The cellular localization of avian influenza virus PB1-F2 protein alters the magnitude of IFN2 promoter and NF k B-dependent promoter antagonism in chicken cells

The accessory protein, PB1-F2, of influenza A virus (IAV) functions in a chicken host to prolong infectious virus shedding and thus the transmission window. Here we show that this delay in virus clearance by PB1-F2 in chickens is accompanied by reduced transcript levels of type 1 interferon (IFN)-induced genes and NF k B-activated pro-inflammation cytokines. In vitro , two avian influenza isolate-derived PB1-F2 proteins, H9N2 UDL01 and H5N1 5092, exhibited the same antagonism of the IFN and pro-inflammation induction pathways seen in vivo , but to different extents. The two PB1-F2 proteins had different cellular localization in chicken cells, with H5N1 5092 being predominantly mitochondrial-associated and H9N2 UDL being cytoplasmic but not mitochondrial-localized. We hypothesized that PB1-F2 localization might influence the functionality of the protein during infection and that the protein sequence could alter cellular localization. We demonstrated that the sequence of the C-terminus of PB1-F2 determined cytoplasmic localization in chicken cells and this was linked with protein instability. Mitochondrial localization of PB1-F2 resulted in reduced antagonism of an NF k B-dependent promoter. In parallel, mitochondrial localization of PB1-F2 increased the potency of chicken IFN 2 induction antagonism. We suggest that mitochondrial localization of PB1-F2 restricts interaction with cytoplasmic-located IKK b , reducing NF k B-responsive promoter antagonism, but enhances antagonism of the IFN2 promoter through interaction with the mitochondrial adaptor MAVS. Our study highlights the differential mechanisms by which IAV PB1-F2 protein can dampen the avian host innate signalling response.


INTRODUCTION
Influenza A viruses (IAVs) are respiratory pathogens of multiple species, including poultry and humans. The high prevalence levels of IAV infection in poultry in the Middle East and South-East Asia, where it is considered to be endemic in some countries, are a significant economic burden to the poultry industry [1,2]. It is estimated that the H7N9 subtype of avian IAV caused upwards of $6.5 billion in losses to the Chinese economy in 2013 alone [3]. Food security, animal welfare and of course the potential spillover into the human population and the resulting possibility of pandemics, are all important reasons for further study of IAV interactions within avian hosts, with the aim of enabling the design and implementation of better control strategies for IAV in poultry.
IAVs have a core repertoire of 10 viral proteins that all subtypes express, allowing successful replication of the virus in infected cells. There is also an ever increasing number of 'accessory' proteins that are not required for replication in a cell, but which contribute and modulate IAV infection in vivo [4]. In 2001, PB1-F2 was the first accessory protein to be identified in IAV [5]. PB1-F2 is a small protein that ranges in size from 87 to 101 amino acids (aa) and is expressed from genome segment 2 of IAV in a +1 reading frame relative to that of the PB1 protein, the viral RNAdependent RNA polymerase [5]. The presence of PB1-F2 in the IAV genome is highly conserved (93 %) as a full-length protein in IAV isolates recovered from avian species, which contrasts with a full-length conservation of only 43 % in human-or 48 % in swine-isolated IAV sequences [6,7].
Many functions have been attributed to PB1-F2 since its identification; it has variously been shown to induce apoptosis, antagonize innate immune responses, promote secondary bacterial infection and modulate viral polymerase activity, but many of these functions have been shown to be cell type-and IAV strain-dependent [8,9]. It has also been shown to be rapidly turned over in cells due to degradation by the proteosome [10].
Previously we and others have shown that the absence of a full-length PB1-F2 protein in an avian IAV has the effect of increasing pathogenicity in chickens, in both highly pathogenic and low-pathogenic IAV infection models [6,11]. It has been documented in mammalian models that IAV infections with large innate responses, characterized by extraordinary levels of inflammatory cytokines and chemokines, correlate with enhanced pathogenicity [12,13]. Therefore the absence of PB1-F2 resulting in enhanced innate immune responses is a likely mechanism for this increase in pathogenicity in chickens. It has previously been demonstrated in mammalian cells infected with IAV that PB1-F2 can antagonize anti-viral innate responses, for example Yoshizumi et al. and others report that human interferon beta (IFNb) promoter activation is suppressed by the presence of PB1-F2 [14][15][16][17]. Additionally, in vivo, in mice, type 1 interferon (IFN)-induced transcripts in the lung were reduced when PB1-F2 was present [18]. In chickens, in the context of a highly pathogenic H5N1 IAV, it was found that similar to mice, infection with a virus that contained a full-length PB1-F2 protein resulted in reduced levels of a range of anti-viral innate gene transcripts [11]. A direct interaction between PB1-F2 of the vaccine strain A/ PR/8/34 (PR8) and human mitochondrial antiviral signalling protein (MAVS) has been demonstrated by co-immunoprecipitation and it is thought that PB1-F2 exerts its antagonist function of mammalian IFNa/b through modulation of this critical adapter protein in the IFN induction pathway [19,20]. Reis et al. showed that PB1-F2 from a range of avian IAVs and PR8 was able to interact with human IKKb, resulting in the inhibition of human NFkB activation [18]. The antagonistic functionality of PB1-F2 for the interferon-induced and pro-inflammation pathways in avian cell lines and hosts is less clear, however, with only one study providing data where the effect of PB1-F2 on transcript levels of interferon-stimulated genes in the lungs of chickens infected with a highly pathogenic H5N1 strain was analysed [11]. Leymarie et al. showed that transcripts of the interferon signalling pathway were downregulated when they compared datasets from infections by a PB1-F2-containing virus and the isogenic knock-out of PB1-F2. Differences exist between the signalling induction pathways for innate response cytokines and chemokines in chickens and mammals; for example, chickens lack the pattern recognition receptor RIG-I, which is important for initiating the type 1 IFN induction pathway in mammals [21]. Sequence comparison between human MAVS and chicken MAVS shows only a 39 % conservation at the amino acid level, whereas there is a stronger degree of conservation between human IKKb and that in the chicken genome (83 % conservation at amino acid level). There has been no analysis of the mechanism of immune response suppression in the chicken host by PB1-F2. It is likely that host immune response suppression results in the observation that our group previously reported, that the presence of a PB1-F2 protein in a low pathogenic H9N2 IAV can prolong the virus shedding duration in chickens, resulting in an increased transmission window and environmental contamination [6].
