Passage of influenza A/H3N2 viruses in human airway cells removes artefactual variants associated with neuraminidase-mediated binding

Serological assays with modern influenza A/H3N2 viruses have become problematic due to the progressive reduction in the ability of viruses of this subtype to bind and agglutinate red blood cells (RBCs). This is due to reduced ability of the viral haemagglutinin (HA) glycoprotein to bind to the sialic acid-containing receptors presented by these cells. Additionally, as a result of reduced HA-mediated binding in cell culture, modern A/H3N2 viruses often acquire compensatory mutations during propagation that enable binding of cellular receptors through their neuraminidase (NA) surface protein. Viruses that have acquired this NA-mediated binding agglutinate RBCs through their NA, confusing the results of serological assays designed to assess HA antigenicity. Here we confirm with a large dataset that the acquisition of mutations that confer NA binding of RBCs is a culture artefact, and demonstrate that modern A/H3N2 isolates with acquired NA-binding mutations revert to a clinical-like NA sequence after a single passage in human airway epithelial (HAE) cells.


Passage of influenza A/H3N2 viruses in human airway cells removes artefactual variants associated with neuraminidasemediated binding
InTRoduCTIon Influenza A/H3N2 viruses have circulated in the human population since the Hong Kong pandemic of 1968 and currently co-circulate alongside influenza A/H1N1 viruses descended from the virus responsible for the 2009 pandemic. The seasonal influenza vaccine contains antigenic components from these two subtypes of influenza A, as well as from one or both of the currently circulating lineages of influenza B. Due to the rapid genetic and antigenic evolution of influenza, global surveillance of circulating viruses is continuous and prior to each flu season influenza viruses isolated from human infections worldwide are assessed to see if they have become antigenically distinct from the strains included in the previous season's vaccine. While routine sequencing of influenza haemagglutinin (HA) glycoproteins can help to predict antigenic properties, serological assays are essential to demonstrate whether genetic changes confer enough antigenic drift to necessitate a vaccine strain update. The core assay used to make this assessment is the haemagglutination inhibition (HAI) assay, which determines the specificity of antiserum by measuring its ability to block the interaction between a virus and sialic acid-containing receptors on the surface of red blood cells (RBCs). However, the current use of this assay for the A/H3N2 viruses is complicated by the accumulation of mutations in the HA globular head of these viruses since 1968, which have altered the receptor-binding properties of these viruses and led to a reduction in their ability to bind and agglutinate RBCs. Receptor binding changes have also resulted in reduced isolation rates of A/ H3N2 viruses in the Madin-Darby canine kidney (MDCK) cell line traditionally used to propagate influenza viruses [1,2], reducing the pool of viruses that can be assessed at all and further compromising strain selection. Alongside these changes in HA-mediated binding, the phenomenon of neuraminidase (NA)-mediated binding by acquisition of a D151G mutation in the NA of passaged A/H3N2 isolates has also been observed [3]. G151 receptor-binding variants show reduced sialidase activity [4] and rather than reaching fixation in a viral population, co-exist alongside the D151 genotype, which codes for an enzymatically active NA [5]. In MDCK-SIAT cell culture, a viral population with a mixture of D151 and G151 genotypes is fitter than either variant in isolation, but only when paired with a modern H3 HA with low receptor avidity [5]. Using next-generation sequencing (NGS), G151 variants have recently been shown to be absent from a small panel of nine circulating A/H3N2 viruses collected between 2013 and 2015, which subsequently acquired these variants upon MDCK cell passage [6]. This supports the conclusions of others that NA-mediated binding is a cell culture artefact. Here we analysed a larger set of clinical samples collected in the 2014-15 season to confirm the absence of genetic variants associated with NA-mediated binding. We also reveal that a single passage in human airway epithelial (HAE) cells of isolates bearing genetic variants associated with NA-mediated binding can revert them back to a clinical genotype.

