The COVID-19 pandemic has prompted substantial efforts to develop effective countermeasures against SARS-CoV-2. Preclinical data and phase-III clinical studies indicate that monoclonal antibodies could be effectively deployed for prevention or treatment during the viral symptoms phase of the disease1,2. Cocktails of two or more monoclonal antibodies are preferred over a single antibody as these cocktails result in increased efficacy and the prevention of viral escape. However, this approach requires increased manufacturing costs and volumes, which are problematic at a time when the supply chain is under pressure to meet the high demand for COVID-19 therapeutic agents, vaccines and biologics in general4. Cocktails also complicate formulation5,6 and hinder strategies such as antibody delivery by viral vectors or by nonvectored nucleic acids7,8. One alternative is to use multispecific antibodies, which have the advantages of cocktails and single-molecule strategies.

To this end, we used structural information9 and computational simulations to design bispecific antibodies that would simultaneously bind to (i) independent sites on the same RBD and (ii) distinct RBDs on a spike (S) trimer. We evaluated several designs using atomistic molecular dynamics simulations, and produced four constructs: of these, CoV-X2 was the most potent neutralizer of SARS-CoV-2 pseudovirus, and had a half-maximal inhibitory concentration (IC50) of 0.04 nM (5.8 ng ml−1) (Extended Data Fig. 1). CoV-X2 is a human-derived IgG1-like bispecific antibody in the CrossMAb format10 that is the result of the combination of the Fragment antigen binding (Fab) of the monoclonal antibodies C121 and C135, which are two potent neutralizers of SARS-CoV-23. Structural predictions showed that CoV-X2—but not its parental monoclonal antibodies—can bind bivalently to all RBD conformations on the S trimer, which prevents the binding of ACE2 receptor11 (Fig. 1a, Extended Data Fig. 2).

Fig. 1: Biochemical and in vitro neutralizing properties of CoV-X2 are superior to those of its parental monoclonal antibodies.
figure 1

a, Computational simulations predict bivalent binding of CoV-X2 to all three RBDs on the S trimer (Extended Data Fig. 2). Green and blue, C121 and C135 moieties, respectively; yellow and orange, RBDs. b, c, Surface plasmon resonance (SPR) demonstrates that both arms of CoV-X2 are functional. In b, immobilized RBD complexed with C121 (left) or C135 (right) (first antibody) binds to CoV-X2 (second antibody). In c, RBD–CoV-X2 prevents binding by C121 (left) or C135 (right) single monoclonal antibodies. Paler colours denote controls (second antibody only). d, Both arms of CoV-X2 bind simultaneously to the RBD, as avidity is retained at decreasing RBD concentrations. Top and middle, representative SPR traces indicating the dissociations of Fab (left), IgG (centre) or CoV-X2 (right) binding to RBD immobilized at 150 nM (top) or 15 nM (middle) on the SPR chip (Extended Data Fig. 6). Bottom, plots of normalized ka (left) and kd (right) obtained with different concentrations of immobilized RBD. Increasing normalized kd values indicate loss of avidity. Solid lines, IgG; dotted lines, Fab. e, f, CoV-X2 prevents ACE2 binding to S trimer in an ELISA. ACE2 binding to antibody–S-trimer complexes measured with increasing concentration of indicated antibody and constant ACE2 (e), or at constant antibody concentration with increasing ACE2 (f). Mean of two replicates shown. g, CoV-X2 neutralizes SARS-CoV-2 pseudovirus and escape mutants generated by its parental monoclonal antibodies. Normalized relative luminescence (RLU) for cell lysates after infection with NanoLuc-expressing SARS-CoV-2 pseudovirus in the presence of increasing concentrations of antibodies. Wild-type SARS-CoV-2 pseudovirus (top left) is shown alongside an escape mutant generated in the presence of C135 (R346S; top right), or two escape mutants generated in the presence of C121 (E484G (bottom left) and Q493R (bottom right))14. Dashed lines, parental Fabs; solid lines, IgG. Mean of two independent experiments with two replicates each. h, Neutralization of SARS-CoV-2 isolates with sequences corresponding to wild-type virus (which was first isolated in China), and the B.1 (D614G), B.1.1.7, P.1 and B.1.351 variants. Mean of three experiments with s.d. RBD residues mutated in the variants are indicated in the table and as red spheres on the S trimer structure, on which the epitopes of C135 (blue) and C121 (green) are shown.

CoV-X2 bound at a low nanomolar affinity to the RBD and S trimer of wild-type SARS-CoV-2 and to those of several naturally occurring SARS-CoV-2 variants, including B.1 (which contains D614G, in the S), B.1.1.7 (which contains N501Y, in the RBD) and B.1.351 (which contains K417N/E484K/N501Y, in the RBD)12,13, as well as to escape mutants generated by the parental monoclonal antibodies14 (Extended Data Figs. 35).

CoV-X2 also bound to preformed C121–RBD and C135–RBD complexes, which confirmed that both of its arms are functional (Fig. 1b, c). Next, we used an avidity assay by surface plasmon resonance to experimentally confirm the computational prediction that CoV-X2 can simultaneously engage two sites on the same RBD (Fig. 1d, Extended Data Fig. 6, Methods). Avidity occurs when IgG binds bivalently to antigens, which results in slower dissociation rates (kd) (Extended Data Fig. 6a). Accordingly, C121 and C135 IgG showed avidity at high concentrations of antigen, owing to the intermolecular binding of adjacent RBDs; at lower concentrations of antigen, the kd was faster as intermolecular binding was prevented by the increased distance between RBD molecules, which resulted in loss of avidity. Intramolecular avidity is not possible for C121 and C135 because a single epitope is available on each RBD molecule. By contrast, CoV-X2 maintained avidity even at low concentrations of antigen, which indicates that it undergoes bivalent, intramolecular binding (Fig. 1d, Extended Data Fig. 6). We then performed enzyme-linked immunosorbent assays (ELISAs) to evaluate the ability of CoV-X2 to inhibit the binding of recombinant ACE2 to the S trimer (Fig. 1e, f). Consistent with the structural information9, C135 did not affect the interaction between ACE2 and S. C121, which occupies the ACE2-binding site on the RBD, prevented ACE2 binding only partially. By contrast, ACE2 binding was not detected in the presence of CoV-X2, which suggests a synergistic effect by the two moieties that comprise the bispecific antibody.

