Upf1 regulates neurite outgrowth and branching by transcriptional and post-transcriptional modulation of Arc Free

Activity-dependent changes in synaptic strength, as observed in long-term potentiation (LTP) and long-term depression (LTD), are considered to be the leading cellular mechanisms underpinning learning and memory (Bliss and Collingridge, 1993). Neurons can coordinate activity-dependent processes through the activation of new gene transcription, and the protein synthesis underlying that synaptic plasticity is critically controlled at the level of mRNA translation (Klann and Dever, 2004). Alteration in the expression of immediate early genes (IEGs) is the first genetic response to a variety of cellular stimuli (Lanahan and Worley, 1998). The somatodendritic protein activity-regulated cytoskeleton-associated protein (Arc; also known as Arg3.1), is one such IEG, and has been the focus of several studies that have shown it to be involved in activity-dependent synaptic plasticity, including both LTP and LTD (Bramham et al., 2010, 2008; Plath et al., 2006). In Arc knockout (KO) mice, early-phase LTP is enhanced, whereas late-phase LTP is blocked, and acute knockdown of Arc also results in LTP disruption (Messaoudi et al., 2007). Furthermore, in hippocampal slices from Arc KO mice, mGluR (also known as Grm) protein-dependent LTD is suppressed, and Arc KO neurons fail to decrease surface expression of the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) after dihydroxyphenylglycine application (Park et al., 2008).

A previous study has established a critical role for nonsense-mediated mRNA decay (NMD) in regulation of Arc expression, and this suggests that this pathway could function to prevent improper Arc protein synthesis at undesired times and/or locations (Giorgi et al., 2007). Seemingly confirming the functional impact of NMD, knockdown of the exon junction complex (EJC) core protein, eIF4AIII, enhances Arc mRNA and protein expression. However, unexpectedly, this knockdown also upregulates AMPAR levels at synapses and selectively increases the amplitude of miniature excitatory postsynaptic currents, even though Arc levels are elevated. It is quite a contradictory response because enhanced levels of Arc trigger reduction of surface AMPAR at synapses. Thus, these data suggest that post-transcriptional regulation by NMD is insufficient to fully explain the role of Arc gene regulation in synaptic plasticity, and, therefore, gaining understanding of additional mechanisms for maintaining the tight regulation of Arc gene expression will be critical for elucidating the precise role of Arc expression in neuronal cells.

Here, we investigated the role of Upf1, an essential component of the NMD pathway, in Arc gene regulation to better understand how NMD regulates Arc mRNA stability and expression. Interestingly, we found that Upf1 functions at both the transcriptional and translational level to inhibit Arc gene expression. Upf1 suppresses Arc transcription by downregulating transcription factors that promote Arc expression, including Mef2a. Moreover, we show that Upf1 also binds to the Arc 3′ untranslated region (UTR) and suppresses translation. Intriguingly, however, Arc mRNA can escape from the NMD pathway through the action of Ago2 (also known as miRISC), which binds to the 3′UTR and further represses Arc translation, thus inhibiting NMD. Collectively, these data imply that Upf1 is crucial for maintaining low expression levels of Arc under normal conditions, which likely contribute to rapid and selective Arc induction in response to vigorous neuronal activity.

A simultaneous analysis of the pre-mRNA and mRNA abundance of selected transcripts found that only a minority of UPF1-dependent genes are directly regulated by this protein (Viegas et al., 2007). Tani et al. (2012) additionally measured the genome-wide steady-state abundance and decay rates of mRNAs using RNA sequencing (RNA-seq) and 5′-bromouridine immunoprecipitation chase sequencing (BRIC-seq), and identified indirect UPF1 targets with >2-fold upregulation, but unaltered decay rates, in UPF1-depleted HeLa cells. These results suggest that many genes are regulated by UPF1 at the transcriptional level, independent of NMD. Therefore, in order to determine whether Arc transcription is similarly regulated by Upf1, we measured steady-state Arc mRNA levels, using primer sets specific to the mature and premature transcripts, as illustrated in Fig. 1A. Mouse Neuro-2a (N2a) cells were transfected with a short hairpin RNA (shRNA) that targets Upf1, and we first confirmed that this results in a decreased level of Upf1 protein (Fig. 1B). We then found that levels of both the mature and premature Arc mRNA were significantly increased in Upf1 knockdown cells (Fig. 1C,D).

The promoters and enhancers that contribute to activity-dependent Arc expression have been well characterized (Kawashima et al., 2009). For example, the rapid transcription of Arc as an immediate early gene (IEG) is dependent on a critical synaptic activity-responsive element (SARE), which was identified as a unique, ∼100 bp element, located >5 kb upstream of the Arc transcription initiation site in the mouse genome. Here, we generated an Arc-luciferase reporter vector containing a SARE sequence fused directly upstream of ArcMin, a TATA-containing sequence near the transcription initiation site of the Arc gene (Fig. 1E). This distally located SARE-promoter fusion was found to be sufficient to replicate crucial properties of endogenous Arc transcriptional regulation (Fig. 1F). Moreover, to investigate the possibility of transcriptional regulation by Upf1, we then measured Arc reporter activity in Upf1 knockdown cells. We found that Arc promoter activation was increased in cells with shRNA-mediated Upf1 knockdown, as evidenced by a ∼2-fold increase in luciferase activity (Fig. 1G). Together, these results suggest that Upf1 exerts a negative regulatory effect on Arc transcription.

