N6-methyladenosine promotes induction of ADAR1-mediated A-to-I RNA editing to suppress aberrant antiviral innate immune responses

Among over 150 distinct RNA modifications, N6-methyladenosine (m6A) and adenosine-to-inosine (A-to-I) RNA editing represent 2 of the most studied modifications on mammalian mRNAs. Although both modifications occur on adenosine residues, knowledge on potential functional crosstalk between these 2 modifications is still limited. Here, we show that the m6A modification promotes expression levels of the ADAR1, which encodes an A-to-I RNA editing enzyme, in response to interferon (IFN) stimulation. We reveal that YTH N6-methyladenosine RNA binding protein 1 (YTHDF1) mediates up-regulation of ADAR1; YTHDF1 is a reader protein that can preferentially bind m6A-modified transcripts and promote translation. Knockdown of YTHDF1 reduces the overall levels of IFN-induced A-to-I RNA editing, which consequently activates dsRNA-sensing pathway and increases expression of various IFN-stimulated genes. Physiologically, YTHDF1 deficiency inhibits virus replication in cells through regulating IFN responses. The A-to-I RNA editing activity of ADAR1 plays important roles in the YTHDF1-dependent IFN responses. Therefore, we uncover that m6A and YTHDF1 affect innate immune responses through modulating the ADAR1-mediated A-to-I RNA editing.


Introduction
Posttranscriptional regulation plays a pivotal role in ensuring proper gene expression in almost all organisms. An emerging new mechanism of posttranscriptional regulation is conducted through various RNA modifications. With over 150 distinct chemical modifications known to exist in different RNA species [1], new scientific discoveries and technological advances have prompted rapid development of the field of epitranscriptomics [2,3]. Among mRNA modifications, 2 of the most well-characterized are N 6 -methyladenosine (m 6 A) and adenosine-to-inosine (A-to-I) RNA editing. The profound impact of m 6 A modification on diverse biological functions and diseases have only recently been demonstrated [4,5]. The m 6 A modification exhibits biological functions through a series of m 6 A effector proteins generally referred to as writers, readers, and erasers, which deposit, recognize, and remove m 6 A, respectively. In mammals, the deposition of m 6 A on most mRNA is catalyzed by a large methyltransferase complex (writers) containing the core heterodimer of methyltransferase-like 3 (METTL3) and METTL14 [6][7][8]. m 6 A modification recruits m 6 A-binding proteins (readers) including several members of YTH family [9] that specifically recognize m 6 A to modulate various aspects of RNA metabolism. For example, YTH N 6 -methyladenosine RNA binding protein 1 (YTHDF1) promotes translation efficiency of m 6 A-modified mRNAs through recruiting translation initiation factors [10,11]. Another members of YTH family, YTHDF2, interacts with the CCR4-NOT deadenylase complex and facilitates degradation of its target mRNAs [12,13]. The removal of m 6 A is carried out by 2 demethylases (erasers), Fat mass and obesity-associated gene (FTO) [14], and alkB homolog 5 RNA demethylase (ALKBH5) [15]. The orchestration of writers and erasers functions suggests m 6 A as a reversible modification that can control dynamics of physiological processes [16].
On the other hand, A-to-I RNA editing is catalyzed by adenosine deaminase acting on RNA (ADAR) enzymes, ADAR1 and ADAR2, that specifically bind to double-stranded RNA (dsRNA) [17,18]. Among mammalian ADAR family members, editing activity of ADAR3 has not been detected. Because inosine base pairs preferentially with cytosine, A-to-I RNA editing can cause alteration in amino acid sequences [19], splicing [20], and dsRNA structures [21,22]. Recently, the ADAR1-mediated A-to-I RNA editing has emerged as a key regulatory factor that controls innate immune interferon (IFN) response [23,24]. Both ADAR1 knockout mice and knock-in mice expressing the catalytic-deficient ADAR1 (ADAR1 E861A/E861A ) exhibited embryonic lethality with elevated IFN-stimulated genes (ISGs) signature and apoptosis [24,25]. The lethality can be rescued by concurrent knockout of interferon induced with helicase C domain 1 (IFIH1, also called MDA5), a cytosolic sensor that recognizes viral dsRNA and induces the type I IFN-mediated antiviral immunity [25,26]. The molecular mechanism underlying these phenotypes is that ADAR1 unwinds dsRNA structure by RNA editing and prevents self-activation of the IFN response induced by endogenous dsRNA. Genomic mutations in ADAR1 were identified in patients with Aicardi-Goutières syndrome (AGS), a severe autoimmune disease with a high IFN signature [27]. These studies indicated that the appropriate regulation of IFN signaling by ADAR1 is essential for normal homeostasis of the innate immune system.
Dysregulation in the expression of ADARs and A-to-I RNA editing is associated with a variety of physiological abnormalities including neuronal disorders [28,29], circadian rhythm disruption [30,31], and cancer progression [32]. Growing evidence shows a profound impact of ADAR1 on cancer immunotherapies through regulating IFN signaling. Loss of ADAR1 reduces cell viability in certain types of tumor cells expressing high levels of ISGs [33,34]. In addition, deletion of ADAR1 in tumor cells improves sensitivity to immunotherapy through enhancing IFN sensing and overcomes resistance to immune checkpoint blockage [35]. Therefore, a better insight into regulatory mechanisms of ADAR1 expression may provide further understanding to targeting ADAR1 in cancer therapies. Intriguingly, a recent study discovered a conserved m 6 A site in the ADAR1 transcript among several primates [36], implying potential roles of m 6 A-mediated regulation of ADAR1 expression. However, biological functions of m 6 A on the ADAR1 transcript have not been investigated. Here, we find that binding of YTHDF1 to m 6 A-modified mRNA enhances IFN-mediated ADAR1p150 induction. Knockdown of YTHDF1 attenuated global induction of A-to-I RNA editing upon IFN stimulation, resulting in activation of the dsRNA-sensing pathway. In the absence of YTHDF1, IFN stimulation causes enhanced innate immune responses, and virus replication in cells was suppressed by the elevated expression of ISGs. Our data uncover a new m 6 A-mediated regulation of A-to-I RNA editing and IFN response in innate immunity.

