A SARS-CoV-2 neutralizing antibody selected from COVID-19 patients by phage display is binding to the ACE2-RBD interface and is tolerant to most known recently emerging RBD mutations

The novel betacoranavirus SARS-CoV-2 causes a form of severe pneumonia disease, termed COVID-19 (coronavirus disease 2019). Recombinant human antibodies are proven potent neutralizers of viruses and can block the interaction of viral surface proteins with their host receptors. To develop neutralizing anti-SARS-CoV-2 antibodies, antibody gene libraries from convalescent COVID-19 patients were constructed and recombinant antibody fragments (scFv) against the receptor binding domain (RBD) of the S1 subunit of the viral spike (S) protein were selected by phage display. The selected antibodies were produced in the scFv-Fc format and 30 showed more than 80% inhibition of spike (S1-S2) binding to cells expressing ACE2, assessed by flow cytometry screening assay. The majority of these inhibiting antibodies are derived from the VH3-66 V-gene. The antibody STE90-C11 showed a sub nM IC50 in a plaque-based live SARS-CoV-2 neutralization assay. The in vivo efficacy of the antibody was demonstrated in the Syrian hamster and in the hACE2 mice model using a silenced human IgG1 Fc part. The crystal structure of STE90-C11 Fab in complex with SARS-CoV-2-RBD was solved at 2.0 Å resolution showing that the antibody binds at the same region as ACE2 to RBD. The binding and inhibtion of STE90-C11 is not blocked by many known RBD mutations including N439K, L452R, E484K or L452R+E484Q (emerging B.1.617). STE90-C11 derived human IgG1 with FcγR silenced Fc (COR-101) is currently undergoing Phase Ib/II clinical trials for the treatment of moderate to severe COVID-19. In Brief Human antibodies were selected from convalescent COVID-19 patients using antibody phage display. The antibody STE90-C11 is neutralizing authentic SARS-CoV-2 virus in vitro and in vivo and the crystal structure of STE90-C11 in complex with SARS-CoV-2-RBD revealed that this antibody is binding in the RBD-ACE2 interface. S1 binding of STE90-C11 and inhibition of ACE2 binding is not blocked by many known RBD mutations.


Introduction
The severe pneumonia COVID-19 (coronavirus disease 2019) is a disease caused by the novel coronavirus SARS-CoV-2 and was described at the end of 2019 in Wuhan China . This new human pathogenic coronavirus is closely related to the bat coronavirus RATG13 indicating an animal-to-human transition (Shang et al., 2020a). The spike (S) protein of SARS-CoV-2 binds to the human zinc peptidase angiotensin-converting enzyme 2 (ACE2) as main receptor for cell entry. The S protein has two subunits: the N-terminal S1 harbouring the receptor binding domain (RBD) and the viral membrane-anchored C-terminal S2 subunit, which is required for trimerization and fusion of the virus and host membrane for viral entry (Shang et al., 2020b;Starr et al., 2020;Walls et al., 2020;Wang et al., 2020b;Wrapp et al., 2020;Yan et al., 2020) .
Blocking of the RBD-ACE2 interaction by therapeutic antibodies as a strategy to treat COVID-19  is a promising approach since the successful generation of neutralizing antibodies against S1 and RBD was already demonstrated for SARS-CoV and MERS (Coughlin and Prabhakar, 2012;Widjaja et al., 2019;Zhu et al., 2007). Some of the anti-SARS-CoV antibodies showed also cross-reaction and crossneutralization of SARS-CoV-2 (Huo et al., 2020;Tian et al., 2020;Wang et al., 2020a).
Antibody phage display is a widely used in vitro technology to select human antibody fragments for the development of therapeutic antibodies . Currently, 14 EMA or FDA approved antibodies were generated by antibody phage display (Alfaleh et al., 2020). Human antibodies can be selected from two kind of antibody gene libraries: universal libraries and immune libraries. Universal libraries which have a naïve, semisynthetic or synthetic antibody gene repertoire, allow -in theory -the selection of antibodies against any kind of molecule, while immune libraries derived from immunized donors typically facilitate the antibody generation against the pathogen that affected the 7/57 donors (Bradbury et al., 2011;Frenzel et al., 2017;Kretzschmar and von Rüden, 2002;Kügler et al., 2018;Kuhn et al., 2016). Immune phage display libraries from convalescent patients or immunized human donors were used in the past for the successful generation of antibodies against infectious diseases (Duan et al., 2009;Rahumatullah et al., 2017;Trott et al., 2014;Wenzel et al., 2020a;Zhang et al., 2003). Human antibodies were also selected directly against SARS-CoV-2 using different approaches including single B-cell cloning or antibody phage display Bertoglio et al., 2021;Kreer et al., 2020;Robbiani et al., 2020;Shi et al., 2020;Zeng et al., 2020).
In this work, immune phage display libraries from six COVID-19 convalescent patients were constructed, RBD-binding antibodies have been selected resulting in SARS-CoV-2 inhibiting and neutralizing antibodies. The crystal structure of the best neutralizing candidate STE90-C11 in complex with wt RBD was elucidated. Together with its unique binding pattern that tolerates a large number of RBD mutations, its properties suggest its use as a therapeutic agent against SARS-CoV-2 infections.

