Publisher Correction: C9orf72 arginine-rich dipeptide repeats inhibit UPF1-mediated RNA decay via translational repression

Expansion of an intronic (GGGGCC)n repeat region within the C9orf72 gene is a main cause of familial amyotrophic lateral sclerosis and frontotemporal dementia (c9ALS/FTD). A hallmark of c9ALS/FTD is the accumulation of misprocessed RNAs, which are often targets of cellular RNA surveillance. Here, we show that RNA decay mechanisms involving upstream frameshift 1 (UPF1), including nonsense-mediated decay (NMD), are inhibited in c9ALS/FTD brains and in cultured cells expressing either of two arginine-rich dipeptide repeats (R-DPRs), poly(GR) and poly(PR). Mechanistically, although R-DPRs cause the recruitment of UPF1 to stress granules, stress granule formation is independent of NMD inhibition. Instead, NMD inhibition is primarily a result from global translational repression caused by R-DPRs. Overexpression of UPF1, but none of its NMD-deficient mutants, enhanced the survival of neurons treated by R-DPRs, suggesting that R-DPRs cause neurotoxicity in part by inhibiting cellular RNA surveillance. C9orf72 repeat expansion is the major genetic cause of familial amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). Here, the authors show that transcriptome aberrations commonly found in c9ALS/FTD are a result from defects in cellular RNA surveillance pathways that involve an RNA helicase UPF1.

S ince the identification of a GGGGCC (G 4 C 2 ) repeat expansion within the first intron of C9orf72 as the major cause of both familial ALS and FTD 1,2 , a variety of mechanisms, including haploinsufficiency 3 , RNA toxicity 4 , and dipeptide repeat (DPR) toxicity [5][6][7] , have been proposed to explain the pathogenicity of this autosomal dominant mutation. The expanded G 4 C 2 repeat region is transcribed in both directions, producing the C9orf72 pre-mRNA, which contains intronic G 4 C 2 repeats, and the antisense RNA that contains G 2 C 4 repeats 8 . Both RNAs accumulate in nuclear foci and, after being exported, can be translated into distinct sets of DPR-containing polypeptides 8 . When overexpressed, the two arginine-rich DPRs (R-DPRs), poly (GR) and poly(PR), have been shown to cause cell death both in vitro and in vivo through a variety of potential mechanisms, such as nucleolar dysfunction, nucleocytoplasmic transport defects, changes in stress granule dynamics, and translation inhibition.
In addition to C9orf72, genes encoding RNA-binding proteins such as FUS (fused in sarcoma) and TDP-43 (TAR DNA-binding protein 43) are also enriched in ALS/FTD-associated mutations, suggesting that RNA misprocessing may be a converging point in ALS/FTD pathophysiology 9,10 . Indeed, widespread RNA processing defects have been described by previous high-throughput RNA sequencing (RNA-seq) studies using post-mortem brain tissues from sporadic ALS (sALS) and c9ALS/FTD subjects 4,11 . In normal cells, misprocessed mRNAs are targeted for degradation by multiple RNA surveillance pathways [12][13][14] , such as nonsensemediated decay (NMD), no-go decay (NGD), and nonstopmediated decay (NSD). The accumulation of aberrant RNAs in c9ALS/FTD brains hints at the possibility that one or more mRNA surveillance pathways may be compromised.
Here we show that a main reason for the accumulation of aberrant RNAs in c9ALS/FTD brains is the global defect in UPF1-dependent RNA decay pathways including NMD. In cultured neurons, R-DPRs are sufficient to cause NMD inhibition by reducing global translation. Finally, ectopic expression of UPF1 protects neurons from R-DPR toxicity in an NMD-dependent manner. These results suggest a previously underappreciated role of cellular RNA surveillance in the pathophysiology of c9ALS/ FTD.

