Crystal structure of mouse coronavirus receptor-binding domain complexed with its murine receptor

To prepare an MHV-A59 NTD fragment suitable for crystallization, we designed a series of C-terminal truncation constructs of MHV-A59 NTD based on its secondary structure predictions. One NTD fragment containing residues 1–296 was well expressed in insect cells and stable in solution. It bound to mCEACAM1a[1,4] to form a 1:1 heterodimeric complex with K_d of 21.4 nM (Fig. 1B). We crystallized this complex in space group P6₁22, a = 76.4 Å, b = 76.4 Å, and c = 942.1 Å (Table S4), with two complexes per asymmetric unit (Fig. S1). The structure was determined by single-wavelength anomalous diffraction (SAD) phases using selenomethionine-labeled mCEACAM1a. The phases were subsequently improved by an averaging method (35). We refined the structure at 3.1Å resolution (Table S4). The final model contains residues 15–268 of NTD (except for a disordered loop from residues 40–64) and residues 1–202 of mCEACAM1a. The model also contains glycans N-linked to viral residue 192 and to receptor residues 37, 55, and 70.

The core of MHV NTD is a 13-stranded β-sandwich with two antiparallel β-sheets stacked against each other through hydrophobic interactions (Fig. 1 C and D and Fig. 2A). Surprisingly, this β-sandwich core of MHV NTD has a galectin fold, which will be discussed in detail later. Outside the core structure, MHV NTD contains several peripheral structural elements (Fig. 1C and Fig. 2A). Three loops from the “upper” β-sheet of the core converge with the N-terminal segment to form a distinct, receptor-contacting substructure; these four structural elements are termed receptor-binding motifs (RBMs). Three disulfide bonds, connecting cysteines 21–158, 165–246, and 153–187, reinforce MHV NTD (Fig. 2B).

MHV NTD binds to domain D1 of mCEACAM1a (Fig. 1C and Fig. 2A). There is no significant structural change in mCEACAM1a before and after MHV binding (Fig. S2). The four RBMs of MHV NTD contact two virus-binding motifs (VBMs) on the CC′C′′ face of mCEACAM1a (Fig. 1C and Fig. 2A). A total of 14 residues in NTD interact with a total of 17 residues in mCEACAM1a (Fig. 3 A–C). The binding buries 1,500 Ų at the interface (Fig. 2B). This interface is intermediate between the SARS-CoV/ACE2 interface (1,700 Ų) and the HCoV-NL63/ACE2 interface (1,300 Ų), but all of the three viral receptor-binding domains bind to their respective protein receptors with similar affinities (23). Notably, none of the observed or predicted glycans is involved in MHV-NTD/mCEACAM1a interactions (Fig. 2B), and therefore MHV-NTD/mCEACAM1a binding depends exclusively on protein–protein interactions.

The interface between MHV NTD and mCEACAM1a is dominated by hydrophobic interactions with scattered polar interactions. Two hydrophobic patches stand out. One centers on Ile41 from the CC′ loopof mCEACAM1a. Ile41 is surrounded by the hydrophobic side chains of MHV Tyr15, Leu89, and Leu160 and the aliphatic side chains of MHV Gln159 and Arg20 (Fig. 2C). Additionally, MHV Arg20 forms a bifurcated hydrogen bond with the main chain carbonyl group of receptor Thr39 and another hydrogen bond with MHV Gln159. Receptor Arg96 forms a bifurcated salt bridge with receptor Asp89, while stacking with MHV Arg20. The second critical hydrophobic patch involves multiple hydrophobic residues from both MHV and mCEACAM1a that include MHV residues Ile22 and Tyr162 and receptor residues Val49, Met54, and Phe56 (Fig. 2D). Additionally, MHV Asn26 forms a hydrogen bond with the main chain amide group of receptor Thr57. In protein–protein interactions, hydrophobic interactions contribute more to binding energy, whereas hydrophilic interactions contribute more to binding specificity. The above key hydrogen bonds between NTD side chains and the receptor main chain help bring the adjacent hydrophobic patches into place. These structural analyses suggest that the hydrophobic patches and additional polar interactions provide significant binding energy and specificity to MHV-NTD/mCEACAM1a binding interactions.

