Multiple polymerase gene mutations for human adaptation occurring in Asian H5N1 influenza virus clinical isolates

The role of the influenza virus polymerase complex in host range restriction has been well-studied and several host range determinants, such as the polymerase PB2-E627K and PB2-D701N mutations, have been identified. However, there may be additional, currently unknown, human adaptation polymerase mutations. Here, we used a database search of influenza virus H5N1 clade 1.1, clade 2.3.2.1 and clade 2.3.4 strains isolated from 2008–2012 in Southern China, Vietnam and Cambodia to identify polymerase adaptation mutations that had been selected in infected patients. Several of these mutations acted either alone or together to increase viral polymerase activity in human airway cells to levels similar to the PB2-D701N and PB2-E627K single mutations and to increase progeny virus yields in infected mouse lungs to levels similar to the PB2-D701N single mutation. In particular, specific mutations acted synergistically with the PB2-D701N mutation and showed synergistic effects on viral replication both in human airway cells and mice compared with the corresponding single mutations. Thus, H5N1 viruses in infected patients were able to acquire multiple polymerase mutations that acted cooperatively for human adaptation. Our findings give new insight into the human adaptation of AI viruses and help in avian influenza virus risk assessment.


Analysis of human adaptation mutations in the polymerase complex of Asian H5N1
Nviruses. Bioinformatics allows us to identify novel mutations in avian H5N1 viruses that enable efficient viral replication in mammals. However, identification of putative adaptation mutations that differ from the consensus sequence is challenging when viruses in diverse H5N1 genetic lineages are analyzed, since rare mutations may be masked. In fact, there has been movement and (co) circulation of distinct H5N1 clades in the field in East and Southeast Asia 4,[21][22][23]25,26 .
To avoid such potential problems, we conducted a database search targeting a geographically and chronologically close group of PB2, PB1, and PA polymerase and NP nucleoprotein gene sequences in human and avian H5N1 strains isolated in Southern China, Vietnam and Cambodia during 2008-2012. The nucleotide sequences were downloaded from the National Center for Biotechnology Information (NCBI) Influenza Virus Resource (www.ncbi.nlm.nih.gov/genomes/FLU/). This yielded a total of 24 clade 1.1, clade 2.3.2.1 and clade 2.3.4 human virus PB2, PB1, PA and NP sequences, which included all the public database available sequences from human viruses in Asia during this period (as of October, 2017) and a total of 198 avian virus PB2, PB1, PA and NP sequences in the corresponding clades. Although a few sequences from human influenza viruses in these clades were isolated later and were available in the GISAID EpiFlu database (https://www.gisaid.org/) at the time of our database search, only sequences isolated in 2008-2012 were used in this study.
The consensus sequences of the polymerase and NP genes were determined by aligning all the sequence data of the vRNA segments carrying those genes. We then searched the sequence data set for polymerase and NP genes with amino acid mutations that had been presumably selected in H5N1-infected patients. Mutations in the polymerase and NP gene sequences of human and avian viruses were identified by comparing each gene sequence to its consensus sequence. Based on their prevalence in human and avian viruses, we identified a total of 33 single mutations (9 in PB2, 8 in PB1, 14 in PA and 2 in NP) that were either only in human viruses or were more prevalent in human viruses than in avian viruses as putative human adaptation mutations in clade 1.1, clade 2.3.2.1, and clade 2.3.4 viruses ( Table 1). Some mutations were the only mutation in a vRNA segment in some strains, some mutations were only found together with one or more other mutations in a vRNA segment in some strains, and some mutations were the only mutation in a vRNA segment in some strains and with one or more other mutations in that vRNA segment in other strains (Table S1). Therefore, we searched for multiple mutations in viruses isolated from patients and identified 13 viruses with multiple mutations. In all, 49 single and multiple mutations were investigated in this study.