To assess the neutralizing ability of CoV-X2 in vitro, we first used SARS-CoV-2 pseudoviruses15. The bispecific antibody neutralized pseudovirus that carries wild-type SARS-CoV-2 S at sub-nanomolar concentrations (IC50 = 0.04 nM (5.8 ng ml−1) and 90% inhibitory concentration of 0.3 nM (44 ng ml−1)), which was similar to or better than the parental IgG and an over 100-fold-better IC50 than that of the parental Fabs (Fig. 1g). CoV-X2 remained effective against pseudoviruses that bear escape mutations that make them resistant to the individual monoclonal antibodies14 (Fig. 1g), and against a pseudovirus with RBD mutations that are found in the B.1.351 variant (which was first reported in South Africa) (IC50 = 1.3 nM (191 ng ml−1)) (Extended Data Fig. 5). To confirm the efficacy of CoV-X2, we performed plaque-reduction neutralization assays with infectious virus. CoV-X2 efficiently neutralized wild-type SARS-CoV-2 (IC50 = 0.9 nM); the B.1 variant that carries S(D614G), which was first detected in Europe (IC50 = 0.2 nM); the B.1.1.7 variant, which was first observed in the UK (IC50 = 0.2 nM); the P.1 variant, which was first isolated in Brazil (IC50 = 2.1 nM); and the B.1.351 variant (IC50 = 12 nM) (Fig. 1h). The P.1 and B.1.351 variants have almost identical mutations in the RBD; the only difference is the presence of an Asn (in B.1.351) versus Thr (in P.1) at position 417, which does not interact with CoV-X2. Nonetheless, the neutralization of B.1.351 was lower than that for P.1, which suggests that there are either some conformational differences in the RBD between the two variants or long-range effects that derive from additional mutations in the S. A similar behaviour is seen with the wild-type sequence (D614), which has a lower neutralization than S(D614G) even if no other difference is present; a plausible explanation is that a Gly at position 614 makes the CoV-X2 epitopes more accessible by favouring the ‘up’ conformation of the RBD16. We conclude that the in vitro binding and neutralizing properties of CoV-X2 make it preferable over its parental antibodies.

To assess the clinical potential of CoV-X2, we investigated its ability to protect against infection and disease in an animal model. We developed a mouse model in which human ACE2 (hACE2) is expressed by cells of the upper and lower respiratory tract after inhalation of a modified adeno-associated virus (AAV) (AAV-hACE2) (Fig. 2, Extended Data Fig. 7, Methods).

Fig. 2: CoV-X2 protects mice transduced with AAV-hACE2 against SARS-CoV-2 disease.
figure 2

a, Body weight over time in mice infected with SARS-CoV-2. We transduced 13–15-week-old C57BL/6NCrl wild-type female mice with AAV-hACE2 by forced inhalation, delivering viral particles to the upper and lower respiratory tract. After >7 days, mice were infected with SARS-CoV-2 (1 × 104 plaque-forming units (PFU)) or received vehicle by the intranasal route. Weight was monitored daily for 8 days (SARS-CoV-2, n = 5; control, n = 4). Mean with s.d. b, Kinetics of viral burden in lungs of mice infected with SARS-CoV-2 (n = 5) by plaque assays. Mean with s.d. Dashed line, limit of detection. c, Kinetics of viral RNA levels in lung samples from mice infected with SARS-CoV-2 (n = 5) by quantitative PCR with reverse transcription (RT–qPCR). Mean with s.d. d, Wild-type mice were transduced with AAV-hACE2 by forced inhalation. After >7 days, mice were inoculated intraperitoneally with antibodies one day before (black arrow) or 12 h after (red arrow) being infected intranasally with SARS-CoV-2 (1 × 104 PFU). e, f, Body weight was monitored daily in mice treated 24 h before infection (e, C121, n = 9; C135, n = 5; CoV-X2, n = 13; isotype control, n = 10) or 12 h after infection (f, CoV-X2, n = 4; control, n = 10). Mean with s.d. is shown. g, Lung viral burden by plaque assay at 2 dpi (isotype control, n = 8; CoV-X2, n = 5; C121, n = 5; C135, n = 5) and 5 dpi (isotype control, n = 6; CoV-X2, n = 10; C121, n = 5; C135, n = 5). The dashed line indicates the limit of detection; mean with s.d. P value was calculated with two-tailed Student’s t-test. h, Viral RNA levels in the spleen by RT–qPCR at 5 and 8 dpi (isotype control, n = 6 or 8 (indicated by dots); CoV-X2, n = 8 or 10 (indicated by dots)). Mean with s.d. P value, two-tailed Student’s t-test. i, Photographs of lungs from infected mice (8 dpi). j, Histopathology and F4/80 immunohistochemistry (IHC). Haematoxylin and eosin (H&E)-stained sections of paraffin-embedded lungs from infected mice (8 dpi). Arrowheads, foamy macrophages. F4/80 IHC shows abundant macrophage infiltration in lungs of mice treated with isotype control, but not those with CoV-X2. Each image representative of two separate experiments (n = 3–5 mice per group). Scale bars, 20 μm (inset in H&E), 50 μm (H&E right, F4/80 IHC), 100 μm (H&E middle), 1 mm (H&E, left).

Source data

This approach enables the rapid production of large cohorts of mice, and has the advantage of being applicable to wild-type and mutant mouse colonies, independently of age and sex. Moreover, because AAV vectors are only weakly immunogenic and cytotoxic, the system allows for prolonged expression of hACE217,18,19,20 (Extended Data Fig. 7). SARS-CoV-2 infection of ACE2-humanized mice (hereafter, hACE2 mice) results in progressive weight loss, respiratory pathology and disease that require culling at 8 days post-infection (dpi) (Fig. 2a–c, Extended Data Fig. 7).

