Versatile and multivalent nanobodies efficiently neutralize SARS-CoV-2

 See all authors and affiliations

Science  18 Dec 2020:
Vol. 370, Issue 6523, pp. 1479-1484
DOI: 10.1126/science.abe4747

Nanobodies that neutralize

Monoclonal antibodies that bind to the spike protein of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) show therapeutic promise but must be produced in mammalian cells and need to be delivered intravenously. By contrast, single-domain antibodies called nanobodies can be produced in bacteria or yeast, and their stability may enable aerosol delivery. Two papers now report nanobodies that bind tightly to spike and efficiently neutralize SARS-CoV-2 in cells. Schoof et al. screened a yeast surface display of synthetic nanobodies and Xiang et al. screened anti-spike nanobodies produced by a llama. Both groups identified highly potent nanobodies that lock the spike protein in an inactive conformation. Multivalent constructs of selected nanobodies achieved even more potent neutralization.

Science, this issue p. 1473, p. 1479


Cost-effective, efficacious therapeutics are urgently needed to combat the COVID-19 pandemic. In this study, we used camelid immunization and proteomics to identify a large repertoire of highly potent neutralizing nanobodies (Nbs) to the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike protein receptor binding domain (RBD). We discovered Nbs with picomolar to femtomolar affinities that inhibit viral infection at concentrations below the nanograms-per-milliliter level, and we determined a structure of one of the most potent Nbs in complex with the RBD. Structural proteomics and integrative modeling revealed multiple distinct and nonoverlapping epitopes and indicated an array of potential neutralization mechanisms. We bioengineered multivalent Nb constructs that achieved ultrahigh neutralization potency (half-maximal inhibitory concentration as low as 0.058 ng/ml) and may prevent mutational escape. These thermostable Nbs can be rapidly produced in bulk from microbes and resist lyophilization and aerosolization.


Get the latest issue of Science delivered right to you!

Globally a novel, highly transmissible coronavirus—severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) (12)—has infected more than 30 million people and has claimed almost 1 million lives, with the numbers still rising as of September 2020. Despite preventive measures, such as quarantines and lockdowns that help curb viral transmission, the virus rebounds after social restrictions are lifted. Safe and effective therapeutics and vaccines remain urgently needed.

Like other zoonotic coronaviruses, SARS-CoV-2 expresses a surface spike (S) glycoprotein, which consists of S1 and S2 subunits that form a homotrimeric viral spike to interact with host cells. The interaction is mediated by the S1 receptor binding domain (RBD), which binds the peptidase domain of human angiotensin-converting enzyme 2 (hACE2) as a host receptor (3). Structural studies have revealed different conformations of the spike (45). In the prefusion stage, the RBD switches between a closed conformation and an open conformation for hACE2 interaction. In the postfusion stage, the S2 undergoes a substantial conformational change to trigger host membrane fusion (6). Investigations of sera from COVID-19 convalescent individuals have led to the identification of potent neutralizing antibodies (NAbs) that primarily target the RBD but also non-RBD epitopes (713). High-quality NAbs may overcome the risks of Fc-associated antibody-dependent enhancement and are promising therapeutic candidates (1415).

VHH antibodies, or nanobodies (Nbs), are minimal, monomeric antigen binding domains derived from camelid single-chain antibodies (16). Unlike immunoglobulin G (IgG) antibodies, Nbs are small (~15 kDa), highly soluble and stable, readily bioengineered into bi- or multivalent forms, and amenable to low-cost, efficient microbial production. Owing to their robust physicochemical properties, Nbs can be administered by inhalation, which makes them appealing therapeutic agents for treatment of respiratory viruses (1718). Recently, several SARS-CoV-2 neutralizing Nbs have been identified through the screening of SARS-CoV or Middle East respiratory syndrome (MERS) cross-reacting Nbs or the use of synthetic Nb libraries for RBD binding. However, these synthetic Nbs generally neutralize the virus at or below microgram-per-milliliter concentrations (121922) and thus are hundreds of times less potent than the most effective NAbs, likely due to monovalency and/or lack of affinity maturation (2324). The development of highly potent anti–SARS-CoV-2 Nbs may provide a means for versatile, cost-effective prophylaxis, therapeutics, and point-of-care diagnosis.

To produce high-quality SARS-CoV-2 neutralizing Nbs, we immunized a llama with the recombinant RBD. Compared with the preimmunized serum sample, the postimmunized serum showed potent and specific serologic activities toward RBD binding with a titer of 1.75 × 106 (fig. S1A). The serum efficiently neutralized the pseudotyped SARS-CoV-2 at a half-maximal neutralization titer (NT50) of ~310,000 (fig. S1B), orders of magnitude higher than that of the convalescent sera obtained from recovered COVID-19 patients (78). To further characterize these activities, we separated the single-chain VHH antibodies from the IgGs. We confirmed that the single-chain antibodies achieve specific, high-affinity binding to the RBD and possess subnanometer half-maximal inhibitory concentration (IC50 = 509 pM) against the pseudotyped virus (fig. S1C).

We identified thousands of high-affinity VHH Nbs from the RBD-immunized llama serum by using a robust proteomic strategy that we recently developed (25) (fig. S2A). This repertoire includes ~350 distinct CDR3s (CDRs, complementarity-determining regions). For Escherichia coli expression, we selected 109 highly diverse Nb sequences from the repertoire with distinct CDR3s to cover various biophysical, structural, and potentially different antiviral properties. Ninety-four Nbs were purified and tested for RBD binding by enzyme-linked immunosorbent assay (ELISA), from which we confirmed 71 RBD-specific binders (fig. S2, B and C, and tables S1 and S4). Of these RBD-specific binders, 49 Nbs presented high solubility and high affinity (ELISA IC50 below 30 nM; Fig. 1A) and were promising candidates for functional characterizations. We used a SARS-CoV-2–green fluorescent protein (GFP) pseudovirus neutralization assay to screen and characterize the antiviral activities of these high-affinity Nbs. Of the tested Nbs, 94% neutralize the pseudotype virus below 3 μM (Fig. 1B), and 90% neutralize below 500 nM. Only 20 to 40% of the high-affinity RBD-specific monoclonal antibodies identified from patient sera have been reported to possess comparable potency (78). More than three-quarters (76%) of the Nbs efficiently neutralized the pseudovirus below 50 nM, and 6% had neutralization activities below 0.5 nM. We selected the 18 most potent Nbs on the basis of the pseudovirus GFP reporter screen and measured their potency accurately using the pseudovirus-luciferase reporter assay. Finally, we used the plaque reduction neutralization test (PRNT) assay (26) to evaluate the potential of 14 Nbs to neutralize the SARS-CoV-2 Munich strain. All of the Nbs reached 100% neutralization and neutralized the virus in a dose-dependent manner. The IC50 values range from single-digit nanograms-per-milliliter amounts to below the nanograms-per-milliliter level. Of the most potent Nbs, three of them (89, 20, and 21) showed neutralization of 2.1 ng/ml (0.133 nM), 1.6 ng/ml (0.102 nM), and 0.7 ng/ml (0.045 nM), respectively, in the pseudovirus assay (Fig. 1C) and 0.154, 0.048, and 0.022 nM, respectively, in the SARS-CoV-2 assay (Fig. 1, D and E). Overall, there was an excellent correlation between the two neutralization assays (R2 = 0.92; fig. S3).

Fig. 1 Production and characterizations of high-affinity RBD Nbs for SARS-CoV-2 neutralization.

(A) Binding affinities of 71 Nbs toward the RBD, as determined by ELISA. The pie chart shows the number of Nbs according to affinity and solubility. O.D., optical density. (B) Screening of 49 high-affinity Nbs with high-expression level, as determined by SARS-CoV-2–GFP pseudovirus neutralization assay. n = 1 for Nbs with neutralization potency IC50 ≤ 50 nM; n = 2 for Nbs with neutralization potency IC50 > 50 nM (n indicates the number of replicates). (C) Neutralization potency of 18 highly potent Nbs was calculated on the basis of the pseudotyped SARS-CoV-2 neutralization assay (luciferase). Purple, red, and yellow lines denote Nbs 20, 21, and 89 with IC50 < 0.2 nM. Two different purifications of the pseudovirus were used. The average neutralization percentage was shown for each data point (n = 5 for Nbs 20 and 21; n = 2 for all other Nbs). (D) Neutralization potency of 14 neutralizing Nbs by SARS-CoV-2 plaque reduction neutralization test (PRNT). The average neutralization percentage was shown for each data point (n = 4 for Nbs 20, 21, and 89; n = 2 for other Nbs). (E) Summary table of pseudovirus and SARS-CoV-2 neutralization potencies of 18 Nbs. N/A, not tested. (F) SPR binding kinetics measurement for Nb21. Ka, acid dissociation constant; Kd, dissociation constant; KD, equilibrium dissociation constant; RU, relative units.

We measured the binding kinetics of Nbs 89, 20, and 21 by surface plasmon resonance (SPR) (fig. S4, A and B). Nbs 89 and 20 have affinities of 108 and 10.4 pM, respectively, and the most potent Nb21 did not show detectable dissociation from the RBD during 20 min of SPR analysis. The subpicomolar affinity of Nb21 potentially explains its unusual neutralization potency (Fig. 1F). From the E. coli periplasmic preparations, we determined the thermostability of Nbs 89, 20, and 21 to be 65.9°, 71.8°, and 72.8°C, respectively (fig. S4C). Finally, we tested the on-shelf stability of Nb21, which remained soluble after ~6 weeks of storage at room temperature after purification. No multimeric forms or aggregations were detected by size exclusion chromatography (SEC) (fig. S4D). Together, these results suggest that these neutralizing Nbs have valuable physicochemical properties for advanced therapeutic applications.

We employed an integrative approach by using SEC, cross-linking and mass spectrometry, and structural modeling for epitope mapping (2730). First, we performed SEC experiments to distinguish between Nbs that share the same RBD epitope with Nb21 and those that bind to nonoverlapping epitopes. According to SEC profiles, Nbs 9, 16, 17, 20, 64, 82, 89, 99, and 107 competed with Nb21 for RBD binding (Fig. 2A and fig. S5), which indicates that their epitopes overlap substantially. By contrast, higher-mass species (from early elution volumes) corresponding to the trimeric complexes composed of Nb21, RBD, and one of the Nbs (34, 36, 93, 105, and 95) were evident (Fig. 2B and fig. S6, A to H). Moreover, Nb105 competed with Nbs 34 and 95, which did not compete for RBD interaction, suggesting the presence of two distinct and nonoverlapping epitopes. Second, we used disuccinimidyl suberate (DSS) to cross-link Nb-RBD complexes, and we used mass spectrometry to identify, on average, four intermolecular cross-links for Nbs 20, 93, 34, 95, and 105. The cross-links were used to map the RBD epitopes derived from the SEC data (materials and methods). Our cross-linking models identified five epitopes (I, II, III, IV, and V corresponding to Nbs 20, 93, 34, 95, and 105) (Fig. 2C). The models satisfied 90% of the cross-links with an average precision of 7.8 Å (Fig. 2D and table S2). Our analysis confirmed the presence of a dominant epitope I (e.g., epitopes of Nbs 20 and 21) overlapping with the hACE2 binding site. Epitope II also colocalized with the nonconserved hACE2 binding site. Both epitopes I and II Nbs can compete with hACE2 binding to the RBD at very low concentrations in vitro (fig. S7A). Epitopes III to V colocalized with conserved sites (fig. S7, B and C). Notably, epitope I Nbs had significantly shorter CDR3s (four amino acids shorter; P = 0.005) than other epitope binders (fig. S6I). Despite this, most of the selected Nbs potently inhibited the virus with an IC50 below 30 ng/ml (2 nM) (table S1).