Effect of H5N1 polymerase mutations on viral polymerase activity. To investigate the effect(s) of the mutations on viral polymerase activity, we carried out minigenome assays of polymerases with these mutations in the genetic background of A/Vietnam/HN31432/2008 (VN/HN). VN/HN is a representative H5N1 clade 2.3.4 strain 27 with no known human adaptation mutations (e.g., PB2-E627K and PB2-D701N) in its HA and polymerase genes. VN/HN was used as the genetic background in this study because most of the identified mutations (13/24) (Table S1). Human 293T cells and avian QT-6 cells were used for this study. For 293T cells, polymerase activity at both 33 and 37 °C (the temperature of the human upper and lower respiratory tract, respectively) was assayed. For QT-6 cells, polymerase activity was assayed at 37 °C to allow the results to be compared with those for 293T cells at the same temperature. Among the identified mutations, the PB2-D701N single mutation has been reported to increase the replication of avian H5N1 viruses in mammals 12,28,29 and, therefore, was included in this study for comparison with other single and multiple mutations.
Minigenome assays showed that several single and multiple mutations in PB2 and PA increased H5N1 VN/HN polymerase activity in human 293T cells; e.g., PB2-T81A/V344M/D701N, PB2-Y658H, PB2-Y658H/V344M, PB2-D701N, PA-M86V/A343T, PA-M86V/A343T/E613V, PA-A343T and PA-A343T/E613V significantly increased polymerase activity at 37 °C (Fig. 1A), although a few other mutations reduced the polymerase activity. Several multiple mutations acted synergistically to produce a greater increase in polymerase activity than the single mutations. The PB2-T81A/V344M/D701N triple mutation produced up to 4.1-fold higher polymerase activity than wild-type PB2 and significantly higher polymerase activity than the known PB2-D701N human adaptation mutation (Fig. S1A, P < 0.01 by ANOVA with Tukey's multiple comparison test). The PA-M86V/ A343T double mutation also had a synergistic effect, with a 3.9-fold higher polymerase activity than wild-type PA and significantly higher polymerase activity than PA-A343T, which also increased polymerase activity (Fig. S1A, P < 0.01 by ANOVA with Tukey's multiple comparison test). In each case, the effect of the multiple mutation in increasing polymerase activity was greater than the increase produced by the PB2-D701N single mutation. Similar results were obtained at 33 °C, although the effects were more moderate (Figs 1B and S1B). The mutations in PB2 increased polymerase activity in a human cell-specific manner: the effects were significantly less in avian cells (Figs 1C and S1C). In addition, the mutations increased polymerase activity at both 37  that is geographically unique to Egypt 24,30 ( Fig. S2A-C). Taken together, these results suggested that the mutations selected by viral growth in patients acted, separately and together, to increase H5N1 polymerase activity in human cells.
Effect of H5N1 polymerase mutations on viral replication in human airway epithelial cells. We and PA-M86V/A343T mutants produced progeny virus titers that were up to 4.0-fold higher in A549 cells and 8.3-fold higher in Calu-3 cells, compared to VN/HN (wt) ( Fig. 2A and B). Progeny virus production by these mutants at 33 °C was similar to that at 37 °C, but with greater differences between the mutants and wt at 33 °C. The progeny virus titers produced by the mutants were up to 31.2-fold higher in A549 cells and 298.4-fold higher in Calu-3 cells, with the highest increase produced by the PB2-T81A/V344M/D701N mutant ( Fig. 3A and B). The PB2-T81A/V344M/D701N and PA-M86V/A343T multiple mutations showed synergistic effects on viral replication compared with the corresponding single mutations both at 37 and 33 °C ( Fig. S3A-D, respectively): the maximum progeny virus titers were PB2-D701N < PB2-T81A/V344M/D701N (P < 0.01 by ANOVA with Tukey's multiple comparison test) and PA-A343T < PA-M86V/A343T (P < 0.01 by ANOVA with Tukey's multiple comparison test). The effects of these single and multiple mutations on viral replication in human cells were comparable to that produced by the PB2-D701N mutation and, in some conditions, comparable to that produced by the PB2-E627K mutation. In contrast, viral growth of the PB2-G685R putative negative control mutation was indistinguishable from that of VN/HN (wt), both at 33 and 37 °C, although PB2-G685R is located in the PB2-627 domain. These results indicated that the increased polymerase activity of the selected mutations was significant in their adaptation for virus replication in human cells.
