We first applied the AI metric to human-adapted H1N1 and avian H2 subtypes for two reasons. In the former case, we tested the ability of AI values to discriminate the 1918 and 2009 pandemic HAs from the seasonal strains. In the latter case, we validated AI's potential to highlight the conservation of antigenic sites in avian H211
. For H1N1, the HA of the human-adapted strains were compared to 1918 pandemic H1N1 HA (A/South Carolina/1/18) and the characterized H1 antigenic sites Sa, Sb, Ca, Cb19,20
were used to calculate AI (Methods
). The AI values clearly discriminate the reemerging swine-origin HA of 2009 H1N1 pandemic from the seasonal H1 based on the antigenic identity to the 1918 pandemic H1N1 HA (). The reemerging swine-origin HA of the 2009 H1N1 pandemic and those that circulated during the 1918–40 period are characterized by AI values > 70% and markedly differ from the strains that circulated during 1940–2008 (varies from 48% to 77% with an average of 55%). Two ‘classical' swine viruses (data points marked by black arrows), A/New Jersey/76 (H1N1) and A/Wisconsin/4754/1994 (H1N1), isolated between 1940–2008, also have high AI value and are genetically distinct when compared to the main cluster of human influenza viruses circulating in that period. Both viruses are known to have caused human infections following pig-human interspecies transmission. The A/New Jersey/76 influenza virus is reported to have caused respiratory illness in 13 soldiers with 1 death at Fort Dix, New Jersey21
. The A/Wisconsin/4754/1994 virus was recovered from a 39 year-old man who came in close contact with experimentally infected pigs22
. For the H2 subtype, the HA of the avian H2 strains were compared to the 1957–58 pandemic H2N2 HA (A/Albany/6/58(H2N2)) and the antigenic sites I-A, I-B, I-C, I-D, II-A and II-B23
characterized by hybridoma antibodies generated in BALB/c mice were used to calculate AI. Consistent with the findings of a previous report11
, the AI values indicate that the antigenic sites of the 1957–58 pandemic H2N2 HA are conserved in circulating avian H2 influenza viruses ().
In fact, the antigenic sites of the majority of avian H2 viruses in circulation are 100% identical to the 1957–58 pandemic H2N2 HA (). The conservation of antigenic sites in swine H2 influenza could not be assessed using this method due to lack of sequence information (H2N2 viruses do not circulate in swine; indeed, infection of swine with H2 viruses is rarely recorded). Similar to H1 subtype, the evolution of human H2 is characterized by steady antigenic drift leaning away from the pandemic strain. Although the majority of the viral strains that circulated during the immediate post-pandemic period 1957–68 have AI values > 70%, viral strains with AI ~ 60% appeared after 1967 (). It is reasonable to expect that the AI values would have decreased further had H2 continued to circulate in human population as a seasonal virus after 1968. The above analyses using H1 and H2 subtypes suggest that viruses carrying pandemic HA-like genes can be distinguished from seasonal viruses using a cutoff value AI ~ 70%.
Trends in antigenic evolution over time.
In the case of H3, the 5 antigenic sites (A–E)24,25
were used to calculate AI in reference to the prototypic pandemic strain of 1968 (A/Aichi/2/1968 (H3N2)). Unless stated otherwise henceforth an AI value for a given H3 HA sequence refers to its antigenic identity with the 1968 pandemic H3 HA. A total of 1,103 H3 avian and swine sequences were downloaded from the NCBI Influenza Database26
and analyzed. Of these 1,103 sequences, 756 were of avian origin and 347 were of swine origin. The avian sequences comprised nine different subtypes (H3N1-9), and the swine sequences comprised four different subtypes (H3N1, H3N2, H3N3 and H3N8). The avian and swine H3 amino acid sequences were compared against A/Aichi/2/1968 and AI values were computed for all the sequences (). In addition, a total of 3,632 human-adapted H3N2 HA sequences were downloaded from the NCBI database and compared against 1968 pandemic H3 HA to enable a cross-species comparison of the antigenic drift (). The AI values and phylogeny analysis indicate that, in comparison with recent human H3, avian and swine H3 are genetically and antigenically closer to the 1968 pandemic HA. Thus, we confirmed that avian and swine H3 are indeed antigenically intact ( & S1)
In addition to the amino acids that constitute the antigenic sites, the attachment of complex glycans at specific glycosylation sites (Asn-X-Ser/Asn-X-Thr, where X is not a Proline) is also often part of the antigenic surface. An increase or decrease in the number of N-glycosylation sites therefore critically governs the antigenic properties of HA27
. The 1968 pandemic H3 HA carries only two glycosylation sites on the globular head region (at 81 & 165), whereas HA from seasonal strains carries an average of six sites (at 63, 122, 126, 133, 144, 165)28
. To incorporate glycosylation in the calculation of antigenic identity, the globular head region of the avian and swine HA sequences were examined for the conservation of 1968 pandemic H3-like glycosylation pattern (Methods
). Among the 1,103 avian and swine H3 HA sequences, 359 carried additional glycosylation sites or positional shifts and therefore were removed from further consideration. The remaining 744 HA sequences (~ 67%) were found to possess the 1968 pandemic HA-like glycosylation pattern. Out of the 744 HA sequences, strains corresponding to 449 sequences (all avian) were isolated after 2000—many as recently as 2010—and their AI value exceed 70%.
