Main

More than 140 million people around the world have been affected by coronavirus disease 2019 (COVID-19) caused by SARS-CoV-2. COVID-19 varies from mild to severe, life-threatening respiratory infection with coagulopathy. Most (81%) people infected with SARS-CoV-2 show a mild and self-limited course3, while severe cases of disease are more likely to be present in older patients with underlying comorbidities, such as chronic obstructive pulmonary disease (COPD)4,5,6, compared to patients with mild disease. But even young healthy adults sometimes experience severe illness. Conversely, it has been reported that nearly 40–45% of patients with SARS-CoV-2 infections are asymptomatic7. The wide spectrum of clinical manifestations of COVID-19 suggests that individual immune responses to the underlying pathogen may play some crucial role in determining the clinical course. Currently, no efficient therapies and preventive measures exist for COVID-19, thus studies about host immune response against SARS-CoV-2 infection are required for a better understanding of the pathological processes for the rational development of countermeasures to control SARS-CoV-2 infection. There is also an urgent need to identify biomarkers that can predict which patients will deteriorate.

Microbial invasion in our body is sensed by pattern-recognition receptors (PRRs) that are present in most types of cells, which initiate the activation of cell-intrinsic defense and innate immune responses. During RNA virus infection, virus-derived RNA species are mainly targeted by certain subsets of the PRRs. Among such PRRs, RIG-I, melanoma differentiation-associated gene 5 (MDA5) and toll-like receptor 3 (TLR3) play a central role in activation of the downstream signaling pathways, leading to the inductions of type I/III IFNs and proinflammatory cytokines1,2. In particular, RIG-I is a critical cytosolic RNA sensor in the detection of a variety of RNA viruses such as influenza A virus (FluV)1,2. RNAs carrying a 5′-triphosphate modification (3pRNA) or short-type double-stranded RNAs (dsRNAs), which are found in viral RNA genomes or replication intermediates, are ligands for RIG-I (refs. 1,2). It has been shown that ligand binding to the RIG-I C-terminal domain (CTD) activates the ATPase activity of RIG-I to change its conformation, which enables RIG-I to interact with the adapter protein mitochondrial antiviral-signaling protein (MAVS; also known as IPS-1, VISA or Cardif) through its caspase activation and recruitment domains1,2,8. This, in turn, results in kinase-dependent activation of interferon regulatory factor-3 (IRF-3) and nuclear factor kappa B (NF-κB) transcription factors, leading to the activation of the downstream gene induction programs such as type I/III IFN inductions to confer an antiviral state in cells upon viral infection1,2. The innate immune mechanism operated by host PRRs during SARS-CoV-2 infection remains poorly understood, although reports suggest that SARS-CoV-2 and its predecessor SARS-CoV have subversion strategies against innate immune signaling through SARS-CoV-2-derived proteins9,10,11,12.

SARS-CoV-2 is an enveloped, positive-sense, single-stranded RNA betacoronavirus. Entry of SARS-CoV-2 into its target cells such as lung alveolar or bronchial cells is reported to depend on the binding of the viral spike protein to its host angiotensin-converting enzyme 2 (ACE2; ref. 13), in collaboration with two host-cell proteases, TMPRSS2 and ADAM17 (ref. 14). Recent reports have shown that an exacerbated inflammatory response was observed in patients with severe and critical disease15,16,17. On the other hand, no robust upregulation of IFN or other proinflammatory cytokines was observed in the sera of asymptomatic individuals with SARS-CoV-2 infection18. In this respect, we hypothesized that there must be some distinctive cell-autonomous innate defense mechanism against SARS-CoV-2 in these asymptomatic individuals. While there have been accumulating reports of adaptive immune responses against SARS-CoV-2 infection19,20, questions regarding the recognition and control of this virus by the innate defense system in lung epithelial cells remain largely unanswered.

We first investigated the innate cytokine responses during SARS-CoV-2 infection in primary human alveolar and bronchial epithelial cells, HPAEpiC and HBEpiC, respectively, and the human lung cancer cell line A549. Although both ACE2 and TMPRSS2 were expressed and sufficient infection was detected in these three cell types (Extended Data Fig. 1a,b), we hardly observed the mRNA induction of IFN-β, IFN-λ1, CXCL10, interleukin (IL)-6 and tumor necrosis factor (TNF) in response to SARS-CoV-2 infection (Fig. 1a). A similar observation was also obtained in other primary human cells including HBEpC and HCoEpiC (Extended Data Fig. 1c). Consistently, IRF-3 and IκBα were not phosphorylated either (Fig. 1b). In addition, the interaction of RIG-I with MAVS was detected upon infection with FluV but not SARS-CoV-2 (Fig. 1c). However, viral replication was suppressed in human cells even without activation of the innate response (Fig. 1d and Extended Data Fig. 1d), at least until 24 h after infection, while innate cytokines were highly induced with prominent viral replication upon FluV infection (Fig. 1a,e). Therefore, we speculated that there might be an unconventional innate defense mechanism to restrict viral propagation.

Fig. 1: SARS-CoV-2 replication is suppressed in human lung cells without activation of innate responses.
figure 1

a, Quantitative PCR with reverse transcription (RT–qPCR) analysis of IFNB1, IFNL1, CXCL10, IL6 or TNF mRNA induction by infection with SARS-CoV-2 (upper) or FluV (lower) in HPAEpiC, HBEpiC or A549 cells (n = 3 biologically independent samples). b, Immunoblot analysis of phosphorylated IRF-3 (IRF-3 (p.Ser 396)) and IκBα (IκBα (p.Ser 32)) in A549 cells infected with SARS-CoV-2 or FluV for 0, 2, 4 or 8 h. c, Immunoprecipitation (IP) of endogenous MAVS and RIG-I in A549 cells at 0, 2, 4 or 8 h after infection with SARS-CoV-2 or FluV. d,e, Copy number of SARS-CoV-2 RNA (d) or FluV RNA (e) at 0, 1, 4, 8 or 24 h after infection with SARS-CoV-2 or FluV in HPAEpiC, HBEpiC or A549 cells (n = 3 biologically independent samples). Data are presented as the mean values ± s.d. and are representative of at least two independent experiments with similar results. ND, not detected. RE, relative expression. WCL, whole-cell lysate.

