By using x-ray crystallography, the structure of H7 HA from the NY107 virus was determined to 2.6 Å resolution (Table 1). In addition, we also report three H7 HA receptor complex structures, with avian receptor analog (3′SLN) to 2.7 Å resolution and with human receptor analogs (6′SLN and LSTb) to 3.0 Å and 2.6 Å resolution, respectively (Table 1). The overall structure of NY107 is similar to other reported HA structures with a globular head containing the RBS and vestigial esterase domain, and a membrane proximal domain with its distinctive, central helical stalk and HA1/HA2 cleavage site (Figure 1A). Although five asparagine-linked glycosylation sites are predicted in the NY107 HA monomer, interpretable electron density was observed at only two sites, Asn38 in HA1 and Asn82 in HA2 (all residue numbers are based on H3 numbering). At these sites, only one or two N-acetyl glucosamines could be interpreted.

During viral replication, HA is synthesized as a single chain precursor (HA0) and cleaved by specific host proteases into the infectious HA1/HA2 form. In baculovirus expression systems, highly pathogenic HAs, with a polybasic cleavage site, are expressed as an HA1/HA2 form [24], whereas HAs with monobasic cleavage sites (single Arg) from low pathogenic viruses are expressed as the HA0 form [25]. NY107 is regarded as a low pathogenic virus, and as expected, was produced in the HA0 form (Figure S2). However, subsequent digestion with thrombin protease to remove the His-tag resulted in cleavage to a profile on SDS-PAGE comparable to that of an HA1/HA2 form (Figure S2). A comparison of the NY107 cleavage site with the consensus cleavage pattern in the MEROPS database (http://merops.sanger.ac.uk) suggests it to be a possible thrombin cleavage site.

Based on their molecular phylogenies, HAs are divided into two groups and five clades: group 1 includes H8, H9, and H12; H1, H2, H5, and H6; H11, H13 and H16; group 2 includes H3, H4, and H14; H7, H10 and H15 [26]. Among all available HA structures, we selected ten representative HAs from both avian and human subtypes for structural analysis. As expected, NY107 HA is structurally very similar to the Avian-H7 in all comparisons and closely related to H3, the other group 2 members used in the analyses (Tables S1 and S2).

The RBS is at the membrane distal end of each HA monomer and its specificity for sialic acid and the nature of its linkage to a vicinal galactose residue is a major determinant of host range-restriction. The consensus RBS for all current HAs is composed of three major structural elements: a 190-helix (residues 188–194), a 220-loop (residues 221–228), and a 130-loop (residues 134–138). In addition, highly conserved residues (Tyr98, Trp153, His183, and Tyr195) form the base of the pocket.

Although the NY107 RBS is similar to other subtypes (H1, H2, H3, H5, and H9), a previously observed specific feature of H7 HAs, is also observed in the NY107 150-loop region: two residues inserted at position 158 result in this loop protruding more than 6Å towards the binding site compared to other subtype HAs (Figure 1B and Table S2) [27]. More interestingly, the eight amino acid deletion, only found in the North American lineage H7s, from position 221 to 228 (Figure S1), resulted in a complete loss of the 220-loop (Figure 1B). Sequence alignment shows that Arg220 and Arg229 are conserved in all influenza A HA subtypes (Figure S1), but structural alignment of NY107 HA shows Arg220 occupying the Gly228 position, and the much shorter loop turns at residue Pro217 (Figure 1C). The Cα distance between NY107 Arg220 and its homolog in the Av-H7 structure (PDB: 1TI8) [27] is 5.8Å, and they point in opposite directions (Figure 1C). The side chain direction of Av-H7 Arg220 is almost parallel with the beta sheet after Arg229, whereas the NY107 Arg220 points downward to the binding pocket. The Cα position of Arg229 in both H7 structures remains the same, except the side chain in the NY107 swings away by about 5.9Å (Figure 1C) and could help to stabilize this region by forming a hydrogen bond to the mainchain carbonyl of Gln210 in the neighboring monomer. In the absence of the 220-loop in NY107 HA, upon glycan binding the long side chain of Arg220 compensates for its loss and is displaced 4Å upward to form hydrogen bonds with receptor analogs inside the binding pocket (Figure 1D).

Previously, mutations in the HA receptor binding domains of H1N1 (Glu190Asp/Gly225Asp) and H2N2/H3N2 (Gln226Leu and Gly228Ser) subtypes were responsible for adaptation of these viruses to pandemic strains [24], [28], [29], [30]. Due to missing residues 221–228 in the NY107 HA RBS, neither mechanism for adaptation is possible. Thus, in order to look more closely at the role of the missing loop and its effect on receptor specificity, we first subjected the recombinant HA (recHA) to glycan microarray analyses and compared it to a reverse genetics-derived NY107 virus, and a co-circulating Eurasian virus and recHA, A/Netherlands/219/2003 (NL219), that has the consensus avian sequence in the 220-loop and it also infected a human [15].

