Inhibition of Avian Influenza A Virus Replication in Human Cells by Host Restriction Factor TUFM Is Correlated with Autophagy

ABSTRACT Avian influenza A viruses generally do not replicate efficiently in human cells, but substitution of glutamic acid (Glu, E) for lysine (Lys, K) at residue 627 of avian influenza virus polymerase basic protein 2 (PB2) can serve to overcome host restriction and facilitate human infectivity. Although PB2 residue 627 is regarded as a species-specific signature of influenza A viruses, host restriction factors associated with PB2627E have yet to be fully investigated. We conducted immunoprecipitation, followed by differential proteomic analysis, to identify proteins associating with PB2627K (human signature) and PB2627E (avian signature) of influenza A/WSN/1933(H1N1) virus, and the results indicated that Tu elongation factor, mitochondrial (TUFM), had a higher binding affinity for PB2627E than PB2627K in transfected human cells. Stronger binding of TUFM to avian-signature PB2590G/591Q and PB2627E in the 2009 swine-origin pandemic H1N1 and 2013 avian-origin H7N9 influenza A viruses was similarly observed. Viruses carrying avian-signature PB2627E demonstrated increased replication in TUFM-deficient cells, but viral replication decreased in cells overexpressing TUFM. Interestingly, the presence of TUFM specifically inhibited the replication of PB2627E viruses, but not PB2627K viruses. In addition, enhanced levels of interaction between TUFM and PB2627E were noted in the mitochondrial fraction of infected cells. Furthermore, TUFM-dependent autophagy was reduced in TUFM-deficient cells infected with PB2627E virus; however, autophagy remained consistent in PB2627K virus-infected cells. The results suggest that TUFM acts as a host restriction factor that impedes avian-signature influenza A virus replication in human cells in a manner that correlates with autophagy.

S easonal influenza viruses circulating in human populations infect millions of people annually and have profound and costly public health consequences (1). In the 20th century, several influenza pandemics, such as those that occurred in 1918, 1957, and 1968, devastated human populations. Natural reservoirs of influenza A virus exist in wild waterfowl and domestic poultry (2), and the ability of such avian influenza A viruses to cross host barriers and ultimately adapt to humans is a key factor driving outbreaks. Most avian influenza A viruses are unable to infect humans, but those that have crossed the host barrier can cause serious infections (3); for example, an H5N1 influenza A virus highly pathogenic to both chickens and humans emerged in Hong Kong in 1997 and eventually killed 6 of 18 infected persons (4). Several fatal H5N1 influenza A virus outbreaks have occurred since then. In 2013, new avian-origin H7N9 influenza A viruses were reported in eastern China (5), currently yielding a fatality rate of Ͼ36.8% in humans (Food and Agriculture Organization. H7N9 situation update, 24 March 2017 [http://www.fao.org/ag/againfo/programmes/en/empres/h7n9/Situation _update.html]), on the basis of confirmed cases. These outbreaks underscore the need to understand how influenza A viruses cross species barriers and develop infectivity in humans.
Influenza A viruses are enveloped negative-stranded RNA viruses of the Orthomyxoviridae family that possess segmented genomes. Each ribonucleoprotein (RNP) complex of an influenza virion consists of an RNA strand packaged with four viral proteins, polymerase basic protein 1 (PB1), polymerase basic protein 2 (PB2), polymerase acid (PA) protein, and nucleoprotein (NP). The RNP complex drives viral replication in the host nucleus, enabling the virus to hijack host cell resources (6). A single substitution of glutamic acid (E, avian signature) for lysine (K, human signature) at residue 627 of PB2 is a major determinant for viruses to overcome host restrictions (7), as this substitution restores viral polymerase activity (8) and allows viruses to replicate efficiently in mammalian cells and animal models (9)(10)(11).
The host factors involved in the adaptive mechanism of PB2 627 have been the focus of much research, and several hypotheses have emerged. In the first hypothesis, it is suggested that positive factors control the adaptive mechanism (12); for instance, importin-␣1 and importin-␣7 bind more strongly to the PB2 627 K-RNP complex, and this facilitates viral replication in human cells (13). Importin-␣7 knockout (KO) mice are less susceptible to infection with viruses with PB2 627 K (13,14), and thus, importin-␣7 is considered to be a key positive factor. The second hypothesis posits that no restriction factors exist, but the decrease or disappearance of positive factors has an impact on the PB2 627 E-RNP complex (15). For example, for viruses carrying avian-signature PB2 627 E, chicken ANP32A enhances polymerase activity and viral replication in human cells to levels comparable to those of human-signature PB2 627 K, while human ANP32A lacks 33 key functional amino acids and therefore restricts the replication of avian influenza viruses in human cells (16). The third hypothesis posits that restriction factors selectively inhibit the avian-signature PB2 627 E-RNP complex in mammalian cells (17); for example, RIG-I has greater binding affinity than NP for PB2 627 E, and this disrupts the viral replication machinery in human cells (18); however, RIG-I knockdown failed to rescue PB2 627 E polymerase activity, suggesting that other restriction factors remain to be identified.