The importance of contact residues at the NTD/mCEACAM1a interface has been confirmed by mutagenesis data. Here we compared the efficiency of mCEACAM1a-dependent cell entry by lentiviruses pseudotyped with wild-type or mutant MHV-A59 spike protein (Fig. 3D). Our data, together with published data (Fig. 3E), showed that single mCEACAM1a substitution I41G and single NTD substitutions I22A, Y162A, Y162Q, Y162H, but not Y162F, abrogated viral infectivity (28, 36), underscoring the significance of the two hydrophobic patches. Furthermore, single NTD substitutions R20A, R20K, and N26A significantly decreased viral infectivity, confirming the significance of the hydrogen bonds between Arg20 and mCEACAM1a and between Asn26 and mCEACAM1a. A naturally occurring Q159L mutation in MHV NTD caused small viral plaques (Fig. 3E) (37), suggesting that the hydrogen bond between Gln159 and Arg20 helps position Arg20 to interact with mCEACAM1a. Therefore, through evolution MHV-A59 appears to have optimized many of its interactions with mCEACAM1a, and thus substitutions in MHV NTD that disrupt these specific molecular interactions weaken or abrogate viral infectivity.

Our study provides the structural basis for the viral and host specificities of coronavirus/CEACAM1 interactions. On the basis of structural analyses and mutagenesis data, we determined that MHV NTD contains Arg20, Ile22, Asn26, and Tyr162, all of which form energetically favorable interactions with mCEACAM1a, whereas other group 2a coronavirus NTDs contain residues at the corresponding positions that are expected to disrupt critical hydrophobic or polar interactions with mCEACAM1a (Fig. 3B and Fig. S3). On the other hand, on the basis of structural analyses, we found that mCEACAM1a contains Ile41, Val49, Met54, and Phe56, all of which form energetically favorable interactions with MHV, whereas mCEACAM1b, bovine CEACAM1a and -1b, and human CEACAM1 contain residues at the corresponding positions that likely disrupt these favorable interactions with MHV (Fig. 3C and Fig. S3). For example, hydrophobic residues Ile41 and Phe56 in mCEACAM1a become hydrophilic residues Thr41 and Thr56 in mCEACAM1b (Fig. 3C and Fig. S3). These results reveal the mechanisms whereby MHV uses only mCEACAM1a, and not mCEACAM1b or CEACAM1 from cattle or humans, as its receptor, and whereby other group 2a coronaviruses cannot use mCEACAM1a as a receptor.

The β-sandwich core of MHV NTD shares the same 11-stranded fold as human galectins (S-lectins) and rotavirus VP4 (viral lectin) (38, 39), augmented by two additional β-strands in the “lower” β-sheet (Fig. 4 and Fig. S4). MHV NTD and human galectin-3 have a Dali Z-score of 7.8 and an rmsd value of 2.9 Å over 137 matching Cα atoms (40). Importantly, the topologies of their β-sandwich cores are identical (Fig. S4). This unexpected structural homology between MHV NTD and human galectins suggests that coronavirus NTDs may function as viral lectins. To test this possibility, we designed NTD constructs of other group 2 coronaviruses that correspond to the crystallized MHV-NTD fragment based on the sequence alignment of these spike proteins. We expressed and purified each of these coronavirus NTDs and detected their binding interactions with mucin, a mixture of highly glycosylated proteins containing all of the sugar moieties (Neu5,9Ac2, Neu5Gc, and Neu5Ac) recognized by the coronavirus spike proteins. Results showed that NTDs of HCoV-OC43 and BCoV bound sugars, whereas NTDs of MHV-A59, HCoV-HKU1, and SARS-CoV did not (Fig. 5). Removal of neuraminic acids from mucin by neuraminidase treatment prevented binding of HCoV-OC43 or BCoV NTD.

