Stimulation of dysregulated IFN-β responses by aberrant SARS-CoV-2 small viral RNAs

Patients with severe COVID-19 exhibit a cytokine storm characterized by greatly elevated levels of cytokines during worsening disease. Despite this, the interferon (IFN) response is delayed, contributing to disease progression. Here, we report that SARS-CoV-2 generates excessive amounts of small viral RNAs (svRNAs) encoding exact 5′ ends of positive-sense genes in human cells, whereas signicantly fewer similar svRNAs are produced by endemic human coronaviruses (OC43 and 229E). SARS-CoV-2 5′ end svRNAs are RIG-I agonists associated with IFN-beta expression in later stages of infection. The rst 60-nt ends bearing duplex structures and 5′-triphosphates are responsible for immune-stimulation. The 5′ end svRNAs were also produced during infection ex vivo and in vivo. The delta variant retains the robust 5′ end svRNA production of the parental strain, whereas omicron (BA.1 and BA.2) produces little of these erroneous svRNAs. We propose that RIG-I activation by accumulated 5′ end svRNAs overcomes the initial IFN antagonistic ability of viral proteins and contributes to drive late over-exuberant IFN production leading to the development of severe COVID-19 and suggest that evolutionary modication of SARS-CoV-2 5′ end svRNA production may correlate with the reduced disease severity likely seen with omicron (BA.1 and BA.2).

Like most viral RNAs, coronavirus RNA is detected by host RNA-sensors, cytosolic retinoic acid-inducible gene-I (RIG-I)-like receptors including RIG-I and melanoma differentiation-associated protein 5 (MDA-5) (Li et al., 2010;Sa Ribero et al., 2020;Thorne et al., 2021). Upon activation, RIG-I and MDA-5 transduce signaling cascades, leading to the activation of interferon regulatory factor 3 (IRF3) and NF-kB that are required for type 1 IFN and pro-in ammatory cytokine production, respectively. During SARS-CoV-2 infection, both RIG-I and MDA5 reportedly sense SARS-CoV-2 RNAs to activate innate immune responses (Kouwaki et al., 2021;Thorne et al., 2021), although studies suggest that it is MDA5 that predominantly governs the innate immune response to SARS-CoV-2 (Rebendenne et al., 2021;Yin et al., 2021). However, it is not well understood which viral RNA species are sensed by these molecules on infection. Additionally, longitudinal analyses revealed that SARS-CoV-2 does elicit an IFN response, but this is delayed (Lucas et al., 2020), suspected to contribute to disease progression (Huang et al., 2020;Jose and Manuel, 2020;Mehta et al., 2020). However, the viral mechanism driving the characteristic signature of delayed IFN induction in COVID-19 is poorly understood.
Abortive RNA production has been understood to be a result of RNA transcription by RNA polymerase of host machinery and RNA viruses (double-stranded RNA viruses and single-stranded RNA viruses with both polarities) (Li et al., 2008;Perez et al., 2010;Shim et al., 2002;Te Velthuis et al., 2018;Zhong et al., 2000). RNA viruses are known to exhibit high replication error rates during their reproduction and thereby produce abortive leader RNAs. It was suggested that these are recognized by RIG-I, thereby inducing an innate immune response (Oh et al., 2016;Plumet et al., 2007;Te Velthuis et al., 2018). In particular, the emergent RNA virus SARS-CoV-2 is now in a transitional period moving from bats to human hosts (Zhou et al., 2021) and supposedly faces challenges in hijacking the replication machinery in human cells (Bashor et al., 2021;Benedetti et al., 2020). Thus, SARS-CoV-2 replication in humans may produce erroneous RNAs in large amounts, which may trigger dysregulated host cell transcriptomic and immune responses, including the cytokine storm as seen with other highly pathogenic viruses such as some avian in uenza virus strains (Te Velthuis et al., 2018). Although other human coronaviruses (HCoVs) also originated from non-human CoVs, they have already established stable virus lineages in humans over long adaptation periods. These considerations motivated us to hypothesize that the distinct production of aberrant viral RNAs among different CoVs may represent one factor responsible for the different outcomes in terms of immune responses and disease severity. Interplay between aberrant viral RNAs and COVID-19 pathogenesis in this context remains to be elucidated. Here, we applied deep sequencing, focusing on the small RNAs (<200-nt) produced by SARS-CoV-2 in comparison with two endemic HCoVs (229E and OC43) that cause common colds and more mildly symptomatic respiratory disease. We report that SARS-CoV-2 infection results in the generation of excessive amounts of aberrant small RNAs that act as RIG-I ligands to induce the delayed and over-exuberant IFN-b responses observed in severe COVID-19.