Results
UPF1-mediated RNA decay targets accumulate in c9ALS brains. We re-examined the post-mortem c9ALS, sALS, and control RNA-seq data 11 from the frontal cortex, which shows both RNA foci and DPR pathology in c9ALS/FTD. As shown previously 11 , intron retention events were prevalent among c9ALS subjects, as indicated by an increase in read densities for the majority of introns (Fig. 1a, red, median fold change: 2.1). In contrast, the prevalence of intron retention was overall similar between sALS and control subjects (Fig. 1a, blue, median fold change: 0.87). Intron-retaining mRNAs often contain premature stop codons, which would render them substrates for NMD. To assess whether NMD may be defective in c9ALS/FTD, we next quantified the abundance of regulatory targets of NMD. The overall abundance of a list of putative neuronal NMD targets (orthologs of mRNAs that are upregulated in Upf2 knockout mouse forebrain 15 ) (N = 275) was increased in c9ALS (Fig. 1b, left, red), but not in sALS subjects (Fig. 1b, right, red). Similar changes were observed when we used two other independent lists of NMD targets identified in human HeLa cells 16,17 (Fig. 1b, orange (N = 75) and yellow (N = 1271)), further supporting a global NMD deficit in c9ALS brains. In addition, we found that the mRNAs encoding known NMD factors, including UPF3B, SMG5, and SMG6, were significantly upregulated in c9ALS subjects (Fig. 1c), consistent with a compensatory feedback upon NMD inhibition 18 .
Canonical histone mRNAs lack polyA tails and instead have 3ʹ end stem-loop structures 19 . Histone mRNA decay requires UPF1, but not all NMD factors 19,20 . Similar to NMD targets, we observed overall accumulation of canonical histone mRNAs in c9ALS (Fig. 1d, left), but not in sALS subjects (Fig. 1d, right). In contrast, noncanonical, polyadenylated histone variant mRNAs were largely unchanged between c9ALS and controls (Supplementary Fig. 1), suggesting that accumulation is specific to the UPF1-dependent, canonical histone mRNAs.
To test whether the observed RNA decay deficits in postmortem brain may be recapitulated in c9ALS motor neurons, we compared the abundance of NMD targets between in vitro differentiated motor neurons from iPSCs derived from control, c9ALS, and SOD1 D90A -ALS subjects 21 . Again, we observed higher abundance of NMD targets and histone mRNAs in c9ALS but not SOD1 D90A -ALS motor neurons when compared to control motor neurons ( Supplementary Fig. 2).
Collectively, the accumulation of intron-retaining mRNAs, NMD regulatory targets, mRNAs encoding NMD factors, and canonical histone mRNAs suggest that UPF1-mediated RNA decay mechanisms are broadly inhibited in c9ALS/FTD.

R-DPRs acutely inhibit NMD in cultured cells. The C9orf72
repeat expansion can be transcribed in both directions, producing both sense G 4 C 2 repeat-and antisense G 2 C 4 repeat-containing RNAs 8 , which can be translated into five DPRs: poly(GA), poly (GP), poly(GR), poly(PA), and poly(PR). To test whether any of these gene products may be sufficient to inhibit NMD, we ectopically expressed each of the two repeat RNAs as part of the 5ʹ untranslated regions (UTRs) of a GFP transcript, and each of four codon-optimized DPRs as GFP fusion proteins 22 in HEK293 cells, and quantified the expression levels of several UPF1mediated decay targets that accumulated in c9ALS subjects, including four NMD target mRNAs and a canonical histone mRNA. We did not include the poly(GP) expression construct in our analysis because it could not be validated by Sanger sequencing, presumably due to the exceedingly high structureforming potential of (GGNCCN) n sequence. DPR-GFP fusion proteins were expressed at similar levels ( Supplementary Fig. 3). Within 24 h after transfection, each of the two arginine-rich DPRs (R-DPRs), poly(GR) and poly(PR), increased the abundance of all five tested UPF1 targets (Fig. 2a), whereas no significant changes in target abundance were detected in cells expressing either of the two repeat RNAs nor the two alanine-rich DPRs (Fig. 2a).
To confirm that the selected transcripts are NMD targets in neurons, we treated induced human neurons (iNeurons) with each of two distinct NMD antagonists: caffeine, which nonspecifically inhibits UPF1 phosphorylation by SMG1, and cycloheximide (CHX), which inhibits translation elongation. After 24 h, both caffeine and CHX treatments caused significant increases in abundance of the selected transcripts in iNeurons ( Supplementary Fig. 4), suggesting that these transcripts are indeed regulated by NMD in human neurons.
