H9N2 Influenza Virus Infections in Human Cells Require a Balance between Neuraminidase Sialidase Activity and Hemagglutinin Receptor Affinity

H9N2 avian influenza (AI) virus, one of the most prevalent AI viruses, has caused repeated poultry and human infections, posing a huge public health risk. The H9N2 virus has diversified into multiple lineages, with the G1 lineage being the most prevalent worldwide. In this study, we isolated G1 variants carrying an 8-amino-acid deletion in their NA stalk, which is, to our knowledge, the longest deletion found in H9N2 viruses in the field. The NA stalk length was found to modulate G1 virus entry into host cells, with the effects being species specific and dependent on the corresponding HA binding affinity. Our results suggest that, in nature, H9N2 G1 viruses balance their HA and NA functions by the NA stalk length, leading to the possible association of host range and virulence in poultry and mammals during the evolution of G1 lineage viruses.

T he currently circulating H9N2 subtype of avian influenza (AI) viruses was first detected in South China in 1994 (1). H9N2 viruses then spread throughout Eurasia and Africa and are now among the most prevalent AI viruses. H9N2 infections cause a decline in egg production with substantial lethality in poultry (2) and also cause a mild respiratory disease in humans (2,3). Coinfections of H9N2 viruses and other subtype AI viruses frequently generate novel enzootic virus reassortants, e.g., H5N1 A/goose/ Guangdong/1/96 (Gs/GD) lineage, H7N9, and H5N8 viruses (1,(4)(5)(6). Therefore, it is of great importance to closely monitor possible H9N2 virus genetic and phenotypic changes to reduce the risk of a future pandemic.
The H9N2 viruses that are most prevalent in Eurasia and North Africa are in the G1 and Y280 lineages. G1 lineage viruses have circulated in South China and spread west to Central Asia, the Middle East, and North Africa, where the lineage diversified into four sublineages (sublineages A to D), with sublineages A and B (designated G1-A and G1-B, respectively) being the most prevalent in these areas (7). Y280 lineage viruses have been prevalent throughout China.
In the Middle East, G1-A/B reassortant viruses emerged in Israel and Jordan from 2006 to 2009, with the viral hemagglutinin (HA), neuraminidase (NA), M, and NS genes originating from G1-B and the other viral genes originating from G1-A (7,8). The G1-A/B virus was first identified in Egypt in 2010 and became endemic in a wide range of poultry (9)(10)(11), with four human H9N2 infection cases being reported (2). The G1-A/B virus carried the human adaptation PB2-E627V and -K526R mutations (8), implying that Egypt was a hot spot of H9N2 virus evolution in the Middle East.
Influenza A viruses have two major glycoproteins, hemagglutinin (HA) and neuraminidase (NA), on the virion surface. HA binds to sialylglycan on the surface of a target cell, and NA cleaves the terminal sialic acids (Sia) of sialylglycans to mediate the release of nascent progeny virions from infected cells (12,13). It has recently been reported that NA also enables virus movement on the cell surface toward endocytic sites, resulting in virus internalization for infection (14,15). Since both proteins target the same receptor, the functional activities of HA and NA need to be balanced for both virus entry into and virus release from host cells (16).
NA is a tetrameric, type II transmembrane protein. The enzyme active site is located in a globular head that is connected to an endodomain via a narrow stalk. The NA stalk length and amino acid sequence vary considerably, even in viruses of the same subtype (24), although there are some common structural features in the stalk, including at least one cysteine residue and a potential glycosylation site. NA stalk deletions have been occasionally detected in AI viruses (e.g., H5N1, H6N1, H2N2, H9N2, H7N1, and H7N3 viruses) and have been associated with strains during early host adaptation from aquatic birds to terrestrial poultry (25)(26)(27)(28)(29)(30)(31). The NA stalk length has a significant effect on virus replication, pathogenesis, and even host range for some of these AI viruses.
Experimental passage of a duck H9N2 strain in quail and chickens selected an adapted strain with a 21-amino-acid (aa) deletion in the NA stalk that improved viral replication in two bird species and mice (32). Also, previous studies showed that early isolates of G1 and Y280 lineage viruses in China contained 2-or 3-amino-acid deletions, respectively (33): the 3-amino-acid deletion enhanced the virulence of Y280 lineage virus in chickens and mice (30). However, thus far, the effect of NA stalk length in the G1 lineage viruses now prevalent in Eurasia and Africa on biological changes has not been elucidated.