To evaluate the protective effect of antibodies, we treated hACE2 mice one day before SARS-CoV-2 infection (150 μg) (Fig. 2d, e, g, h) or 12 h after challenge (250 μg) (Fig. 2d, f) and monitored them over time. Upon intranasal infection with 1 × 104 plaque-forming units of SARS-CoV-2 (SARS-CoV-2/human/Czech Republic/951/2020), mice treated with an isotype control (an antibody against the envelope protein of Zika virus) showed weight loss starting at 3 dpi; by 8 dpi, most mice had lost approximately 25–30% of their body weight, reaching the end point for humane culling (Fig. 2e, f). We recovered infectious virus from the lungs (Fig. 2g), and detected viral RNA in the spleen (Fig. 2h) but not in the heart (data not shown). The lung pathology of these mice resembled that associated with severe COVID-19 in humans21, and was characterized by diffuse alveolar damage (50–80% of tissue area), alveolar replacement with infiltrates of immune cells and fibroblasts, thickened septa and infiltrations by activated macrophages with foamy cytoplasm (Fig. 2j). By contrast, mice treated with CoV-X2 maintained their body weight (P < 0.0001 at 4–8 dpi, when compared to mice treated with isotype control) (Fig. 2e; the P values between all groups are given in Extended Data Table 1), had a reduced viral burden in lungs and viral RNA levels in spleen (Fig. 2g, h) and displayed neither macro- nor histopathological changes (diffuse alveolar damage of less than 5–10%) (Fig. 2i, j). Although infectious virus was detected at 2 dpi in all mice that were treated with C121 (n = 5), C135 (n = 5) or control (n = 8), in mice that received CoV-X2 it was detected only at low levels in one mouse (out of five) (Fig. 2g). At 5 dpi, we readily recovered infectious virus from control-treated mice (5 of 6 mice), but only from 1 out of 10 mice treated with CoV-X2 (Fig. 2g). Furthermore, CoV-X2 was also protective when administered at 12 h after SARS-CoV-2 challenge (Fig. 2d, f). As none of the mice treated with CoV-X2 exhibited symptoms at any time, we conclude that CoV-X2 protects mice from the disease.

Because monotherapy with C121 or C135 monoclonal antibodies leads to virus escape in vitro14, we treated hACE2 mice with the individual antibodies and sequenced the virus. Only wild-type RBD sequences were obtained from control mice (n = 10). In mice treated with C121, there was selection for a mutation that results in the E484D substitution (5 of 5 mice that were analysed at 8 dpi). Escape mutations at E484 generated by C121 have previously been observed in vitro14, and changes at this residue (which are also present in the B.1.351 and P.1 variants) reduce neutralization by human sera by more than tenfold22. The E484D substitution affects intermolecular hydrogen bonds at the core of the interface between C121 and the RBD, and it has previously been suggested to increase the affinity of the RBD for ACE223. Virus containing Asp at position 484 is pathogenic, as 7 out of 9 mice treated with C121 developed disease (Fig. 2e) and only this virus was found in their lungs. By contrast (and unlike the in vitro results14), we did not observe virus evasion or pathology in mice treated with C135 (n = 5) (Fig. 2e and data not shown). Even though we did not retrieve infectious virus (at 8 dpi) from mice treated with CoV-X2 (n = 13) and noticed no symptoms in these mice, we detected low levels of residual viral RNA in some mice after 40 cycles of PCR amplification: in 6 of 13 mice, the virus sequence was wild type, and in 2 mice the overlapping sequencing traces were consistent with the coexistence of wild-type virus and a variant with Asp at position 484. Thus, in these 2 (out of 13) mice that contained the E484D variant, CoV-X2 remained protective even if the mutation diluted the effective antibody concentration (presumably leaving only the C135 moiety active).

Monoclonal antibodies that target the SARS-CoV-2 S are in advanced clinical trials and show promise against COVID-191,2. Concomitant use of multiple antibodies is preferred for increased efficacy and adding resistance against viral evasion. Indeed, the virus can escape pressure by a single antibody in vitro and—as shown here—also in mice. Moreover, RBD mutations that threaten the efficacy of single monoclonal antibodies have already been detected in virus that is circulating in minks and humans, including mutations at the C121 and C135 epitopes (Extended Data Fig. 8). One disadvantage of antibody cocktails is the requirement for a development and production capacity twice or more that required for single monoclonal antibodies, which is a considerable challenge in light of the increased demand owing to COVID-19 related vaccines and therapeutic agents (on top of the need to maintain production of biologics for other diseases)4.

Multispecific antibodies offer the advantages of cocktails in a single molecule. We have shown that the CoV-X2 bispecific antibody is more effective than the related monoclonal antibodies at inhibiting ACE2 binding; it has a sub-nanomolar IC50 against a broader array of viral sequences and it protect mice from SARS-CoV-2 even when C121 (its potent parental monoclonal antibody) fails owing to viral escape. C135, the other parental monoclonal antibody, did not generate escape mutants in our mouse experiment, but readily generated them in vitro14. CoV-X2 is expected to be more resistant to viral escape as compared to monoclonal antibodies. We have shown that CoV-X2 binds and neutralizes mutants that are not recognized by its parental monoclonal antibodies, as well as variants of concern that have recently been detected in the UK12, South Africa13 and Brazil24.

Unlike other multispecific antibodies25, CoV-X2 is a fully human IgG-like molecule. As such, it is suitable for further development and could be engineered to alter effector functions. For example, the Fragment crystallizable (Fc) of CoV-X2 has already been modified by introducing LALA-PG mutations26 to modulate its interaction with Fc receptor and complement without affecting its antigen-binding properties. The LALA modification prevents antibody-dependent enhancement of flavivirus infection27 and may be a desirable modification in the context of SARS-CoV-2, as cellular and animal experiments with coronaviruses (including SARS-CoV28,29) support the possibility of antibody-dependent enhancement. Other modifications (such as the LS modification26 for increased half-life) are easily achievable. Finally, CoV-X2 is human-derived and produced in the CrossMab format, which has already shown to be safe in clinical trials30, which further supports the possibility of future development. IgG-like bispecific antibodies are therefore worth adding to the tools that we use to address SARS-CoV-2 and its future mutations.


No statistical methods were used to predetermine sample size. The experiments were not randomized, and investigators were not blinded to allocation during experiments and outcome assessment.