Results
SARS-CoV-2 replicates e ciently in cultured cells at levels similar to HCoVs.
We initially determined the kinetics of replication of HCoV-OC43, HCoV-229E and SARS-CoV-2 parental strain (hereafter referred to as SARS-CoV-2) in different cell lines. This was necessary because there is no single cell line available that is permissive to all the three CoVs. We therefore used human cell lines (HCT9, and infected HCT8 cells with HCoV-OC43, MRC5 cells with HCoV-229E and Calu-3 cells with SARS-CoV-2 at a multiplicity of infection (MOI) of 0.001. The CoVs are generally propagated at different temperatures, namely 33ºC for HCoV-OC43 and HCoV-229E and 37ºC for SARS-CoV-2. We therefore monitored the replication kinetics at 33ºC for HCoV-OC43 and HCoV-229E and at both 33ºC and 37ºC for SARS-CoV-2 to exclude any potential bias due to different culture temperatures.
All CoVs e ciently produced viral progeny, reaching almost 100% infection rates in the host cell populations 72 or 96 hours post-infection (hpi) at either temperature ( Figure 1A) with peak titers at the corresponding times ( Figure 1B). We also tested non-human primate Vero cells in this study, which are permissive to HCoV-OC43 and SARS-CoV-2 infections, monitored in the same manner. HCoV-OC43 achieved 100% infection up to 96 hpi at 33ºC (Figure S1A), whereas SARS-CoV-2 did so up to 72 hpi at 33ºC or 48 hpi at 37ºC, with peak titers at the corresponding times ( Figure S1B). These data show that all CoVs completed a full replication cycle in these cells at either tested temperature and e ciently promulgated infection throughout the whole cell population with slightly different kinetics.
SARS-CoV-2 infection elicits a delayed IFN response in human cells.
Poly (I:C) robustly stimulated IFN-b and IL-6 secretion from HCT8, MRC5 and Calu-3 cells at 24 h posttransfection ( Figure S2A-B), which correlated with the rapid induction of RIG-I expression ( Figure S2C). These data con rm the presence of substantial intact antiviral signaling machinery in these established human cell lines.
We then infected these human cells with CoVs under the same conditions as above and measured the kinetics of IFN-b and IL-6 expression post-infection. The different CoVs evoked distinct antiviral responses in these cells ( Figure 1C-D). HCoV-OC43 induced little IFN-b and IL-6 production over the whole time course, implying the intact antagonistic ability of the host antiviral response (de Wit et al., 2016).
HCoV-229E also suppressed IFN-b production throughout the course of infection but IL-6 expression gradually increased over time. Notably, however, SARS-CoV-2 evoked both IFN-b and IL-6 expression at a later stage of infection, explained by its ability to dampen early IFN-b induction and re ecting a delayed IFN-b response. Delayed IFN-b induction was speci c to SARS-CoV-2 infection regardless of culture temperature, and correlated with the induction of RIG-I which was observed only in SARS-CoV-2-infected cells ( Figure 1E). These data prompted us to identify the IFN-b agonist(s) active in SARS-CoV-2 infection. To this end, we focused on viral RNA species produced during SARS-CoV-2 replication in infected cells.
We puri ed total RNAs from human cells mock-infected or infected with the different CoVs. The RNAs were separated into a large RNA (lRNA) fraction (>200-nt) and a small RNA (sRNA) fraction (<200-nt), which were then transfected into human 293T cells or differentiated THP-1 cells to assess their IFNb stimulatory ability ( Figure S3A). Neither RNA fraction from mock-or HCoV-OC43-infected cells stimulated IFN-b production above basal levels in the transfected cells ( Figure S3B). The sRNA fractions from HCoV-229E-infected cells also failed to increase IFN-b levels, although the lRNA fraction raised them substantially above baseline. In contrast, both RNA fractions from SARS-CoV-2-infected cells stimulated IFN-b production to a similar degree as Poly (I:C). These data show that both sRNAs and lRNAs produced by SARS-CoV-2 are IFN-b stimulatory. Depletion of cytoplasmic RNA-sensors (RIG-I and MDA-5) reduced the in ammatory response after infection, suggesting that RNA sensing is a key driver of SARS-CoV-2induced innate immune activation ( Figure 1F-G). Because the IFN-b stimulatory activity of the small viral RNA (svRNA) fraction was unique to SARS-CoV-2 and likely to be associated with the increased IFN-b response during infection, we then focused on identifying SARS-CoV-2-derived sRNAs that evoke IFNb production.
To investigate which svRNA species are involved in IFN-b production during CoV replication, we harvested sRNAs (<200-nt) and lRNAs (>200-nt) from CoV-infected cells shortly before viral titers had plateaued at which time no cytopathic effects were apparent ( Figure 1A-B). These were then subjected to RNAsequencing (RNA-seq), as established by others (Te Velthuis et al., 2018). The sRNA preparations included a step whereby all 5′ monophosphate RNAs were rst removed by incubation with XRN-1, followed by RppH treatment to enable viral RNAs bearing 5′ PPP to be ligated onto a sequence adapter ( Figure 2A). The advantage of the protocol is that it permits the e cient recovery of svRNAs bearing a 5′ PPP from RNA libraries and reduces cellular RNAs bearing other 5′ modi cations, such as a 5′ P or a cap, despite also recovering common RNA breakdown products derived from hydrolysis during RNA isolation.
We performed two comparative RNAseq analyses, one for human cells (HCT8, MRC5 and Calu-3) and the other for non-human primate cells (Vero).
The sRNA-seq approach provided a high output, similarly for all samples, in the order of 10 million total reads per sample (Table S1). Upon mapping the reads to the human and non-human primate genomes, we found similar read counts for host genes from all the infected cells. We then mapped the reads to the CoV genomes and compared the read counts between them. Total virus reads from SARS-CoV-2 were more abundant than from the HCoVs both in human and Vero cells, up to 27-fold and 4-fold more respectively, despite svRNAs comprising <3% of these libraries. The lRNA-seq also provided in the order of 10 million total reads per sample similarly for all samples (Table S2). However, the relative abundance of virus reads from the lRNA library was less prominent for SARS-CoV-2 than for HCoVs, with up to 9-fold and 2-fold more in human cells and Vero cells, respectively. This implies that SARS-CoV-2 produces svRNAs in greater amounts during RNA synthesis than the HCoVs. Next, we separated the virus reads from the sRNA library by strand speci city and discovered that most reads from all the CoV svRNAs mapped to positive-sense RNA, with >99% occupancy ( Figure 2B). These reads were mapped across the entire CoV genome ( Figure 2C).
