Identification of VNARs against SARS-CoV-2
After four rounds of biopanning against the SARS-CoV-2 RBD with the ELSS libraries each containing ~10 billion clones, we isolated nearly two dozen unique VNAR domains that bound to the RBD by ELISA (Fig. 1a). As a primary method for identifying potent inhibitors of viral entry, we screened VNARs in a luciferase-based infectivity assay. Increasing concentrations of VNARs were used to neutralize pseudotyped SARS-CoV-2 and SARS-CoV-1 in ACE2-expressing HEK293T cells using luciferase activity as a readout for viral infectivity (Fig. 1b). As a positive control for our primary screen, we used VHH-72-Fc, a previously identified bivalent single-domain camelid antibody14. In our hands, VHH-72-Fc potently neutralized both SARS-CoV-2 and SARS-CoV-1 (IC50 = 1.49 nM ± 0.25 nM and IC50 = 20.2 nM ± 2.7 nM, respectively), values consistent with the literature. From this screen, VNARs with IC50 values below 10 nM were prioritized for further characterization (Fig. 1c). This resulted in the identification of three lead VNARs: 3B4, 2C02, and 4C10. For subsequent controls, we selected a VNAR with relatively low neutralization potency (2D01) and a non-targeting naïve VNAR (2V).
a ELISA screen for identification of potential SARS-CoV-2 RBD binders. Negative control wells containing expression media, and positive control wells containing VNAR E06 (anti-serum albumin) and rabbit monoclonal CR3022-RB (anti-SARS-CoV-2 Spike) are indicated. b Primary screen for identification of neutralizing VNAR domains. Concentration-dependent neutralization of pseudotyped SARS-CoV-2 (black) or SARS-CoV-1 (red) in HEK293T cells transiently expressing ACE2. Data represents mean ± s.e.m. relative luminescence units (RLUs) from n = 3 independent biological experiments. c Left, rank-ordered IC50 values for neutralizing VNARs from panel (b). VNARs with IC50 < 10 nM (dashed line) were selected for further characterization. Upper right, depiction of primary sequences of selected VNARs, relative length of complementarity determining regions (CDR1, CDR3) and location of cysteine residues (teal) are shown. d Phylogenetic tree of selected virus taxa, divergent lineages of betacoronaviruses are shown. Glycoproteins encoded by the indicated viruses (*) were used to generate pseudoviruses. e Heatmap summarizing IC50 values for neutralization of the indicated pseudovirus with the indicated VNAR antibody. Values are derived from experiments described in (f). f Secondary validation of selected neutralizing VNAR domains. Concentration-dependent neutralization of viral particles pseudotyped with glycoproteins natively encoded by either SARS-CoV-2 (black), SARS-CoV-1 (red), WIV1-CoV (blue). MERS-CoV (green), or VSV (purple) in Calu-3 cells. Cell viability was also assessed in the presence of increasing concentrations of VNARs (yellow). Data represents mean ± s.e.m. RLU values from n = 3 independent biological experiments. g Concentration-dependent neutralization of replication-competent authentic SARS-CoV-2, strain USA_WA1/2020 in Vero E6 cells. Data represents mean ± s.e.m. RFU values from n = 3 independent biological experiments.
Virus neutralization experiments were next conducted against a series of pseudotyped viruses (Fig. 1D). These secondary validation experiments were performed in Calu-3 cells, an ACE2-expressing human bronchial epithelia cell line that is susceptible to infection by beta-coronaviruses15. Our four VNARs were tested for concentration-dependent neutralization of pseudotyped SARS-CoV-2 and SARS-CoV-1 in this cell line. We also screened for cross-reactivity with the pre-emergent ‘SARS-like’ coronavirus, WIV1-CoV, which currently circulates throughout bat populations and displays human tropism via hACE216. Additionally, we assessed neutralization of pseudotyped Middle East respiratory syndrome coronavirus (MERS-CoV), a phylogenetically distant beta-coronavirus which utilizes human DPP4 as a host-cell receptor17,18. As a negative control, we screened our panel of VNARs for neutralization of viral particles pseudotyped with vesicular stomatitis virus glycoprotein (VSV-G), an unrelated enveloped virus.