In this study, we first isolated three H9N2 G1 strains carrying an 8-amino-acid deletion in the NA stalk. This was, to our knowledge, the longest deletion detected naturally in an H9N2 virus. We then investigated the effect of different NA stalk deletions, including the 8-amino-acid deletion, on the G1 virus phenotype. Our results were the most representative of those of the three H9N2 strains isolated in this study with a deletion in their NA stalk. Using a reverse-genetics system, we rescued a recombinant G1-A/B virus (designated the NA wild type [NA-wt]) with HA and NA genes from the CL80 strain and other genes from A/chicken/CL42/2013 (CL42), which had internal gene sequences representative of the G1-A/B lineage strains in Egypt (8,18). The CL80 NA plasmid was also modified by inserting the 8 amino acids (58-IERNITEI-65) missing in NA-wt (designated NAϩ8aa/gly) ( Fig. 2A). Glycosylation (gly) at residue 61 in NA-wt was confirmed by the mobility shift observed by Western blotting (Fig. 2B). To investigate the effect of glycosylation, a CL80 strain carrying the 8-amino-acid insertion, but with an N61S substitution in the glycosylation site (58-IERSITEI-65, designated NAϩ8aa), was generated. The N61S substitution was selected based on the sequence of A/quail/Egypt/14864V/2014, which contains no glycosylation site. To differentiate between the effect of glycosylation from the effect of the 3-amino-acid insertion with the glycosylation motif (residues 61 to 63), a CL80 strain with a 5-amino-acid insertion without the glycosylation motif (IER---EI, designated NAϩ5aa) was generated. A previous study reported that an old H9N2 laboratory strain had lost 21 amino acids in its NA stalk (residues 58 to 83) during experimental passages (32). To systematically analyze the effect of stalk length on G1-A/B replication, two strains containing NA stalk deletions were generated from CL80. One strain had a 3-amino-acid deletion (residues 66 to 68, designated NAϪ3aa), and the other strain had a 13-amino-acid deletion (residues 66 to 78, designated NAϪ13aa). The deletion in NAϪ13aa corresponded to a 21-amino-acid deletion (residues 58 to 83) reported in an experimental passage study (32). All these NAs had equivalent expression levels (Fig. 2C). Also, the result that all these recombinant viruses were successfully rescued indicated that up to a 21-aminoacid deletion in the NA stalk was not deleterious to G1-A/B replication.
Enzymatic activity of NA stalk length mutants. The NA activity of NA-wt and the stalk length mutants was determined by measuring the cleavage of the small soluble NA-XTD substrate. Virus samples were standardized to an equivalent titer based on the numbers of focus-forming units (FFU) and on the numbers of hemagglutination units (HAU) and used for the cleavage assays. With both standardized virus titers, NA The 8-amino-acid deletion (58-IERNITEI-63) in the NA stalk compared to the consensus sequences of the G1 lineage strains in Egypt, including CL42 (consensus H9N2), was found in three G1-A/B strains. NA stalk length differences are shown relative to that of NA-wt. The blue area in the NA boxes indicates an amino acid insertion. A blue underline in the stalk sequences marks residues 61 to 63, which are a potential NA glycosylation site in the consensus H9N2 and NAϩ8aa/gly strains. A dotted line in an NA box and a black dashed line with an underline in a stalk sequence mark an amino acid deletion. (B) Western blot analysis of the NA mutants with a mobility shift indicating glycosylation in the NAϩ8aa/gly mutant. A representative Western blot is shown. (C) Quantification of NA expression in 293T cells. The cells were transfected with plasmids expressing the indicated NAs, and cell lysates were analyzed by Western blotting using anti-NA antibody. Each data bar shows the mean Ϯ SD from four independent experiments. nt, nucleotide; NS, no statistically significant difference. enzymatic activity varied with the stalk length (Fig. 3A). Relative to the NA activity of NA-wt, NA with the longest NA stalk (NAϩ8aa) had 9-fold higher activity and NA with the shortest stalk (NAϪ13aa) had 2-fold lower activity. However, glycosylation at residue 61 significantly reduced the NA activity in NAϩ8aa/gly, thereby obviating the increase in NA activity produced by the 8-amino-acid insertion in NAϩ8aa (Fig. 3A, NAϩ8aa/gly versus NAϩ8aa).
In addition, the kinetic parameters of the enzymatic activity of the NA stalk length mutants were measured using the fluorescent small 2=-(4-methylumbelliferyl)-␣-D-Nacetylneuraminic acid (MUNANA) substrate. With NA titers standardized by both the numbers of FFU and the numbers of HAU, NA enzymatic activity varied with stalk length ( Fig. 3B and C). NAs with longer stalks had greater enzymatic activity, with the maximum velocity of substrate cleavage (V max ) for NAϩ8aa being about 4-fold higher than that for NA-wt (Table 1), and the V max for NAϪ13aa was the lowest and slightly less than that for NA-wt. Glycosylation at residue 61 significantly reduced both the enzymatic activity and V max . All the NA stalk length mutants had relatively similar Michaelis-Menten constant (K m ) values, reflecting their similar substrate binding affinities. Therefore, the relative enzyme efficacies of the NA stalk length mutants, based on the ratios of their V max and K m values, were similar to their relative V max values.
