SARS-CoV-2 neutralizing human recombinant antibodies selected from pre-pandemic healthy donors binding at RBD-ACE2 interface

COVID-19 is a severe acute respiratory disease caused by SARS-CoV-2, a new recently emerged sarbecovirus. This virus uses the human ACE2 enzyme as receptor for cell entry, recognizing it with the receptor binding domain (RBD) of the S1 subunit of the viral spike protein. We present the use of phage display to select anti-SARS-CoV-2 spike antibodies from the human naïve antibody gene libraries HAL9/10 and subsequent identification of 309 unique fully human antibodies against S1. 17 antibodies are binding to the RBD, showing inhibition of spike binding to cells expressing ACE2 as scFv-Fc and neutralize active SARS-CoV-2 virus infection of VeroE6 cells. The antibody STE73-2E9 is showing neutralization of active SARS-CoV-2 as IgG and is binding to the ACE2-RBD interface. Thus, universal libraries from healthy human donors offer the advantage that antibodies can be generated quickly and independent from the availability of material from recovering patients in a pandemic situation.

I n 2015, Menachery et al. presciently wrote: "Our work suggests a potential risk of SARS-CoV re-emergence from viruses currently circulating in bat populations" 1 . Four years later, a novel coronavirus causing a severe pneumonia is causing a worldwide pandemic and has been named severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). The outbreak was initially noticed in a sea food market in Wuhan, Hubei province (China) at the end of 2019. The disease has been named COVID-19 (coronavirus disease 2019) by the World Health Organization. Genome sequencing shows high identity to bat coronaviruses (CoV, in particular RaTG13), beta-CoV virus causing human diseases like SARS and Middle East respiratory syndrome (MERS), and, to a lesser extent, the seasonal CoV hCoV-OC43 and HCov-HKU1 2,3 . The spike (S) protein of SARS-CoV-2, as well as SARS-CoV, binds to the human zinc peptidase angiotensin-converting enzyme 2 (ACE2), which is expressed on numerous cells, including lung, heart, kidney, and intestine cells, thus initiating virus entry into target cells. S protein consists of the N-terminal S1 subunit, which includes the receptor-binding domain (RBD), and the C-terminal S2 subunit, which is anchored to the viral membrane and is required for trimerization of spike itself and fusion of the virus and host membrane [4][5][6] . The host enzyme furin cleaves the S protein between S1 and S2 during viral formation and the membrane-bound host protease TMPRSS2 is responsible for the proteolytic activation of the S2' site, which is necessary for conformational changes and viral entry [7][8][9][10] .
Antibodies against the spike protein of coronaviruses are potential candidates for therapeutic development 11 . Antibodies against the S1 subunit, especially against RBD, can potently neutralize SARS-CoV and MERS [12][13][14] . Monoclonal human antibodies against SARS-CoV are described to cross-react with SARS-CoV-2; some of them are also able to neutralize SARS-CoV-2 15,16 . Other reports show how monoclonal antibodies against SARS-CoV-2 can be selected by rescreening memory B cells from a SARS patient 17 , selected from COVID-19 patients by single B cell PCR 18,19 or using phage display 20,21 . Human recombinant antibodies are successfully used for the treatment of other viral diseases. The antibody mAb114 22 and the three antibody cocktail REGN-EB3 23 showed a good efficiency in clinical trials against Ebola virus 24 . The antibody palivizumab is European Medicines Agency (EMA)/Food and Drug Administration (FDA) approved for treatment of a severe respiratory infection of infants caused by the respiratory syncytial virus (RSV) 25,26 and can be used as a guideline to develop therapeutic antibodies against SARS-CoV-2.
Antibody phage display is a powerful tool to generate human antibodies against infectious diseases 27 . We successfully used this technology to develop in vivo protective antibodies against Venezuelan encephalitis virus 28 , Western-equine encephalitis 29,30 , Marburg virus 31 , and Ebola Sudan virus 32 .
In this work, we show the generation of human recombinant antibodies against the spike proteins of SARS-CoV-2 from a universal, human naïve antibody gene library that was constructed before the emergence of SARS-CoV-2. Several selected scFv-Fc antibodies efficiently inhibit the binding of the spike protein to ACE2-expressing cells and are blocking SARS-CoV-2 infection of VeroE6 cells. The best antibody in the IgG format is a potential candidate for the clinical development of a passive immunotherapy for therapeutic and prophylactic purposes.