To confirm the effects of the selected mutations on H5N1 virus replication, we carried out focus-forming assays in canine MDCK cells and measured the diameters of the foci due to the mutations. Although MDCK cells are not a human cell line, they have been used to investigate human adaptation of AI viruses 16,31,32 . Most mutants produced significantly larger foci diameters than VN/HN (wt) (Fig. 4A and B), and the PB2-T81A/ V344M/D701N triple mutation and PA-M86V/A343T double mutation had synergistic effects compared to the corresponding single mutations (Fig. S4, P < 0.05 by ANOVA with Tukey's multiple comparison test), which were correlated with their increased polymerase activity and replication in human cells. In contrast, in avian DF-1 fibroblasts infected with one of the mutant viruses or with VN/HN (wt) at a MOI of 0.01, there was little difference between the progeny virus titers produced by the mutants and by VN/HN (wt) at all the time points studied (Fig. 5). These results showed that the H5N1 viral mutations selected in patients had a significant effect, alone or in combination with other mutations, in increasing viral replication in human airway cells but not in avian cells, indicating a species-specific role for these mutations in H5N1 virus adaptation to humans. Structural model of the VN/HN polymerase complex. Models of the VN/HN polymerase complex structure were generated from a crystal structure of the influenza virus polymerase (Protein Data Bank ID code 4WSB) 33 . Our models showed that the mutations that produced increased H5N1 polymerase activity in human cells were located in several domains of the polymerase subunits other than the PB2-627 domain (Fig. 7) [34][35][36][37] .
Of the PB2 mutations, T81A was in the N1-subdomain and close to the vRNA promoter, V344M was in the Cap-binding domain, Y658H was in the PB2-627 domain and D701N was in the PB2-NLS domain. For PA mutations, PA-M86V was in the PA-endonuclease domain, and A343T and E613V were in the PA-C terminal domain. All the mutations were exposed at the protein surface, except PA-M86V which was at the inter-subunit interface between the PA N-terminal region and the PB C-ext domain.

Discussion
Previous studies to identify human adaptation mutations in influenza viruses have typically relied on comparison of viruses with high and low pathogenicity in mammals. High-throughput screening now allows analysis of large numbers of mutations and identification of previously unknown human adaptation mutations 17,24,31 . In this study, we targeted a geographically and chronologically related group of polymerase mutants identified in human H5N1 strains isolated in Southern China, Vietnam and Cambodia during 2008-2012. This targeted bioinformatics approach enabled us to identify a number of polymerase gene mutations in H5N1 viruses that had not been previously reported to be human adaptation mutations, in addition to the previously reported PB2-D701N human adaptation mutation. These mutations increased the polymerase activity and replication of clade 2.3.4 viruses in both human cells and mouse lungs to levels similar to the PB2-D701N and PB2-E627K single mutations, which are known human adaptation polymerase mutations. The identification of previously reported and unreported human adaptation mutations in H5N1 viruses confirmed the applicability of our experimental approach and the design of this study. Although the effects of the mutations identified in this study on polymerase activity were relatively low (i.e., less than 5-fold) in our minigenome assay system compared with other studies 14 have been due to differences in the genetic background used to generate recombinant viruses and to the assay conditions used in these studies. The adaptation of AI virus polymerase in mammals has been studied extensively and several mutations have been reported to be related to viral cross-species transmission and adaptation to humans. Among these mutations, PB2-E627K and, to lesser extent, PB2-D701N have been described as being critical for avian H5N1 virus replication in mammalian hosts 7,10-12,19,20 . However, in most cases, these mutations were detected as single mutations, implying that they did not have to act synergistically with other mutations. In fact, the PB2-D701N and PB2-Q591R mutations alone, but not combined with the PB2-E627K mutation, produced a replication advantage for the avian H5N1 virus polymerase during human cell infection 13 . In contrast, in this study we found synergistic effects with the PB2-T81A/V344M/D701N and PA-M86V/A343T multiple mutations, producing an additional increase in viral replication in human cells and mouse