Although PB2 is localized primarily in the host nucleus and most PB2-interacting human proteins are nucleus related, PB2 signals have also been detected in the mitochondria, and a mitochondrial-targeting signal is present at the N terminus of PB2 (19,20). Previous research has reported that PB2 can interact with the mitochondrial antiviral signaling (MAVS) protein to disrupt type I interferon (IFN) induction (21). It is possible that host mitochondrial factors can also interact with PB2 to disrupt viral adaptive mechanisms, and in this study, we found that Tu elongation factor, mitochondrial (TUFM, also known as EF-Tu, P43, or COXPD4), can act as a selective PB2 627 E restriction factor. TUFM is a fundamental mitochondrial protein that has been implicated in protein translation, GTPase activity, and RNA binding (22), and it has also been reported to act as an NLRX1-interacting partner that enhances autophagy while inhibiting MAVS protein-induced IFN-␤ expression in vesicular stomatitis virus (VSV)infected cells (23). Host defense against microbes and viruses, including influenza A virus, is known to be one of the triggers of autophagy (24,25). Influenza A viruses have also been shown to reduce autophagy via direct interactions between viral matrix 2 protein and the microtubule-associated protein 1 light chain 3 (LC3) autophagic protein (26).
Our previous research on species-associated genomic signatures sought to delineate the genetic boundary between avian and human influenza A viruses (27). In this study, we identified TUFM, a novel host restriction factor that demonstrated a higher binding affinity for avian-signature PB2 627 E than for human-signature PB2 627 K and that selectively inhibited PB2 627 E viral replication in human cells, potentially through the mediation of cellular autophagy.

RESULTS
Proteomic analysis of differential interactions with human cellular proteins for influenza A/WSN/1933(H1N1) virus PB2 627 K and PB2 627 E. To obtain differentially expressed PB2 constructs, FLAG-tagged PB2 of human influenza A/WSN/1933(H1N1) (WSN) virus was generated and named WSN PB2 627 K. A K627E substitution was then generated by site mutation, and the result was named WSN PB2 627 E. WSN PB2 627 E is defined as avian-like because its backbone was derived from human influenza A virus with only a single substitution in PB2 (K627E). Human proteins associated with WSN PB2 627 K and PB2 627 E were collected by FLAG immunoprecipitation (FLAG-IP), separated by gradient SDS-PAGE, subjected to silver staining (Fig. 1A), and identified by matrixassisted laser desorption ionization (MALDI-TOF) mass spectrometry (MS) analysis. For detailed proteomic results see Table S1 in the supplemental material. A total of 168 putative associated proteins were identified, including 91 PB2 627 K-associated and 105 PB2 627 E-associated proteins. After 28 common proteins (gray) associated with both PB2 proteins were excluded, 63 PB2 627 K-specific and 77 PB2 627 E-specific proteins were classified (light blue and pink, respectively, in Fig. 1B).
For details regarding the putative interacting proteins in each group that are enriched in biological processes, see Table S2. The protein-protein interactions show highly cohesive networks and reveal abundant functional interactions within networks ( Fig. 1C and D). The results indicate that PB2 627 K-and PB2 627 E-associated proteins have different protein interaction profiles; however, interactions among nuclear proteins related to transcriptional regulatory functions, such as NCL, NPM1, PRMT5, ZNF224, and WDR77, are similar in both networks of PB2 627 K-and PB2 627 E-associated proteins. Interestingly, the interactions among the RPL3, RPS3, and EEF2 proteins in both networks are, respectively, connected by RPL8 and TUFM ( Fig. 1C and D). TUFM is known to participate in various biological processes and possesses translation factor activity, translation elongation factor activity, GTPase activity, and nucleotide binding capability (see Table S2). Mitochondrial factors involved in influenza virus PB2 627 adaptive mechanisms have not been well characterized to date, and therefore, TUFM was selected for further validation, as this study hypothesized that host restriction factors may interact exclusively with PB2 627 E to inhibit avian influenza virus replication in human cells.
Demonstration of higher binding affinity of TUFM for WSN PB2 627 E. The proteomic profile of host factor TUFM is presented in Fig. 2A, and MS-matched TUFM peptides are listed in Fig. 2B. Total cell lysates were harvested from PB2-transfected 293T cells, and Western blotting was performed after FLAG-IP. The results indicate that TUFM has higher binding affinity for avian-signature PB2 627 E than for human-signature PB2 627 K in transfected 293T cells (lanes 7 and 8 of Fig. 2C), thus validating the proteomic analysis results.