Source data

Full size image

Because SARS-CoV-2 is an RNA virus, we next examined the involvement of innate RNA sensors in viral suppression in A549 cells. SARS-CoV-2 replication was significantly observed upon knockdown of RIG-I but not MDA5 and TLR3 (Fig. 2a). A similar observation was also obtained in HPAEpiC and HBEpiC cells (Fig. 2b). Consistently, RIG-I deficiency allowed SARS-CoV-2 to replicate after entry into the cell, without any effect on viral infection in both lines of RIG-I knockout (KO) A549 cells (RIG-I KO8 and RIG-I KO35; Fig. 2c and Extended Data Fig. 2a,b). The expression of SARS-CoV-2 spike protein and viral particles was clearly detected in RIG-I KO35 cells but not wild-type (WT) A549 cells (Fig. 2d,e). On the other hand, because RIG-I protein levels were highly decreased in A549 cells that were deficient in the RIG-I adapter MAVS (Extended Data Fig. 3a,b), SARS-CoV-2 replication could also be detected in MAVS KO cells comparably to RIG-I KO cells (Extended Data Fig. 3c). In fact, the re-expression of RIG-I alone in MAVS KO A549 cells restored suppression of viral replication in a dose-dependent manner (Fig. 2f). These data indicate that the underlying mechanism for this viral suppressing activity is based upon RIG-I but not MAVS. Conversely, viral suppression was not restored by the re-expression of MDA5 in MAVS KO cells (Fig. 2f), suggesting that this activity is specific to RIG-I itself.

Fig. 2: RIG-I-mediated suppression of SARS-CoV-2 replication does not require its canonical MAVS-dependent signaling.
figure 2

a,b, Copy number of SARS-CoV-2 RNA at 24 h after infection with SARS-CoV-2 in A549 (a), HPAEpiC or HBEpiC (b) cells transfected with short interfering RNAs (siRNAs) against gene products of RIG-I (siRIG-I), MDA5 (siMDA5) or TLR3 (siTLR3; n = 3 biologically independent samples). c, Copy number of SARS-CoV-2 RNA at 0, 1, 4, 8 or 24 h after infection with SARS-CoV-2 in A549 WT, RIG-I KO8 or RIG-I KO35 cells (n = 3 biologically independent samples). d, Immunostaining of spike protein in A549 WT or RIG-I KO35 cells uninfected or after 72 h of infection with SARS-CoV-2. Nuclei were counterstained with Hoechst 33342. Scale bar, 50 μm. e, Viral titers at 5 d after infection with SARS-CoV-2 in A549 WT or RIG-I KO35 cells (n = 3 biologically independent samples). f, Copy number of SARS-CoV-2 RNA at 24 h after infection with SARS-CoV-2 in A549 MAVS KO21 cells transfected with empty vector (–), Flag-tagged RIG-I or MDA5 expression vector (upper) (n = 3 biologically independent samples). Immunoblot analysis of Flag-tagged RIG-I and MDA5 (lower). g, Viral titers at 5 d after infection with SARS-CoV-2 in HBEpiC, HBEpC, HBEpC-COPD or PBEC-COPD (upper; n = 3 biologically independent samples). Immunoblot analysis of RIG-I, ACE2 or TMPRSS2 with the lysates of indicated cells (lower). h, Immunoblot analysis of RIG-I, ACE2 or TMPRSS2 with the lysates of indicated cells treated with ATRA (0, 0.01, 0.1 or 1 μM) for 24 h. i, Viral titers at 5 d after infection with SARS-CoV-2 in HBEpC-COPD or PBEC-COPD pretreated with ATRA (0, 0.01, 0.1 or 1 μM) for 24 h (n = 3 biologically independent samples). Data are presented as the mean values ± s.d. and are representative of at least two independent experiments with similar results. Statistical significance was determined by two-tailed Student’s t-test. PFU, plaque‐forming unit.

Source data

Full size image

It has been proposed that COPD is a risk factor for severe COVID-19 that leads to hospitalization and intensive care unit admission4,5,21,22. Another interesting report showed that RIG-I expression levels are downregulated in most lung cells derived from patients with COPD23. Consistently, we found that RIG-I expression was barely detected in human primary bronchial epithelial cells (HBEpC-COPD and PBEC-COPD) derived from two independent patients with COPD (Fig. 2g). Of note, viral particles could be detected at 5 d after infection only in HBEpC-COPD and PBEC-COPD but not healthy donor cells (Fig. 2g). Next, we tried to test the translatability of RIG-I protein upregulation for the control of SARS-CoV-2 infection. RIG-I was first identified as an upregulated gene in the acute promyelocytic leukemia cell line NB4 upon all-trans retinoic acid (ATRA) stimulation24, as described by its name. We found that the protein levels of RIG-I could be remarkably upregulated in a dose-dependent manner upon treatment with ATRA in these cells derived from patients with COPD, while ATRA did not affect the protein expressions of ACE2 and TMPRSS2 (Fig. 2h). Treatment with ATRA significantly reduced levels of SARS-CoV-2 viral titers in these cells (Fig. 2i), which was dependent on RIG-I (Extended Data Fig. 3d). These data suggest that RIG-I protein expression is an important parameter that modulates the anti-SARS-CoV-2 innate defense.