Results SARS CoV2 spike domains or subunits and human ACE2 were produced in insect cells and mammalian cells. SARS-CoV-2 RBD-SD1 (aa319-591) according to Wrapp et al. 33 , S1 subunit (aa14-694), S1-S2 (aa14-1208, with proline substitutions at position 986 and 987 and "GSAS" substitution at the furin site, residues 682-685), and extracellular domain of ACE2 receptor were produced in insect cells using a plasmid-based baculovirusfree system 34 as well as in Expi293F cells. All antigens with exception of S1-S2 were produced with human IgG1 Fc part, murine IgG2a Fc part, or with 6xHis tag in both expression systems. S1-S2 was only produced with 6xHis tag. The extracellular domain of ACE2 was produced with human IgG1 Fc part or mouse IgG2a in Expi293F cells and 6xHis tagged in insect cells. The yields of all the produced proteins are given in Table 1. A graphical overview of all the produced proteins is given in Supplementary Fig. 1. The expressed proteins were analyzed by size exclusion chromatography (SEC; Supplementary Fig. 2). S1 as well as S1-S2 were more efficiently produced in insect cells compared to Expi293F cells. RBD-SD1 was produced well in both production systems. The binding of the produced spike domains/proteins to ACE2 was validated by enzyme-linked immunosorbent assay (ELISA) and flow cytometric analysis on ACE2-positive cells (Table 1).
Antibodies were selected by phage display. Antibodies were selected against SARS-CoV-2 spike S1 subunit in four panning rounds in microtiter plates (MTPs). The following single-clone screening was performed by antigen ELISA in 96-well MTPs, using soluble monoclonal scFv produced in Escherichia coli. Subsequently, DNA encoding for the binders was sequenced and unique antibodies were recloned as scFv-Fc fusions.
In detail, three panning strategies were compared. In a first approach (STE70), the lambda (HAL9) and kappa (HAL10) libraries were combined and the antigen S1-hFc (with furin site, produced in High Five cells) was immobilized in phosphatebuffered saline (PBS). Here only seven unique antibodies were identified. In a second approach, the selection was performed separately for HAL10 (STE72) and HAL9 (STE73) using S1-hFc as antigen (with furin site, SEC purified, immobilized in carbonate buffer). Here 90 unique antibodies were selected from HAL10 and 209 from HAL9. In a third approach (STE77 and STE78), S1-hFc produced in Expi293F cells was used (immobilized in carbonate buffer). Here the panning resulted in only three unique antibodies that were not further analyzed in inhibition assays. An overview is given in Table 2.
The antibody subfamily distribution was analyzed and compared to the subfamily distribution in the HAL9/10 library and in vivo (Fig. 1). The phage display-selected antibodies mostly originated from the main gene families VH1 and VH3. Only few antibodies were found using VH4. In 96 of the 309 selected antibodies (31%), the V-gene VH3-23 was used. The V-gene distribution in the lambda light chains was similar to the distribution in the original library. Only antibodies comprising the V-gene VL6-57 were selected from the lambda library HAL10. In antibodies selected from the kappa library, VK2 and VK4 were underrepresented.
Anti-SARS-CoV-2 scFv-Fc were produced transiently in mammalian cells. In the interest of rapid throughput to quickly address the growing impact of the COVID-19 pandemic, only a selection of the unique antibodies was chosen for production as scFv-Fc and characterization. Antibodies with potential glycosylation sites in the complementarity determining regions (CDRs), identified by in silico analysis, were excluded. A total of 109 scFv-Fc antibodies were produced in 5 mL culture scale, with yields ranging from of 20 to 440 mg/L.
Antibodies inhibit the binding of spike to ACE2-positive cells in the scFv-Fc format. To further select potential therapeutic candidates, an inhibition assay was established using flow cytometry of ACE2-positive cells, measuring competition of S1-S2 trimer binding by scFv-Fc antibodies. The entire spike protein ectodomain was used for this inhibition assay for optimal representation of the viral binding. In a first screening, the 109 scFv-Fc were tested at 1500 nM (molar ratio antibody: S1-S2 30:1). Seventeen antibodies with inhibition better than 75% were selected for further analysis ( Fig. 2A, Table 3, and Supplementary  Fig. 3). The complete V-gene sequences of these 17 antibodies are given in Supplementary Fig. 4.
The inhibition of the 17 antibodies was further validated on human Calu-3 cells, which naturally express ACE2 9 using RBD-mFc ( Supplementary Fig. 5A) and S1-S2-His ( Supplementary  Fig. 5B) showing a stronger inhibition on Calu-3 compared to the transiently overexpressing ACE2-positive Expi293F cells. The Expi293F system allowed an improved estimation of inhibition potency when using the complete S1-S2 spike protein, because the S1-S2 was directly labeled with a fluorophore and the signals were not amplified in comparison to RBD with a murine Fc and a fluorophore-labeled secondary antibody. Further, ACE2-expressing Expi293F cells present a much higher amount of ACE2 receptor on their surface compared to Calu-3, due to the CMV-mediated expression (data not shown). Taken together, these data show that all 17 inhibiting antibodies selected against S1 directly interfered with RBD-ACE2 binding.