Fig. 2 Nb epitope mapping by integrative structural proteomics.

(A) Summary of Nb epitopes on the basis of SEC analysis. Purple, Nbs that bind the same RBD epitope; green, Nbs of different epitopes; gray, not tested. (B) Representation of SEC profiling of the RBD, RBD-Nb21 complex, and RBD-Nb21-Nb105 complex. The y axis represents ultraviolet 280 nm absorbance units (mAu). (C) Cartoon model showing the localization of five Nbs that bind different epitopes: Nb20 (purple), Nb34 (green), Nb93 (dark pink), Nb105 (yellow), and Nb95 (light pink) in complex with the RBD (gray). Blue and red lines represent DSS cross-links shorter or longer than 28 Å, respectively. (D) Top-10-scoring cross-linking–based models for each Nb (cartoons) on top of the RBD surface.

To explore the molecular mechanisms that underlie the potent neutralization activities of epitope I Nbs, we determined a crystal structure of the RBD-Nb20 complex at a resolution of 3.3 Å by molecular replacement (materials and methods, table S3, and fig. S13). Most of the residues in the RBD (Asn334 to Gly526) and in the entire Nb20, particularly those at the protein interaction interface, are well resolved. There are two nearly identical copies of RBD-Nb20 complexes in one asymmetric unit, with a root mean square deviation of 0.277 Å over 287 Cα atoms. In the structure, all three CDRs of Nb20 interact with the RBD by binding to its large extended external loop with two short β strands (Fig. 3A) (31). Glu484 of the RBD forms hydrogen bonding and ionic interactions with the side chains of Arg31 (CDR1) and Tyr104 (CDR3) of Nb20, whereas Gln493 of the RBD forms hydrogen bonds with the main-chain carbonyl of Ala29 (CDR1) and the side chain of Arg97 (CDR3) of Nb20. These interactions constitute a major polar interaction network at the RBD and Nb20 interface. Arg31 of Nb20 also engages in a cation π interaction with the side chain of Phe490 of the RBD (Fig. 3B). In addition, Met55 from the CDR2 of Nb20 packs against residues Leu452, Phe490, and Leu492 of the RBD to form hydrophobic interactions at the interface. Another small patch of hydrophobic interactions is formed among residues Val483 of the RBD and Phe45 and Leu59 from the framework β sheet of Nb20 (Fig. 3C).

Fig. 3 Crystal structure analysis of an ultrahigh-affinity Nb in complex with the RBD.

(A) Cartoon presentation of Nb20 in complex with the RBD. CDR1, -2, and -3 are in red, green, and orange, respectively. (B) Zoomed-in view of an extensive polar interaction network that centers on R35 of Nb20. (C) Zoomed-in view of hydrophobic interactions. (D) Surface presentation of the Nb20-RBD and hACE2-RBD complex (PDB ID 6M0J). Single-letter abbreviations for amino acid residues are as follows: A, Ala; E, Glu; F, Phe; L, Leu; M, Met; Q, Gln; R, Arg; V, Val.

The binding mode of Nb20 to the RBD is distinct from those of other reported neutralizing Nbs, which generally recognize similar epitopes in the RBD external loop region (3234) (fig. S8). The extensive hydrophobic and polar interactions (Fig. 3, B and C) between the RBD and Nb20 stem from the notable shape complementarity (Fig. 3D) between the CDRs and the external RBD loop, leading to ultrahigh affinity (~10 pM). On the basis of our crystal structure, we further modeled the structure of the best neutralizer Nb21 with the RBD (materials and methods). Only four residues vary between Nbs 20 and 21 (fig. S9A), all of which are on CDRs. Two substitutions are at the RBD binding interface. Ser52 and Met55 in the CDR2 of Nb20 are replaced by two asparagine residues (Asn52 and Asn55) in Nb21. In our superimposed structure, Asn52 forms a new H bond with Asn450 of the RBD (fig. S9B). Although Asn55 does not engage in additional interactions with the RBD, it creates a salt bridge with the side chain of Arg31, which stabilizes the polar interaction network among Arg31 and Tyr104 of Nb21 and Gln484 of the RBD (fig. S9B). All of those likely contribute to a slower off-rate of Nb21 (Fig. 1F and fig. S4A) and stronger neutralization potency. Structural comparison of RBD-Nb20 or RBD-Nb21 and RBD-hACE2 [Protein Data Bank (PDB) ID 6LZG] (31) clearly showed that the interfaces for Nb20 or Nb21 and hACE2 partially overlap (Fig. 3D and fig. S9C). Notably, the CDR1 and CDR3 of Nb20 or Nb21 would clash with the first helix of hACE2, the primary binding site for the RBD (fig. S9D).

To understand the antiviral efficacy of our Nbs, we superimposed RBD-Nb complexes on different spike conformations according to cryo–electron microscopy (cryo-EM) structures. We found that three copies of Nb20 or Nb21 can simultaneously bind all three RBDs in their “down” conformations (PDB ID 6VXX) (4) that correspond to the inactive spike (Fig. 4B). Our analysis indicates a potential mechanism by which Nbs 20 and 21 (epitope I) lock RBDs in their down conformation with ultrahigh affinity. Combined with the steric interference with hACE2 binding in the RBD open conformation (Fig. 4A), these mechanisms may explain the exceptional neutralization potencies of epitope I Nbs.

Fig. 4 Potential mechanisms of SARS-CoV-2 neutralization by Nbs.

(A) hACE2 (blue) binding to spike trimer conformation (wheat, beige, and gray colors) with one RBD in the “up” conformation (PDB IDs 6VSB and 6LZG). (B) Nb20 (epitope I, purple) partially overlaps with the hACE2 binding site and can bind the closed spike conformation with all RBDs “down” (PDB ID 6VXX). (C) Summary of spike conformations accessible (+) to the Nbs of different epitopes. (D) Nb93 (epitope II, dark pink) partially overlaps with the hACE2 binding site and can bind to spike conformations with at least one RBD up (PDB ID 6VSB). (E and F) Nb34 (epitope III, green) and Nb95 (epitope IV, light pink) do not overlap with the hACE2 binding site and bind to spike conformations with at least two open RBDs (PDB ID 6XCN).

Other epitope binders do not fit into this inactive conformation without steric clashes and appear to use different neutralization strategies (Fig. 4C). For example, epitope II (Nb93) colocalizes with the hACE2 binding site and can bind the spike in the one RBD “up” conformation (Fig. 4D; PDB ID 6VSB) (3). This epitope may neutralize the virus by blocking the hACE2 binding site. Epitope III and IV Nbs can bind only when two or three RBDs are in their up conformations (PDB ID 6XCN) (24), in which the epitopes are exposed. In the all-RBDs-up conformation, three copies of Nbs can directly interact with the trimeric spike. Through RBD binding, epitope III (Nb34) can be accommodated on top of the trimer to lock the helices of S2 in the prefusion stage, possibly preventing their large conformational changes for membrane fusion (Fig. 4E). When superimposed onto the all-up conformation, epitope IV (Nb95) is proximal to the rigid N-terminal domain (NTD) of the trimer, presumably restricting the flexibility of the spike domains (Fig. 4F). Future high-resolution structural studies (e.g., by cryo-EM) of these Nbs in complex with the viral S protein will be needed to better understand the neutralization mechanisms.

Epitope mapping enabled us to bioengineer homo- and heterodimeric and homotrimeric Nbs. Homodimers and -trimers based on Nb20 or Nb21 were designed to increase the antiviral activities through avidity binding to the trimeric spike. Heterodimers pairing Nb21 with Nbs that bind a different epitope were designed to prevent viral escape. The homodimers and -trimers used flexible linker sequences of 25 (GS) or 31 (EK) amino acids (materials and methods). The heterodimers used flexible linkers of 12 amino acids.

Through a pseudovirus luciferase assay, we found up to ~30-fold improvement for the homotrimeric constructs of Nb213 (IC50 = 1.3 pM) and Nb203 (IC50 = 4.1 pM) compared with the respective monomeric form (Figs. 1, C and E, and 5, A and C). Similar results were obtained from the SARS-CoV-2 PRNT (Fig. 5, B and C, and fig. S11A). The improvements are likely greater than indicated by these values, as the measured values may reflect the assay’s lower detection limits. For the heterodimeric constructs (i.e., Nb21-Nb34), we observed up to a fourfold increase of potency. The multivalent constructs retained similar physicochemical properties to those of the monomeric Nbs (including high solubility, yield, and thermostability) and remained intact (nonproteolyzed) under the neutralization assay condition (fig. S10). They remained highly potent for pseudovirus neutralization after lyophilization and aerosolization (materials and methods and fig. S11, B to G), indicating the marked stability and potential flexibility of administration. Most of the RBD mutations observed in the Global Initiative on Sharing Avian Influenza Data (GISAID) (35) are very low in frequency (<0.0025), which may increase under Nb selection. Therefore, a cocktail consisting of ultrapotent, multivalent constructs that simultaneously bind a variety of epitopes with potentially different neutralization mechanisms will likely efficiently block virus mutational escape (Fig. 5E and fig. S12) (93638).

Fig. 5 Development of multivalent Nb cocktails for highly efficient SARS-CoV-2 neutralization.

(A) Pseudotyped SARS-CoV-2 neutralization assay of multivalent Nbs. The average neutralization percentage of each data point is shown (n = 2). ANTE-CoV2-Nab20TGS/EK: homotrimeric Nb20 with the GS or EK linker; ANTE-CoV2-Nab21TGS/EK: homotrimeric Nb21 with the GS or EK linker. (B) SARS-CoV-2 PRNT of monomeric and trimeric forms of Nbs 20 and 21. The average neutralization percentage of each data point is shown (n = 2 for the trimers; n = 4 for the monomers). (C) Summary table of the neutralization potency measurements of the multivalent Nbs. N/A, not tested. (D) Mapping mutations to localization of Nb epitopes on the RBD. The x axis corresponds to the RBD residue numbers (333 to 533). Rows in different colors represent different epitope residues. Epitope I: 351, 449 to 450, 452 to 453, 455 to 456, 470, 472, 483 to 486, and 488 to 496; epitope II: 403, 405 to 406, 408, 409, 413 to 417, 419 to 421, 424, 427, 455 to 461, 473 to 478, 487, 489, and 505; epitope III: 53, 355, 379 to 383, 392 to 393, 396, 412 to 413, 424 to 431, 460 to 466, and 514 to 520; epitope IV: 333 to 349, 351 to 359, 361, 394, 396 to 399, 464 to 466, 468, 510 to 511, and 516; epitope V: 353, 355 to 383, 387, 392 to 394, 396, 420, 426 to 431, 457, 459 to 468, 514, and 520.