m 6 A regulates interferon-induced expression of ADAR1p150
A previous genome-wide analysis suggested a negative correlation between m 6 A methylation and A-to-I RNA editing, with depletion of m 6 A increasing editing [37]. In addition to this report, the observation of a conserved m 6 A site in the ADAR1 transcript [36] prompted us to investigate functional connection between these 2 main types of mRNA modifications. The transcriptome-wide distributions of m 6 A in various cell types have been mapped using immunoprecipitation-based high-throughput sequencing approach [9,38]. We analyzed the previously published m 6 A-seq data and found that all of the 3 mammalian A-to-I RNA editing enzymes (ADAR1-3) have m 6 A sites in their mRNAs (Figs 1A and S1A). ADAR1 is expressed as 2 isoforms: a constitutively expressed 110-kDa isoform ADAR1p110, which primarily localizes to the nucleus, and an IFN-inducible 150-kDa isoform ADAR1p150, which localizes to both the cytoplasm and nucleus. RIP-qPCR with antibodies against YTHDF1 and YTHDF2 showed that both YTHDFs strongly bound to ADAR1p110, ADAR1p150, and ADAR2 mRNAs in A172 glioblastoma cells, whereas ADAR3 was not expressed in A172 cells (Fig 1B). We next performed knockdown of m 6 A writers and readers and examined their effects on ADARs. Although METTL14 knockdown slightly reduced ADAR1p150 protein levels, knockdown of METTL3, YTHDF1, or YTHDF2 did not cause notable changes of both ADAR1p110 and ADAR1p150 protein expression levels in A172 cells under normal culture conditions (S1B and S1C Fig). On the other hand, YTHDF2 knockdown increased expression levels of both ADAR2 protein (S1B and S1C Fig) and mRNA (S1D Fig). YTHDF2 is known to facilitate mRNA degradation [12,13], suggesting that m 6 A methylation could suppress ADAR2 expression through controlling its transcript stability. m 6 A modification enables rapid tuning of gene expression in response to cellular stress conditions such as heat shock [39,40] and oxidative stress [41]. Because IFN stimulation is known to induce ADAR1p150 transcription from IFN-stimulated response element [42], we examined potential role of m 6 A on ADAR1p150 expression in response to human IFN-α stimulation. Immunoblot analysis revealed that knockdown of METTL3, METTL14, and YTHDF1 attenuated the IFN-mediated induction of ADAR1p150 (Fig 1C), whereas YTHDF2 knockdown did not noticeably alter ADAR1p150 protein expression upon IFN stimulation (S1E Fig). In contrast, the expression level of ADAR1p150 mRNA was increased by knocking down METTL3 or YTHDF1 in IFN-treated cells (S1F Fig), which could be due to activation of upstream IFN pathway. The detailed mechanisms will be described in the latter part of this study. YTHDF1 promotes translation efficiency of target transcripts through interaction with translation initiation factors [10,43,44]. Our recent analysis indicated that depletion of YTHDF1 leads to an overall reduction of translation efficiency in transcripts harboring YTHDF1-bound m 6 A peaks [11]. Furthermore, we measured degradation rate of ADAR1p150 protein after treatment of cycloheximide (CHX) inhibiting new protein translation. YTHDF1 knockdown did not affect ADAR1p150 protein stability under the condition of IFN treatment (S1G Fig). The decreased ADAR1p150 protein level by IFN stimulation, despite the increased mRNA level and no change in the protein stability, clearly suggests that YTHDF1 promotes translation efficiency of ADAR1p150 mRNA via m 6 A in IFN-treated cells (S1H Fig). Of note, m 6 A immunoprecipitation (m 6 A-IP) followed by quantitative reverse transcription PCR (RTAU : Pleas -qPCR) revealed that the ratio of m 6 A-modified/total mRNA between mock and IFN-α treatment was not altered (Fig 1D). However, because IFN-α treatment increased the expression levels of both modified and unmodified ADAR1p150 mRNA, more m 6 A-modified

PLOS BIOLOGY
ADAR1p150 transcripts were induced by IFN stimulation to produce more ADAR1p150 protein.
We next monitored the time course of the ADAR1p150 protein expression after IFN stimulation with YTHDF1 knockdown. We observed that the knockdown of YTHDF1 reduced and delayed the induction of the ADAR1p150 protein level in response to IFN (Figs 1E and S1I), indicating that YTHDF1 enables rapid expression of ADAR1p150 upon IFN stimulation.

