Human coronaviruses OC43 and HKU1 bind to 9-O-acetylated sialic acids via a conserved receptor-binding site in spike protein domain A

To test the validity of the current model, we measured the effect of substitutions in the proposed RBS using S1^A–Fc fusion proteins. The binding properties of the mutated proteins were studied by hemagglutination assay (HAA) with rat erythrocytes and by solid-phase lectin-binding enzyme-linked immune assay (sp-LBA) with bovine submaxillary mucin (BSM) as ligand. These assays are complementary: HAA is the more sensitive of the two, while the results of sp-LBA more precisely reflect differences in binding-site affinity (10). HAA and sp-LBA performed with S1^A of BCoV strain Mebus confirmed the binding to 9-O-Ac-Sias (Fig. 1 B–D). For comparison, the S1^A domains of related β1CoVs OC43 strain ATCC and porcine hemagglutinating encephalomyelitis virus (PHEV) strain UU were included. Again, 9-O-Ac-Sia–dependent binding was observed, but the affinity of the S1^A domain of OC43 was ∼32-fold lower than that of BCoV as measured by sp-LBA and that of PHEV was lower still (∼1,450-fold). Substitution in PHEV and OC43 S1^A of Ala for Tyr¹⁶², Glu¹⁸², Trp¹⁸⁴, and His¹⁸⁵ (Fig. 2A), residues deemed critical for binding of BCoV S1^A (17), indeed resulted in decreased binding (Fig. 2 B and C), but importantly and as in the original report, none of the mutations gave complete loss-of-function. The strongest effect was observed for Trp¹⁸⁴, but its substitution in the context of OC43 S1^A merely reduced binding affinity to that of wild-type PHEV S1^A. In a reverse approach, we attempted to identify substitutions that result in gain-of-function of OC43 S1^A (i.e., mutations that would raise its binding to that of BCoV S1^A). To this end, we systematically replaced residues within or proximal to the proposed binding site in OC43 S1^A by their BCoV orthologs (Fig. 2A; see also SI Appendix, Fig. S1). Individual substitutions Arg¹⁴³His, Lys¹⁸¹Val, Leu¹⁸⁶Trp, and Ile¹⁴⁵Thr (Fig. 2D), however, did not increase binding affinity, and neither did the replacement of residues 147–151 (Ser-Thr-Gln-Asp-Gly) by Leu, which resulted not only in a large deletion but also in the removal of an N-glycosylation site. In fact, an OC43 S1^A mutant, in which we combined all substitutions and thus essentially reconstructed the proposed BCoV S1^A RBS in the OC43 background, did not differ in its binding affinity from the parental wild-type OC43 protein as measured by HAA and sp-LBA (Fig. 2D). From the combined results, we conclude that the actual RBS is located elsewhere. In further support, the RBS proposed by Peng et al. (17)—henceforth referred to as “site A”—does not conform to the typical anatomy of O-Ac-Sia binding sites (18–24).

The most obvious and direct approach to identify the RBS would be by crystallographic analysis of an S1^A–ligand complex. In their report, Peng et al. (17) stated that all efforts to determine the S1^A holo-structure had been unsuccessful. We considered that the complications the authors encountered might be related to crystal packing, and therefore opted to perform crystallization trials not with BCoV or OC43 S1^A, but with that of PHEV. Because its sequence varies from that of BCoV S1^A by 22%, we hoped that crystals of different packing topology would be produced, as was indeed the case. Crystals formed under a variety of conditions, but all with space group P3₁21 with two S1^A molecules per asymmetric unit. These crystals consistently disintegrated within seconds during soaking attempts with receptor analog methyl-5-N-acetyl-4,9-di-O-acetyl-α-neuraminoside. Crystals flash-frozen immediately upon ligand addition showed poor diffraction that did not extend beyond 10-Å resolution. The observations are suggestive of ligand-binding interfering with crystal stability (SI Appendix, Fig. S2).

