A neutralizing human antibody binds to the N-terminal domain of the Spike protein of SARS-CoV-2

Hitting SARS-CoV-2 in a new spot A key target for therapeutic antibodies against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is the spike protein, a trimeric protein complex with each monomer comprising an S1 and an S2 domain that mediate binding to host cells and membrane fusion, respectively. In addition to the receptor binding domain (RBD), S1 has an N-terminal domain (NTD). In searching for neutralizing antibodies, there has been a focus on the RBD. Chi et al. isolated antibodies from 10 convalescent patients and identified an antibody that potently neutralizes the virus but does not bind the RBD. Cryo–electron microscopy revealed the epitope as the NTD. This NTD-targeting antibody may be useful to combine with RBD-targeting antibodies in therapeutic cocktails. Science, this issue p. 650

as the function of NTD is not well understood. In some coronaviruses, the NTD may recognize specific sugar moieties upon initial attachment and might play an important role in the prefusion-to-postfusion transition of the S protein (23)(24)(25)(26). The NTD of the MERS-CoV S protein can serve as a critical epitope for neutralizing antibodies (26).
The SARS-CoV-2 S protein-targeting monoclonal antibodies (mAbs) with potent neutralizing activity are a focus in the development of therapeutic interventions for COVID-19 (27)(28)(29). Many studies reported the functions and structures of SARS-CoV-2-neutralizing antibodies that target the RBD and inhibit the association between the S protein and ACE2 (28)(29)(30)(31)(32)(33)(34). The RBD-targeting antibodies, applied individually, might induce resistance mutations in the virus (26). Antibodies that target non-RBD epitopes might be added to antibody cocktail therapeutics for SARS-CoV-2. We thus sought to identify antibodies to different regions of the S protein and to the Nucleocapsid (N) protein.

Isolation of human mAbs from memory B cells and plasma B cells
To isolate mAbs and analyze the humoral antibody responses to SARS-CoV-2, we collected plasma and peripheral blood mononuclear cells (PBMCs) from 10 Chinese patients who had recovered from SARS-CoV-2 infection. The age of donors ranges from 25 to 53 years. The interval from disease confirmation date to blood collection date ranged from 23 to 29 days for patients 1 to 5 and 10 to 15 days for patients 6 to 10 (table S1). We evaluated the titers of binding antibodies in plasma to different fragments of the SARS-CoV-2 S proteinincluding the full ECD, S1, S2, and the RBDand to the N protein. Plasma from all the patients except donor 2 bound to all five SARS-CoV-2 protein segments, whereas that from donor 2 recognized S-ECD and S2 only (Fig. 1A). The neutralizing capacities of plasma against authentic SARS-CoV-2 and HIV-vectored pseudotyped SARS-CoV-2 are correlated [correlation coefficient (r) = 0.6868, P < 0.05] (Fig. 1B). These results indicate that humoral immune responses were specifically elicited for all 10 patients during their natural infection with SARS-CoV-2.
To isolate S protein-specific mAbs, we first sorted the immunoglobulin G-positive (IgG + ) memory B cells from PBMCs of convalescent patients 1 to 5 with flow cytometry, using S-ECD as the probe (Fig. 1C). The percentage of S-ECDreactive IgG + B cells ranges from 0.56 to 11%, as revealed with fluorescence activating cell sorting (FACS). To avoid losing B cells with low copies of S-ECD-specific receptors on cell surfaces, we sorted plasma B cells from mixed PBMCs derived from another five convalescent patients (patients 6 to 10) without using S-ECD protein as the probe in flow cytometry. The percentage of plasma B cells in CD3-CD19 + B cells was 12.8%, which is higher than the percentage of memory B cells in CD3-CD19 + B cells (Fig. 1C).
From the sorted B cells, we identified 9, 286, 43, 12, and 26 clones of single B cell from patients 1 to 5, respectively, and 23 clones of single B cell from the mixed PBMCs of patients 6 to 10 (Fig. 1D). The distribution of the sequenced heavy (IgH) gene families was comparable among the 10 donors, with VH3 being the most commonly used VH gene, whereas different donors displayed variable preferences for the light chain (IgL) gene families (Fig.  1D). The combination of V3 and J4, V3 and D3, and D3 and J4 were the most common usage for the IgH gene family (fig. S1). The average mutations of amino acids per mAb from memory B cells ranged from 17.50 to 48.04 for donors 1 to 5, respectively, whereas mAbs from plasma B cells possessed an average of 13.99 amino acid mutations for donors 6 to 10 ( Fig. 1E). Human antibodies elicited through repeated exposures to different antigens confer an average of 26.46 amino acid mutations per Ab, as previously reported (35). These results indicate that natural SARS-CoV-2 infection elicited high levels of somatic hypermutation (SHM) in memory B cells. The lengths of complementaritydetermining region 3 (CDR3) for antibodies were similar among the donors, with average lengths of these CDR3 ranging from 13.9 to 17.7 for VH and 9.3 to 10.1 for VL (Fig. 1F). The CDR3 lengths of these mAbs were longer than that in antigen-specific immune receptors (means of 12.7 for VH and 6.5 for VL, respectively) reported previously (36).

