Novel Mutations Evading Avian Immunity around the Receptor Binding Site of the Clade 2.3.2.1c Hemagglutinin Gene Reduce Viral Thermostability and Mammalian Pathogenicity

Since 2007, highly pathogenic clade 2.3.2 H5N1 avian influenza A (A(H5N1)) viruses have evolved to clade 2.3.2.1a, b, and c; currently only 2.3.2.1c A(H5N1) viruses circulate in wild birds and poultry. During antigenic evolution, clade 2.3.2.1a and c A(H5N1) viruses acquired both S144N and V223I mutations around the receptor binding site of hemagglutinin (HA), with S144N generating an N-glycosylation sequon. We introduced single or combined reverse mutations, N144S and/or I223V, into the HA gene of the clade 2.3.2.1c A(H5N1) virus and generated PR8-derived, 2 + 6 recombinant A(H5N1) viruses. When we compared replication efficiency in embryonated chicken eggs, mammalian cells, and mice, the recombinant virus containing both N144S and I223V mutations showed increased replication efficiency in avian and mammalian hosts and pathogenicity in mice. The N144S mutation significantly decreased avian receptor affinity and egg white inhibition, but not all mutations increased mammalian receptor affinity. Interestingly, the combined reverse mutations dramatically increased the thermostability of HA. Therefore, the adaptive mutations possibly acquired to evade avian immunity may decrease viral thermostability as well as mammalian pathogenicity.


Introduction
Highly pathogenic H5N1 avian influenza A (HP A(H5N1)) viruses are fatal to poultry and cause high human fatality after dead-end transmission from infected poultry [1,2]. HP A(H5N1) viruses spread from Asia to Africa and Europe by migratory birds, and antigenic evolution has continued under immune pressure by vaccination and natural infection in Asia [1,3,4]. The ancestral HP A(H5N1) virus A/goose/Guangdong/1/96 (clade 0) has evolved into multiple clades from clade 1 to 9 [5]. Some

Data Mining and Analysis of HA Genes and In Silico Analysis of HA Trimer Structure and N-Glycan Profiles
The HA gene sequences of clade 2.3.2, clade 2.3.2.1, and clades 2.3.2.1a, b, and c HP A(H5N1) viruses were collected from the Influenza Virus Database (https://www.ncbi.nlm.nih.gov/genomes/FLU/ Database/nph-select.cgi?go=database) and Global Initiative on Sharing All Influenza Data (GISAID, https://www.gisaid.org/) (n = 647) on April 14, 2018. The collected nucleotide sequences were translated and compared with the BioEdit program (ver. 7.2.5). Additionally, HA sequences of all A(H5) viruses were collected (n = 4189), and amino acid sequences and frequencies of 144NGS and 158NGS and residue 223 (V or I) were analyzed. In addition, HA genes (n = 513) of HP A(H5N1) strains from laboratory-confirmed human cases were analyzed as above. Residues 144N and 223V/I were localized and analyzed for intermolecular interactions with other residues in the 3D structure of the H5 trimer (modified 4juk.pdb and 6e7g.pdb) using PyMOL (Molecular Graphis System version 2.3.1, Delano Scientific LLC, South San Francisco, CA, USA). The N-glycosylation prediction at position 144 was performed by the NetNGlyc program (DTU Bioinformatics, Lyngby, Denmark).

