The influenza NS1 protein modulates RIG-I activation via a strain-specific direct interaction with the second CARD of RIG-I

A critical role of influenza A virus nonstructural protein 1 (NS1) is to antagonize the host cellular antiviral response. NS1 accomplishes this role through numerous interactions with host proteins, including the cytoplasmic pathogen recognition receptor, retinoic acid–inducible gene I (RIG-I). Although the consequences of this interaction have been studied, the complete mechanism by which NS1 antagonizes RIG-I signaling remains unclear. We demonstrated previously that the NS1 RNA-binding domain (NS1RBD) interacts directly with the second caspase activation and recruitment domain (CARD) of RIG-I. We also identified that a single strain-specific polymorphism in the NS1RBD (R21Q) completely abrogates this interaction. Here we investigate the functional consequences of an R21Q mutation on NS1's ability to antagonize RIG-I signaling. We observed that an influenza virus harboring the R21Q mutation in NS1 results in significant up-regulation of RIG-I signaling. In support of this, we determined that an R21Q mutation in NS1 results in a marked deficit in NS1's ability to antagonize TRIM25-mediated ubiquitination of the RIG-I CARDs, a critical step in RIG-I activation. We also observed that WT NS1 is capable of binding directly to the tandem RIG-I CARDs, whereas the R21Q mutation in NS1 significantly inhibits this interaction. Furthermore, we determined that the R21Q mutation does not impede the interaction between NS1 and TRIM25 or NS1RBD's ability to bind RNA. The data presented here offer significant insights into NS1 antagonism of RIG-I and illustrate the importance of understanding the role of strain-specific polymorphisms in the context of this specific NS1 function.

The influenza A virus (IAV) 3 is a serious public health concern that causes annual epidemics and occasional pandemics, resulting in significant levels of morbidity and mortality each year (1,2). The ability of IAV to adapt to various hosts and undergo genetic reassortment ensures constant generation of unique strains with unknown degrees of pathogenicity, transmissibility, and pandemic potential. Currently, our knowledge of the precise combination of factors that drives the emergence of strains with pandemic potential is incomplete. However, several proteins expressed by IAV have been identified as key determinants of virulence during infection. One such protein encoded by IAV is nonstructural protein 1 (NS1).
NS1 is highly multifunctional and ranges from 215-237 amino acids in length, depending on the strain from which it is derived. During infection, NS1 plays a critical role in antagonizing the cellular innate immune response (3)(4)(5). It consists of two independently folding functional domains: the N-terminal RNA-binding domain (NS1 RBD ) and the C-terminal effector domain (NS1 ED ). Both domains facilitate the immunosuppressive action of NS1 through interactions with numerous host proteins involved in the cellular innate immune response (3)(4)(5). Specifically, NS1 is well-known for its abrogation of the type I IFN response through interactions with pathogen recognition receptors and their activation partners, such as retinoic acid inducible gene I (RIG-I) and the E3 ubiquitin ligase TRIM25 (6 -11).
RIG-I is an ATP-dependent cytoplasmic helicase whose primary function is to induce an antiviral signaling cascade in response to RNA virus infections (12,13). It contains two N-terminal caspase activation and recruitment domains (2CARDs), a central helicase domain consisting of three subdomains (Hel-1, Hel-2i, and Hel-2), a linker domain (Br), and a regulatory C-terminal domain. RIG-I is predominantly activated by short 5Ј-triphosphorylated RNAs (5Јppp dsRNAs), such as those produced by the partial complementarity in the 5Ј and 3Ј UTRs of IAV genomic RNAs (13)(14)(15)(16). Therefore, RIG-I is considered the main cytoplasmic sensor of IAV infection (17). Briefly, RIG-I undergoes a significant conformational change upon recognition of IAV RNAs that results in exposure and subsequent ubiquitination of the N-terminal 2CARDs by TRIM25 (18,19). Upon ubiquitination, RIG-I translocates to the mitochondria, where it participates in a CARD-CARD interaction with the mitochondrial antiviral signaling protein (MAVS, also known as IPS-1, VISA, and CARDIF) (20 -22). This results in recruitment of the kinases IKK and TBK1, which function to phosphorylate the antiviral transcription factor IRF3 (23)(24)(25). Upon phosphorylation, IRF3 translocates to the nucleus and promotes transcription of type I IFNs, such as IFN-␣/␤.
Ubiquitination of the RIG-I 2CARDs by the E3-ubiquitin ligase TRIM25 is a critical step in activation of the RIG-I signaling pathway (20,26,27). This step in the signaling pathway is inhibited during influenza infection by a direct interaction between NS1 and TRIM25 (6,26). Although it is has been shown that this interaction is a critical mechanism through which NS1 antagonizes RIG-I 2CARD ubiquitination (6,26), it has not been shown whether additive antagonistic effects could result from additional direct interactions between NS1 and RIG-I. Recently, we reported that the NS1 RBD from 1918 H1N1 IAV (A/Brevig Mission/1/1918) interacts directly with the second CARD of RIG-I and that a naturally occurring mutation in the NS1 RBD (R21Q) abrogated this direct interaction (8). Because of the critical role the 2CARDs play in RIG-I activation, we hypothesized that this direct interaction would enhance the ability of NS1 to inhibit ubiquitination of the RIG-I 2CARDs. Furthermore, this inhibition would be independent of other previously studied NS1 interactions involving TRIM25 (6 -11) and dsRNA structures (3) also known to inhibit activation of RIG-I.
In this study, we demonstrate that the naturally occurring R21Q mutation in NS1 (NS1 R21Q ) significantly impacts NS1's ability to control RIG-I signaling. Specifically, we used reverse genetics to generate WT A/Puerto Rico/8/1934 IAV (rPR8 WT ) and a mutant IAV encoding the R21Q mutation in NS1 (rPR8 R21Q ) to test the effects of the mutation on aspects of the viral life cycle. We observed that rPR8 R21Q induced significantly more IRF3 phosphorylation and IFN-␤ expression compared with rPR8 WT . These data indicate that rPR8 R21Q is a significantly more potent RIG-I activator compared with rPR8 WT . Further analysis revealed that NS1 R21Q is significantly less efficient at inhibiting TRIM25-dependent ubiquitination of the RIG-I 2CARDs, a critical step in activation of RIG-I signaling. Additionally, we determined that the R21Q mutation markedly diminished the ability of NS1 to interact with the RIG-I 2CARDs. Finally, we confirmed that the R21Q mutation did not have an effect on any of the other NS1 functions known to inhibit the RIG-I pathway. Taken together, this is the first study to identify a natural polymorphism in the NS1 RBD that mitigates its ability to control RIG-I 2CARD ubiquitination by TRIM25. This study demonstrates that strain-specific polymorphisms in NS1 not only have a significant impact on its ability to efficiently antagonize RIG-I activation but that the effects of these polymorphisms may be conditional to speciesspecific activation of RIG-I.