YTHDF1 affects A-to-I RNA editing levels
The effects of YTHDF1 on A-to-I RNA editing events were assessed in YTHDF1 knockdown cells. As inosine base pairs with cytosine, A-to-I RNA editing levels can be measured as an Ato-G conversion in the cDNA by reverse transcription (RT) and PCR. We quantified RNA editing levels [G / (A + G) (%)] at selected editing sites (Fig 2A) from direct sequencing chromatograms of the RT-PCR products (Fig 2B). In the transcripts coding TNKS, EIF2AK2, and PSMB2, A-to-I RNA editing levels were increased by IFN treatment due to the induction of ADAR1p150 (Fig 2B and 2C). YTHDF1 deficiency significantly reduced the IFN-mediated induction of editing levels at these sites ( Fig 2B and 2C), which is consistent with the attenuation of ADAR1p150 induction by YTHDF1 knockdown (Fig 1C). To illustrate a comprehensive picture of IFN-induced changes in A-to-I RNA editing levels, we performed RNA-seq analysis in YTHDF1 knockdown cells with 3 biological replicates. For quantification of editing levels, the numbers of A and G bases at the known editing sites were counted from the aligned RNA-seq reads. Previous studies identified tens of thousands of A-to-I RNA editing sites in the human transcriptome, which can be found in the databases of REDIportal and RADAR [45,46]. Among these sites, 488 sites were edited in A172 glioblastoma cells according to our criteria (S2 Table), and about half of those editing sites (261 sites) showed more than 5% increase of the average editing levels after IFN stimulation (Fig 2D and 2E). About half of them (130 sites) exhibited more than 5% decrease of their average editing levels with YTHDF1 knockdown (Fig 2D and 2E), indicating that a large part of the IFN-induced editing events depend on the YTHDF1-mediated ADAR1p150 induction.

YTHDF1 enhances interferon responses
Recent studies have shown that ADAR1-mediated A-to-I RNA editing disrupts secondary structures of dsRNA to prevent self-activation of cytosolic viral dsRNA sensors that induce IFN production and downstream ISGs, contributing to antiviral immunity [23]. These observations prompted us to explore potential roles of YTHDF1 in the IFN signaling. YTHDF1 knockdown significantly enhanced the induction of not only interferon beta 1 (IFNB1) but also interferon lambda 1 (IFNL1) and IFNL3 mRNAs following stimulation with IFN-α ( Fig 3A). Consistent with this, the IFN-α-induced secretion of the IFN-β protein was higher in YTHDF1-deficient cells than that in control cells, whereas IFN-β was not detected in unstimulated cells (Fig 3B). IFNB1 mRNA is known to be m 6 A modified, and its stability could be affected by m 6 A in certain cells [47,48]. However, decay rates of IFNB1 and IFNL1 transcripts were not altered with YTHDF1 knockdown in A172 cells (S2A Fig). Thus, the up-regulation of IFN-β is not caused by the m 6 A-dependent regulation of the IFNB1 mRNA stability, but rather by signaling pathways upstream of IFNB1 transcription in this cell line. During dsRNA-sensing response, the transcription of IFNB1 is under the control of the transcription factor IFN regulatory factor 3 (IRF3) that is activated by TANK-binding kinase 1 (TBK1). We found that phosphorylation of TBK1, a hallmark of TBK1 activation, was drastically increased by YTHDF1 knockdown following stimulation with IFN-α (Figs 3C and S2B), demonstrating Reduced IFN-induced A-to-I RNA editing was observed with YTHDF1 knockdown. Two-tailed Student t tests were performed to assess the statistical significance of differences between groups, � p < 0.05, �� p < 0.01, ��� p < 0.001, n.s. p ≧ 0.05, N.D. means not detected. n = 3 for all experiments. Data are presented as the mean ± SEM. (D) A heatmap of over editing levels from 3 biological replicates at 488 selected editing sites. Editing level changes were shown in each row covering all editing sites, ranked by the editing level. High editing levels were displayed in yellow and low in blue. An overall IFN-induced A-to-I RNA editing levels were reduced with YTHDF1 knockdown. (E) A flowchart for identification of IFN-induced editing sites. The numerical values for this figure are available in S1 Data. A-to-I RNA editing, adenosine-to-inosine RNA editing; IFN, interferon; KD, knockdown; n.s., not significant; RT-PCR, reverse transcription PCR; SEM, standard error of the mean; siRNA, small interfering RNA. https://doi.org/10.1371/journal.pbio.3001292.g002

PLOS BIOLOGY
that YTHDF1 depletion could activate dsRNA-sensing pathway that induces IFNB1 transcription.
The type I IFNs including IFN-α and IFN-β trigger STAT1 phosphorylation and subsequently initiate transcription of various ISGs. Our RT-qPCR analysis and RNA-seq data revealed that the expression levels of ISGs transcripts were highly induced by IFN-α treatment in YTHDF1 knockdown cells (Fig 3D and 3E, S3 Table), as a result of high IFN-β secretion. As expected, enhanced induction of STAT1 phosphorylation by IFN-α treatment was observed in YTHDF1 knockdown cells (Figs 3C and S2B). The cytosolic dsRNA-sensing pathway also causes activation of NF-κB, which induces transcription of a variety of genes involved in immune and inflammatory responses. We found that the expression levels of several NF-κB-regulated genes were increased by YTHDF1 knockdown with IFN-α stimulation in our RNA-seq data ( Fig 3F and S3 Table) and RT-qPCR analysis ( Fig 3G). The transcriptional activation of NF-κB is regulated by its interaction with inhibitory modulator IκBα that retains NF-κB in the cytoplasm. In response to various pathways including dsRNA signaling, IκBα is phosphorylated and degraded via the ubiquitin-proteasome system. Phosphorylation of NF-κB is also known to contribute to its activation. The phosphorylation of IκBα and NF-κB were increased in YTHDF1 knockdown cells following stimulation with IFN-α (Figs 3H and S2B), indicating that NF-κB signaling pathway is also activated by YTHDF1 deficiency under the condition of IFN treatment.