To characterize the CoV-derived svRNAs, their fragments were categorized into three classes ( Figure 3A). First, svRNA fragments present inwards of the 5′ untranslated region (UTR) (here termed "5′ UTR svRNAs"). Second, svRNA fragments containing the leader sequence and transcriptional regulatory sequence (TRS) precisely jumping to the start sites of the major N, M and S open reading frames (ORFs) (here termed "N/M/S svRNAs"). Third, svRNA fragments harboring the leader sequences and TRS fused to other minor ORFs and all fragments that excluded the 5′ UTR (here termed "other svRNAs"). Notably, counts for the 5′ UTR svRNA fragments were much higher in SARS-CoV-2 than in the HCoVs, up to 44-fold and 9.6-fold more in human and Vero cells, respectively ( Figure 3B and Table S3). Patterns of the svRNA classes were dependent on the CoV species but independent of host cell type. Thus, 5′ UTR svRNAs represented a major fraction of the SARS-CoV-2 svRNAs, but less so in the HCoVs svRNAs, where N ORF svRNAs were abundant fractions ( Figure 3C). Nonetheless, a group formed by 5′ UTR svRNAs and N/M/S ORF svRNAs, which constitute most of the 5′ end-containing svRNA species, were always in the majority (about >54% occupancies) in the svRNAs of all these CoVs.
The coverage of svRNAs at 5′ end genomes was then further elucidated. Most strikingly, the high abundance of SARS-CoV-2 svRNAs started from the precise 5′ end (here termed "5′ end svRNAs") ( Figure  4A-E, left panels). In contrast, substantially less coverage at the 5′ end was seen with the HCoVs, despite the svRNAs mostly harboring exact 5′ termini. The peak sizes of the 5′ UTR svRNA fragments were around 60 to 80-nt (SARS-CoV-2), bimodal at 50 to 80-nt and 140 to 150-nt (HCoV-OC43) and relatively broad at 50 to 80-nt (HCoV-229E) ( Figure 4A-E, right panels). Concordantly, the svRNA fragments with the highest counts had the precise 5′ ends of the CoVs genomes with the rst 63-nt for SARS-CoV-2, the rst 72-nt for HCoV-OC43, and the rst 53-nt for HCoV-229E, with the fragments identical in human and Vero cells ( Figure 4F). Collectively, these data suggest that svRNA production is speci c to the CoV species but not to host cell type, and that svRNAs, particularly 5′ end svRNAs, are produced in extremely large amounts by SARS-CoV-2 relative to HCoVs.
Interestingly, svRNAs formed by both 5′ and 3′ ends of CoVs genomes (termed "5′ UTR-3′ UTR svRNAs") were also present in the sRNA libraries, at levels higher in SARS-CoV-2 ( Figure S4A), and with a peak size of 81-85-nt ( Figure S4B). However, these were all present at a level >1,000-fold lower than the levels of the 5′ end svRNAs. The identities of the 5′ UTR-3′ UTR svRNAs were veri ed by PCR, gel isolation and Sangersequencing using complementary primers to both termini ( Figure S4C-D). Moreover, representative 5′ UTR-3′ UTR svRNAs of different lengths induced IFN-b secretion from transfected 293 cells and the 5′ PPP end was essential for this stimulatory activity ( Figure S4E-G). However, given the paucity of their production, they are hypothesized to make a much lower contribution to IFN-b activation in SARS-CoV-2-infected cells, compared to the 5′ end svRNAs.
The rst 60-nt sequence of the SARS-CoV-2 5′ end svRNAs bearing 5′ PPP and with a duplex structure is responsible for their immune-stimulatory ability.
To determine the exact sequence of the 5′ end svRNA that is involved in IFN and cytokine production during SARS-CoV-2 replication, in vitro transcribed (IVT) RNAs corresponding to the rst 40, 60, 80 and 100-nt of the 5′ end of the SARS-CoV-2 genome ( Figure 5A) were transfected into 293T cells or into differentiated THP-1 cells, and IFN-b and IL-6 secretion was quanti ed. IVT RNAs originating from the 60, 80 and 100-nt 5′ ends induced markedly high IFN-b and IL-6 production ( Figure 5B), whereas the 40-nt RNA had only a slight effect, correlating with the RIG-I response in cells transfected with IVT RNAs of the corresponding lengths ( Figure 5C). Calf intestine alkaline phosphatase (CIP) treatment or capping with a Cap 1 analog resulted in a complete loss of the immune-stimulatory activity, indicating dependence on 5′ PPP for stimulating host antiviral signaling. We also hypothesized that the 5′ end svRNAs would form high order RNA structures ( Figure 5D), recognized by the J2 antibody that detects duplex RNAs. Indeed, we found that the J2 antibody recognized IVT RNAs of >60-nt in length ( Figure 5E-F) that formed highly structured epitopes as expected, contributing to the high IFN-b and IL-6 stimulatory activity ( Figure 5B). These data show that induction of IFNs and cytokines by SARS-CoV-2 5′ end svRNAs can be attributed to the rst 60-nt sequence that bears 5′ PPP and to the corresponding secondary structure.
We next asked whether exclusively SARS-CoV-2 5′ end sequences had IFN stimulatory ability, and not those from other coronaviruses. For this, IVT RNAs corresponding to HCoV-OC43 and HCoV-229E 5′ ends with the same lengths as tested for SARS-CoV-2 were transfected into cells, and IFN-b secretion was quanti ed. The two HCoV IVT RNAs induced IFN-b production maximally at 60-nt in length, although the peak levels were different among the three CoVs ( Figure S5A). These data suggest that the IFNstimulatory ability of SARS-CoV-2 5′ end svRNAs largely depends on the high quantity of material produced and not on the sequence of the 5′ end region. The 5′ end sequences of CoV species have the highest homology among the genomes (Madhugiri et al., 2014) and, indeed, the two HCoV 5′ ends would be expected to form secondary structures similar to the SARS-CoV-2 5′ end ( Figure S5B), supporting this notion.
SARS-CoV-2 5′ end svRNAs produced in the cytoplasm are thought to be recognized by host RNA-sensors.
To determine the role of cytosolic RIG-I and MDA-5 in svRNA sensing during SARS-CoV-2 infection, we transfected IVT RNA corresponding to SARS-CoV-2 5′ end svRNAs into 293T RIG-I-or MDA-5-knockdown cells. Silencing was shown to be effective in that speci c siRNAs reduced the levels of RIG-1 and MDA-5 mRNAs by 90% compared to mock-transfected cells ( Figure 5G). RIG-I silencing reduced IFN-b production to background levels on IVT RNA stimulation, whereas knocking down MDA-5 had little effect ( Figure 5H). Furthermore, we sequenced svRNAs co-puri ed with RIG-I from Calu-3 cells infected with SARS-CoV-2 and found that RIG-I largely recognized the 5′ end svRNAs ( Figure 5I-K). Moreover, this signature was similar to that seen in the cytoplasmic fraction of the infected cells, where the svRNAs mostly start from the exact 5′ end of the genome, on average being 53-nt in length and with positive polarity. These data document that RIG-I is the primary cytosolic sensor of 5′ end svRNAs originating from SARS-CoV-2 to trigger IFN and cytokine responses.