Glycan microarray analysis of recombinant NY107 (Figure 2A and Table 2) revealed a highly restricted binding profile with strong binding to only α2-3 sulfated (#4–8), α2-3 branched (#9–11) and mixed α2-3/α2-6 branched sialosides (#60–64) as well as to the long linear sialyl di- and tri-lactosamines (#22, 24). Weak binding was also observed (above background) to other α2-3 glycans on the array. The recombinant NY107 also revealed a strict glycan binding preference to only one α2-6 glycan, the internal structure, Galβ1-3(Neu5Acα2-6)GlcNAcβ1-3Galβ1-4Glc (#58; LSTb) (Figure 2A), a glycan highlighted in a previous study [22]. The virus with higher valency and avidity revealed stronger binding to all α2-3 groups, in addition to the branched di-sialyl α2-6 biantennary structures (#46–48) as well the LSTb (#58) (Figure 2B and Table 2). In contrast, the NL219 recHA (Figure 2C and Table 2) bound well to only the avian α2-3 containing sialyl-glycans (sulfated, branched, linear and fucosylated). Its corresponding virus also reflected this specificity although it also revealed strong binding to α2-3 N-glycolylneuraminic acid (Neu5Gc) containing glycans (#66–70) (Figure 2D and Table 2).

To further assess the effect of the missing 220 loop on HA structural stability and receptor specificity it was essential to evaluate these functions on the ancestral HA containing the full length 220-loop. To this end, we engineered an HA with an avian H7 consensus (PQVNGQSG) 220-loop re-introduced (NY107-220ins) into the NY107 HA and recovered this virus by reverse genetics. Compared to the NY107 virus (Figure 2A) glycan microarray analyses of the resulting NY107-220ins virus (Figure 3A and Table 2) revealed a decrease in binding to branched (#9–11) and linear (#12–27) α2-3 sialosides and a loss of binding to the branched di-sialyl α2-6 biantennary structures (#46–48), LSTb (#58) as well as the mixed α2-3/α2-6 branched sialosides (#60–64). In addition, sequence analysis of the NY107-220ins HA revealed the presence of quasispecies in the second position of the inserted loop, P(Q/K)VNGQSG, suggesting that re-introduction of the loop alone is not tolerated and does not create an avian-type binding profile. Thus other amino acid substitutions in the HA might have co-evolved with the deletion of the 220 loop to help stabilize the RBS/HA to maintain functionality.

When viruses containing this 220-loop deletion emerged in North America in the mid 90's, four additional amino acid substitutions, Gly114Arg, Asp119Gly, Gly186Glu and Gly205Arg, in the HA1 as well as an Asp19Asn in the HA2 chain were also introduced to most of the circulating isolates. Of these, Gly186Glu and Gly205Arg in the HA1 are close to the RBS, at the monomer interface, and could potentially modulate its structure and/or function. NY107 viruses with a restored consensus 220-loop and a single Glu186Gly (NY107-ins-186) or Arg205Gly (NY107-ins-205) substitution as well as the Glu186Gly/Arg205Gly double substitution (NY107-ins-186/205) were derived by reverse genetics and evaluated. Glycan microarray analysis for the three resulting viruses revealed similar glycan binding profiles with increased binding to α2-3 sialosides, including mixed α2-3/α2-6 branched sialosides (#60–64), α2-3 Neu5Gc (#66–70), but limited binding to the α2,6 sialosides (Figures 3B, 3C, 3D), resulting in a binding profile virtually identical to that of the NL219 virus and other avian influenza viruses (Figure 2D) [30]. Sequence analysis of the three reverse genetics derived viruses revealed no mutations/quasispecies in the HAs of either the NY107-ins-186 or the NY107-ins-186/205 virus stocks, indicative of replication fitness. For the NY107-ins-205 virus however, a Glu186Gly substitution emerged in the HA after only two passages in eggs following recovery from DNA transfection, indicating the importance of the co-variant position 186 with respect to HA functionality/glycan specificity. Altogether, the data indicates that the H7 subtype avian influenza viruses that were circulating in aquatic birds and poultry in North America before 1996 exhibited a classic avian α2-3 sialoside binding preference. In order for the 220-loop deletion to be tolerated, concurrent Gly186Glu and Gly205Arg substitutions in the vicinity of RBS of HA emerged to achieve a restricted α2-3 binding profile and only a moderate/limited increase in binding to branched di-sialyl α2-6 biantennary structures (#46–48) as well the α2,6 internal sialoside, LSTb (#58).

