Glycosylation of the Hemagglutinin Protein of H5N1 Influenza Virus Increases Its Virulence in Mice by Exacerbating the Host Immune Response

ABSTRACT The highly pathogenic avian influenza (HPAI) H5N1 viruses continue to circulate in nature and threaten public health. Although several viral determinants and host factors that influence the virulence of HPAI H5N1 viruses in mammals have been identified, the detailed molecular mechanism remains poorly defined and requires further clarification. In our previous studies, we characterized two naturally isolated HPAI H5N1 viruses that had similar viral genomes but differed substantially in their lethality in mice. In this study, we explored the molecular determinants and potential mechanism for this difference in virulence. By using reverse genetics, we found that a single amino acid at position 158 of the hemagglutinin (HA) protein substantially affected the systemic replication and pathogenicity of these H5N1 influenza viruses in mice. We further found that the G158N mutation introduced an N-linked glycosylation at positions 158 to 160 of the HA protein and that this N-linked glycosylation enhanced viral productivity in infected mammalian cells and induced stronger host immune and inflammatory responses to viral infection. These findings further our understanding of the determinants of pathogenicity of H5N1 viruses in mammals. IMPORTANCE Highly pathogenic avian influenza (HPAI) H5N1 viruses continue to evolve in nature and threaten human health. Key mutations in the virus hemagglutinin (HA) protein or reassortment with other pandemic viruses endow HPAI H5N1 viruses with the potential for aerosol transmissibility in mammals. A thorough understanding of the pathogenic mechanisms of these viruses will help us to develop more effective control strategies; however, such mechanisms and virulent determinants for H5N1 influenza viruses have not been fully elucidated. In this study, we identified glycosylation at positions 158 to 160 of the HA protein of two naturally occurring H5N1 viruses as an important virulence determinant. This glycosylation event enhanced viral productivity, exacerbated the host response, and thereby contributed to the high pathogenicity of H5N1 virus in mice.

poultry industry in many countries, including Bangladesh, China, Egypt, India, Indonesia, and Vietnam (2). Since the first recognized human cases of HPAI H5N1 virus infection in Hong Kong in 1997, the World Health Organization has reported 854 confirmed human infection cases as of July 2016, with a fatality rate of over 50%. HPAI H5N1 viruses were reported to have limited potential for human-to-human transmission (3)(4)(5). Recent studies, however, have demonstrated that HPAI H5N1 viruses can be transmitted in ferrets or guinea pigs by respiratory droplets if they acquire certain mutations or reassort with pandemic 2009 H1N1 virus (6)(7)(8). Accordingly, it is essential for us to gain a thorough understanding of the pathogenic mechanisms of these viruses in order to develop more effective control strategies.
Genetic manipulation of influenza viruses and studies with different mammal models have led to the identification of several viral markers that are associated with virulence and pathogenesis (9)(10)(11)(12)(13)(14)(15)(16). Most notably, the reconstruction of the 1918 Spanish pandemic influenza virus and studies of HPAI H5N1 viruses have clearly established the complexity and multigenic characteristics of influenza virus pathogenesis. Hemagglutinin (HA), neuraminidase (NA), and the viral RNA polymerase complex play important roles in the high virulence of the 1918 pandemic influenza virus in mice (17,18). The cleavage site of HA plays crucial roles in the systemic replication and lethal infection of the H5 and H7 subtypes of influenza viruses in chickens (19) and mammals (10,20). Mutations in the M1 protein also contribute to the virulence of H5N1 influenza viruses in mice (9). Certain amino acids or regions of the NS1 protein subvert the antiviral immune response of the host and are essential to the pathogenicity of H5N1 influenza viruses in mice (11,15,21). The amino acids at positions 627 and 701 of PB2 are principal determinants of the high virulence of H5N1 influenza viruses in mammals (10,12); 30% and 5% of human H5N1 isolates have been reported to possess 627K and 701N in their PB2 proteins, respectively (http://www.ncbi.nlm.nih.gov/). Moreover, the amino acid at position 591 of PB2 is reported to be important for the efficient replication of pandemic H1N1 viruses in humans (22) and to substantially increase the pathogenicity of an avian H5N1 virus in mice (23). The PA protein directly contributes to the virulence of H5N1 avian influenza viruses in domestic ducks (24) and mice (25,26). Thus, although we are acquiring important data in this area, the mechanism of host range and the virulent determinants of H5N1 influenza viruses have yet to be fully explored.