Elution of NA stalk length mutants from chicken and turkey erythrocytes. To examine the ability of NA-wt and the NA stalk length mutants to cleave Sia from substrates more complex than MUNANA, we carried out elution assays with viruses bound to chicken and turkey erythrocytes. Chicken erythrocytes express significantly more ␣2,3 Sia than ␣2,6 Sia, whereas turkey erythrocytes express significantly more ␣2,6 Sia than ␣2,3 Sia (34). Therefore, erythrocytes from both avian species were used to assay NA substrate specificity. The elution efficacy from these two erythrocyte species was in agreement with our results on the effect of NA stalk length on NA enzymatic cleavage of the MUNANA substrate ( Fig. 4A and B); i.e., NAϩ8aa, with the longest stalk, eluted the most quickly, whereas NAϪ13aa, with the shortest stalk, eluted the most slowly. Glycosylation of residue 61 largely blocked the increased elution produced by the 8-amino-acid insertion ( Fig. 4A and B, NAϩ8aa/gly versus NAϩ8aa). All the viruses eluted more slowly from turkey erythrocytes, which mostly expressed ␣2,6 Sia, than from chicken erythrocytes, which mostly expressed ␣2,3 Sia. This difference was greater when turkey erythrocytes were treated with ␣2,3 sialidase, so that only ␣2, 6 Sia was on the cell surface (Fig. 4C). This result was in agreement with the results of previous studies showing that AI virus NAs have a cleavage preference for ␣2,3 Sia (35). The time to elute half of the bound viruses was calculated by curve fitting the nonlinear elution data and expressing the data relative to those for NA-wt. The relative elution times were almost identical for all three types of erythrocytes (Fig. 4D), implying that NA stalk length did not affect the substrate specificity for Sia cleavage. Collectively, these results suggest that NA stalk length is correlated with enzymatic activity and elution efficacy from erythrocytes but not with substrate specificity.
Replication of NA stalk length mutants in avian cells and eggs. The effect of stalk length on the replication of NA stalk length mutants in avian cells and eggs was investigated. Chicken-derived DF-1 cells and embryonated eggs were infected with NA-wt and the NA stalk mutants, and their replication kinetics at 37°C were measured. Virus replication in DF-1 cells and eggs was closely correlated with NA stalk length: the longer that the stalk was, the better that the replication was ( Fig. 5A and C). The virus titers early postinfection were NAϩ8aa Ͼ NAϩ5aa Ͼ NA-wt Ͼ NAϪ3aa Ͼ NAϪ13aa. At 48 h postinfection (hpi), NAϩ8aa, with the longest NA stalk length, produced 24.4and 5.7-fold higher progeny virus titers in DF-1 cells and eggs, respectively, than NA-wt ( Fig. 5B and D). In contrast, NAϪ13aa, with the shortest NA stalk length, produced less than half the progeny virus yield of NA-wt. Glycosylation at residue 61 significantly decreased the increase in G1-A/B replication due to the 8-amino-acid insertion ( Fig. 5B and D, NAϩ8aa/gly versus NAϩ8aa).

Replication of NA stalk length mutants in a human cell line and primary cells.
To investigate the relationship between NA stalk length and G1-A/B mutant replication in human airway cells, human airway-derived Calu-3 cells and primary bronchial epithelial cells were infected and viral replication kinetics at 33°C were measured for 96 h. In contrast to the growth kinetics in avian cells, the replication of NA stalk length mutants in human cells was negatively correlated with the NA stalk length ( Fig. 6A and C).   6B and D). In contrast, NAϩ8aa had the lowest progeny virus yields in Calu-3 cells and primary cells, 3.0-and 2.3-fold lower than those for NA-wt, respectively. Glycosylation at residue 61 blocked the increase in G1-A/B replication in human cells due to the 8-amino-acid insertion ( Fig. 6B and D, NAϩ8aa/gly versus NAϩ8aa). Taken together, these results indicate that NA stalk deletions lead to an increase in virus replication in human airway cells but a decrease in avian cells infected with the G1-A/B mutants. These results also indicate that the NA stalk length has an appreciable effect on host adaptation of the G1-A/B virus in vitro.
Pathogenicity of NA stalk length mutants in mice in vivo. To assess the effects of changes in NA stalk length on in vivo infections, BALB/c mice were inoculated intranasally with 10 3 to 10 7 FFU of either NA-wt or one of three representative stalk length mutants, i.e., NAϩ8aa/gly, NAϩ8aa, or NAϪ13aa. The body weight and survival of each infected mouse were monitored daily for 14 days.