To understand from a structural perspective how NY107 interacts with host receptors, we solved the structure of NY107 in complex with an avian and two human receptor analogs. For the avian receptor analog, 3′SLN, the electron density maps revealed well-ordered features for the Sia-1, Gal-2, and GlcNAc-3 in the NY107 HA complex structure (Figure 4A). Structural comparison of NY107 HA binding to other, H1, H2, H3, H5, and H9 subtypes (Figure S2A) revealed that 3′SLN binding to NY107 resembled binding of the other published HAs. Indeed, the terminal Sia-1 moiety is positioned almost identically in all structures, and forms the majority of hydrogen bonds and contacts with residues in the RBS (Figure 4A and Table S3).

Published avian HA structures with an intact 220-loop form very close interactions with Gal-2 of 3′SLN via residue Gln226 which is important in receptor specificity and host adaptation. For example, in the avian H7/3′SLN HA structure it interacts with Gal-2 O4 [31]. In the NY107 HA structure, although Gln226 is absent and no other residue occupies the same space as Gln226 (Figure 1B), Arg220 does forms a hydrogen bond between Arg220 NH2 and Gal-2 O4 (Figure 4A). Interestingly, although there was interpretable density for the GlcNAc-3 (Figure 4A and Figure S4B), no hydrogen bonding was apparent between the HA and the GlcNAc-3, which is consistent with other reported structures [32]. Thus, for binding to avian receptors, the trans conformation of α2-3 linkages is essential and perhaps only the first two saccharides are required. Indeed, due to the absence of 220-loop in the NY107 HA structure, the “aperture” of the RBS formed by 220-loop and 130-loop in regular HAs is increased by ∼10 Å, so that the branched, internal, and perhaps more complicated glycans might be accommodated more efficiently.

In the NY107/6′SLN complex, only Sia-1 and Gal-2 are ordered (Figure 4B). The Sia-1 remains in the same position as previously analyzed glycan/HA complexes from H1, H2, H3, H5, and H9 (Figure S3B), whereas the Av-H7 complex structure with Sialyllacto-N-tetraose c (LSTc) did not reveal any density for the Sia-1 in the receptor binding site [31]. The Gal-2 position varies significantly among different subtypes. Compared to the human-adapted H1 HA [32], Gal-2 in the NY107 HA is 3Å higher, and thus is further from the protein (Figure S3B). In NY107, the Gal-2 only forms an intramolecular, saccharide-saccharide interaction with Sia-1. The poor electron density map and fewer interactions with protein residues suggest that the cis conformation of α2-6 linkages in 6′SLN trisaccharides show a reduced binding affinity with NY107.

Glycan array results with NY107 revealed a strong binding signal for the internal α2-6 sialoside, LSTb. To further investigate this interaction, we solved the structure of the NY107/LSTb complex. The final model contained Sia-1, NAG-2, Gal-3, and Gal-5 in the RBS. Although glycan microarray data indicated NY107 to have a specific affinity for LSTb, few interactions were apparent from the crystal structure. Sia-1 still forms multiple hydrogen bonds with residues in the RBS (Table S3 & Figure 4C). The branched Gal-5 interacts with Ser137, to help stabilize the LSTb binding. However, Arg220 and Lys193, the two residues showing close binding with 3′SLN, did not form any hydrogen bonds with LSTb. In the structure, Gal-5 also interacts with a crystal packing symmetry mate and thus the flexibility of whole LSTb may be restricted. In solution, with more freedom, the LSTb should be able to tilt closer to the RBS, and thus Glc-4 may have more interactions with the 190-helix than seen in the crystal structure.

Based on H3 numbering [42], cDNA corresponding to residues 11–329 (HA1) and 1–176 (HA2) of the ectodomain of the hemagglutinin (HA) from A/New York/107/2003 (H7N2; Genbank:ACC55270) and A/Netherlands/219/2003 (H7N7; Genebank: AAR02640) was cloned into the baculovirus transfer vector, pAcGP67-A (BD Biosciences), incorporating a C-terminal thrombin cleavage site, a “foldon” sequence [43] and a His-tag at the extreme C-terminus of the construct to enable protein purification [25], [44]. Transfection and virus amplification were carried out according to the baculovirus expression system manual (BD Biosciences Pharmingen).