Site-Directed Mutagenesis and Generation of Viruses by Reverse Genetics
To generate mutated H5N1 recombinant viruses, the cloned HA genome of K10-483 was mutated with a Muta-direct Site Directed Mutagenesis Kit (iNtRON, Gyeonggi, Korea) and specific primer sets (Table 1). A Hoffmann's reverse genetics system with a few modifications was used for recombinant virus generation [32]. Briefly, 300 ng of each of the eight plasmids were transfected together into confluent 293T cells in 6-well plates (10 6 cells/well) with Lipofectamine 2000 and PLUS reagents (Life Technologies Co.). After overnight incubation, 1 mL of Opti-MEM (Life Technologies Co.) and 1 µg/mL of TPCK-treated trypsin (Sigma-Aldrich, St. Louis, MO, USA) were added to transfected cells. The supernatant was harvested after another overnight incubation, and 200 µL of the supernatant was inoculated into ten-day-old ECEs. The presence of the recombinant viruses in allantoic fluids was checked by HA assay according to the World Health Organization (WHO) Manual on Animal Influenza Diagnosis and Surveillance, and the genome segments were confirmed by RT-PCR and sequencing as previously described [33]. Table 1. Primers used in this study for site-directed mutagenesis.

Comparative Replication Efficiency in ECEs and Growth Kinetics in MDCK and A549 Cells
The generated recombinant viruses (E1) were passaged in ten-day-old SPF ECEs, and the titers of the recombinant viruses (E2) were measured to obtain the 50% chicken embryo infection dose (EID 50 ). Each virus was diluted 10-fold and inoculated into ECEs. Virus replication was confirmed by a plate hemagglutination test with 1.0% chicken Red Blood Cells (RBCs), and the EID 50 was calculated by the Spearman-Karber method [34]. To compare the replication efficiency of the recombinant viruses, the same number of viruses were used to infect ECEs and mammalian cells. One hundred EID 50 of recombinant virus (E2) were inoculated into ten-day-old SPF ECEs, and after 3 days, the EID 50 of harvested allantoic fluid was measured as above. In MDCK and A549 cells, 10 5 EID 50 of each virus were used to infect confluent MDCK and A549 cells in 12-well plates. The supernatant of the infected cells was harvested at 0, 12, 24, 48, and 72 h post inoculation (hpi), and 10-fold diluted supernatant was used to inoculate confluent MDCK cells in a a 96-well plate to measure the 50% tissue culture infectious dose (TCID 50 ) at each time point. Virus replication was confirmed by a hemagglutination assay, and TCID 50 was calculated using the Spearman-Karber method as above.

Mouse Pathogenicity Test
The mouse pathogenicity test was approved on January 10, 2019 by the Institutional Animal Care and Use Committee (IACUC) of Seoul National University (IACUC-SNU-171214-1-1). The approved experiment was performed in a biosafety level 2 facility at the Animal Center for Pharmaceutical Research of Seoul National University (Seoul, Korea) according to the national guidelines for the care and use of laboratory animals. Six-week-old female BALB/c mice (n = 8) (KOATEC, Pyeongtaek, Korea) were anesthetized by Zoletil 50 (15 mg/kg, IP) (Virbac, Carros, France), and 10 6 EID 50 /50 µL of each recombinant virus was inoculated intranasally. Five mice from each group were weighed for 2 weeks, and three mice were euthanized to obtain lung samples at 3 days post inoculation (dpi). During the experiment, mice with 20% or more body weight loss were euthanized. The sampled lungs were ground with TissueLyzer 2 and 5 mm stainless steel beads (Qiagen, Valencia, CA, USA) and suspended in PBS. The virus titer (EID 50 ) was measured as previously described [35].

HA and Hemagglutination Inhibition (HI) Tests
The HA test and HI test with chicken RBCs and guinea-pig RBCs were performed according to the WHO manual for the laboratory diagnosis and virological surveillance of influenza. The recombinant virus was serially diluted 2-fold in 96-well plates, and chicken RBCs (1%) or guinea-pig RBCs (1%) were added. After 40 min of incubation at 4 • C, the hemagglutination unit (HAU) of each virus was recorded. Chicken RBCs have similar amounts of sialic acids bound to galactose by α2,3 linkage (SAα2,3Gal) and sialic acid linked to galactose by α2,6 linkage (SAα2,6Gal), and guinea pig RBCs have more SAα2,6Gal than SAα2,3Gal [38]. The HI test of recombinant viruses was conducted with chicken egg white to compare the resistance of the recombinant viruses against egg white. Chicken egg white was serially diluted as in the HA test, and 4 HAU of each virus were added to each well and incubated for 30 min at 4 • C. Then, chicken RBCs or guinea pig RBCs were added, and the HI titer was recorded after 40 min of incubation at 4 • C. All experiments were repeated three times independently.