In mammalian cells, PB1-F2 proteins display one of two sub-cellular localizations; predominantly mitochondrialassociated, as is the case for the PB1-F2 protein from the laboratory-adapted H1N1 A/PR8/8/34 (PR8), or diffuse in the cytoplasm, with a small proportion of mitochondrial localization, as is the case for H5N1 A/HK/156/1997 [5,[22][23][24][25]. However, no study of cellular localization in avian cell types has been undertaken. It has been noted in several reports that the C-terminus sequence of the PB1-F2 protein determines cellular localization in mammalian cells, although the residues that confer mitochondrial targeting appear to vary between PB1-F2 proteins [22]. The ability of PB1-F2 to antagonize mammalian IFNa/ b promoters has been linked to the C-terminus of PB1-F2 [19,20].
The PB1-F2 protein of IAV is variable in sequence, which has been shown to impact on cellular localization and function in mammalian cells. There is a paucity of published information about PB1-F2 in the context of IAV infection of avian cells and IAV isolates that infect avian species are more likely to have a full-length PB1-F2 protein encoded in their genome. Therefore there is a particular gap in our knowledge of the specific in vitro functions of PB1-F2 in avian cells and how these might be correlated with the effects of PB1-F2 on IAV pathogenicity and viral shedding in chickens.
Here we analysed the cellular localization and ability of avian IAV PB1-F2 proteins to antagonize type 1 IFN (chicken IFN-1 and IFN-2) and pro-inflammation induction pathways in chicken cells. We demonstrated that the C-terminal sequence of PB1-F2 can alter the cellular localization and that mitochondrial association of PB1-F2 affects the antagonistic function and stability of PB1-F2 in chicken cells. We confirmed interactions with chicken orthologues of MAVs and IKKb, suggesting that PB1-F2 acts to suppress innate responses in chickens and this is the mechanism of suppressed pathogenicity and prolonged viral shedding of IAV strains that carry full-length PB1-F2 proteins. . Lines were drawn transecting objects in the 'red' channel (displayed as white lines) and the pixel intensities along these transects for both the 'red' and 'green' channels were displayed graphically. Pearson's correlation coefficients for of transects were calculated and the R 2 values were displayed. Displayed are representatives of these lines, which were drawn randomly at three locations per image.

PB1-F2 proteins from different strains of avian influenza virus have distinct cellular localizations in chicken cells
We analysed the association with mitochondria of PB1-F2, from a wide panel of different avian influenza virus subtypes (Table 1), in a chicken fibroblast cell line (Fig. 1). DF1 cells were transfected with plasmids encoding PB1-F2 proteins with an N-terminal V5 tag and immune-stained for the V5 tag along with Mitotracker dye to highlight the mitochondria before imaging by confocal microscopy. Co-localization between mitochondria and PB1-F2 was quantified by drawing transects through the mitochondria and determining the correlation of pixel intensity for PB1-F2 and mitochondrial staining using Pearson's coefficient. Both visually and by quantification, the PB1-F2 proteins could be categorized into two groups; those with strong mitochondrial co-localization, with Pearson's coefficients of >0.7 (H5N1 50-92, H3N8a and H10N7), and those that showed a diffuse cytoplasmic localization with a weak mitochondria association, with Pearson's coefficients of <0.7 (H9N2 UDL01, H8N4, H11N9, H3N8b and H6N2).
Localization of PB1-F2 in the cytoplasm is determined by amino acids in the C-terminus of the protein The PB1-F2 proteins of the viruses H9N2 UDL01 and H5N1 50-92 are examples that fell into the two different localization patterns. H9N2 UDL01, R 2 =0.243 [±0.032 standard error of the mean SEM)], is predominantly cytoplasmic, whereas H5N1 50-92, R 2 =0.749 (±0.024 SEM), is predominantly mitochondria-associated (Figs 1 and 2b). A comparison of the primary amino acid sequences of H9N2 UDL01 and H5N1 50-92 PB1-F2 proteins revealed that there were 11 amino acid differences between them, 4 in the N-terminal region (aa  and seven in the C-terminus (aa 41-90) (Fig. 2a). Previous reports have demonstrated that the association of PB1-F2 with mitochondria in mammalian cells is determined by the PB1-F2 C-terminus [14,22,23]. Therefore we generated chimeric PB1-F2 proteins, whereby a swap of the C-terminus (aa 41-90) was made. We generated UDL01 : 5092 PB1-F2, which contained the N-terminus of H9N2 UDL01 and the Cterminus of H5N1 50-92, and the reciprocal swap 5092:UDL01 PB1-F2 (Fig. 2a). Transects drawn through mitochondria on confocal microscopy images confirmed that V5-PB1-F2 localization to the cytoplasm was determined by the C-terminus of the UDL01 PB1-F2 protein (Fig. 2b). However, the H5N1 5092 C-terminus (UDL01 : 5092) did not enable localization at the mitochondria to the same extent as the H5N1 5092 WT PB1-F2. The UDL01 : 5092 PB1-F2 protein was, however, more efficiently targeted to the mitochondria, with a mean R 2 correlation value of 0.648 (±0.026 SEM), than 5092:UDL01, which was predominantly cytoplasmic, with the mean R 2 correlation value with mitochondria being 0.273 (±0.045 