation (Figs. 4 and 5). To extend these findings to mammalian cells, we transfected mouse NSC-34 motor neuronlike cells with a monocistronic construct carrying 75 G4C2 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26303-x repeats and flanking C9orf72 intronic sequences (Fig. 6a), similar to the one used to generate the C9 ubi animals (Fig. 1a). shRNAmediated knockdown of eIF2D in NSC-34 cells resulted in a strong reduction of poly-GA (Fig. 6b-d), similar to the effects observed with C. elegans eif-2D mutants. Compared to control shRNA, eIF2D shRNA did not affect poly-GP and poly-GR steady-state levels in these cells (Fig. 6e, f).
Next, we employed an in vitro platform to differentiate iPSCderived motor neurons from a healthy individual and two ALS patients with G4C2 repeat expansions (see "Methods" section), ( Supplementary Fig. 25). Despite using a sensitive ELISA method, we were unable to detect poly-GA expression above background levels in the iPSC-derived motor neurons from these specific ALS patients, preventing us from testing the role of eIF2D in poly-GA translation in these cells ( Supplementary Fig. 26). Future experiments with additional iPSC lines from other ALS patients carrying long G4C2 repeats could help clarify the role of human eIF2D in poly-GA production. However, we did detect poly-GR and poly-GP in these iPSC-derived motor neurons. Similar to our findings in C. elegans and NSC-34 cells, siRNA-mediated knockdown of eIF2D had no effect on poly-GR ( Supplementary  Fig. 26). As in NSC-34 cells, we also observed no difference in poly-GP expression after eIF2D knockdown ( Supplementary  Fig. 26), which together with the small effect on poly-GP steadystate levels observed in C. elegans eif-2D mutants suggest that multiple translation initiation factors control poly-GP expression (see "Discussion" section and Table 1).
Our data with monocistronic G4C2 constructs in C. elegans (Figs. 4 and 5), and mouse NSC-34 cells (Fig. 6a-f) are complemented with in vitro data in human HEK293 cells transfected with a bicistronic construct carrying 75 G4C2 repeats and flanking intronic sequences in the second cistron ( Fig. 6g and Table 1). To monitor the expression of different DPRs, we generated three different versions of this construct by inserting nLuc in the poly-GA, poly-GP, or poly-GR reading frame (Fig. 6g). In the transfected cells, we robustly detected all three DPRs (Fig. 6h-m). Similar to our findings in C. elegans eif-2D mutants, eIF2D knockdown decreased poly-GA and poly-GP (Fig. 6h-k), whereas eIF2D overexpression increased DPR production ( Supplementary Fig. 27). Importantly, the protein levels of fLuc, which is expressed by the first cistron via canonical AUG-dependent translation, were similar in cells transfected with either control or eIF2D shRNAs (Fig. 6n), suggesting a specific role for eIF2D in non-canonical translation in human cells. Interestingly, we also observed an effect on poly-GR upon eIF2D knockdown in HEK293 cells transfected with the bicistronic construct (Fig. 6k). We surmise the way DPR translation is initiated in this construct is likely quite different from the monocistronic constructs used in the worm and NSC-34 cells, and from the endogenous human C9ORF72 locus. Taken together, our in vivo (C. elegans) and in vitro (NSC-34, HEK293) data uncovered a conserved role for eIF2D in controlling the steady-state levels of at least one DPR (poly-GA). Table 1 summarizes the observed effects on DPRs upon eif-2D/eIF2D knockdown in vivo (C. elegans) and in vitro (NSC-34, HEK293, human iPSC-derived motor neurons).
eif-2D/eIF2D does not modulate phosphorylation of eukaryotic initiation factor-2α (eIF2α). A growing body of evidence suggests that phosphorylation of eIF2α as a result of activation of the integrated stress response (ISR) leads to increases in DPR production 29,42 . We, therefore, wondered whether eIF2D is involved in stress-activated phosphorylation of eIF2α. However, we found no significant differences in phosphorylated eIF2α levels between wild-type and eif-2D mutant animals carrying the C9 ubi transgene (Supplementary Fig. 28). Furthermore, eIF2D knockdown does not affect the level of p-eIF2α in HEK293 cells following administration of a known ER stressor (sodium arsenite) (Supplementary Fig. 29). Lastly, administration of the proteasome inhibitor MG132 in this cellular system did not affect the results of eIF2D knockdown on RAN translation (Supplementary Fig. 30). Altogether, our findings suggest eif-2D/eIF2D controls DPR steady-state levels without altering the levels of p-eIF2α.