Results
Previously, we demonstrated that an R21Q mutation in the NS1 RBD abrogated its ability to interact with the second CARD domain of RIG-I. Residue 21 is distal to other residues in the NS1 RBD (Arg 35 , Arg 38 , and Lys 41 ) that are known to have func-tions in RNA binding and NS1 cellular localization ( Fig. 1) (28 -30). Disrupting these residues (Arg 35 , Arg 38 , and Lys 41 ) by mutation is well-known to cause significant defects in viral fitness, including decreased replication and inability to control the cellular antiviral response. What is not known, however, is the effect of the R21Q mutation on the viral life cycle, as mutations in this region of the NS1 RBD have yet to be studied.
The R21Q mutation has no effect on the overall structure of the NS1 RBD Given that residue 21 is in an unstudied region of the NS1 RBD , we first wanted to ensure that there were no gross structural changes in the NS1 RBD associated with the R21Q mutation. To assess any potential structural changes caused by the mutation, we used NMR to analyze the 1 H-15 N heteronuclear single-quantum coherence (HSQC) spectra of both the WT NS1 RBD (RBD WT ) and the R21Q mutation (RBD R21Q ). Both HSQC spectra showed a single set of dispersed peaks, indicative of a well-folded protein with a single, major conformation ( Fig. 2A). Because the NS1 RBD is a symmetric homodimer, only 72 amide resonances were expected and ultimately observed (not including the side-chain amides of glutamine and asparagine) for both the RBD WT and RBD R21Q . We then obtained resonance assignments for RBD R21Q using threedimensional NMR experiments such as HNCACB and CBCA-(CO)NH for further analysis. High-quality triple-resonance spectra allowed backbone resonance assignments to be obtained for 70 of the 72 (97%) possible assigned residues of the RBD R21Q . Analysis of the 13 C ␣ chemical shifts indicated that each monomer of the RBD R21Q mutant is composed of ␣ 1 -turn-␣ 2 -turn-␣ 3 (Fig. 2B) (31). This six-helix dimeric structure of the RBD R21Q homodimer is consistent with both 13 C ␣ chemical shift analysis of the RBD WT (Fig. 2B) and our previously solved solution structure of the 1918 H1N1 NS1 RBD .

Characterization of rPR8 WT and rPR8 R21Q replication and NS1 cellular localization
Because the NS1 R21Q mutation had not been characterized previously, it was critical to determine whether the mutation impacted basic aspects of the viral life cycle, such as IAV replication and NS1 cellular localization. To accomplish this, we used a previously described IAV reverse genetics system (32) to generate and rescue recombinant A/Puerto Rico/8/1934

Modulation of RIG-I activation by the influenza NS1 protein
viruses encoding WT NS1 (rPR8 WT ) and R21Q NS1 mutant (rPR8 R21Q ). Upon infecting adenocarcinomic human alveolar basal epithelial (A549) cells at m.o.i. 0.01, we observed no significant differences in viral replication between rPR8 WT and rPR8 R21Q (Fig. 3A). Similarly, we observed no significant change in the cellular localization of NS1 in A549 cells infected with rPR8 WT or rPR8 R21Q 3B). Given that R21Q is a naturally occurring mutation, it is not surprising that replication and intracellular localization are not affected; numerous viruses capable of productive infection in humans also possess a Gln at position 21. However, the observation of similar replication kinetics does not adjudicate the question of whether the two viruses will result in differential activation of the RIG-I signaling pathway upon infection, nor is it prognostic of potential differences in virulence between the two viruses, as influenza replication levels do not necessarily correlate with virulence (33,34).

Infection with rPR8 R21Q results in significantly increased IRF-3 phosphorylation and IFN-␤ expression compared with rPR8 WT infection
Activation of the RIG-I pathway results in phosphorylation of the antiviral transcription factor IRF3, which then translocates to the nucleus and induces transcription of type I inter-ferons (e.g. IFN-␤). IAV NS1 has been shown previously to inhibit IRF3-mediated signaling through its antagonism of the RIG-I pathway (9,35). Based on our previous data demonstrating that NS1 R21Q is unable to interact with the second CARD of RIG-I, we hypothesized that rPR8 R21Q would be less efficient at antagonizing RIG-I signaling. To test this hypothesis, we infected A549 cells with rPR8 WT and rPR8 R21Q (m.o.i. 2) to determine the respective levels of IRF3 phosphorylation 6, 12, and 24 h post-infection. We observed that infection with rPR8 R21Q resulted in significantly increased IRF3 phosphorylation (Fig. 4, A and B) 12 and 24 h post-infection relative to rPR8 WT levels. Furthermore, these observed differences are not due to disparities in NS1 expression levels between the two viruses, as NS1 expression was determined to be equal across all time points measured (Fig. 4A, quantification not shown). These data suggest that a Gln at position 21 in NS1 results in significantly less efficient antagonism of the RIG-I signaling pathway without altering NS1 expression.
Given that phosphorylated IRF3 acts as a transcription factor for up-regulation of type I IFNs, a corresponding increase in IFN-␤ expression upon infection with rPR8 R21Q compared with rPR8 WT was expected. We observed significantly increased IFN-␤ mRNA expression at all time points (Fig. 4C) and significantly increased IFN-␤ protein expression 12 and 24 h postinfection ( Fig. 4D) using identical experimental conditions as those described previously. Increased IRF3 phosphorylation and IFN-␤ mRNA and protein expression suggest that rPR8 R21Q is a significantly more potent activator of the RIG-I pathway during infection.