The neutralization potency of the VNARs (3B4, 2C02, 4C10) against pseudotyped SARS-CoV-2 and SARS-CoV-1 was faithfully reproduced in Calu-3 cells and HEK293T-hACE2 cells (Fig. 1d–f, Supplementary Fig. 1 Supplementary Table 1). As expected, VNAR-2D01 displayed ~10-fold lower potency compared to 3B4, 2C02 or 4C10 (Fig. 1d, Supplementary Table 1). High concentrations of non-targeted VNAR-2V had no effect on viral infectivity (Fig. 1f). Each VNAR effectively neutralized viral infectivity of pseudotyped WIV1-CoV suggesting that each VNAR likely binds to epitopes conserved among ‘lineage B’ beta-coronaviruses17. In contrast, VNAR-3B4 was uniquely effective at neutralizing cellular entry of viral particles pseudotyped with MERS-CoV spike proteins, a ‘lineage C’ beta-coronavirus, despite sharing only 32% sequence homology with SARS-CoV-2 Spike (Fig. 1d–f)17. This observation suggests that VNAR-3B4 likely binds to an interface that is evolutionarily conserved among beta-coronavirus lineages. Importantly, the inhibitory activity observed by these VNAR antibodies was specific to coronaviruses and was not due to deleterious effects on cell health, as demonstrated by the lack of effect on either pseudotyped VSV infectivity or cell viability (Fig. 1e, f).
Finally, we tested the neutralizing efficacy of the VNARs against replication-competent SARS-CoV-2 (strain USA_WA1/2020) in Vero E6 cells (Fig. 1g). In this system, VNAR-3B4 displayed a modest loss of potency (IC50 = 11.5 nM ± 5.4 nM), while VNARs-2C02 and 4C10 were both found to be ~10-fold more potent compared to data collected from pseudovirus experiments (IC50 = 0.84 nM ± 0.15 nM, and IC50 = 0.61 nM ± 0.26 nM, respectively). As expected, 2D01 was the least potent anti-SARS-CoV-2 RBD VNAR antibody (IC50 = 4.6 μM ± 1.2 μM), and VNAR-2V failed to have any impact on viral infectivity. IC50 values for neutralization of replication-competent SARS-CoV-2 are collated in Table 1. Altogether, these data demonstrate that VNARs 3B4, 2C02, and 4C10 are potent and effective monomeric anti-viral VNARs.
Structural basis for SARS-CoV-2 neutralization by VNARs
To understand the mechanism of neutralization, we determined the crystal structure of SARS-CoV-2 spike RBD in complex with VNARs 3B4 and 2C02 at 1.92-Å and 1.96-Å resolution, respectively. The structures show that VNARs 3B4 and 2C02 recognize distinct epitopes on the RBD surface (Fig. 2a, Supplementary Fig. 2), neither of which overlaps with the ACE2 receptor interface. VNAR-3B4 binds distal to the ACE2 binding interface with no direct interaction with the residues involved in ACE2 recognition. This epitope is only accessible to 3B4 when the RBD is in the “up” conformation and is blocked by the N-terminal domains (NTD) of other spike protomers when the RBD is in the “down” position (Fig. 2b, upper inset). Alignment of the 3B4 crystal structure with an available cryo-EM structure of the full spike protein bound to the host receptor ACE2 shows that the framework of 3B4 likely clashes with ACE2 when the RBD is in the “up” conformation. This suggests that the mode of neutralization for 3B4 is through steric occlusion rather than direct competition with the ACE2 binding site (Fig. 3a). This mechanism is similar to Class 1 NAbs as characterized by Barnes, et al, which bind only to “up” RBDs and also block ACE2 binding in some form19.
a Transposed structures of VNAR 3B4 (blue) and 2C02 (yellow) complexes aligned to the RBD colored white as a surface representation. b SARS-CoV-2 RBD shown as a surface representation in gray with the ACE2 binding interface colored purple. A cartoon representation of VNAR-3B4 depicted in blue bound to the RBD. Inset shows a surface view of the top of the full-spike trimer with the RBD colored red in the “RBD-down” conformation and colored green in the “RBD-up” conformation with VNAR-3B4 in blue. c A 180-degree rotation of the SARS-CoV-2 RBD is shown as in Panel A with a cartoon representation of VNAR-2C02 (yellow) bound to the RBD. Inset shows a surface view of the top of the full-spike trimer with the RBD colored as in Panel A with two VNARs colored in yellow.