Discussion
The discovery of RAN translation of C9orf72 nucleotide repeats combined with the reported toxicity of DPRs in model systems have focused the ALS/FTD field on this unconventional form of translation, as well as on DPRs as potential therapeutic targets. To study the molecular mechanisms underlying RAN translation, we initially developed a C. elegans model that carries a transgene encoding 75 copies of the G4C2 repeat flanked by intronic C9orf72 sequences under the control of a ubiquitous promoter. These C9 ubi animals produce three DPRs from the sense transcript, namely poly-GA, poly-GP, and poly-GR, and show signs of neurodegeneration, as well as locomotor and lifespan defects. Such phenotypes are likely due to DPR toxicity because mutation of a non-canonical CUG initiation codon selectively decreased poly-GA production and ameliorated locomotor and lifespan defects in C9 ubi animals without affecting C9orf72 mRNA levels or the formation of G4C2 RNA foci. Moreover, a second C. elegans model (C9 neuro ) that carries a similar transgene exclusively expressed in neurons displayed a similar phenotype. Importantly, we found that three loss-of-function mutations in the eukaryotic translation initiation factor eif-2D/eIF2D suppressed DPR expression (poly-GA, poly-GP) and ameliorated lifespan defects in both C9 ubi and C9 neuro models. Mechanistically, we found the conserved SUI1 domain at the C-terminus of eIF-2D/eIF2D is required for DPR (poly-GA) expression. Lastly, our in vitro findings in mouse and human cells combined with our C. elegans data (Table 1) suggest a conserved role of eif-2D/eIF2D in RAN translation.
research is the relative contribution of each of the five DPRs (poly-GA, poly-GP, poly-GR, poly-PR, poly-PA) to disease pathogenesis. In this paper, two lines of evidence indicate that poly-GA, the most readily detected DPR in the brain and spinal cord of ALS/FTD patients with C9orf72 expanded repeats 43,44 , significantly contributes to disease phenotypes in C. elegans: (1) poly-GA aggregates and appears to be the most abundantly expressed DPR in our worm models, and (2) mutation of the non-canonical CUG initiation codon selectively decreases poly-GA production, and ameliorates locomotor and lifespan defects in C9 ubi and C9 neuro animals. Importantly, these in vivo results significantly extend previous reports demonstrating that this same CUG is required in vitro for poly-GA translation 20,29-31 . Of note, mutation of this CUG to UAG both in C. elegans (Fig. 1) and an in vitro cellular system (Supplementary Fig. 5) did not affect poly-GP and poly-GR translation, suggesting different near-cognate translation initiation codons are used for each DPR. Indeed, an AGG located 1 nt upstream of the repeat has been reported to initiate translation of poly-GR 31 . Despite the above findings, we cannot rule out the possibility that a frameshift occurs in DPR translation upon CUG mutation, as found by other investigators 30 .
In support of our C. elegans findings, recent studies in mice reported neurodegeneration, motor deficits, and cognitive defects following ubiquitous expression of poly-GA 45 , as well as behavioral abnormalities following spinal cord expression of poly-GA 46 . A prevailing hypothesis is that poly-GA causes cellular toxicity through proteasomal impairment and protein sequestration 5 . A recent publication reports that poly-GA spreads cell-to-cell in an in vitro system and inhibits proteasomal function both cell-autonomously and non-cell-autonomously, leading to ubiquitination of the nuclear localization signal of TDP-43 and its subsequent cytoplasmic mislocalization 47 . Importantly, treatment with an antibody against poly-GA ameliorated TDP-43 mislocalization 47 , as well as toxicity in mouse models of mutant C9orf72-induced ALS/FTD 48 . A zebrafish model that produces poly-GA did not exhibit motor axon toxicity 49 , however, this may be due to low poly-GA expression.
Our C9 ubi worm model also produces poly-GP and poly-GR. poly-GP is thought to be relatively nontoxic 5 compared to poly-GR, which has been demonstrated to be toxic in mouse neurons 34 . In C. elegans, codon-optimized constructs driving expression of poly-GR or poly-PR caused toxicity in muscles and neurons 50 . Furthermore, initiation of poly-GR translation has been reported to occur at the same upstream CUG codon used for poly-GA production, but with subsequent frameshifting into the GR reading frame 30 . However, two recent papers state that translation initiation of poly-GR did not occur from this upstream CUG codon 31,35 . Because we detected poly-GR in our C. elegans model ( Fig. 1 and Supplementary Fig. 15), it is possible that poly-GR contributes to the observed disease phenotypes. Supporting this notion, very low levels of poly-GR in neurons have been reported to impair synaptic function and induce behavioral deficits in both Drosophila and mice 34 .