Disease outcomes reflect the battle between pathogens and hosts. In addition to viral intrinsic factors, host-specific traits and differences in host immune responses can ameliorate or exacerbate both the infection and prognosis. Likewise, host responses contribute to the pathogenesis of H5N1 viruses in mammals. Many studies have demonstrated that dysregulation of the innate immune responses during H5N1 influenza virus infection increase disease severity (27)(28)(29)(30).
Previously, we characterized the genetic and biological diversity of HPAI H5N1 viruses in Vietnam (31), and found that two viruses, A/chicken/Vietnam-Bac Lieu/1214/ 2007 (CK/1214) and A/chicken/Vietnam-Ca Mau/1180/2006 (CK/1180), shared similar genetic backgrounds, belonging to the same genotype, but displayed substantially different levels of virulence in mice (31,32). In the present study, we used reverse genetics to identify the molecular determinants for the different virulence of these two viruses in mice and explored the possible underlying mechanisms.

RESULTS
Rescued H5N1 viruses retain the biological properties of wild-type viruses. In our previous studies (31,32), we characterized the pathogenicity and viral replication of CK/1180 and CK/1214 in mice and found that CK/1180 systemically replicated and showed high pathogenicity in mice, with a 50% mouse lethal dose (MLD 50 ) of 2.5 log 10 50% egg infectious doses (EID 50 ), but CK/1214 replicated only in lung tissue and displayed low virulence in mice. In this study, we found that even 10 7 EID 50 of CK/1214 did not kill mice (MLD 50 Ͼ 7.5 log 10 EID 50 ) (Fig. 1). By using constructed plasmids, we rescued the CK/1180 and CK/1214 viruses, designated R-CK/1180 and R-CK/1214, respectively. After confirming the sequences of the rescued viruses, we tested their replication and lethality in mice. Similar to the wild-type virus, the rescued R-CK/1180 replicated systemically ( Fig. 1A and Table 1) and was highly pathogenic in mice ( Fig. 1C and E). R-CK/1214 virus, like its wild-type counterpart, did not kill mice and replicated only in the lungs of the organs tested (Fig. 1B, D, and F). These results indicate that the rescued viruses maintained the biological properties of the wild-type viruses. These two viruses share the same NP gene product at the amino acid level and differ by only 24 amino acids in their PB2, PB1, PA, PA-X, HA, NA, M1, M2, NS1, and NS2 proteins ( Table 2).
The HA gene determines the high virulence of the CK/1180 virus in mice. To avoid any "gain of function" concerns (33), we only used the lethal virus CK/1180 as the backbone to generate the reassortants, each containing one gene derived from CK/ 1214, as described previously (10), and tested their replication and pathogenicity in mice. The virulence of the reassortants that contained the PB2, PB1, PA, NA, M, or NS gene of CK/1214 was attenuated 5 to 50 times (MLD 50 , 3.5 to 4.5 log 10 EID 50 ) relative to that of the rescued CK/1180 virus (MLD 50 , 2.8 log 10 EID 50 ) ( Fig. 1C and Table 1). Viruses were detected in all four tested organs of the mice infected with the reassortants that contained the PB2, PB1, PA, NA, or M gene of CK/1214 and in the lungs, spleens, and kidneys (but not brains) of the mice infected with the reassortant con-taining the NS gene of CK/1214 ( Table 1). The reassortant bearing the HA gene of CK/1214 in the CK/1180 background mainly replicated in the lungs (a low level of virus was detected in the spleens of two of three infected mice) and was dramatically attenuated (3,981-fold), with an MLD 50 of 6.4 log 10 EID 50 compared with that for the R-CK/1180 virus (Table 1). These results indicate that the HA gene plays an important role in the high virulence of the CK/1180 virus in mice.  