Extrapolating from H1 and H2 pandemic scenarios, the above strains are likely to pose a threat should they acquire the mutations necessary to crossover into human population. Of note, a novel H3N8 avian influenza virus acquired the ability to infect harbor seals in New England recently29
. The AI of the seal H3N8 HA is 78%, which is the habitual AI range of avian H3 influenza viruses. Given the high AI value, the history of the spread of avian influenza to humans and the fact that seal H3N8 has already acquired potential to bind sialic acid receptors that are commonly found in the mammalian respiratory tract29
, seal H3N8 virus could jump, directly or via reassortment, to humans with pandemic consequences. More recently, the CDC reported the outbreak of a triple reassortant H3N2 swine-origin influenza virus (SOIV) and released a set of sequences at Global Initiative on Sharing All Influenza Data (GISAID) following this event. The HA of a prototype outbreak strain, A/Minnesota/11/2010 (referred as Minn10), shares very high homology (approx. 98%) with the HA of swine A/swine/Minnesota/7931/2007(H3N2) (SwMinn10), and has good binding and transmission properties30
. Although the AI value of SwMinn10 (approx. 39%) is comparable to that of a typical seasonal H3 HA, they share very low antigenic identity between them (only 15 out of the 27 [approx. 55%] antigenic positions are conserved). More importantly, the glycosylation pattern appears to be very different between SwMinn10 and seasonal H3 HA. The SwMinn10 HA contains only three glycosylation sites in the globular head region, compared to 6 for the seasonal HA. The swine predecessor was not part of the 581 sequences identified by the analysis. This is due to its low AI value and the extra (third) glycosylation site in the head region. Although Minn10 cannot be regarded as a strain resembling the 1968 pandemic strain, the outbreak caused by this virus supports our theory that avian and swine strains that are divergent enough from the seasonal HA, both antigenically and with respect to their glycosylation pattern, need to be considered as potential threats. Consequently, based on the above observations, we relaxed the criteria used to identify potential pandemic strains and considered those HAs isolated after 2000, having matching glycosylation pattern as pandemic H3 and whose AI was equal to or greater than 49%, the maximum AI value of recent seasonal H3 (2000 or after) (). This yielded 581 sequences (549 avian, 32 swine).
If a virus carrying a HA similar to any one of the 581 sequences acquires the potential to crossover into humans, it would likely have a major impact on both immune recognition and vaccine efficacy. The efficacy of the influenza vaccine in humans is thought to correlate well with the ‘antigenic relatedness' metric (reciprocal of antigenic distance) obtained from ferret antisera hemagglutinin inhibition (HI) assays31,32
between the vaccine strain and the circulating epidemic strains33,34
. We tested the degree of correlation between the AI values and the HI-derived antigenic relatedness to: (1) assess the potential of AI in predicting vaccine-induced cross-reactive antibody responses; and (2) to evaluate the cross-protective capacity of the current vaccine strain, A/Victoria/361/2011 (H3N2), against potential threats. For this exercise, we analyzed three sets of ferret serum HI cross-reactivity data where amino acid sequences of the HA1 polypeptide were present. The first set contained 7 viral strains (21 pairwise comparisons) isolated from 2008 to 201035
. The second set contained 9 viral strains (36 pairwise comparisons) isolated from 1970 to 197936
. The third set contained 6 viral strains (15 pairwise comparisons) isolated from 1994 to 199937
. Antigenic relatedness between two viral strains based on ferret anti-serum was determined using the method described by Lee et al.32
. Briefly, the antigenic relatedness between two viral strains is directly proportional to the ratio of the product of the heterologous titers against each other to the product of the homologous titers. In total, 72 pairwise comparisons among 22 viruses were available for analysis. Among the 72 pairwise comparisons, 5 (7%) have an antigenic relatedness > 70% (i.e., similar antigenicity), and 67 (93%) have an antigenic relatedness < 70% (i.e., antigenic variant). Results indicate that the AI values have significant correlation with the HI-based antigenic relatedness metric (,,), indicating that AI values could be applied to predict vaccine-induced cross-reactive antibody responses and