Soluble NY107 was recovered from the cell supernatant by metal affinity chromatography using Ni-NTA resin (Qiagen Inc.). Fractions containing NY107 were pooled and dialyzed against 10 mM Tris-HCl, 50 mM NaCl, pH 8.0, then subjected to ion-exchange chromatography (IEX) using a Mono-Q HR 10/10 column (GE Healthcare). IEX purified NY107 was subjected to thrombin digest (3 units/mg protein; overnight at 4°C) and purified by gel filtration chromatography using a Superdex-200 16/60 column (GE Healthcare) and 50 mM Tris-HCl, 100 mM NaCl, pH 8.0 as running buffer. Protein eluting as a trimer was buffer exchanged into 10 mM Tris-HCl, 50 mM NaCl, pH 8.0 and concentrated to 14.5 mg/ml for crystallization trials. At this stage, the protein sample still contained the additional plasmid-encoded residues at both the N (ADPG) and C terminus (SGRLVPR).

Initial crystallization trials were set up using a Topaz Free Interface Diffusion (FID) Crystallizer system (Fluidigm Corporation, San Francisco, CA). Crystals were observed in several conditions containing PEG 3350 or PEG 4000. Following optimization, diffraction quality crystals for NY107 were obtained at room temperature using a modified method for microbath under oil [45], by mixing the protein with reservoir solution containing 20% PEG 3350, 0.2 M magnesium chloride at pH 7.2. For receptor analog complexes, crystals were soaked for 3 hours in the crystallization buffer containing 10 mM 3′SLN or 6′SLN (V-labs Inc., Covington, LA), or overnight in 10mM LSTb (Sigma, St. Louis, MO). All crystals were flash-cooled at 100K using 20% glycerol as the cryo-protectant. Datasets were collected at Advanced Photon Source (APS) beamlines 22 ID and BM at 100K. Data were processed with the DENZO-SACLEPACK suite [46]. Statistics for data collection are presented in Table 1.

The structure of NY107 was determined by molecular replacement with Phaser [47] using the structure of the avian H7 (Av-H7) from A/turkey/Italy/2002, pdb:1TI8 (HA1, 78% identity; HA2, 90% identity) as the searching model. One HA trimer occupies the asymmetric unit with an estimated solvent content of 58% based on a Matthews' coefficient (Vm) of 2.9 ų/Da. Rigid body refinement of the trimer led to an overall R/Rfree of 28.6%/37.4%. The model was then “mutated” to the correct sequence and rebuilt by Coot [48], then the protein structures were refined with REFMAC [49] using TLS refinement [50]. The final models were assessed using MolProbity [51]. The three complex structures were refined and evaluated using the same strategy. All statistics for data processing and refinement are presented in Table 1. Electron density maps (2fo-fc) were generated in Refmac [49] while simulated annealing omit maps were generated by sa-omit-map, a part of the Crystallography and NMR System (CNS) software [52].

Wild type and mutant viruses of NY107 (H7N2) and A/Netherland/219/2003 (H7N7) were generated from plasmids by a reverse genetics approach [53]. To generate viruses with amino acid insertion or substitution in the HA, mutations were introduced into plasmid DNA with an overlap extension PCR approach [54]. Viruses derived by plasmid transfection of HK293 cells were propagated in eggs. The genomes of resulting virus stocks were sequenced to detect the emergence of possible variants during amplification.

Glycan microarray printing and recHA analyses have been described previously [24], [30], [44], [55] (see Table 2 for glycans used for analyses in these experiments). Virus were analyzed on the microarray as described previously [30].

The atomic coordinates and structure factors of NY107 are available from the RCSB PDB under accession codes 3M5G for the unliganded NY107, 3M5H for the NY107 with 3′-SLN and 3M5I and 3M5J for NY107 with 6′SLN and LSTb, respectively.

A/New York/107/03 (H7N2), Genbank: ACC55270; A/Netherlands/219/03 (H7N7), Genbank: AAR02640; A/Hong Kong/1-9/68 (H3N2), 2HMG; A/Duck/Ukraine/1/63 (H3N8), PDB: 1MQL; A/South Carolina/1/18 (H1N1), PDB: 1RD8; A/Puerto Rico/8/34 (H1N1), PDB: 1RU7; A/Swine/Iowa/15/30 (H1N1), PDB: 1RUY; A/Singapore/1/1957 (H2N2), PDB: 2WRC; A/Viet Nam/1203/04 (H5N1), PDB: 2FK0; A/Duck/Singapore/3/97 (H5N3), PDB: 1JSM; A/Swine/Hong Kong/9/98 (H9N2), PDB: 1JSD; A/Turkey/Italy/8000/02 (H7N3), PDB: 1TI8; C/Johannesburg/1/66, 1FLC.