C9orf72 Dipeptide Repeats Inhibit UPF1-Mediated RNA Decay Independent of Stress Granule Formation

Expansion of an intronic (GGGGCC)n repeat region within the C9orf72 gene is a major cause of familial amyotrophic lateral sclerosis and frontotemporal dementia (c9ALS/FTD). A pathological hallmark in c9ALS/FTD is the accumulation of misprocessed RNAs, which are often targets of RNA surveillance pathways in normal cells. Here we show that nonsense-mediated decay (NMD) and other RNA decay mechanisms involving upstream frameshift 1 (UPF1), collectively referred to as UPF1-mediated RNA decay (UMD), are broadly inhibited in c9ALS/FTD brains. These effects are recapitulated in cultured cells by the ectopic expression of arginine-rich dipeptide repeats (DPRs), poly(GR) and poly(PR). Despite these two DPRs causing the recruitment of UPF1 to stress granules, stress granule formation is neither sufficient nor necessary for UMD inhibition. Our results suggest that UMD inhibition may accelerate the accumulation of deleterious RNAs and polypeptides in c9ALS/FTD.


INTRODUCTION
Since the identification of a GGGGCC (G4C2) 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 G4C2 repeat region is transcribed in both directions, producing the C9orf72 pre-mRNA, which contains intronic G4C2 repeats, and the antisense RNA that contains G2C4 repeats 8 . Both RNAs accumulate in nuclear foci and, once exported, can be translated into distinct sets of DPRcontaining polypeptides 8 .
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 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 both sporadic ALS (sALS) and c9ALS/FTD patients 4,11 . In normal cells, misprocessed mRNAs are targeted for degradation by multiple RNA surveillance pathways [12][13][14] , such as nonsense-mediated decay (NMD), no-go decay (NGD), and nonstop-mediated 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.

UPF1-mediated RNA decay targets accumulate in c9ALS brains
We began by re-examining 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, intron retention events were prevalent among c9ALS subjects, as indicated by a right-shift in the cumulative distribution function (CDF) of the fraction of reads mapped to the introns of each gene ( Figure 1A, red curve), but not in sALS subjects ( Figure 1A, blue curve).
Intron-retaining mRNAs typically contain premature stop codons, which would render them substrates for NMD. To assess whether NMD may be inactivated in c9ALS/FTD, we next quantified the expression of endogenous regulatory targets of NMD. When we compared the abundance of previously defined (N=76) endogenous NMD targets 15 between ALS and controls, we found that they were broadly up-regulated in c9ALS ( Figure 1B, compare red and black curves), but not sALS subjects ( Figure 1B, compare blue and black curves). These observed changes were replicated when we used an expanded list (N=1,271) of NMD targets 16 (Figure 1B, orange and light blue curves). In addition, we found that the mRNAs for key NMD factors, UPF3B, SMG5, and SMG6 were significantly up-regulated in c9ALS subjects ( Figure 1C), consistent with the auto-regulatory feedback upon NMD inhibition 17 .
Canonical histone mRNAs lack polyA tails and instead have 3ʹ end stem-loop structures 18 .
Histone mRNA decay requires UPF1, but not other NMD factors 18,19 . Similar to NMD targets, we observed an overall up-regulation of canonical histone mRNAs in c9ALS, but not in sALS subjects ( Figure 1D). In contrast, noncanonical, polyadenylated histone variant mRNAs were largely unchanged between c9ALS and controls (Figure 1−figure supplement 1), suggesting that accumulation is specific to those UPF1-dependent, canonical histone mRNAs.
Collectively, the accumulation of intron-retaining mRNAs, NMD regulatory targets, NMD factor mRNAs, and canonical histone mRNAs suggest that multiple UPF1-mediated RNA decay mechanisms, hereafter collectively referred to as UMD, are broadly inactivated in c9ALS/FTD.

Arginine-rich DPRs acutely inhibit UMD in cultured cells
The C9orf72 repeat expansion can be transcribed in both directions, producing both sense G4C2 repeat-and antisense G2C4 repeat-containing RNAs 8 , which can be translated into a total of 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 UMD, we ectopically expressed each of the 2 repeat RNAs as part of the 5ʹ untranslated regions (UTRs) of a GFP transcript, and each of the 4 codonoptimized DPRs as GFP fusion proteins 20 in HEK293 cells, and quantified the expression levels of several UMD targets that accumulated in c9ALS subjects, including three NMD targets, two NMD factor mRNAs, and a canonical histone mRNA. We failed to validate the codon-optimized poly(GP) expression construct, and excluded it from our analysis. Interestingly, poly(GR) and poly(PR) increased the expression levels of six and five out of six tested UMD targets, respectively, within 24 hours after transfection (Figure 2A), whereas no significant changes in UMD target expression were detected in cells expressing repeat RNAs nor the two alanine-rich DPRs (Figure 2A), To validate this finding on a transcriptome scale, we examined the previously published RNAseq data from K562 leukemia cells treated with synthetic PR20 peptides 21 . Again, we observed a substantial upregulation of both NMD targets ( Figure 2B, red and orange curves) and canonical histone mRNAs ( Figure 2B, purple curves). Multiple NMD factors were up-regulated ( Figure  2C). Collectively, these results suggest that poly(GR) and poly(PR) are sufficient to cause acute UMD inhibition in cultured cells.