Next, we employed the CRISPR/Cas9 system to develop a method for efficient KO of Upf1. A single-guide RNA (sgRNA) was designed to target the Cas9 endonuclease to sites within exon 1 of the Upf1 gene (Fig. S1A). We tested several cell lines containing a homozygous deletion in Cas9+our sgRNA, and found two, #3 and #13, with decreased levels of Upf1. Of these, only the #3 line showed complete Upf1 KO with no residual expression, as well as strong induction of Arc protein levels; we therefore used the Cas9+sgRNA #3 cell line (referred to as KO cells) for all studies past this point (Fig. S1B). We then validated efficient KO of Upf1 using different antibodies (Fig. S1C) and dual-color fluorescent protein-based reporters (Pereverzev et al., 2015) for quantification of NMD activity using fluorescence microscopy (Fig. S2D,E). We also performed a rescue experiment by re-expressing Upf1 in Upf1-depleted cells to ensure proper selection (Fig. S2F). Consistent with our results from experiments employing transient shRNA-mediated Upf1 knockdown, the levels of both mature and premature Arc mRNA were found to be significantly increased in Upf1 KO cells (Fig. 2A–D). Also, Arc promoter activity was higher in KO cells than in wild-type (WT) cells (Fig. 2E,F).

The transcription factor myocyte enhancer factor 2a (Mef2a) has been found to suppress the number of excitatory synapses via induction of its target genes, Arc and synaptic Ras-GTP-activating protein 1 (SynGAP1) (Flavell et al., 2006). We therefore investigated the potential role of Upf1 in Mef2a mRNA regulation. Our data showed that the level of mature Mef2a mRNA was significantly higher in Upf1 KO cells; however, Mef2a pre-mRNA levels were unaffected (Fig. 2G,H). Based on these results, we hypothesized that Upf1 regulates Mef2a transcript stability. We conducted an in vitro binding assay using in vitro transcribed, biotin-conjugated Mef2a 3′UTR with N2a cell lysate. Fig. 2I shows that Upf1 binds to Mef2a 3′UTR in vitro. To determine whether Upf1 is associated with Mef2a mRNA in vivo, we performed RNA immunoprecipitation (RNA-IP) followed by quantitative RT-PCR (qRT-PCR). Consistent with our hypothesis, Mef2a transcripts were found to be enriched in Upf1-specific RNA immunoprecipitates, compared with IgG controls (Fig. 2J). Moreover, Mef2a protein levels were significantly increased in the Upf1 KO cells (Fig. 2K,L). These data suggest that Upf1 triggers NMD of the Mef2a transcript, resulting in the downregulation of Mef2a expression. Thus, our results imply that Upf1 suppresses Arc expression, at least in part, by enhancing destabilization of Mef2a mRNAs, resulting in decreased Mef2a protein levels in neuronal cells.

Because Arc has a long 3′UTR that contains two introns, it is thought that Arc mRNA is a natural target of NMD (Giorgi et al., 2007). Consistent with this, RNA interference (RNAi)-mediated knockdown of the NMD regulator Upf1 increases the level of Arc mRNA and protein in PC12 cells. To independently confirm whether Upf1 is involved in the regulation of Arc mRNA stability, we measured the half-life of endogenous Arc transcript in Upf1 knockdown N2a cells. Notably, despite an almost 3-fold increase in Arc mRNA steady-state levels (Fig. 1C) in cells with the Upf1 knockdown, the measured difference in actual mRNA stability in knockdown versus control cells was not statistically significant (Fig. S2A). This suggests that Upf1 affects this particular target mostly at the level of transcription, rather than mRNA stability. To rule out the possibility that our results were influenced by low residual levels of the Upf1 protein, we used Upf1 KO cells obtained with the CRISPR/Cas9 system. Consistent with previous results, the Arc mRNA half-lives in Upf1 KO and WT cells were found to be approximately equal (Fig. 3A). An additional test of mRNA stability involving qRT-PCR with specific primers targeting the Arc pre-mRNA also did not show any significant differences in mRNA decay in Upf1 KO versus WT cells (Fig. S2B).

We then developed a luciferase reporter assay based on the Arc 3′UTR to determine whether the NMD surveillance pathway functions normally to control Arc gene expression. We built and tested two different Arc 3′UTR reporter constructs. In the first design, we inserted the Arc 3′UTR derived from genomic DNA (Arc 3′U-G), which contains two short introns, downstream of Renilla luciferase of psiCHECK2 vector (Fig. 3B). The EJC binds to this Arc 3′U-G after splicing occurs, and that spliced mRNA might be susceptible to degradation by NMD. We observed decreased luciferase mRNA levels from the reporter containing Arc 3′U-G, compared with the psiCHECK2 mock vector (Fig. 3C), but, surprisingly, we also found decreased reporter mRNA levels in Upf1 KO compared with WT cells (Fig. 3F). These results suggest the possibility that Arc mRNA becomes unstable through other novel mechanisms. To test this hypothesis, we inserted the complementary DNA (cDNA)-type Arc 3′UTR (Arc 3′U-C) into our reporter (Fig. 3D), and also observed decreased luciferase mRNA levels from the reporter containing Arc 3′U-C, compared with the control reporter (Fig. 3E). Moreover, we found that Upf1 deficiency did not significantly enhance the levels of Arc 3′U-C reporter mRNA (Fig. 3F). We next measured luciferase activity from this construct, compared with the Arc 3′U-G. Interestingly, the Arc 3′U-C reporter showed a decrease in luciferase activity when compared with control reporters lacking 3′UTR sequences (Fig. S2C,D); however, activity from this construct was not significantly different from that of the Arc 3′U-G reporter. Together, these results suggest that Upf1 does not act on the Arc 3′UTR to regulate the stability of this RNA transcript.