lungs. The PB2-T81A, PB2-V344M and PB2-T81A/V344M mutations alone had little effect but increased polymerase activity when the two mutations were combined with the PB2-D701N mutation (Figs 1 and S1), implying that both the PB2-T81A and PB2-V344M mutations probably need to cooperate with the PB2-D701N mutation to produce a synergistic effect. Likewise, the PA-M86V mutation alone had little effect on polymerase activity but had a synergistic effect with the PA-A343T mutation. This is in agreement with recent studies showing that several polymerase mutations act cooperatively with PB2-E627K to increase viral growth in mammals 8,17 . These findings suggested that avian H5N1 viruses could acquire multiple human adaptation polymerase mutations in infected patients and that these mutations may act synergistically in optimizing viral fitness to infect humans.
In this study, we identified human adaptation mutations that acted to increase polymerase activity in the genetic background of both clade 2.3.4 and clade 2.2.1 viruses, although clade 2.2.1 viruses are geographically and phylogenetically far from Asian H5N1 clades. This implied that the mutations enabled viruses in different H5N1 genetic lineages to increase viral replication in human cells, although not all the mutations may have the same effects in other clades. However, most of the mutations that significantly increased clade 2.3.4 virus replication in human cells were in the PB2 and PA genes, which was in contrast to our previous study showing that human clinical isolates of clade 2.2.1 viruses had acquired multiple human adaptation polymerase mutations in the PB1 gene 24 . The mechanism(s) underlying the differences in the location of human adaptation polymerase mutations remains unknown. However, the different locations may be due to the different genetic backgrounds of the The PB2-627 domain has been reported to be a frequent host-adaptation site in the AI polymerase 7,10 , but the exact mode-of-action of the polymerase domains for host adaptation remains unknown. In contrast, our database search approach identified a number of human-adaptation mutations that were scattered over the entire polymerase complex, besides the PB2-627 domain. Although one of the human adaptation mutations identified in this study, PB2-Y658H, was located within the PB2-627 domain, it was at a substantial distance from PB2-627 and did not form a structural site with the PB2-627 residue. In addition, mutations that acted synergistically to increase viral replication were not located near one another in the PB2 structure. This was consistent with previous reports showing that H5N1 viruses could acquire human adaptation polymerase mutations at sites other than in the well-characterized PB2-627 domain 8,17 . The mutations identified in this study were also located on the surface of discrete domains of the polymerase complex or at the inter-subunit interface of the polymerase subunits. In addition, a number of the mutations in PB2 were clustered within the NLS and the Cap-binding domains. This suggested that the mutations may mediate the interaction of the polymerase complex with host factors or among the polymerase subunits. Previous studies reported that PB2-E627K in the 627 domain and PB2-D701N in the NLS domain affected the interaction between the viral polymerase and host importin-α isoforms to adapt mammalian machinery 19,20 . Also, some mutant residues may not make direct protein contacts but instead may affect protein flexibility, to help other protein regions maintain polymerase activity or to promote interactions with other protein domains or host factors. These considerations implied that the molecular basis underlying human adaptation by H5N1 polymerase mutations may be multifactorial. However, our structural models may be incomplete, because the viral polymerase complex has been found to be flexible, which allows it to adopt different conformations. Therefore, the exact mechanism(s) by which mutations affect human adaptation of AI viruses needs to be further investigated. mutagenesis libraries, the PB2-Y658S amino acid change was identified as a mutation that increased H5N1 replication in human cells 14 . PB2 residue 658 also forms a host-specific structural pocket along with residues 613 and 661 that may be involved in avian to human adaptation 43 . PB2-V344M has been reported to be one of the mutations in influenza virus A/Indonesia/UT3006/2005 (clade 2.1.3) that increased viral replication in human cells 13 . However, in these previous studies, the effect of these mutations on host adaptation was not assessed because of their relatively low prevalence in H5N1 viruses.