The R21Q mutation in NS1 results in increased IFN-␤ promoter activation
Although increased IRF3 phosphorylation and IFN-␤ induction during infection suggest that rPR8 R21Q is less efficient at antagonizing RIG-I signaling, it does not provide specific information regarding the mechanistic underpinnings by which NS1 antagonizes the RIG-I pathway. To overcome this limitation, we designed a series of experiments to probe how NS1 WT and NS1 R21Q affect specific steps of the RIG-I activation pathway. In addition to our hypothesis that NS1 directly inhibits ubiquiti-

Modulation of RIG-I activation by the influenza NS1 protein
nation of the RIG-I 2CARDs, NS1 has been shown previously to suppress two additional steps in the RIG-I signaling pathway. These steps include NS1 potentially shielding dsRNA structures from cellular factors responsible for detecting viral infection (3) as well as inhibition of RIG-I ubiquitination through a direct interaction with TRIM25 (6,26). Completion of these experiments would result in a more mechanistic understanding of the phenotypic differences observed between rPR8 WT and rPR8 R21Q .
We first sought to determine whether the difference in the ability of rPR8 WT and rPR8 R21Q to inhibit IRF3 phosphorylation and IFN-␤ induction during viral infection could be recapitulated in a plasmid-based system. This was accomplished by transfecting HEK293T cells with a reporter construct containing a minimal IFN-␤ promoter driving firefly luciferase expression (p125-Luc), NS1 WT or NS1 R21Q expression constructs, and a Renilla luciferase control vector. IFN-␤ promoter activity was stimulated by infection with the Cantell strain of Sendai virus, a well-known RIG-I agonist (14,36), or cotransfection with a construct encoding GST-2CARDs. It is important to note that overexpression of the RIG-I 2CARDs alone induces IFN-␤ promoter activity (27). In agreement with our virus results and previously published reports, overexpression of NS1 WT resulted in a marked decreased in IFN-␤ promoter activity in cells stimulated with Sendai virus (Fig. 5A) and RIG-I GST-2CARDs (Fig. 5B). Compared with NS1 WT , overexpression of NS1 R21Q led to significantly increased IFN-␤ promoter activity under the same conditions. This suggests that introduction of the R21Q mutation reduces the efficiency of this antagonism because all other amino acids that have been demonstrated to significantly affect this pathway (Arg 38 , Lys 41 , Glu 96 , and Glu 97 ) remain intact (6).

The R21Q mutation decreases the ability of NS1 to inhibit ubiquitination of RIG-I 2CARDs by TRIM25
Our previous study showing that the R21Q mutation abrogated the interaction between the NS1 RBD and the second CARD domain of RIG-I led to our hypothesis that NS1 R21Q would be less efficient at blocking ubiquitination of the RIG-I 2CARDs by TRIM25. To test this hypothesis, we used pulldown Cell monolayers were lysed, and total RNA was collected 6, 12, and 24 h post-infection. IFN-␤ mRNA expression was determined using qPCR, with -fold change values calculated via the ⌬⌬CT method. D, IFN-␤ levels in supernatants harvested from A549 cells infected with rPR8 WT or rPR8 R21Q (m.o.i. 2) were determined 6, 12, and 24 h post-infection using an IFN-␤-specific ELISA. Data were analyzed using Student's t test, and error bars represent the mean Ϯ S.E. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001; ****, p Ͻ 0.0001; NS, not significant. All data are representative of three independent experiments, with rPR8 WT indicated in black and rPR8 R21Q indicated in lavender.

Modulation of RIG-I activation by the influenza NS1 protein
studies in which 293T cells were transfected with plasmids encoding GST-fused human RIG-I 2CARDs (GST-2CARDs), TRIM25, and either NS1 WT or NS1 R21Q . The GST-2CARDs fusion protein was pulled down using GSH-agarose beads and visualized by Western blotting to assess the relative ubiquitination level of 2CARDs in the presence of either NS1 WT or NS1 R21Q . Use of the GST-2CARDs fusion protein as a method to determine the effect of viral proteins on RIG-I 2CARD ubiquitination has been described previously (6,20,26,37). Consistent with previous studies (6,26), there was a marked decrease in GST-2CARD ubiquitination in the presence of NS1 WT (Fig.  6A). We also observed that NS1 R21Q was less efficient at inhibiting ubiquitination of 2CARDs by TRIM25 relative to NS1 WT , confirming our hypothesis (Fig. 6B). However, the ability to inhibit 2CARD ubiquitination was not completely abrogated, as there was a significant decrease in ubiquitination of 2CARDs in the presence of NS1 R21Q relative to the GFP control. From these data, it is clear that the R21Q polymorphism significantly decreases the ability of NS1 to inhibit ubiquitination of RIG-I 2CARDs.
The data presented so far provides strong evidence that NS1 R21Q is a less efficient RIG-I antagonist because of its inability to control RIG-I 2CARD ubiquitination. However, it is important to confirm that the molecular basis of the observed differences is due to NS1 R21Q being deficient in binding to the RIG-I 2CARDs. To accomplish this, 293T cells were transfected as above, and GST-2CARD was pulled down. Probing for 3ϫFLAG-NS1 WT and NS1 R21Q in pulldown fractions revealed that, indeed, NS1 WT is capable of interacting with the RIG-I 2CARDs, but this interaction is markedly diminished for the NS1 R21Q mutant (Fig. 6, C and D). In addition, the interaction between NS1 WT and the RIG-I 2CARDs remained unchanged in the absence of TRIM25 (Fig. 6C, sixth and seventh lanes), indicating a direct interaction between NS1 WT and RIG-I 2CARDs. To ensure that the observed interaction between NS1 and 2CARDs was not due to nonspecific interactions with GST or the GA beads, we incorporated the necessary controls into the GST pulldown experiments (Fig. 6E). These data agree with our previous study showing that NS1 WT is capable of interacting directly with the second CARD of RIG-I and that the R21Q mutation disrupts the interaction. Furthermore, the inability of NS1 R21Q to bind the RIG-I 2CARDs as efficiently as NS1 WT provides a straightforward explanation for the observed differences in ubiquitination of 2CARDs. Specifically, direct interaction between NS1 WT and RIG-I 2CARDs could physically obstruct ubiquitination of the second CARD of RIG-I either by TRIM25 or its ability to bind free-floating ubiquitin chains. In contrast, NS1 R21Q binding to RIG-I 2CARDs is largely abolished relative to NS1 WT , resulting in significantly less suppression of RIG-I 2CARD ubiquitination.

