Glycan remodeled erythrocytes facilitate antigenic characterization of recent A/H3N2 influenza viruses

Glycan microarray analysis to determine A/H3N2 receptor specificity

Although glycan microarray technology has been used to examine receptor requirements of HAs²⁴, these were not populated with biologically relevant glycans to establish minimal receptor requirements. This information is, however, critical to understand how receptor binding has evolved over time and how a lack of expression of specific glycans by erythrocytes or laboratory hosts may have resulted in a loss of agglutination or a lack of propagation, respectively.

We have constructed a glycan array that is populated with biologically relevant bi-antennary N-glycans having different numbers of LacNAc repeating units in various structural configurations. They resemble structures found on human respiratory tissue, which abundantly expresses N-glycans having multiple consecutive LacNAc repeating units that can be capped by sialic acid^25,26. The synthetic glycans are either unmodified (compounds 1–3), capped by avian α2,3-linked (compounds 4–6), or human α2,6-linked sialosides (compounds 7–17). Most naturally occurring N-linked glycans have asymmetrical architectures in which the various antennae are modified by oligo-LacNAc moieties of different lengths²⁷. To probe the importance of such architectures for HA recognition, we prepared symmetrical as well as asymmetrical glycans (8, 9, 11, 12, 13, 15, 16, and 17) that are modified by either one or two sialosides linked to LacNAc chains of different length. The collections of compounds made it possible to probe the importance of mono- vs. bidentate binding interactions, and a possible preference or requirement for a sialoside at specific antenna or at an extended LacNAc chain. All glycans contain an anomeric asparagine moiety and its α-amine facilitated immobilization on amine reactive, NHS-activated glass slides. The quality of the printing was validated by probing the array with the lectins ECA, SNA, and MAL1 as well as an anti-H3 antibody (Fig. 1a and Supplementary Fig. 1).

a Binding was visualized using a human anti-H3 stalk antibody (CR8020). Bars represent the background-subtracted average relative fluorescence units (RFU) of four replicates ±SD. Values for all individual datapoints are represented in the Supplementary Source Data file. b The 30 most abundant N-glycans on erythrocytes from chicken and turkey organized by the number of LacNAc repeating units and relative intensity. High-mannose type N-glycans are not shown and their abundance is presented in Supplementary Fig. 5. The structures of all detected glycans are shown in Data S1.

The glycan array was probed with various A/H3N2 viruses, representing distinct evolutionary time points and clades and having different abilities to agglutinate erythrocytes derived from species commonly used in HI assays (Supplementary Fig. 2). In this respect, chicken and turkey erythrocytes are widely employed in HA assays because they are nucleated, which unlike mammalian erythrocytes, sediment rapidly thereby greatly facilitating the visual readout of the assay⁷. They express α2,3- and α2,6-linked sialosides and can be agglutinated by avian as well as human influenza viruses. Other types of erythrocytes have been employed for HI assays²⁸ and in particular those of guinea pig erythrocytes have been proven to be useful because they exhibit a somewhat broader agglutination ability and can for example be employed to antigenically characterize H3N2 viruses of the 3C.2 clade which cannot be agglutinated turkey and chicken erythrocytes²⁹.

In the studies, we included A/NL/816/91 (NL91) which can agglutinate chicken, turkey, and guinea pig erythrocytes, A/NL/109/03 (NL03) which only agglutinates turkey and guinea pig erythrocytes²⁹ and A/NL/761/09 (NL09) which only agglutinates α2,6-resialylated turkey and guinea pig erythrocytes^30,31,32. During the past decade, A/H3N2 viruses have evolved into distinct, cocirculating antigenic groups, referred to as clades (Supplementary Fig. 2). We examined A/NL/1797/17 (NL17) and A/NL/371/19 (NL19) as recent examples of the 3C.2a clade that insufficiently hemagglutinate all commonly used erythrocytes for HI assays, and poorly infect MDCK cells³³. The 3C.3a clade is represented by A/NL/10006/19 and A/NL/384/19, and their evolutionary predecessor A/NL/622/12 (3C.3). Although 3C.3 viruses cannot agglutinate any erythrocyte type, 3C.3a viruses are unique as they regained an ability to agglutinate turkey and guinea pig erythrocytes (Supplementary Fig. 2).