As an alternative, we performed extensive cocrystallization screenings with PHEV S1^A, both with native and EndoH_f-treated protein samples, together with methyl-5,9-di-N-acetyl-α-neuraminoside (Neu5,9Ac₂α2Me), a receptor analog chemically more stable than the 9-O–acetylated compound (22, 34). These attempts also remained unsuccessful as no crystals were formed. In a final effort to obtain crystals of altered packing that perchance would be compatible with ligand soaking, each of the four N-glycosylation sites in PHEV S1^A were systematically removed, separately and in combination, by Asn-to-Gln substitutions. However, the resulting proteins were prone to aggregation and hence unsuitable for crystallization.

The diffraction data obtained for noncomplexed PHEV S1^A eventually allowed us to solve the apo-structure to 3.0-Å resolution (PDB ID 6QFY; SI Appendix, Fig. S1). While the results do not provide direct clues to the location of the β1CoV S RBS, they do permit a side-by-side comparison of the S1^A domains of two divergent β1CoVs. The difficulties met by us and others to identify the RBS by crystallography led us to switch strategy and to follow an alternative approach based on comparative structural analysis and visual inspection of S1^A domains using the general design of HE 9-O-Ac-Sia binding sites as a query.

As explicated by Neu et al. (35), structural comparison of viral attachment proteins in complex with their sialoglycan-based ligands may aid to define common parameters of recognition, and in turn these “rules of engagement” may be used to predict the location of Sia-binding sites in viral proteins for which such structural information is still lacking. Studies by us and others on corona-, toro-, and orthomyxoviral HE(F) proteins provide a wealth of structural data on how proteins recognize O-Ac-Sias (18–24). The O-acetyl Sia binding sites as they occur in HE lectin domains and esterase catalytic sites—despite major differences in structure and composition—conform to a common design in which ligand/substrate recognition is essentially based on shape complementarity and hydrophobic interactions, supported by protein–sugar hydrogen bonding involving characteristic Sia functions, such as the glycerol side chain, the 5-N-acyl moiety and, very often, the carboxylate. Typically, the critical O-acetyl moiety docks into a deep hydrophobic pocket (P1) and the 5-N-acyl in an adjacent hydrophobic pocket/depression (P2). P1 and P2, ∼7 Å apart in 9-O-Ac-Sia RBSs, are separated by an aromatic side chain, placed such that in the bound state the side chain intercalates between the O- and N-acyl groups, often with the ring structure positioned to allow for CH/π interactions with the O-acetyl-methyl moiety (18–20, 22–24). Site A previously proposed as the BCoV S1^A RBS (17) clearly does not conform to this signature. Visual inspection of the PHEV and BCoV S1^A structures, however, identified a region distal from site A that in many aspects does resemble a typical O-Ac-Sia binding site, with two hydrophobic pockets separated by the Trp⁹⁰ indole (Fig. 3A). We will refer to this location as site B. Interestingly, in the BCoV S1^A apo-structure, at the rim of site B, a sulfate ion is bound through hydrogen bonding with Lys⁸¹ and Thr⁸³. Its presence is of considerable significance as oxoanions in apo-structures often are indicative of and informative on interactions between the RBS and ligand-associated carboxylate moieties (36–38). Thus, in apo-structures of other Sia-binding proteins, sulfate and phosphate anions were found to mimic the Sia carboxylate in topology and sugar–protein hydrogen bonding (37, 38). Automated docking of 9-O, 5-N-Ac-Sia with the Sia carboxylate anchored at the position of the sulfate ion showed that the ligand would fit into site B, with the 9-O-acetyl group, most critical for ligand recognition (15, 39), docking into the more narrow and deeper pocket of the two (Fig. 3B). The 5-N-acyl moiety would be accommodated by a hydrophobic patch within the adjacent pocket, which is wide enough to also accept a 5-N-glycolyl group.