SEM) (Fig. 2b). Previous publications have demonstrated that putative mitochondrial targeting sequences lie between amino acid residues 46 and 75. There were four residues that were different between our two prototypic strains in this aa 46-75 region, so we generated expression constructs that swapped all four mutations in both the H9N2 UDL01 and H5N1 5092 PB1-F2 proteins to analyse in greater detail those residues that contribute to PB1-F2 localization away from the mitochondria. Site-directed mutagenesis at aa 60, 62, 66 and 68 (from R, H, N and I respectively) in H9N2 UDL01 to Q, L, S and T, as is the sequence in H5N1 5092, generated the UDL01-4M PB1-F2 and the reverse mutations in H5N1 5092 generated the 5092-4M PB1-F2. Co-localization of UDL01-4M and 5092-4M PB1-F2 proteins that were V5-tagged with mitochondria via co-staining with Mitotraker Red in DF-1 cells demonstrated that the 5092-4M PB1-F2 was relocalized in a diffuse pattern to the cytoplasm (R 2 =0.399) (Fig. 2c), similarly to UDL WT [R 2 =0.024 (±0.032 SEM)] and 5092:UDL01 [R 2 =0.273 (±0.045 SEM)]. Conversely, UDL01-4M PB1-F2 had no effect on mitochondrial localization (R 2 =0.265) (Fig. 2c), maintaining a similar diffuse cytoplasmic localization to H9N2 UDL, in contrast to the UDL01 : 5092 mutant [R 2 =0.647 (±0.045 SEM)]. This suggests that mitochondrial targeting is more easily disrupted than the retargeting of PB1-F2 to the mitochondria, which may require features in the full-length protein.
Sequence analysis of the region between aa 46-75 of the other avian influenza virus PB1-F2 proteins from Fig. 1 indicated that the four amino acids identified as being important for the cytoplasmic localization of the H5N1 5092-4M PB1-F2 are not common in cytoplasmic-targeted PB1-F2s. Indeed, the PB1-F2 protein from H8N4 A/Mallard/Alberta/192/92 had a cytoplasmic localization and shared identical residues at these four positions with H5N1 5092, which was mitochondrial-targeted. None of the other PB1-F2 proteins

PB1-F2 suppresses interferon-stimulated gene transcripts upon avian influenza virus infection in chicken cells
To provide robust data on the functionality of PB1-F2 to antagonize the type I interferon signalling pathway in chickens we used several different methods. Firstly, we analysed and compared the transcript levels of ISGs in nasal tissue taken from chickens at 2 and 5 days post-infection with isogenic H9N2 viruses that contain the coding capacity for either a full-length PB1-F2 protein (H9N2 UDL01) or a truncation to the initial 11 aa of PB1-F2 (H9N2 UDL01-KO), deemed to be non-functional and therefore a knock-out (Fig. 3a). These two viruses had previously been reported on in James et al. and replicated in chicken tissues to similar levels on day 2 post-infection but with the PB1-F2 knock-out H9N2 infectious virus being cleared more quickly on day 5 than the H9N2 with a full-length PB1-F2 protein (Fig. S1) [6]. We observed significantly increased transcript levels for several ISGs and cytokines in the nasal tissue from birds infected by the H9N2 PB1-F2 knock-out virus on both day 2 (Mx, OAS, STAT-1 and IL-6) and day 5 (Mx, IL-6 and IL-1b) post-infection (Fig. 3a).
We utilized primary chicken embryonic fibroblast (CEF) cells and chicken bone marrow-derived monocyte (BMDM) cells to see whether the same upregulation of ISG and cytokine transcripts was observed in vitro by avian influenza viruses lacking a functional PB1-F2 protein compared to those possessing a full-length PB1-F2 protein (Fig. 3b, c). In both cell types, the H5N1 5092 virus pair (H5N1 5092 and H5N1 5092-KO) showed a significant increase in Mx and OAS transcript levels when PB1-F2 was absent during infection and the trend was evident for the H9N2 UDL01 virus pair, although it was not statistically significant. In the BMDM, the transcript levels for IL-6 and IL-8 were significantly elevated when PB1-F2 was absent in the H9N2 UDL01-KO virus and again this trend was evident for the H5N1 5092 virus pair, but it was not statistically significant. The H9N2 UDL01 and H5N1 5092 virus pairs were equally able to infect both CEFs and BMDMs at a multiplicity of infection (m.o.i.) of 3 (Fig. 3d, e). The growth curves of the virus pairs in primary chicken kidney cells (Fig. S4a) showed that for both pairs of viruses there was no difference in infectious virus release at early time points post-infection, up to 24 h. We did observe significantly more virus at 48 h post-infection in the growth curve for the H5N1 5092 PB1-F2 KO virus compared to H5N1 5092 WT. In the chicken fibroblast DF-1 cell line, the reconstituted viral polymerase activity of both the H5N1 5092 and H9N2 UDL01 viruses was not affected by the ability of the PB1 segment to encode a full-length PB1-F2 (Fig. S4b). Our in vitro infections suggested that the PB1-F2 of both H9N2 UDL01 and H5N1 5092 was able to suppress ISG and pro-inflammation cytokines transcripts, but the potency of the antagonism varied.

Co-localization with the mitochondria and consequently MAVS enhances chicken IFN-2 promoter antagonism by PB1-F2
We performed a co-localization study in chicken DF-1 cells by overexpression of a V5-tagged chicken MAVS construct (ckMAVS-V5) and a FLAG-tagged version of the H9N2 UDL01 and H5N1 5092 PB1-F2 proteins (Figs 4a and S2).