Immune library construction
For the immune library construction, we collected blood samples of 16 q-RT-PCR confirmed COVID-19 convalescent patients. To analyze the immune answer, sera of all patients were titrated on RBD (Supplementary Data 1). The patients 1,2,5,6,9 and 14 showed the highest IgG titers against RBD and were chosen for library construction.
PBMCs were isolated and CD19+/CD138+ plasma cells were enriched by FACS from some of these samples (Supplementary Data 2). Both PBMCs and plasma cells were used for library construction. Six individual libraries were constructed from PBMCs of patients 2, 6 and 14, while four additional libraries were made from CD19+/CD138+ plasma cells pools (pool 1: patients 2,6,14; pool 2: patients1,5,9). All libraries were constructed cloning kappa and lambda antibody genes separately, resulting in diversities ranging from 0.7x10 7 to 1.7x10 8 estimated independent clones.

Antibody fragments were selected against SARS-CoV-2
Antibodies were selected by panning in microtiter plates against SARS-CoV-2 S1 subunit using S1-S2 in the first panning round, RBD in the second panning round and S1-S2 in the third panning round. This approach focused on the antibody selection on epitopes located within the RBD while being presented in the correct conformation of the intact spike protein trimer. The two plasma cell-derived immune libraries for kappa, respectively lambda, were combined and the panning was performed separately for lambda and kappa libraries. All kappa and lambda libraries originating from the PBMCs were pooled together. The following single clone screening was performed by antigen ELISA in 96 well MTPs, using The most abundantly used V-genes were VH3-66, VK1D-39 und VL3-21. Antibody sequences that showed potential developability liabilities, i.e. glycosylation sites or single cysteines in the CDRs were excluded from further analysis. 135 antibodies were recloned into the bivalent scFv-Fc format and produced in Expi293F cell in 5 mL culture scale, with yields ranging from of 70 to 260 mg/L.

Screening for inhibition in a cell-based assay
To screen for SARS-CoV-2 blocking antibodies, an inhibition assay was performed by flow cytometry on ACE2-expressing EXPI293F cells, measuring direct competition between S1-S2 trimer and 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 135 antibodies were tested at 1500 nM S1-S2 (molar ratio antibody:S1-S2 30:1) ( Figure 1). 30 antibodies that inhibited S1-S2 binding more than 80% were selected for further analysis.
The germlines genes of these inhibiting scFv-Fc are given in Table 1. VH3-66 was the most abundantly used heavy chain V-gene of the well inhibiting scFv-Fc. All VH3-66 derived antibodies had closely related VH CDRs amino acid sequence while the corresponding light chains were different.

Antigen binding and cell-based inhibition by IgG
The 30 antibodies were screened in a cytopathic effect (CPE)-based neutralization screening assay (data not shown) to select antibodies for further characterization as IgG.

Antibodies from COVID19 patient libraries neutralize active SARS-CoV-2 virus
To analyze the neutralization capacity of the antibodies, the 19 IgGs were first screened in a plaque titration assay using active ~15 pfu SARS-CoV-2 virus. The antibody Palivizumab was used as control antibody ( Figure 3A). The antibody STE90-C11 was chosen to be further analyzed. To confirm its neutralizing activity, a plaque assay using ~150 pfu was performed and the determined IC50 was 0.56 nM in the IgG format ( Figure 3B).

STE90-C11 is specific for SARS-CoV-2 and binds most RBD mutations including B.1.617
A detailed characterization of the best neutralizer was performed in respect to crossreaction to other coronaviruses, binding to known RBD mutations, as well as biochemical and biophysical properties. The binding of STE90-C11 to other coronaviruses SARS-CoV-11/57 1, MERS-CoV, HCov-HKU1, HCoV-229E and HCoV-NL63 Spike proteins was tested by titration ELISA (Supplementary Data 7). STE90-C11 bound specifically SARS-CoV-2 and did not show any cross reactions to other human-pathogenic coronaviruses.
Because first SARS-CoV-2 mutants in patients are being identified (www.gisaid.org and ) and more mutations within the RBD have been characterized to arise under antibody selection pressure , S1 subunits harboring single point mutations described in GISAID in the RBD were produced and binding of STE90-C11 to those mutants was tested. In addition a variant with seven point mutations (V367F, N439K, G476, V483A, E484K, G485R and F486V) was analyzed (S1-7PM). The EC50 values of STE90-C11 are given in Figure 4 and the corresponding ELISAs are given in Supplementary Data 8, including further binding assays comparing STE90-C11 to REGN10933, REGN10987, CR3022 and CB6. All S1 or S1-S2 constructs were still able to bind recombinant ACE2 (Supplementary Data 9), indicating correct folding. The antibody STE90-C11 showed reduced binding to N501Y and lost binding to K417N/K417T but was able to bind to all other investigated RBD mutations including a S1-7PM mutant and more relevant to E484K, N439K and L452R which are the RBD mutations in the emerging B.1.525, B.1.526, B.1.1.33, B.1.258 and B.1.429/B.1.427 variants. Most important is the binding to B.1.617 (L452R+E484Q) which is currently emerging in India and other parts of the world. Subsequently, the inhibition of S1 binding to ACE2 for the RBD mutants bound best by STE90-C11 was confirmed in the cell based inhibition assays (Supplementary Data 10). The inhibition of B.1.617 was as slightly better than the inhibition of the wildtype.