This study was limited by the relatively small number of sequences of clinical human clade 1.1, clade 2.3.2.1 and clade 2.3.4 strains isolated in 2008-2012 in databases, although all of the available sequences of human strains isolated during 2008-2012 were analyzed in this study. Furthermore, host adaptation of influenza viruses involves polygenetic traits, implying that AI viruses probably require adaptation mutations in addition to those in PB2 (e.g., in other genes such as the HA gene 4,44 and the NS gene 45,46 ) to become pandemic. Future database searches should be done using larger sequence data sets, since larger numbers of AI sequences from human isolates have recently been deposited in multiple public databases and should be available for bioinformatics analyses.
In conclusion, we have identified a number of putative human adaptation polymerase mutations that were selected in infected patients using a database search targeting geographically and chronologically close sequences of H5N1 viruses from Southern China, Vietnam and Cambodia. Since all the mutations identified here have been found in natural isolates, these human adaptation mutations may (re) emerge in novel strains and perhaps facilitate virus adaptation to mammals. In fact, novel H5N6 viruses carrying clade 2.3.4 internal genes have emerged since 2013 and caused outbreaks in East Asia and Southeast Asia 42,47 . Some viruses isolated from these outbreaks had the PB2-Y658H polymerase human adaptation mutation that was identified in this study. The results presented here should help in influenza virus surveillance efforts and contribute to understanding the mechanism of AI virus adaptation to new hosts.

Materials and Methods
Ethics statement. All    in which all experimental work is carried out in biosafety cabinets. Air exhausted from the class 3 units is filtered by HEPA filters and then leaves the facility via a second set of HEPA filters. The BSL3+ has a dedicated electrical generator in the event of power loss.
Only authorized personnel that have received appropriate training can access the BSL3+ facility. All personnel working in the BSL3+ facility wear a disposable protective, FFP3 facemasks and multiple pairs of gloves. Furthermore, all personnel conducting this study were vaccinated against seasonal and H5N1 influenza viruses. Antiviral drugs are directly available to further mitigate risks upon incidents.
In this study, all the mutations introduced into the recombinant VN/HN virus (clade 2.3.4) have been detected as single or multiple mutations in H5N1 viruses isolated from patients, except the PB2-E627K mutation. Also, the mouse infection study was performed after no increase was observed in the progeny virus titers of the selected viruses carrying the mutations in this study compared to previously published studies. The PB2-E627K mutant virus was excluded from the mouse infection experiments in this study.

Database search.
A database search was conducted as described previously 17,24 . Briefly, sequences of PB2, PB1, PA and NP genes from 222 influenza A virus subtype H5N1 strains isolated in Southern China, Vietnam and Cambodia from 2008-2012 were obtained from the National Center for Biotechnology Information (NCBI) Influenza Virus Resource (www.ncbi.nlm.nih.gov/genomes/FLU/). Duplicate sequences from the same strain and sequences with questionable amino acid translation were removed. Nucleotide sequences of genes with more than 100 amino acids missing at either end were also excluded. These sequences were aligned using the MAFFT program 48 . Polymerase mutations in 24 human and 198 avian H5N1 virus strains were identified by comparing these sequences to a consensus sequence of each protein that was determined from the aligned sequences of all the H5N1 strains downloaded. The prevalence of the mutations in the human and avian virus strains was then calculated and compared between viruses isolated from human and avian hosts. To increase the likelihood of identification of relevant human-adaptation mutations, polymerase mutations that were either in human viruses or were more prevalent in human viruses than in avian viruses (arbitrary cutoff value of over 4 fold) were included.