The R21Q mutation has no effect on the interaction between NS1 and TRIM25
Although we demonstrated that the R21Q mutation in NS1 inhibits its ability to suppress ubiquitination of RIG-I 2CARDs, other potential mechanisms underlying the differences in IRF3 phosphorylation and IFN-␤ induction observed during rPR8 WT and rPR8 R21Q infection remain. Consequently, it was necessary to determine the effect of the mutation on other known mechanisms by which NS1 interferes with RIG-I activation to obtain more mechanistic insight. One example of how NS1 interferes with activation of the RIG-I pathway is through its direct interaction with TRIM25 (38). It was therefore neces-  (Promega). B, 293T cells were transfected with p125-Luc, pCMV-rLuc, HA-TRIM25, and GST-2CARDs along with a GFP control (white), 3ϫFLAG-NS1 WT (black), or 3ϫFLAG-NS1 R21Q (lavender) expression vectors. At 24 h, IFN-␤ promoter activity was measured using a Dual-Luciferase assay (Promega). Representative Western blots are shown to demonstrate relative transfection efficiency. All data are representative of three independent experiments. Densitometric analysis of expression levels of NS1 WT and NS1 R21Q normalized to actin expression indicates that the differences observed in both experiments are not due to differences in expression levels (data not shown). Data were analyzed using Student's t test, and error bars represent the mean Ϯ S.E. ***, p Ͻ 0.001; ****, p Ͻ 0.0001.

Modulation of RIG-I activation by the influenza NS1 protein
sary to determine whether our initial observations were independent of this interaction or whether this interaction contributed to the mitigation of NS1 function observed for the R21Q mutation. To test this, 293T cells were transfected with HA-TRIM25 along with either a GFP control vector, 3ϫFLAG-NS1 WT , or 3ϫFLAG-NS1 R21Q . TRIM25 was immunoprecipitated using ␣HA-agarose 36 h after transfection. Analysis of the immunoprecipitated fractions revealed that the R21Q mutation in NS1 has no effect on its ability to bind TRIM25 (Fig. 7). This observation offers critical insight into the molecular underpinnings of our initial observation that the R21Q mutant NS1 is less efficient at suppressing RIG-I activation. These data, combined with the data in Fig. 4, suggest that the R21Q mutation in NS1 specifically affects NS1's ability to bind the RIG-I 2CARDs independent of its TRIM25 binding function. Furthermore, these data support our hypothesis that the enhanced RIG-I activation and induction of type I interferons are due to NS1 R21Q 's deficiency in binding 2CARDs and not the interaction between NS1 and TRIM25.

The R21Q mutation has no effect on the interaction between NS1 and RNA
Another potential mechanism by which NS1 antagonizes the RIG-I pathway is through its binding of dsRNA structures. It has been posited that NS1 binds dsRNA structures to shield its presence in the cell from host proteins that have evolved to induce the antiviral signaling cascade in response to RNA virus infection (3). Assessing the R21Q mutation's effect on this interaction would allow us to interrogate another possible contributor to the differences we observed between rPR8 WT and rPR8 R21Q infection. We employed fluorescence polarization anisotropy to measure the binding affinity of a dsRNA structure to both the WT NS1 RDB and the R21Q NS1 RDB mutant. For this analysis, we chose to use a previously described dsRNA (39) to determine its binding affinity to the NS1 RBD (Fig. 8A). Analysis of the isotherms indicated that the R21Q mutation had no effect on the ability of NS1 to bind the dsRNA structure, eliminating it as a potential contributor to our observed phenotype (Fig. 8B). Specifically, the binding affinities (K d ) to the dsRNA structure for NS1 WT and NS1 R21Q were measured to be 27 Ϯ 0.95 nM and 28.6 Ϯ 1.2 nM, respectively. These data reinforce our hypothesis that the differences in IRF3 phosphorylation and IFN-␤ induction observed during rPR8 WT and rPR8 R21Q infection are due to direct inhibition of ubiquitination of 2CARDs by NS1.