SARS-CoV-2 5′ end svRNAs accumulate in cells at later times after infection in vitro and ex vivo.
To determine 5′ end svRNA biogenesis in infected cells, the kinetics of SARS-CoV-2 5′ end svRNA production were studied in Calu-3 cells under the same conditions as in Figure 1. Quantitative reverse transcription-polymerase chain reaction (qRT-PCR) was performed using a stem-loop RT primer speci c for the most common svRNA having 63-nt ( Figure 6A). Speci city of the primer set was con rmed using IVT RNA templates with different 5′ end lengths ( Figure 6B). This showed that the SAR-CoV-2 63-nt-speci c primer set could detect <3% of 43-nt and 265-nt 5′ end svRNAs, with minimal detection at <10% of the 150-nt counterpart. These data indicated that the stem-loop qRT-PCR were able to distinguish different lengths of 5′ end-containing RNAs (i.e. 5′ end svRNA vs. genomic/subgenomic RNAs). The levels of svRNAs were normalized to levels of snRNA-U6, and values were expressed relative to the earliest sampling time (8 hpi). As expected, only minimal amounts of the 5′ end svRNAs were detected at 8 hpi ( Figure 6C), in agreement with the lack of signi cant IFN-b secretion at this time. The svRNAs progressively accumulated up to 96 hpi at 33ºC and 72 hpi at 37ºC, with a >5 log increase compared to levels at 8 hpi. These data suggest that 5′ end svRNAs arise at low levels early in the viral life cycle and accumulate in infected cells over time, reaching high levels at later times after infection.
To further de ne the pro le of SARS-COV-2 5′ end svRNA production ex vivo, reconstituted human nasal epithelia were infected with SARS-CoV-2 or HCoV-OC43 at an MOI of 0.1 ( Figure 6D). The kinetics of production of progeny viruses and 5′ end svRNAs and expression pro les of IFN-b and IL-6 were compared between the two viruses. SARS-CoV-2 and HCoV-2 showed similar replication kinetics in the reconstituted human epithelia, yielding >4 log FFU/ml at 96 hpi ( Figure 6E). In contrast, there was a clear difference between them in terms of 5′ end svRNA production; thus, SARS-CoV-2 accumulated to a high degree at 96 hpi, whereas HCoV-OC43 produced only marginal amounts throughout the culture period ( Figure 6F). This ndings are in accord with the marked elevation of the IFN-b level at 96 hpi ( Figure 6G) and to some extend with the gradual increase in the IL-6 level during at 48 and 96 hpi ( Figure 6H). These ex vivo data further support the interpretation that SARS-CoV-2 abortive svRNAs accumulate excessively during replication in human airway epithelia, compared with endemic HCoVs.
Stoichiometric balance of SARS-CoV-2 IFN antagonists and 5′ end svRNAs determines the host IFN response.
SARS-CoV-2 encodes numerous accessory proteins (e.g. ORF6) that act as IFN antagonists (Kimura et al., 2021;Lei et al., 2020). These are initially translated from the viral genomic RNA after virus entry to swiftly inhibit type 1 IFN production. In contrast, as shown in Figure 6C, svRNAs accumulate in infected cells at later times after infection. We thus hypothesized that IFN activation mediated by accumulated 5′ end svRNAs overcomes the antagonistic ability of the accessory proteins and drives later IFN production. To elucidate counteractions between the IFN antagonists and 5′ end svRNA agonists in host cells, we stimulated 293T cells ectopically expressing SARS-CoV-2 ORF6 with IVT 5′ end svRNAs from SARS-CoV-2. ORF6 dose-dependently suppressed IFN induction mediated by IVT 5′ end svRNAs ( Figure 6I). Thus, these data demonstrate that the stoichiometric balance between IFN antagonists and IFN-stimulatory 5′ end svRNAs determines the IFN response in host cells, consistent with our hypothesis.
The SARS-CoV-2 packaging signal generator is estimated to reside within an internal region encompassing nsp15 and nsp16 (Syed et al., 2021), indicating that 5′ end svRNAs are unlikely to be packaged into released virions as defective interfering particles. Instead, exosomes are recognized as a key vehicle to transfer various genetic materials including small RNAs including miRNAs (Valadi et al., 2007). We thus hypothesized that excessive SARS-CoV-2 5′ end svRNAs are released extracellularly via exosomes. To determine whether SARS-CoV-2-infected cells secrete exosomes carrying 5′ end svRNAs, exosomes were isolated from supernatants of SARS-CoV-2-infected Calu-3 cells at 72 hpi. To separate exosomes from free virions, isolated exosomes were immunoprecipitated using an antibody against CD63, a selective marker of exosomes. Total RNAs from puri ed exosomes were then separated into sRNA (<200-nt) and lRNA (>200-nt) fractions and sRNA fraction was assessed by RNA-seq ( Figure S6A). Immunoblotting con rmed the speci c enrichment of exosomes and exclusion of virions in these samples ( Figure S6B). We found that a substantial proportion of sRNAs from such exosomes mapped to SARS-CoV-2, especially to the positive-sense genomic material (Table S4). These svRNAs were the exact 5′ end genome and had similar signatures of 5′ end svRNAs as found in cell lysates ( Figure S6C-E). These data indicate that IFN-stimulatory 5′ end svRNAs can be released from infected cells, although their contributions to any deterioration of immune responsiveness remain unknown.
The SARS-CoV-2 delta variant produces 5′ end svRNAs in infected cells at levels similar to the parental virus.
Delta variants reportedly have a higher replication e ciency in human airway epithelia (Mlcochova et al., 2021), indicating a better tness in human cells. Thus, we investigated whether their tness in humans had resulted in a reduced production of erroneous svRNAs in infected human cells. The delta strain (termed Delta) e ciently infected Calu-3 cells and produced larger amounts of viral progeny with more rapid replication kinetics than the parental strain ( Figure S7A-B). This correlated with slightly faster IFNb expression induced by the delta strain at 48 hpi at 33ºC or 72 hpi at 37ºC ( Figure S7C-D). RIG-I expression was also induced by infection with Delta ( Figure S7E).
sRNA-seq generated a high output of reads per sample (Table S5) and showed that svRNAs from Delta were more abundant than from the HCoVs, despite the svRNAs being represented to only a minor degree in the library (<0.63%). Again, a large majority of the Delta reads mapped to positive-sense RNA with the 5′ end having the highest coverage ( Figure S7F-G). The Delta 5′ end svRNAs were the major svRNA categories mostly having the exact 5′ end ( Figure S7H) and produced at levels similar to the parental strain (Table S6). svRNA fragment patterns were identical to the parental strain with a peak size of 63-nt ( Figure S7I). qRT-PCR using the above-mentioned looped RT primers revealed that 5′ end svRNAs accumulated in infected human cells with slightly faster and higher kinetics than parental strain ( Figure  S7J). These data show that Delta has evolved in humans without correcting the production of erroneous svRNAs in human cells.
The SARS-CoV-2 Omicron variants BA.1 and BA.2 produce less 5′ end svRNAs in infected cells than the parental virus.
As of February 2022, omicron variants have emerged as the latest lineage with multiple mutations and reportedly have a greater transmissibility at the same time as reduced pathogenicity (WHO, 2022). While the original Omicron lineage BA.1 has become dominant in many countries, BA.2 has been detected in at least 67 countries and has become dominant in several of them. We investigated whether these variants produce less aberrant 5′ end svRNAs in human cells. Compared with the parental strain, the Omicron variants BA.1 and BA.2 (termed BA.1 and BA.2) replicated less well in Calu-3 cells as reported previously (Yamasoba et al., 2022), with the most drastic difference observed at 33ºC rather than 37ºC ( Figure 7A). IFN-b expression was markedly attenuated, although nally achieving a substantial level at 120 hpi at 37ºC but not 33ºC ( Figure 7B), whereas IL-6 expression was less reduced in Omicron-infected cells ( Figure 7C). These correlated with only a modest accumulation of 5′ end svRNAs from BA.1 and BA.2 ( Figure 7D). Additionally, in Vero cells, Omicron (BA.1 and BA.2) again accumulated markedly less 5′ end svRNAs and exhibited less IFN and cytokine responses compared to the parental strain and Delta, even though Omicron and Delta had similarly attenuated replication kinetics ( Figure 7E-F), similar to a prior report (Shuai et al., 2022). These data show that Omicron (BA.1 and BA.2) is less immunestimulatory, associated with lower production of aberrant svRNAs in infected cells.
To address whether 5′ end svRNAs are produced on SARS-CoV-2 infection in vivo, an established hamster infection model was used ( Figure 7G). Infection in this species does not result in severe symptoms or ARDS, but represents a reproducible COVID-19 model with moderate pneumonia and in ammatory in ltrates in the lung (Higuchi et al., 2021;Winkler et al., 2021). We infected hamsters with SARS-CoV-2 parental, Delta and BA.1 Omicron strains at the same inoculation titers and sampled lungs and serum at 5 dpi. Viral titers were measured in the lungs and 5′ end svRNA levels in the lungs and sera by stemloop RT-PCR for the 63-nt 5′ end svRNAs described above. All infected hamsters exhibited su ciently high virus titers in the lungs with >5 log TCID50/g tissue ( Figure 7H), although Delta had a slightly lower titer and Omicron had a moderately lower titer than the parental strain, consistent with prior reports for Omicron BA. 1 (Halfmann et al., 2022).
The 5′ end svRNAs were present in all infected lung at levels of >8 log copies/100 ng RNA in all hamsters infected with parental and Delta strains ( Figure 7I). In marked contrast, these were below the level of detection in lungs infected with Omicron. These data are consistent with the results from cultured human and Vero cells. Taken together, they indicate that SARS-CoV-2 parental virus and Delta clearly produce 5′ end svRNAs during viral replication in target cells in vivo, but their production is markedly diminished in Omicron-infected cells.
We also assessed whether 5′ end svRNAs were released into the sera of SARS-CoV-2-infected hamsters, but the amounts, if any, were below the level of detection in all individuals ( Figure 7J). This is probably due to the mild-to-moderate symptomatic model of SARS-CoV-2 infection in hamsters. Thus, our study has a limitation with regard to this issue and further studies using clinical serum samples from patients with severe COVID-19 are needed to validate the release of 5′ end svRNAs into sera and to carefully document its association with disease aggravation.