The amino acids at positions 158 to 160 in the HA protein of CK/1180 form a glycosylation site. The predicted HA protein sequences of the CK/1180 and CK/1214 viruses differ by a single amino acid at position 158 (Table 2) (H3 numbering). Asparagine (N) at this position of the HA protein of CK/1180 virus forms a potential glycosylation site-NST-at amino acids 158 to 160, whereas there is no such site at the corresponding amino acid positions in the HA protein of the CK/1214 virus. To investigate whether N-linked glycosylation indeed occurs at this potential site, we performed Western blot analysis of HA polypeptides of CK/1180 and CK/1214 treated or not with peptide-N-glycosidase F (PNGase F) enzyme. We found that the HA1 polypeptide of CK/1180 exhibited decreased mobility relative to that of CK/1214, whereas when the HA1 polypeptides of the two viruses were deglycosylated with PNGase F, they showed similar mobilities (Fig. 2). These results confirm that the potential N-linked glycosylation site at amino acid positions 158 to 160 in the HA protein of the CK/1180 virus is glycosylated. Analysis of viral replication in mice showed that CK/1180 virus with this glycosylation in its HA protein systemically replicated in mice, and the loss of this glycosylation (CK/1180-1214HA virus) resulted in mainly viral replication in the lung (Table 1). These findings suggest that glycosylation at positions 158 to 160 in the HA protein promotes H5N1 virus systemic spread in mice and thereby contributes to its higher virulence in mice.
Glycosylation at positions 158 to 160 of HA increase VLP formation and virus yield. Virus-like particles (VLPs) of influenza viruses are extensively used to investigate viral assembly and budding efficiency (34)(35)(36). To examine the effect of the glycosylation at positions 158 to 160 of HA on VLP formation, we produced two different VLPs (designated V-1214HA and V-1180HA) and compared their respective productivities. Of note, the only difference between the two VLP particles was the glycosylation at positions 158 to 160 of the HA proteins. We found that V-1180HA produced a higher level of the hemagglutinin unit in the cell supernatants than did V-1214HA (Fig. 3A); also, more HA and M1 proteins were detected from V-1180HA in the cell supernatants ( Fig. 3B) than from V-1214HA, indicating that the productivity of V-1180HA was considerably higher than that of V-1214HA.
Next, we compared the levels of replication of CK/1180, CK/1214, and CK/1180-1214HA in MDCK cells. We found that CK/1180 replicated more efficiently than CK/1214 and CK/1180-1214HA at the desired times at a multiplicity of infection (MOI) of 0.001 (Fig. 4A), even though there was no notable difference in viral growth at an MOI of 5 (Fig. 4B). Analysis of viral proteins, including HA, NP, and M1, in the cell lysates showed that HA expression was much higher in MDCK cells infected with CK/1180 than in those infected with CK/1180-1214HA at 12 h postinfection (hpi) at an MOI of 5, but there was no difference in NP and M1 expression ( Fig. 5A and B). Flow cytometric analysis of HA expression on the membrane of infected MDCK cells showed that there was no difference among CK/1180-, CK/1214-, and CK/1180-1214HA-infected MDCK cells (Fig.  5C). We checked the viral plaque phenotypes of CK/1180 and CK/1180-1214HA in MDCK cells by using an immunofluorescence assay and found that CK/1180 formed much bigger rocket-like plaques in MDCK cells than CK/1180-1214HA and displayed stronger diffusivity during replication in cells (Fig. 4C). The plaque diameters of CK/1180 were significantly larger than those of CK/1214 and CK/1180-1214HA as determined by  using the standard plaque assay (37) (Fig. 4D and 6A). We then examined the MDCK cells infected with CK/1180, CK/1214, and CK/1180-1214HA at an MOI of 5 at 10 hpi by using ultrathin-section electron microscopy as described previously (38), and we found no difference in the morphology of the virions (Fig. 6B). Together, these results demonstrate that glycosylation at positions 158 to 160 of HA increases the plaque size and replication capacity of H5N1 virus in mammalian cells.