Determination of EC50 of the inhibiting antibodies to RBD, S1, and S1-S2. The EC50 of the inhibiting scFv-Fc on RBD, S1 (without furin site), and S1-S2 spike was measured by ELISA. All inhibiting antibodies bound the isolated RBD (Fig. 3), identifying it as their target on the viral surface. Most of the inhibiting antibodies showed a half-maximal binding in the subnanomolar range for RBD. While STE72-2G4 showed subnanomolar EC50 values for RBD and S1, it was discarded due to noticeable crossreactivity to mFc. The EC50 on the S1-S2 spike trimer was reduced for most of the antibodies, in comparison to the isolated RBD or S1.
ScFv-Fc combinations show synergistic effects in inhibition assays. Combinations of best-inhibiting scFv-Fc were tested in the flow cytometric inhibition assay using 1500 nM antibody and 50 nM S1-S2 spike ( Supplementary Fig. 6). Some of the combinations showed an increase of inhibition compared to the same amount of individual antibodies.
Anti-RBD scFv-Fc neutralize active SARS-CoV-2. All 17 inhibiting scFv-Fc were screened in a cytopathic effect (CPE)-based neutralization assay using 250 plaque-forming units (pfu)/well SARS-CoV-2 Münster/FI110320/1/2020 and 1 µg/mL (~10 nM) scFv-Fc ( Fig. 4A and Table 3) to select antibodies for further characterization as IgG. VeroE6 cells showed pronounced CPE characterized by rounding and detachment clearly visible in phase-contrast microscopy upon SARS-CoV-2 infection within 4 days, while uninfected cells maintained an undisturbed confluent monolayer. Virus inoculum preincubated with anti-RBD antibodies led to decreased CPE in varying degrees quantified by automated image analysis for cell confluence. The scFv-Fc format allowed rapid production for neutralization testing and allowed faster functional pre-screening. All 17 antibodies showed neutralization in this assay. Figure 4B shows examples for strong (STE73-6C8) and weak (STE73-2C2) neutralizing antibodies and controls.
Binding, inhibition, and SARS-CoV-2 neutralization in the IgG format. Eleven scFv-Fc showing a neutralization efficacy of 90% according to the CPE-based neutralization assay were converted into the IgG format. First, their binding was analyzed by titration ELISA on RBD, S1, and S1-S2 (Supplementary Fig. 7 and Table 3). Three antibodies lost binding after conversion to IgG (STE72-8A2, STE72-8A6, and STE73-6B10, data not shown), others showed reduced binding of different degrees, while three antibodies retained their binding (STE70-1E12, STE72-4E12, and STE73-2E9). In the next step, the antibodies were tested in the cell-based inhibition assay using S1-S2 ( Fig. 5A) or RBD (Fig. 5B).
Here the inhibition was confirmed for STE73-2E9, -9G3, and -2G8. STE72-1B6 showed inhibition of RBD comparable to the latter antibodies, but its activity was almost absent on S1-S2, thus it was not further considered.
Inhibiting IgGs are binding at the RBD-ACE2 interface. The efficiently inhibiting IgGs STE73-2E9, STE73-2G8, and STE73-9G3 were analyzed for their binding to various S1 subunit variants harboring a panel of recently reported mutations in RBD region and the D614G mutation. Three assays were employed: (1) ELISA ( Fig. 6A, B), (2) surface plasmon resonance (SPR) (Fig. 6B), and (3) bScreen protein array ( Fig. 6B) with S1 and/or RBD-SD1 proteins from different sources. All three antibodies lost binding to RBD mutations in the region aa483-486 directly at the RBD-ACE2 interface, showing that mutations in that region affect their epitope on the antigen. There were only minor differences between different approaches, e.g., at positions aa439 and aa476 for STE73-2E9. We then used this information to guide and validate computational docking simulations followed by atomistic molecular dynamics simulations according to protocols developed and well established in our group 35 , obtaining threedimensional atomic models of the antibody-RBD interaction for these three antibodies (Fig. 6C). The binding models of the three antibodies to the spike ectodomain are shown in Supplementary  Fig. 8.
Analysis of STE73-2E9 cross-reactivity with other coronaviruses. The neutralizing antibody STE73-2E9 was further characterized by titration ELISA on SARS-CoV-2 spike recombinant constructs (Fig. 7C) and S1 subunits from different coronaviruses ( Fig. 7D) showing that STE73-2E9 is binding specifically SARS-CoV-2 spike protein in the RBD region. The specific binding to S1/RBD of SARS-CoV-2 was further confirmed by the bScreen protein array-binding analysis (data not shown).