In our study, in vivo antibody affinity maturation followed by advanced proteomics (25) enabled the rapid discovery of a diverse repertoire of high-affinity RBD Nbs, including an ultrapotent neutralizer with subpicomolar affinity, which is unusual for natural, single-domain antibody fragments. We demonstrated the simplicity and versatility of Nb bioengineering and the desirable physicochemical properties of the monomeric Nbs and their multivalent forms. To our knowledge, the multivalent constructs represent the most potent SARS-CoV-2 neutralizers to date. Flexible and efficient administration, such as inhalation, may further improve their antiviral efficacy while minimizing the dose, cost, and potential toxicity for clinical applications. The high sequence similarity between Nbs and human IgGs may restrain the immunogenicity (39). It is possible to fuse the antiviral Nbs with highly stable albumin-Nb constructs (40) to improve pharmacokinetics. These high-quality Nbs can also be applied as rapid and economic point-of-care diagnostics. We envision that the Nb technology described here will contribute to curbing the current pandemic and possibly a future event.



Globally there is an urgency to develop effective, low-cost therapeutic interventions for coronavirus disease 2019 (COVID-19). We previously generated the stable and ultrapotent homotrimeric Pittsburgh inhalable Nanobody 21 (PiN-21). Using Syrian hamsters that model moderate to severe COVID-19 disease, we demonstrate the high efficacy of PiN-21 to prevent and treat SARS-CoV-2 infection. Intranasal delivery of PiN-21 at 0.6 mg/kg protects infected animals from weight loss and substantially reduces viral burdens in both lower and upper airways compared to control. Aerosol delivery of PiN-21 facilitates deposition throughout the respiratory tract and dose minimization to 0.2 mg/kg. Inhalation treatment quickly reverses animals’ weight loss post-infection and decreases lung viral titers by 6 logs leading to drastically mitigated lung pathology and prevents viral pneumonia. Combined with the marked stability and low production cost, this novel therapy may provide a convenient and cost-effective option to mitigate the ongoing pandemic.



By January 2021, a year after the outbreak of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) was first reported (1), close to 100 million people have been infected by this highly transmissible virus resulting in significant morbidity and mortality worldwide. In addition to vaccines, there is an unparalleled quest to develop innovative and cost-effective therapeutics to combat the COVID-19 pandemic (2, 3). Early treatment using high-titer convalescent plasma (CP) may reduce the risk of severe disease in seniors (4), although CP is limited by supply. Potent neutralizing monoclonal antibodies (mAbs), predominantly isolated from COVID-19 patients for recombinant productions, have been developed for passive immunotherapy (5–14). In vivo evaluations of mAbs in animal models of COVID-19 disease such as murine, hamster, and nonhuman primates (NHPs), have provided critical insights into efficacy and the mechanisms by which they alter the course of infection (15–24). While mAb therapy lifts hopes to treat mild symptom onset in patients, they nevertheless require exceedingly high administration doses-typically several grams for intravenous (i.v.) injection (25, 26). The requirement of high doses for efficient neutralization may reflect SARS-CoV-2 virulence, pathogenesis, and the notoriously low efficiency of i.v. delivering these relatively large biomolecules across the plasma-lung barrier to treat pulmonary infections (27). Moreover, the associated high costs and challenges in bulk manufacturing may further limit the broad clinical use of mAbs worldwide (2).


In parallel efforts, we and others have recently developed camelid single-domain antibody fragments or nanobodies (Nbs) that primarily target the receptor-binding domain (RBD) of the SARS-CoV-2 spike (S) glycoprotein for virus neutralization (14, 28–31). Highly selected Nbs and the multivalent forms obtain high neutralization potency comparable to, or even better (per-mass) than, some of the most successful SARS-CoV-2 neutralizing mAbs. In particular, an ultrapotent homotrimeric construct, Pittsburgh inhalable Nanobody 21 (PiN-21), efficiently blocked SARS-CoV-2 infectivity at below 0.1 ng/ml in vitro (28). Compared to mAbs, Nbs are substantially cheaper to produce. Moreover, affinity-matured, ultrapotent Nbs are characterized by high solubility and stability that facilitate drug scaling, storage, and transportation, all of which are critical in response to pandemics. The excellent physicochemical properties and small sizes of Nbs raise an exciting possibility of efficient pulmonary delivery by aerosolization with characteristics of rapid onset of action, high local drug concentration/bioavailability, and improved patient compliance (needle-free) that may benefit a large population of SARS-CoV-2 infected patients (27–29, 32). However, despite the promise, no successful in vivo studies have been reported to date. The inferior pharmacokinetics of monomeric Nbs due to their small size (~15 kDa) and a lack of Fc-mediated immune effectors’ function, which is often required to augment the in vivo neutralizing activities of mAbs (33–35), drive potential concerns for Nb-based therapy. It remains unknown if the high in vitro neutralization potency of SARS-CoV-2 Nbs can be translated into in vivo therapeutic benefits.


In this study, we systematically evaluated the efficacy of PiN-21 for prophylaxis and treatment of SARS-CoV-2 infected Syrian hamsters which model moderate to severe COVID-19 disease. We provided direct evidence that ultra-low administration of PiN-21 efficiently treats the virus infection. Notably, PiN-21 aerosols can be inhaled to target respiratory infection which drastically reduces viral loads and prevents lung damage and viral pneumonia. This novel Nb-based therapy shows high potentials for the treatment of early infection and may provide a robust and affordable solution to address the current health crisis.



PiN21 efficiently protects and treats SARS-CoV-2 infection in Syrian hamsters

To assess the in vivo efficacy of PiN-21, 12 hamsters were divided into two groups and infected with 9 x 104 plaque-forming units (p.f.u.) of SARS-CoV-2 via the intratracheal (IT) route. Shortly after infection, Nb was delivered intranasally (IN) at an average dose of 0.6 mg/kg (Fig 1A). Animals were monitored daily for weight change and clinical signs of disease. Half of the animals were euthanized 5 days post-infection (d.p.i.) and the remaining were euthanized 10 d.p.i. Virus titers in lung samples from the euthanized animals were measured by plaque assay. Nasal washes and throat swabs were collected at 2 and 4 d.p.i. to determine viral loads in the upper respiratory tract. Consistent with published studies (36, 37), IT inoculation of hamsters with SARS-CoV-2 resulted in a robust infection, rapid weight loss in all animals up to 16% at 7 d.p.i. and resulting recovery and reversal of weight loss by 10 d.p.i. before recovery. However, concurrent IN delivery of PiN-21 eliminated any significant weight loss in the infected animals (Fig 1B). This dramatic protection was accompanied by a reduction of viral titer in the lungs, with an average decrease of 4 order of magnitude in the lung tissue, respectively, compared to control on 5 d.p.i. (Fig 1C). Consistently, a 3-log reduction of the viral genomic RNA (gRNA) by reverse transcriptase (RT)-qPCR was evident on 5 and 10 d.p.i. (Fig S1A-B). Infectious virus was essentially cleared by 10 d.p.i.


Figure 1.

Download figure

Open in new tab

Figure 1.

PiN-21 protects Syrian hamsters from SARS-CoV-2 infection.

A. Overview of the experimental design. 9 x 104 p.f.u. of SARS-CoV-2 was IT inoculated followed by IN delivery of 100 μg PiN-21 or a control Nb. Animal weight changes were monitored daily. Nasal washes and throat swabs were collected on 2 and 4 d.p.i. Animals were euthanized for necropsy on 5 (n=3) and 10 d.p.i (n=3).


B. The protection of weight loss of infected hamsters treated with PiN-21. *** indicates a p-value of < 0.001.


C-E. Measurement of viral titers by the plaque assay. ** indicates a p-value of < 0.01. The dashed line indicates the detection limit of the assay.


Notably, the virus was undetectable in the upper respiratory tract (URT) including both nasal washes and throat swabs of all PiN-21-treated animals on 2 d.p.i. This is significantly different from the control group, where varying levels of infectious virus were present (Fig 1D-E). Furthermore, five out of six PiN-21 treated animals remained protected from detectable infection 4 d.p.i. The results were further supported by a drastic decrease of gRNA in the URT (Fig S1C-D). Together, this demonstrates that the high in vitro neutralization potency of PiN-21 can be translated into therapeutic benefits in vivo independent of Fc-mediated immune responses. PiN-21 can efficiently protect SARS-CoV-2 infection in hamsters by rapidly and drastically suppressing viral replication in both the URT and lower respiratory tract (LRT).


Previous studies reveal that clinical mAbs are less effective for COVID-19 treatment (post-infection) than for prophylaxis (pre-infection) in animal models, possibly reflecting the virulence of SARS-CoV-2, speed of virus replication, and rapid symptom onset (15, 22, 38). Therefore, we evaluated the therapeutic potential of PiN-21 since it was highly effective when co-administered. To explore the second route of infection, hamsters were inoculated IN with 3 x 104 p.f.u. of SARS-CoV-2. PiN-21 or a control Nb (0.6 mg/kg) was IN-delivered to animals 6 hours post-infection (h.p.i.). Animal weights were monitored daily, throat swabs and nasal washes were collected, before euthanized on 6 d.p.i. (Fig S2A). Similar to the IT route, IN-infection of hamsters with SARS-CoV-2 resulted in precipitous weight losses in the control animals. Encouragingly, intranasal treatment using PiN-21 significantly reduced weight loss throughout the assessment period (Fig S2B), paralleling the results of clinical mAbs in the same model albeit using substantially higher doses. Less than 100-fold reduction in virus titers were found in nasal washes and throat swabs on 2 and 4 d.p.i. (Fig S2C-D). Moreover, infectivity was undetectable in lung tissues 6 d.p.i., indicating that the virus has been predominantly cleared. Analysis of early time points will be needed to better understand virus suppression by Nb treatment.


PiN21 aerosolization effectively treats SARS-CoV-2 infected hamsters at an ultra-low dose

The marked physicochemical properties of PiN-21 prompted us to evaluate pulmonary delivery by inhalation. To evaluate the impact of construct size and pharmacokinetics on lung uptake, we fused monomeric Nb21 and PiN-21 to an Nb that binds serum albumin (Alb) of both human and rodents with high affinity to generate two serum-stable constructs (Nb-21Alb and PiN-21Alb) (39). Using a portable mesh nebulizer, we aerosolized Nb21Alb, PiN-21, and PiN-21Alb and evaluated their post-aerosolization neutralization activities by pseudovirus neutralization assay. All constructs retained high neutralization potency in vitro (Fig S3B). The amount of Nbs recovered post-aerosolization was inversely correlated with the size of constructs (Fig S3A). Moreover, while Nb-21Alb had the highest recovery, the post-aerosolization in vitro neutralization activity was substantially lower than for other constructs, Nb-21Alb was therefore excluded from downstream therapeutic analyses.


Next, we compared two ultrapotent constructs PiN-21 and PiN-21Alb for targeted aerosolization delivery into hamsters. Nbs were aerosolized using a nebulizer (Aerogen, Solo) that produces small aerosol particles with a mass median aerodynamic diameter of ~ 3 μm (Table S1). Animals were sacrificed 8 and 24 hours post-administration to assess Nb distribution and activities when recovered from various respiratory compartments and sera (Fig 2A). Consistent with the result using the portable nebulizer, we found the inhalation dose of PiN-21 was approximately two times PiN-21Alb (at 8 h, 41.0 μg or 0.24 mg/kg for PiN-21 v.s. 23.7 μg or 0.13 mg/kg for PiN-21Alb) (Table S1) while neutralization activity as assessed by plaque reduction neutralization test (PRNT50) of SARS-CoV-2 remained essentially unchanged after aerosolization of both Nb constructs (Fig 2B).


Figure 2.

Download figure

Open in new tab

Figure 2.