Enzymatic activity of ADAR1 is required to enhanced interferon responses in YTHDF1-deficient cells
To test whether the enhanced IFN responses in YTHDF1 knockdown cells were caused by the reduction of ADAR1p150 expression, we generated stable A172 cell lines that express control EGFP, ADAR1p150, and catalytically inactive mutant of ADAR1p150 E912A , which is previously established [49] (Fig 4A), respectively. Overexpression of wild-type ADAR1p150 exhibited significantly enhanced A-to-I RNA editing activity, while overexpression of ADAR1p150 E912A did not increase editing activity (S2C Fig). Strikingly, the enhanced IFNB1 induction was completely abolished by overexpression of ADAR1p150 ( Fig 4B). Similarly, overexpression of ADAR1p150 suppressed IFNL1, IFNL3, and other ISGs following stimulation with IFN ( Fig 4B). These inhibitory effects were abrogated by overexpression of a catalytic inactive mutant ADAR1p150 E912A ( Fig 4B). These data indicate that the aberrant activation of IFN pathways by YTHDF1 knockdown following IFN treatment is largely due to the defects of ADAR1p150 induction and its A-to-I RNA editing activity.
Furthermore, we investigated the contribution of downstream components of ADAR1-regulated signaling pathway. It is known that ADAR1 destabilizes secondary structure of endogenous RNA to prevent self-activation of the dsRNA sensors, retinoic acid-inducible gene I (RIG-I)-like receptors (RLRs) such as MDA5 [23]. RLR activation recruits and activates mitochondrial antiviral-signaling protein (MAVS), leading to phosphorylation of TBK1 and downstream transcription of IFN genes. To examine the effect of this signaling pathway, we used a TBK1 inhibitor (BX795) that effectively inhibits the induction of IFNs genes upon stimulation with synthetic analog of dsRNA, poly (I:C), in A172 cells (S2D Fig). Pretreatment of YTHDF1 knockdown cells with BX795 attenuated the induction of IFNs genes and downstream ISGs expression following IFN-α stimulation (Fig 4C). In addition, concomitant knockdown of MDA5 and YTHDF1 markedly suppressed the induction of IFNs genes and ISGs (Figs 4D and S2E). Taken together, our results indicate that the YTHDF1-mediated induction of ADAR1p150 and its editing activity are critical for preventing undesirable activation of MDA5, which could cause phosphorylation of TBK1 and excessive downstream IFN production during IFN response.

YTHDF1 affects cellular interferon-inducible cell growth and apoptosis
To elucidate the role of YTHDF1 in cellular response to IFN stimulation, YTHDF1 was stably knocked down by short hairpin RNA (shRNAAU  The signals were normalized to GAPDH and then normalized to control siRNA samples. Two-tailed Student t tests were performed to assess the statistical significance of differences between groups, � p < 0.05, �� p < 0.01, ��� p < 0.001. n = 3 for all experiments. Data are presented as the mean ± SEM. The numerical values for this figure are available in S1 Data. Ato-I RNA editing, adenosine-to-inosine RNA editing; IFN, interferon; ISG, IFN-stimulated gene; RT-qPCR, quantitative reverse transcription PCR; SEM, standard error of the mean; siRNA, small interfering RNA.  (S4D-S4F Fig). On the other hand, IFNs genes were not expressed in HeLa and HEK293T cells even though treated with IFN-α, suggesting that YTHDF1 does not affect IFN pathway in HeLa and 293T cells. Taken together, our results imply the presence of a cell type-specific mechanism that YTHDF1 promotes IFN-mediated ADAR1p150 induction and regulates IFN responses. Although the reason that only A172 and LN229 glioblastoma cells exhibit this regulation remain elusive, it might be possible that these differences are due to cell context differences such as expression levels of YTHDF1 protein, m 6 A modification levels of ADAR1p150 mRNA, or sensitivity to dsRNA detection.

YTHDF1 knockdown inhibits viral replication in cells
We next investigated a YTHDF1 knockdown effect on viral replication in A172 cells to demonstrate a physiological role of YTHDF1 in antiviral activities of IFN pathway. YTHDF1 knockdown cells were infected with recombinant GFP-expressing vesicular stomatitis virus (rVSV-GFP), a prototypic nonsegmented negative-stranded RNA virus whose replication is highly sensitive to the antiviral activities of IFN. We found that GFP expression was markedly reduced in YTHDF1 knockdown cells (S5A Fig). This reduction in viral replication was also observed in the decreased expression levels of viral G protein (S5B and S5C Fig), viral genome and the replicative intermediate antigenome RNA (Fig 5A), and the release of infectious virus (Fig 5B). YTHDF1 knockdown significantly increased the expression levels of IFN genes, ISGs, and NF-κB-regulated genes at early time points after VSV infection (Fig 5C). These results indicated that YTHDF1 depletion enhances cellular antiviral activities through increasing IFN responses during viral infection.
We also examined VSV replication in stable cell lines expressing ADAR1p150 or ADAR1-p150 E912A upon YTHDF1 knockdown to investigate contributions of ADAR1 and A-to-I RNA editing activity to this antiviral activity. Overexpression of ADAR1p150 significantly attenuated the YTHDF1-mediated suppression of viral genome RNA expression (Fig 5D) and the release of infectious virus (Fig E). In addition, overexpression of ADAR1p150 slightly recovered the reduced expression levels of viral G protein caused by YTHDF1 knockdown, though the effect was marginal and observed only at 12 h after VSV infection (S5D and S5E Fig). It is

PLOS BIOLOGY
possible that YTHDF1 can directly affect viral protein expression because a very recent study showed that VSV viral transcripts have m 6 A sites [50], and our study revealed that YTHDF proteins affect the expression levels of viral G protein through binding to m 6 A sites in their mRNA [51]. The enhanced mRNA expression of IFN genes, ISGs, and NF-κB-regulated genes were partially suppressed by ADAR1p150 overexpression (Fig 5F), which is consistent with the increased viral replication in ADAR1p150-overexpressing cell lines. These effects were not observed or diminished by overexpression of mutant ADAR1p150 E912A (Figs 5D-5F and S5D and S5E). Our results demonstrate ADAR1p150 and its A-to-I RNA editing activity as a significant mediator for proviral function of YTHDF1.