Discussion
Here, we found that SARS-CoV-2 generates signi cantly higher levels of aberrant svRNAs that originate from the exact 5′ ends in infected human and Vero cells than do HCoV-OC43 and HCoV-229E. The 5′ end svRNAs activated RIG-I and led to high expression of IFN-b and IL-6, both of which can contribute to the hyperin ammatory responses commonly observed in severe COVID-19 (Pedersen and Ho, 2020;Suryawanshi et al., 2021). Most studies investigating this issue have used total RNA libraries for RNA-seq to map viral RNA transcripts, where sRNAs species have rarely been captured. In contrast, RNA preparation technique we opted for here, that was previously established by others (Te Velthuis et al., 2018), included a step whereby only sRNAs (<200-nt) bearing 5′ PPP were enriched, which enabled us to capture unexpectedly large amounts of svRNAs bearing 5′ PPP by RNA-seq. Our study showed substantial utility of this method to faithfully detect 5′ termini of viral RNA genomes. Also, this is, to the best of our knowledge, the rst report showing that aberrant svRNAs, produced during SARS-CoV-2 replication, can be associated with exacerbated immune responses in severe COVID-19.
Our analyses resulted in the discovery of small RNAs derived from SARS-CoV-2. There is growing evidence that support the generation of small RNAs originating from RNA viruses including in uenza virus (Perez et al., 2010), HCV, Dengue virus, West Nile virus (Parameswaran et al., 2010) and SARS-CoV-1 (Morales et al., 2017). Most svRNAs, recovered both from SARS-CoV-2-infected cells themselves and the exosomes that they release, were the precise 5′ end of the positive-sense genome under uncapped conditions. The strand speci city indicated that they were synthesized on an antigenomic intermediate template during replication. One could envisage that the svRNAs were RNA degradation products. However, we found that small numbers of svRNAs with random sizes mapped equally across the entire genome of the CoVs investigated here, implying that these random svRNAs are viral RNA breakdown products and support the notion that the 5′ end svRNAs were produced during viral replication. The production of similar 5′ terminal viral RNAs was rst described during in uenza virus replication (Perez et al., 2010;Umbach et al., 2010) despite their shorter length (22-27-nt). The in uenza svRNAs correspond to the exact 5′ end of the RNA genome and are supposedly involved in transcription-to-replication switching, but not IFN induction. In contrast, our results suggest that SARS-CoV-2 5′ end svRNAs can drive robust immune responses. This may be explained by the distinct length spectrum of the 5′ end svRNAs (average 25-nt in in uenza viruses vs. 65-nt in SARS-CoV-2). Indeed, whereas 5′ end svRNAs of SARS-CoV-2 ≥63-nt in length effectively activated the IFN response, shorter 43-nt 5′ end svRNAs had a lesser ability to do so ( Figure 5B). Another difference is that CoVs replicate in the cytoplasm, whereas in uenza viruses replicate in the nucleus, so that cytosolic RNA sensors barely have access to intracellular in uenza svRNAs. Many details regarding the biogenesis and function of CoV 5′ end svRNAs in the virus life cycle remain to be determined. However, interestingly, SARS-CoV-2 5′ end svRNAs mostly end just before the TRS ( Figure 4A-B), which is actively implicated in the discontinued transcription that is characteristics of CoVs. Furthermore, despite being quantitatively lower, qualitatively similar signatures of 5′ end svRNAs were also recovered from cells infected with HCoV-C43 or HCoV-229E ( Figure 4C-E). Taken together, these ndings suggest that these 5′ end svRNAs may have certain functional roles in the CoV life cycle, but the high level of their production is abortive in SARS-CoV-2 replication. Unfortunately, our current data can not completely exclude the possibility that CoV 5′ end svRNAs are degraded by-products of replication. Nonetheless, regardless of how they are generated, the presence of abundant cytoplasmic SARS-CoV 5′ end svRNAs is, at least partially, associated with host immune activation in a late stage of the infection.
In replicating in uenza A virus, small aberrant viral RNAs containing both the 5′ and 3′ ends of viral RNAs were shown to activate RIG-I (Te Velthuis et al., 2018); erroneous or dysregulated replication by avian in uenza virus causes high production of these aberrant RNAs, underlying the high IFN and cytokine inductions in mammals. Similarly, SARS-CoV-2 probably has not completed human adaptation, whereas endemic HCoVs are fully human-adapted viruses. Consistent with its poorer human adaptation, highly abundant svRNAs were present in human/Vero cells or reconstituted human airway epithelia infected with SARS-CoV-2 compared to those infected with HCoV-OC43 and/or HCoV-229E. We thus propose that 5′ end svRNAs might be generally produced by dysregulated replication/transcription of newly-emerging CoVs in humans.
CoVs replicate in the cytoplasm of host cells, wherein cytosolic RNA-sensors, including RIG-I and MDA-5, are mostly responsible for monitoring viral RNAs. RIG-I preferentially recognizes short structured RNAs with 5′ PPP ends, whereas MDA-5 ligands are much less well-characterized and are presumed to be long structured RNAs, with no requirement for 5′ PPP (Chazal et al., 2018). Thus, SARS-CoV-2 5′ end svRNAs that possess 5′ PPP and duplex structures are ideal RIG-I ligands and, consistent with this, they activate RIG-I-dependent signaling pathways. In addition to the avian in uenza svRNAs (Te Velthuis et al., 2018), RIG-I was also shown to sense leader-containing short RNAs of other positive-sense RNA viruses such as avivirusess (Chazal et al., 2018), where RIG-I recognized the 5′ PPP ends of nascent transcripts before capping, inducing IFN secretion. Both RIG-I and MDA5 mediate innate immune responses during SARS-CoV-2 infection (Kouwaki et al., 2021;Thorne et al., 2021), although MDA5 is regarded as a central mediator of IFN production (Rebendenne et al., 2021;Yin et al., 2021). These results support the notion that SARS-CoV-2 5′ end svRNAs are RIG-I ligands, and likely contribute to excessive immune responses.
CoVs have developed diverse strategies to counteract IFN responses, in particular the type 1 IFN pathway (Konno et al., 2020;Li et al., 2020a;Sa Ribero et al., 2020). Numerous nonstructural proteins and accessory ORF proteins from various different CoVs were shown to prevent type 1 IFN induction and its downstream STAT1 signaling pathway in human cells (Frieman et al., 2007;King and Sprent, 2021;Xia et al., 2020). After infection by SARS-CoV-2, viral positive-sense genomic RNA facilitates rapid translation of these proteins, which in turn antagonizes viral RNA-induced RIG-I-IFN signaling. The antagonistic capability of CoV proteins was shown to be more selective for type 1 IFN signaling than NF-kΒ signaling during CoV infections (Konno et al., 2020;Lei et al., 2020;Sa Ribero et al., 2020). This may promote activation of RIG-I-NF-kΒ signaling, leading to release of other cytokines in the absence of IFN responses at early disease stages, as seen with HCoV-229E infection in the present study. Unlike the endemic HCoVs, SARS-CoV-2 infection results in the accumulation of 5′ end svRNAs and reaches high levels at later stages of disease; the threshold for RIG-I activation can then be achieved by overcoming the antagonistic ability of viral defense proteins, which in turn drives exuberant IFN production and multiple ISGs, reported to be associated with the immunopathogenesis of COVID-19 (Huang et al., 2020;Li et al., 2020b). Therefore, it is feasible that the stoichiometric balance between antagonistic viral proteins and agonistic 5′ end svRNAs would at least to some extent be associated with the disease development of COVID-19 and delayed IFN-b activation. Our data also demonstrate that Delta has not corrected the erroneous svRNA production that the parent virus exhibited, whereas the Omicron variant (BA.1 and BA.2) may have done so, implying an association with a decreased immune stimulatory signature. However, SARS-CoV-2 evolutionary modi cation of 5′ svRNA production requires further in-depth monitoring to con rm its association with COVID-19 disease progression.
The 5′ end sequences of CoVs have the highest similarity among any part of their genomes (Madhugiri et al., 2014). RNA-based antisense therapy is now widely used. Our studies thus extend the understanding of SARS-CoV-2 immunopathology and shed light on the design of drug targets against COVID-19 and future emerging CoV variants of concern. Declarations Y.A. and Y.W. performed the in vitro experiments. I.Y., N.N., and N. K. performed the RNA-seq analysis. T.O. and T. S. performed the hamster infection experiments. Y.A., I.Y., T.O., A.I., N.N., N. K., T.D., T.O., T.N., K.M., D.O., and Y.W. interpreted the results. Y.A., and Y.W. conceptualized the study and designed the experiments. Y.A., and Y.W. wrote the manuscript. All authors reviewed and proofread the manuscript.