STE90-C11 is protective in vivo in the hamster challenge model
The SARS-CoV-2 neutralizing effect of STE90-C11 was tested in vivo in a Syrian hamster challenge model. In this model, hamsters were intranasally infected with 1x10 5 pfu genuine 12/57 SARS-CoV-2 and 2 h later treated with 3.7 mg/kg or 37 mg/kg STE90-C11 or with PBS in the control group. The virus titer in the lung of treated animals showed a dose dependent reduction of viral load on day 3 and 5 after infection ( Figure 5A). The measured mean viral load was 8.3x10 3 pfu with 37 mg/kg IgG compared to 1.5x10 6 pfu in the PBS control at day 3. The hamster model is further characterized by a rapid weight loss in the first days after SARS-CoV-2 infection but also rapid recovery of body weight from around one week post infection on. Animals treated with the higher STE90-C11 dose showed a reduced loss of weight and a faster weight recovery compared to the PBS control ( Figure 5B).

STE90-C11 is protective in vivo in the transgenic ACE2 mice challenge model
The SARS-CoV-2 neutralising effect of STE90-C11 was further tested in vivo in the transgenic human ACE2 mice model (K18hACE2). Mice were first treated with 6, 30, 60 or 120 mg/kg STE90-C11 or with PBS in the control group. After 1h, they were inoculated intranasally with 1,000 or 2,000 pfu genuine SARS-CoV-2, or with PBS in the control group. The virus titer was measured with 4.7x10 4 pfu in the first experiment with 1,000 pfu and 7.4x10 5 pfu in the experiment with 2,000 pfu. Only in the two experiments, one with 60 mg/kg and one with 6 mg/kg a remaining viral load of 5.6x10 2 , respectively 3.1x10 2 pfu, was detected. In all other mice treated with STE90-C11, the virus was completely removed from the lungs ( Figure 5C and 5D). In the control group, a significant weight loss was

SARS-CoV-2 escape mutant screening leads to increased K417x but not N501x mutations
An in vivo escape mutant screening experiment was performed according to Baum et al. (Baum et al., 2020) to determine RBD mutants which are not neutralized by STE90-C11.

13/57
Here, serial dilutions of STE90-C11, respectively Palivizumab as control antibody, were used to infect Calu-3 cells with active wt SARS-CoV-2. The supernatant with the propagated virus particles were used for the second round of neutralization and infection.
After the second passage, the viruses were analyzed by DNA sequencing. In the experiment with STE90-C11 4.7% of the remaining viruses were wt, 81.9% had the K417T mutations, 3.35 K417T+A475V, 0.3% A475V, 0.3% N481H, 0.12% N481Y and 0.11% N501T. In the control experiment with a non related antibody and no selection pressure 89.9% of the remaining viruses were wt, 0.32% N481H, 0.14% K417T, 0.13% N501T and 0.1% Q414K. This experiment shows that the K417x mutation was not neutralized by STE90-C11, but no selection pressure was found on the N501x mutations, indicating a sufficient neutralization of these variants by STE90-C11.

Biochemical characterization of STE90-C11
Aggregation propensity is an important quality parameter in therapeutic antibody nM. In the second setup, the IgG was immobilized using a Fab2G capture tip and the KD of monovalent S1-His was determined to be 1.6 nM. In the third assay, the IgG was 14/57 immobilized using a protein A capture tip, resulting in a measured KD of 6.5 nM to S1-His, confirming the affinity of STE90-C11 is in the low nanomolar range. STE90-C11 did not bind to RBD on immunoblots after treatment at 95°C and SDS-gel separation under reducing conditions, while it was able to detect RBD when separated under non-reducing conditions heated either at 95°C or 56°C (Supplementary Data 14). These data suggest that the antibody recognizes a conformational epitope.

