An ultrapotent synthetic nanobody neutralizes SARS-CoV-2 by stabilizing inactive Spike

1Over the past two decades, three zoonotic β-coronaviruses have entered the human population, causing severe respiratory symptoms with high mortality. The COVID-19 pandemic is caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the most readily transmissible of these three coronaviruses. No preventive treatment has been approved for any coronavirus to date, and the timeline for an effective and broadly available vaccine for SARS-CoV-2 remains uncertain. The development of new therapeutic and prophylactic approaches thus remains essential.

2Coronavirus virions are bounded by a membrane that contains the homotrimeric transmembrane glycoprotein Spike, which is responsible for virus entry into the host cell. The surface-exposed portion of Spike is composed of two domains, S1 and S2. S1 binds the host cell receptor angiotensin-converting enzyme 2 (ACE2), whereas S2 catalyzes fusion of the viral and host cell membranes. Contained within S1 is the receptor binding domain (RBD), which directly binds to ACE2, and the N-terminal domain (NTD). The RBD is attached to the body of Spike by a flexible region and can exist in an inaccessible down state or an accessible up state. Binding to ACE2 requires the RBD to occupy the up state and enables cleavage by host proteases, triggering a conformational change in S2 required for viral entry. In SARS-CoV-2 virions, Spike exchanges between an active, open conformation with at least one RBD in the up state and an inactive, closed conformation with all RBDs in the down state.

3We isolated single-domain antibodies (nanobodies) that neutralize SARS-CoV-2 by screening a yeast surface-displayed library of >2 × 109 synthetic nanobody sequences for binders to the Spike ectodomain. We used a mutant form of SARS-CoV-2 Spike (SpikeS2P) as the antigen. SpikeS2P lacks one of the two proteolytic cleavage sites between the S1 and S2 domains and introduces two mutations and a trimerization domain to stabilize the prefusion conformation. We labeled SpikeS2P with biotin or with fluorescent dyes and selected nanobody-displaying yeast over multiple rounds, first by magnetic bead binding and then by fluorescence-activated cell sorting (Fig. 1A).

4Three rounds of selection yielded 21 distinct nanobodies that bound SpikeS2P and showed decreased binding in the presence of a dimeric construct of the ACE2 extracellular domain (ACE2-Fc). These nanobodies fall into two classes. Class I binds the RBD and competes directly with ACE2-Fc (Fig. 1B). A prototypical example of this class is nanobody Nb6, which binds to SpikeS2P and to RBD alone with a dissociation constant (KD) of 210 and 41 nM, respectively (Fig. 1C and table S1). Class II, exemplified by nanobody Nb3, binds to SpikeS2P (KD = 61 nM) but displays no binding to RBD alone (Fig. 1C and table S1). In the presence of excess ACE2-Fc, binding of Nb6 and other class I nanobodies is blocked entirely, whereas binding of Nb3 and other class II nanobodies is moderately decreased (Fig. 1B). These results suggest that class I nanobodies target the RBD to block ACE2 binding, whereas class II nanobodies target other epitopes. Indeed, surface plasmon resonance (SPR) experiments demonstrate that class I and class II nanobodies can bind SpikeS2P simultaneously (Fig. 1D).

5Class I nanobodies show a consistently faster association rate constant (ka) for nanobody binding to the isolated RBD than to SpikeS2P (table S1), which suggests that RBD accessibility influences the KD. We next tested the efficacy of class I and class II nanobodies to inhibit binding of fluorescently labeled SpikeS2P to ACE2-expressing human embryonic kidney (HEK) 293 cells (Fig. 1E and table S1). Class I nanobodies Nb6 and Nb11 emerged as two of the most potent clones, with half-maximal inhibitory concentration (IC50) values of 370 and 540 nM, respectively. Class II nanobodies showed little to no activity in this assay. We prioritized two class I nanobodies, Nb6 and Nb11, that combine potent SpikeS2P binding with relatively small differences in ka between binding to SpikeS2P or RBD. For class II nanobodies, we prioritized Nb3 because of its relative yield during purification (table S1).