To validate the effect of R-DPRs on NMD on a transcriptome scale, we re-examined RNA-seq data from K562 leukemia cells treated with synthetic PR 20 peptides 23 . We observed substantial accumulation of both NMD targets (Fig. 2b, red, orange, and yellow) and canonical histone mRNAs (Fig. 2b, purple). In addition, multiple NMD factor mRNAs were upregulated in PR 20 -treated K562 cells (Fig. 2c).
To examine the effect of R-DPRs in cell types that are more relevant to FTD, we treated mouse primary cortical neurons with synthetic PR 20 peptides, which are readily taken up by cells and cause neuronal cell death 23 . Within 24 h, PR 20 caused the accumulation of mouse mRNAs orthologous to the tested human UPF1 targets (Fig. 2d). Similarly, 24 h of PR 20 treatment also caused the accumulation of UPF1 targets in human iNeurons (Fig. 2e). Collectively, our results suggest that R-DPRs are sufficient to inhibit NMD in both neuronal and non-neuronal cell types.
R-DPRs cause the recruitment of UPF1 to stress granules. Previous studies have linked the cellular toxicity of R-DPRs to their influence on stress granule dynamics [24][25][26] . To test whether NMD inhibition may be linked to stress granule formation, we examined the localization of UPF1 and G3BP1, a stress granule marker, in HeLa cells expressing each repeat RNA or DPR. In control cells, UPF1 exhibited diffuse cytoplasmic localization, with G3BP1-negative small foci that resembled P-bodies (Fig. 3a). In contrast to previous studies 27, 28 , we did not observe a significant increase in stress granules in cells expressing either G 4 C 2 or G 2 C 4 repeat RNAs (Fig. 3a, b), suggesting that stress granule formation may not be a direct consequence of repeat RNA expression. In contrast, poly(GR) and poly(PR) induced the concentration of UPF1 in G3BP1-positive stress granules in 45 and 24% of transfected cells, respectively (Fig. 3a, b). To confirm that UPF1 is a constitutive component of stress granules, we induced stress granules by applying oxidative stress using sodium arsenite (NaAsO 2 ). Indeed, UPF1 and, to a lesser extent, UPF3B, were both recruited to arsenite-induced stress granules (Supplementary Fig. 5). These results are consistent with previous studies on the influence of R-DPRs on stress granule assembly [24][25][26] and the presence of UPF1 in stress granules 29 , raising the possibility that NMD inhibition may be due to the sequestration of UPF1 and other NMD factors in stress granules.
To test whether DPRs can also induce stress granules in neurons, we expressed each DPR in mouse primary cortical neurons. Compared to those in cell lines, Upf1-positive stress granules were exceedingly rare in neurons ( Supplementary Fig. 6). We detected Upf1-positive, Ataxin-2-positive stress granules only in a small fraction (<5%) of poly(GR)-expressing neurons ( Supplementary Fig. 6), consistent with a recent study showing that poly(GR) co-localizes with stress granule-like inclusions in (G 4 C 2 ) 149 -expressing mouse brain 30 . These results suggest either that the neuronal cytoplasm may be less permissive to stress granule formation, or that neuronal stress granules may exhibit distinct morphology and/or dynamics from those in cell lines.
Stress granules are not required for NMD inhibition. To directly assess the role of stress granule formation in DPRinduced NMD inhibition, we expressed GFP, poly(GR), or poly (PR) in wild-type (WT) or G3BP1/2 double-knockout (G3BP-DKO) U2OS cells 31 (Supplementary Fig. 7a). Consistent with previous studies 24, 25 , while R-DPRs were expressed at similar levels between G3BP-WT and -DKO cells ( Supplementary  Fig. 7b), R-DPR-induced stress granule formation was substantially reduced in G3BP-DKO cells (Fig. 3c, Supplementary  Fig. 7c). Despite the large reduction in stress granules in G3BP-DKO cells, R-DPRs still increased the abundance of NMD targets (Fig. 3d), suggesting that stress granule formation is not required for the inhibitory effect of R-DPRs on NMD.  In an orthogonal approach to reduce stress granule formation, we treated HEK293 cells with ISRIB (integrated stress response inhibitor), which has been shown to block the downstream effects of eIF2α phosphorylation including stress granule formation 32 . As expected, ISRIB treatment partially reduced stress granules in poly(PR)-expressing cells (Supplementary Fig. 8a). However, the increase in abundance of NMD targets was largely unchanged (Supplementary Fig. 8b). These results, together with those from G3BP-DKO cells, indicate that recruitment of UPF1 to stress granules is not the cause of NMD inhibition.