With an inoculation dose of 10 5 FFU, the NAϩ8aa virus, with the longest stalk length, caused a small weight loss in all 6 infected mice, and all of the infected mice survived (Fig. 7A and B). In contrast, the same dose of NA-wt and NAϪ13aa viruses , and turkey erythrocytes that had been treated with ␣2,3 neuraminidase (C). The samples were then transferred to 33°C, and their HA titers were assayed as a function of time. The data are expressed relative to the percentage of the initial HA titer at time zero at 4°C. Each data point is the mean Ϯ SD from three independent experiments. The Sia expressed on each type of erythrocyte is shown in parentheses at the left axis. (D) The time when half of each virus eluted from each type of erythrocyte was calculated by fitting the HA elution data with the Boltzmann equation. The data are expressed relative to the results for NA-wt. Each data point is the mean Ϯ SD from three independent experiments. NS, no statistically significant difference. A schematic of the NA stalk in each strain, as shown in Fig. 2A, is shown at the left. caused severe weight loss; 4 of the 6 mice infected with NA-wt survived, and 2 of the 6 mice infected with NAϪ13aa survived. With a dose of 10 6 FFU, there was significant weight loss in all the infected mice; 3 of the 6 mice infected with NAϩ8aa survived, but none of the mice infected with NA-wt or NAϪ13aa survived, with death being faster for NAϪ13aa-infected mice than for NA-wt-infected mice. NAϪ13aa was the most pathogenic in mice, with its 50% mouse lethal dose (MLD 50 ) being 3.2-fold lower than that of NA-wt, whereas NAϩ8aa was the least pathogenic, with its MLD 50 being 5.6-fold higher than that of NA-wt. Glycosylation at residue 61 increased the pathogenicity effect of the 8-amino-acid stalk insertion, resulting in rapid weight loss and the reduced survival of infected mice (NAϩ8aa versus NA-wt).
To determine the virus titers in infected mice, groups of five mice were inoculated intranasally with 10 5 FFU of each virus and their lungs were collected at 4 days postinfection (dpi). The virus titers in the infected mice were in agreement with their clinical signs and with the in vitro replication ability of the virus in human airway cells. The progeny virus yield in the lungs of mice infected with NAϪ13aa was considerably higher than that in the lungs of mice infected with NA-wt (Fig. 7C). In contrast, NAϩ8aa Pathology studies were carried out on the lungs of infected mice collected at 4 dpi. NAϩ8aa produced only a limited inflammatory response in the lungs of infected mice (Fig. 7D). In contrast, NAϪ13aa induced more severe pathological changes in the lungs than NAϩ8aa did, with bronchiolar necrosis and alveolitis characterized by hemorrhage and infiltration of inflammatory cells. NA-wt and NAϩ8aa/gly caused moderate pneumonia in infected mice, with slightly less severe bronchiolar hemorrhage and inflammatory cell infiltrates being seen in NAϩ8aa/gly-infected mice than in NA-wt-infected mice. Together, these results indicate that the NA stalk deletion enables increased viral replication in human airway cells in vitro and contributes to increased replication in infected mice in vivo, which suggest that the NA stalk length has appreciable effects on the mammalian adaptation of G1-A/B NA mutants.