Discussion
IFN family members are central regulatory cytokines associated with protective immune response to infection, whereas an aberrant activation of IFN pathways cause autoimmune diseases such as AGS. Therefore, fine-tuned production of IFNs and balanced activation of the subsequent responsive genes are necessary for proper physiological activity. The present study showed that YTHDF1 is an important suppressor of IFN responses, which may prevent cells from the harmful effects of overactivated IFN signaling and downstream ISGs. YTHDF1 promotes expression levels of ADAR1p150 whose mRNA is m 6 A modified and is induced by IFN stimulation. Some IFNinduced transcripts have A-to-I RNA editing sites, indicating the presence of double-stranded structure in these transcripts (Figs 2B and S6A-S6C). Although we have not yet clarified key dsRNA-containing transcripts, which are responsible for MDA5 recognition, these IFN-inducible dsRNAs can be potential activator for dsRNA-sensing pathway, resulting in positive feedback of IFN responses. Our data revealed that the ADAR1-mediated A-to-I RNA editing, facilitated by the YTHDF1-mediated up-regulation, is important for inhibiting this feedback and preventing excessive activation of IFN responses. Therefore, functional regulation of YTHDF1 may serve to a promising therapeutic strategy for autoimmune diseases associated with high IFN signatures.
Translation up-regulation by YTHDF1 has been confirmed by various groups in diverse biological systems [10,43,44,[52][53][54][55]. It can be induced by stimulation as we have shown recently [43]. A recent study has questioned whether YTHDF1 affects translation of m 6 Amodified mRNA in HeLa cells [56]. However, the cumulative fraction analysis from the overall ribosome profiling data would have missed YTHDF1 regulation on its individual targets. Indeed, our recent QTL analysis revealed that depletion of m 6 A causes not only reduction but also increase of translation efficiency in transcripts harboring m 6 A sites [11]. Translation regulation of m 6 A-modified mRNA can be heterogeneous in HeLa cells; however, translation upregulation of YTHDF1-bound transcript in HeLa cells is clear [10].
One of the important physiological functions of IFN is the antiviral immune responses to viral infection. Recent studies indicated that m 6 A modification can play either proviral [47,48,50,51,57] or antiviral [44,58] roles in virus life cycle. A series of m 6 A writers, readers and erasers were increased upon human cytomegalovirus (HCMV) infection [47,48]. In A172 cells examined in this current study, the protein expression levels of METTL3, METTL14, YTHDF1, and YTHDF2 were not affected by IFN-α stimulation (Figs 1C and S1E), suggesting that the induction of m 6 A effectors may not be caused by virus-induced type I IFN pathways in A172 cells. In addition, several papers showed that the expression levels of not only IFNB1 [47,48] but also various ISGs [44] are regulated by the m 6 A-mediated mechanisms. Based on the results that the antiviral effects of YTHDF1 knockdown were not completely recovered by overexpression of ADAR1p150 (Fig 5D and 5E), there could be other m 6 A-regulated ISGs that are responsible for the effects of YTHDF1 knockdown upon viral infection. It should be noted that YTHDF1 does not directly affect IFNB1 mRNA stability in A172 cells (S2A Fig). We also demonstrated that A172 and LN229 glioblastoma cells showed YTHDF1-mediated suppression of IFN responses, while HeLa and 293T cells did not (S4A-S4C Fig). We believe that these differences are due to cell context differences such as expression levels of YTHDF1 protein, m 6 A modification levels of ADAR1p150 mRNA, or sensitivity to dsRNA detection. In fact, our ongoing work does suggest that the relative expression levels of YTHDF1 and other RNA-binding proteins critically impact translation regulation through YTHDF1, which we will report in the future.
A recent study found that m 6 A modification negatively regulates A-to-I RNA editing through inhibiting the association of m 6 A-modified mRNAs with ADAR enzymes [37]. They showed that knockdown of m 6 A writers did not affect expression of ADAR enzymes under normal cellular conditions. We revealed that, upon IFN stimulation, m 6 A appears to affect ADAR1p150 translation and A-to-I RNA editing. We have previously proposed that YTHDF1 may enable generation of a pulse of protein synthesis through translation promotion [4]. The present study suggested a possible mechanism that this type of m 6 A-mediated regulation on A-to-I RNA editing restricts the extent and length of the IFN responses to balance toxic effects on cells during viral infection or IFN stimulation. Interestingly, the m 6 A regulation of ADAR1p150 expression does not seem to be facilitated by YTHDF2 upon IFN stimulation (S1E Fig). In contrast, we observed that YTHDF2 affects ADAR2 expression but not ADAR1, though its physiological role remains to be investigated in the future. We currently do not have any clue about how YTHDF proteins achieve selectivity of functions among transcripts. While m 6 A could affect A-to-I RNA editing, an examination of the A-to-I RNA editing databases REDIportal and RADAR [45,46] also revealed numerous RNA editing sites in the transcripts encoding m 6 A-related genes such as METTL3, METTL16, YTHDF3, YTHDC1, YTHDC2, and ALKBH5. A-to-I RNA editing is known to regulate expression, stability, and alternative splicing of target mRNA, suggesting that it is possible that ADAR enzymes could conversely affect m 6 A methylation through modifying these m 6 A-related transcripts. Crosstalk between A-to-I RNA editing and m 6 A methylation might be more complicated, in particular upon immune stimulation or other stress responses.
The IFN family members play an important role in not only antiviral innate immunity but also cell proliferation, adaptive immunity, and cancer therapy [59]. In certain types of cancers, elevated ADAR1 expression and consequent hyperediting are caused by enhanced IFN signatures due to inflammatory environment of cancers, which drives cancer progression [60]. Several recent reports found that tumor cells exhibit a vulnerability to ADAR1 deficiency [33,34]. Loss of ADAR1 overcomes a resistance to a checkpoint that normally reduces sensing of endogenous dsRNAs, leading to IFN-mediated inflammation and increase in the efficacy of immunotherapy [35]. Furthermore, Our previous report found that YTHDF1 suppresses CD8 + T cell antitumor response through controlling cross-presentation of tumor antigens [55]. These reports and the present study together suggest that developing selective inhibitors of YTHDF1 function may also be immunotherapy strategies for certain tumors through enhancing IFN responses in tumor cells.