One key concern regarding these findings is whether a TUFM homolog exists in avian species, and if so, whether such an avian TUFM homolog would demonstrate similar binding affinity. We sought to address this question in the most extensively studied avian species to date (as well as the species often hardest hit by avian influenza outbreaks), the chicken (Gallus gallus). Previously, the G. gallus genome database (version 6.5; updated on 4 January 2016) contained partial nucleotide sequences listed as chicken TUFM (chTUFM; XM_015274224), but owing to insufficient nucleotide sequence information, the generation of constructs to enable chTUFM protein expression or small interfering RNA (siRNA) design was very difficult. We therefore resolved the complete coding sequence (CDS) of chTUFM (GenBank accession no. KY769204). For a detailed sequence analysis of the key domains in human TUFM and its homologs in chickens and other species, see Fig. S1. We further investigated the binding affinity of chTUFM for WSN PB2 627 E in avian cells. After successfully constructing full-length chTUFM with a Myc tag expression plasmid, we performed Myc immunoprecipitation (Myc-IP) to compare the binding affinity of chTUFM with WSN PB2 627 K or PB2 627 E in DF-1 cells. We found that chTUFM exhibited no associations with either WSN PB2 627 K or PB2 627 E (see lanes 6 and 7 of Fig. S2A), in contrast to results derived from cotransfected Myc-tagged human TUFM and WSN PB2 627 E plasmids (see lane 8 of Fig. S2A) in DF-1 cells. From sequence alignment and phylogenetic analysis of TUFM homologs (see Fig. S1), it was observed that TUFM homologs in avian species appear to have undergone extensive evolution compared to mammalian species (see Fig. S1B to D). In addition, substitutions of 20, 7, and 3 amino acids (aa) were, respectively, noted in  Table S1). (B) Venn diagram of proteins associated with PB2 627 K and PB2 627 E. Each number is the number of proteins in each category. (C, D) Protein-protein interaction networks of PB2 627 K-associated (C) and PB2 627 E-associated (D) proteins. Common proteins are shown as gray nodes, while PB2 627 K-and PB2 627 E-specific proteins are shown as light blue and pink nodes, respectively. The different colors of the connecting lines represent the evidence backing each interaction, such as experiments (pink) and curated databases (cyan) for known interactions, gene fusion (red) for predicted interactions, coexpression (brown), and text mining (yellow). Interactions and connections with at least two proteins are presented. domain 1 (D1), D2, and D3 of avian TUFM compared with mammalian TUFM (see Fig. S1B), suggesting that avian TUFM may be fundamentally different from mammalian TUFM, and thus, chTUFM may have less affinity for the WSN PB2 627 binding site (see Fig. S2A).
Higher binding affinity of TUFM for avian-signature PB2s from the 2009 pdmH1N1 and 2013 H7N9 influenza viruses. We further assessed whether TUFM also has higher binding affinity for avian-signature PB2s derived from the 2009 pandemic H1N1 (pdmH1N1) and 2013 H7N9 influenza A viruses. In the 2009 swine-origin The data are representative of three independent experiments. Total cell lysates (2 mg) were assayed by FLAG-IP, with results shown at the right and input proteins (50 g) shown at the left. ␣-PB2 and ␣-FLAG represent FLAG-tagged PB2 proteins. ␣-Actin represents an NC, as it was not found to interact with PB2 (see Table S1). TUFM protein bands were quantitatively measured by ImageJ software. (F) Domain mapping of TUFM. Domain structures of human TUFM and predicted molecular masses of wild-type and truncated proteins are illustrated on the left. Myc-tagged wild-type TUFM or truncated TUFM Δ1 or TUFM Δ2&3 was cotransfected into 293T cells with FLAG-tagged WSN PB2 627 E. Cells transfected with the empty vector were used as an NC. Cell extracts were immunoprecipitated by FLAG-IP, and then PB2 627 E associations with truncated TUFM proteins were probed with anti-Myc antibody, and the results are shown on the right. TUFM Inhibition of Avian Influenza in Human Cells ® pdmH1N1 influenza A viruses, the substitutions in PB2, G590S and Q591R, are known genomic signatures that facilitate escape from host restriction mechanisms in human cells (28). We found that TUFM similarly demonstrated higher binding affinity for avian-signature PB2 590 G/PB2 591 Q than for human-signature PB2 590 S/PB2 591 R in transfected human cells (lanes 7 and 8 of Fig. 2D). In 2013 avian-origin H7N9 viruses that crossed species barriers to infect humans, we previously discovered that humansignature PB2 627 K was crucially associated with human infectivity in such viruses as well (29). We therefore transfected FLAG-tagged PB2 627 K or PB2 627 E from influenza A/Anhui/1/2013(H7N9) virus into 293T cells, and TUFM was also found to have higher affinity for avian-signature PB2 627 E than for human-signature PB2 627 K (lanes 7 and 8 of Fig. 2E). These results indicate that TUFM favors binding with avian-signature PB2 over humansignature PB2 in the WSN, pdmH1N1, and H7N9 influenza A virus strains.