Glycosylation at positions 158 to 160 of HA enhances the immune and inflammatory processes in infected mouse lung. To assess the effect of glycosylation of HA on the host response, we performed gene expression profiling in the lungs of CK/1180-, CK/1180-1214HA-, and phosphate-buffered saline (PBS)-inoculated mice by means of a microarray analysis. A total of 3,347 genes were observed to be significantly differentially expressed (P Ͻ 0.05; fold change Ͼ 2) during CK/1180 infection; the corresponding number was 2,936 for CK/1180-1214HA (Fig. 7A). A total of 2,348 differentially expressed (DE) genes were common to the two viruses (Fig. 7B). Canonical pathway analysis demonstrated that CK/1180 preferred to regulate higher immune response pathways than did CK/1180-1214HA, which included innate immune and adaptive immune responses (Fig. 7C). Thirty-five DE genes whose regulation was diametrically opposed between the CK/1180 and CK/1180-1214HA infections were identified (Fig. 8A); these included some mucin-related genes. For example, Clca3, a putative calcium-activated chloride channel involved in the regulation of mucus production and/or secretion (39,40), and Muc5ac were upregulated after CK/1180 infection but downregulated after CK/1180-1214HA infection (Fig. 8A). By using quantitative PCR, we validated the expression of Clca3, Muc5ac, and Muc5b; Fig. 8B shows their fold changes in expression compared with control levels. These expression patterns are consistent with the results of the microarray analysis.

DISCUSSION
By using two HPAI H5N1 viruses (CK/1180 and CK/1214) with similar genomes but different pathogenicities in mice (31,32), in this study, we demonstrated that a single amino acid mutation at position 158 of the HA protein had fundamental effect on the systemic replication and pathogenicity of these H5N1 influenza viruses in mice. We further found that the substitution G158N introduced an N-linked glycosylation at positions 158 to 160 of the HA protein and that this N-linked glycosylation enhanced viral productivity in infected mammalian cells and exacerbated host immune and inflammatory responses to viral infection.
The pathogenicity of influenza viruses in mammals is complex and determined by multiple viral genes (9-16, 21, 41). The HA and NS genes of human H5N1 influenza A virus contribute to its high virulence in ferrets (42), while mutations in the PA, NP, and HA genes of pandemic (H1N1) 2009 influenza virus contribute to its adaptation to mice (43). Moreover, the PB2 and HA genes affect host range and pathogenicity in mouseadapted influenza A virus (44). In the present study, we found that a single mutation at position 158 of HA has a major effect on the pathogenicity of H5N1 viruses in mice, with the other genes contributing only slightly to the virulence and to different extents (Table 1).
Protein glycosylation is gradually being recognized as an important means of evolution for influenza viruses (45,46). Sun et al. stated that the number of glycosylation sites in HA protein increases during the evolution of human seasonal influenza viruses (47). The acquisition of potential glycosylation sites is an effective way for influenza viruses to escape positive selective pressures from the host (45,48,49). In the present study, we found that CK/1180, with an additional glycosylation site at positions 158 to 160 of HA, showed higher pathogenicity in mice, consistent with a previous report (50). Natural cytotoxicity receptors (NCRs), such as NKp46 (NCR1), NKp44 (NCR2), and NKp30 (NCR3), are important natural killer (NK) cell triggering receptors (51). NKp46 and NKp44 have been shown to functionally interact with the HA proteins of different influenza viruses during NK cell-mediated virus clearance (52,53). We therefore spec- The glycosylation at positions 158 to 160 in the HA protein has been documented to affect the biological properties of influenza virus in several aspects. The loss of this glycosylation site contributes to higher viral replication in the upper respiratory tract of ferrets (6), bestows upon the HA protein the ability to bind to ␣-2,6 glycans, and facilitates the direct contact transmission of H5N1 virus in guinea pigs (54) and the airborne transmission of influenza virus in ferrets (6,7). In our present study, although the loss of the glycosylation at positions 158 -to 160 in the HA significantly attenuated our H5N1 virus in mice, it did not alter the receptor binding preference (Fig. 9) or the viral replication in mouse nasal tissues of these viruses (Table 1, CK/1180 and CK/1180-1214HA). Moreover, we found that deletion of this glycosylation site just slightly reduced the virulence in mice of another highly pathogenic H5N1 virus A/duck/Hunan/ 49/05 (26) and decreased its systemic replication (Table 1). We therefore speculate that the role of glycosylation of the HA protein in viral pathogenicity to mammals and in receptor binding specificity may differ depending on the individual virus subtype or strain.