To explain the architecture of site B, SI Appendix, Fig. S1 presents the overall structure of BCoV S1^A and the designation of secondary structure elements. Central to S1^A is a β-sandwich scaffold, with one face packing against a separate domain containing the C-terminal region and the other covered by loop excursions that form a topologically distinct layer. The putative binding site is located within this layer at an open end of the β-sandwich with at its heart the β5-3₁₀1 region (residues 80–95). Leu⁸⁰, Leu⁸⁶, Trp⁹⁰, Phe⁹¹, and Phe⁹⁵ form a well-packed hydrophobic core that on one side interacts with the underlying β-sandwich. On the other side, along the protein surface, it is wrapped by residues from the N-terminal loop 1–β1-loop 2 segment (L1β1L2) to form site B pocket P1. L1β1L2 is locked in position through disulfide bonding of Cys²¹ and β12-residue Cys¹⁶⁵, and through extensive intersegment hydrogen bonding with β5-3₁₀1 residues, involving Ser²⁵ and Asn²⁷ via main-chain interactions with Leu⁸⁶ and Arg⁸⁸, Val²⁹ via main-chain–side-chain interaction with Ser⁸⁷, and Thr³¹ via side-chain–side-chain interaction with Trp⁹⁰.

Site B is conserved in all three β1CoVs and, from S cryo-EM structures obtained for other lineage A betacoronaviruses (26, 27), should be readily accessible also in the context of the fully folded intact spike (SI Appendix, Fig. S3). In the PHEV and BCoV S1^A crystals (17), however, site B—but notably not site A—is occluded by packing contacts, in the case of BCoV S1^A even via coincidental intermonomeric site B–site B interactions (SI Appendix, Figs. S2 and S4). Thus, the rapid disintegration of PHEV S1^A crystals during soaking may well be explained by disruption of crystal contacts in result of ligand-binding.

To test whether site B represents the true RBS, we again performed structure-guided mutagenesis of PHEV and OC43 S1^A (Fig. 3). Of note, the mutant proteins tested were expressed and secreted to wild-type levels, arguing against gross folding defects (for a quantitative comparison of the expression levels of key mutant proteins and their melting curves, see SI Appendix, Fig. S5). From the docking model, Trp⁹⁰, Lys⁸¹, and Thr⁸³ (Ser⁸³ in OC43 S1^A) are predicted to be critically involved in ligand binding. In accordance, their replacement by Ala led to complete loss of detectable binding by sp-LBA (Fig. 3C). Upon substitution of Trp⁹⁰ by Ala, binding was no longer observed for either PHEV or for OC43 S1^A, not even by HAA, be it conventional or high-sensitivity nanoparticle (NP-)HAA (40) (Fig. 3 D and E). Substitution of either Lys⁸¹ or Thr⁸³ in PHEV S1^A led to loss of detectable binding also as measured by conventional HAA at 4 °C, but for OC43 S1^A-Ser⁸³Ala and OC43 S1^A-Lys⁸¹Ala residual binding was detected. Saliently, HA by OC43 S1^A-Lys⁸¹Ala fully resolved after a shift-up to room temperature, indicating that the mutation renders the receptor–ligand interaction thermolabile and that binding affinity is more severely affected under physiological conditions (Fig. 3D). Importantly, combined substitution of Lys⁸¹ and Ser⁸³ in OC43 S1^A resulted in complete loss of detectable binding even by NP-HAA (Fig. 3E). Also, Ser⁸⁷Ala substitution, which disrupts the hydrogen bond with Val²⁹ and thereby weakens the association between L1β1L2 and β5-3₁₀1, caused loss of binding as detected by sp-LBA, and a considerable decrease in binding as measured by HAA. Similar results were obtained upon substitution of Leu⁸⁶ and Leu⁸⁰, which line the proposed P1 and P2 pockets, respectively (Fig. 3). Finally, replacement of L1β1L2 residue Thr³¹, thus abolishing the hydrogen bond with the β5-3₁₀1 residue Trp⁹⁰, reduced binding affinity in sp-LBA by more than a 1,000-fold, as demonstrated for OC43 S1^A (SI Appendix, Fig. S6).