The PB2 D701N substitution is a key determinant of the host range of influenza A viruses, and our previous research found that the PB2 D701N substitution restored polymerase activity in H7N9 viruses that retained avian signature PB2 627 E (29). In this study, we also examined the relevance of this substitution to the binding affinity of TUFM and found that H7N9 wild-type (701D; see lane 7 of Fig Fig. S3B), which are known to associate with TUFM ( Fig. 2C to E). On the basis of these findings, interactions between TUFM and the PB2 701D/N site are considered to be unlikely.
TUFM D1 is required for association with avian-signature PB2 627 E. To better elucidate the interaction between TUFM and avian-signature PB2 627 E, we sought to identify the essential TUFM domain involved in the binding of these two proteins. TUFM D1, D2, and D3 and the corresponding start and end amino acid residues are shown in Fig. S1A. GTP-binding D1 was previously shown to be essential for Atg5-Atg12 conjugate recruitment (23), and therefore, we generated TUFM Δ1 (deletion of D1) and TUFM Δ2&3 (deletion of D2 and D3) truncations for further investigation. We found that the interaction between PB2 627 E and TUFM required D1, given that its deletion (TUFM Δ1) abolished this interaction (lane 7 of Fig. 2F); however, D2 and D3 did not appear to be as critical to the interaction (lane 8 of Fig. 2F). A similar assay was performed with wild-type and truncated chTUFM, but none of these proteins demonstrated any association with PB2 627 E (see lanes 6 to 8 of Fig. S2B), compared to a positive control with human TUFM (see lane 5 of Fig. S2B; similar to lane 6 of Fig. 2F). Sequence alignment comparison of avian TUFM and mammalian TUFM (see Fig. S1B) showed that extensive polymorphisms, including 20 aa substitutions, occurred in D1 (as known as GTP-binding D1), and thus, the lack of association between chTUFM and PB2 627 E may be due to these in sequence and structure differences in D1.
TUFM inhibits avian-signature PB2 627 E viral replication in human cells. We sought to investigate the impact of TUFM on the replication of influenza A viruses in human cells. To obtain viruses carrying avian-signature PB2 627 E instead of PB2 627 K, reverse genetics was used to generate the K627E substitution of PB2 in the rWSN PB2 627 K virus to derive the rWSN PB2 627 E virus. Human A549 cells were used for initial viral growth kinetic studies, and A549 cells treated with TUFM siRNA (si-TUFM), si-TUFM plus the FLAG-tagged TUFM plasmid (TUFM-FLAG, for overexpression of TUFM), or a negative control (NC) were, respectively, infected with either rWSN PB2 627 K or PB2 627 E at a multiplicity of infection (MOI) of 0.001 or 2 ( Fig. 3A to F). The growth curves of rWSN PB2 627 K virus showed no significant difference in A549 cells in either multicycle or single-cycle experiments ( Fig. 3A and D), indicating that TUFM had no effect on rWSN PB2 627 K. Conversely, viral yields of rWSN PB2 627 E in TUFM-deficient (treated with si-TUFM) A549 cells significantly increased from 4.4-to 9.3-fold over controls at 36 to 60 h postinfection (hpi) with an MOI of 0.001 (Fig. 3B) and increased 5.0-to 5.9-fold over controls at 6 to 12 hpi with an MOI of 2 (Fig. 3E). Viral yields in A549 cells transfected with si-TUFM plus TUFM-FLAG were similar to controls (Fig. 3A, B, D, and E), indicating an absence of off-target effects with si-TUFM, while also confirming that rWSN PB2 627 E replication increases in TUFM-deficient A549 cells.
We further used normal human bronchial epithelial (NHBE) cells for investigation, but as NHBE cells became more fragile after siRNA transfection by nucleofection, only single-cycle experiments were conducted, and additional TUFM-FLAG plasmid transfections were not performed. Results showed that the viral growth kinetics of rWSN PB2 627 K at 3 to 12 hpi with an MOI of 2 were not affected in TUFM-deficient NHBE cells, but rWSN PB2 627 E viral yields significantly increased from 5.2-to 8.0-fold over controls at 6 to 12 hpi with an MOI of 2 ( Fig. 3G to I). These results indicate that the presence of TUFM specifically impeded the replication of the avian-signature PB2 627 E virus but not that of the human-signature PB2 627 K virus. TUFM Inhibition of Avian Influenza in Human Cells ® A stable TUFM heterozygotic ϩ/Ϫ KO MDCK cell line was generated with CRISPR/ Cas9 endonuclease (see Fig. S4A), and it contained a second allele with a 2-nucleotide (nt) deletion in comparison with the wild type (see Fig. S4B). The deletion at nt 859 to 860 engendered a frameshift from aa 106 and ultimately resulted in a stop codon that ended translation at aa 114, as indicated by sequencing results (see Fig. S4C). Viral yields of rWSN PB2 627 E substantially increased from 12.4-to 14.5-fold over the wild type in the TUFM ϩ/Ϫ KO MDCK cell line at 48 to 72 hpi with an MOI of 0.001 (see Fig. S4D), indicating that TUFM inhibits avian-signature PB2 627 E viral replication in canine cells. Protein levels of the stable TUFM ϩ/Ϫ KO MDCK cell line were confirmed by Western blotting to be successfully knocked down (see Fig. S4E). No significant difference between the viability of TUFM-deficient human cells and TUFM ϩ/Ϫ KO MDCK cells was observed in 3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide (MTT; Millipore) assays (see Fig. S4F).