Cells. Human embryonic kidney 293T cells, human lung carcinoma A549 cells, canine kidney MDCK cells
and quail fibroblast QT-6 cells were maintained in Dulbecco's Modified Eagle's Medium or Ham's F-12K medium supplemented with 10% fetal calf serum at 37 °C in 5% CO 2 . Human bronchial epithelial Calu-3 cells and chicken fibroblast DF-1 cells were maintained in Dulbecco's Modified Eagle's Medium supplemented with 10% fetal calf serum at 37 °C in 5% CO 2 . Chicken embryo fibroblasts (CEFs) were prepared from 10-day-old embryonated eggs as described previously 49 . Virus preparation. Influenza viruses were grown in 10-day-old embryonated chicken eggs that had been purchased from Shimizu Laboratory Supplies, Japan. The allantoic fluids and culture supernatants were then harvested and stored as seed viruses at −80 °C. Virus purification to produce working stocks was described previously 30,50 . Virus titers were assayed as FFU by focus-forming assays using MDCK cells as described below.
Focus-forming assays. MDCK cells (90% confluent in 96-well plates) were infected and, after 1 h at 37 °C, the virus inoculum was removed and the cells were washed with PBS and overlaid with 1% methylcellulose in Modified Eagle's Medium. At 16 h post-infection at 37 °C, the cells were fixed with 4% paraformaldehyde in PBS, incubated with anti-H5N1 virus polyclonal antibody for 1 h at 37 °C, washed 3 times with PBS, and reacted with Alexa Fluor 488 anti-rabbit antibody for 1 h at 37 °C. After 3 washes with PBS, the plates were coated with a 50% glycerol solution. Digital images were taken using an inverted fluorescence microscope ECLIPSE Ti2 system (Nikon) and foci sizes were measured using NIS-Elements software (Nikon).

Minigenome assays.
Minigenome assays based on a dual-luciferase system were performed as described previously 17,50 . Briefly, 293T and QT-6 cells were transfected with plasmids each expressing the PB2, PB1, PA or NP gene of A/Vietnam/HN31432/2008 (VN/HN) or A/duck/Egypt/D1Br/2007 (EG/D1), and a human or chicken polymerase I-driven plasmid expressing firefly luciferase from a virus-like RNA. Cells were also transfected with a plasmid expressing Renilla luciferase to monitor transfection efficiencies. Cells were transfected with these plasmids using TransIT-LT1 (Mirus) and incubated at 33 or 37 °C. Firefly luciferase activity values at 48 h post-transfection were normalized relative to the Renilla luciferase activity.
Reverse genetics. Recombinant viruses were generated with a plasmid-based reverse genetics system in the VN/HN (wt) virus genetic background as described previously 51,52 . All viruses were sequenced to ensure the absence of unwanted mutations. Experimental infections in mice. Five-week-old female BALB/C mice (Japan SLC), under mixed anesthesia (medetomidine-butorphanol-midazolam), were intranasally inoculated with 10 −1 to 10 4 FFU of viruses in PBS. The mice body weight and survival were monitored daily for 14 d. Mice that lost more than 30% of their original weight were euthanized. In addition, at 3 and 6 d after inoculation with 1 × 10 2 FFU virus, mouse lungs were collected and virus titers were assayed as FFU in MDCK cells.

Viral growth kinetics in cultured cells. A549 and
Homology modeling of VN/HN polymerase mutants. Homology modeling of VN/HN polymerase used the crystal structure of the heterotrimeric polymerase complex of influenza virus A/little yellow-shouldered bat/Guatemala/060/2010 (H17N10) (Protein Data Bank ID code 4WSB) 33 as a template, as described previously 17 . Statistical analysis. Statistical analysis was carried out using GraphPad Prism Version 6 software (GraphPad Software Inc.).