nA-mediated agglutination of guinea pig RBCs is a result of a d151G mutation in the nA of A/ Sydney/71/2014
To investigate the phenomenon of NA-mediated agglutination of RBCs, A/H3N2 isolates that had been egg-passaged (n=29) or passaged in MDCK and/or MDCK-SIAT1 cells (n=22) were tested for the sensitivity of their HA titres to the presence of 40 nM oseltamivir (Fig. 1). Cell-passaged isolates showed significantly higher sensitivity than egg passaged strains, for example cell-passaged influenza A/ Sydney/71/2014 (hereafter called Sydney/2014) demonstrated the greatest drop in HA titre from 128 to 2 in the presence of oseltamivir. Sanger sequencing revealed the presence of a mixed GAT/GGT genotype at the codon for residue 151 of the NA gene in Sydney/2014, which was quantified by NGS as 78 % GAT and 22 % GGT, resulting in 78 % aspartic acid (D) and 22 % glycine (G) amino acids at this position. Variants at residue 151 have previously been shown to alter the activity of NA, with the G151 resulting in a sialic acid-binding, sialidase-inactive phenotype [3,4]. We generated variants of Sydney/2014 with either a D151 or G151 NA together with the HA of Sydney/2014 and the six remaining genes of A/Puerto Rico/8/34 (PR8) by reverse genetics. The G151 virus was generated and propagated by the addition of exogenous bacterial NA. We tested the sensitivity of the HA titres of the RG viruses to oseltamivir. Sydney/2014 with purely NA D151 showed no drop in titre in the presence of oseltamivir, while a purely NA G151 virus demonstrated an HA titre of 64, which was entirely abrogated in the presence of oseltamivir, demonstrating that this mutation conferred NA-mediated binding (Table 1).

Recent clinical A/H3n2 isolates do not contain the minor nA variants associated with nA-mediated binding
The appearance of NA variants in A/H3N2 viruses that mediate the agglutination of RBCs has implications for carrying out and interpreting the results of serological assays, but these variants may not be of clinical significance. To investigate whether such variants are found in circulating viruses, NGS data from 251 clinical A/H3N2 samples collected during the 2014-15 influenza season in the UK were assessed for the presence of NA variants at codon 151. The presence of variants was also assessed at codon 148. A T148I recombinant NA has demonstrated reduced catalytic   (Fig. 2). Selection of variants still occurred in MDCK-SIAT1 cells, but to a lesser extent (Fig. 2). T148K reached 20 % in one harvest and in the other T148K reached 8 % with an accompanying D151G variant at 1 %. Passaged viruses did not acquire any fixed mutations in HA or NA. No minor variants were detected in the HA receptor-binding site (RBS) or at positions other than 148 or 151 in the NA catalytic site.

Historical H150R mutation was not a prerequisite for the acquisition of nA 151 variants
It was shown recently [9] that an H150R mutation in the NA of A/H3N2 viruses was responsible for the NA-mediated agglutination of turkey RBCs. This mutation became fixed in the majority of circulating viruses around 2008. We were interested to see whether this was a prerequisite for the acquisition of NA 148/151 mutations that mediate haemagglutination in the guinea pig RBC assays that are now routinely carried out. Viruses were generated with the internal genes of PR8, the HA of Sydney ) and the G151 variant whose titre was below the level of detection of the plaque assay. Seventytwo hour harvests from the mixed D151: G151 infections were sequenced and showed that whereas both MDCK and MDCK-SIAT1 cells retained mixtures of genotypes at codon 151, HAE cells selected for a population in which G151 variants were undetectable by Sanger sequencing (Fig. 4b).

A single passage in HAE cells removes variants associated with nA-mediated binding
Based on the observation that genotypic variants associated with NA-mediated binding were not replicated in HAE cultures, we hypothesized that passage through these cells could remove these variants from MDCK/MDCK-SIAT1-passaged viruses, allowing their subsequent use in serological assays. HA assays were conducted on A/H3N2 isolates from 2013-2016 (n=38) and a panel of seven isolates spanning subclades 3C.2a, 3C.2a1 and 3C.3a were chosen based on the sensitivity of their HA titres to oseltamivir (    Fig. 5). Post-HAE passage, in the majority of replicates (13/24) these laboratory-acquired mutations were undetectable (Fig. 5). Mutations elsewhere in the enzymatic site of NA or on the HA globular head, particularly in the RBS, may result in changes to the HA : NA functional balance. No non-synonymous mutations arose in the HA RBS of HAEpassaged isolates and the highest frequency mutation in the NA enzymatic site other than at 148/151 was G405A at 3.6 % in Sydney/2014 (Fig. 6).