In addition to what appears to be DPR toxicity in our C. elegans models, there remains the question as to whether the repeat RNA itself rather than, or in addition to, its protein product could be toxic 51 . A role for RNA toxicity does not appear to be the case in our nematode experiments since mutation of the non-canonical initiation codon CUG to UAG led to a significant decrease in poly-GA expression and rescue of worm survival to wild-type levels, but had no effect on C9orf72 mRNA levels nor G4C2 RNA foci formation. It is possible, however, that the CUG > UAG mutation, albeit minimal in nature, can affect RNA structure.
One question that arises is whether our transgenic C9 ubi and C9 neuro animals constitute an appropriate model for the patient with C9orf72 expanded repeats. It is clear that the RNA template (repeat RNA plus flanking intronic sequence) produced in our C. elegans animals differs from the RNA template actually expressed in the central nervous system of ALS/FTD patients with a C9orf72 expanded repeat. The human G4C2 expanded repeat varies in length in patients and even in different cells within the same patient. The latter finding suggests that a number of potential toxicities-besides one related to DPR production-may be present in the human disease state, including toxicity related to repeat RNA and/or synthesis of other DPRs. Toxicity may also occur as a consequence of a decrease in the authentic C9orf72 gene product, consistent with a recent study reporting that a decrease in C9ORF72 protein leads to a decrease in autophagic clearance of toxic accumulations of the DPRs 31 .
eif-2D/eIF2D regulates poly-GA expression from G4C2 nucleotide repeats. eIF2D is known to initiate translation in vitro at both AUG and non-AUG initiation codons 38,52 . The present study demonstrates a key role of eIF2D in poly-GA expression since eIF2D knockdown in two cellular systems (mouse NSC-34 and human HEK293 cells) and three different eif-2D loss-of-function alleles in C. elegans led to a prominent decline in poly-GA steady-state levels ( Table 1). It is noteworthy that genetic removal of eif-2D in C. elegans does not appear to have an effect on (a) canonical AUG-initiated translation of three tested mRNAs ( Supplementary Fig. 22), (b) animal locomotion, Fig. 6 Knockdown of eIF2D suppresses DPR expression in NSC-34 and HEK293 cells. Monocistronic or bicistronic constructs were cotransfected along with either control or anti-eIF2D shRNA and cultured for 48 h. a Schematic diagram showing C9-HA construct. AG promoter: CMV enhanced chicken βactin promoter. HA is in poly-GA frame. FLAG and Myc are in poly-GR and poly-GP frames, respectively, but are not schematically depicted. BGH bovine growth hormone. b A representative western blot of NSC-34 lysates to detect poly-GA, eIF2D, and β-actin. c, d Quantification of eIF2D (c) and poly-GA (d) expression from the western blots. e-f Quantification of poly-GP (e) and poly-GR (f) by ELISA in NSC-34 cells. The experiments were repeated four times. mean ± s.e.m. Turkey's multiple comparisons test was used to calculate P-values. One-way ANOVA with Tukey's multiple comparisons test was performed. g Schematic diagram showing the bicistronic constructs. CMV cytomegalovirus promoter; fLuc firefly luciferase; nLuc nanoluciferase; BGH bovine growth hormone. h The cell lysates were processed for western blotting and immunostained with poly-GA, fLuc, eIF2D, and α-tubulin antibodies. i, j, n Quantification of eIF2D (i), poly-GA(j), and fLuc (n) on western blots. The experiments were repeated three times. mean ± s.e.m. Two-tailed unpaired t-test was performed. k The levels of luciferase activity were assessed by dual-luciferase assays. The experiments were repeated three times. mean ± s.e.m. l, m Poly-GP and poly-GR detected with ELISA in HEK293 cells transfected with a ΔC9 or C9 bicistronic construct. The experiments were repeated three times. mean ± s.e.m. Two-tailed unpaired t-test was performed. and (c) lifespan. The absence of detrimental phenotypes in C. elegans animals globally lacking eif-2D/eIF2D makes this factor an attractive therapeutic target. What is the mechanism through which eif-2D/eIF2D