In accordance with these observations, RNA-immunoprecipitation (RIP) assay, as well as viral cross-linking and solid-phase purification (VIR-CLASP), revealed a specific interaction of RIG-I but not MDA5 with viral RNA in human lung cells after SARS-CoV-2 infection (Fig. 3a and Extended Data Fig. 3f). Further RIP and functional analyses showed that the interaction with viral RNA and viral suppressing activity were detected upon the re-expression of the C-terminal fragment of RIG-I (C-RIG-I) or the helicase domain (HD), but not the CTD in RIG-I KO A549 cells (Fig. 3b,c). In addition, these phenotypes were not affected upon re-expression of RIG-I K270A mutant, which has no ATPase activity26, whereas RIG-I Q299A mutant, which lacks dsRNA-binding activity27, could not bind to viral RNA with promoted viral replication (Fig. 3d,e). Moreover, the interaction of RIG-I with the SARS-CoV-2 RNA genome failed to stimulate RIG-I ATPase (Fig. 3f), which was also consistent with the observation that SARS-CoV-2 does not activate the conventional RIG-I signaling pathway (Figs. 1b,c and 2f). These results suggest that the RIG-I HD but not the CTD is required for the interaction with SARS-CoV-2 RNA to induce an inhibitory effect on viral replication, without activating the conventional downstream signaling pathways.

Fig. 3: The dsRNA-binding activity of RIG-I HD is required for SARS-CoV-2 suppression in a MAVS-independent manner.
figure 3

a, RIP assay with HPAEpiC, HBEpiC or A549 cell lysates prepared after 6 h of infection with SARS-CoV-2 by using anti-RIG-I, anti-MDA5 or control immunoglobulin. Immunoprecipitated SARS-CoV-2 positive-strand RNA was quantified by RT–qPCR and is represented as fraction of input RNA before immunoprecipitation (percentage of input; upper; n = 3 biologically independent samples). Immunoblot analysis of whole-cell lysate and immunoprecipitated RIG-I and MDA5 (lower). In addition, encephalomyocarditis virus (EMCV)-derived RNA was specifically detected in MDA5 immunoprecipitates but not RIG-I immunoprecipitates, which were derived from EMCV-infected A549 cells (Extended Data Fig. 3e), as previously reported25. b, RIP assay with A549 RIG-I KO35 cell lysates expressing Flag-tagged deletion mutants of RIG-I prepared after 6 h of infection with SARS-CoV-2 by using anti-Flag antibody. The immunoprecipitated SARS-CoV-2 positive-strand RNA was measured by RT–qPCR as described in a (upper; n = 3 biologically independent samples). Immunoblot analysis of immunoprecipitated Flag-tagged proteins (lower). c, Copy number of SARS-CoV-2 RNA at 24 h after infection with SARS-CoV-2 in A549 RIG-I KO35 cells expressing Flag-tagged deletion mutants of RIG-I (upper; n = 3 biologically independent samples). Immunoblot analysis of Flag-tagged deletion mutants of RIG-I (lower). d, RIP assay with A549 MAVS KO21 cell lysates expressing Flag-tagged point mutants of RIG-I prepared as described in b. The immunoprecipitated SARS-CoV-2 positive-strand RNA was measured by RT–qPCR as described in a and b (upper; n = 3 biologically independent samples). Immunoblot analysis of immunoprecipitated Flag-tagged proteins (lower). e, Copy number of SARS-CoV-2 RNA at 24 h after infection with SARS-CoV-2 in A549 MAVS KO21 cells expressing Flag-tagged point mutants of RIG-I (upper; n = 3 biologically independent samples). Immunoblot analysis of Flag-tagged point mutants of RIG-I (lower). f, ATPase activity of recombinant RIG-I protein after the addition of SARS-CoV-2 RNA genome or 3pRNA (n = 3 biologically independent samples). Data are presented as the mean values ± s.d. and are representative of at least two independent experiments with similar results. Statistical significance was determined by two-tailed Student’s t-test.

Source data

Full size image

We further tried to find the detailed mechanism for how RIG-I exerts a suppressing effect on SARS-CoV-2 replication. SARS-CoV-2 has a positive-sense, single-stranded RNA, which should be transcribed by viral RNA-dependent RNA polymerase (RdRp) into a negative-sense RNA in the first step of viral replication. Therefore, we hypothesized that RIG-I interaction with SARS-CoV-2 may block this RdRp-dependent process. As we expected, any negative-strand RNA was not detected in all the primary human cells tested as well as A549 cells (Fig. 4a,b and Extended Data Fig. 4a), whereas we detected transcribed negative-sense viral RNAs, together with increased levels of positive-sense viral RNAs that were further transcribed from them, in RIG-I KO cells but not IRF-3 KO cells, at 8 h or later after infection (Fig. 4a and Extended Data Fig. 4b,c). In support of this, positive-sense viral RNA was coprecipitated with RdRp only in the absence of RIG-I (Fig. 4c), which did not affect RdRp protein expression levels (Extended Data Fig. 4d). Recombinant RIG-I (rRIG-I) but not rMDA5 competitively inhibited the binding of RdRp to viral RNA, in a dose-dependent manner (Fig. 4d). We next determined which region(s) of the SARS-CoV-2 RNA genome is responsible for the interaction with RIG-I. The access of the RdRp to the 3′-untranslated region (3′ UTR) of the genome, which is conserved in SARS-CoV-2 in terms of RNA secondary structure and sequence28, is reported to initiate the negative-strand synthesis in other SARS-related betacoronaviruses29,30. This prompted us to check whether RIG-I binds to the 3′ UTR of positive-strand SARS-CoV-2 RNA, by using in vitro-transcribed (IVT) 3′ UTR and three other RNA fragments. As expected, rRIG-I protein was preferentially pulled down together with the 3′ UTR RNA fragment, as compared to the other IVT RNA fragments (Fig. 4e). To actually show the interaction of endogenous RIG-I with the 3′ UTR region of the viral genome, we performed two different experiments. The interaction of endogenous RIG-I with SARS-CoV-2 genome during infection in primary lung cells as well as A549 cells was selectively suppressed in the presence of the 3′ UTR IVT RNA but not the other IVT RNAs (Extended Data Fig. 4e). Consistently, RNase H protection was detected preferentially in the presence of the oligonucleotides targeting the 3′ UTR (P5, PK1, PK2, s2m and Oct; Fig. 4f). These results indicate that endogenous RIG-I detects SARS-CoV-2 through the interaction of its HD with the 3′ UTR of positive-strand viral RNA, which blocks the access of RdRp to viral RNA, shutting off SARS-CoV-2 replication.