UMD inhibition is independent of stress granule recruitment of UPF1
Previous studies have linked the cellular toxicity of poly(GR) and poly(PR) to their influence on stress granule dynamics [22][23][24] . To test whether UMD 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 ( Figure 3A). In contrast to previous studies 25, 26 , we did not observe a significant increase in stress granules in cells that expressed either G4C2 or G2C4 repeat RNAs ( Figure 3A  expressing mouse brain 28 . 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. To directly test the impact of stress granule assembly on UMD, we induced stress granules by applying oxidative stress or osmotic stress using sodium arsenite (NaAsO2) or sorbitol, respectively. We confirmed that UPF1 and, to a lesser extent, UPF3B, were both recruited to arsenite-induced stress granules ( unchanged and in some cases (e.g. UPF3B) even slightly increased ( Figure 3E). These results indicate that stress granule assembly is neither sufficient nor required for DPR-induced UMD inhibition, rejecting the notion that stress granule assembly may contribute to UMD inhibition 30 .

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 aberrant RNAs 4,11 . Our results suggest that decreased degradation rates resulting from NMD inhibition also contribute to the accumulation of potentially deleterious RNAs in c9ALS/FTD.
Aside from targeting misprocessed RNAs such as intron-retaining mRNAs, a separate function of NMD is lowering the abundance of many endogenous, properly processed RNAs. 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 31 (data not shown). Since many NMD factors including UPF1 are conserved in all eukaryotes and required for survival in most metazoans [32][33][34] , chronic NMD inhibition would likely be detrimental to neuronal viability 13 .
Our analysis shows that inhibition of RNA decay is not limited to NMD, but also affects other UPF1-dependent decay substrates such as canonical histone mRNAs 18,19 . Notably, aberrations in heterochromatin have been recently linked to poly(PR) expression in mouse models 35 . It remains to be determined whether the accumulated histone mRNAs are translated, and how they may impact chromatin modifications and neuronal survival in c9ALS/FTD. Ectopically expressed arginine-rich DPRs are sufficient to recapitulate UMD inhibition in cultured cells, suggesting that in c9ALS/FTD patients UMD inhibition may result from low levels of poly(GR) and/or poly(PR) accumulation over a long period. Considering that argininerich DPRs are strong stress granule inducers and that UPF1 is a known stress granule component, it is tempting to hypothesize that these DPRs inhibit UMD by sequestering UPF1 in stress granules and away from P-bodies and the rest of cytoplasm, where UMD can occur efficiently.
However, we present 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 DPRs inhibited UMD to similar degrees. (ii) DPR expression rarely caused stress granule assembly in neurons. (iii) Although chemically induced stress granules also recruited UPF1, they did not broadly inhibit UMD. (iv) Blocking stress granule assembly by deleting G3BP1/2 did not alleviate UMD inhibition. These results collectively reject the notion that stress granule assembly contributes to UMD inhibition. We speculate that UMD inhibition may result from the inhibitory effect of arginine-rich DPRs on global translation 36,37 . Without translation, NMD targets would not be distinguishable from non-targets. It remains to be tested whether the inhibition of other UMD mechanisms may result secondarily from translational arrest-induced NMD inhibition.
A recent study has reported that ectopic expression of NMD factors can alleviate poly(GR)-and poly(PR)-induced toxicity in SH-SY5Y neuroblastoma cell line and in flies 30 . In that study, overexpression of UPF1 shows stronger suppression of toxicity than that of UPF2, consistent with our finding that RNA decay pathways other than NMD are also involved. Considering that UPF1 and UMD have now been implicated in multiple ALS/FTD models 30,38,39 , restoring UMD activity to a normal level may represent a promising therapeutic approach.

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/intron reads for all annotated genes were quantified using BEDTools. After excluding genes with <1 read/sample on average, reads per million uniquely Expression changes of each gene set were compared to all genes using one-sided Mann-Whitney-Wilcoxon tests.

RT-qPCR analysis
Expression constructs for repeat RNAs or GFP-tagged codon-optimized DPRs were transfected in HEK293 cells or U-2 OS cells with Lipofectamine 2000 (Invitrogen) or FuGENE HD (Promega), respectively, according to the manufacturers' instructions. 24−48 hrs after transfection, total RNA was extracted with TRIzol and treated with Turbo DNase (Invitrogen).
RT-qPCR was performed on a CFX96 RT-PCR system (Bio-Rad) with Luna Universal One-Step RT-qPCR reagents (New England Biolabs). 50 ng total RNA and 250 nM primers were added in each reaction. Ct values were averaged across 2 technical replicates and normalized to GAPDH internal controls. 3 biological replicates were performed. Sun,Eshov & Guo,page 11 Transfected HEK293 cells, HeLa cells, U-2 OS cells, or E18 mouse cortical neurons on coverslips were fixed at room temperature with 4% paraformaldehyde in PBS, washed three times with PBS, and permeabilized with 0.5% Triton X-100. After washing and 1 hr blocking with 10% goat serum, cells were incubated with primary antibodies (rabbit anti-UPF1, Cell