STE90-C11 binds at the ACE2-RBD interface
To get further insight into the neutralizing mechanism of STE90-C11, a complex of STE90-C11 Fab and SARS-CoV-2 RBD (22 kDa fragment) was prepared and subjected to crystallization screening. X-ray diffraction images collected from resulting crystals yielded a dataset to an overall resolution limit of 2.0 Å (Supplementary Data 15). After solving the structure by molecular replacement a model was built into the electron density ( Figure 6). Figure 6A shows the binding of STE90-C11 to RBD and Figure 6B the binding of ACE2 to RBD. STE90-C11 binds SARS-CoV2 RBD with an interface area of 1113 Å 2 ( Figure 6C, Supplementary Data 16). Roughly 60% of this area can be contributed to the VH segment forming up to ten hydrogen bonds at the same time and region dominated by hydrophobic interaction (V99, A100, Y33, Y52) ( Figure 6D). The remaining 40% are provided by the VL segment contributing eight additional hydrogen bonds to stabilizing the interaction ( Figure   6E). All six CDR loops contribute to the interaction between STE90-C11 Fab and SARS-CoV-2 RBD. The superposition of the RBDs of the STE90-C11:SARS-CoV-2 RBD complex with a ACE2:SARS-CoV-2 RBD complex (Lan et al., 2020) resulted in a low Cα root mean square distance of 0.532 Å (183 atoms) indicating that binding of STE90-C11 did not induce substantial changes in the RBD. The superposition further revealed that the neutralization mechanism of STE90-C11 is based on directly competing for the ACE2 15/57 binding side, as the interaction interface on the RBD of both molecules almost completely overlap. Further superposition of the STE90-C11:SARS-CoV-2 RBD complex model onto cryo-EM models of the whole SARS-CoV-2 spike protein  indicate that due to steric hindrance STE90-C11 like ACE2 is expected to only bind to the open conformation of the spike protein (Supplementary Data 17).

Discussion
For the treatment of COVID-19 but also to protect risk groups, human anti-SARS-CoV-2 antibodies are a promising therapeutic option. Human recombinant antibodies were successfully used for the treatment of other viral diseases. The antibody mAb114 (Ansuvimab-zykl) (Corti et al., 2016) and three-antibody cocktail REGN-EB3 (Atoltivimab/maftivimab/odesivimab) (Pascal et al., 2018) were FDA approved in 2020 and showed a good efficiency in clinical trials against Ebola virus especially in comparison to Remdesivir (Mulangu et al., 2019). For the treatment of a severe respiratory infection of infants caused by the respiratory syncytial virus (RSV), the antibody Palivizumab is approved by EMA and FDA (van Mechelen et al., 2016;Subramanian et al., 1998). These anti-viral antibodies can be used as blueprint to develop therapeutic antibodies against SARS-CoV-2. Currently, monoclonal antibodies against SARS-CoV-2 were selected by rescreening memory B-cells from a SARS patient (Pinto et al., 2020), selected from COVID-19 patients by single B-cell PCR (Cao et al., 2020;Shi et al., 2020) or FACS sorting (Kreer et al., 2020), selected from transgenic mice followed and from patients by single B-cells FACS sorting (Hansen et al., 2020) or using phage display using universal or patient libraries including different antibody formats (Bertoglio et al., 2021;Chi et al., 2020;Li et al., 2020a;Liu et al., 2020;Ma et al., 2021;Noy-Porat et al., 2020;Parray et al., 2020;Zeng et al., 2020).
In this work, antibody phage display immune libraries were constructed using lymphocytes from six local convalescent COVID-19 patients. 197 unique antibodies were selected against RBD. We intently focused on RBD aiming to select antibodies which directly interfere with the interaction between spike and ACE2, to avoid potential antibody dependent enhancement (ADE), especially in patients with severe symptoms. ADE during coronavirus entry has been described for both MERS (Wan et al., 2020) or SARS (Wang 17/57 et al., 2014). It is defined as "enhancement of disease severity in an infected person or animal when an antibody against a pathogen...worsens its virulence by a mechanism that is shown to be antibody-dependent" (Arvin et al., 2020). Furthermore, immune dysregulation and lung inflammation was also caused by anti-spike antibodies in an acute SARS-CoV infection (Liu et al., 2019). On the other hand, Quinlan et al. (Quinlan et al., 2020) reported that animals immunized with RBD SARS-CoV-2 did not mediate ADE. A therapeutic antibody directed against RBD and the usage of a Fc-part which is not binding to Fc-gamma receptors ("silent Fc") would reduce the risk of ADE, in particular because ADE cannot be fully predicted from in vitro studies or from animal models (Arvin et al., 2020). Winkler and et al. argue, that an intact Fc part is needed for optimal therapeutic protection (Winkler et al., 2021). The ACTIV-3/TICO LY-CoV555 Study Group describes no efficacy of LY-COV-555 in hospitalized patients and discussed also ADE as potential explanation for the lacking efficacy (ACTIV-3/TICO LY-CoV555 Study Group et al., 2021).
Because, the role of ADE is not fully deciphered in case of SARS-CoV-2  and LY-CoV555 was not effective in patients with severe symptoms, it is advised to use a silent Fc part to address hospitalized patients with moderate to severe symptoms. Therefore, all assays with STE90-C11 IgGs were performed with a silenced Fc part.
The selected anti-RBD antibodies were analyzed for RBD::ACE2 inhibition in the scFv-Fc format. The advantage of the IgG like bivalent scFv-Fc format is that it speeds up analysis by requiring only one cloning step and provides high expression yields in small scale format (Bertoglio et al., 2021;Jäger et al., 2013;Wenzel et al., 2020a). 30 scFv-Fc antibodies inhibited the binding of the spike protein to ACE2-expressing living cells.
Interestingly, the majority of the VH V-genes of the selected anti-RBD antibodies and 28 of these 30 inhibiting antibodies is the VH3-66 V-gene showing strong enrichment of this Vgene in the immune response of COVID-19 patients against RBD. This is in accordance 18/57 with Cao et al. (Cao et al., 2020) who found that enriched use of VH3-53 or VH3-66. VH3-53 and VH3-66 V-genes are closely related. Robbiani et al. (Robbiani et al., 2020) showed an enrichment of VH3-30, VH3-53 and VH3-66 and Hansen et al. (Hansen et al., 2020) reported a bias towards VH3-53. The VH3-53 and VH3-66 V-genes are closely related. In contrast, Ju et al. (Ju et al., 2020) identified mainly VH1-2, VH3-48 and VH3-9 V-genes for their selected anti-RBD antibodies from one patient and Brouwer et al. (Brouwer et al., 2020) identified a strong enrichment of VH1-69, VH3-30 and VH1-24 in three patients, but also an enrichment of VH3-53 and VH3-66. VH3-66 antibodies against RBD were also selected from the naive HAL9/10 antibody gene libraries made long before the SARS-CoV-2 outbreak (Kügler et al., 2015) but not significantly enriched (Bertoglio et al., 2021). The selected inhibiting antibodies in our study (Table 1) all have a high germinality index and are therefore very similar to the not hypermutated human germline, promising low immunogenicity when used as a therapeutic agent. Only some light chains showed a major difference to their closest related V-gene. Kreye et al. (Kreye et al., 2020) also described that antibodies they selected from COVID-19 patients were very close to the germline genes. Interestingly, the VH V-genes of these immune libraries are closer to their germline compared to the anti-RBD antibodies selected from the naive HAL9/10 antibody gene libraries (Bertoglio et al., 2021).
In the cell-based inhibition assay, some of the antibodies showed a molar antibody:spike monomer ratio lower then 1:1. Similar behaviour was also found by Bertoglio et al.  and may be explained by the observation that three RBDs on the same spike trimer can be in different conformations ("up" and "down"), while only the RBDs in "up" positions are able to bind ACE2  and therefore, the "down" RBDs are not fully accessible.

