Assessing Antigenic Drift of Seasonal Influenza A(H3N2) and A(H1N1)pdm09 Viruses

Under selective pressure from the host immune system, antigenic epitopes of influenza virus hemagglutinin (HA) have continually evolved to escape antibody recognition, termed antigenic drift. We analyzed the genomes of influenza A(H3N2) and A(H1N1)pdm09 virus strains circulating in Thailand between 2010 and 2014 and assessed how well the yearly vaccine strains recommended for the southern hemisphere matched them. We amplified and sequenced the HA gene of 120 A(H3N2) and 81 A(H1N1)pdm09 influenza virus samples obtained from respiratory specimens and calculated the perfect-match vaccine efficacy using the p epitope model, which quantitated the antigenic drift in the dominant epitope of HA. Phylogenetic analysis of the A(H3N2) HA1 genes classified most strains into genetic clades 1, 3A, 3B, and 3C. The A(H3N2) strains from the 2013 and 2014 seasons showed very low to moderate vaccine efficacy and demonstrated antigenic drift from epitopes C and A to epitope B. Meanwhile, most A(H1N1)pdm09 strains from the 2012–2014 seasons belonged to genetic clades 6A, 6B, and 6C and displayed the dominant epitope mutations at epitopes B and E. Finally, the vaccine efficacy for A(H1N1)pdm09 (79.6–93.4%) was generally higher than that of A(H3N2). These findings further confirmed the accelerating antigenic drift of the circulating influenza A(H3N2) in recent years.


Introduction
Influenza A virus is a major cause of acute respiratory disease in humans and is responsible for~250,000-500,000 deaths annually worldwide [1]. Pandemic influenza A virus infection resulted in significant morbidity and mortality in 1918 (H1N1), 1957 (H2N2), 1968 (H3N2), and 2009 (H1N1) [2]. Subtypes of influenza A viruses are defined by the surface proteins hemagglutinin (HA) and neuraminidase, two major viral targets for the host immune system [3]. The HA protein of the influenza virus is cleaved by the protease enzyme in the host cells into two subunits: HA1 and HA2. The HA1 subunit plays a major role in binding to host receptor or neutralizing antibodies and represents major antigenic sites (defined as epitopes A, B, C, D, The Outstanding Professor of Thailand Research Fund (DPG5480002), Siam Cement Group, and MK Restaurant Company Limited. The authors thank the 100th anniversary Chulalongkorn University Fund for doctoral scholarship awarded to Nipaporn Tewawong. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors received funding from commercial sources, Siam Cement Group and MK Restaurant Company Limited. However, this does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. The authors also declare that no competing interests exist.
products were visualized on a 2% agarose gel and purified using the HiYield gel DNA fragment extraction kit (RBC Bioscience, New Taipei City, Taiwan). DNA sequencing was performed by First BASE Laboratories Sdn Bhd (Selangor, Malaysia).

Nucleotide sequence accession numbers
The HA sequences of influenza A(H3N2) and A(H1N1)pdm09 obtained in the 2010 season in Thailand had previously been deposited in GenBank (S2 Table), while the HA sequences of the influenza A(H3N2) (in 2011-2014) and A(H1N1)pdm09 (in 2011-2014) isolates were submitted to GenBank under accession numbers KP335865 to KP335981 and KP941680 to KP941741, respectively. Moreover, the HA sequences of reference and southern hemisphere vaccine strains from influenza A(H3N2) and A(H1N1)pdm09 viruses included in phylogenetic analysis were obtained from the GenBank and GISAID databases. Their accession numbers were also included in the S2 Table. Phylogenetic analysis and antigenic characterization The HA sequences were edited and assembled using SeqManPro (DNASTAR, Madison, WI). The ClustalX v.2.1 was used for the alignment of protein and nucleotide sequences [15]. The Akaike information criterion and maximum likelihood value indicated that the HKY+G model was the best fit model [16]. A phylogenetic tree of the HA1 coding nucleotide sequences was generated by Molecular Evolutionary Genetic Analysis (MEGA) version 6.06 [17] using a maximum likelihood tree by the HKY+G model with 1,000 bootstrap replicates; only bootstrap values over 50 were shown. The amino acid residues in the five epitopes (A-E) of A(H3N2) (A/Aichi/2/1968) and A(H1N1)pdm09 (A/California/04/2009) viruses were previously identified [18,19]. The relative amino acid frequency in the epitope of HA1 was performed using WebLogo [20].