Similarly to the co-localization with Mitotraker, we performed immune staining against the FLAG and V5 tags and drew transect lines through the confocal images along the ckMAVS-V5 staining and the correlation of pixel intensity along these transects for each stain was then calculated using Pearson's coefficient. We observed that mitochondrial-targeted H5N1 5092 PB1-F2 protein had a strong colocalization with ckMAVS (R 2 =0.851±0.022 SEM), whereas the H9N2 UDL01 PB1-F2 protein only had a weak correlation (R 2 =0.332±0.042 SEM); a GFP-FLAG expression construct was used as a control and showed no co-localization with ckMAVS-V5 (R 2 =À0.002±0.020 SEM) (Fig. 4a). Coimmunoprecipitation using an anti-V5 antibody from DF-1 cells expressing ckMAVS-V5 and either FLAG-tagged H5N1 5092 or H9N2 UDL01 PB1-F2 demonstrated that both PB1-F2 proteins could interact with ckMAVS, but that the efficiency of the binding differed, with the H5N1 5092 PB1-F2 associating to a greater extent with ckMAVS than H9N2 UDL01 (Fig. 4b).
Having already detailed the cellular localization difference of the PB1-F2 proteins, H9N2 UDL01 and H5N1 5092, we hypothesized that cellular localization could influence the potency with which type 1 IFN was induced and consequently ISGs during infection. Therefore we used a welldescribed IFN bioassay that makes use of type 1 IFN inhibition of a GFP vesicular stomatitis virus (VSV-GFP) to determine the ability of the PB1-F2 proteins to antagonize the production of bioactive chicken IFN from DF-1 chicken cells [26]. We applied equal volumes of virus-inactivated supernatant taken from chicken BMDM cells infected with H9N2 UDL01 and H5N1 5092 isogenic virus pairs that contained a full-length PB1-F2 or PB1-F2 KO, at a high m.o.i. to DF-1 cells, as a primer, prior to infection of the DF-1s by VSV-GFP. The level of GFP fluorescence measured in the cells was therefore inversely proportional to the relative levels of IFNs secreted during the initial influenza infection. We found that in both virus pairs the parental viruses that contained a full-length PB1-F2 protein induced less type 1 IFN following infection than the viruses that were knocked out for PB1-F2 (Fig. 4c). This suggested that the level of IFN induction was suppressed more by the PB1-F2 from H5N1 5092, which is located predominantly at the mitochondria, co-localized with ckMAVS, than the PB1-F2 from H9N2 UDL01, which has a more modest co-localization with MAVS and a predominantly cytoplasmic cellular localization.
To increase the clarity of the relationship between subcellular localization of PB1-F2 and the antagonism of IFN and the subsequently induced ISGs, we utilized a chicken IFN-2 promoter firefly luciferase reporter construct in DF-1 cells.
The reporter was co-transfected with poly I : C, which can induce the IFN-2 promoter, a constitutively expressed Renilla reporter and PB1-F2 plasmids. We compared the ability of H5N1 5092 PB1-F2 with that of H9N2 UDL01 PB1-F2 and found that, in line with our other results, H5N1 5092 PB1-F2 was more potent in the antagonism of the chicken IFN-2 promoter (Fig. 4d). In conjunction, we utilized the chimeric PB1-F2 proteins containing the reciprocal C-terminus swaps, 5092:UDL01 and UDL01 : 5092, which show altered cellular localizations with respect to the mitochondria, compared to the parental PB1-F2 proteins. Our results demonstrated that possession of the H9N2 UDL01 C-terminus, which disrupted mitochondrial localization, resulted in a reduced antagonism of the chicken IFN-2 promoter compared to those PB1-F2 proteins containing the H5N1 5092 C-terminus (Fig. 4d).
Co-localization with the mitochondria did not correlate with enhanced PB1-F2 antagonism of an NFkB responsive promoter Influenza virus infection is known to activate NFkB transcription of cytokines involved in inflammation, such as IL-8 and IL-1b. In chickens infected by a H9N2 virus that possessed a full-length PB1-F2 protein, some of the transcripts for pro-inflammatory genes, e.g. IL-6 and IL-1b, were supressed compared to infection where the PB1-F2 was absent (Fig. 3a). To determine whether PB1-F2 from the avian influenza strains H9N2 UDL01 and H5N1 5092 specifically antagonized NFkB activation, we utilized a NFkB response element luciferase reporter construct in chicken DF-1 cells, and this was co-transfected with a constitutively active Renilla luciferase plasmid and PB1-F2 plasmids. Cells were stimulated by transfection with poly (poly I : C), which activated NFkB, leading to luciferase induction. Co-expression of H9N2 UDL01 PB1-F2 significantly inhibited the luciferase signal, whereas co-expression of H5N1 5092 PB1-F2 did not, although there was a trend in multiple assays for the luciferase signal to be lower than for no PB1-F2 expression (empty) (Fig. 5a). H9N2 UDL01 PB1-F2 inhibition of the NFkB response element was due to the inhibition of NFkB nuclear translocation (Fig. S3).
We utilized our C-terminus-swapped PB1-F2 constructs to determine the contribution of the C-terminus sequence, and by inference cellular localization, to the ability of PB1-F2 to antagonize NFkB activity. We found that both 5092:UDL01 and UDL01 : 5092 PB1-F2 proteins were able to significantly inhibit NFkB-responsive promoter activity in this set-up (Fig. 5a). UDL01 : 5092 was only partially mitochondrial- localized compared to H5N1 5092, which may explain this result.