a Inset shows the crystal structure of the spike trimer with the RBD in the “up” orientation (green) bound to ACE2 (purple) (PDB ID: 7DF4). An alignment of the RBD (white) from the ACE2 (purple) and the VNAR-3B4 (blue) crystal structures. b Inset shows the crystal structure of the spike trimer with the RBD in the “up” orientation bound to ACE2 (purple) (PDB ID: 7BF4). An alignment of the RBDs (white) from the ACE2 (purple) and the VNAR-2C02 (yellow) crystal structures. c Figure inset shows the orientation of the main panel with a top-surface view of the spike trimer with numbered protomers and all RBDs in the “down” position (aligned from PDB: 7BF4). The figure shows the VNAR-2C02 structure (yellow) aligned to the “down” RBD from protomer 1 (red) and proximal to the NTD from protomer 3 (cyan).
The structure of the RBD in complex with VNAR-2C02 revealed that it binds to the opposite side of the RBD than VNAR-3B4. This epitope is accessible to VNAR-2C02 when the RBD is in the “up’ or “down” conformation (Fig. 2c, lower inset). In contrast to VNAR-3B4, superposition of the VNAR-2C02 crystal structure with the cryo-EM structure of the spike trimer in complex with ACE2 shows that VNAR-2C02 does not come into close contact with ACE2, suggesting that neutralization is not a result of a steric clash with ACE2, but rather through allosteric effects that decrease the population of “up” RBDs that are available to bind ACE2 for viral entry (Fig. 3b). This mechanism is in line with Class 3 NAbs that bind outside of the ACE2 interface19. Interestingly, alignment of the VNAR-2C02 structure to an RBD in the “down” conformation in the full-trimer reveals that it binds in a cleft that is formed between the RBD from protomer 1 and the NTD of protomer 3 (Fig. 3c). It is possible that the enhanced efficacy of VNAR-2C02 observed in the viral neutralization assays is a result of additional interactions between VNAR-2C02 and the NTD in this cleft. In this mode, VNAR-2C02 would act to pin the “down” RBD and NTD together and prevent the RBD from sampling the “up” conformation that is necessary for attachment to ACE2, though this remains to be confirmed structurally. This indirect mechanism for blocking ACE2 binding would pair well with other NAbs that directly block ACE2, such as 3B4. The combined neutralization mechanisms of 3B4 and 2C02 would therefore be most therapeutically beneficial when co-administered.
The small binding profile of this class of antibody allows VNARs to pack tightly against the RBD and access binding motifs not accessible to conventional antibodies. The primary interaction interface of VNARs 3B4 and 2C02 covers 734 Å2 and 792 Å2, respectively. Each VNAR covers less area than a standard bivalent antibody, yet still maintains exquisite target specificity. For VNAR-3B4, its interaction was dependent on only 5 residues in the CDR3 region of the VNAR and 7 total residues in the spike RBD. Residues Glu122, His124, Asp126 of CDR3 form an anti-parallel β-sheet with residues Ser375, Phe377, and Cys379 in the β2-strand of the RBD (Fig. 4a). The extensive hydrogen bond (h-bond) pairing of these backbone residues likely accounts for much of the affinity of the interaction, and because it is not dependent on sidechain participation, it makes this part of the interaction primarily residue independent. All three residues of 3B4 that form direct β-sheet interactions, Glu122, His124, and Asp126, also participate in some form of sidechain h-bonding. In addition to the backbone-backbone interactions above, there is also an elaborate h-bond network formed between the terminal amines of Arg103 of 3B4 with the backbone carbonyls of Ala372 and Phe374 that is supported by a side-on intramolecular interaction between Asp126 and Arg103 of 3B4 (Fig. 4b). Another sidechain-backbone interaction occurs between His124 of VNAR-3B4 and the backbone carbonyl of RBD Tyr369. Two sidechain-sidechain h-bonds complete the extent of the interaction – one between Asp107 of 3B4 and Tyr369 of the RBD and a second between Glu122 of VNAR-3B4 and Ser383 of the RBD (Fig. 4b).