19/57
The monovalent affinity of STE90-C11 was measured with 1.6 nM-8.1 nM depending on the BLI setup. This is in the same range as BD-368-2 with 0.82 nM affinity to RBD (Cao et al., 2020), 3.3 nM for REGN10933 in the setup with the monovalent RBD (Hansen et al., 2020). 70 nM for B38  or 0.8-4.7 nM for CB6 .
The analysis of the binding STE90-C11 to other human coronaviruses showed specificity for SARS-CoV-2. Comparison between the epitope of STE90-C11 and other published antibodies ( Figure 5F) revealed that it binds in the same region as CB6  and B38 . The selectivity of STE90-C11 and CB6 for SARS-CoV-2 over SARS-CoV RBD can mainly be attributed to the interaction with residues 473-476, as they are not involved in ACE2 binding and SARS-CoV RBD folds into a different conformation of this loop causing steric hindrance for antibody binding (Lan et al., 2020).
In addition to the in vitro neutralization of authentic SARS-CoV-2, the in vivo efficacy of STE90-C11 was demonstrated in the Syrian Hamster Challenge model (Kreye et al., 2020) and in the model using transgenic mice with human ACE2 receptor  in a dose dependent manner.
The binding of STE90-C11 to S1 containing RBD mutations observed in strains from COVID-19 patients was analyzed, showing a binding to most variants. As the cell-based inhibition analysis demonstrate, STE90-C11 is able to inhibit most analyzed mutants, validating a tolerance for RBD mutations. REGN10933 (Hansen et al., 2020), which is also binding at the RBD-ACE2 interface, showed a loss of neutralization in an assay using pseudoviral particles for the F486V and a reduced neutralization for both G485D and E484K mutations . Loss of binding to F486A is also described for VH-Fc ab8 . In the epitope analysis, REGN10933 has more molecular interactions within the region aa483-aa486 compared to STE90-C11. Currently, the variant B.1.1.7 is widely spread and STE90-C11 has reduced binding to N501Y and lost binding to 20/57 mutants with the K417N/T mutations in combination with N501Y which occur in B.1.351 and B.1.1.28.1 variants. The structure suggest that the exchange N501Y may push the light chain CDR1-loop of the antibody further away while the exchange of K417N/T leads to an abolishment of a salt bridge to D101 of the heavy chain, similar to observations done for the COVOX-269 and COVOX-222, respectively Supasa et al., 2021). In COVOX-269 binding to the RBD-N501Y leads to a conformational change of light chain CDR3 loop around Y94. In the escape mutant experiment, no N501x mutants were enriched, indicating a sufficient neutralization of the N501x but not K417x mutation. In summary, the patient-derived SARS-CoV-2 neutralizing anti-RBD antibody STE90-C11 binding to the RBD-ACE2 interface maintains high similarity to the human germline Vgenes VH3-66, the same family of many isolated anti-SARS-CoV-2 neutralizing antibodies.
STE90-C11 is tolerant to most known RBD mutants especially those of the mutants 21/57

Resource Availability
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Michael Hust (m.hust@tu-bs.de)

Materials Availability
All requests for resources and reagents should be directed to the Lead Contact author.
This includes antibodies, plasmids and proteins. All reagents will be made available on request after completion of a Material Transfer Agreement.

