Humanized single domain antibodies neutralize SARS-CoV-2 by targeting the spike receptor binding domain

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spreads worldwide and leads to an unprecedented medical burden and lives lost. Neutralizing antibodies provide efficient blockade for viral infection and are a promising category of biological therapies. Here, using SARS-CoV-2 spike receptor-binding domain (RBD) as a bait, we generate a panel of humanized single domain antibodies (sdAbs) from a synthetic library. These sdAbs reveal binding kinetics with the equilibrium dissociation constant (KD) of 0.99–35.5 nM. The monomeric sdAbs show half maximal neutralization concentration (EC50) of 0.0009–0.07 µg/mL and 0.13–0.51 µg/mL against SARS-CoV-2 pseudotypes, and authentic SARS-CoV-2, respectively. Competitive ligand-binding experiments suggest that the sdAbs either completely block or significantly inhibit the association between SARS-CoV-2 RBD and viral entry receptor ACE2. Fusion of the human IgG1 Fc to sdAbs improve their neutralization activity by up to ten times. These results support neutralizing sdAbs as a potential alternative for antiviral therapies.

5. The pseudotype particle generation should be explained in more detail in the material and methods section: "filtered through a syringe filter"? "Vero cells or other cells seeded .." Which other cells, cell density? 6. It is interesting that 3F11 and 5F8, which only partially suppress ACE2-RBD interaction can not completely neutralize the pseudotype particles. The authors should provide an explanation for this finding.
7. The SARS-CoV-2 neutralization assay has to be better described (line 315). How was the virus isolated from the BAL of the patients? Propagated on which cells? How many passages? How many vero cells were seeded? The amount of viral RNA that was present in the cell supernatant was used as a proxy for viral replication. Again essential protocol details are missing. How was RNA extracted, cDNA synthesized, which primer and probes were used, how was standard RNA generated and quantified? Figure 2b, Y-axis: percentage of what? 8. Figure 4. A Coomassie stained gel should be included to document the purity of the preparations. Line 337: why was a goat anti-mouse IgG antibody used to stain the human Fc fusions in the Western blot? 9. Supplementary figure 2 is not visible.
Reviewer #2 (Remarks to the Author): Xiaojing Chi and colleagues report the isolation and characterization of nanobodies targeting the receptor-binding domain of SARS-CoV-2. This is a very timely and important study but major revisions are needed before it could be published.

Major comments:
The authors describe the selection of RBD binders using a phage display, but they do not provide any data showing the selection/improvement process during the cycles they performed. This data should be included, even if it goes to the supplementary information.
It is not clear what the authors mean by saying that they "maximally humanized" the framework pf the binders. Nanobodies are inherently non-human, and it is not clear how "human-like" they become after this design. The authors will need to provide additional data. Also, the authors mention several times that some residues were mutated, while others were kept as appear in camelids to maintain stability. A major question relates to the actual stability of the designs, as low stability will make them irrelevant for further development. It will be important to evaluate their thermal stabilities and compare them to non-modified nanobodies.
The neutralization/infectivity data that are shown in Figure 2 are poorly presented, and perhaps also inadequately treated. The authors present infectivity data, but there are huge variations in the infectivity levels in low reagent concentrations. A dashed line is shown at "50%", but since some of the reagents plateau at ~60%, this line is meaningless. The authors do not specify how IC50 values were calculated, and thus the validity of these values is questionable. To make sense, the data should be inversed to show "neutralization" rather than "infectivity". Also, each neutralization data set should be normalized by the "background" (no virus) control to define the real value for 100% neutralization. Then, a 4-parameters neutralization model should be fitted and shown with the experimental data points. When fitting this model, it will be acceptable to force it to reach 0% neutralization at the lowest concentration of the reagents. From this model, the IC50, as well as IC90 values, should be reported. There is a good chance that the reported IC50 values will change following this treatment.
For all the SPR data shown in the manuscript, the authors need to present along with the experimental data (overlaid) the fitted model from which the kinetic parameters were derived. For clarity, it will be convenient to use a two-color scheme. Residual plots (in the supplementary) will also be desired. Lastly, there is a convention for how to present chemical quantities like using italics for the 'K' and subscript for the 'D' in the "KD". The same goes true for the on and off rate parameters.
The authors present the production of Fc-fusion constructs in figure 4 and they report the IC50 and IC80 values for these constructs. It is imperative that the authors will also present the neutralization data (could be in the supplementary data), treated as mentioned above.