Fig. 4: RIG-I directly counteracts the access of RdRp to the viral RNA genome to shut off viral replication.
figure 4

a, Copy number of SARS-CoV-2 positive-strand or negative-strand RNA at 0, 1, 4, 8 or 24 h after infection with SARS-CoV-2 in A549 WT, RIG-I KO8, KO35 or IRF-3 KO145 cells (n = 3 biologically independent samples). b, Copy number of SARS-CoV-2 positive-strand or negative-strand RNA at 0, 1 or 24 h after infection with SARS-CoV-2 in HPAEpiC (left) or HBEpiC (right) cells transfected with indicated siRNAs (n = 3 biologically independent samples). c, RIP assay with A549 WT or RIG-I KO35 cell lysates prepared after 6 h of infection with SARS-CoV-2 by using anti-RdRp. The immunoprecipitated SARS-CoV-2 positive-strand RNA was quantified by RT–qPCR and is represented as fraction of input RNA before immunoprecipitation (percentage of input; upper; n = 3 biologically independent samples). Immunoblot analysis of immunoprecipitated RdRp (lower). d, A549 RIG-I KO35 cell lysates prepared after 6 h of infection with SARS-CoV-2 were incubated with rRIG-I or rMDA5. The interaction of SARS-CoV-2 positive-strand RNA with RdRp was analyzed by RIP assay and RT–qPCR analysis as described in c (upper; n = 3 biologically independent samples). Immunoblot analysis of immunoprecipitated RdRp and Coomassie Brilliant Blue (CBB) staining of recombinant proteins (lower). e, RNA pull-down assay showing the binding activity of SARS-CoV-2 RNA genome or IVT RNAs to rRIG-I. f, RNase H protection of SARS-CoV-2 positive-strand RNA in RIG-I immunoprecipitates from A549 cell lysates prepared after 6 h of infection with SARS-CoV-2. RNA products were quantified by RT–qPCR and were represented as a fraction of RNase H–untreated RNA (percentage of input; upper). Purified total RNAs were used as a positive control (lower; n = 3 biologically independent samples). Oligonucleotide-targeted regions were designed in the 5′ UTR (SL2, SL3 and SL5), the 3′ UTR (P5, PK1, PK2, s2m and Oct; 35) or regions A–D (11670–11681, 11758–11769, 12303–12314 or 12382–12393 nucleotide regions of the SARS-CoV-2 RNA genome). Data are presented as the mean values ± s.d. and are representative of at least two independent experiments with similar results. Statistical significance was determined by two-tailed Student’s t-test.

Source data

Full size image

In this study, we identified RIG-I as an innate sensor for SARS-CoV-2 in primary human pulmonary epithelial cells. RIG-I recognizes the 3′ UTR of the viral RNA genome in a CTD-independent manner, but this hitherto unknown recognition misfires the activation of the downstream antiviral innate immune signaling, such as the induction of type I/III IFNs and proinflammatory cytokines, which could be explained by the finding that RIG-I ATPase is not stimulated by the interaction with the SARS-CoV-2 RNA genome. However, it is noted that this RIG-I-mediated sensing is sufficient to finally suppress viral propagation after entry into the cell, in an IFN-independent manner. Mechanistically, RIG-I directly exerts its antiviral activity via its competitive inhibition of the recruitment of viral RdRp to viral genome, which blocks the first step of the RdRp-dependent transcription process. In support of this, SARS-CoV-2 replication was detectable upon RIG-I deficiency or downregulation.

Deletion of RIG-I but not MDA5 also allowed primary lung cells and A549 cells to induce innate cytokines (IFNB1, IFNL1 and IL-6) and viral replication upon SARS-CoV-2 infection (Extended Data Fig. 5a–f). Such cytokine inductions observed in RIG-I KO A549 cells or RIG-I knockdown of primary lung cells were significantly suppressed by MDA5 knockdown (Extended Data Fig. 5d,g), which was associated with enhanced viral replication (Extended Data Fig. 5f,h). In relation to this, it was reported by Yin et al. that SARS-CoV-2-induced innate immune response is dependent on MDA5 in human lung adenocarcinoma cell line Calu-3 cells31, which was consistent with our results using Calu-3 cells (Extended Data Fig. 6a,b). In this respect, we speculated that the reason why we see no IFN and cytokine induction in the primary cells and A549 cells (whereas innate cytokines are induced in Calu-3 cells) may be possibly because of the different expression levels of RIG-I protein. In fact, the levels of RIG-I protein in these cells used in our study were much higher (about ten times) than those in Calu-3 cells (Extended Data Fig. 6c). Our data suggest that in the primary lung cells and A549 cells that have sufficient levels of RIG-I protein, viral replication is blocked by the RIG-I-mediated signaling-abortive anti-SARS-CoV-2 defense mechanism (Fig. 1). On the other hand, SARS-CoV-2 can replicate with innate cytokine induction in cells with low levels of RIG-I expression, including lung cells derived from patients with COPD (Fig. 2g), as observed in Calu-3 cells31 (Extended Data Fig. 6a,b). Consistent with this, knockdown-based suppression of RIG-I protein expression resulted in a dose-dependent upregulation of IFN-β mRNA induction and viral replication upon SARS-CoV-2 infection (Extended Data Fig. 6d). Exogenous expression of the RIG-I HD but not the CTD in Calu-3 cells suppressed IFN-β mRNA induction and viral replication in a dose-dependent manner (Extended Data Fig. 6e). Therefore, it is likely that RIG-I but not MDA5 triggers a signaling-abortive anti-SARS-CoV-2 defense in primary lung cells with sufficient expression of RIG-I, and that in some cells with downregulated RIG-I expression or possibly stability, SARS-CoV-2 can propagate to induce MDA5-dependent innate cytokine responses.