Data and Code Availability
The structure data and the antibody sequence is available at PDB 7B3O.

SARS-CoV-2-infected individuals and sample collection
For the generation of human immune antibody libraries against SARS-CoV-2, blood from

Production of antigens in insect cells
The antigens were produced and purified as described before (Bertoglio et al., 2021;Korn et al., 2020). In brief, different domains or subunits of the Spike protein (GenBank:

Protein purification
Protein purification was performed as described before (Bertoglio et al., 2021) 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 Profinia System (BIO-RAD

COVID-19 convalescent patient library construction
For the generation of human immune antibody libraries against SARS-CoV-2, blood from 6 donors showing a good antibody titer against RBD was used. The library construction was performed as described previously with minor modifications (Kügler et al., 2018). In brief, the peripheral blood mononuclear cells (PBMC) were extracted from the blood via Ficoll RNA was converted into cDNA using Superscript IV (Thermo Fisher Scientific, Waltham, USA) according to the manufacturer's instructions. Library construction was achieved as described previously (Wenzel et al., 2020a) with slight modifications. In brief, specific primers for the VH chain, kappa light chain and lambda light chain were used to amplify antibody genes from the cDNA. These resulting PCR products were purified and again amplified with primers adding specific restriction sites for further cloning into the E. coli expression vector pHAL52. The vector pHAL52 is derived from the vector pHAL30 (Kügler et al., 2015) with an AscI restriction site in the VL stuffer and a SalI restriction site in the VH stuffer for removal of uncut vector backbone. First, the light chain was cloned between the restriction sites MluI-HF and NotI-HF. In a second step the VH chain was cloned with the restriction sites HindIII-HF and NcoI-HF into previously cloned pHAL52-VL, additionally digested with AscI. The efficiency of library cloning was tested by colony PCR and the rate of complete scFv insertion was determined. The libraries were packaged with Hyperphage (Rondot et al., 2001;Soltes et al., 2007). Antibody phage were precipitated with PEG-NaCl and resuspended in phage dilution buffer (10 mM TrisHCl pH7,5, 20 mM NaCl, 2 mM EDTA). Resulting phage titer was determined by infection of E. coli XL1 blue MRF'.

Antibody selection using phage display
Antibody selection was performed as described previously with modifications (Russo et al., 2018). In brief, 5 µg of of S1-S2-His or RBD-His (produced in High Five cells) was diluted in carbonate buffer (50 mM NaHCO3/Na2CO3, pH 9.6) and coated onto the wells of a High 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 1x10 10 cfu M13KO7 helper phage. Subsequently, the infected bacteria were incubated 30 min at 37°C followed by 30 min at 37°C and 650 rpm before centrifugation for 10 min at 3220xg. 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 (20x in panning round 2, 30x in panning round 3) and the amount of antigen was reduced (2.5 µg in panning round 2, 1.5 µg in panning round 3). After third panning round single clones were analyzed for production of RBD specific scFv by 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 (Russo et al., 2018;Wenzel et al., 2020a). (30%)) were mixed). After stopping the reaction by addition of 1 N H2SO4, absorbance at 450 nm with a 620 nm reference was measured in an ELISA plate reader (Epoch, BioTek).
Monoclonal binders were sequenced and analyzed using VBASE2 (www.vbase2.org) (Mollova et al., 2010) and possible glycosylation positions in the CDRS were analyzed according to Lu et al (Lu et al., 2019).

ScFv-Fc and IgG production
Unique scFv sequences isolated by antibody-phage display were subcloned into pCSE2.7-hIgG1-Fc-XP using NcoI/NotI (New England Biolabs, Frankfurt, Germany) for mammalian 31/57 production in Expi293F cells as scFv-Fc (Wenzel et al., 2020a). For IgG production, the variable domains were recloned into the IgG vectors human IgG1 format by subcloning of VH in the vector pCSEH1c (heavy chain) and VL in the vector pCSL3l/pCSL3k (light chain lambda/kappa) (Steinwand et al., 2014) adapted for Golden Gate Assembly procedure with Esp3I restriction enzyme (New England Biolabs, Frankfurt, Germany). For the antibodies CB6 (pdb 7C01), CR3022 (pdb 6W41) and REGN10933 and REGN10987 (pdb 6XDG) the public available amino acid sequences were used and the V-Genes were ordered as GeneArt Strings DNA fragments (Thermo Fisher Scientific, Schwerte, Germany) and recloned in the above indicated vectors. A "silenced" Fc part with following point mutations described Armour et al (Armour et al., 1999) and Shields et al (Shields et al., 2001) were Inhibition of S1-S2 and RBD binding to ACE2 expressing cells using flow cytometry