Previous studies have shown that artificial tethering or loading of Ago2 onto the 3′UTR of premature termination codon (PTC)-containing mRNAs inhibits CBP80/20 (also known as NCBP1/2)-dependent translation efficiency and NMD (Choe et al., 2011, 2010). Ago2-mediated NMD inhibition is also dependent on the ability of Ago2 to associate with the cap structure. In this manner, microRNAs (miRNAs) could serve as a surveillance system against nonsense codon-containing mRNAs. To test the possibility that this system is involved in Arc mRNA regulation, we first measured endogenous Arc mRNA levels in cells with knockdown of known miRNA-processing enzymes. Specifically, a small interfering RNA (siRNA) knockdown approach was used to downregulate expression of Ago2, Dicer (also known as Dicer1) and Drosha, key proteins required for miRNA biogenesis and function, and we found that downregulation of any one of these destabilizes Arc mRNA levels (Fig. 4A,B; Fig. S2E–G). To obtain more direct evidence for the regulatory role of miRNAs on Arc mRNA regulation, we verified an interaction between the Arc 3′UTR and Ago2 by streptavidin-biotin RNA affinity purification analysis. Overexpressed green fluorescent protein (GFP)-Ago2 was co-precipitated with biotin-labeled Arc 3′UTR mRNA (Fig. 4C). In contrast, 3′UTR-bound Ago2 was dramatically decreased by the addition of unlabeled Arc 3′UTR, suggesting that Ago2 specifically interacts with the 3′UTR of Arc. Endogenous Ago2 is also bound by the biotin-labeled full-length Arc 3′UTR, and its specific binding to Arc 3′UTR was confirmed by competition analysis with unlabeled 3′UTR (Fig. 4D). Interestingly, we found that CBP20 binding to Arc mRNA was significantly increased in Ago2 knockdown cells, suggesting that Ago2 inhibits CBP80/20-dependent translation of Arc mRNA (Fig. 4E). Collectively, these results suggest that Ago2 binds to the Arc 3′UTR and could serve as a surveillance system to protect against NMD through CBP80/20-dependent translation efficiency.

Our measurement of endogenous Arc mRNA turnover rates and mRNA levels of the Arc 3′UTR reporter construct indicated that Upf1 was not a major determinant of Arc mRNA stability (Fig. 3). Therefore, we next tested whether Upf1 can bind to the Arc 3′UTR by performing RNA pulldown assays; our data confirmed that Upf1 can bind to Arc 3′UTR, and this binding was independent of the EJC (Fig. S3A). The 3′UTR also contains a repository of regulatory elements for translation. Previously, we demonstrated that the 3′UTR of chryptochrome 1 (Cry1) was important for its rhythmic translation, and binding of AU-rich element RNA-binding protein 1 (Auf1; also known as Hnrnpd) to the Cry1 3′UTR recruits the 40S ribosomal subunit to the 5′ end of mRNA (Lee et al., 2014). To determine whether Upf1 similarly modulates Arc translation via binding to the Arc mRNA 3′UTR, we used the same reporter vectors containing the complete Arc 3′UTR from genomic DNA (Arc 3′U-G) or the Arc 3′UTR from cDNA (Arc 3′U-C) from previous experiments, as well as the negative-control vector psiCHECK2 mock (Fig. 5A, schematic). Interestingly, luciferase activities produced from both Arc 3′UTR-containing reporter constructs were enhanced in Upf1 KO cells, suggesting that Upf1 suppresses translation of Arc through its 3′UTR (Fig. 5A).

To confirm the role of Upf1 in translation of endogenous Arc mRNA, we applied sucrose gradient analysis to monitor the association of Arc mRNA with polysomes and non-polysomes in Upf1 KO or WT cells. We found that the overall ribosome profiles were similar in Upf1 KO and WT cells (Fig. 5B). Additionally, Upf1 KO did not affect the distribution of the control ribosomal protein L32 (Rpl32) mRNA. Critically, however, the distribution of Arc mRNA was shifted from the monosome to the polysome fraction in Upf1 KO cells, reflecting an induction of Arc mRNA translation (Fig. 5C; Fig. S3B). To further assess the potential translation-suppressing activity of Upf1, we transfected N2a cells with an mRNA reporter construct containing the Arc 3′UTR fused to luciferase containing an m⁷GpppG-cap (Fig. 5D). We observed that luciferase synthesis from the Arc 3′UTR-containing reporter was considerably higher in Upf1 KO cells than in WT cells (Fig. 5E; Fig. S3C,D).

To clarify the mechanism involved in the Upf1-mediated translational regulation of Arc, we measured Ago2-binding affinities to the Arc 3′UTR in WT and Upf1 KO cells, because Ago2 was proposed to have a role in Arc translational regulation as described above (Fig. 4). Importantly, Ago2 binds more weakly to the Arc 3′UTR and, subsequently, increased CBP20 binding in Upf1 KO cells, suggesting that RNA helicase Upf1 regulates Ago2-binding affinity to the Arc mRNA and subsequent Ago2-mediated Arc translation (Fig. S4A,B). Next, we tested whether newly synthesized Arc in neuronal cells is induced by Upf1 depletion using a puromycin-proximity ligation assay (Puro-PLA) followed by immunostaining. The number of Arc Puro-PLA puncta increased in Upf1 KO compared with WT cells (Fig. 5F). Then, we tested whether newly synthesized Arc exploited 3′UTR-mediated translation mechanism. To this end, we prepared a vector that was designed to express the FLUC peptide under Arc 3′UTR (Fig. 5G) and either FLUC mock or Arc 3′UTR, which were co-transfected with EGFP vectors into WT or Upf1 KO cells prior to puromycin treatment of the cells. As expected, FLUC-Arc 3′U Puro-PLA puncta were increased in Upf1 KO compared with WT cells. However, there were no differences between Upf1 KO and WT cells in FLUC-mock Puro-PLA puncta (Fig. 5H,I; Fig. S4C,D). Taken together, these results imply that Upf1 induces translational repression by facilitating Ago2 binding towards 3′UTR of Arc mRNA.