Cytosolic localization of PB1-F2 provides increased opportunity to directly interact with IKKb but reduces PB1-F2 cellular stability We tested the ability of H5N1 5092 and H9N2 UDL01 PB1-F2 to interact with chicken IKKb using co-immunoprecipitation in DF-1 cells (Fig. 5b). Both V5-tagged H9N2 UDL01 and H5N1 5092 PB1-F2 proteins were shown to co-precipitate with a FLAG-tagged chicken IKKb. The pulldown was more efficient with H9N2 UDL01 PB1-F2 than with H5N1 5092 PB1-F2, despite less H9N2 UDL01 PB1-F2 protein being present in the input samples (Fig. 5b). We also saw that in DF-1 cells chicken IKKb localized more closely with the PB1-F2 of H9N2 UDL01 than that of H5N1 5092 (Fig. 5c, d). A general feature of our work with the H9N2 PB1-F2 protein was the instability of the protein in cells; we required the use of MG132, a welldescribed proteasome inhibitor in order to obtain good visualization in cells and by Western blot. Despite the PB1-F2 proteins being cloned into the same plasmid background, with identical promoters and Kozak sequences, the H5N1 5092 PB1-F2 protein does not have the same instability seen with H9N2 UDL01 PB1-F2 (Fig. 6a). We hypothesized that the difference in cellular location of the two PB1-F2 proteins may result in different exposure to proteasomal-mediated degradation, with interaction with MAVS at the mitochondria somehow shielding the H5N1 5092 PB1-F2 from degradation. To provide a quantified measure of protein stability in chicken DF-1 cells, we performed a cycloheximide chase assay and levels of PB1-F2 expression were determined post-application of cycloheximide by Western blot. The H9N2 UDL01 PB1-F2 degrades significantly from the levels detected at the point of cycloheximide application (time=0) within 30 min, whereas by 8 h post-cycloheximide application, the H5N1 5092 PB1-F2 is still visible (Fig. 6a-c).
In Fig. 2 we demonstrated that swapping the H5N1 5092 C-terminus (5092:UDL01), or the four amino acids termed '4M', for those in the H9N2 UDL01 (5092-4M) resulted in a disruption of the mitochondrial localization of H5N1 5092, although the reciprocal mutations swaps did not completely relocalize H9N2 UDL01 PB1-F2 from the diffuse cytoplasmic staining pattern to the equivalent level of H5N1 5092. We used the panel of mutant PB1-F2 proteins in the cycloheximide chase assay and looked at the time point 1 h after the addition of cycloheximide to transfected DF-1 cells. We found that disrupting the mitochondrial localization (5092:UDL01 and 5092-4M) resulted in a significant difference in the stability of the PB1-F2 proteins (Fig. 6b, c).

DISCUSSION
The capacity of IAV to encode a full-length PB1-F2 is conserved in 93 % of all avian host-isolated IAVs, compared to 43 and 48 % from human and swine host isolates. We have previously demonstrated that the presence of a full-length PB1-F2 in an avian IAV resulted in a longer virus shedding profile in chickens and increased viral tissue dissemination in infected birds, and that these outcomes had a cumulative effect of prolonging the transmission window of the virus from an infected chicken to a naïve sentinel chicken [6]. The absence of a full-length PB1-F2 also resulted in an increase in viral pathogenicity in chickens in both our previous study and that by Leymaire et al. (2014) [6,11]. PB1-F2 has been described previously to have cell-type-specific and viral strain-specific activities. Here we show that PB1-F2 proteins from two avian IAVs of the subtypes H5N1 and H9N2 inhibit the production of type I IFN in chicken cells and can interact with chicken MAVS upstream of the type 1 interferon promoter (Figs 3 and 4). In addition, these two PB1-F2 proteins can antagonize NFkB activation and interact with chicken IKKb (Fig. 5). PB1-F2-induced suppression of type 1 IFN and NFkB must occur in vivo in chickens, since lower levels of ISG and pro-inflammation transcripts, which are stimulated by onward signalling from type 1 IFN induction and NFkB activation, were observed in nasal tissue from birds that were infected by H9N2 IAV containing an intact PB1-F2 protein. Specifically, the abundance of transcripts for the ISGs MX, OAS and STAT-1 and the pro-inflammatory-associated cytokines IL-6 and IL-1b were all significantly reduced when infection was by a virus possessing a full-length PB1-F2 protein as compared to infection by an isogenic virus that did not encode a functional PB1-F2. This suppression of ISGs and pro-inflammation signals was also observed in a PB1-F2-dependent manner in vitro in primary chicken fibroblasts and macrophages. Some differences in the magnitudes of the transcript levels were observed between the CEFs and chicken BMDM upon infection by isogenic viruses that differed only by the presence of PB1-F2, but this was likely due to different sensitivities of fibroblast and myeloid cells to these induction signals, since the entry by the viruses at high m.o.i. appeared to be equal. There was little difference in the magnitude of apoptosis caused by the individual viruses in the two different cell lines as measured using a capsase3/7 activity reporter (data not shown). Leymarie et al. also report for another H5N1 influenza isolate that the presence of PB1-F2 suppresses innate responsive transcripts during chicken infection [11].