6To define the binding sites of Nb6 and Nb11, we determined their cryo–electron microscopy (cryo-EM) structures bound to SpikeS2P (Fig. 2, A and B; figs. S1 to S3; and table S2). Both nanobodies recognize RBD epitopes that overlap the ACE2 binding site (Fig. 2E). For Nb6 and Nb11, we resolved nanobody binding to both the open and closed conformations of SpikeS2P. We obtained a 3.0-Å map of Nb6 bound to closed SpikeS2P, which enabled modeling of the Nb6-SpikeS2P complex (Fig. 2A), including the complementarity-determining regions (CDRs). We also obtained lower-resolution maps for Nb6 bound to open SpikeS2P (3.8 Å), and Nb11 bound to open and closed SpikeS2P (4.2 and 3.7 Å, respectively). For these lower-resolution maps, we could define the nanobody’s binding orientation but not accurately model the CDRs.

7Nb6 bound to closed SpikeS2P straddles the interface between two adjacent RBDs. Most of the contacting surfaces are contributed by CDR1 and CDR2 of Nb6 (Fig. 2C). CDR3 contacts the adjacent RBD positioned counterclockwise when viewed from the top (Fig. 2C). The binding of one Nb6 therefore stabilizes two adjacent RBDs in the down state and likely preorganizes the binding site for a second and third Nb6 molecule to stabilize the closed Spike conformation. By contrast, Nb11 bound to down-state RBDs only contacts a single RBD (Fig. 2D).

8The structure of Nb6 bound to closed SpikeS2P enabled us to engineer bivalent and trivalent nanobodies predicted to lock all RBDs in the down state. We inserted flexible Gly-Ser linkers of either 15 or 20 amino acids to span the 52-Å distance between adjacent Nb6 monomers bound to down-state RBDs in closed SpikeS2P (fig. S4). These linkers are too short to span the 72-Å distance between Nb6 molecules bound to open Spike. Moreover, steric clashes would prevent binding of three RBDs in open Spike with a single up-state RBD even with longer linker length (fig. S4). By contrast, the minimum distance between adjacent Nb11 monomers bound to either open or closed SpikeS2P is 68 Å. We predicted that multivalent binding by Nb6 constructs would display substantially slowed dissociation rates owing to enhanced avidity.

9In SPR experiments, both bivalent Nb6 with a 15–amino acid linker (Nb6-bi) and trivalent Nb6 with two 20–amino acid linkers (Nb6-tri) dissociate from SpikeS2P in a biphasic manner. The dissociation phase can be fitted to two components: a fast phase with kinetic rate constants kd1 of 2.7 × 10−2 s−1 for Nb6-bi and 2.9 × 10−2 s−1 for Nb6-tri, which are close to that observed for monovalent Nb6 (kd = 5.6 × 10−2 s−1), and a slow phase that is dependent on avidity (kd2 = 3.1 × 10−4 s−1 for Nb6-bi and kd2 < 1.0 × 10−6 s−1 for Nb6-tri) (Fig. 3A). The relatively similar kd for the fast phase suggests that a fraction of the observed binding for the multivalent constructs is nanobody binding to a single SpikeS2P RBD. By contrast, the slow dissociation phase of Nb6-bi and Nb6-tri indicates engagement of two or three RBDs. We observed no dissociation for the slow phase of Nb6-tri over 10 min, indicating an upper boundary for kd2 of 1 × 10−6 s−1 and subpicomolar affinity. This measurement remains an upper boundary estimate because the measurement is limited by the intrinsic dissociation rate of SpikeS2P from the SPR chip imposed by the chemistry used to immobilize SpikeS2P. The true dissociation rate, therefore, may be considerably lower.

10Biphasic dissociation could be explained by a slow interconversion between up- and down-state RBDs, with conversion to the more stable down state required for multivalent binding: A single domain of Nb6-tri engaged with an up-state RBD would dissociate rapidly. The system would then reequilibrate as the RBD flips into the down state, eventually allowing Nb6-tri to trap all RBDs in closed SpikeS2P. To test this directly, we varied the association time for Nb6-tri binding to SpikeS2P. Indeed, we observed an exponential decrease in the percentage of fast-phase dissociation with a half-life (t1/2) of 65 s (Fig. 3B), which, we surmise, reflects the time scale of conversion between the RBD up and down states in SpikeS2P. Taken together, dimerization and trimerization of Nb6 afforded 750-fold and >200,000-fold gains in KD, respectively.