Whole viruses were applied to the microarray and detection of binding was accomplished by a human anti-H3 stalk antibody (CR8020) (Supplementary Fig. 1). NL91 recognized most of the human-type receptors, including compounds that have an α2,6-sialoside on a mono-LacNAc residue (glycans 7–9, Fig. 1a and Supplementary Fig. 3). Compound 8 exhibited a substantial greater responsiveness compared to 9 indicating that this virus has a preference for a sialoside at the α1,3-arm. Interestingly, the sialyltransferase, ST6Gal1, which is solely responsible for installing human-type receptors, preferentially modifies the α1,3-arm of N-linked glycans³⁴. Compounds 7 and 8 did bind similarly demonstrating that an additional sialic acid at the α1,6-arm does not substantially contribute to binding. Another unanticipated observation was that compounds 12 and 16 did not exhibit binding whereas 9 showed responsiveness highlighting that an extended and unmodified LacNAc moiety at the α1,3-arm can block recognition of the other arm. Collectively, the results show that the minimal receptor for NL91 is a bi-antennary N-glycan having two LacNAc moieties modified by a single sialoside (glycan 8).

NL03 and NL09 recognized far fewer glycans and did not bind to structures having their α2,6-sialosides at a mono-LacNAc moiety (7-9, 12 and 16). This observation indicates that the minimal receptor for these viruses is a bis-sialylated N-glycan having at least one di-LacNAc moiety (glycan 13). NL17 and NL19 (3C.2a) showed only strong responsiveness to 14, 15, and 17. These glycans have in common that at least one of the arms is extended by three consecutive LacNAc units that is further modified by an α2,6-sialoside. Thus, a glycan having four LacNAc units arranged in an asymmetrical manner (15) represents the minimal receptor for these viruses. Mono-sialylated derivative 15 gave a similar responsiveness compared to the bis-sialosides 14 and 17 indicating that a bidentate binding event does not substantially contribute to recognition as previously suggested²². Instead, it appears that reduction in binding of an α2,6-sialyl-Gal moiety, which is widely regarded as the prototypic human receptor, has been compensated by recognition of sialosides at extended LacNAc chains. 3C.3a viruses (A/NL/10006/19 and A/NL/384/19) exhibited a similar binding profile as NL03 and 09 and bound to bis-sialosides 10 and 13 having the Neu5Ac residue at a di-LacNAc chain. Interestingly, their ancestor (A/NL/622/12, 3C.3) required the Neu5Ac residue to be presented on a tri-LacNAc structure similar to the requirement of 3C.2a viruses. Thus, recent 3C.3a viruses have regained an ability to recognize shorter structures.

Glycomic analysis of chicken and turkey erythrocytes

Next, we examined structures of N-linked glycans expressed by chicken and turkey erythrocytes and compared the data with the receptor requirements of the various A/H3N2 viruses. Membrane fractions of the cells were treated with PNGase F to release the N-glycans which were isolated by solid phase extraction using C18 and Porous Graphitized Carbon (PGC) cartridges, and then analyzed by liquid chromatography mass spectrometry (LC-MS)³⁵. The 30 most abundant complex type N-glycan compositions for the two cell types are presented in Fig. 1b. The compounds are organized according to an increasing number of LacNAc moieties (indicated by different color coding) and include hybrid-type and core structures having none (black bars) or 1 LacNAc moiety (yellow bars) and complex N-glycans having variable numbers of LacNAc repeating units (2–4 LacNAc, green, blue and purple bars, respectively). High-mannose type N-glycans were also detected and their structures and abundance are shown in Supplementary Fig. 5. Strikingly, chicken erythrocytes do not substantially express N-glycans having 4 LacNAc units which is the minimal epitope requirement for contemporary non-agglutinating A/H3N2 viruses. Turkey erythrocytes do express some glycans with this number of LacNAc units (purple bars in Fig. 1b), but the majority was assigned as tri- and tetra-antennary glycans because of substitution with three or four sialic acids. The latter was supported by selective release of bi-antennary N-glycans by Endo F2, and in this case LC-MS analysis did not detect glycans having four LacNAc moieties (Supplementary Fig. 4). Thus, turkey erythrocytes also do not substantially display sialylated epitopes having three consecutive LacNAc moieties. The majority of the glycans released by Endo F2 lacked fucose indicating that the fucosides observed in some of the structures depicted in Fig. 1 can be assigned to core modification. Chicken erythrocytes express substantial quantities of high mannose glycans (Supplementary Fig. 5) whereas turkey cells display almost none of these structures. The greater abundance of complex type glycans on turkey erythrocytes offers a possible rationale for the ability of the NL03 and NL09 A/H3N2 viruses to agglutinate unmodified or α2,6-resialylated turkey erythrocytes, respectively.