a Stick representation of the peptide backbone in the primary interaction interface. Interactions between residues Ser-375, Phe-377 and Cys-379 of the RBD (white) and Glu-122, His-124, and Asp-126 of 3B4 (blue) are shown as black dashes. Inset shows the orientation of the zoomed in view. b A 180-degree view of the interaction interface. Interactions between residues of the RBD (white) and 3B4 (blue) are shown as yellow dashes with black dashes. Inset shows the orientation of the zoomed in view. c Transparent spheres surrounding the stick representations of residues involved in the hydrophobic interactions between the RBD (white) and VNAR-2C02 (yellow). Inset shows the orientation of the zoomed in view. d A 180-degree view of the interaction interface between the RBD (white) and 2C02 (yellow). Residues are shown as sticks with yellow dashes representing hydrogen bonds between atoms. Rounded insets show an alternate view of interactions that are partially obscured from view in the main panel. Top inset shows the orientation of the zoomed in view.
The interaction of VNAR-2C02 is dependent on 9 residues separated between the HV2 and CDR3 regions of the VNAR and 12 total residues in the spike RBD. While the 3B4 interaction depended primarily on h-bonding, the interaction of 2C02 largely relies on hydrophobic interactions. Hydrophobic residues Ala348, Ala352, Leu452, Ile468, Phe490, and Leu492, along with the aromatic residue Tyr351, create a nonpolar patch on the RBD which then interfaces with Tyr54, Leu105, and Phe116 of VNAR-2C02 (Fig. 4c). Though Thr53 of VNAR-2C02 is considered polar, the nonpolar methyl group of its side chain appears to be taking part in the hydrophobic core of this interaction. A number of h-bonds help to stabilize the binding of VNAR-2C02 to the RBD (Fig. 4d). RBD residues Asn354 and Arg466 form an h-bond network with Asn115 of 2C02, though the electron density of the crystal structure indicates that multiple rotameric states of Arg466 are possible. Multiple rotamers were also revealed in the density for Arg346 on the RBD indicating that Arg346 transiently h-bonds with Tyr118 or the backbone carbonyl of His117 in VNAR-2C02 (Fig. 4d, oval inset). Arg67 of VNAR-2C02 takes part in two interactions: first, it forms a salt bridge with RBD Glu484, and second, it forms a cation-pi interaction with RBD Phe490. Lastly, Ser63 forms a backbone h-bond with the carbonyl of RBD Gly446.
VNARs cross-react with closely related coronaviruses
In order to understand how our VNARs might bind to the RBD of related coronaviruses, we performed sequence alignments of the three coronaviruses and homology modeling of VNAR-3B4 bound to SARS-CoV-1 and MERS, and VNAR 2C02 bound to SARS-CoV-1. Alignment of the SARS-CoV-1 and SARS-CoV-2 sequences show that there is a high degree of sequence homology between the RBDs, with almost complete conservation in the VNAR-3B4 epitope (Fig. 5a, blue box). A surface view of the RBD shows that most of the sequence variation occurs primarily at the ACE2 binding interface while the VNAR-3B4 interface remains similar (Fig. 5b, top). Among the critical interacting residues, there is a threonine in SARS-CoV-1 in place of the alanine in SARS-CoV-2 of which the backbone carbonyl, and not the sidechain, is the key interactor. Homology modeling of the complex between VNAR-3B4 and the SARS-CoV-1 RBD shows that the critical interactions, including the alanine to threonine change, are likely similar in the SARS-CoV-1:VNAR-3B4 complex (Fig. 5c, top). Crucially, the anti-parallel β-sheet between CDR3 of 3B4 and β2-strand of the RBD as well as the arginine h-bond network are also a key part of the interaction in the model. Alignment of the modeled SARS-CoV-1 RBD with a crystal structure of the SARS-CoV-1 RBD bound to ACE2 (PDB id: 2AJF) shows that the modeling imposed very little structural distortion to align the residues, indicating that only a modest rearrangement of the RBD framework is needed to accommodate binding of 3B4 (Fig. 5d, top).
