Molecular and viral characterization of the hemagglutinin protein (HA) from different hosts has increased in the last three decades, in response to three worldwide outbreaks of influenza in the years 1918, 1957, and 1968 [1]. The H3N2 antigenic subtype responsible for the 1968 pandemic was first isolated in July 1968 in Hong Kong, and supplanted the H2N2 virus responsible for the 1957 Asian flu pandemic[2,1].

Bioinformatics and computational approaches towards molecular understanding of HA have largely focused on the determination of mutation levels and evolution of the HA gene, and identification and prediction of antigenic variants of H3N2 by locating potential immunodominant positions on the HA protein. Phylogenetic analysis of H3N2 genomes illustrates that the H3N2 virus is composed of multiple and distinct clades, which exhibit genetic variation by interacting with minor lineages through reassortment events [2]. Whole-genome alignments, statistical analysis with construction of evolutionary trees were used to identify locations of mutations within H3N2, predict their yearly frequency, and determine modes of antigenic drift and positive selection [2]. Using a parsimonious tree to map the HA1 domain of 254 H3N2 viral genes, Fitch and coworkers determined that HA1 evolves at an average rate of 5.7 nucleotide substitutions/year, and indicated the presence of six hypervariable codons of the HA gene which accumulate replacement substitutions at a rate that is 7.2 times that of other codons [3]. Some studies have concluded that H3 hemagglutinin gene exhibits positive selection in key regions of the HA molecule such as the receptor-binding site and antibody-binding sites [4], which result in new antigenic and resistant strains. Several studies used bioinformatics approach to predict antigenic strains of the H3N2 virus [5-7]. One study generated a model based on 131 positions in the five antigenic sites of the protein, and which could predict antigenic variants of H3N2 with an agreement rate of 83% to existing serological data [5]. Later studies also identified twenty amino acids positions, which are potential immunodominant positions and contribute to antigenic difference between strains [6].

To the best of our knowledge, few bioinformatics publications have addressed motif search in segments of the H3N2 genome where mutations have been observed. A recent study by Ahn and Son [7] aimed to detect relative synonymous codon usage (RSCU) and codon usage patterns (CUP) in HA and Neuraminidase (NA) from H3N2, H9N2, and H5N1 subtypes within human, avian, and swine populations. They established a unique CUP for each subtype, and observed a possible divergence within human H3N2 isolates based on their synonymous CUPs. A study published earlier this year [8] has focused specifically on the H3N2 subtype, using nucleotide co-occurrence networks of human H3N2 strains to predict H3N2 evolution. However, analysis of H3N2 nucleotide and protein genomes to discover patterns and motifs yet remains to be elucidated. In this study, we report motifs and assign potential functional characteristics within the HA protein sequences of the gene of H3N2 human influenza isolates from Hong Kong between 1968 and 1999. We identify motifs within the HA protein, and interrelate these motifs with amino acid substitutions frequency, co-mutating pairs, potential post-translation modification sites, antigenic sites, receptor-binding sites. We focus our analysis on motifs with varying mutation frequency and correlate the variable motif with a high number of potential post-translational modification sites that overlap antigenic and receptor binding sites. We speculate that mutation in these motifs results in the emergence of viral strains that are highly pathogenic and has the intrinsic character to overcome that host defense mechanisms.

Figure 1 Selected 14 MEME Blocks in the HA1 consensus sequence from 1968–1999. Combined block diagram of non overlapping sites with p value < 0.0001 was generated from the MEME server which are common to the entire data set, with the exception (more ...)

Table 1 MEME blocks positions, size and genetic distance

Table 2 Amino acid substitutions in the different isolates from 1969–1999 used to extrapolate the genetic distance in the different MEME blocks

Figure 2 Number of aminoacid substitutions in each MEME block over the period from 1968–1999. (A) Bar graph of amino acid substitutions within MEME blocks for each of the years. (B) Behavior of the substitutions in MEME block 7; frequency of amino acid (more ...)