controls poly-GA steady-state levels? We and others found that poly-GA translation is initiated from a CUG (−24nt upstream of G4C2) 20,[29][30][31] , and a recent study showed that Met-tRNA is used for initiation of poly-GA translation at this CUG 31 . A previous study showed that both Met-tRNA and non-Met-tRNA are delivered by eIF2D to the P-site of the 40S ribosomal subunit during translation initiation at the HCV IRES in HEK293 cells 38 . Additional cryo-electron microscopy results showed that the SUI1 domain of eIF2D contacts the codon-anticodon duplex on the HCV IRES, and also contacts Met-tRNA, presumably stabilizing its orientation 53,54 . Besides eIF2D, the SUI1 domain is also present in two other non-canonical initiation factors, eIF1 and DENR 38 . It has been proposed that eIF2D may act by changing the conformation of the 40S ribosomal subunit in a manner similar to eIF1, but unlike eIF1, eIF2D stabilizes tRNA binding within the P-site 38 . Using genome engineering in C. elegans, we found the SUI1 domain is required for poly-GA steady-state levels. This observation together with the similar effects on DPR expression observed in eif-2D (-); C9 ubi and UAG ubi animals favor a model in which eIF-2D acts at the CUG codon to deliver the Met-tRNA (which is necessary for poly-GA initiation) to the P-site of the 40S ribosomal subunit. However, it is important to acknowledge that eIF2D has additional functions besides translation initiation. For example, eIF2D was implicated in recycling 40S ribosomal subunits [55][56][57] , which enabled ribosomes and mRNAs to participate in multiple rounds of translation, leading to efficient protein expression. It is thus possible that genetic removal of eif-2D in our worm models caused inefficient DPR production by more than one mechanism.
Multiple factors are involved in DPR translation from G4C2 nucleotide repeats. As mentioned in Introduction, recent studies suggest multiple factors (e.g., RPS25, eIF3F, eIF4B, eIF4H) are involved in RAN translation of DPRs from G4C2 transcripts. It remains unclear though to what extent the production of each DPR is controlled by these factors, in part because studies often focus on a single DPR. We provide evidence both in C. elegans and two cell culture systems (mouse NSC-34 and human HEK293 cells) that eif-2D/eIF2D controls the steady-state levels of poly-GA. Moreover, our C. elegans data suggest a minor role for eif-2D/eIF2D in poly-GP expression and no effect on poly-GR, suggesting that the production of each DPR may involve additional factors. Consistent with this notion, RPS25 knockdown in vitro reduced poly-GA by 90%, poly-GP by 50%, and poly-GR by 30% 13 . A promising candidate to work together with eIF2D is DENR, another initiation factor shown in a recent fly study to act redundantly with eIF2D during development and integrated stress response 58 . Of note, it is possible that these factors do not control each other's expression based on our findings with a C. elegans ortholog of eIF4B and eIF-2D (Supplementary Fig. 31). Collectively, these and other studies 16 suggest that a number of different factors play critical roles in RAN translation of DPRs.
Our previous study demonstrated that knockdown of eIF2A in HEK293 cells and neural cells in the chick embryo led to a modest (30%) decrease in poly-GA expression from G4C2 expanded repeats 20 . A recent study also implicated eIF2A in RAN translation from CCUG and CAGG repeats in myotonic dystrophy type 2 (DM2) 59 . In the present study, however, eif-2A/eIF2A deficiency had no effect on poly-GA expression in C. elegans, while eIF2D knockdown in NSC-34 and HEK293 cells and eif-2D knock-out in C. elegans had a prominent effect on poly-GA expression. These results suggest that eIF2D plays a more important role in RAN translation than eIF2A. It is possible, however, that eIF2A also plays a significant role in RAN translation from G4C2 expanded repeats, but only upon cellular stress since it can translate a subset of mRNAs when eIF2α is phosphorylated 17,18,60 . We note that eIF2D does not control the levels of phosphorylated eIF2α in C. elegans and HEK293 cells ( Supplementary Figs. 27 and 28).