a Sequence alignment between SARS-CoV-1, SARS-CoV-2, and the MERS RBDs. Residues different from SARS-CoV-2 are highlighted in red and the interaction interface for VNAR-3B4 is marked with a blue box. Bolded letters indicate residues critical for the interaction between the RBD and 3B4 with arrows indicating the residues that form a backbone beta-sheet. The sequence alignment is numbered above according to SARS-CoV-2. b Surface representations of SARS-CoV-1 (pink, above) and MERS (green, below) with variant residues colored red. The ACE2 binding interface is highlighted in purple for SARS-CoV-1 and the DPP4 binding interface in orange for MERS. The homology-modeled interaction interface for 3B4 is colored blue for both structures. c Zoomed in view of the modeled interaction interface between 3B4 (blue) and SARS-CoV-1 (pink, above) and MERS (green, below), with 3B4 colored blue in both pictures. Interacting residues are highlighted as in Fig. 2, showing the backbone interactions in black dashes and sidechain to backbone or sidechain to sidechain interactions shown as yellow dashes. Insets show the orientation of the zoomed in view. d Overlays of the 3B4 interface from modeled RBDs and their matching RBDs obtained by x-ray crystallography. The panel above shows the modeled SARS-CoV-1 RBD, colored pink, aligned with the crystal structure of SARS-CoV-1 RBD bound to ACE2 (PDB id: 2AJF), colored magenta. The panel below shows the modeled MERS RBD, colored light green, aligned with the crystal structure of MERS RBD bound to DPP4 (PDB id: 4L72), colored dark green.
In contrast to the conservation of the VNAR-3B4 epitope, the VNAR-2C02 epitope is less conserved. Among the 12 interacting residues in SARS-CoV-2, only 5 interactions are conserved in SARS-CoV-1 (Supplementary Fig. 3A, B). Interestingly, homology modeling of the SARS-CoV-1 RBD:VNAR-2C02 complex indicated that similar or replacement interactions take place and that the bulk of the hydrophobic core interactions were intact. Among the 7 residues that are part of this key hydrophobic interaction, four were conserved, two were replaced by residues with similar properties, and only one (L452) is lost (Supplementary Fig. 3A, see arrow annotations). For example, Ala348 has been replaced by Pro335, which is a non-polar like residue when cyclic, and the aromatic residue Phe490 has been replaced by Trp476 (Supplementary Fig. 3C). Conserved residues Tyr338 and Arg453 can still h-bond with His117 and Asn115 of 2C02, respectively. Fortunately, most residues that changed are still able to form similar h-bonds. Among these changes, a variation of Asn354 to Glu341 can still accept an h-bond from the backbone amine but it can no longer donate an h-bond to the carbonyl of the Asn115 of VNAR-2C02. The final major difference between the SARS-CoV-2 structure and the SARS-CoV-1 model is a result of the deletion of a residue equivalent to Glu484. This change results in the loss of an important salt bridge that forms with Arg67 of VNAR-2C02. However, the change of a similar aromatic residue, Phe490 to Trp476, allows Arg67 of VNAR-2C02 to maintain its cation-pi interaction.
Greater variation exists between the SARS-CoV-2 and the MERS RBDs, with only 3 of the 7 interacting residues in the VNAR-3B4 epitope remaining (Fig. 5a, blue box). A surface view of the MERS RBD shows that the sequence variation occurs throughout the RBD (Fig. 5b, bottom). The MERS RBD binds DPP4 instead of ACE2 and a large degree of structural and sequence variation would be expected. Homology modeling of VNAR-3B4:MERS RBD complex found that many of the backbone interactions, including those of the antiparallel β-sheet and arginine h-bond network could still interact (Fig. 5c, bottom). Though the h-bonding between the conserved Ser429 and Glu122 of 3B4 remained, the change from a tyrosine to a leucine certainly eliminates one h-bond formed with Asp107 of VNAR-3B4. It was much less likely that all these interactions remain the same when comparing the modeled MERS RBD to the crystal structure of the MERS RBD bound to its receptor DPP4 (PDB id: 4L72). This alignment has an RMSD of 4.137, indicating that the model and x-ray structure are very dissimilar and that a large structural rearrangement of the MERS RBD framework is needed to fully accommodate the binding of 3B4 (Fig. 5d, bottom). Visual inspection of the alignment shows that the modeling constraints forced both the α1 and α2 helices that surround the β2-strand of the interaction to unwind a single helical turn in order to allow the β2-strand flexibility to interact with 3B4. Such a large structural rearrangement would not be favorable, and it is unlikely that the homology modeling accurately represents the interaction between the MERS RBD and VNAR-3B4. Realistically, the conserved Phe, Cys, and Ser residues of the VNAR-3B4 epitope are the only likely interactions, and the remaining β-sheet and arginine network hydrogen-bonding are lost. Deletion of the h-bonding that occurs between the VNAR and the RBD would explain why VNAR-3B4 weakly binds the MERS RBD and is moderately effective at neutralization of pseudovirus.