As shown in a previous study by Messaoudi et al. (2007), sustained translation of newly induced Arc mRNA is necessary for Cofilin (also known as Cfl1) phosphorylation. Given the evidence supporting a role for Upf1 in Arc expression, we next tested the level of phosphorylated (p)-Cofilin in Upf1 KO cells. Western blot analyses indicated that, as expected, Cofilin phosphorylation was robustly increased in Upf1 KO cells (Fig. 6A). Cofilin is one of the major regulators of F-actin dynamics, which controls actin polymerization as well as depolymerization (Bamburg and Bernstein, 2010). Its phosphorylated state promotes actin polymerization. Thus, regulation of Cofilin activity is critical in actin dynamics, which alters cell morphology and neuronal outgrowth. Interestingly, compared with WT cells, Upf1 KO cells generated longer neurite-like processes under normal conditions (Fig. 6B), indicating that the Upf1-dependent Arc regulation is essential for appropriate neurite outgrowth. Neurite outgrowth is a fundamental neural process in formation of proper neuronal connections in the central nervous system. Abnormal or excessive connections between neurons contribute to many neurodevelopmental diseases (Belmonte et al., 2004). We, hence, examined the effect of Upf1 on neurite outgrowth in primary hippocampal neurons. To examine the role of Upf1 in hippocampal neurons, we used shRNA encoding GFP and shRNA targeting Upf1. Two days after transfection on hippocampal neuron cultures, levels of Upf1 were significantly decreased compared with levels in neurons transfected with a mock shRNA, as measured by immunofluorescence (Fig. 6C,D). By analyzing individual neurons, we confirmed that neurons with Upf1 knockdown exhibited increased neurite length and number of primary branches compared with mock shRNA-transfected neurons (Fig. 6E–G). Concordantly, the results from Sholl analysis clearly showed that dendritic arborization was enhanced in shRNA targeting Upf1 (sh_Upf1)-transfected neurons, suggesting that Upf1 is required for proper neurite outgrowth and branching (Fig. 6H,I).

Genome-wide microarray and RNA-seq expression analyses have suggested a regulatory role for Upf1 (Mendell et al., 2004; Pan et al., 2006). However, most studies aimed at identifying Upf1 targets have not distinguished Upf1 indirect versus direct targets. In a previous investigation, whole-transcriptome decay rates were measured in Upf1-depleted cells using a combined RNA-seq and BRIC-seq approach (Tani et al., 2012). This analysis identified 76 direct Upf1 target transcripts that show altered decay rates and are upregulated in Upf1-depleted cells. Interestingly, however, hundreds of transcripts were at least 2-fold upregulated, but did not show altered decay rates, in response to Upf1 depletion. These transcripts were referred to as indirect NMD targets. Another study involving the simultaneous analysis of pre-mRNA and mRNA abundance for an array of selected transcripts strongly suggested that Upf1 depletion promotes transcriptional activation (Viegas et al., 2007). That is, a subset of 16 transcripts was affected by Upf1 depletion in an NMD-independent and non-post-transcriptional fashion. The results of our study are consistent with these data and indicate that Arc expression is also regulated by Upf1 at the level of transcription.

We first analyzed Arc mRNA and pre-mRNA levels in Upf1 knockdown and KO cells by qRT-PCR and found that these did not differ significantly (Figs 1C,D and 2B–D), strongly suggesting that Arc mRNA is transcriptionally modulated. The Arc promoter has been well characterized and is widely used as a tool to facilitate neuronal activity mapping (Kawashima et al., 2009; Waltereit et al., 2001). In addition, this promoter has been employed to successfully track an active neuronal circuit in living mice (Kawashima et al., 2013). Here, we utilized an Arc promoter-luciferase reporter to investigate the potential for gene expression at the transcriptional level by Upf1. Consistent with our results measuring endogenous transcripts, Arc promoter activity was found to be upregulated in Upf1-depleted cells (Figs 1G and 2F).

The transcription factors Mef2a and Mef2d are highly expressed in the brain and are tightly regulated by several distinct calcium signaling pathways (McKinsey et al., 2002). These Mef2 transcription factors negatively regulate excitatory synapse number in an activity-dependent manner, in part, through the induction of Arc transcription (Flavell et al., 2006). Here, we found that Upf1 binds to endogenous Mef2a mRNA, and both Mef2a mRNA and protein levels are significantly elevated in Upf1 KO cells (Fig. 2G–L). This suggests that Upf1 can modulate Arc levels indirectly via its effect on Mef2a.

The length of the 3′UTR has been suggested as an important feature that influences NMD, and a number of studies have shown that long 3′UTRs strongly promote NMD and increase Upf1 assembly in targeted mRNAs (Hogg and Goff, 2010; Hurt et al., 2013; Yepiskoposyan et al., 2011). It has further been found that the 3′UTRs of Upf1 indirect target transcripts (group A mRNAs) are significantly longer than those of other transcripts (Tani et al., 2012). Here, we demonstrated that a 3719-nucleotide (nt) region at the 3′UTR of Mef2a is critical for NMD (Fig. S5A). A recent study identified important features of 3′UTRs targeted by Upf1 and found that these have GC-rich motifs embedded in high GC-content regions (Imamachi et al., 2017). Notably, the Mef2a 3′UTR contains one such GC-rich motif (Fig. S5A, highlighted in red). We further showed Arc promoter activity using Mef2a knockdown to establish a causal mechanism (Fig. S5B). Consistent with this observation, transcription factors were enriched among potential Upf1 targets (Imamachi et al., 2017), and, indeed, several transcription factors, including activating transcription factor 3 (ATF3) and CAMP response element-binding protein 2 (Creb2; also known as ATF4), have been identified as Upf1 targets (Mendell et al., 2004). The SARE located upstream of the Arc gene is a major neuronal activity-dependent element, which consists of binding sites for Creb, Mef2 and serum response factor (Srf) proteins. As expected, and consistent with our previous results, gene expression profiles for Creb2 obtained in Upf1 KO cell lines indicated that this transcript is also regulated by Upf1 (Fig. S5C).

Regulation of gene transcription is one of the main mechanisms for modulating the levels of specific mRNAs and proteins. Both the selective localization and accumulation of Arc mRNA at synapses are mediated by activity-dependent regulation of this transcript throughout dendrites (Farris et al., 2014; Steward et al., 1998). These findings suggest that differential localization of Arc mRNA is the combined result of targeting newly synthesized Arc mRNA to active domains and degradation throughout dendrites in the hippocampus. When transcription is inhibited, there is no accumulation of Arc mRNA in the activated lamina, although Arc mRNA was already present throughout dendrites. In this study, we provide evidence for a novel role for Upf1 in the precise regulation of Arc transcription, which we propose allows for a more exquisite activity-dependent control of Arc expression.