Clinical genotype is maintained upon subsequent passage in MdCK-SIAT1 cells for some isolates
The use of HAE cells to return passaged A/H3N2 isolates to a clinical sample-like genotype proved effective for the majority of viruses in the panel. However, HAE cells only yielded a small volume of virus harvest, which must be scaled up in order to make using these cells a viable method for the production of virus stocks that can be shared between laboratories and used for serological assays. Two of the HAE harvests from each starting isolate were used to infect MDCK-SIAT1 cells at 0.01 p.f.u./cell, harvested after 72 h and sequenced by NGS. Fig. 7

dISCuSSIon
This work demonstrates that NA-mediated binding of influenza A/H3N2 isolates in serological assays, which arises as an artefact of MDCK/MDCK-SIAT1 cell passage, may be removed by passage in HAE cells. MDCK cells are commonly used for the propagation of influenza viruses; however, poor growth, reduced isolation rates and the selection of adaptive mutations in the HA and NA surface proteins when recent A/ H3N2 viruses are passaged in these cells prohibits their use [2,10,11]. The use of the MDCK-SIAT1 cell line, modified to express more of the α-2,6linked sialic acid-containing receptors used by human-adapted influenza viruses, has helped to tackle these challenges [1,2]. However, adaptive mutations may still be selected for in MDCK-SIAT1 cells and the infectivity of modern A/H3N2 viruses may be compromised when passaging at multiplicities of infection that prevent them from arising [8,12].
Cell passage of A/H3N2 viruses can select for mutations in NA, such as D151G, which confer receptor-binding properties with a concomitant reduction in sialidase activity [3,4]. Such artefactual NA variants were recently shown to be absent from a set of nine A/H3N2 clinical samples from 2013-15, which later selected for them in cell culture [6].  (Fig. 2) . Interestingly, in MDCK cells T148I and D151G rose separately to >50 % frequency in the viral population in the duplicate serial passages of this virus, demonstrating the stochastic nature by which de novo mutations arise and are then positively selected for to carry out the function of NA-mediated binding (Fig. 2). MDCK-SIAT1 cells selected for 148 and/or 151 variants to <20 % of the viral population (Fig. 2), supporting the idea of their use in maintaining the clinical sequence of A/H3N2 viruses more faithfully than MDCK cells, while highlighting that adaptations can still occur [8,12].
Mögling et al. [9] identified that an H150R mutation hat fixed in A/H3N2 viruses around 2008 confers the ability of NA to agglutinate turkey RBCs. Sydney/2014 carries R150 but RG variants demonstrate that it is the identity of the amino acid at position 151 that confers the sensitivity of its HA titre to oseltamivir when using guinea pig RBCs (  (Fig. 3). Interestingly, the HA of Sydney/2014 was able to drive the selection of NA 148/151 variants in the NA of Sydney/2014 (Fig. 2) whereas the older HA of Wisconsin/2005 was not (Fig. 3). This difference in the ability of HAs from different eras to select NA mutations shows that it is the declining ability of the HA of A/H3N2 viruses to bind cellular receptors in tissue culture that drives the selection of NA variants to assist with receptor binding.
Xue et al. [5] demonstrated that virus populations with a 50 : 50 mixture of the D151 and G151 genotype had a growth advantage in MDCK-SIAT1 cells over viruses with either genotype alone. However, for Sydney/2014 such cooperation was not observed in MDCK-SIAT1 cells but did occur in MDCK cell culture (Fig. 4). The 2007 virus used by Xue et al. [5] found an equilibrium at approximately 50 : 50 D151 : G151 after serial passage experiments in MDCK-SIAT1 cells, suggesting that this was the optimal ratio to achieve a balance between HA and NA functions in this cell line. Subsequent infection with a 50 : 50 mixture of variants of this virus verified a cooperative effect in MDCK-SIAT1 cells. In contrast, Sydney/2014 selected for a population skewed towards a D151 genotype during serial passage in MDCK-SIAT1 cells (Fig. 2). This suggests that the 50 : 50 mixture of genotypes used in Fig. 4 was sub-optimal in terms of the HA : NA functional balance preferred by this virus in MDCK-SIAT1 cells and explains why complementation was not observed. In MDCK cells, passage of Sydney/2014 selected for a mixture closer to 50 : 50 D151 : G151 (Fig. 2) and cooperation was observed at this ratio in this cell line (Fig. 4). These findings highlight that HAs with subtly different binding properties will likely select for different D151 : G151 NA ratios and that this balance will also depend on cell type due to differences in receptor types between lines. The need for a balance between the functions of HA and NA is essential for influenza virus fitness and perturbations to this balance are rapidly redressed by mutations to modulate the binding function of HA and/or the enzymatic activity, stalk length or efficiency of cell surface trafficking of NA [15][16][17][18][19][20]. While under the selection pressure of MDCK cell culture, the acquisition of a mixed genotype at NA 148/151, which confers receptor binding while maintaining some sialidase function, is another example of a mechanism by which influenza viruses may find a functional balance between HA and NA.
Here, in an attempt to return cell-passaged H3N2 isolates to a clinical genotype that lacks NA 148/151 receptor-binding variants, a panel of seven isolates was passaged in HAE cells.
Genotypes associated with NA-mediated receptor binding were reduced in all three HAE-passage replicates for six of the eight H3N2 isolates tested and entirely removed in two of these six (Fig. 5). Interestingly, it was the isolates previously used as prototype strains for A/H3N2 vaccines, A/ Hong Kong/4801/2014 and A/Switzerland/9715293/2013, whose variants at NA 148 were maintained through HAE passage (Fig. 7). This may be a result of the properties of the HAs of these viruses. For the other isolates, seven out of eight HAE passage replicates that were free of NA 148/151 variants remained free of these variants after a subsequent passage in MDCK-SIAT1 cells (Fig. 7). Infection in the presence of mucus did not result in more selection against NA-mediated binding variants for Sydney/2014 (Fig. 5) or a significant difference in titre between the isolates harvested from the two culture conditions. This finding suggests that in this experimental model of airway infection through mucus, the mucus layer did not significantly inhibit phenotypically mixed particles from accessing the cell surface. Phenotypically mixed virions express both D151 sialidase and G151 binding NA molecules on their surface and therefore would still have some ability to cut through mucins in the mucus barrier and enter cells. This allows the delivery of D151 or G151 genomes to generate single phenotype particles, or in the case of co-infection of a cell, both genomes to generate more phenotypically mixed particles carrying either D151 or G151 genomes. However, the reduced frequency of NA-mediated binding variants in the majority of replicates showed that HAE cells exerted negative selection pressure in multi-round replication and demonstrated the utility of these cells in returning A/H3N2 isolates to a more clinical genotype.