TUFM-deficient A549 cells transfected with a TUFM-FLAG plasmid were found to have restricted viral replication compared to that of controls, as shown by viral titers (blue lines in Fig. 3B and E). However, wild-type A549 cells transfected with the TUFM-FLAG plasmid did not display lower viral titers than controls (see Fig. S5A to E), presumably because TUFM is already abundant in A549 cells. Growth curves of rWSN PB2 627 E viruses in different cell systems (see Fig. S5F to J) showed that higher viral titers could be seen in avian DF-1 cells (see Fig. S5G and I) than in human A549 cells (see Fig. S5B and D), indicating that avian-signature influenza A virus does not replicate efficiently in human cells.
We further sought to investigate the impact of chTUFM on the replication of both rWSN PB2 627 K and PB2 627 E viruses in chTUFM-deficient avian cells. The growth curves of both viruses showed no significant difference in either multicycle or single-cycle experiments conducted with chTUFM-deficient DF-1 cells (see Fig. S6), indicating that chTUFM had no effect on both viruses in avian cells. No significant differences were observed in the viability of chTUFM-deficient avian cells and control cells (see Fig. S6F). In addition, the growth curves of both viruses also showed no significant difference in human and avian cells exogenously expressing chTUFM (see Fig. S5K to T). Therefore, the inhibition of avian-signature PB2 627 E viral replication by TUFM appears to be specific for human cells (Fig. 3) but not chicken cells (see Fig. S6).
Colocalization and copurification of TUFM-PB2 in mitochondria Since increased polymerase activity is one of the characteristics of viruses bearing the PB2 E627K and G590S/Q591R signatures and is also believed to be one of the key mechanisms that facilitate escape from host restriction mechanisms in humans, we sought to investigate the effects of TUFM on the polymerase activity in human cells. However, no significant difference in polymerase activities from both WSN PB2 627 K/E and pdmH1N1 PB2 590S/G-591R/Q RNP-NP complexes at both 37°C and 33°C in TUFM-deficient 293T cells was observed (see Fig. S7), although slight variations in the polymerase activity of WSN PB2 627 E at 37°C were noted (see Fig. S7A). We speculated that polymerase activity assays that mimic the viral RNP-NP complex in its nuclear functions may have a limited ability to measure the effect of mitochondrial proteins on viral replication, and therefore, we further investigated the interaction between TUFM and PB2 in mitochondria.
The subcellular localization of TUFM and PB2 in cells infected with either rWSN PB2 627 K or rWSN PB2 627 E was examined by immunofluorescence assay (IFA). At 9 hpi at an MOI of 10, colocalization of TUFM with either PB2 627 K or PB2 627 E was detected in mitochondria (yellow spots in the merged images of Fig. 4A). No significant localization differences were observed between PB2 627 K and PB2 627 E, and TUFM signals were detected in mitochondria regardless of whether or not there was viral infection (Fig. 4A). Qualitative analysis of the interaction between TUFM and PB2 was subsequently performed via mitochondrial fractionation, followed by TUFM immunoprecipitation (TUFM-IP) in 293A cells. 293A cells were selected for this experiment because of their abundant mitochondrial fraction, their genetic background similar to that of the 293T cells used in the initial experiments of this study ( Fig. 1 and 2), and their strong adherence to plastic dishes for facilitation of viral infection. Mitochondrial fractions were isolated from 293A cells infected with either rWSN PB2 627 K or rWSN PB2 627 E (Fig. 4B), and the mitochondrial marker COX4 was detected at equal levels in input controls (lanes 4 to 6 of Fig. 4B). Quantitative analysis of copurified TUFM-PB2 627 K and TUFM-PB2 627 E in mitochondrial fractions was performed, and TUFM-PB2 627 E levels were found to be 1.4-fold higher than TUFM-PB2 627 K levels (lanes 9 and 10 of Fig. 4B), suggesting that TUFM has higher levels of interaction with avian-signature PB2 627 E than with human-signature PB2 627 K in the mitochondria of infected human cells. Taken together, these results indicate selectively higher binding affinity for avian-signature PB2 627 E by TUFM in mitochondria (Fig. 4B), and this may be correlated with the inhibitory effect of TUFM on avian-signature influenza virus replication in human cells (Fig. 3). , and anti-lamin B1 (as a nuclear marker) antibodies. Mitochondrial fractions used for input control (25 g, lanes 4 to 6) prior to immunoprecipitation (IP) were also immunoblotted with the antibodies listed above. Fractions were subject to IP with either IgG (lane 7, as an NC) or anti-TUFM antibody (lanes 8 to 10). Immunoprecipitated mitochondrial fractions were then subjected to immunoblotting with anti-PB2 and anti-TUFM antibodies, and quantitative results of PB2 were divided by TUFM levels and are presented as the mean Ϯ the standard error of the mean. Three independent experiments were performed.