One-third of human genes have 3′UTRs longer than 1000 nts (Pesole et al., 2000). This raises the question of how a large subset of naturally occurring mRNAs containing long 3′UTRs have acquired mechanisms to evade the NMD pathway. A recent study identified a cis-acting element located within the first 200 nts of the 3′UTR, which inhibits NMD when positioned downstream of the translation termination codon (Toma et al., 2015). Additionally, binding of Ago2 to the 3′UTR of PTC-containing mRNAs abrogates NMD by inhibiting CBP80/20-dependent translation (Choe et al., 2011). Uniquely, Arc pre-mRNA contains two introns in the 3′UTR, strongly suggesting that this transcript is a natural target of NMD. To test whether the post-transcriptional decay of Arc mRNA occurs via NMD, we measured both endogenous Arc mRNA stability and the levels of an mRNA reporter construct containing the Arc 3′UTR region in cells depleted for Upf1, a critical component of the NMD pathway. Interestingly, we found that Arc mRNA stability was not increased in Upf1-depleted cells (Fig. 3A; Fig. S2A,B), and, similarly, Arc 3′UTR reporter mRNA levels were unaffected by Upf1 depletion (Fig. 3F; Fig. S2C,D).

It has previously been determined that coactivator-associated arginine methyltransferase 1 (Carm1) associates with the major NMD factor Upf1 and promotes its occupancy on PTC-containing transcripts (Sanchez et al., 2016). As predicted, growth arrest and DNA damage inducible alpha (Gadd45a) and asparagine synthetase (glutamine-hydrolyzing) (Asns), which are established NMD targets, show stabilized mRNA levels in Carm1 KO mouse embryonic fibroblasts (MEFs). Surprisingly, however, in the same study, Arc mRNA decay rates were found to be similar in Carm1 KO and WT MEFs. Moreover, other reports have suggested that some plant genes containing introns in their 3′UTR are not subject to NMD (Gadjieva et al., 2004; Rose, 2004). Here, we found that Ago2 binds to 3′UTR of Arc, and, interestingly, downregulation of this protein, as well as those involved in miRNA biogenesis, including Drosha and Dicer, results in reduced Arc mRNA levels (Fig. 4; Fig. S2E–G), suggesting that Ago2 might inhibit Upf1-mediated NMD. These data demonstrate that Arc is an endogenous Ago2-mediated surveillance target gene and thus provide evidence for the existence of a novel mechanism for post-transcriptional regulation of gene expression, involving regulation of NMD by miRNAs, in neuronal cells. It is reported that a set of miRNAs, which are predicted as binding candidates of Arc 3′UTR, inhibit Arc protein expression in cultured hippocampal neurons (Wibrand et al., 2012). A previous study identified differential regulation of Ago2-associated and Arc-targeting miRNA following LTP induction using Ago2 immunoprecipitation (IP) (Pai et al., 2014). Interestingly, Ago2 IP/input ratios of miR-34a were enhanced, but Ago2 IP/input ratios of miR-326 were significantly decreased, following high frequency stimulation. Given the central role of Arc expression in LTP, Ago2 might be a potential effector in synaptic plasticity.

NMD requires both splicing and translation, and the EJC also plays a critical role in this process. Here, we found that a reporter construct containing the cDNA-type Arc 3′UTR was destabilized to a similar extent as the intron containing Arc 3′UTR, indicating that Arc mRNA is degraded through a novel mechanism, distinct from NMD (Fig. 3D,E; Fig. S2C,D). This finding is also supported by the results of Ninomiya et al. (2016). To analyze the mechanism for human ARC mRNA degradation, the authors constructed an ARC genome-type 3′UTR fused to an EGFP reporter and then removed the termination codons, in order to inhibit NMD targeting of this construct. Surprisingly, however, termination codon removal from the ARC mRNA 3′UTR did not stabilize reporter mRNA. This implies a role for other mRNA decay processes in ARC mRNA regulation. Importantly, ARC mRNA localization is mediated by a dendritic-targeting element (DTE) in the ARC 3′UTR, which also has a cis-acting element for destabilizing ARC mRNA, independent of the NMD pathway (Ninomiya et al., 2016). Besides, this DTE alone was able to promote degradation of the reporter RNA, dependent on translation. In this study, we do not address the potential for additional Arc mRNA degradation and destabilization mechanisms, including trans-acting elements, and the identity of any cis- and trans-acting factors, as well as the region of the mouse Arc mRNA transcript required for degradation and binding of destabilizing factors, remains to be elucidated.

The 3′UTR contains regulatory elements that post-transcriptionally influence gene expression, and these regions can affect mRNA stability (Kim et al., 2011), localization and translation efficiency (Lee et al., 2014). It is thought that longer UTRs may contain regulatory motifs necessary to specify complex temporal and spatial translational regulation. Arc has a long 3′UTR, consisting of 1670 nts, so we tried to reveal another possible role for Upf1 in the translation of Arc. Additionally, the Arc mRNA 3′UTR region was reported to contribute to the modulation of Arc expression upon BDNF treatment (Paolantoni et al., 2018). By in vitro binding assay, we first confirmed that Upf1 binds to the 3′UTR of Arc regardless of the EJC, which shows the possible effects of Upf1 on post-transcriptional regulation of Arc. It is reported that the binding of Upf1 to the EJC triggers Upf1 phosphorylation, and phospho-Upf1 can mediate translational repression (Isken et al., 2008). Here, we demonstrated translational repression of Arc in the presence of Upf1 using a luciferase reporter activity analysis and polysome profiling of reporter mRNA (Fig. 5; Fig. S3A,B). Importantly, regardless of the presence of the EJC, binding of Upf1 to the Arc 3′UTR represses translation. LTP consolidation in the dentate gyrus of the hippocampus induced a shift in Arc mRNA from monosomes and messenger ribonuclearprotein particles to heavy polysomes (Panja et al., 2014). This shift in Arc mRNA may also be caused, in part, by the changes in the binding properties of Upf1 to Arc mRNA, although the specific mechanism underlying this effect is still to be found. Upf1 is widely recognized as an RNA helicase and, once recruited on target mRNAs, Upf1 can move along the mRNA to remodel the messenger ribonucleoproteins (Fiorini et al., 2015; Jankowsky, 2011). We showed that Ago2 also binds to 3′UTR of Arc and might inhibit Arc translation and NMD. To confirm the role of Ago2 in the Upf1-mediated translational repression of Arc, we evaluated Ago2-binding affinity to Arc 3′UTR in the presence or absence of Upf1. We found that the binding affinity of Ago2 to Arc 3′UTR was decreased in the Upf1 KO cells (Fig. S4A), which correlates with translational repression of Arc in the presence of Upf1. The discovery of numerous mRNA transcripts in dendrites has suggested that synapses could be modified directly through regulation of local protein synthesis (Bramham and Wells, 2007; Sutton and Schuman, 2006). It is thought that synaptic mRNAs are transported in a translationally dormant state via interactions with RNA-binding proteins. Following synaptic stimulation, local mRNA translation is then activated in the synaptic region by neutralizing the repressive RNA-binding proteins. In our study, we found that Arc mRNA is not simply degraded by NMD. Rather, Upf1 suppresses Arc mRNA translation through regulation of Ago2-binding affinity towards the Arc 3′UTR, which may aid in precise spatiotemporal gene regulation.