In an attempt to establish the relationship between blocks and amino acids substitutions over the time period between 1968–1999, a line graph was drawn to examine the mutation rate of each of the MEME blocks, in order to infer the evolutionary behavior of the motifs (i.e. whether they were acted upon by positive selection or neutral genetic drift evolution). The frequency of amino acid substitutions within the highly variable MEME block 7 (Figure 2b) largely follows the occurrence pattern of substitutions within the entire protein (Table 2), reaching a peak in 1980, which corresponds to a year with a high number of mutations in the alignment, and following a similar zenith in 1985. However, for the intermediately variable MEME block 2, not all the mutations within each year of the alignment occur in the block, resulting in a zigzag behavior from 1982 onwards (Figure 2c). Some blocks only undergo amino acids substitutions in one or two years of the cohort, as is the case with motifs 8 and 12 (data not shown). MEME block 5 undergo amino acids substitutions from 1968 to 1984, then remain conserved after this period (data not shown). Similarly, MEME block 10 is conserved after 1984 with an exception of an amino acid substitution in 1992 (data not shown). Additionally, certain blocks remain conserved for a few years of the cohort, but undergo amino acids substitutions towards the later years of the study. Notable examples include blocks 6, 9, and 13 (data not shown). The MEME program lists the HA MEME blocks in descending order based on their e-value, as such, MEME blocks 2 and 7 are quite significant and plausible for further analysis.

To confirm these finding, we correlated hot spots of variability with MEME blocks, using an entropy plot of the HA alignment (Figure 3). Hot spots of variability are clustered around amino acid position 140–190, and 200–240. Through out this study, we define a hot spot cluster as a 40 amino acid block containing more than 35% of amino acid substitutions. The first part of hot spot cluster I between amino acid position 140–154, is included within MEME block 7 (130–170). The second part of hot spot cluster I, between position 170–180, overlaps MEME block 11 entirely (171–178) and with one of the repetitive MEME block 2 (179–207). Hot spot cluster II overlaps entirely MEME block 13 (209–214) and almost entirely MEME block 3 (215–249). The two significant hot spots of variability were confirmed by looking at conserved regions generated by BIOEDIT, with a minimum length of 15 amino acids and maximum entropy 0.2, and this region did not overlap with the conserved region analysis (data not shown).

Figure 3 Entropy plot of the protein consensus ClustalW alignment. Amino acid positions that do not exhibit any changes over the years have entropy of 0, whereas positions of high variability are represented by peak in the plot. Two hot spots of variability were (more ...)

Table 3 Positions of potential post-translational modification sites

Of the 12 ASN glycosylation sites found under PPSearch 7 ASN glycosylation sites have been have been cross-referenced to potential sites of HA in the Uniprot KnowledgeBase, UniProtKB/Swiss-Prot Entry Q91MA7. Of these 7 ASN glycosylation, 5 remain conserved in all years of the data set. Interestingly, 4 ASN glycosylation sites noted by Skehel and co-workers [11] overlap our 2 prominent MEME blocks; 4 ASN glycosylation sites (amino acids 24–27, 38–41, 181–184 and 499–503) overlaps MEME 2 block, and 3 overlaps MEME 7 block (amino acids 138–141, 142–148 and 149–152).

Additionally, 9 of the 14 N-myristylation sites are in MEME blocks 1, 2 and 7. Four sites overlap with MEME block 7, three sites with MEME block1, and two sites with MEME block 2. Interestingly, some of these post-translational modification sites are conserved over the years as is the case with the majority of the phosphorylation sites (>70%), while more than 50% of the glycosylation and myristylation sites are observed in selected years (Table 3). Experimental studies need to be performed to confirm these potential post-translational modification sites.

Figure 4 Frequency of specific potential post-translational modification (prosite) motifs implicated in each of the MEME blocks. MEME block 7 has the highest number of post-translational modification sites, followed by MEME block 2, 1 and 3 respectively. High (more ...)

CKII sites were detected in MEME blocks 1, 2, 5, 7, 9 and 12; MEME blocks 1, 5 and 9 CKII sites have zero entropy. Unlike other MEME blocks, nearly all of CKII sites at MEME block 2 and 7 have non-zero entropy. One CKII site (position 205-entropy value 1.2) at MEME block 2 is also involved in the co-mutating pair (see below). These results illustrates that despite the high number of potential CKII sites at the highly conserved MEME 1 these sites remain conserved (Figure 5a) and the variable MEME block 2 and 7 undergo amino acid substitutions in CKII sites.