Aggregation behavior of STE73-2E9. The aggregation behavior of biologicals is a key factor for therapeutic development. STE73-2E9 shows no relevant aggregation under normal conditions (pH 7.4, room temperature (RT) in PBS), heat stress conditions (pH 7.4, 45°C, 24 h in PBS), and pH stress (pH 3, 24 h, RT), implicating that it has favourable general physicochemical properties that are a prerequisite for the development into a passive vaccine ( Supplementary Fig. 9).

Discussion
For 130 years, antibodies in animal sera or convalescent human plasma were successfully used for the treatment of infectious diseases, starting with the work of Emil von Behring und Shibasaburo Kitasato against diphtheria 35 . However, the efficacy of human plasma derived from convalescent donors depends on the viral pathogen. In case of Ebola, the survival upon treatment with    72 . The EC50 were measured on 30 ng immobilized RBD-mFc, S1-mFc, S1-S2-His (trimer) by ELISA. The IC50 was measured by flow cytometry using 50 nM (in relation to monomer) S1-S2 trimer, respectively, 10 nM RBD, and ACE2-positive cells. The molar ratio of antibody-binding site: S1-S2 or RBD is given for 50% inhibition. CPE-based neutralization assay was performed with 250 pfu/well SARS-CoV-2 and 1 µg/mL (~100 nM) scFv-Fc (median neutralization %). EC50 were calculated with GraphPad Prism Version 6.1, fitting to a four-parameter logistic curve. IC50 values were calculated using Logistic5 fit of Origin.
convalescent human plasma was not significantly improved over the control group 36 . On the other hand, reduced mortality and safety was shown for convalescent plasma transfer in case of influenza A H1N1 in 2009 37,38 . This approach was also used against emerging coronaviruses. While the outcomes were not significantly improved in a very limited number of MERS patients 39 , the treatment was successful for SARS 40,41 . This approach was also used for COVID-19 with promising results 42 .
The mode of action of these polyclonal antibody preparations may vary, including virus neutralization, Fcγ receptor-bindingmediated phagocytosis, or antibody-dependent cellular cytotoxicity (ADCC) as well as complement activation [43][44][45] . In any serum therapy, the composition and efficacy of convalescent plasma is expected to differ from donor to donor, as well as from batch to batch, and sera must be carefully controlled for viral contaminations (e.g., HIV, hepatitis viruses) and neutralization potency. A convalescent patient can provide 400-800 mL plasma, with 250-300 mL of plasma typically needed per treatment. With two rounds of treatment per patient, this is a grave limitation, since one donor can only provide material for 1-2 patients 42,43 . Human or humanized monoclonal antibodies are a powerful alternative to polyclonal antibodies derived from convalescent plasma. Following this approach, the humanized antibody Palivizumab was approved in 2009 for treatment and prevention of RSV infections 46 . Other antibodies against viral diseases successfully tested in clinical studies are mAb114 and REGN-EB3 against Ebola disease 24 . Phage display-derived antibodies are typically well-established medications: 12 such antibodies are approved by EMA/FDA at the time of writing, a significant increase compared to the six such antibodies approved in 2016 47 . In this work, we used phage display to isolate monoclonal human antibodies capable of neutralizing SARS-CoV-2 from a universal, naïve antibody gene library that was generated from healthy donors before the emergence of the SARS-CoV-2 virus. This allowed selection of human antibodies against this virus without the necessity to obtain material from COVID-19-infected individuals. While most antibodies against SARS-CoV-2 were obtained from convalescent patients with few exceptions 20,21,48,49 , our approach demonstrates that human antibodies can be generated without the necessity to wait for material from COVID-19-infected individuals. With SARS-CoV-2, the whole world invested an unprecedented amount of resources toward a single goal. This is not common for other diseases and infective outbreaks. The fact that research groups and companies were able to rapidly recruit many individuals and quickly find potentially good antibodies should not be held as standard for every situation. Therefore, this strategy offers a very fast additional opportunity to respond to future pandemics.
As the human receptor of SARS-CoV and SARS-CoV-2 is ACE2 3 , we focused on antibodies which directly block the interaction of the spike protein with this receptor and antibodies preventing ACE2 binding were shown to potently neutralize the closely related SARS-CoV virus 16 . Three hundred and nine unique fully human monoclonal antibodies were generated using different panning strategies. The S1 subunit produced in insect cells was better suited for antibody selection than the S1 subunit produced in mammalian cells. The V-gene distribution of the selected anti-Spike antibodies is largely in accordance with the Vgene subfamily distribution shown by Kügler et al. 50 for antibodies selected against 121 other antigens from HAL9/10. Only the VH1 subfamily was overrepresented and VH4 and Vkappa4 subfamilies were rarely selected despite their presence in the HAL libraries. The most frequently used V-gene was VH3-30. Interestingly, an increased use of this V-gene in anti-SARS-CoV-2 antibodies was also described by Robbiani et al. 51 for anti-RBD B cells selected from COVID-19 patients. By contrast, the second most selected V-gene was VH3-53, which was selected in our approach only once. Robbiani et al. also described an overrepresented use of VL6-57, as found in our antibodies as well. However, it has to be noted that VL6 is also overrepresented in our naïve library compared to its in vivo occurrence.