11Unable to determine the binding site of Nb3 by cryo-EM, we turned to radiolytic hydroxyl radical footprinting. We exposed apo- or Nb3-bound SpikeS2P to synchrotron x-ray radiation to label solvent-exposed amino acids with hydroxyl radicals, which we subsequently quantified by mass spectrometry of protease-digested SpikeS2P. Two neighboring surface residues on the S1 NTD of Spike (Met177 and His207) were protected in the presence of Nb3 at a level consistent with prior observations of antibody-antigen interactions by hydroxyl radical footprinting (fig. S5). Previously discovered coronavirus neutralizing antibodies bind an epitope within the NTD of Spike with Fab fragments that are noncompetitive with the host cell receptor. Further SPR experiments demonstrated that Nb3 can bind SpikeS2P simultaneously with monovalent ACE2 (fig. S6). We hypothesized that the multivalent display of Nb3 on the surface of yeast may account for the partial decrease in SpikeS2P binding observed in the presence of ACE2-Fc. Indeed, a trivalent construct of Nb3 with 15–amino acid linkers (Nb3-tri) inhibited SpikeS2P binding to ACE2 cells with an IC50 of 41 nM (fig. S6). How Nb3-tri disrupts Spike-ACE2 interactions remains unclear.

12We next tested the neutralization activity of monovalent and trivalent versions of our top class I (Nb6 and Nb11) and class II (Nb3) nanobodies against SARS-CoV-2 pseudotyped lentivirus using a previously described assay. Nb6 and Nb11 inhibited pseudovirus infection with IC50 values of 2.0 and 2.4 μM, respectively. Nb3 inhibited pseudovirus infection with an IC50 of 3.9 μM (Fig. 3C and table S1). Nb6-tri shows a 2000-fold enhancement of inhibitory activity, with an IC50 of 1.2 nM, whereas trimerization of Nb11 and Nb3 resulted in more modest gains of 40- and 10-fold (51 and 400 nM), respectively (Fig. 3C). We confirmed these neutralization activities with a viral plaque assay using live SARS-CoV-2 virus infection of VeroE6 cells. Here, Nb6-tri proved exceptionally potent, neutralizing SARS-CoV-2 with an average IC50 of 160 pM (Fig. 3D). Nb3-tri neutralized SARS-CoV-2 with an average IC50 of 140 nM (Fig. 3D).

13We further optimized the potency of Nb6 by selecting a saturation mutagenesis library targeting all three CDRs. Two rounds of selection identified high-affinity clones with two penetrant mutations: I27Y (Ile27→Tyr) in CDR1 and P105Y (Pro105→Tyr) in CDR3. We incorporated these mutations into Nb6 to generate matured Nb6 (mNb6), which binds with 500-fold increased affinity to SpikeS2P (Fig. 4A). mNb6 inhibits both pseudovirus and live SARS-CoV-2 infection with low nanomolar potency, a ~200-fold improvement compared with Nb6 (Fig. 4B and table S1).

14A 2.9-Å cryo-EM structure shows that mNb6 binds to closed SpikeS2P (Fig. 4C and fig. S7). mNb6 induces a slight rearrangement of the down-state RBDs as compared with SpikeS2P bound to Nb6, inducing a 9° rotation of the RBD away from the central threefold-symmetry axis. This deviation likely arises from a different interaction between CDR3 and SpikeS2P, which nudges the RBDs into a new resting position (Fig. 4D). Although the I27Y substitution optimizes local contacts between CDR1 in its original binding site on the RBD, the P105Y substitution leads to a marked rearrangement of CDR3 in mNb6 (Fig. 4, E and F). This conformational change yields a different set of contacts between mNb6 CDR3 and the adjacent RBD. An x-ray crystal structure of mNb6 alone revealed dramatic conformational differences in CDR1 and CDR3 between free and SpikeS2P-bound mNb6 (Fig. 4G and table S3). Although differences in loop conformation in the crystal structure may arise from crystal lattice contacts, they are suggestive of conformational heterogeneity for unbound mNb6 and induced-fit rearrangements upon binding to SpikeS2P.