RAN translation occurs in C. elegans and offers the opportunity to discover conserved regulators. Two previous studies reported DPR-associated toxicity in C. elegans by using either a codon-optimized construct that carries no G4C2 repeats, or a construct that carries 66 G4C2 repeats 22,50 . The latter study strongly indicated that RAN translation occurs in the worm. However, the non-canonical (non-AUG) initiating codon and translation initiation factor remained elusive. By generating C. elegans animals that carry 75 G4C2 repeats flanked by intronic C9orf72 sequences and subsequently blocking poly-GA expression by mutating a non-canonical initiation codon (CUG) upstream of the repeats, we advanced the molecular understanding of how RAN translation occurs in the worm. This finding in C. elegans, together with previous reports in yeast and flies 16,22 , strongly suggest that the molecular machinery required for RAN translation is evolutionary conserved, offering an opportunity to use simple model systems for the discovery of RAN translation regulators.
In summary, we leveraged the specific strengths of the C. elegans model and discovered that the non-canonical initiation factor eif-2D/eIF2D is required for DPR expression (poly-GA, poly-GP) from expanded G4C2 repeats in the C9orf72 gene. Our in vitro findings argue for a conserved role of eif-2D/eIF2D in poly-GA expression, which could be potentially used to pursue directions for C9orf72-associated ALS and FTD therapy. In fact, because RAN translation of other nucleotide repeat expansions (not G4C2) occurs in several neurodegenerative diseases 4 , eIF2D could be investigated as a therapeutic target in these disorders as well.

Methods
C. elegans strains. Worms were grown at 15, 20, or 25°C on nematode growth media (NGM) plates supplied with E. coli OP50 as food source 61 . Mutant alleles used in this study include eif-2A (gk358198), eif-2D (gk904876), and eif-2D (gk561128), which were backcrossed three times to the wild-type (N2) strain. The following strains were generated using CRISPR/Cas9 genome editing by Suny-Biotech. The strain PHX3364 with genotype eif-2D (syb3364) is an endogenous reporter for eif-2d because the GFPnovo2-3xFLAG sequence has been inserted before the stop codon. The strain PHX3432 with genotype eif-2D (syb3432) carries a replacement of the SUI1 domain with a 3xFLAG sequence. A complete list of all primers used in this study is provided in Supplementary Table 1.
Generation of transgenic C. elegans animals. The template for the generation of the plasmids used in the C9 ubi , ΔC9 ubi , UAG ubi and C9 5xG4C2 ubi animals has been previously described in an in vitro study 20 . In C9 ubi animals, the plasmid has 75 G4C2 repeats with 113 nucleotides 5′ and 99 nucleotides 3′ of intronic sequences that normally flank the G4C2 repeats in the human C9orf72 gene; this sequence is followed by nanoluciferase (nLuc) reporter in the reading frame of poly-GA, which is followed by the unc-54 3′ UTR. In ΔC9 ubi animals, the construct lacks the G4C2 repeats and the flanking intronic sequences. In UAG ubi animals, the construct is identical but the non-canonical translation-initiating codon CUG is mutated to UAG. The promoter sequences of the snb-1 and unc-11 genes are shown in Supplementary Data 1. The unc-54 3′ UTR was cloned into the XhoI-SacI site of the plasmids.
To generate the ΔG4C2 repeats ubi construct, ΔG4C2-1-F/R and ΔG4C2-2-F/R fragments were phosphorylated, annealed, and then ligated into restriction sites of HindIII and NotI of the C9 ubi construct. To generate the C9 ΔnLuc ubi construct, the NotI-ΔnLuc-BamHI-F/R fragment was inserted into restriction sites of NotI and BamHI of the C9 ubi construct. For the 3xFlag-eif-2D ubi construct, the 3xFlag sequence (synthesized by Integrated DNA Technologies) was inserted into restriction sites of HindIII and NotI, and eif-2D cDNA was inserted into the restriction sites of NotI and BamHI of the C9 ubi construct. For the C9-GFP ubi NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26303-x ARTICLE NATURE COMMUNICATIONS | (2021) 12:6025 | https://doi.org/10.1038/s41467-021-26303-x | www.nature.com/naturecommunications construct, GFP novo2 62 was inserted into restriction sites of NotI and BamHI of the C9 ubi construct.