From the initial 309 scFv, 109 were recloned in the scFv-Fc IgG-like bivalent format. Their ability to inhibit binding of fluorescently labeled S1-S2 trimer to ACE2-expressing cells was assessed by flow cytometry. The half-maximal inhibition of the best inhibiting 17 scFv-Fc was measured both with the spike In the trimeric spike protein, the RBD can be in an "up" (open) or "down" (close) position. The "down" conformation cannot bind to ACE2, in contrast to the less stable "up" conformation 33 . The RBDs can be in different conformations on the same spike trimer, which offers a possible explanation for the observed effective antibody to spike molar ratios lower than 1:1. This is in accordance with the cryo-electron microscopic images recorded by Walls et al. 4 , where they could find half of the recorded trimers with one RBD in the open conformation. We observed that molar ratios for half maximal inhibition were lower for RBD compared to spike protein. For some antibodies, approximately 0.5 antigen-binding sites were needed to achieve a 50% inhibition. The fact that the antibodies are more efficient at inhibiting RBD binding to ACE2 rather than S1-S2 trimer binding can be explained with the higher affinity of the antibodies for the isolated RBD compared to the trimeric spike, which in turn points to the presence of partially or completely inaccessible epitopes on the trimer, an occurrence seen in other viruses. This is similar to what was reported by Pinto et al. 17 , who also showed a lower affinity of the antibody S309 for spike compared to RBD.
Inhibition of ACE2 binding was stronger on the human lung cells Calu-3, which better represent the in vivo situation than transiently ACE2-overexpressing cells. Nevertheless, we did the titration assays on ACE2-overexpressing Expi293F cells because these seemed to allow a better quantitative discrimination of inhibiting potency.
Antibody combinations can have a synergistic effect as previously described for toxins and viruses 32,[53][54][55] . This approach may also avoid formation of viral escape mutants. Here the best combinations showed a significantly improved inhibition efficacy, at least when using an excess of antibodies (Ab:Ag molar ratio 30:1).
All of the 17 scFv-Fc were tested in neutralization assays using a SARS-CoV-2 strain isolated from a patient and all antibodies showed a degree of neutralization in this assay. While this study did not aim to define the lowest effective concentration of individual antibodies in limiting dilution conditions, all tested antibodies showed a clear and measurable effect at a relatively low concentration. Therefore, our approach provided a rapid selection of antiviral antibodies.
In a next step, we converted 11 antibodies with the best neutralization efficacy according to the cytopathic assay into the IgG format. It was completely unexpected that most antibodies lost efficacy in the inhibition assay after conversion from scFv-Fc to IgG including antibodies like STE70-1E12 without loss of affinity according to the titration ELISA. These results are in contrast to former results where none 54 or only a low percentage 55,56 of antibodies lost efficacy after conversion from scFv-Fc to IgG. Nevertheless, three antibodies showed a good inhibition in the cell-based assay and did not bind to the region of aa483-486 known for RBD mutations from publications 19,57 and from GISAID database (www.gisaid.org). Experimentally validated computational docking shows that these antibodies still at least partially occupy the ACE2-binding site on the RBD, thus likely achieving direct inhibition of virus-ACE2 interaction. The binding sites of antibody BD368-2 18 , B38 58 , and REGN10933 59 also overlap with the RBD-ACE2 binding interface. The best neutralizing antibody STE73-2E9 showed an IC50 of 0.41 nM, which is higher compared to antibodies derived from COVID-19 patients. Cao et al. 18 reported an IC50 of 33 ng/mL (~0.22 nM) for BD368-2 in a comparable live virus plaque assay. Other publications reports better IC50 efficacies, e.g., Kreye et al. 60 for CV07-209 with 3.1 ng/mL (~0.02 nM) or Rogers et al. 61 for CC6.33 with 1 ng/mL. Some assays are often not directly comparable, e.g. Rogers et al. used SARS-CoV-2 pseudovirus instead of authentic virus.