The distinct usage of RIG-I and MDA5 as innate sensors against SARS-CoV-2 appears to depend on the progression of viral replication. We observed the preferential interaction of RIG-I with the positive-strand RNA in the A549 and primary lung cells (Fig. 3a). In contrast, MDA5 preferentially interacts with the negative-strand RNA of the SARS-CoV-2 RNA genome but not the positive-strand RNA (Extended Data Fig. 7a,b). Based on these findings, the possible scenario is that, in the first step, RIG-I recognizes the positive-strand RNA, which was released from viral particles, and blocks viral polymerase-mediated transcription of negative-strand RNA, and in the situation where negative-strand RNA initiates to be transcribed by using the positive-strand RNA as a template, MDA5 in turn plays a role as an innate sensor to induce type I IFNs and other cytokines. This switch seems to be regulated at least partly by the expression levels of RIG-I protein in infected cells.

The current data regarding mechanistic insights also remind us of our previous finding that RIG-I plays a dual role as a hepatitis B virus (HBV) sensor not only to activate antiviral innate signaling but also to inhibit HBV replication in human hepatocytes, wherein RIG-I functions as an antiviral factor that counteracts viral polymerase, in an IFN-independent manner1,32. A similar observation was also reported by other groups in the case of FluV infection: The binding of RIG-I to FluV nucleocapsids can directly restrict viral infection in an innate signaling-independent and IFN-independent manner33,34. These reports indicate that RIG-I recognizes viral RNAs through the CTD and doubly functions both as an innate immune activator and as a direct antiviral effector at least during HBV and FluV infection. In the case of SARS-CoV-2 infection, RIG-I-mediated interaction with viral RNA does not seem to trigger the switch in both operational modes. Our data revealed that the RIG-I HD but not the CTD selectively interacts with the 3′ UTR of the SARS-CoV-2 RNA genome, which resultantly failed to activate the ATPase activity and to interact with MAVS. These results suggest a unique viral recognition mode of RIG-I.

Furthermore, it would be intriguing to consider that this abortion of RIG-I signaling is a viral strategy to evade the innate immune system, which might be evolved by SARS-CoV-2. At this moment, we could not clearly rule out the possibility that some viral protein(s) might block the activation of the conventional IRF–NF-κB pathway downstream of RIG-I, as previously reported9,10,11. However, our data showed that the RIG-I interaction with the SARS-CoV-2 RNA genome does not require the RIG-I CTD, which is critical to trigger the activation of the conventional RIG-I pathway. Actually, an IFN-β response was not observed upon stimulation with the 3′ UTR of SARS-CoV-2 RNA genome (Extended Data Fig. 8a) and the 3′ UTR still interacted with rRIG-I (Lys888Glu) mutant, which cannot bind to conventional RIG-I ligands such as 3pRNA and poly(rI:rC; Extended Data Fig. 8b). Therefore, we posit that the unique interaction between RIG-I HD and viral RNA occurs upon SARS-CoV-2 infection, which enables an unconventional, signaling-abortive mode of viral suppression. This strategic mechanism could partly explain why many individuals with SARS-CoV-2 infection tend to show asymptomatic or mild illness, which may be associated with the failure of efficient induction of long-lasting adaptive immunity35. Regarding possible viral evasion from the RIG-I-mediated antiviral defense mechanism that we have found in this study, the SARS-CoV-2-derived papain-like protease SCoV2-PLpro has recently been reported to decrease ISGylation of cellular proteins including IRF-3 (ref. 36). Since RIG-I was reported to be ISGylated upon intracellular poly(I:C) stimulation37, we speculate that RIG-I may also be one of the target proteins by SCoV2-PLpro.

Taken together, our data have defined RIG-I expression levels as one of the intrinsic determinants for the defense in human lung cells at least during the initial process of SARS-CoV-2 infection. Actually, lung cells derived from patients with COPD, which showed reduced levels of RIG-I protein, were more susceptible to SARS-CoV-2 infection (Fig. 2g). Our data thus suggest the availability of RIG-I expression levels as one of the potential parameters for the prediction of patient outcomes. Our findings also suggest that upregulation of RIG-I expression by some agents such as ATRA (Fig. 2h) or IFNs38,39,40 may be available to enhance the preventive and/or therapeutic potential of COVID-19. Uncovering factors or conditions that modulate RIG-I expression levels are key for better understanding of SARS-CoV-2 pathogenicity and for the development of a new strategy to control SARS-CoV-2 infection.