Why do the NTDs of HCoV-OC43 and BCoV, but not that of MHV, bind sugars, and where is the sugar-binding site located in coronavirus NTDs? In human galectin-3 that binds galactose, the sugar-binding site (site A) is located above the β-sandwich core and involves the 10–11 loop (loop connecting β10 and β11) (Fig. 4 B and E) (39). In rotavirus VP4, which binds sialic acids, site A is blocked by a two-stranded β-sheet; instead, the sugar-binding site (site B) is located in a groove between the two β-sheets of the β-sandwich core (Fig. 4 C and F) (38). In MHV NTD, site B is blocked due to the narrowed groove between the two β-sheets of the β-sandwich core, whereas site A is open and available (Fig. 4 A and D). However, compared with human galectin-3, MHV NTD has a markedly shortened 10–11 loop that may be responsible for its lack of lectin activity. HCoV-OC43 and BCoV NTDs likely share the same galectin fold as MHV NTD due to their high sequence similarities (Fig. S1), but they both contain longer 10–11 loops than MHV NTD (Fig. 3B and Fig. S3) and thus may use site A for sugar binding. To test this hypothesis, we modified the 10–11 loops in both BCoV and HCoV-OC43 NTDs, using MHV NTD as a reference (Fig. 3B and Fig. S3). For both BCoV and HCoV-OC43 NTDs, the mutant and wild-type proteins were equally well expressed and stable in solution, but the mutant proteins (OC43* and BCoV*) lacked sugar-binding activities (Fig. 5). These observations confirm that the 10–11 loops are critical for sugar binding in both BCoV and HCoV-OC43 NTDs. A more refined description of the sugar-binding site in BCoV and HCoV-OC43 NTDs awaits future structural and biochemical studies.

To date, three crystal structures are available for RBDs of coronavirus S1: group 2a MHV NTD, group 2b SARS-CoV C domain (24), and group 1 HCoV-NL63 C domain (23) (Fig. 6). Because of the significant sequence similarities of the S1 subunits of the spike proteins within each coronavirus group, the six-stranded β-sandwich core structure of the HCoV-NL63 C domain likely exists in other group 1 coronaviruses (23), and the 5-stranded β-sheet core structure of the SARS-CoV C domain likely exists in other group 2 coronaviruses (24). Similarly, the galectin-like NTD of MHV likely exists in other group 2 coronaviruses. The folds of group 1 and group 3 coronavirus NTDs are less clear. However, because both TGEV NTD and IBV S1 have lectin activities, the galectin-fold core structure of group 2a coronavirus NTDs may also be found in both group 1 and group 3 coronaviruses in similar or variant forms. The present study advances our understanding of the structures and functions of coronavirus spike proteins and the complex receptor-recognition mechanisms of coronaviruses.

How did coronavirus spike NTDs originate and evolve? We propose that an ancestral coronavirus acquired a galectin-like domain from its host. Subsequently, an ancestral group 2a coronavirus incorporated a HE gene into its genome to aid viral detachment from sugars on infected cells. Later, the galectin-like NTD of MHV evolved additional novel structural elements that allowed it to bind mCEACAM1a. Using a protein receptor instead of sugar receptors greatly enhanced the attachment affinity between MHV and murine cells, making sugar-binding functions dispensable. Accordingly, MHV-A59 underwent changes in the sugar-binding site of its NTD, lost its sugar-binding activity, and stopped expressing its HE gene. In contrast, the galectin-like NTDs of some contemporary coronaviruses such as HCoV-OC43, BCoV, and TGEV retain the lectin activity, although their sugar specificities have diverged in three coronavirus groups and differ from those of contemporary human galectins. Some TGEV strains deleted their NTD to become PRCoV after their C domain acquired APN-binding affinity. In addition to coronaviruses, there is evidence that paramyxoviruses may also have acquired their RBD from a host (although with a β-propeller fold) and used it to bind sugar or protein receptors (41, 42). It seems that viruses use a common evolutionary strategy by acquiring host proteins and evolving them into viral RBDs with different receptor specificity. This strategy allows viruses to explore novel cellular receptors and expand their host ranges. The current study provides critical structural information that illustrates how coronaviruses have successfully used this strategy.

Both MHV-A59 S1 NTD (residues 1–296) and mCEACAM1a[1,4] (residues 1–202) were expressed in Sf9 insect cells using the bac-to-bac system (Invitrogen) and then purified as previously described (24). Briefly, the proteins containing C-terminal His tags were harvested from Sf9 cell supernatants, loaded onto a nickel-nitrilotriacetic acid (Ni-NTA) column, eluted from the Ni-NTA column with imidazole, and further purified by gel filtration chromatography on Superdex 200 (GE Healthcare). To purify the NTD/mCEACAM1a complex, NTD was incubated with excess mCEACAM1a before the complex was purified by gel filtration chromatography and concentrated to 10 mg/mL. Crystals of the NTD/mCEACAM1a complex were grown in sitting drops at 4 °C over wells containing 10% PEG6000 and 0.2 M CaCl2. Crystals were harvested in 2 wk; stabilized in 10% PEG6000, 0.2 M CaCl2, and 30% ethylene glycol; and flash-frozen in liquid nitrogen.