Figure 6. NS1 R21Q is less effective at inhibiting RIG-I 2CARD ubiquitination and exhibits diminished interaction with RIG-I 2CARD.
A, 293T cells were transfected with expression plasmids encoding the indicated combination of TRIM25, GFP (control), GST (control), RIG-I 2CARDs fused with GST (GST-2CARDs), 3ϫFLAG-NS1 WT , and 3ϫFLAG-NS1 R21Q . 36 h post-transfection, GST-2CARDs was pulled down, and ubiquitinated GST-2CARDs (red arrows) was visualized by Western blotting. B, densitometric analysis was used to quantify the ratio of ubiquitinated GST-2CARDs to free GST-2CARDs. NS1 WT is indicated in black, and NS1 R21Q is indicated in lavender. C, 293T cells were transfected with the indicated combination of expression plasmids. 36 h post-transfection, GST-2CARDs was pulled down and free 2CARDs, ubiquitinated 2CARDs, 3ϫFLAG-NS1 WT , and 3ϫFLAG-NS1 R21Q in whole-cell lysates (WCL) as well as pulldown fractions were visualized by Western blotting. D, 293T cells were cotransfected with either 3ϫFLAG-NS1 WT or 3ϫFLAG-NS1 R21Q along with a GST expression vector or GFP control. All samples were subjected to pulldown with GSH-agarose, and pulldown fractions were probed for 3ϫFLAG and analyzed by Western blotting. Data were analyzed using Student's t test and error bars represent the mean Ϯ S.E. *, p Ͻ 0.05. Figure 7. The R21Q mutation has no effect on TRIM25 binding. 293T cells were transfected with pHA-TRIM25 along with a GFP control vector, p3ϫFLAG-NS1 WT , or p3ϫFLAG-NS1 R21Q . 36 h post-transfection, lysates were harvested and subjected to immunoprecipitation (IP) with ␣HA-agarose (Pierce, 26181). HA-TRIM25, 3ϫFLAG-NS1 WT , and 3ϫFLAG-NS1 R21Q in whole cell lysates (WCL) and immunoprecipitated fractions were visualized by Western blotting.

Modulation of RIG-I activation by the influenza NS1 protein In vivo characterization of the R21Q mutation in the mouse model
Given our observations that the R21Q mutation decreases the ability of NS1 to inhibit ubiquitination of RIG-I 2CARDs by TRIM25, we sought to characterize the effect of the R21Q mutation in vivo. Two groups of 8-to 12-week old C57BL/6 mice were inoculated intranasally with 100 pfu of rPR8 WT or rPR8 R21Q . Each group of mice was monitored for 14 days postinfection for mortality (Fig. 9A), weight loss (Fig. 9B), changes in body temperature (Fig. 9C), and signs of illness (Fig. 9D). Following Institutional Animal Care and Use Committee (IACUC) guidelines, when body weight was reduced to 75% of the starting weight, mice were deemed moribund and were euthanized. At the end of the study, 60% and 50% of the mice infected with the rPR8 WT and rPR8 R21Q viruses, respectively, were euthanized by day 11 post-infection. In addition, both viruses caused nearly identical percentages and rates of weight loss that stopped on day 8. However, mice infected with rPR8 R21Q regained this weight ϳ30% faster than mice infected with rPR8 WT , weighing an average 99% of their starting weight days 12 post-infection. Mice infected with rPR8 R21Q also experienced less severe clinical signs of illness compared with those infected with rPR8 WT . Taken together, the in vivo data suggest that the rPR8 R21Q virus may be moderately less pathogenic compared with rPR8 WT , albeit at levels not determined to be statistically significant. We note that the mechanism of RIG-I activation is fundamentally different when comparing humans and mice (26). It is this species-specific difference in RIG-I acti-vation that offers a potential explanation for the lack of significant effect in vivo despite our observation that rPR8 R21Q results in significantly increased IRF3 phosphorylation, type I interferon induction, and TRIM25-dependent ubiquitination of RIG-I 2CARDS relative to rPR8 WT . The potential speciesspecific nature of the R21Q mutation will be addressed under "Discussion."

Discussion
To date, several strain-specific polymorphisms in IAV NS1 have been described to have significant effects on NS1's function as an innate immune antagonist. However, the strain-specific polymorphisms that have been described previously are primarily involved with the ability of NS1 to bind RNA and its interaction with CPSF30 (30-kDa subunit of the cleavage and polyadenylation specificity factor) (28, 40 -42). Although both of these interactions play a large part in the overall ability of NS1 to control the host cellular antiviral response, NS1 takes part in a large number of other interactions where the effect of strain-specific polymorphisms is not well-understood. In this study, we investigated whether a previously described (8) strain-specific polymorphism at residue 21 in the IAV NS1 has a measurable effect on its ability to inhibit RIG-I signaling (Fig.  10). Using reverse genetics, we demonstrated that infection with rPR8 R21Q results in significantly increased IRF3 phosphorylation and type I interferon compared with rPR8 WT . In agreement with these results, we observed that NS1 R21Q results in significantly increased TRIM25-dependent ubiquitination of the RIG-I 2CARDs, a critical step in the activation of RIG-I. Furthermore, we observed that NS1 WT is indeed capable of interacting with the RIG-I 2CARDs in a TRIM25-independent manner and that an R21Q mutation in the NS1 RBD significantly interferes with this interaction. Finally, we determined that the R21Q mutation had no observable effect on the ability of NS1 to bind dsRNA structures or TRIM25. Collectively, these data indicate that NS1 R21Q is less efficient at blocking ubiquitination Figure 8. The R21Q mutation has no effect on RNA binding. A, graphical depiction of a previously described dsRNA structure (39) that was used to determine the effect of the R21Q mutation on the known interaction between the NS1 RBD and RNA. B, fluorescence polarization measurements of increasing concentrations of NS1 WT and NS1 R21Q in the presence of 5Ј 6-FAMlabeled shDM03 (5 nM) determined that the binding affinities of NS1 WT (black) and NS1 R21Q (lavender) to shDM03 were 27 Ϯ 0.95 and 28.6 Ϯ 1.2 nM, respectively. Change in polarization was calculated by subtracting FP measurements of shDM03 buffer-only controls from experimental FP measurements. All data are representative of three independent experiments. Error bars represent the mean Ϯ S.E. Binding affinities were determined by fitting data to one-site-specific binding with a Hill slope (GraphPad).