Lifespan assay. The lifespan assay was performed as previously described 22 . Worms were transferred at the L4 larval stage to NGM plates coated with OP50 bacteria in the presence or absence of 50 μM 5-fluoro-2'deoxyuridine (FUDR, Sigma). Worms were kept at 20°C and scored every 2 or 3 days. If the worms did not respond to prodding the head with the wire spatula, they were scored as dead worms. In the assay with FUDR, the worms were moved to new plates before bacterial food was depleted. In the assay without FUDR, the worms were moved to new plates every 2 days. If we were unable to find the same number of worms on the plate after 2 days, they were scored as censored worms. Similar results were obtained when lifespan assays were performed with ( Microscopy. Worms were anesthetized using 100 mM of sodium azide (NaN 3 ) and mounted on 5% agarose pads on glass slides. Images were taken using an automated fluorescence microscope (Zeiss, Axio Imager Z2) equipped with a Zeiss Axiocam 503 mono using the ZEN software (Version 2.3.69.1000, Blue edition). Representative images are shown following max-projection of 2-5 Z-stack images (each~0.5 μm) using the maximum intensity projection type. Image reconstruction was performed using Image J software (version 2.0.0-rc-59/1.51k) 63 .
Quantification of microscopy images. The number of neurons expressing the respective fluorescent reporter was quantified manually using Image J software (version 2.0.0-rc-59/1.51k) 63 . Quantification was performed in a blinded fashion. Details related to the fluorescent reporter, as well as the number and stage of animals analyzed are provided in the respective figure legends.
Immunocytochemistry C. elegans. Animals of the respective genotypes were grown at 20°C on nematode growth media (NGM). We followed an immunocytochemical staining procedure described previously 64 . In brief, worms were prepared for staining following the freeze-crack procedure and subsequently fixed in ice-cold acetone (5 min) and icecold methanol (5 min). Worms were transferred using a Pasteur pipette from slides to a 50 ml conical tube that contained 40 ml 1 × PBS. Following brief centrifugation (2 min, 1800 × g), worms were pelleted and 1 × PBS was removed. Next, worms were incubated with 300 μl of blocking solution (1 × PBS, 0.2% Gelatin, 0.25% Triton X100) for 30 min at room temperature (rolling agitation). Following removal of the blocking solution, worms were incubated overnight with a mouse anti-poly-GA antibody (1:25 in PGT solution, EMD Millipore). Next, the primary antibody solution was removed and worms were washed five times with washing solution (1 × PBS, 0.25% Triton X100). Worms were incubated with an Alexa Fluor 594 goat anti-mouse IgG secondary antibody (1:1000 in PGT solution, A-11020, Molecular Probes) for 3 h at room temperature. Following 5 washes, worms were mounted on a glass slide and examined with an automated fluorescence microscope (Zeiss, AXIO Imager Z1 Stand).
RNA fluorescent in situ hybridization (FISH). RNA FISH was carried out as previously described 65 . For each genotype, well-fed worms at various stages and grown on 4-5 Petri dishes were fixed with fixation buffer (3.7% formaldehyde in PBS) for 1 h, and then resuspended in 70% ethanol overnight. The samples were washed with wash buffer (10% deionized formamide and 2 × SSC) and hybridized with 5 μM RNA probe (CCCCGGCCCCGGCCCCGG) conjugated with Quasar 670 at the 3′ end (LGC Biosearch Technologies) in hybridization buffer (10% (w/w) dextran sulfate, 10% deionized formamide, and 2 × SSC) overnight at 37°C. The samples were then washed with wash buffer for 30 min at 37°C, and subsequently resuspended with 5 ng/ml DAPI diluted in wash buffer for 30 min at 37°C. After washing with 2 × SSC, the samples were resuspended in GLOX buffer (0.4% glucose, 10 mM Tris-HCl, pH8.0, 2 × SSC, 3.7 μg/ml glucose oxidase, and >300 units catalase), placed onto a slide, and examined with an automated fluorescence microscope (Zeiss, AXIO Imager Z1 Stand).
Automated worm tracking. Worms were maintained as mixed stage populations by chunking on NGM plates with E. coli OP50 as the food source. Worms were bleached and the eggs were allowed to hatch in M9 buffer to arrest as L1 larvae. L1 larvae were refed on OP50 and allowed to grow to day 2 or adulthood. On the day of tracking, five worms were picked from the incubated plates to each of the imaging plates (see below) and allowed to habituate for 30 min before recording for 15 min. Imaging plates were 35 mm plates with 3.5 ml of low-peptone (0.013% Difco Bacto) NGM agar (2% Bio/Agar, BioGene) to limit bacteria growth. Plates are stored at 4°C for at least 2 days before use. Imaging plates were seeded with 50 μl of a 1:10 dilution of OP50 in M9 the day before tracking and left to dry overnight with the lid on at room temperature.