R-DPRs inhibit NMD via translational repression. G 4 C 2 repeats and R-DPRs have also been associated with nucleocytoplasmic transport defects 33 . If nuclear export of RNA is delayed, both NMD targets and non-targets would be expected accumulate in the nucleus. We measured the relative abundance of UPF1 targets in the nuclear and cytoplasmic fractions, and found that both poly(GR) and poly(PR) caused UPF1 targets to accumulate predominantly in the cytoplasm (Fig. 4a). Therefore, nuclear RNA retention is not the cause of the accumulation of NMD targets either. In addition, R-DPR expression did not lead to significant changes in UPF1 phosphorylation levels ( Supplementary Fig. 9). R-DPRs have previously been shown to inhibit translation both in vitro 34,35 and in vivo 36 . Because NMD is translationdependent, global translational repression would be expected to inhibit NMD. We confirmed that poly(GR) and poly(PR) substantially inhibited global translation in HEK293 cells, as indicated by reduced puromycin incorporation into nascent peptides (Fig. 4b). To measure R-DPR-induced translation repression more quantitatively, we treated primary cortical neurons expressing firefly luciferase with several different concentrations of PR 20 , and after 24 h, measured the ratio between luciferase activity and luciferase mRNA abundance. Indeed, significant translational repression was observed starting from the lowest PR 20 concentration (1 μM) (Fig. 4c).
Using this quantitative assay, we assessed the extent to which translational repression could explain the inhibitory effect of R-DPRs on NMD. We treated primary cortical neurons with either PR 20 or CHX at several different concentrations. After 24 h, we measured for each sample, the degree of translational repression and the increase in NMD target abundance (Fig. 4d). The relationship between the changes in NMD target abundance and the degree of translational repression was fitted by log-log models. For each of the three tested NMD targets, PR 20 -induced NMD target accumulation was almost fully (92−122%) explained by translation repression (Fig. 4d). Therefore, the observed NMD inhibition is primarily a consequence of global translational repression by R-DPRs.
UPF1 overexpression protects neurons from R-DPR toxicity. Consistent with the notion that R-DPRs cause neurodegeneration a (G 4 C 2 ) 100 (G 2 C 4 ) 100 GA 50 -GFP GR 50 -GFP PA 50 -GFP PR 50  in part through inhibiting NMD, recent studies have shown that overexpression of UPF1 and, to a lesser extent, UPF2 could reduce R-DPR neurotoxicity in flies 37,38 . To determine whether the UPF1 is also protective in mammalian neurons, we overexpressed UPF1 in primary cortical neurons before treating them with 2 μM PR 20 . Consistent with previous studies 23 , PR 20 caused cell death in >80% neurons within 48 h (Fig. 5). Compared to the control group, overexpression of wild-type UPF1 significantly increased neuronal survival by an average of 71% (Fig. 5). UPF1 have multiple functions beyond NMD 20 . To determine whether the neuroprotective effect of UPF1 requires its NMD activity, we tested a variety of NMD-deficient UPF1 mutants 39 , including C126S (deficient UPF2 binding), R854C (deficient helicase activity), G506R/G508E (GGRE, deficient ATPase/helicase activity), and S1084A/S1089A/S1100A/S1107A (4SA, lacking four phosphorylation sites 40 ). Using an NMD reporter assay, we confirmed that only wild-type UPF1, but none of the four tested mutants could rescue NMD activity in UPF1-deficient cells ( Supplementary Fig. 10). Correspondingly, none of the four NMD-deficient UPF1 mutants increased survival of PR 20 -treated neurons (Fig. 5), suggesting that the neuroprotective effect of UPF1 requires its NMD function.