LTP and LTD typically require a rapid induction of gene expression. In Arc KO mice, early-phase LTP is enhanced, whereas late-phase LTP is blocked, and LTD, in general, is reduced (Plath et al., 2006). Synaptic depression in both homeostatic plasticity and LTD is mediated by AMPAR endocytosis. There is strong evidence for Arc involvement in excitatory synaptic transmission via interaction with components of the dynamin and endophilin 2/3 complex, which leads to internalization of surface AMPAR (Chowdhury et al., 2006; Park et al., 2008; Waung et al., 2008). It has also been found that knockdown of the EJC factor, eIF4AIII, increases Arc mRNA and protein levels, which should decrease the levels of surface AMPAR at synapses. However, unexpectedly, eIF4AIII knockdown resulted in increased amounts of surface AMPAR (Giorgi et al., 2007).

Previous reports have suggested that Arc performs a dynamic function in LTP in vivo (Messaoudi et al., 2007). Initial Arc synthesis is necessary for early stages of LTP, whereas late synthesis is required to generate stably modified synapses. Local infusions of antisense oligodeoxynucleotides against Arc during LTP results in Cofilin dephosphorylation and the loss of F-actin from synaptic sites. Here, we show that a number of Upf1 KO N2a cells have neurite-bearing morphology (Fig. 6B), and these structural changes suggest the possibility that Upf1-mediated Arc regulation affects F-actin expansion. Importantly, sustained hyperphosphorylated Cofilin can be observed in Upf1 KO cells, as shown in Fig. 6A, and we also tested whether Arc knockdown blocks the increase in Cofilin phosphorylation (Fig. S6A).

Our work reveals a physiological role of the Upf1-mediated Arc gene expression in controlling neurite outgrowth and dendritic branching (Fig. 6E–I), which are critical processes for proper wiring of the complex neuronal networks in the brain.

In this study, we provide evidence that Upf1 suppresses Arc transcription and translation via a number of distinct mechanisms. At the transcriptional level, Upf1 regulates Arc activity indirectly by regulating the transcription factors Mef2a and Creb2. Additionally, Upf1 suppresses Arc mRNA translation through its binding to the Arc 3′UTR, although Arc transcripts, which had been thought to be targeted for NMD, escape from this pathway and are stabilized by miRNA-mediated gene silencing. Finally, we show that depletion of Upf1, which leads to sustained Arc expression, promotes hyperphosphorylation of Cofilin and triggers neurite extension. Thus, based on these data, we propose that the Upf1-mediated downregulation of Arc at the level of transcription and translation in basal conditions is necessary for activity-induced Arc synthesis and subsequent activity-dependent changes in synaptic strength.

To generate psiCHECK2 Arc 3′U-C, mouse Arc (accession no. NM_018790.3) 3′UTR was amplified using Pfu polymerase (SolGent) with specific primers and the sequence was confirmed by sequencing. To generate psiCHECK2 Arc 3′U-G, Arc 3′UTR genomic DNA was amplified from Institute for Cancer Research (CrljBgi:CD-1) mice brain genomic DNA using the specific primers. For the in vitro binding analysis, full-length Arc 3′UTRs were amplified as previously described. PCR products were digested with specific enzymes EcoRI/XbaI and subcloned into pBluescript SK(+) (pSK) to generate pSK-Arc 3′U-G or -C.

N2a and NB41A3 cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum and 100 U/ml penicillin G and streptomycin. All cell lines were authenticated prior to experiments and they are not listed as commonly misidentified cell lines by the International Cell Line Authentication Committee. Cell identity and status are regularly checked. To block transcription, cells were treated with 5 μg/ml actinomycin D (Sigma-Aldrich, A9415) or 100 μg/ml 5,6-dichloro-1-beta-D-ribofuranosylbenzimidazole (Sigma-Aldrich, 53-85-0) and then harvested at the indicated time points. To block translation, cells were treated with 100 μg/ml cycloheximide (CHX; EMD Millipore, 239764).

Hippocampal cultures were prepared from embryonic day 17 mouse embryos and treated with DNase (Sigma-Aldrich, DN25) and trypsin (Sigma-Aldrich, T6567) at 37°C. Hippocampal primary neurons were seeded on 12-well plates with round glass coverslips or six-well plates without round glass coverslips coated with poly-L-lysine. Neurons were cultured in neurobasal medium (Gibco, Thermo Fisher Scientific, A3582901) with 1% glutamax (Gibco, Thermo Fisher Scientific, 35050061), 1% penicillin/streptomycin and B27 supplement (Gibco, Thermo Fisher Scientific, 17504044). Neurons at days in vitro (DIV)1 or DIV12 were transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol.