Effect of the balance between HA and NA functions on virus infection of human cells. The balance between HA and NA functions, which has been reported to be critical for influenza virus replication and host range (14,15), was investigated using the NA mutants in this study. We first measured HA binding affinity to the ␣2,3 Sia and ␣2,6 Sia analogues Neu5Ac␣2,3Gal␤1-4GlcNAc (␣2,3 SLN) and Neu5Ac␣2,6Gal␤1-4GlcNAc (␣2,6 SLN), respectively, and to ␣2,3 SLN sulfated on its antepenultimate sugar [(Neu5Ac␣2,3Gal␤1-4(6-HSO 3 )GlcNAc (Su-␣2,3 SLN)], which has been reported to be recognized by H9N2 G1 viruses (21,36). In this study, we used three H9N2 G1 strains that were isolated in the Middle East (i.e., CL80, CL42 and A/chicken/Emirates/R66/02 [designated Em/R66]), each of which had an HA with a distinct Sia binding pattern (described below). Survival was calculated for all mice, including mice that were humanely sacrificed if they lost more than 30% of their body weight within a few days. A schematic of the NA stalk of each strain (as shown in Fig. 2A), the MLD 50 of that strain, and its value relative to that for NA-wt (in parentheses) are shown at the right. CL80 HA, a representative G1-A/B mutant with a shortened NA, had strong binding to Su-␣2,3 SLN, moderate binding to ␣2,6 SLN, and low binding to ␣2,3 SLN (Fig. 8A). However, relative to CL80 HA, CL42 HA had significantly increased binding to ␣2,6 SLN and slightly increased binding to Su-␣2,3 SLN (Fig. 8B). The CL42 HA binding pattern was similar to that of contemporary G1 viruses (20). The ectodomain sequence of the CL80 HA (without the signal peptide) differed from that of the CL42 HA, with substitutions at eight HA amino acid residues. Among these differences, only the P229S mutation was located in the receptor binding domain, implying its putative association with the Sia binding difference between the two viruses. To confirm the effect of the P229S mutation in enabling increased ␣2,6 Sia binding of CL42 HA, viruses carrying CL80 HAs with a single P229S mutation were rescued and the effects of the mutation on Sia binding were measured. As controls, we arbitrarily selected three of the remaining seven mutations that were located in the HA stalk domain (i.e., D384G, S388A, and D428N) and rescued viruses with these three mutations. The D384G, S388A, and D428N reverse mutations had little effect on Sia binding by CL80 HA (Fig. 8A and  B). However, the P229S reverse mutation considerably increased ␣2,6 SLN binding and slightly increased Su-␣2,3 SLN binding, resulting in a binding pattern resembling that of CL42 HA. This indicated that the S229P mutation appreciably reduced CL80 HA binding to ␣2,6 SLN. However, Em/R66 HA showed an atypical Sia binding pattern, with strong binding to both ␣2,3 SLN and Su-␣2,3 SLN and low binding to ␣2,6 SLN (Fig. 8C), as previously reported (21). These results suggest that G1-A/B variants with a shortened NA have a fairly restricted ␣2,6 Sia binding affinity but strong binding to ␣2,3 Sia (mostly sulfated ␣2,3 Sia).
Our data indicate that the G1-A/B variants had HAs with a poor ␣2,6 Sia affinity and, therefore, needed NAs with reduced sialidase activity for more efficient growth in human cells (Fig. 3, 6, and 8). These results suggest that reduced NA activity, due to the stalk length deletion, compensated for the limited HA binding affinity to improve virus growth in human cells. Recent studies reported that the HA and NA functions needed to be modulated to produce a balance in their activities that was required for virus entry into target cells (14,15). Therefore, we investigated whether the effect of NA stalk length on the infection of human airway cells by G1-A/B variants was related to the relatively low ␣2,6 Sia binding affinity of the HAs of these viruses.
To investigate how the balance of HA and NA functions affected virus entry into host cells, we generated viruses carrying either NA-wt or 1 of our 5 NA stalk length mutants and an HA with a distinct Sia affinity from CL80, CL42, or Em/R66. The multiplicities of infection (MOIs) for the virus infections were equalized based on FFU assays on MDCK cells. The infectivity of these viruses was measured with an IN Cell analyzer automated microscope, which enabled us to accurately quantify high numbers of foci in image fields (37,38). In this study, after virus adsorption, the cells were washed and then overlaid with 1% methylcellulose in medium without trypsin to prevent the release and spread of progeny virus, thereby enabling us to observe only foci that were a result of the initial input virus infection. Most H9N2 G1 lineage viruses, including CL80, CL42, and Em/R66, had HAs with mono-or dibasic cleavage sites, and their growth depended on trypsin (20,39). The viruses showed identical infectivity on MDCK cells (Fig. 9A), which confirmed the equalization of the MOIs used for this study. DF-1 cells were then incubated with equal virus MOIs, and infectivity was quantified. The results showed that the infectivity of viruses with an HA from either CL42, CL80, or Em/R66 was correlated with the NA stalk length in these avian cells (Fig. 9B). This was in agreement with our previous result that these viruses had substantially strong binding to ␣2,3 Siaterminated glycans (␣2,3 SLN and/or Su-␣2,3 SLN) (Fig. 8). These results indicate that efficient host cell infection by viruses carrying HA with a high Sia affinity needs to be balanced by an NA with high sialidase activity. Calu-3 cells were then incubated with these viruses, and infectivity was quantified. As was found for DF-1 cells, the infection of human Calu-3 cells by viruses carrying CL42 HA with a high ␣2,6 Sia binding affinity was correlated with the NA stalk length in those viruses: the highest infectivity was with viruses carrying an NA with the longest stalk (Fig. 9C). In contrast, the infectivity of Calu-3 cells with CL80 or Em/R66, which have an HA with a low binding affinity, was negatively correlated with the NA stalk length: virus infectivity was the highest with an NA with a short stalk length and the lowest with an NA with a long stalk length. These results indicate that the effect of NA stalk length on infectivity depends on the binding affinity and/or Sia specificity of the corresponding HA. The results also suggest that a low HA binding affinity can be compensated for by reduced NA activity due to an NA stalk deletion, thereby increasing the infectivity of G1-A/B mutants in human airway cells. As was observed in our previous results ( Fig. 3 and 7), glycosylation at NA residue 61 largely canceled the effect of NA stalk length on viral infectivity. Since the effect of NA stalk length on infectivity was similar in viruses carrying CL80 HA or Em/R66 HA, infectivity was the result of viral HA affinity for an ␣2,6 Sia-or ␣2,3 Sia-terminated glycan but not for a sulfated Sia. In summary, these results indicate that the balance between HA and NA activities determines virus entry and the host range of G1 lineage AI viruses.