Virus titration
Haemagglutination assays were carried out on virus isolates using 0.7 % guinea pig RBCs in PBS (0.1 % BSA, with or without 40 nM oseltamivir carboxylate) in 96-well V-bottom plates. Plates were incubated for 1 h at RT before haemagglutination was determined visually.

Virus reverse genetics
The HA and NA genes of Sydney/2014 were extracted and the NA gene modified by the cloning and recombinant PCR techniques described in Nicolson et al. [22] to carry a GGT codon for glycine at position 151. A 12-plasmid rescue system in Vero cells was used to generate isogenic viruses with the HA and D151 or G151 NA of Sydney/2014 together with the 6 internal genes of A/Puerto Rico/8/34 (H1N1) as described by Nicolson et al. [22].

Sequencing
Sanger sequencing of PCR products or gene-containing plasmids was carried out by Eurofins Genomics. For NGS, the one-step RT-PCR method of Zhou et al. [23] was used to generate PCR products for all eight influenza gene segments. RT-PCR reactions were performed in triplicate for each viral RNA sample. DNA library preparation was carried out using the Nextera XT protocol (Illumina) according to the manufacturer's instructions and samples were run on an Illumina MiSeq platform at the National Institute for Biological Standards and Control (NIBSC). Downstream processing of . fastq. gz files at the NIBSC was performed using R and Geneious 10.0 software. Data processing at Colindale was performed with in-house protocols on Galaxy. The GISAID accession numbers of sequences derived from clinical samples containing A/H3N2 influenza viruses are as follows: EPI_ISL_168782; EPI_ISL_186565 to EPI_ISL_186616; EPI_ISL_188738 to EPI_ISL_188863; EPI_ISL_192193 to EPI_ISL_192263; EPI_ISL_240905.

Funding information
Funding for this work was provided by NIBSC.