Assessment of Nb delivery in the hamster respiratory system

A. Schematic design of PiN-21 and PiN-21Alb aerosolization in hamster models.


B. Nb neutralization potency before and after aerosolization measured by PRNT50 assay.


C and D. Normalized overall neutralization activity by plaque assay of PiN-21 and PiN-21Alb of different time points post-aerosolization.


The neutralization activities of both Nb constructs were detected throughout the respiratory tract and in sera. As expected, within the airways the neutralizing activities were predominantly associated with bronchoalveolar lavage (BAL) fluid, followed by tracheal aspirate, larynx wash, and nasal wash samples (Fig 2C-D). Compared to 8-hour post-inhalation, we found that the amounts and activities of Nbs in BAL, but not in sera, were substantially lower 24-hour post-inhalation, possibly indicating more rapid clearance. In addition, Nb conjugation to serum albumin did not seem to impact the activities in the airways, whilst stability was enhanced in the serum. These data underscore the requirement of an ultra-low dose of the ultrapotent PiN-21 construct to neutralize SARS-CoV-2 infectivity in vivo efficiently. Finally, PiN-21 was preferentially selected for further evaluation owing to the high stability and resistance to aerosolization, which are likely critical for clinical applications.


To assess the therapeutic efficacy of PiN-21 by inhalation, 12 hamsters were IN inoculated with SARS-CoV-2 (3 x 104 p.f.u.) followed by single-dose aerosolization treatment of either PiN-21 or the control Nb at 6 h.p.i. Animals were monitored for weight loss, throat swabs and nasal washes were collected daily. Animals were euthanized (3 d.p.i.) and lungs and trachea were collected for virological, histopathology, and immunohistochemical analysis (Fig 3A). Notably, pulmonary delivery of PiN-21 aerosols, despite only a minute amount, led to a remarkable reverse of weight loss in the treated animals. The average weight gain was 2% in PiN-21 versus 5% loss in the control on 3 d.p.i. (Fig 3B). The weight loss in the control group was highly reproducible when compared with the above experiments. Critically, aerosolization treatment diminished infectious viruses in lung tissue by 6 orders of magnitude (Fig 3C). The treatment also substantially decreased virus gRNA in the lungs (Fig S4C). Moreover, we observed a substantial reduction of viral titers in nasal washes and throat swabs (Fig S4A-B). This indicates that Nb administration by aerosolization may limit human-to-human transmission of SARS-CoV-2.


Figure 3.

Download figure

Open in new tab

Figure 3.

Treatment efficacy of aerosolized PiN-21 in the hamster model of SARS-CoV-2.

A. Overview of the experiment design. 3 x 104 p.f.u. of SARS-CoV-2 was IN inoculated. PiN-21 or a control Nb was aerosolized to hamsters in the cage 6 h.p.i. Animal weight changes were monitored, nasal washes and throat swabs were taken daily. Animals were euthanized for necropsy on 3 d.p.i.


B. The percentage of body weight change of PiN-21 aerosol-treated animals compared to the control (n=6).


C. Reduction of viral titers in hamster lungs (3 d.p.i.). Significant differences were observed between treated and control groups. **, P < 0.01; *, P < 0.05. The dashed line indicates the detection limit of the assay.


D. Lung pathology scores of treated and control groups. Significant difference was denoted by ****, P < 0.0001.


E. H&E staining of necrotizing bronchointerstitial pneumonia affiliate with abundant SARS-CoV-2 S antigen in bronchiole epithelium and alveolar type 1 and 2 pneumocytes in the control group. All images acquired at 20 x, scale bar = 100 μm. Areas marked by boxes are shown at higher magnification in the right-most panel (scale bar = 25 μm).


F. Immunostainings of bronchointerstitial compartments (3 d.p.i.). Orange, SARS-CoV-2 S; magenta, CD68/macrophages; red, CD3e+ T cells; teal, ACE2; Grey, DAPI. The bronchiole is outlined by white hash. Total magnification 200x, scale bar = 100 μm.


Effective control of SARS-CoV-2 infection in the LRT of PiN-21 aerosolized animals

To understand the mechanisms by which Nb aerosols prevent and/or ameliorate lower respiratory disease caused by SARS-CoV-2 infection better, we performed whole lung semi-quantitative ordinal histologic analysis of control (n=6) and PiN-21 (n=6) treated animals euthanized at 3 d.p.i (Table S2-3). Cumulative scores encompassed pathologic features of airways, blood vessels, and alveoli/pulmonary interstitium. PiN-21 aerosolization protected most animals (5/6) from severe COVID-related histopathologic disease reflected by decreased ordinal scores (P < 0.0001) when compared to Nb treated controls (Fig 3D, Table S4). Histopathologic findings in the control group resembled previous reports of SARS-CoV-2 inoculation in Syrian hamsters (40, 41). Pulmonary disease observed in the PiN-21 treated animals was very mild (Fig 3E, Fig S5) being characterized by the absence of severe necrotizing bronchiolitis in the majority of animals (5/6), a pathological finding ubiquitously observed in all control animals. Furthermore, the single PiN-21 treated animal with necrotizing bronchiolitis had localized disease, compared to the multifocal and bilateral distribution observed in most Nb controls (4/6). Bronchiolitis was also affiliated with less severe bronchial hyperplasia and hypertrophy and absence of syncytial cells when compared to Nb controls. The predominant histologic finding in PiN-21 treated animals was minimal-to-mild perivascular and peribronchial mononuclear inflammation consisting of macrophages and lymphocytes. Furthermore, aside from the one animal already mentioned, the PiN-21 group had considerably less interstitial inflammation with decreased vascular permeability, as indicated by the absence of perivascular and intra-alveolar edema, hemorrhage, and fibrin exudation (Fig S5).


In control animals, S antigen was abundant in the cytoplasm of the bronchiolar epithelium, with less common detection in alveolar type 1 and 2 pneumocytes. Interstitial and peribronchiolar infiltrates were composed of large numbers of CD3e+ T cells and CD68+ macrophages, with a complete absence of angiotensin-converting enzyme 2 (ACE2) in the apical cytoplasm of bronchiole epithelium in areas with abundant viral S (Fig 3F, upper panel). Consistent with a striking 6-log virus reduction after aerosolization, S antigen was extremely sparse (<1% of permissive cells) in all PiN-21 treated animals, with decreased T cell and macrophage immune cell infiltrate, and retention of native apical bronchiole ACE2 expression (Fig 3F, lower panel).


To determine the impact of PiN-21 on the upper airways of the lower respiratory tract we also examined the trachea histologically. In PiN-21 treated animals, tracheas for all animals were within normal limits, while mild to moderate neutrophilic and lymphohistiocytic tracheitis, with variable degrees of degeneration and necrosis, and segmental hyperplasia and hypertrophy were observed in all the control animals (Fig 3F). In summary, our data clearly demonstrate that PiN-21 aerosolization given during early disease course is highly effective in decreasing SARS-CoV-2 entry, subsequent replication in permissive epithelial cells of the lower respiratory tract and this, in turn, has a major impact on viral shedding. The result is the prevention of disease, including decreased cytopathic effect on permissive epithelial cells, retention of ACE2 expression on permissive bronchioles, and decreased recruitment of inflammatory cells to sites of replication.


Figure S8.Cryo-EM analysis of Nb105:S and Nb105:RBD: Nb21 complexes (related to Figure 3).

(A) Representative micrograph and 2D class averages of Nb105:S complex.

(B) Gold-standard Fourier shell correlation (FSC) and Euler angular distribution.

(C) Representative micrograph and 2D class averages of Nb105:RBD:Nb21 complex.

(D) Gold-standard Fourier shell correlation (FSC) and Euler angular distribution.

(E) Local resolution estimation for Nb105:RBD:Nb21 complex.

(F) Rigid docking of Nb105:RBD complex to the interface of the dimeric S. The interface highlighted with the green line is between the Nb framework and RBS.

Figure 3.Structures of class II Nbs (95, 34, and 105)

3A: Cryo-EM structures of Nbs 95 and 34 in complex with S.

3B: Cryo-EM structure of the Nb105:Nb21:RBD complex.

3C: Nb95: RBD interactions. Residues in pink denote Nb95 for RBD binding.

3D: Nb105: RBD interactions. Residues in yellow denote Nb105 for RBD binding.

3E: Class II Nb: RBD interactions are predominantly mediated by CDR3. Nbs are represented as ribbons. The CDR3 loops are shown as surface presentations.

3F: Steric effects of class II Nbs on hACE2:RBD interactions. N322 glycosylation (ACE2) is presented in red density.

Class II Nbs bind RBD with high specificity primarily through the hydrophobic residues on a long CDR3 loop (17 or more residues). Notably, they have evolved divergent CDR3 sequences to target largely overlapping epitopes (Figure 3E). Major interactions of Nb95:RBD and Nb105: RBD were resolved by local refinements (Methods). For Nb95, the CDR3 side chains of D99, K100, and S108 form ionic or hydrogen-bonding interactions with the side chains of K378 and Y380 and the main chain carbonyl group of S375 of RBD, respectively (Figure 3C). R408 of RBD also forms a hydrogen bond with the main chain amine group of CDR3 E112. CDR3 residues P110 and F109 form hydrophobic interactions with V407, V503, and Y508 of RBD. Additionally, the Y106 side chain extends to interact with a hydrophobic patch formed by residues between Y369 and F374 (RBD). This interaction is further supported by a framework residue Y59, which forms hydrogen bonds with carbonyl groups of A372 and F374 of RBD (Figure 3C). Similar to Nb95, Nb105 recognizes RBD with CDR3 W104 and Y106 residing in two hydrophobic patches of M379-P384 and Y369-F377, respectively (Figure 3D). Another hydrophobic residue F111 is clamped between V407 and R408 (RBD), forming a cation-π stacking interaction. The three patches of hydrophobic interactions surround an electrostatic interaction between E112 and K378 of RBD (Figure 3D). Moreover, other CDRs of Nb105 help stabilize the CDR3:RBD interactions. For example, R53 (CDR2) forms a cation-π stacking on W103 (CDR3) (Figure 3D). While the Nb34:RBD interface was only partially resolved, it is evident that the CDR3 also plays a major role in RBD binding similar to Nb95 and Nb105 (Figure 3E). Superposition of three Nb: RBD structures reveals small yet different binding orientations of Nbs with respect to the RBD (Figure 3E).

While class II Nbs do not interact with the ACE2 directly, they can still efficiently block ACE2 binding at low nM concentrations (Figure 4H). Superposition of ACE2:RBD into the Nb: RBD complexes reveal that binding of Nbs 105 and 95 to the RBD overlaps with the subdomain II of ACE2 (residues 308 – 326) and the N322 glycan, wherever Nb34 can clash with the glycan (Figure 3F). Recent crystal structures of camelid Nbs 5 and synthetic constructs (sybodies)9 have revealed a similar mode of binding with substantially lower neutralization potency (i.e., over 100-fold) compared to affinity matured class II Nbs (95, 34, and 105). Potentially, this is due to the marked difference of these Nbs in their abilities (e.g., affinity, stability, and orientation with respect to RBD) to block ACE2 binding.

Figure 4.Structures of class III Nbs (17 and 36)

4A: Cryo-EM structures of Nb17 in complex with S.

4B: Nb17:RBD interactions are mediated by all three CDRs.

4C: Cryo-EM structure of the Nb17:Nb105:RBD complex.