Figure 5 Average entropy of specific post-translational modification sites in each of the MEME blocks is demonstrated using boxplot. (A) Average entropy of potential CKII phosphorylation sites in the MEME blocks. Blocks 1, 5 and 9 have zero entropy at all CKII (more ...)

PKC sites were detected in MEME blocks 1, 2, 3, 4, 5, 6, 7, 10 and 11. The conserved MEME blocks 1 and 4 posses PKC sites with zero entropy. The majority of MEME blocks 2 and 3 PKC sites have zero entropy. One amino acid position at MEME blocks 2 and 7 posses the highest entropy of all of PKC's sites. Unsurprisingly, none of the PKC sites at MEME block 11 have zero entropy. The highly variable MEME block 11 has the highest average PKC entropy followed by MEME block 7 (Figure 5b). Four of the PKC sites at MEME block 7 are a part of the co-mutating pairs (see below).

ASN glycosylation sites were detected in MEME blocks 1, 2, 4, 5, 6, 7 and 9. MEME blocks 4 and 5 have zero entropy at all of their ASN sites. MEME block 2, 6 and 9 have nonzero entropy at the majority of their ASN sites. MEME block 1 and 7 are the only blocks with the majority of their glycosylation sites possessing nonzero entropy. Surprisingly, the conserved MEME block 1 also contains the amino acid (position 99) with the highest entropy (Figure 5c); this position is also the amino acid participating in the co-mutation pairs (see below). Additionally, one of the highly variable MEME block 7 N-glycosylation site is also involved in the co-mutation pairs (see below).

Myristylation sites were detected in MEME block 1, 2, 4, 7, and 9. MEME block 1, 2, 4, and 9 have the majority of their myristylation sites possessing zero entropy, in fact all myristylation sites at MEME block 4 have zero entropy, while all but 1 and 2 sites in MEME block 1 and 2, respectively have nonzero entropy (Figure 5d). One of the myristylation sites at MEME block 1, with a relatively high entropy (0.87), is involved in co-mutating pairs (see below).

Table 4 List of antigenic sites observed in the hemagglutinin structure.

In addition, we correlated the receptor binding sites described by Skehel and Wiley (2000) with MEME blocks. Interestingly, 4 of these receptor binding sites overlap the variable MEME block 7 and the intermediately variable MEME block 2 (Table 5). The receptor binding sites described by Skehel and Wiley (2000) and their overlapping MEME motifs 1, 2, and 7 are presented in Table 5.

Table 5 Position of receptor binding sites and their overlap with MEME blocks

Based on overlapping MEME blocks with hot spots, frequency of amino-acid substitutions, potential post-translational modification sites, receptor-binding sites and antigenic sites we mapped MEME blocks 1, 2, 3 and 7 onto the 3D hemagglutinin structure determined by Fleury and co-workers [13]. Antigenic sites A-D were also mapped for comparison and clarity [11]. Mapping MEME blocks 1, 2, 3 and 7 onto the existing 3-D hemagglutinin structure revealed that these blocks lie on the surface of the protein (Figure 6), specifically on the characteristic 8 beta antiparallel strands of the protein.

Figure 6 Graphical representation of MEME blocks and antigenic sites on the 3-D hemagglutinin structure. The HA1 and HA2 are represented in yellow and blue, respectively. A) MEME blocks on HA: MEME2 (Magenta), MEME7 (Red), MEME3 (Bright Green), MEME1 (Orange (89–129 (more ...)

Table 6 Co-mutating pairs and their position with respect to MEME motifs.

As opposed to previous molecular and computational approaches to understanding the dynamic nature of the human H3N2 influenza strain, our approach is one of few that attempts to understand and determine the functional importance of variable and conserved motifs in the hemagglutinin protein over time. To the best of our knowledge, this is the first study that addresses different regions in detail, and recognizes novel motifs and identifies their key functional significance with respect to potential post-translational modification sites, co-mutating amino acid pairs, antigenic and receptor binding sites.