Modulation of RIG-I activation by the influenza NS1 protein
of RIG-I 2CARDs, leading to more potent activation of the RIG-I signaling pathway. A schematic of RIG-I activation in the context of influenza infection with specific steps at which NS1 suppresses activation of RIG-I is shown in Fig. 10.
A critical step in the activation of the RIG-I signaling pathway is ubiquitination of RIG-I 2CARDs by TRIM25. This predominantly occurs by TRIM25 covalently attaching ubiquitin chains to Lys 172 in the second CARD of RIG-I via a direct interaction between the SPRY domain of TRIM25 and the 2CARDs of RIG-I (20). Alternatively, free-floating ubiquitin chains synthesized by TRIM25 can bind directly to RIG-I 2CARDs (43). Given that the R21Q mutation in NS1 has no effect on TRIM25, the differences observed in NS1 WT and NS1 R21Q function should be independent of the ability of NS1 to abrogate TRIM25 function. Therefore, these data can be explained by two distinct mechanisms in which a strain-specific, direct interaction between NS1 and RIG-I 2CARDs enhances NS1's ability to mitigate Lys 172 ubiquitination. The first is that NS1 proteins capable of binding to the RIG-I 2CARDs obstruct access of TRIM25 to the CARDs, resulting in blockage of ubiquitination. Alternatively, the NS1-2CARD interaction interferes with the 2CARDs ability to bind free-floating ubiquitin chains. It should be noted that these possibilities are not necessarily mutually exclusive. For example, it is possible that NS1 proteins capable of binding the RIG-I 2CARDs could interfere with both of these mechanisms. In either scenario, NS1 proteins encoding a Gln at position 21 would rely solely on their ability to inhibit TRIM25, resulting in significantly less efficient suppression of RIG-I 2CARD ubiquitination. Future studies aimed at fully characterizing the structural interplay between NS1 and the RIG-I 2CARDs will reveal the mechanism by which this direct interaction interferes with activation of the RIG-I signaling pathway.
Although this study and our previous work provide information on the molecular basis for how polymorphisms at residue 21 affect NS1's interaction with the RIG-I 2CARDs, they do not address a particularly outstanding question: why does this polymorphism occur when it results in an increased antiviral response during infection? Collective analysis of all available Figure 10. Schematic of RIG-I activation in the context of influenza infection. NS1 interferes with three steps in RIG-I activation, which allows it to perform its overall function of antagonizing the innate immune response. The effect of the R21Q mutation on these three specific functions of NS1 was assessed in this work. These assessments are indicated in the schematic by their respective steps in the RIG-I activation pathway. A, NS1 is posited to antagonize RIG-I activation by shielding the presence of dsRNA structures from host proteins by directly interacting with these dsRNA structures (3). The structural basis of dsRNA recognition by NS1 of IVA is illustrated (28) (PDB: 2ZKO). As indicated in Fig. 8, the R21Q mutation has no effect on the ability of the NS1 RBD to bind dsRNA. B, NS1 was previously determined to interact directly with the second CARD of RIG-I, with the R21Q mutation abolishing this interaction (8). Additional data presented in this work supports this observation (Fig. 6C) and indicate that the R21Q mutation also reduces NS1's ability to inhibit ubiquitination of 2CARDs (Fig. 6, A and B). A hypothetical model of the interaction between the NS1 RBD and CARD2 is illustrated, using previously solved structures of 2CARDs (PDB: 4NQK) (49) and the NS1 RBD (PDB: 2N74) (8). The most critical position of 2CARD ubiquitination (Lys 172 ) is indicated in yellow. C, TRIM25 is the E3 ubiquitin ligase responsible for Lys 63 -specific polyubiquitination of 2CARDs of RIG-I at position Lys 172 . NS1 interacts directly with TRIM25 to inhibit RIG-I ubiquitination, suppressing activation of the RIG-I pathway (38). The structural basis of NS1-mediated TRIM25 (PDB: 5NT2) inhibition is illustrated. As indicated in Fig. 7, the R21Q mutation does not abolish the interaction between NS1 and TRIM25.

Modulation of RIG-I activation by the influenza NS1 protein
NS1 sequences indicated that arginine is the predominant amino acid at position 21 (Fig. 11A). However, this predominance was not maintained when sequences were analyzed as a function of host species or serotype (Fig. 11B). NS1 proteins derived from serotypes that usually result in seasonal epidemics (H1N1/H3N2) have a clear disparity in the identity of residue 21. NS1 proteins derived from H1N1 serotypes predominantly have an Arg, whereas those derived from H3N2 strains predominantly have a Gln at position 21. Additionally, residue 21 identity in IAV subtypes generally regarded as highly pathogenic (H5N1/H7N9) is heavily biased toward Arg. It is plausible that these differences are the result of species-specific adaptation, as it has already been determined that NS1's interference with RIG-I ubiquitination differs significantly between humans and mice (26). It is this difference in RIG-I activation that offers a potential explanation for our in vivo data. Specifically, it has been reported previously that NS1 does not antagonize the RIG-I signaling cascade in mice by preventing 2CARD ubiquitination during IAV infection (44,45). Instead, NS1 prevents ubiquitination of RIG-I in regions of the protein distal to the 2CARDs, which ultimately leads to RIG-I activation in mice (26). This is because activation of human RIG-I requires Lys 63linked ubiquitination of Lys 172 by TRIM25 ubiquitin E3 ligase (20), whereas murine RIG-I lacks Lys 172 in 2CARDs (26). Because the R21Q mutation specifically abrogates the NS1 RBD 's interaction with 2CARD of RIG-I (8), it is likely that the speciesspecific differences in RIG-I activation underlie the marginal differences observed in the murine infection model used in this study.
Going forward, we believe it will be important to address this question by assessing how efficiently NS1 proteins from multiple strains of IAVs isolated from multiple species are able to antagonize RIG-I 2CARD ubiquitination. Of particular interest for future studies will be to determine how differential antagonism of RIG-I 2CARD ubiquitination contributes to virulence during IAV infection. Taken together, this study demonstrates that NS1 is capable of additional sequence-specific interactions with RIG-I to facilitate its antagonism of the cellular antiviral response. It stresses the importance of how strain-specific polymorphisms in NS1 can affect its ability to antagonize the host cellular immune response in ways that are yet to be appreciated. This knowledge provides a solid foundation for further investigation into the structural and molecular mechanisms that define how these strain-specific polymorphisms in NS1 affect its ability to inhibit the RIG-I pathway. Understanding strainspecific differences in coping with the host antiviral response may better prepare us to predict the pandemic potential of IAV and aid in the development of vaccines and antivirals.