Host-specific traits and differences in host immune responses contribute to both infection and clinical outcomes. An exaggerated innate response, with early recruitment of inflammatory leukocytes to the lung, contributed to the morbidity associated with infection with the 1918 influenza virus (55). Recently, a significant association was reported between excessive early cytokine responses to avian H5N1 infection, immune cell recruitment, and poor outcome (30). In the present study, we found that CK/1180 was highly pathogenic in mice but that the loss of HA glycosylation at positions 158 to 160 significantly attenuated this pathogenicity. Gene expression profiling demonstrated that CK/1180 induced a stronger host response in the lungs of mice than that induced by CK/1180-1214HA, including immune and inflammatory responses. Mucin-related proteins modulate the rheological properties of airways and participate in lung defense (56). However, hypersecretion of airway mucus can increase the mortality associated with disease (57,58). Influenza virus infections may induce profound disturbances of the mucociliary system (59). In the present study, CK/1180 infection induced higher expression of mucin-related proteins in mouse lungs than CK/1180-1214HA, which may explain, in part, the increased lethality of CK/1180 to mice. These findings further our understanding of the determinants of pathogenicity of H5N1 viruses in mammals.

Cells and viruses.
Human embryonic kidney cells (293T) were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum plus antibiotics, and Madin-Darby canine kidney (MDCK) cells were maintained in minimal essential medium (MEM) supplemented with 5% newborn calf serum. All cells were incubated at 37°C in 5% CO 2 . Two HPAI H5N1 viruses, CK/1180 and CK/1214, were propagated in 10-day-old specific-pathogen-free (SPF) embryonated chicken eggs and stored at Ϫ80°C until use.
Construction of plasmids. The construction of plasmids for virus rescue was performed as described previously (12). The protein expression plasmids for HA, NA, and M1 of influenza virus were generated by inserting their cDNAs between the SacI and NheI restriction enzyme sites of the pCAGGS plasmid vector for the production of virus-like particles (VLPs). The primer sequences are available upon request. All of the constructs were completely sequenced to ensure the absence of unwanted mutations.
Generation of reverse genetic reassortant viruses. Reassortant viruses were generated by using an eight-plasmid reverse genetics system as described previously (12,24). The rescued viruses were detected by using a hemagglutination assay and were fully sequenced to ensure the absence of unwanted mutations.
Mouse experiments. Groups of eight 6-week-old female BALB/c mice (Beijing Experimental Animal Center) were lightly anesthetized with CO 2 and inoculated intranasally with 10 6 50% egg infectious doses (EID 50 ) of H5N1 influenza viruses in a 50-l volume. Three mice in each group were euthanized on day 3 postinfection. Organs, including the lungs, kidneys, spleens, and brains, were collected and titrated for virus infectivity in eggs as described previously (12). The remaining five mice were monitored daily for 14 days for weight loss and mortality. To determine the MLD 50 of viruses that caused lethal infection of mice, six groups of five mice were inoculated intranasally with 10-fold serial dilutions containing 10 1 to 10 6 EID 50 or 10 7 EID 50 of virus in a 50-l volume. The MLD 50 was calculated by using the method of Reed and Muench (60).