Erythrocyte glycoengineering to install functional receptors

We embarked on a strategy to enzymatically remodel glycans of fowl erythrocytes to install receptors for A/H3N2 viruses of the 3C.2 clade to make them suitable for HI assays (Fig. 2a). Treatment of erythrocytes with a neuraminidase was expected to remove sialic acids and reveal terminal galactosides which are appropriate acceptors for installing additional LacNAc moieties. The latter residues can be introduced by the concerted action of the enzymes B4GalT1 and B3GnT2, which sequentially install β1,4-linked galactoside and β1,3-linked N-acetyl-glucosamines, respectively. The terminal galactosides of the resulting extended LacNAc moieties can then be modified by the sialyltransferase ST6Gal1 to install terminal α2,6-linked sialosides³⁶. The enzymatic remodeling was conveniently performed by incubating the erythrocytes with the neuraminidase from Arthrobacter ureafaciens for 6 h after which B4GalT1, B3GnT2³⁷, UDP-Gal, and UDP-GlcNAc were added followed by incubation overnight. Next, the cells were pelleted by centrifugation, washed to remove the enzymes and sugar nucleotides, and then incubated with ST6Gal1 in the presence of CMP-Neu5Ac for 4 h. Glycomic analysis of the resulting cells, which were denoted as 2,6-Sia Poly-LN cells, confirmed that the antennae of the N-linked glycans had been extended by additional LacNAc moieties, and both cell types expressed sialylated structures having four LacNAc units (Fig. 2b, 4 LacNAc units are indicated in purple bars). Analysis of glycans on turkey erythrocytes released by Endo F2 treatment confirmed the presence of bi-antennary glycans that have 4 LacNAc moieties and are potentially suitable receptors for contemporary A/H3N2 viruses (Supplementary Fig. 4, glycans having 4 LacNAc units are indicated by purple bars). Chicken and turkey erythrocytes express a mixture of α2,3- and α2,6-linked sialosides. To examine whether an increase in the abundance of α2,6-sialosides would improve agglutination, control cells were prepared by treatment with neuraminidase and resialylation with ST6Gal1 (denoted as 2,6-Sia cells). As negative control, we employed cells that have extended LacNAc moieties but lack sialic acids (denoted as Poly-LN cells).

a Neuraminidase A.U.: Neuraminidase from Arthrobacter ureafaciens; B3GnT2: β-1,3-N-acetylglucosaminyltransferase 2; B4GalT1: β-1,4-galactosyltransferase 1; ST6Gal1: α-2,6-sialyltransferase 1. b The 30 most abundant N-glycans on the enzymatically modified erythrocytes from chicken and turkey (based on relative intensity, excluding high-mannose type N-glycans, for all structures refer to Data S2) sorted by abundance and number of LacNAc units. Proposed structures are assigned to detected glycan compositions. c A/NL/816/91, A/NL/109/03, A/NL/761/09, A/NL/1797/17, A/NL/371/19, A/NL/622/12, A/NL/10006/19, and A/NL/384/19 tested with modified erythrocytes (2,6-Sia Poly-LN) from chicken (blue) and turkey (red). Unmodified, 2,6 resialylated (2,6-Sia) and extended desialylated (Poly-LN) erythrocytes were added as controls. Assays were performed in biological triplicates in the presence of oseltamivir and the means ± SEM were plotte.