Plasmids and cloning
The pEBG-GST and pGST-2CARD plasmids were kind gifts from Dr. Michaela Gack, and their construction and purpose have been described previously (20,26,27). The HA-TRIM25 construct was generated by RT-PCR from total RNA isolated from primary normal human bronchiolar epithelial cells with the following primers: forward, 5Ј-GAT CAT GCT AGC ATG GCA GAG CTG TGC CCC CTG-3Ј; reverse, 5Ј-GAT CAT GAA TTC CTA CTT GGG GGA GCA GAT GGA-3Ј. TRIM25 was inserted into an existing vector (Addgene, 18712) using the NheI and EcoRI restriction sites, giving TRIM25 an N-terminal HA tag. The p125-Luc construct was a kind gift from Dr. Takashi Fujita, and its construction has been described previously. pCMV-rLuc was a kind gift from Dr. Mengxi Xiang (University of Alabama at Birmingham). Construction of the influenza reverse genetics plasmids for the A/Puerto Rico/8/1934 strain has been described previously (32). The R21Q mutant NS segment was generated using inverse mutagenesis PCR with the following primers: forward, 5Ј-Phos-CTT TGG CAT GTC CGC AAA CAG GTT GCA GAC CAA GAA CTA-3Ј; reverse, 5Ј-Phos-AAA GCA ATC TAC CTG AAA GCT TGA CAC AGT GTT TGG ATC-3Ј. 3ϫFLAG-NS1 WT and 3ϫFLAG-NS1 R21Q expression constructs were generated by amplifying NS1 from the IAV reverse genetics plasmids used to rescue the viruses to ensure continuity between overexpression and infection studies. NS1 WT and NS1 R21Q were cloned into the p3ϫFLAG vector (Sigma) between the EcoRI and SmaI restriction sites using the following primers: forward, 5Ј-GAT CAT GAA TTC AAT GGA TCC AAA CAC TGT GTC AAG CTT T-3Ј; reverse, 5Ј-GAT CAT CCC GGG GTC AAA CTT CTG ACC TAA TTG TTC CCG CCA T-3Ј.

NMR spectroscopy
NMR experiments were carried out at 25°C using Bruker Avance III spectrometers equipped with TCI cryoprobes operating at 600 MHz and 850 MHz 1 H frequencies. All NMR data were processed using NMRPipe (46) and analyzed using NMR-View (47) software that was compiled on Linux workstations. Backbone 1 H, 13 C, and 15 N resonances were assigned using standard triple-resonance assignment experiments: HNCACB and CBCA(CO)NH (48).

Viruses and infections
Recombinant A/Puerto Rico/8/1934 (rPR8) viruses were rescued as described previously (32). Stocks of rPR8 WT and rPR8 R21Q were generated by low-m.o.i. propagation on Madin-Darby canine kidney cells (MDCK) in 1ϫ serum-free (SF) eagle's minimum essential medium (EMEM) supplemented with 3 g/ml L-1-tosylamido-2-phenylethyl chloromethyl ketone-treated trypsin (Worthington Biochemical). Stock IAVs were titered using traditional plaque assays on MDCKs. All recombinant viruses were verified by Sanger sequencing. A549 type 2 alveolar epithelial cells were infected with rPR8 WT and/or rPR8 R21Q at the given m.o.i. for 1 h in SF EMEM. After the 1-h adsorption, the inoculum was removed and replaced with fresh SF EMEM. Sendai virus (Cantell strain, ATCC, VR-907) was propagated in 10-day-old SPF chicken embryos in accordance with established IACUC protocols. Stock Sendai virus was titered using hemagglutination assays.

Pulldown and immunoprecipitation
For GST pulldown experiments, 293T cells were transfected with a control vector, pEBG-GST or pGST-2CARDs, along with pHA-TRIM25 and either 3ϫFLAG-NS1 WT or 3ϫFLAG-NS1 R21Q . 36 h after transfection, cells were lysed in GST lysis

Modulation of RIG-I activation by the influenza NS1 protein
buffer (50 mM Tris, 200 mM NaCl, and 1% Triton X-100 (pH 7.5)), and an equal amount of protein was added to GSH-agarose (GA) beads (Pierce, 16100). Lysates were incubated with the GA beads for 2 h at room temperature in a tube rotator. After incubation, the beads were collected at 1000 ϫ g for 2 min and washed with GST wash buffer (50 mM Tris and 200 mM NaCl (pH 7.5)). Washing was repeated three times. Bound protein was eluted in 1ϫ Laemmli buffer with ␤-mercaptoethanol by heating at 95°C for 5 min. Samples were then assayed by immunoblot analysis.
For HA-specific immunoprecipitation, 293T cells were transfected with pHA-TRIM25 and either a GFP control vector, p3ϫFLAG-NS1 WT , or p3ϫFLAG-NS1 R21Q . 36 h post-infection, cells were lysed in Tris-buffered saline lysis buffer (50 mM Tris, 200 mM NaCl, and 1% Triton X-100 (pH 7.5)), and an equal amount of protein was added to anti-HA-agarose beads (Pierce, 26181). Lysates were incubated with anti-HA beads for 6 h at 4°C. After incubation, the beads were collected at 12,000 ϫ g for 3 s and washed with Tris-buffered saline wash buffer (50 mM Tris, 200 mM NaCl, and 0.05% Tween 20 (pH 7.5)). The washing steps were repeated three times. Beads were resuspended in 2ϫ Laemmli buffer without ␤-mercaptoethanol and heated at 95°C for 5 min. Samples were then assayed by immunoblot analysis.