To further test our model, we asked whether we might also identify gain-of-function mutations (i.e., substitutions within or proximal to site B that would increase S1^A binding affinity). Among the three β1CoVs, the proposed RBS itself is highly conserved. OC43 and BCoV S1^A, for example, differ at site B only at position 83, which is either a Thr (in BCoV) or Ser (in OC43). Despite this near identity, BCoV S1^A is the stronger binder of the two as indicated by a consistent 32-fold difference in binding efficiency as measured by sp-LBA (Fig. 1C). The difference at position 83 is particularly intriguing, as this residue is predicted to be involved in ligand binding through hydrogen bond formation with the Sia carboxylate (Fig. 3B). Indeed, Ser⁸³Thr substitution in OC43 S1^A resulted in a profound increase in binding affinity almost to that of BCoV S1^A (Fig. 4A).

Comparative sequence analysis of BCoV and PHEV S1^A revealed several amino acid differences proximal to site B—among which are Thr⁸⁸Arg, Thr²²Asn, and Val²⁴Ser—that would alter hydrogen bonding within L1 and between L1 and β5-3₁₀1 (Fig. 4 and SI Appendix, Fig. S1). Moreover, the latter two together create an additional N-glycosylation site in PHEV S1^A. To test whether and how these differences affect receptor binding, PHEV residues were replaced by their BCoV orthologs (Fig. 4B and C). The mutations again led to gain-of-function. Thr⁸⁸Arg substitution, disrupting a Thr⁸⁸-Asp²⁸ hydrogen bond absent in BCoV S1^A, resulted in a consistent twofold increase in binding affinity as measured by sp-LBA. Asn²²Thr substitution, eliminating the glycosylation site in PHEV S1^A, raised binding affinity even 50-fold. However, Ser²⁴Val substitution, also destroying the glycosylation site, did not have an effect in isolation. Yet, Ser²⁴Val and Asn²²Thr in combination raised PHEV S1^A-binding affinity by a further 30-fold almost to that of BCoV (Fig. 4B). Apparently, the difference in S1^A-binding affinity between PHEV and BCoV can be ascribed to the architecture of site B as determined by the L1–β5-3₁₀1 hydrogen bonding network rather than to the absence or presence of a glycan at position 22 (Fig. 4B). Changes in S1^A L1–β5-3₁₀1 association may alter pocket P1 and thereby affect ligand binding (SI Appendix, Fig. S7; note the difference in the topology of Leu⁸⁶ and Trp⁹⁰ side chains in PHEV and BCoV S1^A).

One remaining question is how Tyr¹⁶², Glu¹⁸², Trp¹⁸⁴, and His¹⁸⁵ in site A affect the RBS and why their substitution reduces ligand-binding affinity, albeit modestly compared with mutations within site B. Within the structure the sites are spaced relatively closely together and indirect effects of site A mutations can be envisaged. One possible explanation is the location of strand β12′, comprising Glu¹⁸² and Trp¹⁸⁴ right next to the first residues of L1. As we show, mutation of L1 residues Asn²², Val²⁴, Thr³¹, or of residues that interact with L1 like Ser⁸⁷ have a drastic effect on ligand binding. Through a similar mechanism, mutations that destabilize strand β12′ may in turn affect L1 and ligand binding at site B.

In summary, our analyses revealed both loss-of-function and gain-of-function mutations within or in close proximity of site B, each of which in accord with our model for S1^A binding of 9-O-Ac-Sia in β1CoVs. Moreover, they provide a plausible explanation for the modest yet reproducible loss of binding affinity upon substitution of residues previously identified by Peng et al. (17). The results therefore lead us to conclude that site B identified here represents the true S1^A RBS.