Protein expression and purification

To express SARS-CoV-2 S, a codon-optimized gene encoding residues 1–1208 (GenBank ( MN908947) with proline substitutions at residues 986 and 987, a ‘GSAS’ substitution at the furin cleavage site (residues 682–685), a C-terminal T4 fibritin trimerization motif and a C-terminal 8× HisTag was synthesized and cloned into the mammalian expression vector pcDNA3.1(+) by Genscript. Codon-optimized nucleotide sequences encoding SARS-CoV-2 RBD (residues 319–591; GenBank: MN908947) with a C-terminal 8× HisTag was obtained from Genscript. hACE2 fused to the Fc region of mouse IgG at the C terminus was also synthesized and cloned into the mammalian expression vector pcDNA3.1(+). All proteins were produced by transient PEI transfection in Expi293F cells (ThermoFisher), purified from the cell supernatants 6 days after transfection by HiTrap Chelating HP (Cytiva) or HiTrap Protein A HP (Cytiva), analysed by SDS–PAGE and dynamic light scattering on a DynaPro NanoStar (Wyatt Technology, software Dynamics v. RBD and S trimer mutations were introduced in the plasmids by gene synthesis (Genscript) and purified. All proteins underwent quality control and biophysical characterization to ensure functionality, stability, lack of aggregation and batch-to-batch reproducibility.

Monoclonal antibody production

Plasmids containing the coding sequence for the production of the monoclonal antibodies C121, C135, and C144 were prepared as previously described3. These monoclonal antibodies were produced by transient PEI transfection in Expi293F cells (ThermoFisher), purified from the cell supernatants 6 days after transfection by HiTrap Protein A HP (Cytiva) and HiLoad Superdex 200 16/60 column (Cytiva).

Design, expression and purification of bispecific antibodies

Bispecific antibodies in the single-chain Fv format (CoV-scB1 and CoV-scB2) were designed from the sequences of the variable regions of monoclonal antibodies C144 and C135 (CoV-scB1) or C121 and C1353 (CoV-scB2), following a previously described method31, N-terminal signal peptides (residues 1–19; UniProt ( P01743) and a C-terminal 6× HisTag were added. The constructs were synthesized and subcloned into the mammalian expression vector pcDNA3.1(+) by Genscript. The single-chain bispecific antibodies were produced by transient PEI transfection in Expi293F cells (ThermoFisher), purified from the cell supernatants 6 days after transfection by HiTrap Chelating HP (Cytiva) and HiLoad Superdex 75 16/60 column (Cytiva).

Bispecific antibodies in the CrossMAb format were designed from the sequences of the variable regions of antibodies C144 and C135 (CoV-X1) or C121 and C1353 (CoV-X2). Light and heavy chain constant region sequences (UniProt P01834 and UniProt P01857) were added. The CrossMAbs were designed, as previously described10, with CH1-CL crossover in the C135 moiety. Four constructs, one for each light chain (LC(C144), LC(C121) and LCCH1-CL(C135)) and heavy chain (HC(C144), HC(C121) and HCCH1-CL(C135)) were synthesized and subcloned into the mammalian expression vector pcDNA3.1(+) by Genscript. Signal peptides were included at the N terminus of the variable sequences for expression purposes (residues 1–19; UniProt P01743 for the heavy chains and residues 1–20; UniProt P06312 for the light chains). The antibodies were produced by PEI transient transfection, plasmid ratio 1:1:1:1, in Expi293F cells (ThermoFisher), purified from the cell supernatants 6 days after transfection by HiTrap Protein A HP (Cytiva), and HiLoad Superdex 200 16/60 column (Cytiva).

All antibodies underwent quality control and biophysical characterization to ensure functionality, stability, lack of aggregation and batch-to-batch reproducibility.

Computational modelling

CoV-scB1 and CoV-scB2 were modelled starting from the variable fragment of available experimental structures of the parental antibodies (Protein Data Bank (PDB) ( identifiers: 7K8X for C121, 7K8Z for C135 and 7K90 for C144); the connecting linkers were manually added (Pymol); the stability and feasibility of the bispecific constructs on the S trimer were manually and visually investigated according to structural biology considerations.

CoV-X1 and CoV-X2 were assembled starting from the experimental structures of individual Fab antibodies in complex with SARS-CoV-2 (PDB 7K8X for C121, 7K8Z for C135 and 7K90 for C144). The Fc moiety was manually placed in proximity to the C terminus of each Fab heavy chain; CH1 domains were then connected to the Fc using the ALMOST toolkit32, thus obtaining the full antibody structures.

The S trimer coordinates were obtained from available experimental structures; loop regions not present in the structure were modelled using the I-TASSER suite33. PDB 6VXX served as a basis for the ‘three down’ conformation; PDB 6VYB for the ‘one RBD up’ conformation; and PDB 7A93 for the ‘two RBD up’ conformation. Conformations not directly available (for example, trimer with ‘three RBD up’) were obtained by structural alignment and repetition of the appropriate conformation in the S monomer structures (for example, monomeric ‘RBD up’) using the PyMol software suite34

All the combinations of S trimer (three RBD down, one up, two up and three up) and antibodies (bispecific antibodies formed by combinations of C121, C135 and C144) were subjected to 400 ns or 200 ns of fully atomistic molecular dynamics simulations to obtain energetically favourable and stable conformations using GROMACS35. Calculations were performed on the CINECA-Marconi100 supercomputer.

The system was initially set up and equilibrated through standard molecular dynamics protocols: proteins were centred in a triclinic box, 0.2 nm from the edge, filled with SPCE water model and 0.15 M Na+Cl ions using the AMBER99SB-ILDN protein force field. Energy minimization was performed to let the ions achieve a stable conformation. Temperature and pressure equilibration steps, respectively, at 310 K and 1 bar of 100 ps each were completed before performing the full molecular dynamics simulations with the above-mentioned force field. Molecular dynamics trajectory files were analysed after removal of periodic boundary conditions. The stability of each simulated complex was verified by root mean square deviation and visual analysis.


The antibody binding properties were analysed at 25 °C on a Biacore 8K instrument (GE Healthcare) using 10 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA and 0.005% Tween-20 as running buffer.

SARS-CoV-2 RBD or full S, and their mutants, were immobilized on the surface of CM5 chips (Cytiva) through standard amine coupling. Increasing concentrations of antibody (3.12, 6.25, 12.5, 25 and 50 nM) were injected using a single-cycle kinetics setting; analyte responses were corrected for unspecific binding and buffer responses. Curve fitting and data analysis were performed with Biacore Insight Evaluation Software v.