thus selection of vaccine strains. Particularly, antigenically related viral strains (> 70%) have AI > 80%, hence we employed an 80% cutoff to determine protection, or lack thereof, between a vaccine strain and a challenge viral strain. The current H3N2 vaccine strain A/Victoria/361/2011 has AI values of 92% with the seasonal viral strain A/Brisbane/10/2007, 29% with the A/Aichi/2/68, 56% with a typical H3N2 SOIV and 44% with a representative swine H3 strain from the group of swine viral strains having AI > 49%. These data indicate that the current vaccine strain is unlikely to offer cross-protection against the circulating swine or SOIV viruses whatsoever. Supporting this, IgG polyclonal raised in rabbit with seasonal vaccine H3 strain (A/Brisbane/10/2007(H3N2)) preferentially bind to current seasonal H3 but have weaker affinity to a representative swine H3 (). More significantly, out of the 581 HA sequences, six swine HAs already contain the prototypic mutations (L226, S228) necessary for HA human adaptation38
, and are thus capable of entering the human population either directly or via reassortment (, )38
. We recombinantly expressed HA derived from two swine isolates, A/swine/Chonburi/05CB2/2005 (H3N2) and A/swine/Nakhon pathom/NIAH586-2/2005 (H3N2), which have high AI value () and characterized their relative binding affinities to representative avian and human receptors on a glycan array platform (Methods
, ). Both swine HAs showed high affinity binding to both human and avian receptors. The high affinity human receptor-binding of these swine HAs appears to be in the same range as that of other seasonal H3 HAs characterized previously39,40
, and are thus capable of entering the human population either directly or via reassortment. The antigenic relationship of these HAs (AI value and glycosylation pattern) to the pandemic 1968 H3N2 HA strongly suggests that the six isolates belong to swine virus lineage and not examples of transient reverse zoonoses. Phylogenetic analysis of the 32 swine isolates revealed that majority of them fall under European and Asian swine lineages.
HI-based antigenic relatedness (upper right) and AI values (lower left) in pairwise comparisons among 7 influenza H3N2 viruses isolated from 2008 to 2010 (R = 0.603314, p-value = 0.002)
HI-based antigenic relatedness (upper right) and AI values (lower left) in pairwise comparisons among 9 influenza H3N2 viruses isolated from 1970 to 1979 (R = 0.523472, p-value = 0.00057)
HI-based antigenic relatedness (upper right) and AI values (lower left) in pairwise comparisons among 6 influenza H3N2 viruses isolated from 1994 to 1999 (R = 0.61, p-value = 0.007)
Binding of anti-A/Brisbane/10/2007(H3N2) pAb to H3 strains measured by ELISA.
Table 4 Avian and Swine HAs antigenically similar to 1968 pandemic H3N2 HA. The AI values and glycosylation pattern of A/Aichi/2/1968(H3N2) and six swine HAs having prototypic mutations (L226, S228) necessary for HA human adaptation are compared alongside. A (more ...)
Genetic, antigenic and glycosylation-pattern relatedness of 1968 pandemic H3N2 HA to seasonal, swine and avian H3 HA.
Glycan microarray analysis of representative H3 HAs.
The analyses presented here portend a vaccine strategy to prevent a future H3 pandemic. Among the WHO recommended vaccine strains of influenza A/H3N2 virus, A/Hong Kong/1/1968(H3N2) will be effective (AI > 80%) against 505 of 581 strains (~ 87%) identified by this study, and thus could be used for the development of pandemic influenza vaccine. Surprisingly, H3N2 vaccine strains that were subsequently used are not capable of being as effective. These data suggest that a cocktail of A/Hong Kong/1/1968(H3N2) and an avian and swine strain each that represent the circulating influenza in birds and pigs can form the components of the pandemic influenza vaccine.
To understand the results from AI calculations in the context of the spatial relationship between glycosylation site and antigenic sites of H3 HA we constructed structural homology models of HA1 globular head of ACHI68, BRBN07 and CHIB05 HAs (see for strain information). These structural models of HA comprised the basic trimannosyl core of N-linked glycan attached to the glycosylation sites (). From the structural comparison it is clear that antigenic shape of HA which includes antigenic sites A-E and the glycosylation pattern of HA1 from the swine strain (CHIB05) closely resembles that of the 1968 pandemic HA. Conversely, the antigenic shape of a more recent seasonal strain (BRBN07) is remarkably different from that of the pandemic strain.