Measurement of selection pressure
The selective pressure on encoding HA1 A(H3N2) and A(H1N1)pdm09 was examined by calculating the ratio of synonymous and non-synonymous substitutions (dN/dS, defined as ω) across lineage on a codon-by-codon basis. The individual site-specific selection pressure and ω were estimated using the single likelihood ancestor counting (SLAC) and fixed effects likelihood (FEL) methods contained in the HYPHY package [21]. All analyses utilized the Datamonkey online tool (http://www.datamonkey.org). The value of ω was estimated based on the neighbor-joining trees under the HKY85 substitution model. The significance level for a positively selected site by either SLAC/FEL or both methods was accepted at 0.1. p epitope is given by E = −2.47 × p epitope + 0.47 for influenza A(H3N2) virus and by E = −1.19 × p epitope + 0.53 for influenza A(H1N1)pdm09 virus [23]. The influenza A(H3N2) vaccine efficacy with p epitope = 0 is 47% as a perfect match between vaccine and virus [10]. For the influenza A(H1N1)pdm09 virus, the vaccine efficacy is 53% when p epitope = 0 [23].
Comparison The overall HA1 nucleotide identities among the A(H3N2) strains compared to the given vaccine strains over the period examined were >97%, while the amino acid identities were >96% ( Table 1). The nucleotide and amino acid similarities between A(H3N2) strains from the 2011 and 2012 seasons and A/Perth/16/2009 were >97.6 and >96.3%, respectively. Meanwhile, the nucleotide and amino acid similarities between the 2013 strains and A/Victoria/361/ 2011 were 98.7% and 97.7%, respectively. The A(H3N2) strains in 2014 were closely related to A/Texas/50/2012 (98.2% nucleotide and 96.9% amino acid identities).

Antigenic characterization
The receptor binding site (RBS) on the HA comprised several highly conserved amino acid residues (Y98, T133, W150, H180, and Q223; numbered according to HA1). Residues at the terminal sialic acid receptor binding sites (RBSs) of all A(H3N2) strains were I226 and S228, while all A(H1N1)pdm09 strains possessed D204 (numbered according to HA0). Additionally, differences in the residues on the A(H3N2) and A(H1N1)pdm09 HA protein were located on the antigenic sites, which comprised epitopes A to E (S3 and S4 Tables). We summarized the relative frequencies of the residues found on the dominant epitope domain on the HA1 of A (H3N2) and A(H1N1)pdm09 (Fig 3). Overall, the A(H3N2) strains displayed more diversity from the accumulated epitope mutations than the A(H1N1)pdm09 strains.

Selection pressure on A(H3N2) and A(H1N1)pdm09
Assuming that the influenza HA protein is subjected to selection pressure in order to evade the host cell recognition, the rate of change was assessed by the ω values in which ω < 1 meant that negative or purifying selection was present, ω = 1 when selection pressure was neutral, and ω >1 when there was positive selection [27]. Analysis showed that the overall ω values of the coding HA1 regions of A(H3N2) and A(H1N1)pdm09 were 0.34 and 0.31, respectively. Since the majority of residues in the HA1 domain showed ω < 1, this suggested that the amino acids in the HA epitope domain were under purifying selection. Although overall positive selection was not present, specific sites of positive selection were found using SLAC and FEL methods ( Table 2). Among the 329 codons in the HA1 domain of the A(H3N2) circulating strains, SLAC showed that only codon 33 (Q/R) was a positively selected site (ω = 3.78, P = 0.04).

Estimation of vaccine efficacy for A(H3N2) and A(H1N1)pdm09
To assess the effect of the accumulated mutations on the HA1 domain on the vaccine efficacy in a given year, the p epitope method was used to evaluate how closely the vaccine strain resembled the circulating strain (Table 3). Theoretically, when p epitope in the dominant epitope is higher than 0.19, the vaccine efficacy becomes negative [10].  (Table 4). In all, 12.3% of the strains obtained between 2010-2014 (10/81) possessed dominant mutation in epitope E at position 83, which gave an estimated worst-case vaccine efficacy against the virus of 93.4% (E = 49.5% of 53%, p epitope = 0) of a perfect match with the vaccine strain. In summary, HA1 sequences of A(H1N1)pdm09 from recent years showed antigenic changes mainly on epitope B and one or two amino acid mutations on other epitopes. Taken together, these results suggested that the past and current vaccine provided optimal protection against A(H1N1)pdm09 strains that circulated in Thailand.