Plasmids, site-directed mutagenesis, viruses, and primers
For production of lentiviral particles, pLJM1-EGFP (19319), packaging vector psPAX2 (12260), and envelope vector pMD2.G (12259) were purchased from Addgene. For the construction of the pLJM1 plasmid encoding ADAR1p150, full-length human ADAR1p150 cDNA was amplified by PCR using the primers containing NheI and EcoRI restriction sites and inserted into pLJM1-EGFP. A mutation within ADARp150 (E912A) was introduced by site-directed mutagenesis as described previously [49]. The MISSION lentiviral shRNA plasmids encoding nontarget shRNA (SHC002), YTHDF1 shRNA (TRCN0000286871), and MDA5 shRNA (TRCN0000050849) were purchased from SIGMA. Lentiviruses were generated by transfection of 293T cells with lentiviral vectors, psPAX2 and pMD2.G. Lentiviral supernatants were collected at 48 and 72 h posttransfection, filtered, and added to A172 cells for 12 h. Stable transformants were selected with 1 μg/ml of puromycin and confirmed by immunoblotting for ADAR1 or YTHDF1. rVSV-GFP was generated by reverse genetics technique from the parental strain of VSV Indiana serotype in Li's lab. A172 cells in 24-well plates were transfected with YTHDF1 siRNA or control siRNA, and 3 d later, the cells were infected with rVSV-GFP with the multiplicity of infection (MOI) of 0.1 at 37˚C for 1 h. The virus inoculum was then removed, and 500 μL of DMEM supplemented with 2% FBS was added in each well. At the indicated time points postinfection, cells were photographed under the microscope, culture medium was collected for virus titration by plaque assay, and cells were lysed with RIPA buffer (Abcam, ab156034) for western blot sampling or TRizol reagent (Life Technologies, Carlsbad, California) for total RNA extraction.

VSV plaque assay
Vero CCL81 cells were seeded in 6-well plates, and, at the next day, the confluent monolayer cells were inoculated with 0.5 ml 10-fold serial dilutions of collected samples in FBS-free DMEM at 37˚C for 1 h. The inoculum was then replaced with 2 ml of DMEM overlay containing low-melting agarose (0.25% w/v), 5% FBS, 0.12% sodium bicarbonate, 100 μg/ml of streptomycin, 100 U/ml penicillin, 25 mM HEPES (pH 7.7), and 2 mM L-glutamine. After incubation at 37˚C for 2 d, cells were fixed with 4% neutral buffered formaldehyde for 2 h. The overlay was then removed, and the cells were stained with 0.05% crystal violet to visualize the plaques. Virus titer was calculated with the formula of: Titer (Log 10 PFU/ml) = Log 10 (Plaque number / 0.5 × dilution).

IFN-β ELISA
Spontaneous IFN-β production was detected by the VeriKine-HS Human IFN Beta Serum ELISA Kit (PBL Assay Science, 41515-1) according to the manufacturer's instructions. A172 cells were seeded at a density of 70,000 cells per well in a 24-well plate. Next day, the cells were transfected with siRNAs by using the Lipofectamine RNAiMAX (Invitrogen). TheAU : PerPLOSstyle; n cells were treated with 1,000 U/ml human IFN-α, 48 h after siRNA transfection. TheAU : PerPLOSstyle; numerals cells were gently washed once with medium, then 350 μl fresh warm medium was replaced in the wells, 30 h after IFN-α stimulation. After 72 h of incubation, 50 μl of the supernatant was collected and the concentration of IFN-β was measured by the kit.