TUFM interactions with PB2 627 K and PB2 627 E do not differentially affect type I IFN promoter activity.
Previous research has reported that TUFM interacts with NLRX1, Atg12, Atg5, and Atg16L1 to form a molecular complex in mitochondria, and this complex not only enhances VSV-induced autophagy but also inhibits type I IFN expression by acting against RIG-I (23). Though NLRX1, Atg5, Atg12, and Atg16L1 were not listed among the 168 putative associated proteins that we identified as interacting with the WSN PB2 627 K and PB2 627 E viral proteins (see Table S1), immunoprecipitates of FLAG-tagged WSN PB2 627 K or PB2 627 E pulled down by FLAG-IP revealed that PB2 also binds to endogenous NLRX1 but probably not endogenous Atg5-Atg12 or endogenous Atg16L1 (see Fig. S8A). However, the amount of endogenous NLRX1 associated with PB2 627 K and PB2 627 E was, respectively, only 3.0% and 32.5% of the PB2-TUFM binding levels (see lanes 7 and 8 of Fig. S8A). We subsequently assessed whether the PB2-TUFM interaction competes with the TUFM-NLRX1 interaction and found that levels of exogenously overexpressed NLRX1 slightly decreased in response to increasing levels of PB2 627 E (see lanes 6 to 8 of Fig. S8B). The PB2-TUFM interaction therefore may compete with the interaction between TUFM and NLRX1. Interestingly, expression of TUFM-FLAG alone was able to reduce IFN promoter activity to 14% of that of cells exogenously expressing the caspase activation and recruitment domain (CARD) of RIG-I (see lane 3 of Fig. S8C), and this confirmed previous research showing that TUFM can inhibit IFN (23). Moreover, IFN promoter activity was, respectively, suppressed to 5.1% and 5.6% of that of cells exogenously expressing RIG-I following cotransfection of TUFM with PB2 627 K or PB2 627 E, with no significant difference (see lanes 5 and 7 of Fig. S8C).
TUFM-dependent autophagy may selectively influence avian-signature PB2 627 E virus replication. As previous research indicated that TUFM can enhance autophagy in conjunction with reduction of type I IFN promoter activity in VSV-infected human cells (23), we also sought to ascertain whether TUFM inhibition of avian-signature PB2 627 E influenza A virus in human cells is associated with autophagy. The subcellular localization of PB2 and LC3, a specific marker of autophagosome formation, was examined in infected A549 cells by IFA. LC3 signals were detected in both rWSN PB2 627 K-and rWSN PB2 627 E-infected cells but were less visible in mock-infected controls (Fig. 5A). LC3 punctate dots (shown in green) were calculated per cell (Fig. 5B), and the results indicated that LC3 levels were lower in rWSN PB2 627 E-infected cells than in rWSN PB2 627 K-infected cells. Yellow dots represent colocalized PB2 and LC3 (right side of Fig. 5A) and show that autophagy occurred in both rWSN PB2 627 K-and rWSN PB2 627 Einfected cells. Quantitative analysis of LC3 was conducted by Western blotting of infected 293A cell lysates, with two forms of LC3 separated, cytoplasmic LC3-I and autophagic LC3-II associated with the autophagosome membrane. As the conversion of LC3-I to LC3-II is reflective of autophagic induction (30), LC3-II levels relative to those of ␣-tubulin controls were used to quantify autophagy levels. Both LC3-II/␣-tubulin and LC3-II/LC3-I ratios in rWSN PB2 627 E-infected cells were 50.6% and 59.7% of those in rWSN PB2 627 K-infected cells (Fig. 5C and D), indicating that autophagy levels were lower in rWSN PB2 627 E-infected cells than in rWSN PB2 627 K-infected cells. rWSN PB2 627 E viral titers were almost 10-fold lower than rWSN PB2 627 K viral titers at 9 hpi with an MOI of 2 (Fig. 5E), which corroborates the fact that avian influenza A virus does not replicate efficiently in human cells.