Heat Stability Test
Recombinant viruses were diluted to the same HA titer (2 4 ) and aliquoted for heat treatment. Each aliquot was incubated at 60 • C for 0, 5, 15, and 30 min, and the HA titer was measured.

SDS-PAGE and Western Blotting
To confirm the 144N glycosylation, 4 µL of each recombinant virus (CE3) treated or untreated with PNGase F enzyme (New England Biolabs, Ipswich, MA, USA) was mixed with Protein 5X Sample Buffer (ELPIS BIOTECH, Daejeon, Korea) to denature it for 5 min at 95 • C, and SDS-PAGE was performed using NuPAGE 4-12% Bis-Tris Protein Gels (Life Technologies Co.). The proteins were transferred to a nitrocellulose membrane (Life Technologies Co.), and the membrane was incubated with anti-H5N1 virus (A/Vietnam/1194/2004), HA rabbit IgG (Sino Biological Inc., Beijing, China), followed by incubation with horseradish peroxidase conjugated-goat anti-rabbit IgG (Bethyl Laboratories Inc., Montgomery, AL, USA). Then, HA proteins were visualized with BioFX TMB One Component HRP Membrane Substrate (SurModics IVD, INC., Eden Prairie, MN, USA) and sulfuric acid stop solution (Sigma-Aldrich).

Statistical Analysis
All data were analyzed with IBM SPSS Statistics version 23 (IBM., Armonk, NY, USA). The statistical significance of viral titers in ECEs, growth kinetics in cells, and receptor binding affinity were evaluated by one-way analysis of variance (p < 0.05). Survival rates were compared by Kaplan-Meier survival analysis, and the differences in frequency were assessed by chi-square test (p < 0.05).

Comparison of Replication Efficiency in Mammalian Cells and Mouse Pathogenicity of the H5N1 Recombinant Viruses
The replication efficiency of H5N1 recombinant viruses was compared in MDCK and A549 cells. The viral titers in MDCK cells were similar among the recombinant viruses at 24 and 48 hpi, but rH5N1-N144S-I223V showed a significantly higher titer than other H5N1 recombinant viruses at 72 hpi ( Figure 1A). In A549 cells, rH5N1-N144S-I223V showed significantly higher viral titers at 48 and 72 hpi ( Figure 1B). All the H5N1 recombinant viruses replicated in the lungs of infected BALB/c mice at 3 dpi, but rH5N1-N144S-I223V showed a significantly higher virus titer than that of other H5N1 recombinant viruses (Table 5). In addition, rH5N1-N144S-I223V infection resulted in apparent body weight loss in all mice, and the mice died (100% mortality) within 4 dpi ( Figure 2). However, rH5N1-N144S, rH5N1-I223V, and rH5N1 did not cause significant body weight loss during the observation period.  Weight loss was monitored for 2 weeks, and mice with more than 20% weight loss were euthanized. The weight loss was calculated based on the body weight measured at 0 dpi, and the data are the average of each group; * significant difference of the rH5N1-I223V and rH5N1-N144S groups compared to the mock groups (p < 0.05).