Generation of KO cell lines was performed as previously described (Lee et al., 2016; Mali et al., 2013). A 19 bp of the selected sequence (5′-TGCTCGGCGCCGACACCCA-3′) targeting the DNA region within exon 1 of Upf1 was selected from in silico tools of predicted protospacer adjacent motif target sites. 5′-CTGCTCGGCGCCGACACCCAGGG-3′ was inserted into two 60-mer oligonucleotides (Insert_F, 5′-TTTCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACCGNNNNNNNNNNNNNNNNNNN-3′ and Insert_R, 5′-GACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAACNNNNNNNNNNNNNNNNNNNC-3′). The two oligonucleotides were annealed and extended to make a 100 bp double-stranded DNA fragment using Phusion polymerase. Then the gRNA cloning vector (http://www.addgene.org/41824/) was linearized using AflII, and the 100 bp DNA fragment was incorporated into the above vector using Gibson assembly. N2a cells were cultured in six-well dishes to 70% confluence and co-transfected with Upf1 sgRNA plasmid plus Cas9 with Lipofectamine 2000. After a 48-h incubation, cells were selected by 500 µg/ml G418 with medium changes every 2 days until single colonies formed. After single colony isolation, stable cells were maintained with 200 µg/ml G418-containing medium for 1 week.

Transient transfections were performed by electroporation using a Microporator MP-100 (Digital Bio) or Neon-Transfection System (Invitrogen) and Lipofectamine 2000 (Invitrogen), according to the manufacturers’ instructions. To knock down Upf1 expression, we selected sequences targeting coding sequence of Upf1. The knockdown-verified sense strand of Upf1 shRNA is 5′-TgatgcagttccgttccatcTTCAAGAGAgatggaacggaactgcatcTTTTTTC-3′ (based on position 2185–2203 of mouse Upf1). The 19-nt Upf1 target sequences are indicated in lower-case letters. The sense strand and antisense strand were annealed and cloned into the pLL3.7 lentiviral vector.

For the reporter assay, N2a cells transfected with reporter or control plasmids were lysed in Reporter Lysis 5× Buffer (Promega, E3971). Renilla and firefly luciferase activities were determined using the Dual-Luciferase^® Reporter Assay System (Promega), according to the manufacturer's instructions.

N2a cells were lysed with RNA-IP buffer [10 mM HEPES (pH 7.5), 100 mM KCl, 5 mM MgCl₂, 0.1% NP-40, protease inhibitor]. Rat IgG or anti-Upf1 antibody was incubated with N2a cell lysates at 4°C overnight and then incubated with Protein G beads at 4°C for 4 h. We washed the beads three times with RNA-IP buffer and isolated RNA using TRI reagent. RNA levels were quantified by qRT-PCR.

RNA isolation and cDNA synthesis were performed as we previously described (Song et al., 2016). N2a cells were lysed with TRI-Solution (Bio Science Technology) and the total RNA was extracted after the addition of 0.2 volume of chloroform. Samples were mixed with vortex, incubated at room temperature for 10 min, and centrifuged at 12,000 g for 15 min at 4°C. Recovery of total RNA was performed by precipitation with 1/2 volume of isopropanol. After 10 min incubation at room temperature, samples were centrifuged at 12,000 g for 8 min at 4°C. The RNA pellet was washed with 75% ethanol, briefly air-dried and dissolved in diethylpyrocarbonate (DEPC)-treated water by incubating for 5 min at 65°C. The yield and purity of RNA were determined by a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific). Following quantification, 1μg of each total RNA sample was reverse transcribed using oligo-dT or random hexamer and the ImProm-II Reverse Transcription System (Promega), according to the manufacturer's instructions. For qRT-PCR analysis, each cDNA sample was diluted five times with nuclease-free water.

qRT-PCR was performed as we previously described (Song et al., 2016). The mRNA levels of endogenous genes or reporter plasmids were detected by qRT-PCR using a StepOnePlus Real-Time PCR System (Applied Biosystems) with FastStart Universal SYBR Green Master (Roche). A 20-µl sample of reaction cocktail was constituted of 3 μl diluted cDNA, 10 μl 2× SYBR Green Master Mix and 0.5 μl of each of the forward and reverse primers. The following amplification program was used: polymerase activation at 95°C for 10 min, 40 repeated cycles of 95°C for 15s and 60°C for 1 min. A comparative Ct method was used for quantification. The sequences of the forward and reverse primers are as follows: Arc (mRNA), 5′-TGAGACCAATTCCACTGATG-3′ and 5′-CTCCAGGGTCTCCCTAGTCC-3′; Arc (mature-mRNA), 5′-CCCCCAGCAGTGATTCATAC-3′ and 5′-GTGATGCCCTTTCCAGACAT-3′; Arc (pre-mRNA), 5′-CTCAGATCCCAGCCACTCTC-3′ and 5′-GTGATGCCCTTTCCAGACAT-3′; Gapdh, 5′-CATGGCCTTCCGTGTTCCTA-3′ and 5′-CCTGCTTCACCACCTTCTTGAT-3′; Mef2a (mRNA), 5′-ATCCGTTTACTGGGCTTGTG-3′ and 5′-TGTCAGGACAGCAGATGAGG-3′; Mef2a (pre-mRNA), 5′-AGAAGGAAGAGTTGCGGTGA-3′ and 5′-ACACGCATAATGGATGAGAGG-3′; Rpl32, 5′-AACCCAGAGGCATTGACAAC-3′ and 5′-CACCTCCAGCTCCTTGACAT-3′; Ago2, 5′-AAGTCGGACAGGAGCAGAAA-3′ and 5′-GAAACTTGCACTTCGCATCA-3′; Dicer, 5′-CTCGTCAACTCTGCAAACCA-3′ and 5′-CAGTCAAGGCGACATAGCAA-3′; Drosha, 5′-AACAGTTCAACCCCGAAGTG-3′ and 5′-CTCTGAGCCAGCTTCTGCTT-3′; β-actin, 5′-TATTGGCAACGAGCGG-3′ and 5′-CGGATGTCAACGTCAC-3′ and Creb2, 5′-CCTAGGTCTCTTAGATGACTATCTGGAGG-3′ and 5′-CCAGGTCATCCATTCGAAACAGAGCATCG-3′.