In this study we have utilized 17 HA consensus sequences generated from 32 Hong Kong H3N2 isolates spanning the years from 1968 and 1999. We identified 14 MEME blocks, with the clustering of blocks 1, 2, 3 and 7 between positions 85–250 and 430–550 (Figure 6). We correlated the MEME blocks with rates of amino acid substitution and genetic distance. We also utilized entropy plots to determine the clustering of hot spot variability sites. We determined potential post-translational modification sites and correlated their positions and frequencies to MEME blocks, frequency of amino acid substitutions, antigenic sites and receptor binding sites. Out of the 14 MEME blocks, MEME blocks 1, 2 and 3 co-occur more than once within the HA protein and MEME block 7 is a single block. These blocks have different amino acid substitution frequency and encompass different hot spot clusters, post-translational modification sites, antigenic sites and receptor-binding sites. Of these highlighted blocks, MEME 2 had multiple interesting characteristics. This block (29 amino acids) is repeated three times at positions 14–42, 179–207 and 478–506 of the HA protein, and was characterized as an intermediate mutation frequency block (Figure 1). The repetitive nature of this motif could represent multiple binding pockets and could infer specificity to different proteins. Alternatively, such repetitive motif in the HA1 and HA2 subunits suggest common function in the 2 subunits possibly in guiding receptor binding and membrane fusion. A time course analysis to determine the frequency of substitution over the years was performed and lacked a distinct pattern in its amino acid substitution resulting in a zigzag behavior from 1982 onwards (Figure 2c). Additionally, MEME block 2 had one of the highest post-translational modification frequency; having the highest ASN-glycosylation frequency. It was previously reported that the addition of new oligosaccharides to the HA of the H3N2 viruses contributes to the virus ability to elude antibody pressures by changing its antigenic potential [15]. Alterations in HA glycosylation may affect NK cell recognition of influenza virus-infected cells [16]. Additionally, recently circulating avian influenza viruses (H5 and H9 subtypes) mutate at selected N-linked glycosylation sites [14].

MEME block 2 also encompasses the entire length of antigenic site B, and 1/5 of antigenic sites C and D in HA are present in this block (Table 4). Three receptor binding sites overlap this block (Table 5). A high number of co-occurring pairs of mutation was also observed in this block (Table 6). Mutation of glycosylation sites near receptor binding sites of HA1 was proposed to be an adaptation mechanism of the H7 viruses to a new host [18]. These associations suggest that MEME block 2 is a dynamic block in this protein that contributes to the ability of HA1 to mutate, modify its activity by post-translational modification, enhance pathogenicity by mutating receptor binding sites and escaping the host immune response by mutation in antigenic sites.

Additionally, we have identified MEME block 7 (41 amino acids) at position 130–170 (Table 1) as high mutation frequency block (Figure 1). Contrary to MEME 2 block, MEME block 7 revealed a peak frequency of substitution in 1980, corresponding to one of the years with a high mutation rate and therefore this block largely follows the occurrence pattern of substitutions within the entire protein (Figure 2b). However, the overlap between this block and one of the largest hot spots of variability revealed by the entropy plot, namely, the second cluster of hot spots, indicates that increased numbers of mutations within this block is not coincidental (Figure 3). MEME block 7 contained more than 35% of co-mutating pairs (Table 6). This block had the highest post-translational modification frequency (Figure 4), with the highest number of N-myristylation sites (Figure 5b). The entire length of antigenic site A is contained within MEME block 7 (Table 4) and therefore its rapid mutation is a mechanism by the virus to hide from the immune system.

The prevalence of post-translational sites in MEME blocks of high variability, and the lack of conservation observed within post-translational modification sites indicate their importance in sustaining the virus against environmental factors, contribution to viral spread and pathogenicity, and ultimately increasing viral virulence. Increased mutations within ASN-glycosylation sites between our 1968–1999 cohort, is consistent with previous studies, which suggest that increased glycosylation sites attenuates H3N2 viral activity [14]. It was previously reported that the addition of new oligosaccharide to the HA of the H3N2 virus contributes to the virus ability to elude antibody pressure by changing its antigenic potential [15]. Indeed, several studies have addressed the effect of glycosylation on the binding affinity of HA with sialic acid (SA)-containing receptors [16]. A study of reassortment viruses with different H3 HA on naïve mice has shown that, when matched to HK68 NA molecules, viruses with HK68 (7 potential sites) were more virulent than viruses with 12 potential sites (Pan99), and required 3 logs less virus to kill the mice [19]. Another study showed that mutant HA with 3–6 glycosylation sites decreases receptor-binding activity [15]. Additionally, a balance of glycosylation is needed to mediate the interaction of HA and NA molecules for receptor binding activity and viral release [15].