Comparison of Binding Affinity of the H5N1 Recombinant Viruses to Avian (3 -SLN) and Mammalian (6 -SLN) Receptors and Egg White
We compared the receptor binding affinity of recombinant viruses to 3 -SLN and 6 -SLN ( Figure 3). All the recombinant viruses bound to 3 -SLN more strongly than to 6 -SLN. rH5N1 and rH5N1-N144S-I223V showed significantly higher binding affinities than other viruses, and rH5N1-N144S had the weakest binding affinity to the avian receptor ( Figure 3). When chicken RBCs were used to measure HI titers in egg white for the H5N1 recombinant viruses, only rH5N1-N144S showed slightly lower HI titers than other viruses (64 vs. 128, Table 6). However, the difference was much higher when we used guinea pig RBCs (<8 vs. 64). Therefore, rH5N1-N144S was significantly less inhibited by egg white, and 223I may play a role in the resistance to egg white.

Comparison of Thermostability of the H5N1 Recombinant Viruses
The HA titers of the H5N1 recombinant viruses decreased to zero within 5 (rH5N1-N144S), 15 (rH5N1), and 30 min (rH5N1-I223V) after heat treatment, but that of rH5N1-N144S-I223V was maintained even after heat treatment for 30 min (Figure 4). Therefore, the V223I mutation may decrease HA thermostability.

Intra-and Intermolecular Interactions of Residue 223 in the HA Trimer and Confirmation of the 144N-Glycosylation
Residue 223 is located in the 220-loop, and both 223I and 223V interact intramolecularly with 226Q via hydrogen bonding. Interestingly, residue 223 is located close to 207S of another neighboring HA monomer. Considering the larger side chain of I than V, the V223I mutation may affect the integrity of the HA trimer ( Figure 5). So far, 144NGS is considered to be a real N-glycosylation site, and our Western blotting data also agree with previous reports (Figure 6) [44].  HA proteins of rH5N1 and rH5N1-I223V had 144N-glycan, and they had higher molecular weight than rH5N1-N144S and rH5N1-N144S-I223V in the absence of PNGase F enzyme treatment (−). However, the difference disappeared after treatment of PNGase F enzyme (+).  [7,45]. In Viet Nam, clades 2.3.2.1a, b, and c viruses used to cocirculate, but clade 2.3.2.1b viruses have disappeared [10]. Clade 2.3.2.1 viruses acquired 144NGS, and this mutation was conserved in most of the clade 2.3.2.1a, b, and c viruses. Considering the significantly higher frequency of 158NGS than 144NGS among A(H5) viruses, given the same possibility of acquiring either 144NGS or 158NGS (Table S1), the acquisition of 144NGS may reflect the presence of certain selection pressure or other circumstances.