15The binding orientation of mNb6 is similar to that of Nb6, suggesting that multivalent design would likewise enhance binding affinity. Unlike Nb6-tri, trivalent mNb6 with a 20–amino acid linker (mNb6-tri) bound to SpikeS2P with no observable fast-phase dissociation and no measurable dissociation over 10 minutes, yielding an upper bound for the dissociation rate constant kd of 1.0 × 10−6 s−1 (t1/2> 8 days) and a KD of <1 pM (Fig. 4A). mNb6-tri displays further gains in potency in both pseudovirus and live SARS-CoV-2 infection assays with IC50 values of 120 pM (5.0 ng/ml) and 54 pM (2.3 ng/ml), respectively (Fig. 4B and table S1). Given the subpicomolar affinity observed by SPR, it is likely that these viral neutralization potencies reflect the lower limit of the assays. mNb6-tri is therefore an exceptionally potent SARS-CoV-2 neutralizing molecule.

16We next tested whether viral neutralization by the class I nanobody mNb6 is potentially synergistic with the class II nanobody Nb3-tri. In pseudovirus neutralization assays, we observed an additive effect when combining Nb3-tri with mNb6 (fig. S8). However, the potency for mNb6 viral neutralization was unchanged with increasing concentrations of Nb3-tri, suggesting minimal synergy between these two nanobodies.

17We next tested Nb6 and its derivatives for stability. Circular dichroism revealed melting temperatures of 66.9°, 62.0°, 67.6°, and 61.4°C for Nb6, Nb6-tri, mNb6, and mNb6-tri, respectively (fig. S9). Moreover, mNb6 and mNb6-tri were stable to lyophilization and to aerosolization, showing no aggregation by size exclusion chromatography, and preserved high-affinity binding to SpikeS2P (Fig. 5, A and B, and fig. S9). Finally, mNb6-tri retains potent inhibition of pseudovirus and live SARS-CoV-2 infection after aerosolization, lyophilization, or heat treatment for 1 hour at 50°C (Fig. 5C and fig. S9).

18Strategies to prevent SARS-CoV-2 entry into the host cell aim to block the ACE2-RBD interaction. Although high-affinity monoclonal antibodies are leading the way as potential therapeutics, they are expensive to produce by mammalian cell expression and need to be intravenously administered by health care professionals. Large doses are needed for prophylactic use because only a small fraction of systemic antibodies cross the epithelial cell layers lining the airways. By contrast, nanobodies can be inexpensively produced in bacteria or yeast. The inherent stability of nanobodies enables aerosolized delivery directly to the nasal and lung epithelia. Indeed, aerosol delivery of a trimeric nanobody targeting respiratory syncytial virus (ALX-0171) was recently demonstrated to be effective in substantially decreasing measurable viral load in hospitalized infants. Finally, potential immunogenicity of camelid-derived nanobodies can be mitigated by established humanization strategies.

19Nanobody multimerization has been shown to improve target affinity by avidity. In the case of Nb6 and mNb6, structure-guided design of a multimeric construct that simultaneously engages all three RBDs yielded profound gains in potency. Furthermore, because RBDs must be in the up state to engage with ACE2, conformational control of RBD accessibility serves as an added neutralization mechanism. Indeed, when mNb6-tri engages with Spike, it prevents ACE2 binding both by directly occluding the binding site and by locking the RBDs into an inactive conformation.

20Our discovery of class II neutralizing nanobodies demonstrates potentially new mechanisms of disrupting Spike function. The pairing of class I and class II nanobodies in a prophylactic or therapeutic cocktail could provide both potent neutralization and prevention of escape variants. The combined stability, potency, and diverse epitope engagement of our anti-Spike nanobodies therefore provide a distinctive potential prophylactic and therapeutic strategy to limit the continued toll of the COVID-19 pandemic.