Autophagy levels and viral titers remained constant with no significant difference in TUFM-deficient human cells infected with rWSN PB2 627 K virus (Fig. 6A to C). Interestingly, LC3-II/␣-tubulin and LC3-II/LC3-I ratios were, respectively, reduced to 44.3% and 52.1% in TUFM-deficient 293A cells (treated with si-TUFM) infected with rWSN PB2 627 E, compared to infected cells treated with control siRNA. LC3-II/␣-tubulin and LC3-II/LC3-I ratios were, respectively, restored to 104.3% and 112.5% of the control ratios in si-TUFM plus TUFM-FLAG transfected cells infected with rWSN PB2 627 E (Fig. 6A and B). rWSN PB2 627 E viral titers were increased 6.1-fold in TUFM-deficient cells at 9 hpi with an MOI of 2, while rWSN PB2 627 E viral titers were comparable to those of controls in si-TUFM plus TUFM-FLAG transfected cells (Fig. 6D; similar to Fig. 3E). The results indicate that TUFM-dependent autophagy was reduced in TUFM-deficient cells infected with PB2 627 E virus, and this correlated with an increase in viral titers (Fig. 6A, B, and D).

DISCUSSION
Avian influenza A viruses generally do not replicate efficiently in human cells, and this was also observed in this study, as viral titers of rWSN PB2 627 E were almost 10-fold lower than those of rWSN PB2 627 K (Fig. 5E), suggesting that host restriction factors inhibit the viral replication of avian influenza virus in human cells. TUFM has a higher binding affinity for PB2 627 E than for PB2 627 K in both transfected human cells (Fig. 2C) and the mitochondrial fraction of virus-infected human cells (Fig. 4B). Accordingly, we propose a model (Fig. 7) in which human TUFM interacts with and binds aviansignature PB2 627 E of influenza virus in mitochondria. Binding between TUFM and PB2 627 E is impaired in TUFM-deficient human cells (indicated by a dotted ellipse in Fig. 7), and this may allow free PB2 627 E to facilitate influenza virus replication in human cells; as a result, rWSN PB2 627 E viral growth was observed to increase (Fig. 3B, E, and H). In contrast, in human cells expressing TUFM or in TUFM-deficient human cells reconstituted to express TUFM-FLAG, TUFM-bound PB2 627 E is more prevalent, and this may either reduce free PB2 627 E or disrupt its ability to participate in viral replication, thus leading to lower viral yields (Fig. 3). We therefore propose that TUFM can be considered a host restriction factor for the inhibition of avian influenza virus replication in human cells. In addition, TUFM has been reported to promote autophagy in VSV-infected cells (23), and this study also found that TUFM promoted autophagy in avian-signature PB2 627 E-infected human cells, whereas TUFM deficiency decreased autophagy (Fig. 6). This TUFM-dependent autophagy may serve as a defensive mechanism against avian influenza virus replication in human cells.
Although autophagy is known to act as a critical host defense mechanism against viral infection in cells (24,25), influenza A virus is known to subvert autophagy for self-benefit. For example, influenza A viruses have been shown to reduce autophagy and maintain virion stability via direct interactions between viral matrix 2 protein and LC3 that block autophagosome fusion with lysosomes (26,31); however, in this case, viral titers were not significantly affected (31). Another study showed that autophagy and lung inflammation were suppressed by 3-methyladenine (an autophagy inhibitor) or siRNA of autophagy-related genes in mouse lungs and A549 cells infected with avian influenza virus H5N1 carrying PB2 627 K, but viral titers were also not significantly altered (32). Similarly, viral titers in TUFM-deficient human cells infected with rWSN PB2 627 K were not significantly affected (Fig. 6C), even though autophagy levels of rWSN PB2 627 K virus remained consistently stronger than those of rWSN PB2 627 E virus (Fig. 6A and B and 5C and D). These results suggest that TUFM-dependent autophagy differs greatly between avian-signature PB2 627 E and human-signature PB2 627 K viruses and imply that differences in TUFM specificity can affect the host response to influenza virus infection.
Proteomic analysis ( Fig. 1C and D; see Table S1) of PB2 627 K-specific proteins identified importin-␣4 and importin-␤1, which facilitate nuclear transport (12,14), while common proteins included HSP90, which is involved in the assembly and nuclear import of the RNP complex (33); ␣-tubulin, which assists in transport of the RNP complex from the nucleus to the cytoplasmic membrane (34); and NPM1, NCL, and DDX3X, which are upregulated in both H1N1 and H5N1 influenza A virus-infected cells (12). This indicates that the current experimental design is effective and captures previously documented proteins.