DECLARATION OF INTERESTS
The authors declare no competing interests exist.

STAR METHODS
Detailed methods are provided in the online version of this paper and include the following:

Lead Contact
Further information and requests for resource and regent should be directed to and will be ful lled by the Lead Contact, Yohei Watanabe (nabe@koto.kpu-m.ac.jp).

Materials Availability
All requests regenerated in this study are listed in the Key Resource Tables and are available from the Lead Contact with a completed Materials and Transfer Agreement.

Data and code availability
The results presented in the study are available from the lead contact upon request.

Hamster
The hamster experiments were approved by the Institutional Committee of Laboratory Animal Experimentation of Research Institute for Microbial Diseases, Osaka University (R02-08-0). All efforts were made during the study to minimize animal suffering and to reduce the number of animals used in the experiments. 37ºC. HCoV-OC43 and HCoV-229E were propagated once in HCT8 cells and MRC5 cells respectively in DMEM-F12 containing 0.2% BSA at 33ºC. All viral stocks used in this study were prepared by limiting dilutions of the provided original stocks to eliminate potential inclusion of defective interfering particles in the stocks. Virus titration was performed by measuring focus-forming units (FFU) in focus-forming assays or the median tissue culture infective dose (TCID 50 ) method on Vero cells (SARS-CoV-2 and HCoV-OC43) and MRC5 cells (HCoV-229E) as described previously (Arai et al., 2020a;Arai et al., 2019;Watanabe et al., 2018).

Viral infection of cells in cultures
Human cells and Vero were infected with CoVs at a multiplicity of infection (MOI) of 0.001. After 1 h of incubation at 37ºC, the cells were washed twice with phosphate-buffered saline (PBS), maintained in DMEM-F-12 medium containing 0.2% BSA, and incubated at 33ºC or 37ºC.
Immunostaining and confocal microscopy Cells infected with CoVs were xed at the indicated times post-infection with 4% paraformaldehyde in PBS for 15 min followed by permeabilization with 0.2% Triton-X for 20 min at room temperature. The infected cells were stained with rabbit antibodies against SARS-CoV-2 NP (GENETEX), HCoV-OC43 NP and HCoV-229E NP (Sino Biological) for 1 h at 37°C, followed by incubation with a secondary antibody conjugated with Alexa Fluor-488 (Invitrogen) at 37°C for 1 h. To con rm the IVT RNA secondary structures, 293T cells were transfected with IVT RNAs and xed 24 h later, followed by permeabilization as described above. The cells were stained with mouse anti-dsRNA J2 antibody (SCICONS) for 1 h at 37°C, followed by incubation with Alexa Fluor-488 secondary antibody at 37°C for 1 h. Hoechst 33342 (Invitrogen) was used for the counterstaining of nuclei. Immuno uorescence images were captured using an FV3000 confocal laser scanning microscope (OLYMPUS). Foci for dsRNA were quanti ed using cellSens imaging software (OLYMPUS) from eight randomly selected image elds.