To directly test the relevance of sites A and B for infectivity, we pseudotyped G protein-deficient, luciferase-expressing recombinant vesicular stomatitis virus (VSV) with Flag-tagged BCoV S or with mutant derivatives (see Fig. 5A for a schematic outline of the experiment). Considering that for β1CoVs HE is an essential protein (10, 16), we expressed wild-type BCoV S either exclusively or together with a secretory HE–Fc fusion protein. Because HE–Fc is not membrane-associated, it will not be included in the VSV envelope, but would deplete endogenous 9-O-Ac-Sia pools in the exocytotic compartment and from the cell surface. Thus, virus release would be facilitated and inadvertent receptor association during S biosynthesis and receptor-mediated virion aggregation would be prevented. Wild-type S in the absence of HE–Fc was incorporated in VSV particles in its noncleaved 180K form (Fig. 5B, Upper). However, the resulting particles—purified and concentrated from the tissue culture supernatant by sucrose cushion ultracentrifugation—were noninfectious as measured by luciferase assay (Fig. 5C). VSV particles, produced in cells that coexpressed S and HE–Fc, carried spikes that resembled those described for BCoV and OC43 virions in that a large proportion had been proteolytically cleaved into subunits S1 and S2 (Fig. 5B, Lower) (41, 42). Moreover, these particles were infectious. Values of relative light units (RLUs) detected in inoculated cells were more than 40-fold above background. Strikingly, their infectivity was increased by a further 5.7-fold upon addition of “exogenous” HE–Fc to the inoculum (Fig. 5 A and C). Hence, these conditions were chosen to determine the effects of mutations in BCoV S. Site B mutations caused complete loss of infectivity (Trp⁹⁰Ala), or significantly reduced infectivity to extents correlating with the effect of each mutation on S binding affinity (Trp⁹⁰Ala > Lys⁸¹Ala > Thr⁸³Ala) (Figs. 3 and 5C). Particles pseudotyped with site A mutants were as infectious as S^wt-VSV. The combined data reinforce our conclusion that site B is the RBS and provide direct evidence that this site is essential for BCoV infectivity.

HKU1 is a lineage A betacoronavirus but separated from the β1CoVs by a considerable evolutionary distance. To illustrate, S1^A of HKU1 is only 55–60% identical to those of BCoV and OC43. However, HKU1, like the β1CoVs, uses 9-O-Ac-sialoglycans for attachment and the binding to these receptors is essential for infection (14). According to recent publications, the RBS would not be in S1^A, but in a downstream domain (31, 32), the main argument being that HKU1 S1^A does not detectably bind to glycoconjugates (30). Indeed, we also did not detect binding of HKU1 S1^A–Fc, at least not by standard sp-LBA and conventional HAA (Fig. 6 A and E). However, when HKU1 S1^A–Fc was tested by high-sensitivity NP-HAA with multivalent presentation of the HKU1 S1^A-Fc proteins, agglutination of rat erythrocytes was observed (Fig. 6A). Moreover, depletion of cell surface 9-O-Ac-Sia receptors by pretreatment of the erythrocytes with BCoV HE completely prevented HA (Fig. 6A). These findings conclusively demonstrate that, in fact, HKU1 S1^A does bind to 9-O-Ac-Sia in a 9-O-Ac–dependent fashion apparently through low-affinity high-specificity interactions. The detection of such interactions critically depends on multivalency of receptors and receptor-binding proteins (10, 40), and thus they are easily missed. This requirement for multivalency may also explain why preincubation of cells with soluble S1^A–Fc fusion proteins did not block HKU1 infection (32).

Interestingly, the RBS identified for β1CoV S1^A is conserved in HKU1 in sequence and structure (Fig. 6B and SI Appendix, Figs. S1B and S8), raising the question as to whether binding of HKU1 S to its sialoglycan receptor occurs through the same site. Indeed, Ala substitution of HKU1 S1^A Lys⁸⁰, Thr⁸² or Trp⁸⁹—orthologs of Lys⁸¹, Thr⁸³, and Trp⁹⁰ in BCoV S1^A—resulted in a complete loss of detectable binding (Fig. 6A; see SI Appendix, Fig. S9 for an analysis in the context of S1–Fc). To further study whether this site is involved in 9-O-Ac-Sia binding and to understand the structural basis for the considerable difference in affinity between HKU1 and β1CoV S1^A domains, we performed a side-by-side structure comparison. While in β1CoV S1^A domains the RBS is readily accessible, the corresponding site in HKU1 S1^A is located at the bottom of a canyon because it is flanked by protruding parallel ridges comprised of HKU1 S1^A residues 28 through 34 [element (e)1 corresponding to BCoV L2β2 residues 29–35], and 243 through 252 (e2 corresponding to BCoV residues 253–256 in the β18-β19 loop), each decorated with an N-linked glycan (Fig. 6B; see also SI Appendix, Fig. S1B) (27). Conceivably, this assembly would hamper binding particularly to short sialoglycans, such as the O-linked STn sugars (Sia-α2,6GalNAc-α1-O-Ser/Thr) predominantly present on BSM (43). Individual exchange of e1 and e2 in HKU1 S1^A by the corresponding segments from BCoV indeed increased receptor-binding as measured by NP-HAA by 16- or 256-fold for e1 and e2, respectively (Fig. 6C). Exchange of e2 in fact enhanced receptor affinity to levels that now also allowed detection of binding by conventional HAA, while simultaneous replacement of both elements increased binding affinity even further (Fig. 6D). Importantly, pretreatment of the erythrocytes with BCoV HE to deplete receptors by Sia-de-O-acetylation prevented hemagglutination (Fig. 6C), indicating that the enhanced binding was still 9-O-Ac-Sia–dependent. Replacement of e2 increased receptor affinity to such extent that also binding to BSM by sp-LBA assay became detectable, albeit only with nanoparticle-bound and not with free S1^A (Fig. 6E).