There are clear differences in the components of the type 1 interferon pathway between mammals and chickens, such as the lack of the pattern recognition receptor (PRR) RIG-I and transcription factor IRF-3 in chickens. Instead, the influenza pathogen-associated molecular pattern is recognized by the 'RIG-I like receptor' MDA5 in chicken cells, which then signals via chicken MAVS located at the mitochondria to the transcription factor IRF-7, which induces the type 1 interferon promoter in the nucleus [27]. In this study we have demonstrated for the first time an interaction between chicken MAVS, the mitochondrial-located interferon induction pathway adapter and PB1-F2 of two avian IAV strains. The interaction of PB1-F2 with human MAVs has been demonstrated, previously, but there is only 39 % homology at the amino acid level between chicken and human MAVS [19]. Our coimmunoprecipitation results suggested that the interaction with chicken MAVS was more efficient with the PB1-F2 from H5N1 5092 as compared to H9N2 UDL01, although both did display the interaction. In human 293T cells, Varga et al. showed that PB1-F2 of the vaccine strain A/PR/8/34 interacted with MAVS at the mitochondria and this resulted in a decreased mitochondrial membrane potential and antagonism of type 1 interferon production [19,20]. It is likely that a similar mechanism is the cause of the type 1 interferon induction antagonism seen in chicken cells by PB1-F2, since we can demonstrate that interaction does occur with chicken MAVS and that greater antagonism is seen by PB1-F2 proteins that co-localize at the mitochondria with MAVS, e.g. the H5N1 5092 PB1-F2 protein compared to the H9N2 UDL01 PB1-F2 (Figs 1,  2 and S3). Additionally, we have demonstrated interaction between PB1-F2 and chicken IKKb for the first time. When NFkB is released from inhibition it translocates into the nucleus and acts as a transcription factor stimulating the transcription of innate response genes, and the phosphorylation action of the serine kinase, IKKb, on Ikb releases NFkB inhibition. It has been shown previously that PB1-F2 is able to interact directly with human IKKb [28]. There is an 83 % amino acid conservation between human and the chicken IKKb homologue. We found that the H9N2 UDL01 PB1-F2 had a more efficient interaction with IKKb than H5N1 5092. We hypothesize that interaction with chicken MAVS results in the inhibition of onward interferon induction pathway simulation and is likely how PB1-F2 elicits an effect on ISG levels, and similarly, the interaction of PB1-F2 with IKKb may prevent the phosphorylation of Ikb and thus inhibit the translocation of NFkB to the nucleus, resulting in the inhibition of pro-inflammation cytokines. The variability in the strength of the interaction of the avian influenza virus PB1-F2 proteins with either chicken MAVS or chicken IKKb corresponded to the potency of the two PB1-F2 proteins to antagonize type I IFN induction and NFkB activation. More H5N1 5092 PB1-F2 was precipitated with chicken MAVS and it also had a greater ability to antagonize the type I IFN induction pathway as compared to H9N2 UDL01 PB1-F2. In contrast, H9N2 UDL01 PB1-F2 had a more substantial interaction with IKKb and was able to antagonize the activation of an NFkB dependent promoter significantly in comparison to the H5N1 5092 PB1-F2 protein. The differences in antagonism potency correlated with the degree of PB1-F2 localization to the mitochondria. Significant colocalization with mitochondrial markers, including MAVS, was associated with enhanced antagonism of type I IFN production, but reduced NFkB activation inhibition. Taken together, this work suggests a mechanism by which PB1-F2 can interact with chicken MAVS and chicken IKK-b to downregulate type 1 IFN and the downstream pro-inflammatory signals in infected cells in vivo, which results in reduced pathogenicity and impaired viral clearance in chickens.
Cellular localization was established for a panel of avian influenza PB1-F2 proteins, which clearly established two phenotypes: predominantly mitochondrial or predominantly cytoplasmic (Fig. 1). The two PB1-F2 proteins investigated here, H5N1 5092 and H9N2 UDL01, were representative of each group, with H5N1 5092 being preferentially localized to the mitochondria, the location of chicken MAVS (Fig. S2), and H9N2 UDL01 having a diffuse cytoplasmic localization that was the same as chicken IKKb (Fig. 5 c). It has previously been demonstrated that the Cterminus is responsible for the mitochondrial localization of PB1-F2 in mammalian cells [23] and indeed when we swapped the C-terminus portions of the H5N1 5092 for that of H9N2 UDL01 PB1-F2 proteins we observed a relocalization away from the mitochondria to the cytoplasm in chicken fibroblast cells (Fig. 2). However, a reciprocal swap only partially retargeted the PB1-F2 to the mitochondria, with statistical analysis by Student' s t-test between the Pearson's localization value for H5N1 5092 and UDL01 : 5092 indicating that they were still significantly different (P value=0.008). This suggests that features in the N-terminus of the H5N1 5092 PB1-F2 make an additional contribution to mitochondrial localization. By altering only four amino acids in the C-terminus of the H5N1 5092 to those found in the H9N2 UDL01 PB1-F2 protein (Q60R, L62H, S66N and T68I), producing our 5092-4M mutant, we disrupted the H5N1 5092 mitochondrial association (Fig. 2 c). A triple mutant H5N1 5092 PB1-F2 protein (L62H, S66N, T68I) where residues were changed to those found in the H9N2 UDL01 PB1-F2, as in the 5092-4M mutant, did not result in significant retargeting of the H5N1 5092 PB1-F2 (data not shown). Cheng et al. showed that the introduction of the motif I68, L69, V70 and F71 to H5N1 A/Hong Kong/ 156/97 relocated this PB1-F2 to the mitochondria in mammalian cells, while replacement of this motif in the PB1-F2 of H1N1 A/Puerto Rico/8/34 removed mitochondria association [22]. We found no conservation of this motif in the PB1-F2 proteins we analysed for cellular localization. Similarly, Chen et al. demonstrated an enhancement of the mitochondrial localization of H5N1 A/Hong Kong/156/97 PB1-F2 when two additional leucine residues (Q69L and H75L) were introduced in the C-terminus [23]. Our analysis shows no correlation between the numbers of leucine residues in the C-terminus of PB1-F2 and whether predominant mitochondrial localization occurs in chicken cells (Fig. 2d). Gibbs et al. postulated the presence of an amphipathic helix and a short preceding hydrophobic patch in the C-terminus of PR8 PB1-F2 (aa 65-87) and hypothesized that this was responsible for the targeting of the protein to the mitochondria [24]. We again found no correspondence between this area of sequence; all the sequences we analysed, regardless of observed cellular localization, had a predicted helix structure and four out of the five basic residues observed in the PR8 amphipathic helix. The upstream hydrophobic residue patch was the same between viruses that had different cellular localization.