ELISA
The amounts of IFN-β and IL-6 in cell-culture supernatants were quanti ed using Quantikine kits (R&D Systems), according to the manufacturer's instructions. Optical density at 450 nm was measured with an SH-9000 lab microplate reader (Corona Electric).

Immunoblotting
Cell lysis and immunoblot analysis were performed as described previously (Arai et al., 2020b;Watanabe et al., 2018). An Amersham Imager 680 (GE Healthcare) was used for chemiluminescence detection. The band intensities were quanti ed by Amersham Imager 680 Analysis software (GE Healthcare).

Total RNA transfection
Total RNAs were extracted from Vero cells mock-infected or infected with SARS-CoV-2 using miRNeasy Mini kits (QIAGEN), fractionated into small (<200-nt) RNAs and large (>200-nt) RNAs according to the manufacturer's instructions, and transfected into 293T cells and THP-1 cells using Transit-mRNA (Mirus).
Proper fractionation of small and large RNAs was veri ed by the 2100 Bioanalyzer system (Agilent) before use. Poly (I:C) was included as a positive control. At 24 h post-transfection, cell culture supernatants were collected to measure IFN-b and IL-6 levels by ELISA as described above.
RNA fractionation from CoV-infected cells and enzyme treatment for small RNA sequencing Total RNAs from CoV-infected cells were isolated shortly before the times when titers plateaued at which times cytopathic effects were yet not apparent (72 hpi for CoVs in human/Vero cells at 33ºC except 48 hpi for SARS-CoV-2 in Vero cells and at 37ºC, 48 hpi and 24 hpi for SARS-CoV-2 in Calu-3 cells and Vero cells, respectively). Total RNAs were extracted and fractionated into small and large RNAs as described above and the small RNA fraction was then treated as described previously (Te Velthuis et al., 2018) with some modi cations. Brie y, the small RNA fraction was treated with XRN-1 (New England BioLabs) in NEB buffer 2 and incubated at 37°C for 1 h to digest host-derived miRNAs harboring 5′ P. XRN-1 was inactivated by incubating at 70°C for 10 min. 5′ PPP RNAs derived from viral RNAs were converted to monophosphorylated RNAs by RppH treatment at 37°C for 1 h. The enzyme-treated small RNAs were puri ed using RNA Clean & Concentrator™-25 (ZYMO RESEARCH). The prepared RNAs were applied to RNA-sequencing as described below.
Small RNA sequencing A miRNA library was constructed using the NEBNext Multiplex Small RNA Library Prep Set for Illumina (NEB) following the manufacturer's instructions. For this, 0.05 ng of RNA was reversed transcribed into cDNA after ligation of the multiplex 3′ SR Adaptor, hybridization of the reverse transcription primer, and ligation of the multiplex 5′ SR Adaptor. The RNA library was then ampli ed by 20 PCR cycles using Illumina compatible index primers. The ampli ed library was resolved on a 2% E-Gel EX agarose gel (Thermo Fisher). DNA fragments corresponding to approximately 150-350 bp (small RNA inserts plus 3′ and 5′ adaptors) were recovered using QIAquick Gel Extraction Kits (QIAGEN). The library was quanti ed by Qubit uorometer (Thermo Fisher) and sequenced on the Illumina NovaSeq 6000 platform using paired end reads (100 bp). The sequencing generated >1,000,000 raw reads from the sample. Adaptor sequences were removed from the raw sequencing reads using the Cutadapt program. The trimmed reads were mapped to the SARS-CoV-2 parental strain genome (LC522975) or the Delta strain (EPI_ISL_2158617) using HISAT2 version 2.1.0 (options: --pen-noncansplice 0 --no-temp-splicesite --nosoftclip --pen-canintronlen G,0,0 --pen-noncanintronlen G,0,0). The trimmed reads were also mapped to the host genome (human reference genome sequence (hg19) or Chlorocebus sabaeus strain WHO RCB 10-87 unplaced genomic scaffold, Vero_WHO_p1.0 scaffold-1, whole genome shotgun sequence (NW_023666033)) using HISAT2 version 2.1.0 (option: --no-softclip).
Large RNA-sequencing Full-length cDNA was generated using a SMART-Seq HT Kit (Takara Bio) according to the manufacturer's instructions. An Illumina library was prepared using a Nextera DNA Library Preparation Kit (Illumina, San Diego, CA) according to the SMARTer kit instructions. Sequencing was performed using an Illumina NovaSeq 6000 sequencer (Illumina) in the 100-base paired-end mode. Illumina RTA3 v3.4.4 software was used for base calling. Primer sequences were removed from the raw sequencing reads using the Cutadapt program. The trimmed reads were mapped to the SARS-CoV-2 genome of the parental strain (LC522975) or the delta strain (EPI_ISL_2158617) using HISAT2 version 2.1.0 (options: --pen-noncansplice 0 --notemp-splicesite --no-softclip --pen-canintronlen G,0,0 --pen-noncanintronlen G,0,0).
In vitro production of RNA fragments DNA fragments corresponding to the rst 40, 60, 80 and 100-nt in the 5′ UTR of SARS-CoV-2, HCoV-OC43 and HCoV-229E genomes were ampli ed by PCR using speci c forward primers containing the T7 promoter sequence (ATTGTAATACGACTCACTATAGGG) and a Hind III restriction site, and speci c reverse primers containing an XbaI restriction site. Ampli ed fragments were digested with HindIII and XbaI and cloned into enzyme-digested pUC18. Plasmid DNAs were linearized with XbaI, puri ed with QIAquick PCR Puri cation Kits (QIAGEN) and used as templates for T7 in vitro-transcription using the T7 RiboMAX™ Express Large Scale RNA Production System (PROMEGA). IVT RNAs were gel-puri ed and then further puri ed with miRNeasy mini kits. For preparation of dephosphorylated RNAs, IVT RNAs were treated with CIP for 1 h at 37°C and the treated RNAs were then puri ed with RNA Clean & Concentrator™-25 (ZYMO RESEARCH). Ribo m7G Cap Analog (PROMEGA) was used to synthesize Cap-0 RNA transcripts. Cap-0 RNAs were then converted to Cap-1 RNAs with mRNA Cap 2´-O-Methyltransferase (NEB) and puri ed with RNA Clean & Concentrator™-25. IVT RNAs were transfected into 293T cells and PMA-stimulated THP-1 cells using Transit-mRNA (Mirus). At 24 h post-transfection, cell culture supernatants were collected to measure IFN-b and IL-6 levels by ELISA as described above.
siRNA transfection siRNA for RIG-I (GACUAGUAAUGCUGGUGUAUU with dTdT overhangs) was chemically synthesized by a gene synthesis service (Fasmac). Silencer™ Select Pre-Designed siRNA was used for MDA5 silencing (s34499; Thermo Fisher). siRNAs were transfected using Transit-TKO (Mirus) and incubated for 48 h. To validate silencing, total RNAs were isolated from siRNA-transfected cells using RNeasy Mini Kits (QIAGEN) and qPCR was performed for RIG-I (Hs01061436_m1; Thermo Fisher), MDA5 (Hs00223420_m1; Thermo Fisher) and 18s ribosomal RNA (Hs99999901_S1; Thermo Fisher). Data were normalized to the expression levels of 18S ribosomal RNA for each sample and the DDCt method was used for the relative value quanti cation.