Methods

Viruses, cells, antibodies and reagents

SARS-CoV-2 (JPN/TY/WK-521) was obtained from the National Institute of Infectious Diseases (Tokyo, Japan)41. Influenza virus (strain A/Puerto Rico/8/34) was from T. Miyazaki (Hokkaido University, Japan). EMCV was previously used42. A549 cells (CCL-185), Vero E6 cells (CRL-1586), Calu-3 cells (HTB-55), HEK293T cells (CRL-3216), Vero cells (CCL-81) and Madin-Darby canine kidney (MDCK) cells (CCL-34) were obtained from American Type Culture Collection (ATCC) and maintained in DMEM (Nissui) supplemented with 10% FBS. These cells were incubated at 37 °C with 5% CO2. Sf9 cells (B82501) were purchased from Invitrogen and routinely grown at 28 °C in a Sf-900 II SFM medium (Gibco) supplemented with 5% heat-inactivated FBS. Human Pulmonary Alveolar Epithelial Cells (HPAEpiC; 3200), human bronchial epithelial cells (HBEpiC; 3210) and human colonic epithelial cells (HCoEpiC; 2950) were purchased from ScienCell and cultured in alveolar epithelial cell medium (ScienCell), bronchial epithelial cell medium (ScienCell) and colonic epithelial cell medium (ScienCell), respectively. Human bronchial epithelial cells (HBEpC; CA50205a) and HBEpC-COPD (CA502COPDK05a) cells were obtained from Cell Applications and maintained in bronchial/tracheal epithelial cell basal medium with growth supplement (Cell Applications) except for retinoic acid. Primary bronchial/tracheal epithelial cells-COPD (PBEC-COPD; PCS-300-013) were from ATCC and grown in airway epithelial cell basal medium using a Bronchial Epithelial Cell Growth Kit (ATCC). All cells were incubated at 37 °C with 5% CO2. Antibodies were used as follows: anti-IRF-3 (p.Ser 396; 4D4G; Cell Signaling), anti-IRF-3 (D6I4C; Cell Signaling), anti-IκBα (p.Ser 32; 14D4; Cell Signaling), anti-IκBα (44D4; Cell Signaling), anti-RIG-I (D14G6; Cell Signaling, 1C3; Kerafast), anti-MAVS (ab25084; Abcam), anti-ACE2 (N1N2; GeneTex), anti-TMPRSS2 (N2C3; GeneTex), anti-β-actin (AC-15; Sigma), anti-spike (1A9; GeneTex), anti-Flag (M2; Sigma), anti-MDA5 (D74E4; Cell Signaling), anti-RdRp (4E6; Novus Biologicals) and goat anti-mouse IgG secondary antibody conjugated to Alexa Fluor 488 (A11029; Molecular Probes). Working dilutions of antibodies for immunoblotting were as follows: anti-IRF-3 (p.Ser 396; 4D4G), 1:1,000; anti-IRF-3 (D6I4C), 1:1,000; anti-IκBα (p.Ser 32; 14D4), 1:1,000; anti-IκBα (44D4), 1:1,000; anti-RIG-I (D14G6), 1:1,000; anti-MAVS (ab25084), 1 μg ml−1; anti-ACE2 (N1N2), 1:500; anti-TMPRSS2 (N2C3), 1:500; anti-β-actin (AC-15), 1:1,000; anti-Flag (M2), 1:1,000; anti-MDA5 (D74E4), 1:1,000; and anti-RdRp (4E6), 1:500. Working dilutions of antibodies for immunoprecipitation were as follows: anti-MAVS (ab25084), 1 μg mg−1 total protein; anti-RIG-I (1C3), 1:50; anti-MDA5 (D74E4), 1:100; anti-Flag (M2), 1 μg mg−1 total protein; and anti-RdRp (4E6), 1:50. Working dilutions of antibodies for fluorescence analysis were as follows: anti-spike (1A9; GeneTex), 1:100; goat anti-mouse IgG secondary antibody conjugated to Alexa Fluor 488; 2 μg ml−1. Lipofectamine RNAiMAX (Invitrogen) was used for transfection of siRNA into the cytoplasm. FuGENE HD (Promega) reagent, Lipofectamine RNAiMAX and Lipofectamine 2000 (Invitrogen) were used for gene transfer with lipid transfection. ATRA was purchased from Sigma.

Preparation of viral stocks

SARS-CoV-2 (JPN/TY/WK-521) was obtained from the National Institute of Infectious Diseases (Tokyo, Japan). To generate viral stocks of SARS-CoV-2, Vero E6-TMPRSS2 cells were inoculated with SARS-CoV-2 (JPN/TY/WK-521) with Mynox mycoplasma elimination reagent (Minerva Biolabs)41. Influenza virus (strain A/Puerto Rico/8/34) was from T. Miyazaki (Hokkaido University, Japan). MDCK cells or IFNAR1-deficient mouse embryonic fibroblasts were infected with influenza virus (strain A/Puerto Rico/8/34) or EMCV for propagation, respectively, as previously reported42. These supernatants of infected cells were collected and clarified by centrifugation. Viral titers were determined by plaque assay. These viral stocks were maintained at −80 °C. The experiments with infectious SARS-CoV-2 or influenza virus were performed in a Biosafety Level 3 or 2 laboratory, respectively, and approved by the Hokkaido University Biosafety Committee.

Generation of TMPRSS2-expressing Vero E6 cells

The cDNA of human TMPRSS2 was cloned into the self-inactivating lentiviral vector plasmids, CSII-CMV-MCS-IRES2-Bsd (RIKEN BRC). For lentiviral vector preparation, HEK293T cells were co-transfected with the lentiviral vector plasmid and Lentiviral High Titer Packaging Mix (TAKARA). Vero E6 cells were treated with the culture supernatant containing lentiviral vector, and then the cells stably expressing TMPRSS2 (Vero E6-TMPRSS2) were selected with blasticidin S.

Viral infection and measurement of viral titers

A549, Calu-3, HPAEpiC, HBEpiC, HBEpC, HCoEpiC, HBEpC-COPD or PBEC-COPD cells were infected with SARS-CoV-2 (multiplicity of infection (m.o.i.) of 1) FluV (1 m.o.i.) or EMCV (1 m.o.i.). Cells were infected for 1 h at 37 °C with SARS-CoV-2 in DMEM with 10% FBS, with FluV in serum-free MEM containing BSA and trypsin, or with EMCV in FBS-free DMEM. The supernatants of cells infected with SARS-CoV-2 were collected 5 d after infection. Vero E6-TMPRSS2 cells were used for the plaque-forming assay to determine the titers of SARS-CoV-2. Monolayers of Vero E6-TMPRSS2 cells were inoculated with serial dilutions of the cell supernatants for 1 h at 37 °C, and then, overlaid with DMEM containing 0.5% Bacto Agar (Becton Dickinson). At 3 d after infection, plaques were measured. A plaque-forming assay with MDCK cells or Vero cells was conducted to measure the titers of FluV or EMCV, respectively.

Immunoblotting

Cell lysis and immunoblot analysis were performed as described previously42,43. An Amersham Imager 680 was used for chemiluminescence detection.

RT–qPCR analysis

Total RNAs were isolated from cultured cells by using ISOGEN (Nippon Gene), and were treated with DNase I (Invitrogen). cDNAs were prepared from total RNAs by using ReverTra Ace (TOYOBO). For strand-specific detection of viral RNA, reverse transcription was performed with each forward and reverse PCR primer for negative-strand and positive-strand specific reverse transcription, respectively. Quantitative PCR was performed using SYBR Premix Ex Taq (TAKARA) and analyzed on a StepOnePlus real-time PCR system (Applied Biosystems). Detailed information about the primers used is available in Supplementary Table 1. Data were normalized to the expression levels of ACTB for each sample.

siRNA-mediated gene silencing

Chemically synthesized 21-nucleotide siRNAs, including control siRNA (MISSION siRNA Universal Negative Control), were obtained from Sigma (for sequence information, see Supplementary Table 2). In total, 1 × 105 cells of A549, Calu-3, HPAEpiC or HBEpiC were transfected with 50 nM siRNA in 2.0 μl Lipofectamine RNAiMAX. At 48 h after transfection, the cells were used for further experiments.