Behavioral feature extraction and analysis. All videos were analyzed using Tierpsy Tracker (Version 228) to extract each worm's position and posture over time 36 . These postural data were then converted into a set of behavioral features selected from a large set of features as previously described 36 . For each strain comparison, we performed unpaired two-sample t-tests independently for each feature. The false discovery rate (FDR) was controlled at 5% across all strain and feature comparisons using the Benjamini Yekutieli procedure. The P-value threshold for this analysis (FDR 5%) was 0.0014.
Swimming behavior. Animals were grown at 15°C until they were assayed at three different stages (L4, day 2, day 5).
Swimming rates were measured with 20 animals per genotype. Individual animals were placed on 40 μl of temperature-equilibrated M9 buffer on an agar plate. After 1 min of acclimatization, the number of changes in the direction of bending at the mid-body region was recorded for 30 s. Animals were chosen at random by assaying the first worm in the field of view.
In vitro studies with NSC-34 motor neuron-like cells. shRNA plasmids against mouse eIF2D were constructed by inserting annealed oligos into SgrAI and EcoRI sites of pFB-U6-shRNA-CMV-GFP vector 66 . NSC-34 cells in a 6-well plate were transfected with either C9-3T plasmid alone or in combination with eIF2D shRNA using transfection reagent Attractene (Qiagen) according to the manufacturer's protocol and cultured for 48 h. The NSC-34 cells are a gracious gift from Dr. Neil Cashman 67 .
In vitro studies with HEK293 cells. Bicistronic plasmids were constructed, as previously described 20 , with firefly luciferase (fLuc) in the first cistron and C9orf72 sequences in the second cistron, with or without a UAG replacing the CUG 24 nts upstream of the 75 G4C2 repeats.
shRNA plasmids against human eIF2D were prepared using previously published methods 20 . In brief, oligonucleotides with an siRNA sequence were cloned into the BamHI and HindIII sites of psilencer 2.1-U6 neo vector (ThermoFisher Scientific) according to the manufacturer's protocol. The control shRNA vector was provided in this kit. The eIF2D-V5 Tag plasmid was prepared by inserting human eIF2D cDNA (RC201726, Origene) into restriction sites of EcoRI and NotI of pcDNA6. The GFP-V5 Tag plasmid was prepared by inserting GFP sequence from pFB-U6-shRNA-CMV-GFP plasmids 66 into restriction sites of EcoRI and NotI of pcDNA6.
For western blotting (see below), HEK293 cells (ATCC, CRL-1573) were plated in 6-well plates at 2 × 10 5 per well and cotransfected with 0.5 μg of bicistronic plasmids along with either 2.5 μg control shRNA or anti-eIF2D shRNA using Lipofectamine LTX (ThermoFisher Scientific). For luciferase assays (see below), the cells were plated in 24-well plates at 5 × 10 4 per well and cotransfected with 100 ng of the bicistronic plasmid along with 500 ng of either control shRNA or anti-eIF2D shRNA using Lipofectamine LTX (ThermoFisher Scientific). For the eIF2D overexpression assay, cells were cotransfected with 20 ng of bicistronic plasmids along with 380 ng of either GFP-V5 Tag or eIF2D-V5 Tag plasmids using Lipofectamine LTX. The cells were cultured for 48 h in DMEM supplemented with 10% FBS, 2 mM L-Glutamine, 100 U/ml Penicillin and 100 μg/ml Streptomycin, and then processed.
siRNA knockdown experiments. After 35 days in culture, motor neurons were transfected with a siRNA specific to eIF2D mRNA or a scrambled control. First, a lipofectamine-siRNA (ThermoFisher Scientific) complex was prepared separately for both the targeting and nontargeting siRNA in OptiMEM with incubation at room temperature for 20 min. After this period, the lipofectamine RNAiMAX (ThermoFisher Scientific)-siRNA solution was brought up to the appropriate volume with MN culture medium. The culture medium in the dish was then aspirated and replaced with a lipofectamine-siRNA solution at a final concentration of 60 pmol siRNA in 1.5 ml medium per 1,000,000 cells. Twenty-four hours later the medium was replaced with a normal MN medium. This procedure was repeated 2 more times at 40 and 45 days in culture. After 50 days in culture, motor neurons were lysed with RIPA buffer (10 mM Tris-HCL pH 8.0, 140 mM NaCl, 1 mM EDTA, 0.1% SDS) supplemented with protease inhibitors (Millipore) on ice for 10 min and sonicated using a microneedle sonicator. After removing the RIPAinsoluble fraction by centrifugation, the RIPA-soluble fraction was assessed for levels of the DPRs.