Competition experiments were performed to assess the ability of COV-X2 to bind its target with both arms. A low amount of RBD (5 nM) was immobilized on the surface of a CM5 chip through standard amine coupling. C121 or C135 antibodies were injected at a high concentration (1.5 μM) to saturate the corresponding binding sites on the RBD; CoV-X2 (200 nM) was subsequently injected. The same experimental setting was performed with a different injection order as a control: COV-X2 (1.5 μM) injected to saturate RBD binding sites and subsequent injection of C121 or C135 (200 nM).

Analysis and comparison of kinetics parameters at different RBD concentrations were performed as following. RBD was immobilized on the surface of a CM5 chip at 5, 15, 75 and 150 nM. Increasing concentrations of antibody (3.12, 6.25, 12.5, 25 and 50 nM) were injected using a single-cycle kinetics setting; analyte responses were corrected for unspecific binding and buffer responses. Curve fitting and data analysis were performed with Biacore Insight Evaluation Software v.

Binding inhibition of hACE2

ELISAs were used to investigate the ability of antibodies to inhibit the binding of S to hACE2. Each experiment was performed in duplicate, reporting the mean of the two replicates; error bars represent the s.d. of the measured values. We used 96-well ELISA plates, coated at 4 °C with 37 nM S, and washed and blocked with PBS + 10% FCS. Antibodies were then added either at constant saturating concentration (160 nM) or at different dilutions (starting from 340 nM and serially diluted 1 to 3) and incubated 1 h at 25 °C; after washing, hACE2–mouse Fc was added either at constant saturating concentration (160 nM) or at different dilutions (starting from 340 nM and serially diluted 1 to 2) and left 1 h at 25 °C. After further washing, bound hACE2 was detected using standard protocols with goat anti-mouse IgG coupled to alkaline phosphatase (dilution 1:500, SouthernBiotech). ELISA plates were measured with the reader software Gen5 version 1.11.5 (BioTek Instruments). Data were analysed with Microsoft Excel 2016 and GraphPad Prism version 8.4.2.

SARS-CoV-2 pseudotyped reporter virus and pseudotyped-virus neutralization assay

A panel of plasmids expressing RBD-mutant SARS-CoV-2 S in the context of pSARS-CoV-2-SΔ19 have previously been described14,36. The mutant ‘KEN’ (K417N/E484K/N501Y) was constructed in the context of a pSARS-CoV-2-SΔ19 variant with a mutation in the furin cleavage site (R683G). The IC50 of these pseudotypes were compared to a wild-type SARS-CoV-2 S sequence carrying R683G. Generation of SARS-CoV-2 pseudotyped HIV-1 particles was performed as previously described3,15.

The neutralization activity of the bispecific and monoclonal antibodies was measured as previously reported3,15. In brief, fourfold serially diluted antibodies were incubated with SARS-CoV-2 pseudotyped virus for 1 h at 37 °C. The mixture was subsequently incubated with 293T cells expressing ACE2 for 48 h; the cells were washed twice with PBS and lysed with Luciferase Cell Culture Lysis 5× reagent (Promega). NanoLuc activity in lysates was measured using the Nano-Glo Luciferase Assay System (Promega) with the Modulus II Microplate Reader User interface (TURNER BioSystems). The obtained relative luminescence units (which are indicative of virus presence) were normalized to those derived from cells infected with SARS-CoV-2 pseudotyped virus in the absence of antibodies. The IC50 was determined using four-parameter nonlinear regression (GraphPad Prism).

SARS-CoV-2 virus-neutralization assay

The neutralizing activity of CoV-X2 against SARS-CoV-2 wild type, B.1 (D614G), B.1.1.7, P.1 and B.1.351 was investigated by plaque-reduction neutralization tests following a previously reported protocol37. In brief, 50 μl of antibody, starting from a concentration of 12 μg ml−1 or 190 μg ml−1 in a serial threefold dilution, were mixed in a flat-bottomed tissue culture microtitre plate (COSTAR) with an equal volume of 100 median tissue culture infectious dose of infectious virus that was isolated from patients with COVID-19, sequenced, titrated and incubated at 33 °C in 5% CO2. After 1 h, 3 × 104 (100 μl) Vero E6 cells (VERO C1008, Vero 76, clone 18 E6, Vero E6; ATCC CRL-1586) were added to each well. After 3 days of incubation, cells were stained with Gram’s crystal violet solution (Merck) plus 5% formaldehyde 40% m/v (Carlo Erba S.p.A.) for 30 min. Microtitre plates were then washed in water. Wells were analysed to evaluate the degree of cytopathic effect compared to untreated control. Each experiment was performed in triplicate. The IC50 was determined using three-parameter nonlinear regression (GraphPad Prism).

AAV-hACE2 vector design

Plasmid design and construction

The AAV transfer plasmid expressing hACE 2 (AAV-hACE2) was created by replacing the GFP sequence with ACE2 cDNA obtained from the HEK293 cell line (no. CRL-1573, ATCC, mycoplasma-free, population doubling lower than 13) in a pAAV-GFP plasmid (no. AAV-400, Cellbiolabs). HEK293 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (no. D5796, Sigma Aldrich) supplemented with 10% FBS (no. 10082139, Gibco) and 1% penicillin–streptomycin (no. XC-A4122100, BioSera). Cells were kept at 37 °C under 5% CO2 atmosphere and 95% humidity.