IFN-␤ promoter activity assays
293T cells were transfected with p125-Luc and either a GFP expression vector, 3ϫFLAG-NS1 WT , or 3ϫFLAG-NS1 R21Q along with pCMV-rLuc as a control for transfection efficiency. 36 h after transfection, cells were infected with 200 hemagglutinating units of Sendai virus for 16 -18 h. After infection, luminescence was measured using a Dual-Luciferase assay (Promega) on a Cytation 5 microplate reader (Biotek).

Confocal microscopy
A549 cells were grown on coverslips and infected at 80% confluence with rPR8 WT or rPR8 R21Q at m.o.i. 5. 12 h after infection, the coverslips were fixed with 4% paraformaldehyde for 15 min at 25°C. The fixed cells were then blocked and permeabilized in PBS containing 5% goat serum and 0.5% Triton X-100 for 1 h at 25°C. Cells were then stained with an anti-NS1 primary antibody (Kerafast EMB005) at 1:200 dilution in PBS containing 5% goat serum and 0.3% Triton X-100 for 2 h at 25°C and subsequently washed five times with PBS and 0.1% Tween 20 (PBST). Cells were then stained with an Alexa Fluor 488 -conjugated goat anti-mouse IgG 2B secondary antibody (Life Technologies, A21141) for 30 min at 25°C, followed by five PBST washes. Cells were then treated with 4Ј,6-diamidino-2-phenylindole (1:1000 in PBST) for 10 min at 25°C and washed three times with PBST. Finally, coverslips were mounted onto microscope slides using Permafluor Aqueous Mounting Medium (Thermo Scientific) and sealed using nail polish. Images were acquired using an Olympus FV1000 confocal microscope and Olympus Fluoview FV10 ASW 4.2 software. All images were acquired at ϫ100 magnification with an additional ϫ1.3 digital zoom.

Western blots and ELISA
Whole-cell lysates (GST lysis buffer or M-PER, Pierce, 78501) were separated by SDS-PAGE (Bio-Rad Citerion TGX) and transferred to nitrocellulose membranes according to the manufacturer's protocols (Bio-Rad). After blocking in 5% nonfat dry milk in TBST (10 mM Tris, 150 mM NaCl, and 0.5% Tween 20 (pH 8)) for 1 h, the membrane was washed once with TBST and incubated with antibodies against GST (CST, 2624), FLAG (Sigma, X), ␤-actin (SCBT, 47778), IAV NS1 (SCBT, 130568), total IRF3 (CST, 11904), phospho-IRF3 (CST, 4947), and RIG-I (CST, 3743) in TBST at 4°C overnight. Membranes were washed three times in TBST for 15 min, and membranes were incubated with HRP-conjugated anti-mouse or anti-rabbit secondary antibodies for 1 h at room temperature in 5% nonfat dry milk. Blots were washed as before and developed with Clarity ECL (Bio-Rad). Blots were imaged on a Chemi-Doc imager. ELISA for IFN-␤ in cell culture supernatants was done following the manufacturer's instructions (PBL Assay Science). Densitometry of western blots was analyzed using Bio-Rad Image Lab.

Quantitative PCR
A549 cells were lysed with buffer RLT (Qiagen), and total RNA was extracted using the Qiagen RNeasy Plus kit with guide DNA eliminator columns. 60 ng of total RNA was used per reaction. Real-time analysis for IFN-␤ and ␤-actin was performed with the iTaq One-Step SYBR qPCR kit (Bio-Rad) following the manufacturer's instructions. Relative IFN-␤ mRNA values were normalized according to ␤-actin expression levels. The -fold change of mRNA expression was calculated using the ⌬⌬CT method (Bio-Rad CFX Manager). The specificity of each amplicon was determined by analyzing its corresponding melting curve. The sequences of primers used in qPCR reactions were as follows: IFN-␤, 5Ј-TGG GAG GCT TGA ATA CTG CCT CAA-3Ј (forward) and 5Ј-TCT CAT AGA TGG TCA ATG CGG CGT-3Ј (reverse); ␤-actin, 5Ј-ACC AAC TGG GAC GAC ATG GAG AAA-3Ј (forward) and 5Ј-TAG CAC AGC CTG GAT AGC AAC GTA-3Ј (reverse).

Fluorescence polarization
Fluorescence polarization (FP) experiments were carried out using purified WT and R21Q mutant 1918 H1N1 NS1 RBD s and 5Ј 6-FAM-labeled short hairpin RNA (shDM03) (39). Purification of the NS1 RBD s was carried out as described previously (8). 5Ј 6-FAM-labeled shDM03 was commercially synthesized (IDT). Prior to binding experiments, NS1 RBD s and 5Ј 6-FAM shDM03 were dialyzed against FP buffer (50 mM Tris-HCl, 100 mM NaCl, and 1 mM EDTA (pH 8)) to ensure buffer matching across samples. FP experiments were conducted with a constant concentration of shDM03 (5 nM) in the presence of increasing concentrations of NS1 RBD s. When mixed, samples were incubated at room temperature for 1 h, and then FP measurements were made on a Victor X5 microplate reader (PerkinElmer Life Sciences) using the FP-fluorescein protocol with excitation and emission at 485 and 535 nm, respectively, and a counting time of 1 s. Change in polarization was determined by subtracting FP measurements of shDM03 buffer-only controls from experimental samples.