For in vitro binding assays, biotin-UTP-labeled RNA was transcribed from the XbaI-linearized pSK-Arc 3′UTR plasmids or BamHI-linearized psiCHECK2-Arc 3′UTR plasmids using T7 RNA polymerase (Promega, P2077). Streptavidin-biotin RNA affinity purification was performed as described previously (Seo et al., 2017). In brief, cell extracts prepared from N2a cells were incubated with biotinylated-Arc 3′UTR RNA and subjected to streptavidin resin adsorption. Resin-bound proteins were analyzed by SDS-PAGE.

N2a cells were lysed with RIPA buffer containing protease inhibitor tablet (Thermo Fisher Scientific, A32953), followed by sonication. Protein concentrations of lysates were determined using Bradford reagent (AMRESCO). Proteins were resolved by SDS-PAGE and transferred to nitrocellulose membranes (Pall Corporation), incubated with blocking buffer (5% non-fat dry milk in Tris-buffered saline and 0.1% Tween 20) for 30 min. Horseradish peroxidase-conjugated mouse (Thermo Fisher Scientific), rabbit (Promega), rat or goat (Bethyl Laboratories) secondary antibodies were detected with SUPEX ECL reagent (Neuronex) and an LAS-4000 system (Fujifilm), according to the manufacturers’ instructions. Acquired images were analyzed using Image Gauge (Fujifilm) according to the manufacturer's instructions. The integrated blot density was quantified through ImageJ software-based analysis (http://rsb.info.nih.gov/ij/).

Antibodies used in this study included anti-Mef2a (1:500 dilution; Santa Cruz Biotechnology, sc-17785), anti-Ago2 (1:500 dilution; Santa Cruz Biotechnology, sc-376696), anti-GFP (1:1000 dilution; Santa Cruz Biotechnology, sc-8334), anti-CBP20 (1:100 dilution; Santa Cruz Biotechnology, sc-48793), anti-Cofilin (1:100 dilution; Abcam, ab54532), anti-p-Cofilin (1:1000 dilution; Santa Cruz Biotechnology, sc-12912), anti-14-3-3ζ (1:1000 dilution; Santa Cruz Biotechnology, sc-1019), anti-Arc (1:1000 dilution, SYSY, 156 003), anti-Gapdh (1:5000 dilution, Millipore, MAB374), anti-Flag (1:1000 dilution; Sigma-Aldrich, F1804), rabbit monoclonal anti-FLUC (1:10000 dilution, Abcam, ab185924) and mouse monoclonal anti-puromycin (1:1000 dilution, Merck, MABE343). Anti-Upf1 antibodies were generated and purified from rats immunized with carrier protein-conjugated peptide, DTQGSEFEFTDFTLPSQTQ or KDETGELSSADEKRYRALK.

N2a cells were incubated in PBS containing 100 µg/ml CHX for 30 min on ice. The cells were lysed in polysome lysis buffer [300 mM KCl, 5 mM MgCl₂, 10 mM HEPES (pH 7.4), 0.5% NP-40, 5 mM DTT and 100 µg/ml CHX]. Cell lysates were loaded on 15–45% sucrose gradients in polysome buffer [300 mM KCl, 5 mM MgCl₂ and 10 mM HEPES (pH7.4)] and centrifuged at 32,000 rpm (175,117 g) in a SW41Ti rotor at 4°C for 3 h. The gradients were collected using a Brandel gradient density fractionator and analyzed by an Econo UV monitor (Bio-Rad). Fractions (1 ml each) were spiked with 100 ng Renilla luciferase (Rluc) RNA to ensure the technical consistency of RNA isolation. RNAs were isolated from these fractions, and reverse transcription and qRT-PCR were performed.

N2a cells on chip glass were treated with 5 μM puromycin for 90 min at 37°C, 5% CO₂ and washed twice in PBS-MC (1× PBS, 1 mM MgCl₂, 0.1 mM CaCl₂). Cells were fixed in 4% paraformaldehyde (PFA; Sigma-Aldrich)-sucrose for 20 min and permeabilized with 0.5% Triton X-100 solution (Sigma-Aldrich) for 15 min. Duolink PLA reagents (Sigma-Aldrich) were used to detect newly synthesized proteins according to the manufacturer's instructions. The following antibodies were used for puro-PLA: rabbit monoclonal anti-FLUC (Abcam), anti-Arc (SYSY) and mouse monoclonal anti-puromycin (Merck). Nuclei of cells were stained with Hoechst 33342 and chip glasses were mounted with fluorescent mounting medium overnight. Duolink PLA signals were visualized by fluorescence microscopy (Olympus FV1000).

Transfected cells were maintained for 24 h, fixed with 4% PFA for 20 min. Nuclei of cells were stained with Hoechst 33342 and chip glasses were mounted with fluorescent mounting medium (Dako) overnight. The following antibodies were used for immunocytochemistry: rabbit monoclonal anti-FLUC (Abcam), anti-Upf1 (Sigma-Aldrich or Santa Cruz Biotechnology) and mouse monoclonal anti-puromycin (Merck). All images were obtained using a laser-scanning confocal microscope (model FV1000; Olympus) and IX71-Olympus inverted microscope with U-RFL-T (model IX71; Olympus), and FV10-ASW2.0 fluoviewer software was used for image analysis. We also used Alexa Fluor 568-conjugated phalloidin dye to label actin filaments.

All quantitative data are shown as mean±s.e.m. Statistical analyses were performed using Student's t-test, one-way analysis of variance (ANOVA) or two-way ANOVA and post-hoc Tukey's multiple comparison tests or Sidak's multiple comparisons using GraphPad software. P-values less than 0.05 were considered statistically significant. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

We are grateful to Dr Konstantin A. Lukyanov (Institute of Bioorganic Chemistry, Moscow, Russia) for providing dual-color fluorescent protein-based reporters for quantification of NMD activity, and to Dr Yoon Ki Kim (Korea University, Seoul, Korea) for providing the Flag_Upf1 plasmid.