Discussion
Emerging evidence suggests that widespread transcriptomic aberration is a distinguishing feature of c9ALS/FTD. Previous studies have largely attributed it to the deficiency of splicing factors (e.g., hnRNP H), leading to increased production rates of misprocessed RNAs 4,11 . Our results suggest that decreased degradation rates resulting from NMD inhibition also causes the accumulation of potentially deleterious RNAs in c9ALS/FTD.  Fig. 4 Translation repression is the primary cause of poly(PR)-induced NMD inhibition. a Nuclear fractions of NMD targets, U1 snRNA, and GAPDH mRNA (cytoplasmic) in HEK293 cells expressing GFP, poly(GR), or poly(PR). n = 3 independent experiments. Data are presented as mean values ± SD. *P < 0.05; **P < 0.01, two-sided unpaired t tests. b Puromycin-labeled nascent peptides in HEK293 cells expressing GFP, poly(GR), or poly(PR). +CHX, CHX added 1 h before puromycin labeling. Similar results were obtained from three independent experiments. c Global translation levels in primary neurons treated with PR 20 for 24 h, quantified by the activity of a transfected luciferase reporter, normalized to mRNA levels and untreated control. n = 3 independent experiments. Data are presented as mean values ± SD. *P < 0.05; **P < 0.01; ***P < 0.001, two-sided ratio t tests. d Comparisons between the effects of PR 20 (red) and CHX (blue) on NMD targets. Degrees of translation repression and NMD target abundance were quantified in primary neurons treated with increasing concentrations of PR 20 or CHX. Zero-intercept linear functions were fit to normalized, log 2 -transformed values. Source data and exact P values are provided in the Source Data file.
While intron-specific splicing defects may also exist, the observed wide spectrum of increased intron retention in c9ALS subjects suggests that a more global defect in the degradation of misprocessed RNAs via NMD may underlie this phenomenon.
Aside from targeting misprocessed RNAs such as intronretaining mRNAs (i.e., quality control), a separate function of NMD is lowering the abundance of many endogenous, properly processed RNAs (i.e., quantity control). We found in c9ALS subjects an overall upregulation of numerous NMD target mRNAs as well as many noncoding RNAs including pseudogene transcripts and the splice variants of small nucleolar RNA precursors 41 . Since many NMD factors including UPF1 are conserved in all eukaryotes and required for survival in most metazoans [42][43][44] , chronic NMD inhibition would likely be detrimental to neuronal viability 13 . Indeed, loss-of-function mutations in both UPF2 and UPF3B have associated with intellectual disabilities 15,45 , suggesting a heightened vulnerability of the nervous system to NMD deficiencies.
Our analysis showed that inhibition of RNA decay is not limited to NMD, but also affects other UPF1 targets such as canonical histone mRNAs 19,20 . The multiple functions of UPF1 beyond NMD may in part explain the recent observation that ectopic expression of UPF1 shows stronger neuroprotective effect in R-DPR fly models than that of UPF2, an NMD-specific factor 37 . Notably, aberrations in heterochromatin have been recently linked to poly(PR) expression in mouse models 46 . It will be interesting to determine whether the accumulated histone mRNAs are translated, and how they may impact histone turnover, chromatin modifications, and neuronal survival in c9ALS/FTD.
Ectopically expressed R-DPRs are sufficient to recapitulate NMD inhibition in cultured cells, suggesting that in c9ALS/FTD patients, NMD inhibition may be the consequence of low levels of poly(GR) and/or poly(PR) accumulation over a long period. Considering that R-DPRs are strong stress granule inducers and that UPF1 is a known stress granule component, it is tempting to believe that these DPRs inhibit NMD by sequestering UPF1 in stress granules and away from P-bodies and the rest of cytoplasm, where NMD may occur more efficiently. However, we obtained multiple lines of evidence arguing against this hypothesis: (i) In cell lines, poly(GR) promoted stress granule assembly more strongly than poly(PR), but these two R-DPRs inhibited NMD to similar degrees. (ii) R-DPR expression rarely caused stress granule assembly in neurons. (iii) Blocking stress granule assembly by deleting G3BP1/2 did not alleviate NMD inhibition. (iv) ISRIB treatment, which reduced poly(PR)-induced stress granules, had no effect on NMD target accumulation. These results collectively reject the hypothesis that stress granule assembly is a main contributor to NMD inhibition. Instead, stress granule formation and NMD inhibition appear to be two independent consequences of R-DPRs.