32/57
The inhibition tests in flow cytometry on EXPI293F cells were performed based on a previously published protocol . Briefly, Expi293F cells were transfected according to the protocol above using pCSE2.5-ACE2fl-His and 5% eGFP plasmid.
EC50 were calculated with by GraphPad Prism Version 6.1, fitting to a four-parameter logistic curve. Titration ELISAs on other coronaviruses and S1-HIS mutants were performed as described above.
For titration, antibodies were diluted in 1/√10 steps and mixed with a fixed inoculum of SARS-CoV-2/Münster/FI110320/1/2020 (kind gift of Stephan Ludwig, University of Münster, Germany) (10-20, respectively 100-150 pfu) in a total volume of 500 µl of Vero E6 medium (DMEM, 10% FCS, 2 mM glutamine, penicillin, streptomycin). After one hour incubation at 37°C, cells were infected with the antibody/virus mix, incubated for one hour and then overlaid with Vero E6 medium containing 1.5% methyl-cellulose. Three days postinfection, wells were imaged using a Sartorius IncuCyte S3 (4x objective, whole-well scan) and plaques were counted from these images. Image data was quantified with the IncuCyte S3 GUI tools measuring the decrease of confluence concomitant with the cytopathic effect of the virus in relation to uninfected controls and controls without antibody and analyzed with Origin using the Logistic5 fit.

Specificity assay
To test specificity of the antibody candidates an ELISA on DNA, LPS, lysozyme and cell lysate was performed under standard conditions (see above). In brief, 10 µg/mL of the 34/57 respective antigen was immobilized on 96 well microtiter plates (High binding, Costar) in PBS (pH 7.4) overnight at 4°C. After blocking 10 µg/mL of STE90-C11 IgG, Avelumab, Palivizumab and IVIG respectively were incubated and later detected using goat-anti-hIgG(Fc)-HRP (1:70000, A0170, Sigma). TMB reaction took place for 30 min and absorbance at 450 nm with a 620 nm reference was measured in an ELISA plate reader (Epoch, BioTek). All signals were normalized to the absorbance of Avelumab.

Analytical size exclusion chromatography (SEC)
All purified antigens and indicated antibodies were run on Superdex 200 Increase 10/300GL column (Cytiva) on Äkta pure system (Cytiva) according to the manufactures protocol.

Affinity measurement by Bio-Layer Interferometry
The affinity was measured by Bio-Layer Interferometry in three different assays using the Octet qKe (Fortebio/Sartorius GmbH, Göttingen, Germany).
In the first assay, anti-Mouse Fc-Capture (AMC) sensors were activated for 10 min in PBS.
After that, the sensors were equilibrated in assay buffer (PBS containing 1% BSA and 0.05% Tween 20) for 60 s before RBD-mFc (Sino Biologicals) was loaded onto the sensors at 10 µg/ml for 180 s. After a stable baseline measurement was established (60 s), antigen-loaded sensors were transferred to an 8-point antibody dilution series (500, 150, 50, 15, 5, 1.5, 0.5 and 0 nM). Association of the Fab antibody to the antigen was measured for 300 s. After that, the sensors were transferred into assay buffer were the dissociation was measured for 900 s. Significant binding of the antibody to an unloaded sensor was not detected. For data analysis, the reference measurement (0 nM) was 35/57 subtracted from the other measurements and data traces ranging from 150 to 1.5 nM were used for modeling of the kinetic data using a 1:1 binding model.
In the second assay, anti-human Fab (FAB2G) sensors were activated for 10 min in PBS.
After that, the sensors were equilibrated in assay buffer (PBS containing 1% BSA and 0.05% Tween 20) for 60 s before the IgG antibody was loaded onto the sensors at 2.5 µg/ml for 180 s. After a stable baseline measurement was established (60 s), antibodyloaded sensors were transferred to an 8-point S1-HIS antigen dilution series (500, 150, 50, 15, 5, 1.5, 0.5 and 0 nM). Association of the S1 antigen to the antibody was measured for 300 s. After that, the sensors were transferred into assay buffer were the dissociation was measured for 900 s. Significant binding of the antigen to an unloaded sensor was not detected. For data analysis, the reference measurement (0 nM) was subtracted from the other measurements and data traces ranging from 50 to 5 nM were used for modelling of the kinetic data using a 1:1 binding model in the Data Analysis HT 11.0 software tool.
In the third assay, protein A sensors were activated for 10 min in PBS. Before use, the sensors were regenerated for 5 cycles in 10 mM Glycine buffer (pH2.0) followed by neutralization in PBS. Each step was performed for 5 s. After that, the regenerated sensors were equilibrated in assay buffer (PBS containing 1% BSA and 0.05% Tween 20) for 60 s before the IgG antibody was loaded onto the sensors at 2.5 µg/ml for 180 s. After a stable baseline measurement was established (60 s), antibody-loaded sensors were transferred to an 8-point S1-HIS antigen dilution series (500, 150, 50, 15, 5, 1.5, 0.5 and 0 nM). Association of the S1 antigen to the antibody was measured for 300 s. After that, the sensors were transferred into assay buffer were the dissociation was measured for 900 s.
Significant binding of the antigen to an unloaded sensor was not detected. For data analysis, the reference measurement (0 nM) was subtracted from the other measurements 36/57 and data traces ranging from 50 to 0.5 nM were used for modeling of the kinetic data using a 1:1 binding model.