RNA immunoprecipitation
The RNA immunoprecipitation (RIP) assay was performed as described previously [61] with minor modifications. Briefly, protein A Dynabeads (Invitrogen, 40 μl per 5 μg antibody) was washed with PBST buffer (PBS containing 0.05% Tween 20) 3 times and resuspended in 40 μl PBST buffer. The beads were incubated with 5 μg of control rabbit IgG (R&D SYSTEMS, AB105C), YTHDF1 (ProteinTech, 17479-1-AP) or YTHDF2 (ProteinTech, 24744-1-AP) antibody in 150 μl PBST buffer at 4˚C overnight. A172 cells were collected by gentle scraping, pelleted by centrifuge for 5 min at 1,000g. The cell pellet was resuspended with lysis buffer (150 mM KCl, 10 mM Tris-HCl (pH 7.5), 2 mM EDTA, 0.5% NP-40, 0.5 mM DTT, 1% protease inhibitor cocktail, 400 U/ml RNase inhibitor), pipetted up and down several times. The mRNP lysate was incubated with rotation for 1 h at 4˚C and centrifuged at 15,000g for 15 min to clear the lysate. An aliquot from each lysate was saved as input, mixed with TRIzol LS Reagent (Invitrogen). The cell lysate was mixed with antibody-coupled beads and rotated continuously for 3 h at 4˚C. The beads were washed with 500 μl NT2 buffer (200 mM NaCl, 50 mM Tris-HCl (pH 7.5), 2 mM EDTA, 0.05% NP-40, 0.5 mM DTT, 0.1% protease inhibitor cocktail, 40 U/ml RNase inhibitor) 5 times and then resuspended in 90 μl ice-cold NT2 buffer. After the final wash, 10 μl of the beads were used for western blotting, and the remaining 80 μl were digested by proteinase K. The supernatant was mixed with TRIzol LS Reagent and saved as the IP sample. Total RNA isolated by TRIzol LS reagent was analyzed by RT-qPCR.

Quantification of m 6 A level using anti-m 6 A antibody
The m 6 A-IP enrichment assay followed by RT-qPCR was performed as described previously [61]. Briefly, 3 μg purified cytosol mRNA extracted from A172 cells was incubated with 5 μg of control rabbit IgG (R&D SYSTEMS, AB105C) or m 6 A antibody (Synaptic Systems, 202003) for 4 h at 4˚C, then pulled down by protein A Dynabeads (Invitrogen, 10008D) for 2 h at 4˚C. RNA was extracted from the bead fractions by TRIzol LS Reagent, then subjected to RT-qPCR. HPRT1 gene was used as a reference gene when carrying out RT-qPCR.

RNA preparation and RNA-seq
For quantitative RT-PCR and direct sequencing, total RNA was isolated by using RNeasy Mini Kit (QIAGEN) from cell lysate according to the manufacturer's protocol. For quantitative RT-PCR in the samples of VSV-infected cells, total RNA was extracted from TRizol-lysate and dissolved in 50 μl water. For RNA-seq experiments, total RNA was isolated by using TRIzol reagent (Invitrogen) from cell lysate, and poly (A)+ RNA was purified from the total RNA by using Dynabeads mRNA DIRECT Purification Kit (Invitrogen). The RNA libraries were prepared using Truseq stranded mRNA sample preparation kit (Illumina) according to the manufacturer's protocol. Three biological replicates were sequenced on an Illumina HiSeq 4000 sequencer (50 bp, single end) for each cell lines.

Quantitative RT-PCR (RT-qPCR)
To quantify expression levels of transcripts, total RNA was reverse transcribed by Go Script Reverse Transcriptase (Promega, A6010) with anchored (dT)15 primers and random oligo primers. The cDNA was subjected to real-time PCR (StepOnePlus Real-Time PCR Systems: Applied Biosystems) by using GoTaq Master Mix (Promega, A6010) with the gene specific primers (S1 Table). Relative changes in expression were calculated using the ΔΔCt method. To quantify expression levels of VSV genomic RNA, 2 μl of total RNA was used for cDNA synthesis in 20 μl RT system with primer annealed to 3 0 -end leader sequence of VSV genome (VSV genome-RT). To quantify expression levels of VSV antigenomic RNA, 2 μl of total RNA was used for cDNA synthesis with primer annealed to 5 0 -end trailer sequence of VSV genome (VSV antigenome-RT). The same pair of primers annealing to G gene were used for RT-qPCR for VSV genome and antigenome with 2 μl of cDNA. The Ct values were converted into genomic RNA or total viral RNA copies according to the standard curve generated with serial dilutions of VSV full-length genome plasmid. GAPDH (for mRNA expression level) and HPRT1 (for m 6 A-IP) were used as internal controls.

Direct sequencing
A-to-I RNA editing level was measured as described previously [31]. Briefly, total RNA was reverse transcribed by Go Script Reverse Transcriptase with anchored (dT)15 primers and random oligo primers. The cDNAs were amplified by Phusion Hot Start II High-Fidelity DNA Polymerase (Thermo Scientific, F549) with gene specific primers (S1 Table, Fw1 and Rv1). Second PCR reactions were performed with the second sets of gene specific primers (S1 Table, Fw2 and Rv2) and with the first RT-PCR products (diluted 80-fold) as templates. Subsequently, the amplified cDNA fragments from RT-PCR were treated with ExoSAP-IT (USB) and directly sequenced with the gene specific primer, Fw2 or Rv2. The editing levels were quantified by measuring heights of A peaks (unedited) and G peaks (edited) and calculating percentage of the population edited at each site (100% × [G height / (A height + G height)]).

Measuring decay rates of mRNAs
Cells were transfected with control siRNA and YTHDF1 siRNA by using the Lipofectamine RNAiMAX. TheAU : PerPLOSstyle; numeralsarenotallowedatthebeginningofasentence:Therefore; plea cells were treated with 1,000 U/ml IFN-α for 30 h, 72 h after siRNA transfection. The cells were treated with 1 μg/ml Actinomycin D for time periods specified in S2A

Cell proliferation and apoptosis assay
For cell proliferation assay, 2,500 cells of the stable cell lines expressing control shRNA or YTHDF1 shRNA were seeded per well in a 96-well plate with mock treatment or IFN-α stimulation. The cell proliferation was measured by using Cell Counting Kit-8 (DojindoAU : PerPLOSstyle; In Molecular Technologies, CK04) at various time points according to the manufacturer's protocols. For each cell line tested, the signal from the assay was normalized to the value obtained from 24 h after seeding.
For apoptosis assay, 30,000 cells of the stable cell lines expressing control shRNA or YTHDF1 shRNA were seeded per well in a 24-well plate with mock treatment or IFN-α stimulation. After 5 d of incubation with IFN-α, apoptotic cells were stained with a MEBSTAIN Apoptosis TUNEL Kit Direct (MBLAU : PerPLOSstyle; Co:shouldnotbeusedincompanynamesexceptasa International, 8445) according to the manufacturer's protocol, and the stained cells were examined by flow cytometry.