Effect of NA stalk length on progeny virus release. Finally, we investigated the effect of NA stalk length on progeny virus release by electron microscopy of infected MDCK cells. The micrographs showed that all the NA stalk length mutants had relatively comparable levels of progeny virus aggregates with a similar morphology on the surfaces of the infected cells (Fig. 10). These results suggest that the NA stalk length mutants in this study do not appear to affect the release of progeny virions.

DISCUSSION
The balance between the HA and NA functions is a determinant of influenza virus growth, virulence, and host range (40)(41)(42). NA stalk deletions and deglycosylation in the NA stalk can modify the HA/NA balance, thereby changing the virus phenotype. Several studies have reported some effects of HA deletions on the H9N2 properties of an old laboratory A/duck/Hong Kong/702/79 strain and of an ancestral Y280 A/chicken/ Shandong/16/05 strain (30,32). However, there have been no studies, thus far, analyzing the effect of NA deletions on biological changes in the G1 lineage strains that are now the most prevalent AI virus strains in Eurasia and Africa. To our knowledge, this is the first report showing a relationship between the NA stalk length, viral growth, and host adaptation of G1 lineage viruses. A similar type of HA/NA balance may be significant for infections of other AI virus lineages.
The NA stalk length affects its enzymatic activity by modifying the structural flexibility of the NA enzymatic pocket (43) and/or its accessibility to Sia (44,45). NA stalk deletions have been reported in several subtype AI viruses and have been associated with viral adaptation in mammalian species (25-27, 29, 32, 46). The H5N1 Gs/GD lineage has a 19-to 20-amino-acid deletion in its NA stalk, which reduces NA enzymatic activity and increases viral virulence in mice (28,31). Experimental deletion of 21 amino acids in the NA stalk contributed to the higher mouse pathogenicity of an old H9N2 strain (32). In this study, 3-to 21-amino-acid deletions in the NA stalk increased G1-A/B  Fig. 2A, is shown at the left. growth in human cells and mice in a deletion-dependent manner. The NA stalk deletions also reduced both NA enzymatic activity with small substrates and viral elution efficacy from erythrocytes, suggesting that adaptation of virus growth in mammals involves a reduction of NA enzymatic activity due to stalk deletions. This was in agreement with the findings of previous studies of other AI viruses (28,31,32).
The role of NA stalk glycosylation on virus properties still needs to be fully elucidated. NA glycosylation has been suggested to contribute to virus replication via altered antigenicity and/or enzymatic activity (47). This study showed that glycosylation of NA stalk residue 61 reduced NA enzymatic activity and blocked the increase in NA activity in an NA with an 8-amino-acid insertion in its NA stalk (residues 58 to 65). NAϩ8aa/gly had less enzymatic activity than NAϩ5aa, although NAϩ8aa/gly had a 3-amino-acid longer stalk (longer by the NIT residues). This implied that glycosylation at residue 61 blocked the increase in NA activity that should have been due to the 3-amino-acid insertion. Thus, NA stalk glycosylation had a strong impact on NA enzymatic activity. This was in agreement with previous studies showing that a 3-amino-acid deletion (i.e., the glycosylation motif at residues 61 to 63) affected the phenotype of Y280 lineage A/chicken/Shandong/16/05 (30,33). Also, the increased virulence of A/Anhui/01/2013 (H7N9) was attributed to a 9-amino-acid NA stalk deletion (from residues 57 to 65), which included the glycosylation motif (48), but not the adjacent 5 amino acids (residues 69 to 73), which did not include a glycosylation motif (49). There have been reports of the conflicting effects of short stalk deletions on virus properties. However, our data suggest that a short NA stalk deletion, even a 3-aminoacid deletion, has a considerable effect on NA enzymatic activity and on the virus phenotype, if the deletion includes the glycosylation motif. Collectively, these results suggest that both stalk length and (de)glycosylation contribute to NA enzymatic activity and virus properties.