To obtain the hACE2 sequence, total RNA was isolated from a confluent 10-cm2 plate of HEK293 cells using RNeasy Mini Kit (Qiagen, no. 74104) according to the manufacturer’s protocol and reverse-transcribed with M-MLV Reverse Transcriptase (Promega, no. M1701). The generated cDNA was used as a template for PCR amplification with a pair of primers (ACE2 forward: 5′ATGTCAAGCTCTTCCTGG 3‘, ACE2 reverse: 5′CTAAAAGGAGGTCTGAACATC 3′) specific for hACE2 (NM_021804.3) using Phusion High-Fidelity DNA Polymerase (NEB, no. M0530S). The PCR product was separated in a 1% agarose gel (SeaKem LE AGAROSE; East Port, no. 50004); the band of appropriate size (2,418 bp) was extracted using NucleoSpin Gel and PCR Clean-up (Takara, no. 740609) according to the manufacturer protocol. The extracted product was treated with DreamTaq Green DNA Polymerase (Thermo Scientific, no. EP0711) in the presence of dATP to add 3′A overhangs to the PCR product. The product was then subcloned into pGEM-T Easy Vector (Promega, no. A1360). Proper insertion of the product was assessed by HindIII (Thermo Scientific, no. ER0501) and SacI (Thermo Scientific, no. ER1132) double-digestion control. Kozak sequence and SpeI recognition site were added at the 5′ and 3′ end of the ACE2 cDNA PCR product, respectively, by amplification with a specific pair of primers (hACE2_Kozak_Fw: 5′-CAGGGGACGATGTCAAGCTCTTCCTGG-3′, hACE2_Spe_Rv: 5′-ACTAGTGATCTAAAAGGAGGT-3′) using Phusion High-Fidelity DNA Polymerase (NEB, no. M0530S). The amplified product was separated in 1% agarose gel (SeaKem LE AGAROSE; East Port, no. 50004),extracted with NucleoSpin Gel and PCR Clean-up (Takara, no. 740609) according to the manufacturer’s protocol. The extracted ACE2 sequence was PCR-amplified with a pair of primers with microhomology arms (hACE2_IF_Fw: 5′-TTCGAACATCGATTGCAGGGGACGATGTCAAG-3′, hACE2_Spe_IF_Rv: 5′-GCGCTGCTCGAGGCAACTAGTGATCTAAAAGGAGGT-3′) and Phusion High-Fidelity DNA Polymerase (NEB, no. M0530S), and was subsequently purified from agarose gel with NucleoSpin Gel and PCR Clean-up (Takara, no. 740609).

The GFP sequence in the pAAV-GFP control vector (no. AAV-400, Cellbiolabs) was excised by double digestion using EcoRI (Thermo Scientific, no. ER0271)/HindIII (Thermo Scientific, no. ER0501) restriction enzymes and replaced with the hACE2 sequence flanked by microhomology arms using the In-Fusion cloning system (In-Fusion HD Cloning Kit, Takara Bio Europe, no. 639648) according to the manufacturer’s protocol. Proper insertion and presence of the SpeI recognition site were confirmed by double digestion using HindIII (Thermo Scientific, no. ER0501)/MluI (Thermo Scientific, no. ER0562) and SpeI (Thermo Scientific, no. ER1252), respectively. The generated AAV-hACE2 vector was sequenced (Eurofins Genomics) using the following primers: CMV_Fw: 5′-AAATGGGCGGTAGGCGTG-3′, seq_hACE2_start1: 5′-TGGAGATCTGAGGTCGG-3′, seq_hACE2_start2: 5′-TCTTCCTCCCACAGCTCCT-3′, seq_hACE2_1: 5′-CAGTTGATTGAAGATGTGGA-3′, seq_hACE2_2: 5′-AGAAGTGGAGGTGGATG-3′, seq_hACE2_3: 5′-AGAACTGAAGTTGAAAAGG-3′. The produced hACE2 sequence was 100% identical to the reference hACE2 sequence (NM_021804.3).

Transfection of Neuro-2a cells

Neuro-2a cells (CCL-131, ATCC, mycoplasma-free, population doubling lower than 10) used for validation of AAV-hACE2 function were cultured in DMEM (no. D5796, Sigma Aldrich) supplemented with 10% FBS (no. 10082139, Gibco) and 1% penicillin–streptomycin (no. XC-A4122100, BioSera) and kept at 37 °C under 5% CO2 atmosphere and 95% humidity. Three 60-mm2 plates were seeded each with 4 × 105 cells and transfected the next day with the AAV-hACE2 vector using Lipofectamine 2000 Transfection Reagent (Invitrogen, no. 11668027) according to the manufacturer’s protocol for the 6-well plate transfection. Transfection with pAAV-GFP control vector (no. AAV-400, Cellbiolabs) and non-transfected HEK293 cells were used as negative controls. After 48 h, cells were collected for western blotting in RIPA buffer, supplemented with AEBSF protease inhibitor (AppliChem, no. A1421) and cOmplete Mini Protease Inhibitor Cocktail (Sigma Aldrich, no. 04693124001). Expression of hACE2 in Neuro-2a cells was compared to the non-transfected HEK293 cells.

AAV-hACE2 particle production

AAV293 cell transfection

The AAV293 cell line (Cellbiolabs, no. AAV-100, mycoplasma-free, population doubling lower than 8) used for AAV production was cultured in DMEM (no. D5796, Sigma Aldrich) supplemented with 10% FBS (no. 10082139, Gibco), 1% penicillin–streptomycin (no. XC-A4122100, BioSera) and 1% NEAA (no. M7145, Sigma Aldrich). Cells were kept at 37 °C under 5% CO2 atmosphere and 95% humidity. A day before transfection, 6.5 × 106 cells were seeded on 15-cm2 cultivation plates to reach 80–90% confluency on the day of transfection. Vectors pHelper (no. 340202, Cell BioLabs), AAV Rep/Cap 2/9n (no. 112865, Addgene) and AAV-hACE2 were used for transfection in equimolar ratio. The total amount of DNA (28 μg per plate) diluted in 1 ml per plate of DMEM was mixed with linear polyethylenimine hydrochloride, M.W. 40000 PEI (no. 24765-1, Polysciences) in 1:2.7 ratio. After 20 min of incubation at room temperature, the transfection mixture was added to a cultivation plate with FBS-reduced medium (DMEM supplemented with 1% FBS) in a dropwise manner. After 5 h of incubation at 37 °C under 5% CO2 atmosphere, the medium was removed and replaced with fresh complete growth medium (DMEM supplemented with 10% FBS and 1% penicillin–streptomycin).