The importance of other post-translational sites that we have observed, such as myristylation sites, is seconded by several studies on the biological role of oligosaccharides and lipid modifications of proteins involved in protein translocation. Myristylation is one of 3 protein lipid modifications which are evolutionarily conserved in plants, animals, and fungi, and which aid in targeting proteins to the plasma membrane and other sub-cellular compartments [16]. Our study of several potential myristylation sites in MEME block 7 and the exhibition of high variability in these sites imply a mechanism by the virus to escape from neutralizing antibodies. On the other hand, the abundance of myristylation sites in the conserved MEME block 1 and their conservation in the years studied, in addition to its overlap with a single receptor binding site infers block 1 importance in selective and specific receptor binding and host cell attachment and infers conservation of essential functions through evolution. Unsurprisingly, minimal post-translational sites were observed in this un-variable block 1. However, the few glycosylation sites observed in MEME block 1 are not conserved and in fact contains the ASN site with the highest entropy and its involvement in the co-mutating pairs suggests specific and selected base pair substitutions over the years. The relatively high co-mutating pairs in this block remain unexplained. Interestingly, the minimal overlap of this conserved site with previously reported antigenic sites [12] is also in agreement with the conserved nature of this site.

The HA protein is on the surface of the influenza particle and is involved in receptor attachment and binding and antigenic determinants. This study reports unique regions in the protein that undergo either high mutation rates to possibly acquire new characteristics by post-translational modification or remain conserved for specific viral functional characteristics. Such changes are expected to enhance the ability of the virus in receptor binding, increase its infective state and escape the host immune response by modification of its antigenic sites. The dynamic evolution of a potentially functional motifs of the HA protein in H3N2 Hong Kong Influenza A virus strains, as revealed by bioinformatics analysis, paves the way for future experimental analysis to determine the significance of these post-translational modification sites and the effect of these alterations on receptor binding and antigenic determinants functions. Because of the worldwide flow in the H3N2 seasonal virus strain and the recently proposed limited mutation of the out of region circulating virus [22] further experimental analysis to appreciate the importance of the variable MEME motif 2 and 7 and the conserved MEME motif 1 in viral pathogenesis is required.

Consensus sequences were submitted into the PPSearch (Protein Motifs Search) [19] tool available at the European Bioinformatics Institute website. This revealed several post-translation modification sites within the HA1 and HA2 domains. Findings were compared to other motif finding applications including PROSITE under the Expasy Server, PSite, and the ELM database. MEME blocks were also submitted into MAST to determine their functional significance [20]. Additionally, the 1968 consensus sequence was queried against the BLOCKS [21] and PRINTS [21] database to check for the existence of known protein motifs.

HA: Hemagglutinin; NA: Neuraminidase; MEME: Multiple Em for Motif Elucidation; RSCU: Synonymous codon usage; CUP: Codon usage patterns; CKII: Casein kinase II; ASN: asparagines glycosylation; PKC: Protein Kinase C phosphorylation site; Myristyl: myristylation; PPSearch: Protein Motifs Search

The authors declare that they have no competing interests.

DMAG is a major contributor in the material collection, data analysis and implementation, and writing of the manuscript. MME helped in guiding the study design, implementation and analysis of the data and revised the manuscript. MW helped in the analysis of the correlation between post-translational modification and MEME motifs and generated figures 4 and 5. RS contributed in the initial idea, design and guiding of the project and contributed extensively to the analysis and interpretation of data into the formatted manuscript with extensive revision of the analysis and re-writing of the manuscript to elaborate on its scientific content.

All authors read and approved the final manuscript.

This study was supported by a grant from the Yousef-Jameel Science and Technology Research center funding to RS.