Recent studies reported that NA plays a role at the cell surface in virus entry. Virions attach to the cell surface by HA binding to Sia, but most Sia does not mediate virion endocytosis for virus entry. Viruses migrate on the cell surface to target specific domains with preexisting components of the endocytic pathway (14). NA enzymatic activity is needed to drive virion rolling by creating a Sia gradient that forces a virus to roll away from inappropriate Sia receptors (15). HA and NA act cooperatively to move virions on the cell surface. If the HA and NA activities are not balanced and the NA activity is suboptimal, the virion remains stacked on the cell surface, preventing virus movement and entry. Conversely, when NA activity is too strong compared to the HA binding affinity, virion binding is disrupted by NA activity, resulting in detachment (14). Thus, a balance in HA and NA activities drives optimal virion movement on the cell surface, thereby accelerating virus internalization.
In this study, the G1-A/B mutants carried a shortened NA with low sialidase activity that contributed to their efficient growth in human airway cells (Fig. 6), although it has been suggested that lower NA activity is a disadvantage for progeny virus release. The different NA stalk lengths in this study did not have a noticeable effect on progeny virus yield (Fig. 10). In addition, viruses with a shorter NA stalk retained some elution activity from chicken erythrocytes with ␣2,3 Sia abundance (Fig. 4A), which resembled the expression pattern of mucin. This implies that shortened NAs are still capable of penetrating mucin layers.
Our study shows that NA stalk length significantly affects virus entry into human airway cells and that the effect of NA stalk length on viral infection depends on the corresponding HA binding affinity/specificity for target cells (Fig. 9). Collectively, our results suggest that reduced NA activity due to an NA stalk deletion compensated for the reduced HA binding and acted to optimize G1-A/B mutant virus infection of human airway cells, perhaps by modulating virion rolling/uptake on the cell surface. Future studies are needed to fully investigate the effect of NA stalk length on the efficacy of release of progeny G1 viruses with a modified NA stalk length.
In this study, G1-A/B mutants with an 8-amino-acid deletion in their NA were isolated from different, but geographically close, poultry farms in Egypt. This indicates that the G1-A/B mutants with a shortened NA were replication competent and had disseminated in the local poultry population. However, these G1-A/B mutants have not been subsequently detected in the field. Contemporary Egyptian G1-A/B viruses have an NA stalk motif identical to that of NAϩ8aa/gly, i.e., NA-wt with an 8-amino-acid insertion and glycosylation at residue 61. Viruses carrying NAϩ8aa/gly had higher NA sialidase activity (Fig. 3) and slightly more efficient growth in avian cells and eggs than viruses carrying NA-wt (Fig. 5). Thus, the G1-A/B mutants appeared to have no replicative advantage to enable them to replace the prevalent G1-A/B strains in the avian population, which would have led to the loss of the mutant strains. NA stalk glycosylation could implicate an antigenic change (47), suggesting selective pressure for G1-A/B mutants to lose glycosylation to escape immune pressure in the local poultry population. This may explain why a mutant virus with deglycosylated NA emerged in the field, despite its lower replication in poultry.
Since currently prevalent H9N2 G1 lineage viruses can balance their HA and NA functions, their ongoing propagation may promote viral genetic and phenotypic diversification in the field, providing many opportunities to alter the HA/NA balance. During influenza virus adaptation to human hosts in past pandemics, reduced HA activity may have been compensated for by decreased NA activity (50). Therefore, balancing HA and NA functions, particularly if NA contains a stalk deletion, may produce a future G1 lineage virus adapted for mammalian infection. The results presented here provide a novel insight into the functional interdependence of HA and NA on the infectivity and host range of influenza viruses and highlight the importance of closely monitoring the possible emergence of H9N2 G1 mutants with an expanded host range in areas of endemicity.

MATERIALS AND METHODS
Ethics statement. Nine-day-old embryonated chicken eggs were purchased from Shimizu Laboratory Supplies, Japan. All animal experiments were conducted in compliance with Japanese legislation and guidelines under the jurisdiction of the Ministry of Education, Culture, Sports, Science and Technology in Japan. Animal care, housing, feeding, sampling, observation, and environmental enrichment were approved by the Animal Experiment Committee of the Kyoto Prefectural University of Medicine (approval numbers M29-554 and M30-60).