4D: Nb17 structurally does not overlap with ACE2.

4E: Cryo-EM structure of the Nb36:Nb21:RBD complex.

4F: Epitope of Nb36 on the RBD surface.

4G: Nb17 stacks on NTD via its framework, while isolated Nb36:RBD complex indicates Nb36 would clash with neighboring NTD on S.

4H: ACE2 competition assay with the S.

Class III Nbs utilize distinct mechanisms to neutralize the virus efficiently

a. Nb17 locks the spike in all RBD-up conformations

Nb17 neutralizes the virus in vitro at an IC50 of ~25 ng/ml. Notably, all three RBDs on the S trimer were in an open conformation with two having particularly strong densities (Figure 4AS9A-D). Nb17 binds a conserved epitope including a segment spanning residues 345-356 and additional residues that generally do not overlap with the RBS. This epitope is localized on the opposite side of class II Nb epitopes (Figure 4B). Similar to Nb21, Nb17 also utilizes all three CDRs for RBD recognition. Remarkably, no bulky side chains directly involve interface packing and ultrahigh affinity RBD binding. Instead, small hydrophobic residues, such as Ala, Val, and Pro, and polar interactions are the primary contributors at the Nb17: RBD interface (Figure S9G). Interestingly, 3D variability analysis shows that Nb17 density stacks on the adjacent NTD prefer the open conformation of all RBDs when Nb17 is bound (Movie S1).

Figure S9.Cryo-EM analysis of Nb17:S and Nb17:RBD:Nb105 complexes (related to Figure 4).

(A) Representative micrograph and 2D class averages of Nb17:S complex.

(B) Gold-standard Fourier shell correlation (FSC) and Euler angular distribution.

(C) Local resolution estimation for Nb17:S complex.

(D) Focused classification of the flexible region in Nb17:S complex. The density of Nb17 in class 1 (cyan) is smeared due to motion along the y-direction, class 2 (magenta) has well resolved RBD, Nb17, and NTD density, and both densities of RBD and Nb17 is lost due to motion along the x-direction.

(E) Representative micrograph and 2D class averages of Nb17:RBD: Nb105 sample.

(F) Local resolution estimation for Nb105:RBD: Nb21 sample.

(G) Interface residues of Nb17:RBD complex.

(H) Alignment of Nb17:RBD to Nb21:RBD showing the large overlap between Nb17 CDR3 with Nb21 CDR2 and partially Nb21 CDR1.

To validate our structural model, we reconstituted and analyzed the Nb17: Nb105: RBD complex to characterize the interface interactions(Figure 4CS9E-F). Superposition of the Nb17: RBD complex to the ACE2:RBD complex indicates that Nb17 would not interfere with ACE2 interactions (Figure 4D). Structural alignment reveals that the CDR3 of Nb17 overlaps with Nb21 to compete for RBD binding (Figure S9H), although it remains uncertain how Nb17 neutralizes the virus. We speculate that with ultra-high affinity, Nb17 can lock S1 in a rigid and open conformation that prevents the host membrane fusion process.

Nb17 is highly resistant to all the dominant natural RBD mutations that we have tested. Compared to the Nb21:RBD interface, where E484 is buried inside the core of the interface, E484 localizes at the rim of the Nb17:RBD interface (Figure S9H). As such, while E484 directly contacts Nb17, the mutation (E484K) does not affect RBD binding (Figure 1A). The loss of binding to the super variant of RBD62 is likely caused by two point mutations (I468 and T470) (Figure S11).

a. Nb36 destabilizes the spike trimer

While Nb36:S complex is highly soluble, to our surprise, particles were not detected on the EM grids under cryogenic conditions. Therefore, to characterize Nb36:S interactions, we titrated different concentrations of Nb36 with S protein and imaged the complexes by negative stain EM. The increasing concentration of Nb36 coincided with an enhanced blurring of the particles, which compromised contrast in the electron micrographs (Figure S10A). This observation suggests that Nb36 can destabilize the integrity of the spike. To test this assumption, we employed thermal shift melting assays under similar conditions as those used for negative stain EM. Consistently, an increase in Nb36 concentration correlated with a decrease in protein melting temperature, indicating that Nb36 promotes instability of the S complex (Figure S10B). To map the epitope, we reconstituted and imaged the Nb36:Nb21:RBD complex by cryo-EM (Figure 4ES10C-E). The analysis reveals that the Nb36 epitope partially overlaps with Nb17 while exhibiting no overlap with Nb21 (Figure 4F). The epitope covers a small segment on the non-RBS region (residues 353-360 of RBD) as well as distinct, non-RBS epitope residues that contact Nb17. Nb36 binds RBD in an orientation that is markedly different from Nb17. Superposition of the structure onto S reveals that Nb36 would have a significant steric clash with the neighboring NTD in the trimeric S complex (Figure 4G). Potentially, with its small size and convex epitope, Nb36 can efficiently insert between an RBD and the adjacent NTD destabilizing the highly dynamic structure of the S trimer. This destabilization mechanism is a reminiscence of mAb CR3022. However, Nb36 targets a completely different epitope from CR3022 with substantially higher neutralization potency 30,31.

Figure S10.EM Analysis of Nb36 with S and RBD (related to Figure 4).

(A) Representative negative stain EM micrographs of spike protein in the presence of an increased concentration of Nb36. An example of an intact trimeric spike particle is highlighted by a blue arrow, and an example of a disrupted spike particle is highlighted by a red arrow.

(B) Thermal melting profile of S protein in the presence of an increased concentration of Nb36.

(C) Representative micrograph and 2D class averages of Nb36:RBD: Nb21 complex.

(D) Gold-standard Fourier shell correlation (FSC) and Euler angular distribution.

(E) Local resolution estimation for Nb36:RBD: Nb21 complex.

Collectively, high-resolution cryo-EM analyses revealed both class II and III Nbs bind conserved epitopes that are incompatible with the mutational escape, thus less likely to converge into a circulating variant.

Class III RBD Nbs belongs to a novel class of neutralizing Nbs

To investigate the RBD epitopes and Nb neutralization mechanism systematically, we analyzed all the available structures that include Nb:RBD interactions (Figure 5A). Epitope clustering supports the notion of three distinct classes of neutralizing Nbs. As expected, most Nbs bind class I epitopes that mainly cover RBS (Figure 5A-B). Class I and II epitopes are shared between Nbs and mAb (Figure 5C-D). In contrast, our class III Nbs are novel and unique among all the neutralizing Nbs and mAbs that have been characterized (Figure 5E). Class III epitopes are in close proximity to the neighboring NTD. Thus, accessing these epitopes is elusive for mAbs due to steric hindrance imposed by their large sizes. Here, with optimal orientations and substantially smaller sizes, Nbs can target this conserved region where the virus has low mutational tolerance32. As such, class II and III epitopes may serve as optimal targets for the development of pan-coronavirus vaccines and therapeutics.

Figure 5.Class III Nbs bind novel and conserved neutralizing epitopes unique to Nbs

5A: Epitope clustering analysis of RBD Nbs and correlation with RBD sequence conservation and ACE2 binding sites.

5B: Overview of three Nb classes binding to the RBD, RBD surface was colored based on conservation (ConSurf score).

5C-E: Structural comparison of different classes of Nbs with the closest mAbs for RBD binding.

Nbs and mAbs are differently affected by mutations in the circulating variants

To understand how the unique binding modes of neutralizing Nbs are translated into their high resistance against SARS-CoV-2 mutants we compare the three Nb classes with mAbs. Buried surface area (BSA) of RBD interfacing residues from both Nb and mAb-bound structures were calculated and compared systematically (Figure 6A-C). The analysis reveals that the majority of mAbs (83%) use at least one of the mutated RBD residues to bind, with 60% using two or more variant residues for RBD interactions. In contrast, Nbs target these sites substantially less frequently (Figure 6B) with the exception of class I Nbs, which predominantly recognizes the hot spot fostered by E484 (Figure 6D). Other classes do not bind these variant residues directly (Figure 6B6E).

Figure 6.mAbs and Nbs binding to RBD are differently affected by mutations in the circulating variants

6A: Localization of six RBD residues where major circulating variants mutate.

6B: Buried surface area of Nbs by different RBD residues.

6C: Buried surface area of Fabs by different RBD residues.

6D-E: Representative structures of different classes of Nbs with major variants residues shown as spheres.

Two fab structures binding similarly to Class I Nbs were also shown on the side.

6F: The boxplot showing probability of an epitope residue to hit one of the mutations in variants.

The fact that many critical mutations localize on the RBS is intriguing (Figure 5A). Under selection pressures, the virus appears to have evolved a highly efficient strategy to evade host immunity by preferentially targeting this critical functional region. Specific RBS mutations (such as K417N and E484K) may help this novel zoonotic virus to optimize host adaptation (improved ACE2 binding) achieving higher transmissibility 33. In parallel, as RBS is the main target of serologic response, the mutations provide an effective means for the resulting variants to escape the neutralizing pressure from serum polyclonal antibodies efficiently 27,3234. Since most clinical mAbs originate from convalescent plasma, it is not surprising that they are less effective to the convergent circulating variants 19,20. Fundamentally, this is different from neutralizing Nbs which have never been co-evolving with the virus, and therefore are less sensitive to the plasma-escaping variants. Indeed, we found that the probability of neutralizing Nb epitopes coinciding with the variant mutations was substantially lower than that of mAbs (Figure 6F). This is particularly the case for highly potent and in vivo affinity-matured Nbs, as shown in this study. Together with the functional data (Figure 1), our analysis provides a solid structural basis to understand how potent neutralizing Nbs can resist the convergent variants. Nbs may provide additional therapeutic benefits over mAbs for the evolving variants of SARS-CoV-2.

Systematic comparisons of mAbs and Nbs for RBD binding

Development and structural characterization of a rich collection of neutralizing mAbs and Nbs provide an unparalleled opportunity to compare the two distinct types of antibodies for RBD binding and their respective mechanisms of action. We compiled and analyzed all the available structures including 56 distinct mAb-bound complexes and 23 Nb-bound complexes (Table S4). Interface residues were extracted using a 6 Å-distance cutoff to generate the residue contact profiles and epitope-paratope propensity maps. BSA and epitope curvatures were also calculated and plotted (Methods).

Nbs have lower BSA values than Fabs (μNb=779 Å2 v.s. μFab=862 Å2,p=0.055) (Figure 7A). In addition, the distribution of BSA for Nbs is distinct from Fabs and is substantially narrower (σNb = 151 Å2Fab = 210 Å235 (Figure 7A). Despite the smaller size, Nbs have evolved multiple strategies for high-affinity RBD binding. They exploit surface residues (especially using CDR3 loops) significantly more efficiently than Fabs for RBD engagement (Figure 7B-C). Nbs also have higher BSA per-interface residue (Figure 7B). The involvement of the framework (FR) regions in RBD binding, particularly FR2, is also evident probably due to the absence of light chain pairing (Figure 7C). Interestingly, we found that compared to in vivo affinity-matured RBD Nbs, in vitro selected Nbs tend to use highly conserved FR sequences more extensively for interactions. More dominant involvement of conserved FR likely suggests that in vitro selected Nbs may interact less specifically to RBD (Figure 7E7G). Compared to Fabs, Nbs bind more concave surfaces (Methods) to tighten the interactions (Figure 7D7F). Finally, neutralizing Nbs employ electrostatic interactions more extensively while both types of antibodies predominantly use hydrophobic interactions to achieve high specificity (Figure S1213).