AAV293–hACE2 collection

Three days after transfection, both cell medium and cells were collected for AAV particle isolation. These procedures were adapted and modified from a previous publication38. The medium was collected into 50-ml centrifuge tubes and cells were washed twice with 5 ml of PBS. Subsequently, cells were scraped in 1 ml of PBS and centrifuged at 1,000g for 10 min at 4 °C. The supernatant was added to the previously collected cell medium; the cell pellet was kept on ice during subsequent processing. The medium was centrifuged at 3,200g for 15 min at 4 °C and the supernatant was then filtered into a sterile glass bottle using 0.22-μm PES membranes. PEG-8000 (no. V3011, Promega) was added to the medium in a glass bottle in a 1:4 ratio. The mixture was stirred slowly at 4 °C for 1 h and then incubated overnight at 4 °C without stirring to allow full virus precipitation. The following day, the medium was centrifuged at 2,800g for 20 min at 4 °C and the pellet resuspended in 10 ml of PBS solution with 0.001% Pluronic F-68 non-ionic surfactant (no. 24040032, Gibco) and 200 mM NaCl (no. S5886, Sigma Aldrich) and sonicated at 50% amplitude with 4× 1-s on/15-min off pulses on ice. The cell lysate was centrifuged at 3,200g for 15 min at 4 °C. Subsequently, 50 U ml−1 of benzonase nuclease (no. E1014-25KU, Sigma Aldrich) was added to the viral suspension to degrade any residual DNA. After incubation for 1 h at 37 °C, the viral suspension was centrifuged at 2,400g for 10 min at 4 °C and the clarified supernatant was further purified.

AAV-hACE2 purification by iodixanol gradient ultracentrifugation

A gradient consisting of 15% iodixanol (in 1 M NaCl, 2.7 mM MgCl2 2 mM KCl in phosphate buffer), 25% iodixanol (in 2.7 mM MgCl2, 2 mM KCl, and 0.001% phenol red in PBS), 40% iodixanol (in 2.7 mM MgCl2 and 2 mM KCl in PBS) and 0,002% phenol red (no. P3532, Sigma Aldrich) in 60% iodixanol (OptiPrep Density Gradient Medium, no. D1556, Sigma Aldrich) was prepared in QuickSeal tubes according a previous publication38. Five ml of clarified viral supernatant was carefully added on the top of the gradient and the rest of the tube was filled up with PBS. Ultracentrifugation was carried out at 350,000g for 90 min in a pre-cooled T70i rotor at 10 °C. After ultracentrifugation, approximately 750-μl fractions were collected from the 40% iodixanol phase using an 18G needle puncturing the QuickSeal tube at the interface of the 60% and 40% iodixanol.

AAV-hACE2 purity validation and buffer exchange

The purity of the collected fractions from 40% iodixanol containing AAV-hACE2 particles was assessed by SDS–PAGE. Ten μl of each collected fraction was mixed with 3.5 μl of 4× Laemmli buffer and loaded to 4–20% Mini-PROTEAN TGX Precast Protein Gel (no. 4561096, Bio-Rad). The gels were briefly washed in dH2O and stained with silver according to the manufacturer’s protocol (Pierce Silver Stain Kit, no. 10096113, Thermo Scientific).

Selected AAV-hACE2 fractions were pooled and concentrated using Amicon Ultra-0.5 Centrifugal Filter Unit (molecular weight cut-off of 100 kDa). First, Amicon filter membranes were activated by incubation with 0.1% pluronic F-68, 0.01% pluronic F-68 and 200 mM NaCl, followed by 0.001% of pluronic F-68 in PBS and centrifugation at 1,900g for 5 min at 4 °C. Pooled fractions with AAV-hACE2 particles were loaded onto activated Amicon filter membranes and centrifuged at 2,600g for 5 min at 4 °C. The membranes were washed several times with 0.001% pluronic F-68 in PBS (centrifugation at 2,600g for 8 min at 4 °C) until the residual iodixanol was completely removed from the solution. To elute and concentrate the viral suspension, the membranes were covered with about 5 ml of formulation buffer and incubated for 5 min at room temperature. Amicon filters were centrifuged at 2,600g at 4 °C for approximately 1.5 min until around 0.5 ml of the formulation buffer with AAV-hACE2 particles was left. The eluate was transferred into sterile 1.5-ml tubes, quantified and stored at 4 °C for up to 2 weeks for short-term in vivo application or at –80 °C for long-term storage.

AAV-hACE2 titration by qPCR

The protocol for quantification and determination of the number of genome-containing particles of AAV-hACE2 was adapted from a previous publication39, using qPCR. Purified AAV-hACE2 particles were treated with DNase I (no. EN0521, Thermo Scientific) to eliminate contaminating plasmid DNA. Serial dilutions of a AAV-hACE2 viral suspension were used as template in two separate reactions, one detecting viral inverted terminal repeat (ITR) sequences (ITR_Fw: 5′-GGAACCCCTAGTGATGGAGTT-3′, ITR_Rv: 5′-CGGCCTCAGTGAGCGA-3′) and the second hACE2 (hACE2_Fw: 5′-CCATTGGTCTTCTGTCACCCG-3′, hACE2_Rv: 5′-AGACCATCCACCTCCACTTCTC-3′). Data analysis was performed using the LightCycler 480 Software, version 1.5. AAV concentration (the number of viral genomes in 1 μl of AAV sample) was determined by comparison to standard curves of defined concentrations of AAV-hACE2 vector. Each qPCR run was performed in triplicate; six serial dilutions of the AAV-hACE2 vector were used as positive controls and standards

Mouse experiments

This study was carried out in strict accordance with the Czech national laws and guidelines on the use of experimental animals and protection of animals against cruelty (Animal Welfare Act No. 246/1992 Coll.). The protocol was approved by the Committee on the Ethics of Animal Experiments of the Institute of Parasitology, Institute of Molecular Genetics of the Czech Academy of Sciences, and of the Departmental Expert Committee for the Approval of Projects of Experiments on Animals of the Academy of Sciences of the Czech Republic (permits 82/2020 and 101/2020).

Application of AAV-hACE2 viral particles to mice

Thirteen-to-fifteen-week-old C57BL/6NCrl female mice were anaesthetized by intraperitoneal injection of ketamine and xylazine (0.1 mg per g body weight) (Biopharm) and 0.01 mg per g (Bioveta), respectively). Viral particles containing AAV-hACE2 were diluted to a final concentration of 4 × 109 genome copies in 40 μl of PBS. This volume was applied to mice by forced inhalation. The tip of the nose was gently clipped with tweezers and the tongue gently pulled out. After the mouse started breathing through the oral cavity, 40 μl of viral suspension were applied by a 200-μl pipette tip into the oral cavity and inhaled by the mouse through the trachea into the lungs.