Stem-loop qRT-PCR
Total RNA was extracted from virus-infected cells at the indicated time points, or from lungs and sera of hamsters 5 dpi using miRNeasy mini kits. RNA was reverse transcribed using TaqMan MicroRNA Reverse Transcription Kits (Applied Biosystems) and qPCR was performed using TaqMan Fast Advanced Master Mix (Applied Biosystems) according to the manufacturer's instructions. Stem-loop RT primer, primers and probes for qPCR were designed by Custom TaqMan Small RNA Assays (Applied Biosystems) based on the representative sequences of 5′ end svRNA from SARS-CoV-2 (ATTAAAGGTTTATACCTTCCCAGGTAACAAACCAACCAACTTTCGATCTCTTGTAGATCTGTT) and HCoV-OC43 (ATTGTGAGCGATTTGCGTGCGTGCATCCCGTTTCACTGATCTCTTGTTAGATCTTTTTGTAATCTAAACTTT).
The speci city of the SARS-CoV-2 63-nt speci c stem-loop qRT-PCR was tested using corresponding 5′ end svRNA templates with different 5′ end lengths (43, 63, 150 and 265-nt) that were transcribed in vitro, followed by gel-puri cation as described above. For Calu-3 cell samples, TaqMan MicroRNA Assays for snRNA-U6 (001973) were used as controls for normalization of small RNAs from cells and the relative expression was calculated by the DDCt method using the 8 hpi samples as a reference with a set value of one relative unit. For hamster samples, 100 ng total RNA were used for qRT-PCR.

Virus infection of ex vivo reconstituted human nasal epithelia
The MucilAir system is a reconstituted human nasal epithelium, consisting of ciliated, goblet and basal cells. Cultures were maintained under air/liquid interface (ALI) conditions in transwells with 700 µL MucilAir medium in the basal compartment. Prior to viral infection, the apical surface was washed twice with 200 µl MucilAir medium (20 min at 37°C) to remove mucus. Cells were infected with SARS-CoV-2 or HCoV-C43 on the apical side at an MOI of 0.1 in 150 µl MucilAir medium for 1.5 h at 37°C. Viral inoculum was removed and cells were washed twice with MucilAir medium (20 min at 37°C) before continuing culture for 24, 48 and 96 h. Apical supernatants were harvested by adding 200 µl MucilAir medium on the apical side and incubating for 20 min at 37°C prior to collection, after which viral titers were determined by FFU assays. Intracellular RNA of each well was harvested using miRNeasy mini kits and 5′ end svRNA levels were quanti ed by stem-loop qRT-PCR as described above. For SARS-CoV-2, a stem-loop primer speci c for the most common 63-nt 5′ end svRNA was used, whereas for HCov-OC43, a stem-loop primer speci c for the most common 72-nt 5′ end svRNA was used. Basal medium was harvested to measure IFN-b and IL-6 production levels as described above.

RNA immunoprecipitation (RIP) assay
The RIP assay was conducted using the kit for microRNA (MBL) according to the manufacturer's protocol. Calu-3 cells infected with SARS-CoV-2 at 72 hpi were lysed with lysis buffer and pre-cleared with Protein G Dynabeads (Thermo Fisher). The supernatants were incubated with Protein G immobilized with anti-RIG-I antibody (#3743; Cell Signaling) with gentle rotation for 3 h at 4°C. The beads were then washed ×3 with wash buffer. For RNA-seq, the sRNA fraction was extracted from the total RNA bound to the beads using RNeasy Mini Kits (QIAGEN).
At 24 h post-transfection, cells were stimulated with 2.5 × 10 11 copies of IVT-RNA corresponding to the rst 60-nt of the 5′ end from the SARS-CoV-2 genome using Transit-mRNA (Mirus). At 48 h post transfection, the cells and culture supernatants were harvested for Western blotting and ELISA to determine ORF6 expression and IFN-β production.

Isolation of exosomes
Culture supernatants from SARS-CoV-2-infected cells were collected and centrifuged at 3000 × g for 15 min to remove cell debris and then ltered through a 0.45 µm lter. Then, 90 ml of ltered supernatant was concentrated to a nal volume of 1 ml using Amicon Ultra-4 Centrifugal Filter Units with Ultracel-100 membranes (Millipore). The concentrated culture supernatants were mixed with ExoQuick-TC (System Bioscience). Exosomes in the concentrates were immunoprecipitated using anti-CD63 antibodyconjugated Dynabeads Protein G (Thermo Fisher) according to the manufacturer's instructions. Total RNAs were extracted from exosomal samples using miRNeasy mini kits and investigated by next generation sequencing as described above. The concentrated exosomal samples and immunoprecipitated exosomal samples were resolved by SDS-PAGE and transferred onto a polyvinylidene di uoride membrane (Millipore). The S protein of SARS-CoV-2 and exosome marker CD63 were detected with speci c primary antibodies and HRP-conjugated secondary antibodies. The Amersham ECL Select Western Blotting Reagent was used for band visualization. The band intensities were quanti ed by Amersham Imager 680 Analysis Software (GE Healthcare).

Hamster infection model
Four-week-old female Syrian hamsters were purchased from SLC Japan. Under mixed anesthesia (medetomidine-butorphanol-midazolam), the animals were inoculated intranasally with 1.0×10 6 plaqueforming units SARS-CoV-2 (in 60 ml) as described previously (Higuchi et al., 2021). On day 5 postinfection, all animals were euthanized and lungs and sera were collected for viral titration and/or stemloop qRT-PCR.

QUANTIFICATION AND STATISTICAL ANALYSIS
Data analyses were performed using GraphPad Prism Version 6 software (GraphPad Software).
Statistically signi cant differences between virus pairs were determined by ANOVA with Tukey's multiple comparison test. Data are presented as the means ± SD.