The neutralizing antibody STE73-2E9 was specific for SARS-CoV-2, and we conclude that this antibody is a suitable candidate for the development of passive immunotherapy for the treatment of COVID-19. It could be used not only therapeutically to prevent individuals from being hospitalized in intensive care units but also prophylactically to protect health care workers or risk groups who do not respond to vaccination. Before clinical application, the risk of antibody-dependent enhancement (ADE) of disease has to be considered for COVID-19. In contrast to antibodies against Ebola where ADCC is important for protection 22 , antibodies directed against the spike protein of SARS-CoV-2 may lead to ADE [62][63][64] . SARS cause an acute lung injury, which is also driven by immune dysregulation and inflammation caused by anti-spike antibodies 65 . While Quinlan et al. 66 described that animals immunized with RBD SARS-CoV-2 did not mediate ADE and suggested for vaccines the use of RBD, some of the monoclonal antibodies we analyzed in this study lead to an increased binding of the spike protein to ACE2-positive cells. A possible explanation could be multimerization of the spike by   6 Binding to RBD mutants, epitopes, and structure models. A ELISA using STE73-2E9, -9G3, and -2G8 on S1-His with different RBD mutations. B Overview of the binding of STE73-2E9, -9G3, and -2G8 to different RBD mutations analyzed by ELISA, SPR, and protein array. Sequence SARS-CoV-2 (Gene bank QHD43416). ELISA experiments were performed in duplicate and mean values are given. C The three antibodies STE73-2E9, -9G3, and -2G8 are binding to the ACE-RBD interface (docking models based on epitope data from binding to RBD mutations). Experimentally validated computational models of the variable regions of the antibodies (colored cartoons) binding to the RBD (white surface, same orientation in all images) are shown. The cartoon representation of ACE2 is also shown for comparison. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21609-2 antibody "cross-linking" in this assay or the stabilization of an infection-promoting conformation by the antibodies. These aspects need to be carefully considered in any development of therapeutic antibodies; we suggest to focus on RBD and/or the use of silenced Fc parts with deleted Fcγ and C1q binding 67-69 for safety reasons.
In conclusion, we report the successful isolation and characterization of a fully human, recombinant anti-spike neutralizing monoclonal antibody from naïve phage display libraries. Our approach demonstrates how neutralizing antibodies can be efficiently selected in a rapid time frame and without the need of convalescent patient material. Furthermore, the strategy we used efficiently targeted the spike:ACE interface allowing the selection of directly blocking antibodies.

Methods
Design of expression vectors. Production in Expi293F cells was performed using pCSE2.5-His-XP, pCSE2.6-hFc-XP, or pCSE2.6-mFc-XP 70 where the respective single-chain variable fragment of the antibodies or antigens were inserted by NcoI/ NotI (NEB Biolabs) digestion. Antigen production in High Five insect cells was performed using NcoI/NotI compatible variants of the OpiE2 plasmid 71 containing an N-terminal signal peptide of the mouse Ig heavy chain; the respective antigen; and C-terminal 6xHis-tag, hFc, or mFc. Single point mutations in S1-HIS constructs were inserted through site-directed mutagenesis using overlapping primers according to Zheng et al. 72 with slight modifications: S7 fusion polymerase (Mobidiag, Espoo, Finland) with the provided GC buffer and 3% dimethyl sulfoxide was used for the amplification reaction. The used olignucleotide primers for cloning and site-directed mutagenesis are given in Supplementary Table 1.
Production of antigens in insect cells. Different domains or subunits of the spike protein (GenBank: MN908947), S1 subunit mutants, and the extracellular domain of ACE2 receptor (GenBank NM_021804.3) were baculovirus-free produced in High Five cells (Thermo Fisher Scientific) by transient transfection as previously described in Bleckmann et al. 34 . Briefly, High Five cells were cultivated at 27°C, 110 rpm in ExCell405 media (Sigma) and kept at a cell density between 0.3 × 10 6 and 5.5 × 10 6 cells/mL. For transfection, cells were centrifuged and resuspended in fresh media to a density of 4 × 10 6 cells/mL and transfected with 4 µg plasmid/mL and 16 µg/mL of PEI 40 kDa (Polysciences). Four hours up to 24 h after transfection, cells were fed with 75% of the transfection volume. At 48 h after transfection, cell culture medium was doubled. Cell supernatant was harvested 5 days after transfection in a two-step centrifugation (4 min at 180 × g and 20 min at above 3500 × g) and 0.2 µm filtered for purification.
Protein purification. Protein purification was performed depending on the production scale in either 24-well filter plate with 0.5 mL resin (10 mL scale) or 1 mL column on Äkta go (Cytiva), Äkta Pure (Cytiva), or Profina System (BIO-RAD). MabSelect SuRe or HiTrap Fibro PrismA (Cytiva) was used as resins for Protein A purification. For His-tag purification of Expi293F, supernatant HisTrap FF Crude column (Cytiva) and for His-tag purification of insect cell supernatant HisTrap excel column (Cytiva) was used. All purifications were performed according to the manufacturer's manual. Indicated antigens were further purified by SEC by a 16/600 Superdex 200 kDa pg (Cytiva). All antigens, antibodies, and scFv-Fc were run on Superdex 200 Increase 10/300GL (Cytiva) on Äkta or HPLC (Techlab) on an AdvanceBio SEC 300 Å 2.7 µm, 7.8 × 300 mm (Agilent) for quality control.