Clade 2.3.2 viruses were isolated from ducks and wild birds in mainland
HA glycosylation to evade humoral immunity may be the most effective but final choice because it reduces viral fitness [46]. The variability of NGS precursors and their higher frequencies than NGS at 144-146 (86.5% vs. 13.5%) and 158-160 (71.9% vs. 28.1%) may be in line with the above notion (Table S1). The 144N-glycan reduced the replication efficiency in ECEs and the a2,3 SA affinity of rH5N1-I223V in comparison with rH5N1-N144S-I223V. Therefore, 144N-glycan in H5 reduced viral fitness, similar to the findings in a previous report in H1 [46]. However, circumstantial evidence explaining why clade 2.3.2.1 viruses chose 144NGS instead of the more prevalent and preferable 158NGS needs to be discussed further.
Humoral immunity induced by vaccination may facilitate the appearance of mutants evading vaccine immunity [47].  [50]. Therefore, vaccine-induced antibodies might have targeted the shielded epitope site B (group 2, 155-166) rather than epitope site A (group 1, 140-145), and mutant viruses acquiring 144NGS to shield epitope site B might have been selected [44]. Re-6 with 144NGS might have been effective against clade 2.3.2.1a, b, and c viruses not only due to antigenic similarity but also due to the restricted acquisition of additional 158NGS. The very low frequency of A(H5) viruses possessing both 144NGS and 158NGS (0.1%) may reflect the inferior competitiveness of such HAs in nature (  [10,52,53]. The V223I mutation is unlikely to stand alone without 144NGS due to its very low frequency in nature, and stepwise acquisitions of 144NGS and V223I mutations during clade 2.3.2.1a, b, and c diversification from clade 2.3.2.1 are noteworthy (Table 3). Similarly, cooperating mutations with N-glycans, K147 with 144NGS, and N227S with 158NGS have been reported [23,54]. The biological effects and roles of the V223I mutation are unclear, but a single V223I mutation decreased the virus replication efficiency of rH5N1-N144S in ECEs. Additionally, rH5N1-N144S showed resistance to egg white due to a relatively steep decrease in HI titer and significantly lower a2,3 SA affinity than those of the other viruses in this study (Table 6, Figure 3). Ovomucin in egg white is an effector molecule of innate immunity present on the surface of the mucous membrane, and a different mutation reducing the inhibition has been reported [36,55]. Therefore, characterization of egg white resistance may be useful to understand the evolutionary status of IAVs. Mutations such as S223N (S224N according to our H3 numbering), N224K, G225D, Q226L, G228S, and S227N directly or indirectly increase a2,6 SA affinity, and their side chains are located in or near the RBS. However, the side chain of residue 223 is located at the interface of the globular heads of the HA trimer ( Figure 5). The amino acid residues located between the interfaces of the HA trimer affect structure and pH stability, and the electrostatic intermolecular interaction between T212 and N216 and the increased rigidity of the S221P mutation stabilized the HA trimer to increase pH stability [25,26]. The lower thermostability of rH5N1-N144S than rH5N1 and rH5N1-I223V may imply a negative effect of the intermolecular interaction between 223I and 207S on HA thermostability ( Figure 5). V/I223 did not interact with S207 via hydrogen bonding, and the bulkier side chain of I may cause steric hindrance at the interface. Although we did not test pH stability, a pH stability-related mutation (H103Y) showed increased thermostability by stabilizing the HA trimer in previous reports [56,57]. Therefore, the V223I mutation acquired in addition to S144N during adaptation in vaccinated poultry might have improved viral fitness in terms of a2,3 SA receptor affinity at the cost of decreased HA trimer stability.
The mammalian pathogenicity of HP A(H5N1) viruses is a multigenic trait, and the human pathogenicity of clade 2.3.2.1 viruses has been regarded as lower than that of other clades [58]. In our previous studies, a PR8-derived clade 2.3.2.1c recombinant vaccine strain showed less pathogenicity in mice than another PR8-derived recombinant virus containing HA and NA genes from a low pathogenic (LP) A(H5N1) virus [29,59]. Although the amino acid sequences of the two HA proteins are only 89% identical, the HA of the LP A(H5N1) virus has V223 and neither 144NGS nor 158NGS. The higher mouse pathogenicity and replication efficiency of rH5N1-N144S-V223I than other H5N1 recombinant viruses in MDCK and A549 cells may indicate a reduction in mammalian pathogenicity during poultry adaptation to evade immune responses. Poorly glycosylated HA was recognized by an ER stress pathway and induced strong lung injury [60]. The relatively high mammalian pathogenicity of viruses without 144NGS and 158NGS has been verified [23]. As we did not directly compare the pathogenicity of 144NGS-or 158NGS-bearing recombinant viruses, we cannot conclude which virus is more pathogenic in humans. However, a significantly higher frequency of 158NGS-(70.0%) than 144NGS-bearing (1.6%) HA in human cases compared with the frequencies of single 158NGS-(27.2%) or 144NGS-bearing (13.1%) HA in nature may reflect a higher risk of 158NGS than 144NGS in human infection (Table S2).
In conclusion, intensive inoculation of certain types of vaccines may distort natural evolutionary pathways, and the acquired novel adaptive mutations may reduce viral fitness by destabilizing the HA trimer as well as affecting mammalian pathogenicity. Our results may provide clues for future studies to develop more effective vaccine strains and programs that reduce the appearance of antigenic or more human-pathogenic variants.