Phenotypic properties of the glyco-engineered erythrocytes were examined using the hemagglutination (HA) assay (Fig. 2c). As expected, NL91 agglutinated unmodified, 2,6-Sia and 2,6-Sia poly-LN erythrocytes, which was in agreement with the finding that these viruses can employ N-glycans that have simple and extended α2,6-sialylated structures. NL03 agglutinated unmodified turkey erythrocytes, but interestingly also α2,6-resialylated chicken erythrocytes. The latter may be due to an increase in the abundance of α2,6-linked sialosides having two consecutive LacNAc repeatin units, which are present on chicken erythrocytes. α2,6-Resialylation of turkey erythrocytes was sufficient to recover agglutination of NL09, and in this case the greater abundance of α2,6-sialylation on extended structures already present on these cells is probably responsible for the improved agglutination. Importantly, NL17 and NL19 (3C.2a) agglutinated only erythrocytes that were enzymatically remodeled to have extended sialylated LacNAc moieties (2,6-Sia Poly-LN cells). Similar results were obtained for A/NL/622/12 (3C.3), which is in agreement with receptor requirements similar to viruses of the 3C.2a clade. As anticipated, the 3C.3a viruses (A/NL/10006/19 and A/NL/384/19), which have reverted to recognize shorter structures, could also agglutinate α2,6-resialyated turkey erythrocytes.

Next, HA assays were performed with a wider collection of A/H3N2 viruses (Table 1) to validate the robustness of the glycoengineering method with a focus on contemporary A/H3N2 viruses that have lost the ability to hemagglutinate unmodified erythrocytes and do not replicate efficiently in wild-type MDCK cells. Several recent vaccine strains heavily adapted to growing in eggs (X-161B, IVR-147, X-223A, NIB-104, NIB-112, and X-327), an A/H1N1 (A/Singapore/GP1908/15) and an influenza B (B/Maryland/15/15) strain were included as controls. As expected, pre-2000 A/H3N2 (A/Bilthoven/16190/68, A/Beijing/353/89, and A/Netherlands/816/91) strains efficiently agglutinated unmodified chicken and turkey erythrocytes. Importantly, A/H3N2 viruses that emerged after 2010 only agglutinated the 2,6-Sia Poly-LN cells having extended sialylated epitopes. Although some contemporary A/H3N2 viruses can agglutinate turkey erythrocytes when applied undiluted, especially the two 2019 3C3a viruses, extended sialylated LacNAc moieties increased the efficiency of agglutination by 3–10-fold, including the A/H1N1 and influenza B controls. We performed a time course to determine the stability of the glyco-engineered cells, and no loss in titer or autoagglutination for up to three weeks was observed similar to unmodified cells (Supplementary Fig. 6).

The 2,6-Sia Poly-LN cells were employed to antigenically characterize typical recent seasonal A/H3N2 viruses of various clades by HI assay using post-infection ferret sera (Table 2). All antisera showed robust inhibition of the homologous viruses and variable inhibition of heterologous viruses. Egg-derived vaccine strains displayed poor correspondence with data generated with cell-passaged viruses of the same clade. Antisera raised against cell-passaged virus isolates showed greater clade specificity compared to antisera raised against egg-derived vaccine strains. Additionally, egg-derived vaccine strains were inhibited stronger by various antisera than cell-passaged viruses. The HI assay also revealed that antisera raised against recent vaccine viruses, including A/Kansas/14/17 that was selected for the 2019/2020 northern hemisphere influenza vaccine³⁸, exhibited only minimal cross reactivity against circulating viruses from the same clades, indicating that the circulating viruses differ antigenically from the vaccine strains of the same clade.

The results of the HI assay were compared with a focus reduction assay (FRA) using the same sera and viruses (Supplementary Table 1)³⁹. In this assay, the ability of antibodies to block virus infection in mammalian cell culture (MDCK-Siat cells, which overexpress Sia(α2,6)Gal moieties) is quantified. The FRA confirmed the trends observed in the HI assay (Supplementary Fig. 7), indicating that the modified erythrocytes are reliable for antigenic characterization of A/H3N2 viruses.