PB1-F2 has a short half-life within cells and becomes ubiquitinated on conserved lysine residues in the C-terminus as a post-translational modification [10]. Kosik et al. demonstrated that mutation of the C-terminal lysine residues at positions 73, 78 and 85 to arginines reduced ubiquitination and increased the stability of PB1-F2 in mammalian cells. We found that H9N2 UDL01 had one fewer lysine residue in the C-terminus sequence compared to H5N1 5092 (Fig. 2d), but when the stability of both the H9N2 UDL01 and the H5N1 5092 PB1-F2 proteins and the derived C-terminal-altered mutants in chicken cells was assessed, we found that cellular localization, not the presence of lysines in the C-terminal sequence, correlated with protein stability. H5N1 5092 had a strong co-localization with the mitochondria and was the PB1-F2 protein that was most stable in chicken DF-1 cells. In contrast, disruption of the H5N1 5092 mitochondrial localization via C-terminal (5092: UDL01) or four-residue swap (5092-4M) with the H9N2 UDL01 PB1-F2 resulted in significant degradation of the PB1-F2 protein by 1 h after the addition of cycloheximide. We suggest, therefore, that PB1-F2 proteins that are predominantly targeted to the mitochondria are protected from ubiquitin-directed proteosomal degradation compared to PB1-F2 proteins that have a diffuse cytoplasmic localization. Indeed, studies by others also support this hypothesis; Cheng et al. demonstrate that the predominantly mitochondria-located PB1-F2 of A/Puerto Rico/8/34 has a significantly longer half-life in HeLa cells compared to the PB1-F2 from A/Hong Kong/156/97, which is cytoplasmic in localization, and this finding has also been observed by Kosik et al. for the same strains in 293T cells [10,22]. Chen et al.
showed that a H1N1 A/Taiwan/3355/1997 PB1-F2 that localized to the mitochondria had the same stability in 293T cells as the A/Puerto Rico/8/34 PB1-F2 [23]. PB1-F2 has been shown to have a dramatically differing secondary structure that is dependent upon the sequence and the environment in which the proteins reside [29,30]. In aqueous solution, PB1-F2 isdisordered and in the presence of membranes PB!-F2 can form a structured fibre which can disrupt those membranes [8]. The differing predominant localizations of the H5N1 5092 and H9N2 UDL01 PB1-F2 viruses could influence the protein structure in infected cells and consequently their functionality.
Our results suggest that the PB1-F2 protein can operate at different points within innate immune signalling pathways in a strain-specific manner that is reminiscent of the influenza NS-1 protein, and that localization of PB1-F2 in the cell is a driving factor of the protein's functionality. PB1-F2 is encoded from the +1 reading frame of the functionally important RNA-dependent RNA polymerase, PB1. Analyses indicate that the first two positions in the codon of PB1-F2 vary more than position three, which is in line with +1 translation frame from PB1, and this contributes to the high non-synonymous-to-synonymous ratio found for PB1-F2 proteins [31,32]. The fact that PB1-F2 proteins with different amino acid sequences can use various mechanisms to achieve the same innate pathway antagonism indicates that this function is important, at least in avian cells. Whether evolution of the other viral proteins overrides selection towards one mechanism or another is yet to be determined.

Ethics statement
All animal work was approved and regulated by the UK Government Home Office under the project licence PPL 30/ 2952. All personnel involved in the procedures were licensed by the UK Home Office. Euthanasia of chickens was carried out by intravenous administration of sodium pentobarbital and confirmed through cervical dislocation. supplemented with 10 % foetal bovine serum (FBS) (Biosera, Inc.), 1 % penicillin/streptomycin (Sigma-Aldrich) and 1 % non-essential aa (Sigma-Aldrich), and maintained at 37 C in 5 % CO 2 .

Cells
Primary chicken embryo fibroblasts (CEFs) were produced from 10-day-old Rhode Island Red chicken embryos from specific-pathogen-free (SPF) hens. Embryos were homogenized and trypsinized with 0.25 % trypsin/EDTA solution (Sigma-Aldrich). The homogenate was resuspended in complete media and passed through a Falcon Cell Strainer (Fisher) before being seeded onto plates for experimentation. Influenza infections were conducted in Eagle's minimum essential medium (EMEM) supplemented with 0.35 % bovine serum albumin (BSA) (Sigma-Aldrich), 1 % penicillin/streptomycin and 0.25 µg ml À1 TPCK-treated trypsin.
Primary chicken bone marrow-derived macrophage cells (ckBMDM) were produced from male 3-week-old Rhode Island Red chickens from a SPF flock. Femurs were extracted, and the marrow was flushed and liberated with phosphate-buffered saline (PBS). The cells were homogenized, passed through a Falcon Cell Strainer and then isolated via centrifugation on Histopaque 1083 (Sigma-Aldrich). The cells were washed and resuspended in Roswell Park Memorial Institute medium-1640 (RPMI-1640) medium (Invitrogen) supplemented with 5 % FBS and 5 % chicken serum (Pirbright), 2.95 g l À1 TBP (Sigma-Aldrich), 1 mM sodium pyruvate (Sigma-Aldrich) and 1 % penicillin/ streptomycin, and seeded onto plates for experimentation. BMDMs were stimulated to differentiate by the addition of 10 ng ml À1 of chicken colony-stimulating factor 1 (cCSF-1) for 5 days (The Roslin Institute). Influenza infections were conducted in the same media without serum. Chicken gene transcript quantification Nasal tissues were dissected, weighed and added to 700 µl of TRIzol reagent (Life Technologies). Tissue was homogenized with a 5 mm stainless steel bead (Qiagen) for 4 min at

Viral RNA quantification
Quantification of vRNA was performed as previously described [6]. SuperScript III One-Stepreverse transcription-PCR (RT-PCR)system with Platinum Taq  Co-immunoprecipitation of PB1-F2 and ck MAVs or ck IKKb Chicken MAVS-V5 or IKKb-FLAG expression plasmids were co-transfected into DF-1 cells along with PB1-F2 containing a FLAG or V5 tag as appropriate, using lipofectamine 2000 according to the manufacturer's instructions. After incubation for 24 h after transfection, cells were lysed in Pierce Immuoprecipitation Lysis Buffer (Thermo Fisher) and then processed immediately. The lysates were precleared by adding 0.5 mg ml À1 of Pierce Protein G Magnetic Beads (Thermo Fisher) and a 12-tube magnetic rack (Qiagen). The samples were incubated overnight at 4 C with 10 µg ml À1 of primary antibody [mouse anti-FLAG (Sigma M2)] or rabbit anti-V5 antibody (Genetex) and then incubated at 4 C for 6 h with 0.5 mg ml À1 of magnetic beads. The beads were magnetically isolated and washed three times in Pierce Immuoprecipitation Lysis Buffer (Thermo Fisher). The beads were then boiled in SDS-loading dye for 10 min and used for Western blot analysis.