RNA-seq data processing
Sequenced reads were mapped to the human genome by HiSAT2 [62] (version 2.1.0) with parameters "-score-min L,0,-0.6." For estimation of gene expression levels, the mapped reads of each sample were assembled in a reference-based approach, and transcripts per million (TPM) was calculated for each genes by using StringTie [63] (version 2.0). Differentially expressed genes were detected by R package DESeq2 [64] (version 1.24.0) using Wald test. The significantly differentially expressed genes between control siRNA and YTHDF1 knockdown were reported at adjusted p-value cutoff of 0.05. Heatmaps were generated by heatmap2 in gplots package of R.

Quantification of A-to-I RNA editing levels
Sequenced reads were mapped to the human genome by HiSAT2 [62] (version 2.1.0) with parameters "-score-min L,0,-0.6." Known A-to-I RNA editing sites in the human transcripts were obtained from 2 databases, DARNED and REDIportal [45,46]. A-to-I RNA editing levels for the list of known editing sites were calculated by using REDItools [65] algorithm REDItool-Known.py with parameters "-v 0 -n 0 -m 60." Among 4,688,186 editing sites, 6,588 sites fulfilled our first criteria (termed "Depth filter"): minimum depth of coverage >10× (more than 10 reads at every samples in our RNA-seq data). Among the 6,588 sites, 488 sites fulfilled our second criteria ("Editing filter"): maximum average editing level >5%. We also excluded the editing sites exhibiting minimum average editing level <95%, because it could be caused by genomic mutation in this cell line. An average editing level was defined as "increased/ decreased" if a difference between control and IFN-α stimulation/YTHDF1 knockdown was more than 5%. Heatmaps were generated by heatmap2 in gplots package of R.

Quantification and statistical analysis
Two-tailed Student t tests were performed to assess the statistical significance of differences between groups, � p < 0.05, �� p < 0.01, ��� p < 0.001, n.s. p ≧ 0.05. N.D. means not detected. n = 3 for all experiments. Data are presented as the mean ± standard error of the mean (SEM).

Reference sequence and annotation data
The human genome sequence (GRCh38/hg38), the annotated gene models, and the annotations of rRNAs were obtained from the UCSC genome browser database. (G) IFN-induced apoptosis measured by TUNEL assay in stable YTHDF1 and MDA5 knockdown cells following IFN-α stimulation. (B-G) Two-tailed Student t tests were performed to assess the statistical significance of differences between groups, � p < 0.05, �� p < 0.01, ��� p < 0.001, n.s. p ≧ 0.05. n = 3 for all experiments. Data are presented as the mean ± SEM. The numerical values for this figure are available in S1 Data. IFN, interferon; n. s., not significant; RT-qPCR, quantitative reverse transcription PCR; SEM, standard error of the mean; shRNA, short hairpin RNA. showing knockdown effect of YTHDF1 on the expression of VSV-G protein at 12 and 20 h after rVSV-GFP infection in stable cell lines expressing control EGFP, wild-type ADAR1p150, or catalytically inactive mutant of ADAR1p150 E912A , respectively. Immunoblot images are representative of 3 biological replicates. The signals were normalized to control siRNA samples. (C, E) Two-tailed Student t tests were performed to assess the statistical significance of differences between groups, � p < 0.05, �� p < 0.01, ��� p < 0.001, n.s. p ≧ 0.05. n = 3 for all experiments. Data are presented as the mean ± SEM. The numerical values for this figure are available in S1 Data. MOI, multiplicity of infection; n.s., not significant; rVSV-GFP, recombinant GFP-expressing vesicular stomatitis virus; SEM, standard error of the mean; siRNA, small interfering RNA.  -PCR products of RSAD2, STAT1, TRIM56, IFI44L, CD74, TAP1, CFLAR, and ZC3HAV1 mRNA following IFN-α stimulation. (D) A172 cells were transfected with YTHDF1 siRNA or control siRNA and treated with IFN-α. The cells were then stained with J2 antibody and fluorescent secondary antibody and analyzed by flow cytometry. (E) A schematic model showing YTHDF1-mediated regulation of IFN responses. (A, B, D) Two-tailed Student t tests were performed to assess the statistical significance of differences between groups, � p < 0.05, �� p < 0.01, ��� p < 0.001. n = 3 for all experiments. Data are presented as the mean ± SEM. The numerical values for this figure are available in S1 Data. A-to-I RNA editing, adenosine-to-inosine RNA editing; dsRNA, double-stranded RNA; IFN, interferon; IRF3, IFN regulatory factor 3; m 6 A, N 6 -methyladenosine; RT-qPCR, quantitative reverse transcription PCR; SEM, standard error of the mean; siRNA, small interfering RNA; TBK1, TANK-binding kinase 1; TPM, transcripts per kilobase million. (TIF) S1 Table. RT-qPCR primers, siRNAs, antibodies, and direct-sequencing primers. (XLSX) S2 Table. A-to-I RNA editing levels in RNA-seq analysis. (XLSX) S3 Table. Gene expression profiles in RNA-seq analysis.