Immunoblot analysis
For the immunoblot analysis RBD-His was disrupted either by incubation for 10 minutes at 56°C without β-Mercaptoethanol, at 95°C without β-Mercaptoethanol or at 95°C in Laemmli sample buffer (Laemmli, 1970)
Briefly, anaesthetized hamsters received 1x105 pfu SARS-CoV-2 in 60 µL MEM intranasally two hours before treatment. Following infection, the clinical presentation of all animals was monitored twice a day, body weight of all hamsters was recorded daily. On days 3, 5 and 14 post infection, three randomly assigned hamsters per group were euthanized. Euthanasia was applied by exsanguination under general anaesthesia as described (Nakamura et al., 2017). Oropharyngeal swabs and lungs were collected for virus titrations, RT-qPCR and/or histopathological examinations. All organs were immediately frozen at -80 °C or preserved in 4% formaldehyde for subsequent in-depth histopathological investigations.
To assess virus titers from 50 mg lung tissue, tissue homogenates were prepared using a bead mill (Analytic Jena) and 10-fold serial dilutions were prepared in MEM, and plated on 38/57 Vero E6 cells in 12-well-plates. The dilutions were removed after 2 h and cells were overlaid with 1.25% microcrystalline cellulose (Avicel) in MEM supplemented with 10% FBS and penicillin/streptomycin. Three days later, cells were formalin-fixed, stained with crystal violet, and plaques were counted.

Transgenic mice model of SARS-CoV-2 infection
All animal experiments were performed in compliance with the German Animal Welfare Act (TierSchG BGBl. I S. 1206, 1313May 18, 2006) and The mice were sacrificed by CO2 asphyxiation on day 5. Lungs were collected aseptically, homogenized in 500 µL PBS and stored at −80•C. Part of the organ homogenates were used for titration cells and the other part for qPCR Analysis. Organ homogenates were 39/57 serially diluted 1:10 -1:10 5 in medium (DMEM supplemented with 5% FCS, 2 mM glutamine, 100 IU/mL penicillin and 100 µg/mL streptomycin). Vero E-6 cells were then inoculated with 200ul of diluted homogenates and incubated for 1h, 37°C, CO2 incubator.
Cells were then coverd with 1.75% Carboxymethyl-cellulose and incubated at 37°C, CO2 incubator for 3-5 days. Plates were then fixed with 6% Paraformaldehyde for 1h and then stained with 1% Crystal violet. Plaques were then read under microscope.

In vitro evolution of SARS-CoV-2 by antibody co-cultivation
This assay was performed with STE90-C11 and Palivizumab as described by Baum et al. .

Fab and RBD22 production for co-crystallization
The production of STE90-C11 Fab fragment was done by transient co-transfection of plasmids encoding the heavy and the light chain in Expi293F cells cultivated at 37°C, 5% CO2 and 100rpm in Expi Expression Medium. The transfection was performed at a density of 3*10^6 cells/ml by adding 1µg/ml culture mixed plasmids and 4 µg/ml culture PEI 40 kDa (Polyscience). The culture was incubated for 72 hours according to the protocol. The supernatant was harvested by centrifugation (30min, 3000g) and sterile filtration (0.2µm).
The Fab-fragment was purified by affinity chromatography using 1ml HisTrap Excel column (GE Healthcare) followed by a size exclusion chromatography on a 10/300 Superdex200 Pilatus 6M fast detector (Dectris) and processed with XDS (Kabsch, 2010) and Aimless (Evans and Murshudov, 2013) yielding a dataset with a resolution cut off of 2.0 Å based on a CC1/2 value greater than 0.5. Initial phases were determined by molecular replacement with Phaser (McCoy et al., 2007). As a search model the coordinates of a Fab fragment and the RBD was used (PDB: 7BWJ) (Ju et al., 2020). The model was further improved by manual rebuilding in Coot (Emsley et al., 2010) and computational refinement with phenix.
refine (Afonine et al., 2012)   (fusion protein with murine Fc part), (B) S1 (fusion protein with murine Fc part) or (C) S1-S2 (fusion protein with His tag). An unrelated antibody with murine Fc part (TUN219-2C1), human HEK293 cell lysate, BSA and lysozyme were used as controls. Experiments were performed in duplicate and mean ± s.e.m. are given. EC50 were calculated with GraphPad Prism Version 6.1, fitting to a four-parameter logistic curve. pfu SARS-CoV-2 to determine the IC50. Neutralization assays were performed in triplicates, mean ± s.e.m. are given. Palivizumab was used as isotype control in both assays. IC50 were calculated with OriginPro using the Logistic5 Fit. ELISAs are given in Supplementary Data 8. n.d., not determinable. S1-7PM contains the seven mutations V367F, N439K, G476, V483A, E484K, G485R and F486V.  Depicted is a part of the sequence alignment of the SARS-CoV-2 and SARS-CoV Sprotein (UniProt: P0DTC2; P59594). Identical residues are depicted as bold blue letters.