To study whether the N-glycans attached to e1 and e2 hinder receptor-binding and thus contribute to the apparent low affinity of HKU1 S1^A, we expressed the protein in N-acetylglucosaminyltransferase I-deficient HEK293S GnTI⁻ cells (44). Replacement of complex N-glycans by high-mannose sugars resulted in an eightfold increase in binding affinity as measured by NP-HAA (SI Appendix, Fig. S9). In agreement, combined disruption of the N-linked glycosylation sites in e1 and e2 through Gln substituting for Asn²⁹ and Asn²⁵¹ increased binding affinity to a similar extent, with the latter mutation exerting the largest effect (Fig. 6C). The deletion of glycosylation sites, however, did not enhance affinity to that of the HKU1-BCoV S1^A/e1+e2 chimera, indicating that receptor binding is hampered not only by the N-linked glycans, but also by the local protein architecture. Of note, removal of the glycan attached to Asn¹⁷¹, proximal to the binding site originally predicted by Peng et al. (17), did not enhance but actually lowered the binding affinity of HKU1 S1^A (Fig. 6C), possibly by affecting protein folding.

In summary, our results provide direct evidence that HKU1 binds to 9-O-Ac-Sias via S domain A. The data again argue against a role for site A (17), and on the basis of loss- and gain-of-function mutations strongly indicate that HKU1 binds its sialoglycan receptor via site B here identified for β1CoVs. On a critical note, we are aware that in the published HKU1 S apo-structure, the orientation of the Lys⁸¹ side chain differs from that in BCoV S1^A such that hydrogen bonding with the Sia carboxylate would be precluded. Whether this is an inaccuracy in the structure, whether the orientation of the side chain changes during ligand binding, or whether this is indeed a factual difference between HKU1 and β1CoV RBSs remains to be seen. Be it as it may, our findings do imply that in two human coronaviruses, related yet separated by a considerable evolutionary distance, the RBS and general mode of receptor-binding has been conserved. It is tempting to speculate that the apparently low affinity of HKU1 S1^A is a virus-specific adaptation to replication in the human tract. Conceivably, occlusion of the HKU1 S RBS by e1 and e2 might allow selective high-affinity binding to particular sialoglycans with the 9-O-Ac-Sia terminally linked to extended glycan chains as to prevent nonproductive binding to short stubby sugars—such as present on BSM and other mucins—that otherwise would act as decoys and inhibitors. In analogy, influenza A H3N2 variants and 2009 pandemic H1N1 (Cal/04), initially assumed to carry low-affinity HAs, were recently reported to have evolved a preference for a subset of human-type α2,6-linked sialoglycan-based receptors comprising branched sugars with extended poly-N-acetyl-lactosamine (poly-LacNAc) chains (45). Whether such an adaptation would translate in distinctive differences between OC43 and HKU1 with respect to dynamic virion–receptor interactions and cell tropism merits further study.