Reviewer #1:
General remarks: The authors report on 5 single domain antibodies that can neutralize SARS-CoV-2 S pseudotyped viruses as well as SARS-CoV-2. These VHHs originate from a synthetic camelid sdAb phagemid library in which frame work regions were partially humanized and nucleotide diversity was introduced in CDR1 (length 8 aas), CDR2 (length 8 aas) and CDR3 (length 18 aas). The 5 sdAbs were selected after 4 rounds of panning on immobilized SARS-CoV-2 RBD followed by phage ELISA. The sdAbs bind with variable affinity to SARS-CoV-2 RBD or S and can neutralize lentivirus pseudotypes with an estimated IC 50 value that ranges from 3 to 300 ng/ml. In contrast, all 5 sdAbs neutralized live SARS-CoV-2 with a similar IC 50 (0.24-0.51 ug/ml). The 2 sdAbs that could completely prevent the binding of RBD to immobilized ACE2 performed poorest in the pseudotype neutralization assay. The sdAbs were also genetically fused to a human IgG1 Fc domain, expressed in 293-F cells and purified, and shown to be able to neutralizes pseudotyped lentiviruses, with an IC 50 down to 1.6 ng/ml for 2F2-Fc. The work is of interest because it illustrates the potential of synthetic sdAb libraries as a resource for antiviral molecules that can be rapidly accessed, e.g. in case of a pandemic. A major caveat of the manuscript is the poor description of the synthetic library, the neutralization assays and the experiments in general. The sequences of the 5 sdAbs is not shown and should be provided.

General response: Thank you for your detailed and very helpful comments, all of them have been responded to thoroughly and included in the revision presently submitted.
Please find our point-by-point responses as follows.
2 / 8 Specific remarks: Q #1: "The synthetic sdAb library must be better described. What are the sequences of the FRs? The primers used to introduce diversity in the CDRs? What is the complexity of the library. On page 10 the authors write "More than a thousand agar perti dishes…" How many colonies were obtained in total? "Quality control was carried out by sequencing more than 1000 clones, and the error rate and diversity was calculated".
That is essential information that should be shared, because it documents the diversity of the library, uniqueness of individual clones, lack of in frame stop codons, percentage of phagemids without or with a truncated insert." was performed with RBD (biotinylated or not). However, RBD fused to a mouse Fc or RBD with a His-tag are both mentioned. How was sdAb binding to Fc avoided? After 4 rounds of panning 480 individual colonies were obtained and tested in phage ELISA, from which "representative sdAb sequences were chosen for protein expression". What was the exact set up of the phage ELISA and what was the criterion to select those 5 sdAbs?" The process for the selection of representative sdAb sequences is as follows: sequencing was performed for SARS-CoV-2 RBD positive but mouse Fc negative clones; sequence alignment for excluding repeated clones; determination for distinctive clones for protein expression. Q #7: "The SARS-CoV-2 neutralization assay has to be better described (line 315).
How was the virus isolated from the BAL of the patients? Propagated on which cells? How many passages? How many vero cells were seeded? The amount of viral RNA that was present in the cell supernatant was used as a proxy for viral replication. Again essential protocol details are missing. How was RNA extracted, cDNA synthesized, which primer and probes were used, how was standard RNA generated and quantified? This is a very timely and important study but major revisions are needed before it could be published.

Specific comments:
Q #1: "The authors describe the selection of RBD binders using a phage display, but they do not provide any data showing the selection/improvement process during the cycles they performed. This data should be included, even if it goes to the supplementary information." Q #2: "It is not clear what the authors mean by saying that they "maximally humanized" the framework pf the binders. Nanobodies are inherently non-human, and it is not clear how "human-like" they become after this design. The authors will need to provide additional data. Also, the authors mention several times that some residues were mutated, while others were kept as appear in camelids to maintain stability. A major question relates to the actual stability of the designs, as low stability will make them irrelevant for further development. It will be important to evaluate their thermal stabilities and compare them to non-modified nanobodies." Our response: We agree with you for this concern. To make this clear, detailed methodologies are provided. In addition, the design and sequence of humanized sdAb scaffold are summarized in Supplementary Fig. 1 Q #3: "The neutralization/infectivity data that are shown in Figure 2 are poorly presented, and perhaps also inadequately treated. The authors present infectivity data, 7 / 8 but there are huge variations in the infectivity levels in low reagent concentrations. A dashed line is shown at "50%", but since some of the reagents plateau at ~60%, this line is meaningless. The authors do not specify how IC50 values were calculated, and thus the validity of these values is questionable. To make sense, the data should be inversed to show "neutralization" rather than "infectivity". Also, each neutralization data set should be normalized by the "background" (no virus) control to define the real value for 100% neutralization. Then, a 4-parameters neutralization model should be fitted and shown with the experimental data points. When fitting this model, it will be acceptable to force it to reach 0% neutralization at the lowest concentration of the reagents. From this model, the IC50, aswell as IC90 values, should be reported. There is a good chance that the reported IC50 values will change following this treatment." Our response: Thank you very much for your advice. We have conducted new statistical analysis and data processing according to this recommendation, and the relevant results have been transformed from the inhibition efficiency of the virus to the neutralization efficiency of the virus. Also, IC50 was transformed to EC50 though the values changed slightly after this treatment. In this revised version, you will find Figure   2 and Figure 4 have been updated. At the same time, we also give a more detailed description of the Materials and Methods section. Q #4: "For all the SPR data shown in the manuscript, the authors need to present along with the experimental data (overlaid) the fitted model from which the kinetic parameters were derived. For clarity, it will be convenient to use a two-color scheme.
Residual plots (in the supplementary) will also be desired. Lastly, there is a convention for how to present chemical quantities like using italics for the 'K' and subscript for the 'D' in the "KD". The same goes true for the on and off rate parameters." Our response: We appreciate your suggestions very much. Required revision was made to demonstrate the original experimental data and the fitted model by using black and color scheme overlaid (Figure 1b-1g). In addition, the affinity representation has also 8 / 8 been normalized. Thanks again for your advice. Q #5: "The authors present the production of Fc-fusion constructs in figure 4 and they report the IC50 and IC80 values for these constructs. It is imperative that the authors will also present the neutralization data (could be in the supplementary data), treated as mentioned above." Our response: These results were revised accordingly, and the viral neutralization curves of the Fc-fusion antibodies were also shown in Supplementary Fig. 3.