Previous studies have shown that R-DPRs cause global reduction in translation 36,47 . Considering that NMD requires translation to distinguish targets from non-targets, we reasoned that translation repression must at least in part contribute to R-DPR-induced NMD inhibition. From the dose-response relationships between NMD target abundance and translational repression caused by either PR 20 or CHX, we found that the effect of PR 20 on NMD is almost fully accounted for by its effect on translation. Therefore, at least for the several tested targets, translational repression appears to be the primary, if not exclusive, cause of NMD inhibition by R-DPRs. It remains to be tested whether this may be true at a transcriptome-wide level, and to what extent some NMD targets may be more sensitive to R-DPRs than others.
Although the mechanism of non-AUG translation producing DPRs in c9ALS/FTD remains unclear, poly(GA), part of poly (GP), and poly(GR) are potentially translated from the sense C9orf72 mRNA retaining its first intron, which itself may be an NMD target 38 . Therefore, NMD inhibition would in turn allow this intron-retaining mRNA to further accumulate, producing more DPRs. This vicious circle would presumably accelerate the accumulation of more deleterious RNAs and proteins in cells and ultimately cause cell death.
Recent studies has shown that ectopic expression of NMD factors can alleviate poly(GR)-and poly(PR)-induced toxicity in SH-SY5Y neuroblastoma cell line and in flies 38,48 . Our results extended these findings to mammalian neurons and provided new mechanistic insight into the neuroprotective effect of UPF1. Indeed, wild-type UPF1 significantly enhanced neuronal survival in the presence of synthetic PR 20 peptides. However, none of the NMD-deficient UPF1 mutants, including UPF1 C126S , which is deficient in UPF2 binding, could enhance neuronal survival. While additional contributions from the NMD-independent functions of UPF1 may exist, these results suggest a major role of NMD in the neuroprotective functions of UPF1. Considering that UPF1 and NMD have now been implicated in multiple ALS/ FTD subtypes [48][49][50] , restoring UPF1 and NMD activity to a physiological level may represent a new therapeutic approach.

Methods
RNA-seq data analysis. FASTQ files of previously published RNA-seq datasets were downloaded from European Nucleotide Archive (ENA), and uniquely mapped to the human (GRCh38) or mouse (GRCm38) reference genome using STAR. Exon and intron read densities for all annotated genes were quantified using   (1:2000) in 5% milk/PBST at 4°C with slow shaking overnight. After incubation, membranes were rinsed three times with PBST, and incubated with IR680-or IR800-conjugated secondary antibodies (Li-Cor) diluted (1:10,000) in 5% milk/PBST at room temperature for 1 h. After three rinses with PBST, membranes were imaged using an Odyssey CLx system (Li-Cor).
Quantitation of translational repression. E15 mouse primary cortical neurons were transiently transfected with a plasmid expressing firefly luciferase. After 6-8 h, DMSO, CHX (0.0075~0.24 μg/mL), or PR 20 (0.5~8 μM) was added to media. Cells were lysed after 24 h, with half of the lysate used for luciferase assays (Promega) and the half for RNA extraction. Luciferase mRNA and NMD target abundance were measured by qRT-PCR. Translational activity was first calculated as the ratio between luciferase activity and luciferase mRNA abundance, then normalized to the DMSO control sample.
Neuronal viability analysis. Primary cortical neurons were transfected using Lipofectamine 2000 (Invitrogen) with a 1:9 ratio of an mApple-expressing plasmid (Addgene #54567) and an expression construct for BFP only, wild-type or mutant UPF1-BFP. 48 h after transfection, PR 20 peptides were added to the media at a final concentration of 2 μM. Neurons were imaged 0 or 48 h after PR 20 treatment on a BioTek Lionheart FX automated microscope with a ×4 objective and quantified using the Gen5 image analysis software.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support this study are available from the corresponding author upon reasonable request. RNA-seq data are available through Gene Expression Omnibus

Code availability
In-house shell scripts and R codes used for analyzing RNA-seq data are available upon request.