Molecular dynamics simulations of HAs in complex with their receptors

α2,6-Sialyl-Gal is widely considered as the prototypic human receptor for IAVs¹⁷. All viruses we examined, including A/H1N1 viruses (Supplementary Fig. 8), recognized with high avidity N-glycans having an α2,6-linked sialoside on an extended LacNAc moiety. Human respiratory tissue abundantly expresses such extended structures^25,26, and thus we reasoned that mutational changes in recent A/H3N2 viruses led to a reduced binding avidity of the prototypic human receptor (α2,6-sialyl-Gal), which was compensated by making interactions with an extended LacNAc chain.

X-ray crystal structures^31,40,41 have shown that sialic acid is recognized in a conserved hydrophobic pocket (Y98, H183, Y195, and W153) (Supplementary Table 2). It makes further interactions through a hydrogen bonding network with residues 135-137, E190, and S228. Sequence alignments showed that post-2000 strains acquired single point mutations at one of these residues (Supplementary Table 2), which disrupted the hydrogen bonding network, and likely resulted in a reduced binding affinity of the prototypic human receptor, which is a sialoside α2,6-linked to galactose³¹. Furthermore, HAs of early human A/H3N2 strains have Glu at residue 190, which can form a hydrogen bond with O9 of sialic acid, whereas post-2000 strains, favor Asp190 at this position, which due to its shorter side chain, cannot form such an interaction. The 190E/D mutation was accompanied by a 225G/D mutation, which resulted in a rotation of Gal-2 allowing a hydrogen bond interaction with the site chain of 225D^31,40. This rotation places the extended LacNAc chain closer to the 190-helix and potentially allows for addition interactions. To examine this mode of binding, we compared molecular dynamics generated structures of α2,6-sialyl poly-LacNAc in complex with NL91, NL03, and NL17 (Fig. 3). It recapitulated observations made by X-ray crystallography studies and provided insight in how mutational changes allowed for interactions with an extended LacNAc chains. The MD trajectory of NL91 only showed interactions with sialic acid, including the important hydrogen bond between Sia-1 O9 and Glu190 (Fig. 3a)^31,40. In the case of NL03 and NL17, Asp190 is at a distance of 4.5 Å to Sia-1 O9, and thus cannot establish a hydrogen bond (Fig. 3d, g). The G225D substitution resulted in a rotation of the bond between the sialic acid and Gal-2 to form an H-bond with its O3¹⁶, (Fig. 3e, h). As anticipated, the resulting rotation of Gal-2 placed the extended LacNAc moieties near the 190-helix⁴¹ resulting in a hydrogen bond between Asp190 and Gal-4 O2, while either Ser or Asn193 provided a H-bond with the acetamide moiety of GlcNAc-3 (Fig. 3e). In addition, Gal-6 (Fig. 3e, h) makes a CH-π interaction with the side chain of Y159 (Supplementary Fig. 10). Such interactions are often observed in glycan-protein complexes and contribute substantially to binding⁴². Structural analysis of the HAs of NL91 and NL03 showed that mutations distal to the receptor-binding domain (A131T, H155T, and E156H), reorient the side chain of Y159 resulting in an extended receptor-binding site allowing interactions with Gal-6 (Supplementary Fig. 9). The MD simulations support that A/H3N2 of the 3C.2 clade have undergone mutations to create and extended binding site to compensate for reduced binding of the non-extended human receptor (Fig. 3c, f, i). Interestingly, 3C.3a viruses, which can utilize shorter receptors having two LacNAc units (Fig. 1a), have a tyrosine to serine mutation at position 159 (Supplementary Table 2). Thus, the dependence on extended receptors is reversible, and these viruses have found a way to bind shorter structures with sufficient affinity for infection. The observation that distant mutations can alter the position of a side chain of an amino acid in the receptor-binding site, which subsequently can become susceptible to antigenic pressure, indicates that such remote mutations need to be considered for the evolution of receptor binding and antigenic distance.

Details of the sialic acid binding sites are shown for A/NL/816/91 (a), A/NL/109/03 (d), and A/NL/1797/17 (g). The interactions of the poly-LN chain with the protein are shown for A/NL/816/91 (b), A/NL/109/03 (e), and A/NL/1797/17 (h). The surface and spheres representations of HA/glycans complexes are shown for A/NL/816/91 (c), A/NL/109/03 (f), and A/NL/1797/17 (i).