Generation of gene knockout A549 cells

Target sequences G/A-(N19)-NGG were selected with the online CRISPRdirect program44 (https://crispr.dbcls.jp) or from those previously used45. These sequences are shown in Extended Data Figs. 2–5 and inserted into pX330-U6-Chimeric_BB-CBh-hSpCas9 vector (Addgene) according to the manual. A549 cells were transfected with the pX330 vector with Lipofectamine 2000 (Invitrogen) or FuGENE HD (Promega) and, 72 h later, single-cell clones were isolated. The knockout phenotype of these clones was confirmed by genome sequencing (Applied Biosystems) and western blotting (Extended Data Figs. 2–5).

Plasmids and molecular cloning

The cDNAs for human RIG-I, MDA5 and the related mutants32 (C-RIG-I; amino acids 173–925: HD; amino acids 194–791: CTD; amino acids 792–925) of RIG-I were obtained by RT–PCR of total RNAs from HEK293T cells. For Flag-tagged proteins, cDNA was cloned into the XhoI and NotI sites of the pCXN2-Flag vector. The nucleotide sequence of each cDNA was confirmed with the BigDye Terminator v3.1 sequencing kit (Applied Biosystems). For the generation of expression plasmids for recombinant proteins, the cDNAs for glutathione S-transferase (GST)-tagged RIG-I or MDA5 were cloned into BamHI and HindIII sites of the pFastBac1 vector (Invitrogen).

Fluorescence analysis

A549 cells were infected with SARS-CoV-2 (0.1 m.o.i.) for 0 or 72 h. SARS-CoV-2 spike protein was visualized with anti-spike antibody and the appropriate secondary antibody conjugated to Alexa Fluor 488 (Molecular Probes). Hoechst 33342 (Invitrogen) was used for the counterstaining of nuclei. The localization was examined with an IX-81S confocal microscope (Olympus).

RNA-immunoprecipitation assay

The RIP assay was conducted using the RIP assay kit according to the manufacturer’s protocol with minor modification (MBL). A549 WT or RIG-I KO35 cells, and HPAEpiC, HBEpiC or Calu-3 cells infected with SARS-CoV-2 for 6 h or 24 h were lysed with 500 μl of lysis buffer and precleared with Protein-G Sepharose beads. A total of 20 μl of the supernatant was saved as input. Anti-RIG-I, anti-MDA5, anti-Flag, anti-RdRp or control IgG was added to cell lysates with or without recombinant RIG-I or MDA5 protein (0.75, 1.5 or 3 nM) and gently rotated overnight at 4 °C. Then, Protein-G Sepharose beads were added and further incubated with gentle shaking. One hour later, beads were washed three times with wash buffer and divided into two. One was eluted with Laemmli buffer and analyzed by immunoblotting for checking the precipitated protein, and the other was used for the elution of the precipitated RNAs and analyzed by RT–qPCR with specific primers to detect SARS-CoV-2 RNA. The amount of immunoprecipitated RNAs is represented as the percentile of the amount of input RNA (percentage input).

Viral cross-linking and solid-phase purification (VIR-CLASP)

To analyze the specific interaction of endogenous RIG-I and SARS-CoV-2 RNA genome, we used VIR-CLASP that was designed to minimize nonspecific interactions for the characterization of the interactions between incoming viral RNA and cellular proteins46,47. This assay was referred to in the protocol of a previous paper46,47. To generate viral stocks of unlabeled or 4-thiouridine (4SU)-labeled SARS-CoV-2, Vero E6-TMPRSS2 cells were inoculated with SARS-CoV-2 with or without 4SU (100 μΜ). These supernatants of infected cells were collected and clarified by centrifugation. Viral titers were determined by plaque assay. These viral stocks were maintained at −80 °C. Unlabeled 3 × 106 A549 cells were infected with SARS-CoV-2 or 4SU-SARS-CoV-2 (1 m.o.i.) for 1 h and then washed with PBS and maintained in DMEM. At 6 h later, cells were washed with PBS and irradiated with 365 nm of ultraviolet light. Cells were lysed with 900 μl of 1 × denaturation buffer (50 mM Tris-HCl (pH 6.8), 10% glycerol, 2.5% SDS and 0.67% NP-40). Then, 20 μl of the supernatant was saved for checking protein input. The CLASP step was performed according to the protocol of a previous paper32. The precipitated proteins were used for immunoblot with anti-RIG-I or anti-MDA5 antibodies.

Purification of recombinant proteins

GST-tagged RIG-I and MDA5 were expressed in Sf9 cells according to the manufacturer’s instructions for Bac-to-Bac baculovirus expression system (Gibco) and purified with Glutathione Sepharose 4B (GE Healthcare). GST protein of recombinant protein was cleaved with Precision protease (GE Healthcare).

In vitro ATPase assay

ATPase activity of the recombinant RIG-I protein was measured by phosphate-release assay with BIOMOL Green reagent (Enzo Life Sciences). The purified RIG-I protein (0.1 μg) in 20 μl of ATPase reaction buffer (20 mM Tris-HCl (pH 8.0), 1.5 mM MgCl2 and 1.5 mM dithiothreitol (DTT)) were incubated for 15 min at 37 °C with 0, 0.001, 0.1 or 10 μg of SARS-CoV-2 RNA genome or 3pRNA, then further incubated for 15 min at 37 °C with 5 nmol ATP. These samples or a series of phosphate standard (Enzo Life Sciences) were added to a 96-well plate, followed by addition of 180 μl of BIOMOL Green reagent. The malachite green was quantified by reading the absorbance at 620 nm, and then the phosphate was measured.