Discussion
Since its emergence in 1968, influenza A(H3N2) strain has been the predominant circulating influenza subtype between 2011 and 2014 [28]. In contrast, the A(H1N1) subtype was first documented in Thailand in May 2009 and continued to circulate until 2010. Although the (H1N1) pdm09-like A/California/7/2009 has remained the recommended vaccine strain for the past several years, we identified changes on the HA1 of A(H1N1)pdm09 Thai strains belonging to epitopes B (S185T) and E (P83S). Additional observed changes at residues 97 (D97N) and 197 (A197T) previously implicated in the adaptive mutation or immune escape continued to persist (71% and 18%, respectively) [29]. In this study, the vaccine efficacy for A(H1N1)pdm09 strains of 79.6-93.4% was higher than that for the A(H3N2). This is concordant with previous studies, which estimated the vaccine efficacy using serologically based methods and suggested a moderate to high vaccine effectiveness for influenza A(H1N1)pdm09 during the 2010-2014 seasons [30,31].
We found that influenza clusters on the phylogenetic tree is mostly chronological. The influenza A(H3N2) strains found in Thailand belonging to at least two phylogenetic sub-clades Genetic evolution of influenza virus appears gradual, but antigenic changes were found to occur more abruptly [32]. For example, one single amino acid substitution in the case of N145K on the HA1 of A(H3N2) can characterize the difference between clades. It is also known that the antigenic variation of H3 occurs more frequently than H1. The average amino acid substitution rate of the HA protein is 3.6 per year for A(H3N2) and 2.45 per year for A (H1N1) [32][33]. One reason may be that more individuals are susceptible to the relatively novel A(H1N1)pdm09 strain, and therefore the weaker immnue pressure has resulted in the slow rate of the viral evolution.
New influenza variants are thought to drift considerably from the parental strain when they displayed four or more amino acid mutations on at least two epitope domains on the HA1 protein [34]. Alternatively, antigenic drift variants can result from a change in the antigenic site in combination with a mutation in the RBS, which interacts with the sialic acid on the cell surface [35]. We found that the HA1 sequences from the A(H3N2) strains during the 2011-2012 season possessed seven amino acid changes on four epitopes including epitope C (S45N, T48I, N278K, and N312S), epitope B (A198S), epitope D (V223I), and Q33R. In the 2013-2014 seasons, antigenic drift also occurred due to at least four amino acid mutations on epitope B combined with additional mutations on epitope A (R142G), epitope B (T128A), and epitope C (N278K). It is noteworthy that T128A had previously been observed in Fujian strain, which was associated with high mortality rate in children [36].
Further analysis of the HA1 from A(H3N2) strains revealed three positively selected codons (33, 144, and 198), suggesting that these sites were immune-escaped mutants. The N144D substitution was not unique to the strains found in Thailand as it was observed among isolates in Europe and Africa during 2010-2011 [37]. The resulting glycosylation at position 144 was previously implicated in the antigenic change in A/Fujian/411/02-like strains from the 2002-2003 seasons [38]. Compared to A/Perth/16/2009, two amino acid mutations involving A128T and N45S could effectively alter the glycosylation pattern, providing evolutionary advantage to the virus including more effective masking of viral epitopes, stabilization of polymeric HA structures, regulation of the receptor binding domain, and balancing the binding activity of HA with the release activity of neuraminidase [39]. This was evident when it was observed that the loss of the glycosylation site at 128 of HA1 was associated with loss of antibody recognition [40].
The vaccine efficacy between A/Perth/16/2009 vaccine strain and A(H3N2) strains circulated in Thailand in 2010 of 75.96% is consistent with a moderate vaccine efficacy reported for the trivalent inactivated influenza vaccines in 2010-2011 in Thailand [41] and a moderate protection against subtype-specific A/H3 reported in the U.S. [42]. During the 2011-2012 seasons, the dominant epitope change from A to C relative to the A/Perth/16/2009 subsequently resulted in a decline in the percentage of perfect-match vaccine efficacy (44.6% and 22.1%, respectively). Furthermore, the antigenic sites of the circulating A(H3N2) strains in 2013 and 2014 drifted from epitope C to B compared to the vaccine strains. In 2013, the reference vaccine strain had to be changed from A/Perth/16/2009 to A/Victoria/361/2011, which appeared to moderately improve the perfect-match vaccine efficacy (49.9%). This was consistent with the results from an epidemiological cohort study showing vaccine efficacy in the 2011-2012 (55%) and 2012-2013 (64%) seasons among Thai children [9]. Meanwhile, the vaccine strain chosen in 2014 (A/Texas/50/2012-like) did not improve the perfect-match vaccine efficacy, which was fairly low (approximately 24.9%). Moreover, the p epitope values for the majority of the HA1 sequences in 2014 (75%) was > 0.19 [10] and, therefore, the vaccine efficacy became negative. This has occurred in the past whereby the outbreak of the Sydney/5/79 strain yielded the value of p epitope of 0.238; hence, the vaccine efficacy was −17% compared with the 1997-1998 northern hemisphere influenza vaccine [43]. Taken together, the emergence of multiple circulating strains in 2014 contributed to the reduced vaccine efficacy in Thailand that year and was reflected in the weekly morbidity and mortality report from the U. S. Centers for Disease Control and Prevention, which suggested that the 2014-15 influenza vaccine strain A/ Texas/50/2012 was essentially ineffective against the circulating A(H3N2) strains [44].
There are several limitations in this study. Since there is no general consensus on the epitope regions for A(H1N1)pdm09, we estimated the antigenic drift and vaccine efficacy based on the mutation of the dominant epitope by mapping epitopes A-E from H3 onto the pandemic A/California/04/2009 strain. Our results therefore require validation using alternative models with differently defined epitope regions [45][46]. The greater number of A(H3N2) strains analyzed in this study may have contributed to more mutations observed on the epitope domains of the A (H3N2) than the A(H1N1). Finally, the assessment of vaccine efficacy relied on the comparison of the circulating influenza strains to the vaccine strains chosen annually, therefore this measurement is not absolute as antigenic diversity have not always been predictive of the vaccine effectiveness. In conclusion, continued influenza surveillance, molecular evolution analysis, and antigenic distance measurement of the dominant influenza A strains in circulation will help refine the interpretation of vaccine efficacy and improve the yearly influenza vaccine.