Regarding the role of TUFM in other species during influenza A virus infections, mouse TUFM was previously found to be slightly upregulated at 72 hpi in mouse lung tissue infected with avian H9N2 PB2 627 E influenza A virus at 10 4 PFU (35). A partial chTUFM protein sequence (352 aa; UniProt code P84172) identified by MS was previously released (36) and was found to have 68.5% similarity to human TUFM. A comparison of polymorphisms in aligned TUFM proteins and phylogenetic analysis suggests that different versions of TUFM exist in avian and mammalian species (see Fig. S1B to D). Therefore, TUFM-associated host restriction machinery for avian influenza viruses may be present only in mammalian species and not in avian species. The PB2 E627K substitution that facilitates human adaptation of avian influenza viruses is being detected with greater frequency in human samples since its possible inception in a single human host while remaining absent from chicken samples (37). This allows us to hypothesize that the key restriction factor(s) associated with PB2 must be absent from FIG 7 Proposed model for how TUFM-dependent autophagy selectively influences PB2 627 E viral replication in infected human cells. The model shows TUFM has a stronger interaction with avian-signature PB2 627 E than human-signature PB2 627 K in mitochondria, and therefore host factor TUFM may selectively interact in mitochondria with the PB2 627 E protein upon avian influenza virus infection. Binding of TUFM-PB2 627 E was impaired in TUFM-deficient human cells (indicated by a dotted ellipse), and this may free PB2 627 E to facilitate avian influenza viral replication, thus leading to observable increases in viral growth. However, with more TUFM-bound PB2 627 E, either free PB2 627 E is reduced or PB2 627 E ineffectively participates in viral replication, thereby reducing viral growth. In addition, TUFM promotes autophagy in avian-signature PB2 627 E-infected human cells, whereas TUFM deficiency decreased autophagy. This model proposes that TUFM-dependent autophagy selectively influences avian-signature PB2 627 E viral replication in infected human cells, and may serve as an intrinsic defense against avian influenza viruses.
TUFM Inhibition of Avian Influenza in Human Cells ® avian cells; otherwise, the PB2 E627K substitution that facilitates viral adaptation would appear at a greater frequency among avian species as well. This hypothesis may explain why TUFM inhibits avian-signature PB2 627 E viral replication specifically in human cells (Fig. 3) and not chicken cells (see Fig. S6).
In conclusion, we used a differential proteomic approach to identify a novel host restriction factor, TUFM, that preferentially interacts with avian-signature PB2 627 E in the mitochondria of human cells infected with avian-signature influenza A virus. TUFM impedes the replication of avian-signature influenza virus in human cells, possibly through mediation of autophagy. Our findings provide new insight into the role of mitochondria in the host defense against avian influenza viruses and may have important implications for future antiviral research.
Mitochondrial fractionation followed by TUFM-IP. 293A cells were infected with either rWSN PB2 627 K or rWSN PB2 627 E virus at an MOI of 2 for 9 h. Mitochondria were isolated from 293A cells with a mitochondrial/cytosol fractionation kit (Millipore) in accordance with the manufacturer's instructions. Anti-TUFM antibody (3 g; GeneTex) was coupled to protein G-Sepharose beads (GE Healthcare) and then incubated with 1 mg of the mitochondrial fraction in accordance with the manufacturer's instructions. At least ten 15-cm dishes of 293A cells was collected for one reaction.
IFN promoter activity assay. A firefly luciferase reporter plasmid containing an IFN-␤ promoter (a gift from Michael Gale, University of Washington, Seattle, WA, USA), a pRL-TK Renilla luciferase control plasmid (for normalization of transfection efficiency), and a Myc-tagged RIG-I-CARD-expressing plasmid (N-RIG-I-Myc; provided by Helene M. Liu, National Taiwan University, Taiwan) were cotransfected into 293T cells in combination with a TUFM-FLAG, WSN PB2 627 K, or WSN PB2 627 E plasmid. At 24 h posttransfection, the firefly and Renilla luciferase activities of transfected cells were determined with a dualluciferase reporter assay (Promega) in accordance with the manufacturer's instructions. The normalized firefly luciferase activity represented the promoter activity of IFN-␤.
Nucleotide sequence accession number(s). The full-length cDNA sequence of chTUFM, as derived from CEF cells, has been deposited in GenBank under accession number KY769204. The proteomic data set was deposited in the IMEx database (IntAct, IM-25425) (43).

ACKNOWLEDGMENTS
We are grateful to Hiroshi Kida (Hokkaido University Research Center for Zoonosis Control, Sapporo, Japan) for providing a batch of monoclonal antibodies to influenza virus RNP that enlightened this study. We thank Nadia Naffakh for generously providing plasmids and viruses and Robert G. Webster for generously providing plasmids. We thank Kuo-Ming Lee and Rei-Lin Kuo for helpful discussions. We also thank the anonymous reviewers for their insightful comments on this study.
This work was financially supported by the Ministry of Science and Technology of Taiwan (grants MOST102-2321-B-182-014 and MOST104-2321-B-182-001 to Shin-Ru Shih) and partially supported by Chang Gung Memorial Hospital (grants CMRPD1A0081-83, CMRPD1D0041-43, CMRPD1D0311-313, and CMRPD1E0401-403 to Shin-Ru Shih). The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.