VNARs bind SARS-CoV-2 variants
Emerging mutant strains of SARS-CoV-2 remain a threat to the control of the pandemic. Viral variants are more easily transmissible, and it is uncertain if current vaccine formulations will remain efficacious against them. The CDC is currently monitoring several variants of concern (VOC), including the prominent Alpha and Beta variants that first emerged in the UK and South Africa, respectively, as well as the more recent Delta variant that first appeared in India (Fig. 6). The spike protein is highly susceptible to mutation and many prominent mutations that occur within the RBD alter the interaction with the host ACE2 by enhancing the binding properties (affinity) of the spike protein to ACE2. It is also likely that mutations which affect the ACE2/RBD interface would also diminish or prevent binding of neutralizing antibodies that directly compete at this site.
For panels (a) and (b), all reported mutation sites (as curated by CDC.gov) are colored green in each RBD structure and the ACE2 binding interface is annotated in purple. The list of mutation sites is shown in the center of the figure with arrows drawn to indicate the mutation sites that appear in each of the primary variants of concern (V.O.C). a Surface representation of the RBD with the 3B4 binding interface colored in blue. Insets show mutation sites that occur proximal to 3B4, top: (R408) and bottom:(W436, N439, N440, N501). b A 180-degree rotation of the RBD with the 2C02 binding interface colored in yellow. Inset shows mutations that are a part of or proximal to the 2C02 interface (L452, E484, F490). c A cartoon representation of the RBD with an outline of the surface with mutation sites K417, E484, and N501 displayed as orange sticks. d Summary of binding affinities (KD) for each of the VNARs to the WT SARS-CoV-2 RBD in comparison to the triple mutant RBD indicated in (c). Data shown are a result of triplicate experiments using a BLI binding schema with biotin labeled RBDs (see methods).
To visualize the potential influence of RBD mutations on the binding of our VNARs, we mapped mutation sites reported in the growing list curated by the CDC to the RBD of the VNAR-3B4 and 2C02 crystal structures (Fig. 6). Mutation mapping revealed that most of the RBD mutations occur distal to the VNAR-3B4 epitope and are primarily located in the ACE2 binding interface (Fig. 6a). The most proximal mutation sites to VNAR-3B4 (R408, W436, N439, N440, and N501) do not interact with the VNAR, and there are no apparent amino acid changes that would create a clash with VNAR-3B4 at these sites (Fig. 6a, top and bottom insets). This analysis suggests that VNAR-3B4 will not lose its ability to neutralize current SARS-CoV-2 viral variants. Several mutation sites, however, occur directly in the interface of 2C02. Sites L452, E484, and F490 play a direct role in the binding of 2C02, suggesting that 2C02 binding would be altered when binding to viral variants that include these mutation sites (Fig. 6b).
To assess the ability of our VNARs to neutralize variants, we performed biolayer interferometry (BLI) to compare VNAR binding to both WT and mutants RBD (Supplementary Fig. 4). A mutant RBD from the Beta variant, which also contained mutations found in other VOCs, was used (Fig. 6c). Results from the BLI experiments confirmed our initial prediction that these mutations have little to no effect on the affinity of 3B4 (Fig. 6d). Surprisingly, the loss of the salt bridge interaction between VNAR-2C02 and E484 did not have a strong effect on the affinity of VNAR-2C02 (Fig. 6d), further indicating that the VNAR-2C02 interaction is predominantly the result of hydrophobic interactions. While we do not have any structural data for the binding location of VNAR-4C10 on the RBD, the BLI experiments showed that these mutations do not have a substantial effect on VNAR-4C10 affinity as well. Taken together, the mutational mapping analysis and BLI data strongly indicate that the use of VNARs will remain useful neutralizing agents to combat variant strains of the SARS-CoV-2 virus.