Figure S11.Analysis of the interactions between the RBD residues of I468 and T470 and highly potent neutralizing Nbs from class III (related to Figure 4).

Figure S12.Comparison of neutralizing Nbs and mAbs for RBD binding (related to Figure 7).

(A) The heatmap shows the binding difference between Nbs and Fabs in terms of paratope residue utility despite overall similar epitope regions.

(B) Heatmaps show the difference in preference of epitope-paratope residues between Nbs and Fabs. The comparisons were made separately for RBS binders and non-RBS binders. Nbs with at least 30% overlapping residues with ACE2 binding sites were considered RBS binders.

(C) Illustrations of dominated electrostatic interactions formed between arginine from Nb CDRs and RBD residues. RBD was colored in dark gray, Nbs were colored in khaki, E484 (RBD) was colored in red, F490 (RBD) was colored in teal and R (Nb CDRs) was colored in blue.

Figure S13.Analysis of interactions of E484 (RBD) with neutralizing Nbs and mAbs (related to Figure 7).

Superposition of Fab-RBD structures showing E484 (RBD) forms hydrogen and/or hydrophobic interactions with the respective residues of Fabs. The side chains of residues tyrosine, serine, threonine and arginine in close contact E484 are shown in stick representation. RBD: dark gray, Fab VH: light blue, Fab LH: light green, and residue E484 (RBD): red.

Figure 7.Comparisons of RBD neutralizing Nbs and mAbs

7A: Buried surface areas of RBD: Nb and RBD: Fab complexes. VH: heavy chain. VL: light chain.

7B: Buried surface areas per-interface residue for Nbs and Fabs.

7C: The contact contribution of CDRs and FRs of Nbs and Fabs in RBD binding (using a 6 Å cutoff). Contact contribution % was calculated as # of contacting residues on CDR or FR region/total # of contacting residues.

7D: Quantification of interface cavity. Y-axis is the curvature value.

7E: Comparison of contributions from CDRs and FRs for RBD binding between in vivo matured Nbs and in vitro selected Nbs.

7F: Representative structures showing different binding modes (epitope curvature) of Nb17 and a Fab. Nbs target concave RBD surfaces to achieve high-affinity binding.

7G: Representative structures showing the direct involvement of FR2 from an in vitro selected Nb (PDB# 7A29) for RBD interaction.


By March 2021, over 116 million people had been infected by SARS-CoV-2. While the virus continues to rage with hundreds of thousands of daily new infections, the prospect of curbing it rests on the development of effective interventions including vaccines and therapeutics that are broadly neutralizing to resist both the current and future circulating variants. Highly selective neutralizing Nbs and the multivalent Nb forms represent some of the most potent antiviral agents 58,13. They are characterized by marked stability and solubility to facilitate robust and low-cost manufacturing and storage, which are essential during the pandemic. Critically, the stable constructs can be delivered by novel aerosolization route to treat virus pulmonary infection with high efficacy 14.

Compared to neutralizing mAbs that have been systematically characterized, structures and mechanisms for highly potent neutralizing Nbs had yet to be investigated. In this work, by using single-particle cryo-EM, we have solved a repertoire of such Nbs in complex with the spike or RBD. Neutralizing Nbs can be grouped into three epitope classes. Class I Nbs are the most abundant neutralizing Nbs and bind RBS, which is one of the least conserved regions on the spike. They can bind both up and down RBD conformations on the spike (Figure 2). While class I Nbs can neutralize the virus at extremely low concentrations (e.g., sub-ng/ml), their RBD binding can be abolished by a single point mutation E484K. Class II and III bind conserved epitopes that are not accessible in the closed RBD conformation. Class II Nbs can bind 2-up-1-down RBD conformations. They bind non-RBS epitopes yet can still sterically block ACE2 binding (Figure 3). Class III Nbs bind novel and conserved epitopes that have not been previously discovered (Figure 4). The cryptic RBD epitopes are in the vicinity of NTD that are highly challenging to access by large mAbs. While Nb36 can potentially neutralize the virus by destabilizing the spike, it remains unclear how the highly potent class III neutralizer Nb17 functions. Presumably, Nb17 can promote the shedding of S1 to inhibit virus-host entry, with ultra-high affinity. Nevertheless, the unique RBD binding mode of Nb17 and its ease of bioengineering may facilitate the development of robust biosensors to detect specific conformations of the spike. Currently, this process can only be inferred computationally 3639 or by single-molecule techniques which require extensive bioengineering to label the spike with fluorescence dyes 17. Finally, since Nb17 binds extremely tightly to the RBD-up conformations, potentially, it could be used as a novel Nb “adjuvant” for vaccines to facilitate exposure of conserved yet cryptic RBD epitopes that could help elicit broadly neutralizing activities insensitive to the evolving variants.

The convergent variants emerged from persistent serologic pressure in vivo 4042, and current interventions were developed against the initial SARS-CoV-2 strain that emerged in 2019. It is therefore not surprising that the prevalent variants can evade most clinical antibodies and weaken the polyclonal antibody activities from the vaccine-elicited sera. Compared to mAbs, however, Nbs originate from camelids or are identified by in vitro affinity maturation. With small sizes as well as distinct structural and physicochemical properties, Nbs can target conserved and unique epitopes on the highly dynamic spike structure that are likely fundamentally inaccessible by mAbs. Our systematic analysis including both high-resolution structural studies and functional data suggests that potent neutralizing Nbs are highly resistant to the convergent circulating variants, and possibly future variants that may occur under vaccine-elicited herd immunity (Figure 1). Our structure-function investigations provide a framework to map neutralizing epitopes systematically and to understand the mechanisms by which Nbs efficiently inhibit the virus and its variants (Figure 6D-F).

The novel structural information presented here may also help rational design of “pan-coronavirus” therapies and vaccines. The breadth and potency of class II and III Nbs could be further improved by in vitro affinity maturation using phage or yeast display methods. Guided by the structures, it is now possible to generate multivalent constructs by fusing specific class II and III Nbs with a proper length(s) of flexible linkers. Such constructs may enable avidity binding to substantially improve the neutralization potency while locking the virus on the highly conserved sites with a very low tolerance for future mutations. Combining these multivalent constructs into novel cocktails may provide the ultimate protection from the mutational escape 8. Future studies will be needed to test these ideas and critically, evaluate the in vivo efficacies of these novel inhalable constructs in both preclinical and clinical settings 14.

Tables S1-4

Table S1. Summary of all RBD mutants, related to Figure 1.

Table S2. Statistics for 3D reconstruction and model refinement for Nb:S complexes, related to Figures 23 and 4.

Table S3. Statistics for 3D reconstruction and model refinement for 2Nbs:RBD complexes, related to Figures 3 and 4.

Table S4. Summary of structural comparisons between mAbs and Nbs, related to Figure 6 and 7.


Protein expression and purification

The plasmid with cDNA encoding SARS-Cov-2 spike HexaPro (S) was obtained from Addgene. To express the S protein, HEK293-ES cells were transiently transfected with the plasmid using polyethyleneimine and 3.5 mM valproic acid sodium salt to enhance protein production. After 3 hours of transfection, 1 μM kifunensine was added to further boost protein expression. Cell culture was harvested three days after transfection and the supernatant was collected by high-speed centrifugation at 13,000 rpm for 30 mins. The secreted S protein in the supernatant was purified using Ni-NTA agarose columns. Protein eluates were then concentrated and further purified by size-exclusion chromatography using a Superose 6 10/300 column (Cytiva) in a buffer composed of 20 mM Hepes pH 7.5 and 200 mM NaCl. The purified S protein was then pooled and concentrated to 1mg/ml.

The receptor-binding domain (RBD) was expressed and purified as described previously8. Briefly, RBD was expressed in Sf9 insect cells as a secreted protein using the baculovirus method. A FLAG-tag and an 8x His-tag were fused to the N terminus of the RBD sequence, and a TEV protease cleavage site was inserted between the His-tag and RBD. The protein was purified by nickel-affinity resins, followed by overnight TEV protease treatment and size exclusion chromatography (Superdex 75). The purified protein was concentrated in a buffer containing 20 mM Hepes pH 7.5 and 150 mM NaCl.

Nanobody genes were codon-optimized and synthesized by Synbio as previously described8. All nanobody sequences were cloned into pET-21b(+) vectors using EcoRI and Hindlll restriction sites. Plasmids were transformed into BL21 (DE3) cells and plated onto Agar gel media with 50 μg/ml ampicillin. Agar plates were incubated at 37°C overnight, and single colonies were picked for protein purification. The cell culture was allowed to grow at 37°C to an OD600 of 0.5-0.6, at which point the temperature was lowered to 16°C and 0.51mM IPTG was added to induce protein expression overnight. Cells were then pelleted, resuspended in a lysis buffer (1x PBS, 150 mM NaCl, 0.2% Triton-X100, and protease inhibitors), and ultrasonicated on ice. The clarified cell lysate was collected by centrifugation at 15,000 x g for 10 mins. His-tagged nanobodies were captured using cobalt resin and eluted with a pre-chilled buffer containing imidazole (50 mM NaPO4, 300 mM NaCl, 150 mM Imidazole, pH 7.4). Nanobodies eluted from His-Cobalt resin were further purified using a Superdex 75 gel filtration column using filtered 1x PBS. Nanobodies were used fresh or flash frozen and stored at −80°C before use.

Cryo-EM sample preparation and imaging

For S complexed with Nb21, Nb34, and Nb95, each Nb was mixed with the S protein at a molecular ratio of 5 to 1 and incubated at 4 degrees for 30 minutes. The complex was then diluted in a buffer containing 20 mM Hepes pH 7.5 and 200 mM NaCl to reach a concentration of the S protein at 0.2 mg/ml. Then the sample was applied to a 1.2/1.3 UltrAuFoil grid (Electron Microscopy Sciences) that had been freshly glow-discharged and plunge-frozen in liquid ethane using an FEI Vitrobot Mark IV. All cryo-EM data were collected on Titan Krios transmission electron microscopes (Thermo Fisher) operating at 300 kV. For the S and Nb21 complex, images were acquired on a Falcon 3 detector, with a nominal magnification of 96,000, corresponding to a final pixel size of 0.83 Å/pixel. For each image stack, a total dose of about 62 electrons was equally fractionated into 70 fractions with ~0.88 e/Å2/fraction. EPU 2 Software was used to automate data collection. Defocus values used to collect the dataset ranged from −0.5 to −3.5 μm.

For the complexes of S with Nb95 and Nb34, sample preparation and data collection were similar to those for the complex of S with Nb21 except that the data were acquired on a Gatan K3-Summit detector. Further details of data collection parameters are summarized in Table S2 and S3.