Selenomethionine-labeled mCEACAM1a was expressed in Sf9 cells as previously described (43). Briefly, 24 h postinfection, cells were transferred into medium without methionine for methionine depletion. After 4 h, cells were transferred into medium without methionine but supplemented with 50 mg/mL selenomethionine for selenomethionine labeling. After 36 h, cells were harvested and selenomethionine-labeled protein was purified using the same procedure as above.

X-ray data were collected at Advanced Photon Source Northeastern Collaborative Access Team beamlines. The crystal contains two complexes per asymmetric unit. From a selenomethionine-labeled crystal, 12 selenomethionine sites in mCEACAM1a were identified. SAD phases were then calculated. The phases were subsequently improved by twofold noncrystallographic symmetry (NCS) averaging within the crystal and cross-crystal averaging with the mCEACAM1a crystal (35). During both density modification and structure refinement, the twofold NCS restraint was applied to each of the three domains—NTD and domains D1 and D4 of mCEACAM1a—between the two complexes in each asymmetric unit. This is because domains D1 and D4 undergo a hinge movement relative to each other in the two complexes due to the flexibility of the domain linker. The structure was refined to a final R_free of 30.8% and R_work of 24.8%. Data, phasing, and refinement statistics are shown in Table S4. Software used for data processing, structure determination, and refinement is also listed in Table S4.

The kinetics and binding affinity of MHV S1 NTD and mCEACAM1a were measured by surface plasmon resonance using a Biacore 3000. The surface of a C5 sensor chip was first activated with N-hydroxysuccinimide, MHV S1 NTD was then injected and immobilized to the surface of the chip, and the remaining activated surface of the chip was blocked with ethanolamine. Soluble mCEACAM1a was introduced at a flow rate of 20 μL/min at different concentrations. Kinetic parameters were determined using BIA-EVALUATIONS software.

Lentiviruses pseudotyped with MHV-A59 spike protein were produced as previously described (44). Briefly, plasmid encoding wild-type or mutant MHV-A59 spike protein was cotransfected into 293T cells with helper plasmid psPAX2 and reporter plasmid pLenti-GFP at molar ratio 1:1:1 using polyethyleneimine (Polysciences Inc.). Forty-eight hours posttransfection, the resulting pseudotyped viruses were harvested and inoculated onto the 293 cells expressing mCEACAM1a. After overnight, cells were washed twice with PBS and lifted by 100 μL of 0.25% trypsin and 0.38 mg/mL EDTA. After being washed twice with PBS, cells were fixed with 1% paraformaldehyde and analyzed for GFP expression by flow cytometry. All experiments were repeated at least three times. The expression levels of spike proteins in pseudotyped viruses were measured by Western blotting using polyclonal goat antibody to MHV-A59 spike protein (Fig. S5).

Sugar-binding assays of coronavirus S1 NTDs were performed as previously described (13, 33). Briefly, for the dot-blot overlay assay, 10 μg of bovine submaxillary gland mucin (BSM) (Sigma-Aldrich) was spotted onto nitrocellulose membranes. The membranes were dried completely, blocked with BSA at 4 °C overnight, and either mock-treated or treated with 20 mU/mL Arthrobacter ureafaciens neuraminidase (Roche Applied Science) at 25 °C for 2 h. The membranes were then incubated with 1 μM coronavirus S1 NTDs containing a C-terminal His tag at 4 °C for 2 h, washed five times with PBS, incubated with anti-His antibody (Invitrogen) at 4 °C for 2 h, washed five times with PBS again, incubated with HRP-conjugated goat anti-mouse IgG antibody (1:5,000) at 4 °C for 2 h, and washed five times with PBS. Finally, the bound proteins were detected using a chemiluminescence reagent (ECL plus, GE Healthcare). For the ELISA, BSM (60 μg/mL in PBS) was coated at 4 °C overnight in the wells of 96-well Maxisorp plates (Nunc). The wells were treated in the same way as in the dot-blot overlay assay. Femto-ELISA-HRP substrates were added and incubated at room temperature. The reaction was stopped with 1 N HCl, and the absorbance of the resulting yellow color was read at 450 nm.