REVIEWER COMMENTS
Reviewer #1 (Remarks to the Author): The authors have made reasonable efforts to improve the manuscript.
The library is still not very well described. "Sequence analysis of more than 1000 clones, and the error rate was less than 5%". It is important to provide the number of different clones that were represented in this over 1000 clones sample. Were all sequenced clones full length and with a unique sequence?
The clinical sample is not well described. The sequence of the virus was likely determined, which the authors should deposit.
The selection of the 5 clones based on the phage ELISA seems to be more or less random: why were e.g. wells 1.1G, 2.1B, 2.1F, 2.7E, 4.9F, 4.5B, 5.9D, 5.4E and 5.3G not selected for further analysis?
It is unclear why the authors avoid to publish the sequences of their SDAbs, especially since IP has been filed already.
Reviewer #2 (Remarks to the Author): The authors addressed most of my comments but some additional revisions are needed.
Outstanding issues: 1) The SPR data and analysis in Figure 1, should be improved. While for some nanobodies the data and model look OK (i.e. 5F8), for others like 1E2, 2F2, and 5F8 the fitted models do not adequately describe the data. The models were fitted with a refractive index (RI) change when it is clear that the data do not show that. Better models should be provided, perhaps even two-state models if a simple 1:1 binding model cannot be fitted to the data. The authors will need to adjust their reported affinity values accordingly. Figures 1h, 2c, 4d, and throughout the text. It looks like too many significant figures are used to describe some of the values.

2) Another issue is the reported values in
3) Figure 3 is lacking a binding model that should be shown overlaid on the experimental data when a KD is reported. Also, "KD" is not written properly.
4) The ELISA data in table S1 is meaningless to most readers. This data should be presented in a graphical way. Please make separate graphs for each round and perhaps one that demonstrates the round to round improvements.

Specific remarks:
Q #1: "The library is still not very well described. "Sequence analysis of more than 1000 clones, and the error rate was less than 5%". It is important to provide the number of different clones that were represented in this over 1000 clones sample. Were all sequenced clones full length and with a unique sequence?" Our response: We actually picked 1000 clones randomly for sequencing. More than 950 clones are full length and unique sdAbs sequences, and less than 50 clones show various errors, such as vector self-ligation, reading frame shift or fragment deletion. Therefore, we described that the error rate was less than 5% in the previous version.
Revision was made in Method section accordingly in page 10. The isolated SARS-CoV-2 full length sequence was deposited in GISAID with Accession ID of EPI_ISL_402123 (https://platform.gisaid.org/epi3/frontend#57c81d).
GISAID is a globally recognized virus database and more than 56,000 viral genomic sequences of hCoV-19 have been shared via GISAID since the start of the COVID-19 outbreak. I have registered in GISAID. Editors and reviewers are allowed to login GISAID and browse our deposited sequence with the following information (user ID: yangwei, password: qmxVHUKG). In addition, the metadata and FASTA sequence were attached in the end of this point-by-point response to the reviewers. Q #3: "The selection of the 5 clones based on the phage ELISA seems to be more or less random: why were e.g. wells 1.1G, 2.1B, 2.1F, 2.7E, 4.9F, 4.5B, 5.9D, 5.4E and 5.3G not selected for further analysis?" Our response: After phage ELISA, we sequenced all the positive clones (including 1. 1G, 2.1B, 2.1F, 2.7E, 4.9F, 4.5B, 5.9D, 5.4E and 5.3G) and analyzed their sequences.
Most of the positive clones are repeated sequences. Finally, five unique and representative clones with neutralizing activities were selected for further study. They are 1E2, 2F2, 3F11, 4D8 and 5F8. Q #4: "It is unclear why the authors avoid to publish the sequences of their SDAbs, especially since IP has been filed already." Our response: we understand the requirements of the disclosure of antibody sequences.
We are committed to providing antibody sequences and associated reagents to researchers around the world by signing the MTA because, during submission and revision, our antibodies were licensed to a pharmaceutical company for development and optimization, and we cannot disclose the antibody sequence according to the