For S complexes with Nb17 and 105, purified nanobodies were mixed with the SARS-CoV-2 S HexaPro trimer with 2:1 molar ratio nanobodies to a final concentration of 0.1 mg/mL S protein and incubated at room temperature for two hours. Cryo-EM grids (Quantifoil AU 1.2/1.3 300 mesh) were glow-discharged and coated with graphene oxide thin layer flakes following the protocol from reference43 (figshare. Media. The cryo-EM specimens were prepared using an FEI Vitrobot Mark IV with 3.5μl of freshly prepared nanobody:S complex. Grids were blotted for 3 s with blot force −5 in 100% humidity at 4°C prior to plunge freezing. The frozen-dehydrated grids were transferred to a Titan Krios (Thermo Fisher Scientific) transmission electron microscope equipped with a Gatan K3direct-electron counting camera and BioQuantum energy filter for data acquisition. Movies of the specimen were recorded with a nominal defocus setting in the range of −0.5 to −2.0 μm using SerialEM with beam-tilt image-shift data collection strategy with a 3 x 3 pattern and 1 shot per hole. The movie stacks were collected in the correlated double sampling (CDS) super-resolution mode of the K3 camera at a nominal magnification of 81,000 yielding a physical pixel size of 1.08 Å/pixel. Each stack was exposed for 5 s, with each frame exposed for 0.1 s, resulting in a 50-frame movie. For datasets without using CDS mode, the movie stacks were collected in the super-resolution mode at a nominal magnification of 81,000 with an exposure time of 2.5 s, and each frame exposed for 0.05 s. The total accumulated dose on the specimen was 40 e/Å2 for each stack.

For the trimeric Nb complexes (Nb105:RBD: Nb21, Nb17:RBD: Nb105 and Nb36:RBD: Nb21), two purified Nbs were mixed with purified RBD with 1.1:1.1:1 molar ratio and subsequently polished by size-exclusion chromatography (SEC). Peak fraction corresponding to the trimeric complexes was used for cryo-grid preparation. Movies of the specimen were recorded with a nominal defocus setting in the range of −0.5 to −2.5 μm using SerialEM with beam-tilt image-shift data collection strategy with a 3 x 3 pattern and 3 shot per hole. The movie stacks were collected in the correlated double sampling (CDS) super-resolution mode of the K3 camera at a nominal magnification of 165,000 yielding a physical pixel size of 0,52 Å/pixel. Each stack was exposed for 2.8 s, with each frame exposed for 0.1 s, resulting in a 28-frame movie. The total accumulated dose on the specimen was 108 e/Å2 for each stack.

Cryo-EM data processing

For the samples of S protein with Nb21, Nb34, and Nb95, the cryo-EM data processing was performed using Relion 3.1. Beam-induced motion correction was performed using the motion correction program implemented in Relion to generate average micrographs and dose-weighted micrographs from all frames. Contrast transfer function (CTF) parameters were estimated using CTFFIND4 from average micrographs. The loG-based auto-picking procedure was used for reference-free particle picking. Initial particle stacks were subjected to 2D classification and the best class averages that represented different views were selected as templates for second round automatic particle picking from the dose-weighted micrographs.

For the S protein with Nb21 data, approximately 900,000 particles were auto-picked from 2,574 micrographs for further processing. The whole set of particles was cleaned to remove contaminants or junk particles by 2D classification and 3D classification using 2x binned particles. Finally, approximately 135,000 particles were used for 3D auto-refinement with the structure of EMD-22221 (EMBD ID) low-pass filtered to 40Å as the reference. This yielded a map of ~3.4 Å resolution (corrected gold-standard FSC 0.143 criterion). The particles were re-extracted and used for further 3D classification into four classes. The most populated two classes, which contained 53% (1-up-down RBDs) and 29% (2-up-1-down RBDs) of particles were subjected to further 3D auto-refinement. To improve the local density of Nb21 and RBD, focused refinement was performed with a soft mask applied to one down RBD and Nb21, resulting in an improved local resolution ranging from 4.5 Å to 3.3 Å for the RBD and Nb21 interface. All maps were sharpened using the post-processing program in Relion or DeepEMhancer. The local resolution was estimated by ResMap in Relion. Similar approaches were used to solve the structures of S protein with Nb95 and Nb34 as well. The detailed information of data processing is shown in Figures S4 and S610 and Table S2-3.

For other structures, each movie stack was processed on-the-fly using CryoSPARC live (version 3.0.0) 44,45. The movie stacks were aligned using patch motion correction with an F-crop factor of 0.5. The contrast-transfer function (CTF) parameters of each particle were estimated using patch CTF. Particles were auto-picked using a 220 and 100 Å gaussian blob for Nb:S and 2Nbs: RBD complexes respectively. The numbers of bin2 particles selected after 2D classification are included in Table S2. The initial 3D volume and decoys were generated using ab initio reconstruction with a minibatch size of 1000 using a set of rebalanced 2D classes. The particles after 2D clean-up were submitted to one round of heterogeneous refinement with ab initio 3D volume from good 2D classes and decoy 3D volumes from bad 2D classes. Based on the coordinates and angular information of these particles, bin1 particles of the 3D class with well-resolved secondary structure features were re-extracted from the dose-weighted micrographs. For small trimeric complexes, a pixel size that can achieve the resolution limit of the sample, instead of bin1 pixel, was used for the final reconstruction to prevent overfitting. The final particle set was subjected to non-uniform 3D refinements45, followed by local 3D refinements, yielding final maps with reported global resolutions using the 0.143 criteria of the gold-standard Fourier shell correlation (FSC) (Table S2). The half maps were used to determine the local resolution of each map and focused classification was performed using Relion 3.0 46,47. For Nb17:S complex, the final particles (45,362) were aligned to the C3 symmetry axis to expand the particle set to 136,086 (Figure S9C). Then a mask focused on the arc shape including RBD, Nb17 and NTD was created with the binary map extended 10 pixels and a soft-edge of 10 pixels. The cryosparc particle set was converted to relion star file using pyem (, and focused classification was performed by Relion with k=3. The class with densities for all three targeted domains well resolved was selected for further local refinement in CryoSPARC.

Model building and structure refinement

For modeling whole S protein with Nbs, the RBD models were generated by docking the atomic model of SARS-Cov2 RBD (PDB ID 7JVB, chain B) into the refined cryo-EM density using Chimera (UCSF). Nb structures were modeled ab initio in Coot using based on the locally refined cryo-EM maps and refined in Phenix. After refinement, each residue of the sequence-updated models was manually checked and refined iteratively in Coot and Phenix. Structural models were validated by MolProbity. The final refinement statistics are listed in Table S2-3.

CDR3 loop modeling

To optimize the CDR3 loop conformation, it was modeled ab-initio using ‘RosettaAntibody3’ H3 loop modeling with restraints (distance, dihedral, and planar angles 48) generated by NanoNet. NanoNet is a deep residual neural network, similar to DeepH3 49, trained on solved CDR3 loops of antibodies and nanobodies from the PDB. NanoNet uniqueness comes from the fact that it takes as input only the sequence of the CDRs (each in a single one-hot encoding matrix) without the framework region. In addition, it uses MSE loss and predicts the pairwise distances and angles directly (for angles, it predicts the sine and cosine values to overcome cyclic loss), instead of using categorical cross-entropy loss and trying to predict the pairwise probability distributions. NanoNet architecture consists of two 2D residual blocks, followed by two convolutional layers for each output, with the tanh activation function for angles and ReLU (rectified linear activation function) for distances.

For each nanobody, 100 models were generated, and the one that fitted best in the cryo-EM density map was chosen manually. For Nb20, Nb21, Nb95, Nb105, Nb34 the models generated were similar to the ones without the optimization. For Nbs 17 and 36, the models generated from ‘RosettaAntibody3’ with NanoNet fitted better in the cryo-EM map and were further refined in the density map.

Contact heat map

An RBD residue and an Ab/Nb residue were defined in contact if the distance between any pair of their atoms was lower than a threshold of 6 Å. The Ab/Nb contact value of each RBD residue is calculated as the average of all the Ab/Nb contacts. Nb classes were clustered using k-means (k=3). The conservation score was obtained from Consurf by querying the RBD sequence.

Measurement of buried surface area(BSA)

The solvent-accessible surface area(SASA) of molecules was calculated by FreeSASA50. The buried surface area in the case of the Nb-RBD complex was then calculated by BSA = 1/2[SASA(Nb) + SASA(RBD) – SASA (complex)].

Measurement of structural overlap between Nb and corresponding best matched Fab

The best matched Fab for an Nb was obtained using the epitope similarity(Jaccard-index). The Nb-RBD complex structure was superimposed on its best matched Fab-RBD structure and protein volume was calculated using ProteinVolume51. Then the structural overlap was calculated by structural overlap = [Volume(Nb) + Volume(RBD) — Volume(complex)]/Volume(RBD).

Measurement of the interface curvature

The interface curvature was calculated as the average of the shape function of the interface atoms of the antigen or the Nb. For this purpose, a sphere of radius R (6Å) is placed at a surface point of the interface atom. The fraction of the sphere inside the solvent-excluded volume of the protein is the shape function at the corresponding atom 52.

ELISA (Enzyme-Linked Immunosorbent Assay)

Antigens (RBD and RBD variants) were coated onto 96-well ELISA plates, with 150 ng of protein per well in the coating buffer (15 mM sodium carbonate, 35mM Sodium Bicarbonate, pH 9.6) at 4°C for overnight. The plates were decanted, washed with a buffer (1x PBS, 0.05% Tween 20), and blocked for 2 hours at room temperature (1x PBS, 0.05% Tween 20, 5% milk powder). Nanobodies were serially 5x diluted in blocking buffers from 500 nM to 6.4 pM. Anti-T7 tag HRP-conjugated secondary antibodies were diluted at 1:5000 and incubated at room temperature for 1 hour. Upon washing, samples were further incubated in the dark for 10 minutes with freshly prepared 3,3’,5,5’-Tetramethylbenzidine (TMB) substrate. Upon quenching the reaction with a STOP solution, the plates were measured at wavelengths of 450 nm with background subtraction at 550 nm. The raw data was processed and fitted into the 4PL curve using the Prism Graphpad 9.0. IC50s were calculated and fold changes of binding affinity were calculated to generate the heatmap.

For ACE2 competitive ELISA assays, the super stable spike was coated at 80 ng/ml on the plate. Nanobodies were serially 5x diluted in blocking buffers from 500 nM to 32 pM with an addition of 60 ng/well of biotinylated hACE2 for competition. No Nb was used as a negative control. Pierce High Sensitivity Neutravidin-HRP antibodies were used at 1:8000. The hACE2 percentage was calculated by the reading at each Nb concentration divided by the reading at the negative control. Then the data processed and fitted into the 4PL curve using the Prism Graphpad 9.0.

Pseudovirus neutralization assay

The 293T-hsACE2 stable cell line and the pseudotyped SARS-CoV-2 particles (wild-type and mutants) with luciferase reporters were purchased from the Integral Molecular. The B.1.1.7 UK pseudotyped virus contains all of the naturally prevalent mutations for that strain. The SA 501Y.V2 contains all of the naturally prevalent mutations except del241-243, which is replaced by an L242H substitution for the pseudovirus (Figure S2). The neutralization assay was carried out according to the manufacturers’ protocols in duplicates. In brief, 3-fold or 5-fold serially diluted Nbs were incubated with the pseudotyped SARS-CoV-2-luciferase for 1 hour at 37 °C. At least eight concentrations were tested for each Nb. Pseudovirus in culture media without Nbs was used as a negative control. 100 μl of the mixtures were then incubated with 100 μl 293T-hsACE2 cells at 3×10e5 cells/ml in the 96-well plates. The infection took ~72 hours at 37 °C with 5% CO2. The luciferase signal was measured using the Renilla-Glo luciferase assay system with the luminometer at 1 ms integration time. The obtained relative luminescence signals (RLU) from the wells were normalized according to the negative control and the neutralization percentage was calculated at each concentration. The data was then processed by Prism GraphPad 9.0 to fit into a 4PL curve and to calculate the logIC50 (half-maximal inhibitory concentration).