Antibody selection using phage display. The antibody selection was performed as described previously with modifications 73 . In brief, for panning procedure, the antigen was immobilized on a High binding 96-well plate (Corning, Costar). Five micrograms of S1-hFc (produced in High Five cells) was diluted in carbonate buffer (50 mM NaHCO 3 /Na 2 CO 3 , pH 9.6) and coated onto the wells at 4°C overnight. Next, the wells were blocked with 350 µL 2% MBPST (2% (w/v) milk powder in PBS; 0.05% Tween20) for 1 h at RT and then washed 3 times with PBST (PBS; 0.05% Tween20). Before adding the libraries to the coated wells, the libraries (5 × 10 10 phage particles) were preincubated with 5 µg of an unrelated scFv-Fc and 2% MPBST on blocked wells for 1 h at RT, to deprive libraries of human Fc fragment binders. The libraries were transferred to the antigen-coated wells, incubated for 2 h at RT, and washed 10 times. Bound phage were eluted with 150 µL trypsin (10 µg/mL) at 37°C, 30 min and used for the next panning round. The eluted phage solution was transferred to a 96 deep well plate (Greiner Bio-One, Frickenhausen, Germany) and incubated with 150 µL E. coli TG1 (OD 600 = 0.5) first for 30 min at 37°C, then 30 min at 37°C and 650 rpm to infect the phage particles. One milliliter 2xYT-GA (1.6% (w/v) Tryptone; 1 % (w/v) Yeast extract; 0.5% (w/v) NaCl (pH 7.0), 100 mM D-Glucose, 100 µg/mL ampicillin) was added and incubated for 1 h at 37°C and 650 rpm, followed by addition of 1 × 10 10 colony-forming units M13KO7 helper phage. Subsequently, the infected bacteria were incubated for 30 min at 37°C followed by 30 min at 37°C and 650 rpm before centrifugation for 10 min at 3220 × g. The supernatant was discarded and the pellet resuspended in fresh 2xYT-AK (1.6% (w/v) Tryptone; 1 % (w/v) Yeast extract; 0.5% (w/v) NaCl (pH 7.0), 100 µg/mL ampicillin, 50 µg/mL kanamycin). The antibody phage were amplified overnight at 30°C and 650 rpm and used for the next panning round. In total, four panning rounds were performed. In each round, the stringency of the washing procedure was increased (20× in panning round 2, 30× in panning round 3, 40× in panning round 4) and the amount of antigen was reduced (2.5 µg in panning round 2, 1.5 µg in panning round 3, and 1 µg in panning round 4). After the fourth as well as third panning round, single clones containing plates were used to select monoclonal antibody clones for the screening ELISA.
Screening of monoclonal recombinant binders using E. coli scFv supernatant. Soluble antibody fragments (scFv) were produced in 96-well polypropylene MTPs (U96 PP, Greiner Bio-One) as described before 55,73 . Briefly, 150 μL 2xYT-GA was inoculated with the bacteria bearing scFv-expressing phagemids. MTPs were incubated overnight at 37°C and 800 rpm in a MTP shaker (Thermoshaker PST-60HL-4, Lab4You, Berlin, Germany). A volume of 180 μL 2xYT-GA in a PP-MTP well was inoculated with 20 μL of the overnight culture and grown at 37°C and 800 rpm for 90 min (approx. OD 600 of 0.5). Bacteria were harvested by centrifugation for 10 min at 3220 × g, and the supernatant was discarded. To induce expression of the antibody genes, the pellets were resuspended in 200 μL 2xYT supplemented with 100 μg/mL ampicillin and 50 μM isopropyl-beta-D-thiogalacto pyranoside and incubated at 30°C and 800 rpm overnight. Bacteria were pelleted by centrifugation for 20 min at 3220 × g and 4°C.
Inhibition of S1-S2 binding to ACE2 expressing cells using MacsQuant. The inhibition tests in cytometer on Expi293F cells were performed based on the protocol for "validation of spike protein binding to ACE2" (see above) but only binding to S1-S2-His and RBD-mFc antigen (High Five cell produced) was analyzed. The assay was done in two set-ups. In the first set-up, 50 nM antigen was incubated with min. 1 µM of different scFv-Fc and the ACE2-expressing cells. The resulting median antigen fluorescence of GFP-positive living single cells was measured. For comparison of the different scFv-Fc, first the median fluorescence background of cells without antigen was subtracted, second it was normalized to the antigen signal where no antibody was applied. All scFv-Fc showing an inhibition in this set-up were further analyzed by titration (max. 1500-4.7 nM) on S1-S2-His (High Five cell produced), respectively, on RBD-mFc (max. 100-0.03 nM). The IC50 was calculated using the equation f(x)=Amin + (Amax − Amin)/ (1 + (x0/x)^h)^s and parameters from Origin Pro 2019. In addition, pairwise combinations (max. 750 nM of each scFv-Fc) of the different inhibiting scFv-Fc were tested.