C9 #2 (ND10689) iPSC line was reprogrammed by and received from the Ichida Lab (University of Southern California). B-lymphocytes were originally purchased from Coriell. Cat# ND10689; RRID:CVCL_BE81. Female, 51 years old. Pluripotency status and presence of a G4C2 repeat expansion were previously described in Shi et al. 70 . In the latter study, the G4C2 repeat size is~14 kb for this line. Repeat-primed PCR in Ortega et al. 68 also confirmed a large number of G4C2 repeats.
Luciferase assay. The cells were lysed with 1× passive lysis buffer (Promega). Levels of nLuc and fLuc were assessed using Nano-Glo Dual-Luciferase Reporter assay system (Promega) and a Wallac 1420 VICTOR 3V luminometer (Perkin Elmer) according to the manufacturer's protocol. The 2030 Workstation software (version 4.0) was used.
RT-PCR. To analyze the expression levels of C. elegans mRNA, total RNA was extracted using TRIzol (ThermoFisher Scientific) according to the manufacturer's protocol and treated with TURBO DNase (ThermoFisher Scientific) for 30 min at 37°C. RNA was then isolated using RNeasy Mini kit (Qiagen). cDNA was generated using SuperScript IV First-Strand Synthesis System (ThermoFisher Scientific). Quantitative RT-PCR was performed using Power SYBR Green PCR Master Mix (ThermoFisher Scientific) and primer sets (Supplementary Data 2) in a CFX96 Real-Time System (Bio-Rad). The CFX Manager software (version 3.1) was used.
To analyze mRNA expression levels in iPSC neurons, total RNA was extracted from siRNA-treated motor neurons using an RNeasy Mini Kit. The cDNA was generated using iSCRIPT Reverse Transcription Supermix (BIO-RAD). RT-qPCR analysis was performed using the iTaq Universal SYBR Green Supermix (BIO-RAD) on the CFX system. To calculate relative expression, the value of the average cycle of threshold (Ct) for housekeeping genes (GAPDH and ACTIN) was subtracted from the average Ct value of eIF2D. This ΔCt was then used to determine the relative gene expression described by the function 2− ΔCt (ΔΔCt), represented as a value relative to the control sample.
ELISA. Two ELISA methods were used and DPRs were detected in a blinded fashion. (A) The worm pellets were suspended in cold co-IP buffer consisting of 50 mM Tris-HCl, pH 7.4, 300 mM NaCl, 5 mM EDTA, 1% Triton-X 100, 2% sodium dodecyl sulfate, 1 × Halt TM Protease, and Phosphatase Inhibitor Cocktail (ThermoFisher Scientific) and sonicated with 10 cycles of 0.5 s pulse/0.5 s pause. Protein lysates were cleared by centrifugation at 16,000 × g for 20 min at 4°C. The protein concentration was determined using a BCA protein assay kit (Thermo-Fisher Scientific). Lysates were diluted to the same concentration using co-IP buffer, and the levels of poly-GP in lysates were measured with a Meso Scale Discovery-based immunoassay as described previously 71 . (B) C. elegans, HEK293 cells, and human patient iPSC-derived neurons were homogenized in lysis buffer containing 150 mM NaCl, 25 mM Tris, pH 7.5, 1 mM EDTA, 1 mM EGTA,1% Triton-X 100, 0.5% SDS, and 1 × Halt TM Protease Inhibitor Cocktail (Thermo-Fisher Scientific) and sonicated at a 20% pulse rate for 15-20 s on ice. Lysates were centrifuged at 16,000 × g for 15-20 min at 4°C, and the supernatant was collected in a fresh tube. The protein concentration of the lysates was determined with the BCA protein assay (ThermoFisher Scientific). Lysates were diluted to the same concentration with lysis buffer and tested in duplicate wells using Meso Scale Discovery platform. Measurement of DPRs was carried out using a sandwich immunoassay as described earlier 72 .
Statistical analysis. Statistical analysis was performed by an unpaired t-test, oneway ANOVA with Dunnett's multiple comparisons test or Tukey's multiple comparison test, and two-way ANOVA with the Šídák multiple comparison test using GraphPad Prism version 8.2.1. P-values for the worm lifespan assay were calculated by the Mantel-Cox log-rank test using GraphPad Prism 8.2.1. A P-value of <0.05 was considered significant. The data are presented as mean ± standard error of the mean (SEM).