To access whether hACE2 was expressed in lungs, lung tissue was collected and analysed by western blot in RIPA buffer, supplemented with AEBSF protease inhibitor (AppliChem, no. A1421) and cOmplete Mini Protease Inhibitor Cocktail (Sigma Aldrich, no. 04693124001) 1, 2 and 4 weeks after application. Expression of hACE2 in AAV-hACE2 transduced mice was compared to nontreated C57BL/6NCrl mice. Histone H3 antibody (cat. no.: ab1791, Abcam, lot: GR3237685-2; 1:1,000 dilution) was used as a loading control.

Mouse infection

SARS-CoV-2 (strain SARS-CoV-2/human/Czech Republic/951/2020, isolated from a clinical sample at the National Institute of Public Health, Prague), provided by J. Weber, was used for mouse infection. The virus was passaged in Vero E6 cells five times before its use in this study.

At least 7 days after application of the AAV-hACE2 virus particles, mice were infected intranasally with SARS-CoV-2 (1 × 104 plaque-forming units) in a total volume of 50 μl DMEM. Mice were monitored and weighted daily over an eight-day period. Treated mice were injected intraperitoneally with either 150 μg of antibodies 24 h before the infection or 250 μg 12 h after infection. Mice were killed at the indicated times after infection and their tissues collected for analysis.

Measurement of the viral burden

Tissues were weighed and homogenized using Precellys 24 (Bertin Technologies) and prepared as 20% (w/v) suspension in DMEM containing 10% newborn calf serum. The homogenates were clarified by centrifugation at 5,000g and the supernatant medium was used for plaque assay and viral RNA isolation.

Plaque assays were performed in Vero E6 cells (ATCC CRL-1586; mycoplasma-free) grown at 37 °C and 5% CO2 in DMEM (no. LM-D1112/500, Biosera) supplemented with 10% FBS (no. FB-1001G/500; Biosera), and 100 U ml−1 penicillin, 100 μg ml−1 streptomycin (Antibiotic Antimycotic Solution; no. A5955; Sigma), and 1% l-glutamine (no. XC-T1755/100; Biosera) using a modified version of a previously published protocol40. In brief, serial dilutions of virus were prepared in 24-well tissue culture plates and cells were added to each well (0.6 × 105–1.5 × 105 cells per well). After 4 h, the suspension was overlaid with 1.5 % (w/v) carboxymethylcellulose (no. C4888; Sigma) in DMEM. Following a 5-day incubation at 37 °C and 5% CO2, plates were washed with phosphate-buffered saline and the cell monolayers were stained with naphthol blue black (no. 195243; Sigma). The virus titre was expressed as plaque-forming units per ml.

RNA was isolated from tissue homogenates using the QIAmp Viral RNA mini kit (no. 52906; Qiagen) following manufacturer’s instructions.

Viral RNA was quantified using EliGene COVID19 Basic a RT (no. 90077-RT-A; Elisabeth Pharmacon) according to the manufacturer’s protocol. A calibration curve was constructed from four dilutions of a sample that was quantified using Quanty COVID-19 kit (no. RT-25; Clonit), according to the recommendations from the manufacturer. All real-time PCR reactions were performed using a LightCycler 480 (Roche).

For sequencing, isolated RNA was used as a template for one-step RT–PCR (Qiagen OneStep RT–PCR Kit; no. 210212; Qiagen) with primers specific for SARS-CoV-2 RBD sequence (SARS-CoV-2_seq_FW: 5′-GCACTTGACCCTCTCTCAGAAAC-3′; SARS-CoV-2_seq_RV: 5′-GACTCAGTAAGAACACCTGTGCC-3′). The reaction mixture (final volume 25 μl) contained 5 μl of QIAGEN OneStep RT–PCR buffer, 1 μl of dNTP mix, 5 μl of 5× Q-solution, 1 μl of QIAGEN OneStep RT–PCR enzyme mix, 6 μl of RNase-free water, 1.5 μl of each primer (stock concentration, 0.01 mM) and 4 μl of template RNA. The cycling conditions were as follows: reverse transcription (30 min at 50 °C), initial PCR activation (15 min at 95 °C), 3-step cycling: 40 cycles of 94 °C for 30 s, 52.6 °C for 30 s, and 72 °C for 1 min, followed by final extension (10 min at 72 °C). The PCR products were visualized in a 1.7% agarose gel in Tris-acetate-EDTA buffer. The amplified DNA was purified by using Wizard SV Gel and PCR Clean-Up System (no. A9285; Promega), according to the recommendations of the manufacturer.

The purified DNA was prepared for sequencing (Sanger method) by a commercial service (Eurofins Genomics) with the following conditions (final volume 17 μl): 15 μl of PCR product with a concentration of 5 ng μl−1 and 2 μl of primer with a concentration of 10 μM. The sequencing data were analysed using BioEdit Sequence Alignment Editor, version 7.2.0.

Histology and immunohistochemistry

Lungs were fixed in 4% PFA. Tissues were processed using a Leica ASP6025 automatic vacuum tissue processor and embedded in paraffin using a Leica EG1150 H+C embedding station. Two-μm sections were prepared using a Leica RM2255 rotary microtome and sections were stained with H&E using Leica ST5020 + CV5030 stainer and coverslipper.

To assess the presence of macrophages, a rabbit anti-mouse F4/80 monoclonal antibody (D2S9R XPR rabbit monoclonal antibody, cat. no. 70076, Cell Signaling Technology, lot 5, RRID AB_2799771) was used at 1:800 dilution as primary antibody. The histological sections (thickness 4–5 μm) were deparaffinized in a Multistainer Leica ST5020 (Leica Biosystems). Antigens were retrieved by heating the slides in citrate buffer pH 6 (Zytomed Systems). Endogenous peroxidase was neutralized with 3% H2O2. Sections were incubated for 1 h at room temperature with a 1:800 dilution of the primary antibody. After washing they were incubated with anti-rabbit secondary antibody conjugated with HRP (Zytomed, cat. no. ZUC 032-100, lot A0880-4; no dilution). Staining of the sections was developed with a diaminobenzidine substrate kit (DAKO, Agilent) and sections were counterstained with Harris haematoxylin (Sigma Aldrich, Merck) in a Multistainer Leica.

Biosafety statement

All work with infectious SARS-CoV-2 was performed in biosafety level 3 facilities at the Institute of Parasitology, Biology Centre of the Czech Academy of Science (Ceske Budejovice) and Veterinary Research Institute (Brno) using appropriate powered air-purifying, positive pressure respirators and protective equipment.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.