Preparation of nucleic acids

Purification of nucleic acids for RNA pull-down assay was conducted as previously described with minor modification32. Full RNA genome of SARS-CoV-2 was purified from supernatant of Vero E6-TMPRSS2 cells infected with SARS-CoV-2 by TRIzol (Thermo Fisher). To generate the 5′ UTR, 3′ UTR, control 1 and control 2, cDNA of the RNA genome of SARS-CoV-2 was used for PCR to prepare templates of the subsequent in vitro transcription. Detailed information about the primers used here is available in Supplementary Table 3. To avoid the non-templated nucleotide addition to the 3′ terminus of transcribed RNAs, primers for the generation of antisense strands were modified with 2′-O-methylation (2′-OMe) at the first two nucleotides of the 5′ terminus48. About 100 ng of these PCR products was used as a template for in vitro transcription under the control of the T7 promoter with MEGAscript (Invitrogen) in the presence of Ribo m7G cap analog (7.5 mM; Promega). The integrity of in vitro generated RNAs was checked by formaldehyde gel electrophoresis. These RNAs were additionally dephosphorylated with Alkaline phosphatase (Roche) to completely remove triphosphate-modified RNAs. These regions of 5′ UTR and 3′ UTR were designed in accordance with a previous report28. Control 1 and control 2 were designed not to make secondary structure by M-folds software (http://mfold.rna.albany.edu). 3pRNA was prepared as reported previously32,42,43. Poly(rI:rC) was purchased from GE healthcare.

RNA pull-down assay

The full RNA genome of SARS-CoV-2, in vitro-transcribed RNAs described above, 3pRNA and poly(rI:rC) were labeled with biotin using the Label IT Biotin Labeling Kit (Mirus). Biotin-labeled RNA (30 nM) was incubated with recombinant RIG-I protein (3 nM) in 500 μl of buffer (20 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1% NP-40, 1 mM phenylmethyl sulfonyl fluoride and 1 mM DTT and 1 μg ml−1 leupeptin (pH 7.4)) for 1 h with gentle shaking at 4 °C. Subsequently, Dynabeads M-280 Streptavidin (Invitrogen) were added and incubated for 1 h with gentle shaking. Beads were washed three times with wash buffer (20 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA and 1% NP-40 (pH 7.4)). The pull-down complexes were eluted with Laemmli buffer and analyzed by immunoblotting with the antibody for RIG-I.

RIP assay with RNA competition

Around 3 × 106 HPAEpiC, HBEpiC or A549 cells infected with SARS-CoV-2 for 6 h were lysed with 500 μl of lysis buffer of the RIP assay kit according to the manufacturer’s protocol (MBL). The lysates were incubated with or without in vitro-transcribed RNAs (10 μg) described above for 2 h at 4 °C with gentle shaking. The remaining steps were performed per the RIP assay.

ELISA

Human IFN-β protein in cell culture supernatants at 24 h after transfection of 3′ UTR, 3pRNA or poly(rI:rC) were measured by ELISA according to the manufacturer’s protocol (PBL).

Treatment with all-trans retinoic acid

ATRA was dissolved in DMSO to 10 mM. Next, 1 × 105 HBEpC-COPD or PBEC-COPD cells were seeded on 12-well plates and treated with ATRA (0.01, 0.1 or 1 μM) for 24 h, and then subjected to further experiment.

Oligonucleotide-targeted RNase H protection assay

This assay was conducted in part by referring to the protocol of the previous study49. Approximately 1 × 106 A549 cells were infected with SARS-CoV-2. At 6 h after infection, cells were lysed with 500 μl of lysis buffer and precleared with Protein-G Sepharose beads according to the manufacturer’s protocol of the RIP assay kit (MBL). Then, 20 μl of the supernatant was saved for purification of total RNA. Anti-RIG-I or control IgG was added to cell lysates and gently rotated overnight at 4 °C. Then, Protein-G Sepharose beads were added and further incubated with gentle shaking. One hour later, beads were washed three times with wash buffer and divided into two. One was eluted with Laemmli buffer and analyzed by immunoblotting for checking the precipitated RIG-I protein, and the other was incubated in 25 μl of reaction buffer (12 mM HEPES (pH 8.0), 60 mM KCl, 3 mM MgCl2, 1 mM DTT, 20 U RNasin and 40 μg ml−1 DNA oligonucleotide) with or without RNase H (Promega) at 30 °C for 60 min. For analysis of RNA cleavage efficiency, the reaction was terminated by adding 75 μl of distilled H2O, 100 μl of 2× proteinase K buffer (200 mM Tris-HCl (pH 7.5), 300 mM NaCl, 25 mM EDTA and 2% SDS) and 4 μl of proteinase K (10 mg ml−1) and then purified RNA, according to the manufacturer’s protocol of the RIP assay kit (MBL). Purified total RNAs were used as positive controls for evaluation of efficiencies of RNase H cleavage based on each DNA oligonucleotide. Positive-strand SARS-CoV-2 RNA products were quantified by RT–qPCR with specific primers to detect each region of SARS-CoV-2 RNA. The amount of RNase H-treated RNAs is represented as the percentile of the amount of RNase H-untreated RNA (percentage input). The DNA oligonucleotides used are listed in Supplementary Table 4. Oligonucleotide-targeted regions of 5′ UTR (SL2, SL3 and SL5) and 3′ UTR (P5, PK1, PK2, s2m and Oct) were selected as referred to in a previous report28,29,30. Regions A–D were designed in 11670–11681, 11758–11769, 12303–12314 or 12382–12393 nucleotide regions of the SARS-CoV-2 RNA genome, respectively.

Statistical analysis

All statistical analyses were performed by Microsoft Excel (v16.3). For all experiments, at least two independent experiments were performed. Data are presented as the mean values ± s.d. of biological triplicates. Experiments were evaluated by statistical significance with a two-tailed Student’s t-test. Differences were considered significant at P < 0.05.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.