Deglycosylation using PNGase F and Western blot analysis. Deglycosylation was performed by using PNGase F enzyme (New England BioLabs). Virus was concentrated by ultracentrifugation and purified by sucrose gradient centrifugation as previously described (54). The virus concentrates were denatured and then deglycosylated with the PNGase F enzyme according to the manufacturer's instructions. Virus samples were analyzed by use of SDS-polyacrylamide gel electrophoresis (SDS-PAGE) and Western blotted with chicken antisera induced by the pCAGGS-HA DNA vaccine (61) and IRDyeTM700DX-conjugated goat anti-chicken secondary antibody (Rockland). Protein bands were detected by using an Odyssey infrared imaging system (Li-Cor Biosciences, Lincoln, NE). Viral replication in MDCK cells. MDCK cells were infected with viruses at a multiplicity of infection (MOI) of 5 or 0.001. Virus-containing culture supernatants were collected at the indicated time points and titrated in MDCK cells. The infected cells were collected, lysed in SDS loading buffer, and further analyzed by Western blotting. To measure HA expression on the membrane of the infected cells, the cells were infected with viruses at an MOI of 5 and then were digested with trypsin to obtain a single cell suspension at 12 hpi. The samples were stained with a rabbit monoclonal antibody to influenza A virus H5N1 HA protein (11048-RM07; Sino Biological Inc.) and with Alexa Fluor 488-conjugated goat anti-rabbit IgG (HϩL) secondary antibody (A-11034; Thermo Fisher Scientific). Data were acquired on a FACSAria II (BD Biosciences), and the mean fluorescence intensity (MFI) was analyzed with FlowJo ϫ 10.0.7r2 (Tree Star, San Carlos, CA).
Immunofluorescence. MDCK cells were grown in 6-well plates and infected with virus at an MOI of 0.001. At the desired time points postinfection, the cells were fixed with PBS containing 4% paraformaldehyde. After being blocked with 5% bovine serum albumin (BSA) in PBS, the cells were incubated at room temperature for 2 h with chicken antisera induced by the pCAGGS-HA DNA vaccine. The cells were then washed three times with PBS plus Tween 20 (PBS-T) and incubated for 1 h with fluorescein isothiocyanate (FITC)-coupled goat anti-chicken secondary antibody. After that, the cells were washed three times with PBS-T and were observed with a fluorescence microscope (Carl Zeiss, Germany). VLP formation. Influenza VLPs were produced in 293T cells by coexpression of influenza virus HA, NA, and M1 proteins using a plasmid-based VLP system as described previously (62). Briefly, 293T cells were transfected with pCAGGS plasmids expressing the viral proteins HA, NA, and M1 by using Lipofectamine and Plus reagents (Invitrogen) according to the manufacturer's instructions. As described previously (34), the released VLPs in the cell supernatant were harvested by using ultracentrifugation at 48 h posttransfection. The pellet (after ultracentrifugation) and cells were lysed in SDS sample buffer solution (Wako) with 100 mM dithiothreitol and subjected to Western blotting.
Microarray analysis. For all gene expression analyses, groups of three mice were inoculated with PBS or 10 6 EID 50 of CK/1180 or CK/1180-1214HA. Total RNA was extracted from lung tissues on day 3 postinfection by using a Qiagen RNeasy kit. The microarray assay was performed by using a low-RNAinput linear amplification kit (Agilent Technologies, Santa Clara, CA) and Agilent's whole mouse genome microarray kit, 4 ϫ 44 K (G4122F), as described elsewhere (25). Student's t test and significant analysis of microarray (SAM) were performed to determine the genes that had significantly different expression levels (P Ͻ 0.05 and Ͼ2-fold change) upon infection compared with levels in the PBS-inoculated group. Ingenuity Pathways Analysis (IPA) (Ingenuity Systems, Redwood City, CA) was applied for functional and network analyses of significantly differentially expressed genes.
Validation of array expression data by qRT-PCR. Quantitative real-time PCR (qRT-PCR) was performed to validate the changes in expression of selected genes detected by microarray as described previously (32). Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used for normalization. Primer sequences and protocols for the qRT-PCR are available upon request. Tests were performed in duplicate in at least three separate experiments, and relative RNA levels were determined by using the control group as the reference sample.
Laboratory facility. Studies with HPAI H5N1 viruses were conducted in a biosafety level 3 laboratory approved for such use by the Committee on the Ethics of Animal Experiments of the Harbin Veterinary Research Institute, Chinese Academy of Agricultural Sciences (approval number BRDW-XBS-09). All studies were carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals (63) of the Ministry of Science and Technology of the People's Republic of China.
Accession number(s). All primary expression microarray data have been deposited in the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/projects/geo/) under the accession number GSE89246.