Biosecurity and biosafety. All experiments with live H9N2 viruses were performed at enhanced biosafety level 3ϩ facilities in the Kyoto Prefectural University of Medicine. All studies with recombinant DNA were conducted under the relevant laws in Japan and approved by the Biological Safety Committee of the Kyoto Prefectural University of Medicine (approval number 30-104) after risk assessments were conducted by the Living Modified Organisms Committee of the Kyoto Prefectural University of Medicine and, when required, by the Ministry of Education, Culture, Sports, Science and Technology of Japan. each purified virus were incubated with 50 l of 0.5% chicken or turkey erythrocytes in 96-well round-bottomed microtiter plates at 4°C for 1 h, which allowed the viral HA to bind erythrocyte sialylglycan. The microtiter plates were then transferred to 33°C, which allowed the viral NA to cleave Sia, and the HAU titer was assayed for 60 h. Turkey erythrocytes, which have more ␣2,6 Sia than ␣2,3 Sia, were also treated with ␣2,3 neuraminidase S (New England Biolabs), according to the manufacturer's instructions, to produce erythrocytes with only ␣2,6 Sia on their cell surface. PBS containing 6.8 mM CaCl 2 was used as the diluent. Elution data were plotted against the incubation time at 33°C and analyzed using Origin (version 9.1) data analysis and graphing software (Origin Lab). The data were fitted with the sigmoidal Boltzmann equation, and the elution time at which the HAU titer was reduced from 128 to 64 HAU was calculated. The elution times were expressed relative to the NA-wt elution time, which was set at 100%. Viral infection of cultured cells and eggs. DF-1 and Calu-3 cells (90% confluent in collagen-coated 24-well plates) were infected in triplicate with the indicated viruses at an MOI of 0.005 or 0.03, respectively. After 1 h at 37°C, the DF-1 and Calu-3 cells were washed with PBS and maintained in Dulbecco modified Eagle medium-Ham's F-12 medium (DMEM-F12) containing 0.2% bovine serum albumin (BSA) and 2 g acetylated trypsin/ml (Sigma-Aldrich) at 33°C or 37°C, respectively. The supernatants were collected at the times postinfection indicated throughout the text above. Nine-day-old embryonated chicken eggs (5 eggs per group) were inoculated with 1 ϫ 10 5 FFU of the viruses described above, and allantoic fluids were collected at the times postinoculation indicated throughout the text above. Virus titers were determined by FFU assays on MDCK cells as described above.
Viral infection of mice. Four-to 5-week-old female BALB/c mice (Japan SLC) under mixed anesthesia (medetomidine-butorphanol-midazolam) were inoculated intranasally with 50-l samples of 10-fold serial dilutions (10 3 to 10 7 FFU) of viruses in PBS. The body weight and survival of each mouse were monitored daily for 14 days. Mice that lost more than 30% of their original weight within a few days were humanely euthanized. The lungs of mice infected with 10 5 FFU of virus were collected at 4 dpi, and virus FFU titers were assayed.
Quantitative analysis of foci. Semiconfluent cell monolayers in 96-well plates were inoculated with the viruses indicated above at an MOI of 0.75, which was determined by FFU assays on MDCK cells. After 1 h of adsorption followed by washing with PBS, the cells were overlaid with 1% methylcellulose in modified Eagle's medium without trypsin, to prevent the release and spread of progeny virus, thereby enabling the observation of foci that resulted from the initial virus infection. After 8 h of incubation at 37°C, the cells were fixed and analyzed by immunofluorescence assays to quantify foci that resulted from the initial virus infection and Hoechst-stained nuclei. Images were captured with an IN Cell analyzer 2000 automated microscope (GE Healthcare). Fields containing both virus antigen-positive cells and cell nuclei were counted by IN Cell Developer Toolbox software (GE Healthcare), as described previously (37). Infection rates were calculated by dividing the number of infected cells by the total number of nuclei in the same field.
Electron microscopy. MDCK cells were infected with the viruses indicated above at an MOI of 3. After 18 h of incubation in DMEM-F12 containing 0.2% BSA and 2 g acetylated trypsin/ml (Sigma-Aldrich), the cells were chemically fixed as described previously (18). After fixation, the samples were dehydrated, embedded in a resin (Quetol-812; Nisshin EM), and thin sectioned, and ultrathin 70-nm sections were stained with 2% uranyl acetate, followed by secondary staining with a lead stain solution (Sigma-Aldrich). After sample preparation, progeny virus particles were observed at 100 kV with a transmission electron microscope (model JEM-1400Plus; JEOL). Digital images were captured by a charge-coupled-device camera (model EM-14830RUBY2; JEOL) and visualized using Adobe Photoshop software (Adobe Systems).
Statistical analysis. Statistical analysis was carried out using GraphPad Prism (version 6) software (GraphPad Software). Statistically significant differences between virus pairs were determined by analysis of variance with Tukey's multiple-comparison test.
Data availability. The sequences of the HA and NA gene segments in the CL42, CL79, CL80, and CL83 strains in this study have been deposited in the GenBank database under accession numbers LC379966, LC379968, and LC534257 to LC534262.