The 5′-terminal sequences of the S genes of BCoV strain Mebus (GenBank: P15777.1), OC43 strain ATCC VR-759 (GenBank: AAT84354.1), PHEV strain UU (GenBank: ASB17086.1), HKU1 strain Caen1 (GenBank: ADN03339.1), and mouse hepatitis virus (MHV) strain A59 (GenBank: P11224.2), encoding the signal peptide and adjacent S1^A domain, were cloned in expression vector pCAGGS-Tx-Fc (46). Domain A, defined on the basis of the cryo-EM structure of the MHV-A59 S ectodomain (26), corresponds to S amino acid residues 15–294, 15–298, and 15–294 of BCoV, OC43, and PHEV, respectively. The amino acid sequences of these S1^A–Fc fusion proteins were deposited in GenBank (MG999832-35). Recombinant S1^A proteins, genetically fused to the Fc domain of human IgG, were purified by protein A affinity chromatography from the supernatants of transiently transfected HEK293T cells as in Zeng et al. (18).

Standard HAA was performed as in Zeng et al. (18). Twofold serial dilutions of CoV S1^A–Fc proteins (starting at 2.5 μg per well; 50 μL per well) were prepared in V-shaped 96-well microtiter plates (Greiner Bio-One) unto which was added 50 μL of a rat erythrocyte suspension (Rattus norvegicus strain Wistar; 0.5% in PBS). High-sensitivity NP-HAA was performed as in Li et al. (40). In brief, S1^A–Fc proteins were complexed with pA-LS (a self-assembling 60-meric lumazine synthase nanoparticle, N-terminally extended with the Ig Fc-binding domain of the Staphylococcus aureus protein A) at a 0.6:1 molar ratio for 30 min on ice, after which complexed proteins were twofold serially diluted and mixed 1:1 with rat erythrocytes (0.5% in PBS). HA was assessed after 2-h incubation on ice unless stated otherwise. For specific depletion of cell-surface O-Ac-Sias, erythrocytes (50% in PBS pH 8.0) were (mock-)treated with soluble HE⁺ (0.25 μg/μL BCoV-Mebus and 0.25 μg/μL PToV-P4) for 4 h at 37 °C comparable to as described in Langereis et al. (47).

The 96-well Maxisorp microtitre ELISA plates (Nunc) were coated with 0.1-µg BSM (Sigma-Aldrich) per well. Conventional sp-LBAs were performed using twofold serial dilutions of CoV S1^A–Fc proteins, as described previously (20, 47). In nanoparticle sp-LBA experiments, BSM-bound nanoparticles were detected with StrepMAb-HRP via the C-terminally appended Strep-tag of the pA-LS proteins, as described previously (40). Receptor-depletion assays were performed by (mock-)treatment of coated BSM with twofold serial dilutions of soluble HE⁺ in PBS (100 μL per well) starting at 0.6 ng/μL (BCoV-Mebus) for 2 h at 37 °C, as in Langereis et al. 47. Receptor destruction was assessed by sp-LBA with β1CoV S1^A-Fc at fixed concentrations (BCoV 0.3 ng/μL; OC43 1.2 ng/μL).

The starting material Neu5Ac2αMe (48) was treated with p-toluenesulfonyl chloride (1.8 eq.) in pyridine overnight and then with sodium azide (5 eq.) in DMF at 70 °C. The resulting intermediate was subjected to Staudinger reduction by 1 M trimethylphosphine in toluene (3 eq.) in the presence of potassium hydroxide. Upon completion, acetyl chloride (4 eq.) was added directly. After 5 min, 1 M potassium hydroxide solution was added to quench the reaction. The mixture was neutralized with acidic resin. The residue was purified with silica gel and then p-2 biogel to give pure Neu5,9NAc₂2αMe at a yield of 31% over four steps. The product was dissolved in 50 mM ammonium bicarbonate and freeze-dried, which was repeated three times to give the ammonium salt form. The final product was analyzed by NMR spectroscopy (SI Appendix, Fig. S10).