Protein thermal shift assay

Thermal denaturation of S protein in the presence of an increased concentration of Nb36 was monitored by differential scanning fluorimetry using Protein Thermal Shift dye kit53. Briefly, the same protein samples used for negative stain EM were diluted to a final assay concentration of 100 nM in PBS with 1 mM DTT and 1:1000 fluorescence dye (TFS 4461146). The final assay volume was 20 μL, with 1, 5, 10, 100, and 600 nM of Nb36 was added to a final concentration of 100 nM S protein. Heat denaturation curves were recorded using a realtime PCR instrument (StepOne) applying a temperature gradient of 1 °C/min. Analysis of the data was performed using Excel. Melting temperatures of protein samples were determined by the inflection points of the plots of –d(RFU)/dT.

Negative-stain electron microscopy

For negative staining electron microscopy, 3 μl of specified concentration of Nb36 with the S protein was applied to a glow-discharged grid coated with carbon film. The sample was left on the carbon film for 60s, followed by negative staining with 2% uranyl formate. Electron microscopy micrographs were recorded on a Gatan Ultrascan CCD camera at 22,000 × magnification in an FEI Tecnai 12 electron microscope operated at 100 keV

Potent neutralizing nanobodies resist convergent circulating variants of SARS-CoV-2 by targeting novel and conserved epitopes

Dapeng SunZhe SangYong Joon KimYufei XiangTomer CohenAnna K. BelfordAlexis HuetJames F. ConwayJi SunDerek J. TaylorDina Schneidman-DuhovnyCheng ZhangWei HuangYi Shi


There is an urgent need to develop effective interventions resistant to the evolving variants of SARS-CoV-2. Nanobodies (Nbs) are stable and cost-effective agents that can be delivered by novel aerosolization route to treat SARS-CoV-2 infections efficiently. However, it remains unknown if they possess broadly neutralizing activities against the prevalent circulating strains. We found that potent neutralizing Nbs are highly resistant to the convergent variants of concern that evade a large panel of neutralizing antibodies (Abs) and significantly reduce the activities of convalescent or vaccine-elicited sera. Subsequent determination of 9 high-resolution structures involving 6 potent neutralizing Nbs by cryoelectron microscopy reveals conserved and novel epitopes on virus spike inaccessible to Abs. Systematic structural comparison of neutralizing Abs and Nbs provides critical insights into how Nbs uniquely target the spike to achieve high-affinity and broadly neutralizing activity against the evolving virus. Our study will inform the rational design of novel pan-coronavirus vaccines and therapeutics.


Since its emergence in late 2019, the Coronavirus Disease 2019 (COVID-19) pandemic and its causative agent SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2) have caused devastating consequences to global health and the economy. In response to this crisis, remarkable progress has been made by the scientific community and biopharmaceutical industry to develop innovative strategies to help curb this highly transmissible virus. In addition to vaccine development, an impressive number of neutralizing monoclonal antibodies (mAbs) have been isolated, mostly from the convalescent plasma, to facilitate a better understanding of the host immune response to SARS-CoV-21. Highly potent neutralizing mAbs, both singlets and in combinations, have been approved for emergency therapeutic use with more candidates in the pipeline 24.

Likewise, antibody fragments, especially camelid VHH single-domain antibodies, or nanobodies have also been successfully developed for virus neutralization 512. Compared to mAbs, nanobodies (Nbs) are substantially smaller (~ 15 kDa) yet can still bind virus antigens with excellent specificity. Because of the small sizes and structural robustness, they can be easily bioengineered into multivalent forms that bind different neutralizing epitopes, thus blocking the viral mutational escape58,13. In addition, affinity-matured Nbs are generally highly soluble, stable, and can be rapidly produced in microbes such as E.coli or yeast cells at low costs. Stable constructs can be delivered by small aerosolized particles and inhaled for direct and highly efficient treatment of pulmonary infections 7,8. Most recently, this novel inhalation therapy has been successfully evaluated in vivo for the treatment of SARS-CoV-2 infection. At ultra-low doses, aerosolization of an ultrapotent Nb construct drastically suppresses virus infection in both upper and lower respiratory tracts and prevents viral pneumonia 14. Potent neutralizing Nbs represent a convenient and highly cost-effective therapeutic option to help mitigate the evolving pandemic.

Similar to other coronaviruses, the infection of SARS-CoV-2 is mediated by the spike trimeric glycoprotein (S). Each S monomer is composed of two subunits: S1 and S2. The receptor-binding domain (RBD) of S1 is critical for interacting with the host receptor angiotensin-converting enzyme 2 (ACE2). In the prefusion state, the RBD is undergoing highly dynamic switching between closed (“down”) and open (“up”) conformations on the distal tip of the spike trimer 1517. In the post-fusion state, S1 shedding triggers a large conformational change of S2 to facilitate virus binding to the host membrane for infection 18.

The ACE2 receptor binding site (RBS) of RBD is the major target of serologic response in COVID-19 patients. As a direct adaptation against antibody pressure, however, RBS is also the primary region where a number of convergent mutations have arisen in circulating variants of SARS-CoV-2. These variants may enhance ACE2 binding leading to higher transmissibility, elude many neutralizing mAbs, including those under clinical development, and substantially reduce the neutralizing activities of convalescent and vaccine-elicited polyclonal sera 1921. These adaptations led to the global emergence of convergent circulating variants of concern, including the UK strain (B.1.1.7), the South Africa strain (B.1.351), and the Brazil strain (P.1) 2224. Notably, three RBS residue substitutions (K417N, E484K, and N501Y) derived from these clinical isolates have been demonstrated to drastically reduce, or abolish the binding of a large panel of neutralizing mAbs. Other examples include Y453F on the RBS (mink-human “spillover”) and non-RBS mutation N439K, which result in the immune escape from the convalescent sera 25. Long-term control of the pandemic will require the development of highly effective interventions that maintain neutralizing activities against the evolving strains 22.

Recently, we identified thousands of distinct, high-affinity antiviral Nbs that bind RBD and have determined a crystal structure of an ultrapotent one (Nb20) in complex with RBD 8,26. Here we assessed the impact of the convergent variants of concern and the critical RBD point mutations on the ultrapotent Nbs. Subsequent determination of 9 high-resolution structures, involving 6 Nbs bound to either S or RBD by cryo-EM provided critical insights into the antiviral mechanisms of highly potent neutralizing Nbs. Structural comparisons between neutralizing mAbs and Nbs revealed marked differences between the two antibody species.


Potent neutralizing Nbs are highly resistant to the convergent circulating variants of SARS-CoV-2 and a super RBD variant

We performed ELISA to evaluate how 6 critical RBD mutations impact the binding of 7 highly diverse and potent neutralizing Nbs that we have previously identified 8. Surprisingly, the neutralizing Nbs were largely unaffected by the mutations (Figure 1AS1). The only exception was E484K, which almost completely abolished the ultra-high affinity of Nbs 20 and 21. Additionally, we evaluated two circulating variants of global concern (B.1.1.7 UK and B. 1.351 SA) on Nb neutralization using a pseudotyped virus neutralization assay (Methods). These pseudoviruses fully recapitulate the maior mutations of the natural spike variants, including deletions and point mutations (Figure S2, Methods). The initial SARS-CoV-2 strain (Wuhan-Hu-1) was used as a control. Consistent with the ELISA results, we found that the UK strain (B.1.1.7), possessing a critical RBD mutation N501Y, has little if any effect on all the potent neutralizing Nbs that we have evaluated (Figure 1BS3). The SA strain (B.1.351), containing three RBD mutations (K417N, E484K, and N501Y), drastically reduces the efficacy of Nbs 20 and 21, but has a very marginal impact on the efficacies of other Nbs. The results contrast with recent investigations of a repertoire of neutralizing mAbs including those under clinical development, convalescent, and the vaccine-elicited polyclonal sera, which are significantly affected by at least one of these strains (Table S1).

Figure S1:ELISA curves of Nbs for RBD mutant binding (related to Figure 1).

Figure S2:Structure representations of SARS-CoV-2 spike trimer glycoprotein and mutations for two prevalent circulating strains (related to Figures 1 and 5).

Figure S3:Pseudovirus assay results for individual Nbs (related to Figure 1).

Figure 1.The impacts of RBD circulating variants on Nb binding

1A: ELISA binding of the RBD mutants (a summary heatmap). Data shown as fold change of binding affinity relative to that towards RBD WT.

1B: The fold change of neutralizing potencies of the Nbs against two dominant circulating variants (UK and SA strains) compared to wild-type SARS-CoV-2 pseudovirus particles.

To assess the potential to resist future mutations, highly neutralizing Nbs were also evaluated for their binding to a potential future RBD variant (RBD62 with 9 point mutations), which was identified by in vitro evolution for ultrahigh-affinity ACE2 binding 27. While several Nbs were substantially affected by RBD62, Nbs 34 and 105 retained their high affinity against this super variant.

These striking results prompted us to further investigate the structural basis for the broadly neutralizing activities of these Nbs. Our high-resolution cryo-EM maps of Nbs in complex with either the prefusion-stabilized S 28 or the RBD revealed three distinct classes of neutralizing Nbs that are affected differently by the variants and provided insights into the antiviral mechanisms.

Ultrapotent Class I Nbs and the “Achilles heel”

Class I dominates high-affinity RBD Nbs and represents some of the most potent neutralizers for SARS-CoV-2 (with the half-maximal inhibitory concentration or IC50 < 15 ng/ml). This class of Nbs is characterized by short CDR3 residues (typically 10 residues or less). Previously we have determined a crystal structure of a class I Nb (Nb20) with RBD. Nb20 can block ACE2 binding to the RBD at sub-nM concentration 8. Similarly, Nb21 can neutralize a clinical isolate (the Munich strain) of SARS-CoV-2 at sub-ng/ml, which is unprecedented for monomeric antibody fragments without antibody-like avidity binding 29. To understand the neutralization mechanism of class I Nbs better and in the context of the S trimer, we solved the structure of the most potent class I Nb (Nb21) bound to the S using cryo-EM.

Nb21 binds RBDs in both up and down conformations (Figure 2A). There are two major classes of the spike bound to Nb21: i) one up-RBD and two down-RBDs (resolution of 3.6 Å) and ii) two up-RBDs and one down-RBD (3.9 Å) (Figure 2AS4A-C). Due to the high flexibility of RBD, we performed local refinement of one down-RBD with Nb21 to resolve the binding interface (Figure S4C). Nb21 binds the extended external loop region of the RBD with two β-strands. The interactions are mediated by all three CDR loops (Figure 2B). R31 of Nb21 forms cation-π interactions with F490 of RBD. It also forms a polar interaction network with Y104 (Nb21) and E484 (RBD). These four residues are located at the center of the Nb21:RBD interface, constituting a major site of interactions (Figure 2C). In addition, side chains of R97, N52, and N55 (Nb21) form hydrogen bonds with the main chain carbonyl groups of L492 and Y449 and the side chain of T470 (RBD), respectively (Figure 2C). The main-chain carbonyl group of A29 (Nb21) also forms a hydrogen bond with Q493 (RBD). Besides these polar interactions, F45 and L59 of Nb21 and V483 of RBD form a cluster of hydrophobic interactions, together, providing ultrahigh-affinity and selectivity for RBD binding