VSV-GFP interferon bioassay
The supernatants from three independent experiments involving cells infected with each indicated influenza virus at an m.o.i. of 3 were heated at 60 C for 30 min to inactivate influenza virus. This was confirmed by performing an incell enzyme-linked immunosorbent assay (ELISA) for influenza nucleoprotein. In technical triplicate, the supernatants were twofold serially diluted in complete media on DF-1 cells (4Â10 4 cells well À1 ) in a 96-well opaque white plate (Pierce) and incubated for 12 h. Cells were infected with GFP-tagged vesicular stomatitis virus (VSV-GFP) [38] at an m.o.i. of 3 in Opti-MEM with no phenol red (Thermo). Then, 12 h post-infection, the medium was removed and GFP fluorescence was measured using a GloMax multi-plate reader fluorescence module (Promega). SF9-produced chicken interferon-b was used as a positive control (The Pirbright Institute). No VSV-GFP virus was used to establish background fluorescence and media-only incubations were used to establish maximum fluorescence. The percentage antiviral effect was then calculated using the following formula and displayed graphically: % antiviral effect ¼ sample fluor À background fluor ð Þ Â 100 max fluor À background fluor À 100 Luciferase reporter assays For assaying interferon modulation, a pGL3 luciferase (luc) reporter vector containing the promoter regions from chicken IFN-2 upstream of a firefly luc gene was used. For assaying NF-kB activity, a similar reporter vector was used containing six tandem repeats of the NF-kB consensus binding sequence (5¢-GGGACTTTCC-3¢) upstream of a thymidine kinase (TK) minimal promoter. In triplicate, cells were transfected accordingly with 100 ng of Renilla luc pCAGGs expression plasmid, 500 ng chIFN-2-, chMX-or NF-kB luciferase reporter plasmids and 2 µg PB1-F2 protein expression plasmids. For assays involving polyinosinic : polycytidylic acid [poly (I : C)] stimulation, 6 h post-transfection cells were retransfected with 0.25 µg poly I : C (Sigma). After the experimental time point was reached, cells were lysed and read using the Stop and Glo reagents (Promega). Each individual firefly and Renilla value was checked for consistency, and then the firefly luciferase signals were normalized to Renilla signals.
Cycloheximide chase assay DF1s were transfected with 250 ng ul À1 of the appropriate V5-PB1-F2 or GFP plasmid. Twenty-four hours post-transfection, cells were treated with 100 ug ml À1 cycloheximide (Sigma). Samples were collected at appropriate time points by washing with PBS and lysing with Laemmli buffer. Samples were visualized on Western blots using mouse anti-V5 antibody (abcam) and IRDye donkey anti-mouse (800 nm) (LiCor). Actin was used as a loading control and visualized using mouse anti-actin antibody (abcam) with IRDye donkey anti-mouse (800 nm) (LiCor). GFP was used as a stable protein control. Bands were quantified using ImageJ (NIH) and PB1-F2 was normalized to the actin loading control.

Viral polymerase activity assay
In triplicate, DF-1 cells were transfected with 20 ng of PA, 160 ng of NP, and 80 ng of each of the PB1 and PB2 pCAGGs expression plasmids. Transfection reactions also included 250 ng of chicken-specific viral RNA-dependent RNA polymerase firefly luciferase reporter (described previously [39]) and 100 ng of a Renilla luciferase pCAGGs expression plasmid. Cells were incubated for 24 h at either 37 or 41 C before being lysed in 100 µl of 1Â passive lysis buffer (Promega). Plates were frozen for 30 min at À80 C and defrosted before reading. Then 10 µl of lysate was loaded onto a 96-well opaque white plate (Pierce) and analysed on a GloMax multi-plate reader (Promega) with 50 µl of LARII and Stop and Glo reagents (Promega). Each individual firefly and Renilla value was checked for consistency, and then the firefly luciferase signals were normalized to Renilla signals.

Viral growth curves
In triplicate, in six-well plates, chicken kidney cells were infected with a virus m.o.i. of 0.0001 1 h at 37 C. The cells were washed with PBS and incubated at 37 C with 2 ml of appropriate media containing no serum. Then 500 µl of supernatant was taken at the indicated time points after infection and replenished, and the samples were stored at À80 C. All samples were titrated via plaque assay on MDCK cells including the input.

Funding information
This work described herein was funded by a The Pirbright Institute studentship grant; BBS/E/00001759, and BBSRC ISP grants; BBS/E/I/ 00001650, BBS/E/I/00007034 and BBS/E/I/00007038. The funders had no role in study design, data collection, data interpretation, or the decision to submit the work for publication.