Crystallization conditions were screened by the sitting-drop vapor-diffusion method using a Gryphon (Art Robins). Drops were set up with 0.15 μL of S1^A dissolved to 11 mg/mL in 10 mM BisTris-propane, 50 mM NaCl, pH 6.5, and 0.15-μL reservoir solution at room temperature. Diffracting crystals were obtained from the JCSG⁺ screen [200 mM MgCl₂, 100 mM Bis⋅Tris pH 5.5, 25% (wt/vol) PEG3350] (JCSG Technologies) at 18 °C. Crystals were flash-frozen in liquid nitrogen using reservoir solution with 30% (wt/vol) glycerol as the cryoprotectant. Crystals were soaked with methyl-5,9-di-N-acetyl-α-neuraminoside and methyl-5-N-acetyl-4,9-di-O-acetyl-α-neuraminoside, as described, for the latter ligand, with batches previously used to solve HE holo-structures (18–20, 22).

Diffraction data of PHEV S1^A crystals were collected at European Synchrotron Radiation Facility station ID30A-3. Diffraction data were processed using XDS (49) and scaled using Aimless from the CCP4 suite (50). Molecular replacement was performed using PHASER (51) with BCoV S1^A residues 35–410 as template (PDB ID code 14H4). NCS-restrained structure refinement was performed in REFMAC (52), alternated with model building in Coot (53); water molecules were built in difference density peaks of at least 5.0 σ located at buried sites conserved and occupied by water in the 1.5-Å resolution structure of BCoV S1^A. Molecular graphics were generated with PYMOL (pymol.sourceforge.net). See Table 1 for X-ray data statistics for PHEV S1^A.

Molecular docking of 9-O-Sia in the crystal structure of apo-BCoV S1^A (PDB ID code 4H14) was performed with AutoDock4 (22) similar to the procedure described by Bakkers et al. (22). The ligand molecule was extracted from BCoV HE (PDB ID code 3CL5). During docking, the protein was considered to be rigid. An inverted Gaussian function (50-Å half-width; 15-kJ energy at infinity) was used to restrain the Sia-carboxylate near the position occupied by the sulfate ion in the BCoV S1^A crystal structure. The initial ligand conformation was randomly assigned, and 10 docking runs were performed.

Recombinant G protein-deficient VSV particles were pseudotyped as described previously (54), with the S protein of BCoV strain Mebus. To allow transport of S to the cell surface and its incorporation into VSV particles, a truncated version of S was used from which the C-terminal 17 residues, comprising an endoplasmic reticulum-retention signal, had been removed. To facilitate detection, the S protein was provided with a Flag-tag by cloning its gene in expression vector pCAGGS-Flag. HEK 293T cells at 70% confluency were transfected with PEI-complexed plasmid DNA, as described previously (18). For coexpression of BCoV S and HE–Fc, S expression vectors and pCD5-BCoV HE–Fc (18) were mixed at molar ratios of 8:1. At 48 h after transfection, cells were transduced with VSV-G–pseudotyped VSV∆G/Fluc (54) at a multiplicity of infection of 1. Cell-free supernatants were harvested at 24 h after transduction, filtered through 0.45-µm membranes, and virus particles were purified and concentrated by sucrose cushion ultracentrifugation at approximately 100,000 × g for 3 h (10). Pelleted virions were resuspended in PBS and stored at −80 °C until further use. Relative virion yields were determined on the basis of VSV-N content by Western blot analysis by using anti–VSV-N monoclonal antibody 10G4 (Kerafast). Uptake of BCoV S into virus particles was detected by Western blot analysis with monoclonal antibody ANTI-FLAG M2 (Sigma). Inoculation of HRT18 monolayers in 96-well cluster format was performed with equal amounts of S-pseudotyped VSVs, as calculated from VSV-N content (roughly corresponding to the yield from 2 × 10⁵ cells from each transfected and transduced culture), diluted in 10% FBS-supplemented DMEM. For virus infections with “exogenous” HE, 12.5 mU/mL of BCoV HE–Fc protein was added to the inoculum. At 18 h postinfection, cells were lysed using passive lysis buffer (Promega). Firefly luciferase expression was measured using a homemade firefly luciferase assay system, as described previously (55). Infection experiments were performed independently in triplicate, each time with three technical replicates.