SARS-CoV-2 neutralization in cell culture. VeroE6 cells (ATCC CRL-1586) were seeded at a density of 6 × 10 4 /well onto cell culture 96-well plates (Nunc, Cat. #167008). Two days later, cells reached 100% confluence. For neutralization, antibodies (1 µg/mL final concentration) were mixed with the virus inoculum (250 pfu/well), using strain SARS-CoV-2/Münster/FI110320/1/2020 (kind gift of Stephan Ludwig, University of Münster, Germany), in 100 µL full VeroE6 culture medium (Dulbecco's modified Eagle's medium, 10% fetal calf serum, 2 mM glutamine, penicillin, streptomycin) in technical quadruplicates or sixfold replicates and incubated for 1 h at 37°C. Then cells were overlaid with the antibody/virus mix and phase-contrast images were taken automatically using a Sartorius Incu-Cyte S3 (×10 objective, 2-h image intervals, 4 images per well) housed in a HeraCell 150i incubator (37°C, 100% humidity, 5% CO 2 ). Image data were quantified with the IncuCyte S3 GUI tools measuring the decrease of confluence concomitant with the CPE of the virus in relation to uninfected controls and controls without antibody and analyzed with GraphPad Prism 8. Given is the median of the inhibition.
For titration, antibodies were diluted in 1/√10 steps and mixed with a fixed inoculum of SARS-CoV-2 (100-150 pfu) in a total volume of 500 µL of Vero E6 medium. After 1 h incubation at 37°C, cells were infected with the antibody/virus mix, incubated for 1 h, and then overlaid with Vero E6 medium containing 1.5% methyl-cellulose. Three days postinfection, wells were imaged using a Sartorius IncuCyte S3 (×4 objective, whole-well scan) and plaques were counted from these images.
Cloning and production of IgG. For IgG production, selected antibodies were converted into the human IgG1 format by subcloning of VH in the vector pCSEH1c (heavy chain) and VL in the vector pCSL3l/pCSL3k (light chain lambda/ kappa) 76 , adapted for Golden Gate Assembly procedure with Esp3I restriction enzyme (New England Biolabs). Expi293F (Thermo Fisher Scientific) cells were transfected with 12.5 µg of both vectors in parallel in a 1:1 ratio. For production, the transfected Expi293F cells were cultured in chemically defined medium F17 (Thermo Fisher Scientific) supplemented with 0.1% pluronic F68 (PAN-Biotech) and 7.5 mM L-glutamine (Merck) for 7 days. A subsequent protein A purification was performed as described above.
Affinity determination by SPR. The antibody-binding properties were analyzed at 25°C on a Biacore™ 8 K 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-CoV2 RBDs, wild type and different mutants, or full S-protein, were immobilized on the surface of a CM5 chip through standard amine coupling. Increasing concentration of antibodies (6.25-12.5-25-50-100 nM) were injected using a single-cycle kinetics setting and analyte responses were corrected for unspecific binding and buffer responses. Curve fitting and data analysis were performed with the Biacore™ Insight Evaluation Software (version 2.0.15.12933).
anti-human step were used for data generation. This was done by subtracting the signal mean value from 30 to 20 s before the association phase from the signal mean value from 20 to 30 s after the association phase.
Antibody structures and computational docking studies. The antibody structures were modeled according to the canonical structure method using the RosettaAntibody program 77 as previously described 78 and docked to the experimental structure of SARS-CoV-2 S protein RBD (PDBid: 6M17) 6 .
Docking was performed using the RosettaDock 3.12 software 79 as previously described 80 . Briefly, each antibody was manually placed with the CDR loops facing the RBD region containing the residues identified by the peptide mapping experiment. The two partners were moved away from each other by 25 Å and then brought together by the computational docking algorithm, obtaining thousands of computationally generated complexes (typically 15,000). The antibody/RBD complexes were structurally clustered and then selected according to the scoring function (an estimate of energetically favorable solutions) and agreement with the peptide mapping data. Selected complexes were further optimized by a docking refinement step and molecular dynamics simulations.
The MD simulations were performed using GROMACS 81 with standard MD protocol: antibody/antigen complexes were centered 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 followed. Temperature (298 K) and pressure (1 Bar) equilibration steps of 100 ps each were performed. Five hundred-nanosecond MD simulations was run with the abovementioned force field for each protein complexes. MD trajectory files were analyzed after removal of periodic boundary conditions. The overall stability of each simulated complex was verified by root mean square deviation, radius of gyration, and visual analysis according to standard procedures.