Binding profiles of SARS-CoV-2 S proteinspecific human mAbs
To screen for S protein-specific antibodies, we determined the binding specificity using enzyme-linked immunosorbent assay (ELISA) for the 399 human mAbs sorted above. From donors 1 to 5, respectively, 1, 16, 1, 3, and 9 S-ECD-specific mAbs were identified. A total of 35 S-ECDspecific mAbs were identified from donors 6 to 10 ( Fig. 2A). We further characterized domain specificities of the 35 mAbs with different fragments of the S protein, including S1, S2, and RBD ( Fig. 2A). The S-reactive mAbs are classified into four major groups on the basis of their medium effective concentration (EC 50 ) values ( Fig. 2A). Group 1 recognizes only S-ECD. Group 2 recognizes S-ECD and S1, with subgroup 2A binding S-ECD and S1 and subgroup 2B binding S-ECD, S1, and RBD. Group 3 interacts with both S1 and S2, where subgroup 3A targets the RBD and subgroup 3B fails to bind the RBD. Group 4 recognizes S-ECD and S2. Only four mAbs recognize the RBD among the 35 S-specific mAbs (Fig. 2, A and B).
We performed a competition-binding assay using ELISA for several representative mAbs to determine whether there are overlapping antigenic sites between different mAbs, with CR3022 being used as a positive control mAb that reported to bind the SARS-CoV-2 RBD (Fig. 2C) (37). Among these mAbs, 4A8 in group 2A competed with 1M-1D2 in group 2B. Another RBD-reactive mAb, 2M-10B11 in group 2B, competed with CR3022, suggesting overlapped epitopes on RBD for these two mAbs. These results indicate that antibody responses elicited by natural SARS-CoV-2 infection were diverse in epitope recognition of S proteins.
To characterize the diversity in gene usage and affinity maturation, the phylogenetic trees of these S-ECD-specific mAbs were analyzed on the basis of the amino acid sequences of VHDJH and VLJL by using a neighbor-joining method in MEGA7 Software (38). Results indicate that the VH gene usage is very diverse among the 35 mAbs from 10 donors, with VH3-30 being the most frequently used germline gene. There was no particularly favored VH gene identified among S1, S2, or RBDreactive mAbs (Fig. 2D). The percentages of heavy chain variable gene sequence identity ranged from 40.9 to 97.6% in the 35 S-ECDspecific mAbs ( fig. S2 and table S2).
We next performed luciferase reporter gene assays for all 35 S-binding mAbs using HIVvectored pseudotyped SARS-CoV-2 (39), among which three mAbs exhibited neutralizing ac-tivity against the pseudotyped virus (Fig. 3C). 4A8 protected ACE2-293T cells with an EC 50 of 49 mg/ml. Although mAb 2M-10B11 and 9A1 did not neutralize authentic SARS-CoV-2, 2M-10B11 protected against pseudotyped virus with an EC 50 of 170 mg/ml, and 9A1 provided weak protection. To our surprise, neutralization by 0304-3H3 and 1M-1D2 was not observed (Fig.  3C). The inconsistency between the results for pseudotyped SARS-CoV-2 compared with authentic SARS-CoV-2 were also observed for mAbs against MERS-CoV (40, 41) and may be caused by the different presentation of S protein resulted from the different environmental factors the viruses underwent, such as the cells used for the neutralizing assays or for the production of the pseudotyped or authentic virions (42). On the basis of these results, 4A8 is a potential candidate for the treatment of SARS-CoV-2 because it displayed strong neutralizing capacities against both authentic and pseudotyped SARS-CoV-2.

Binding characterization of candidate mAbs
To determine the possible neutralizing mechanism of the mAbs, we determined the binding affinities of the five mAbs with potential neutralizing activity against different segments of the S protein-including the full S-ECD and domains S1, S2, and RBD-using biolayer interferometry (BLI). All five tested mAbs bound to S-ECD with high affinity; equilibrium dissociation constants (K d ) were less than 2.14 nM (Fig. 4A). 4A8 and 1M-1D2 bound to S1 with K d of 92.7 and 108 nM, respectively, whereas 0304-3H3 and 9A1 targeted S2 with K d of 4.52 and <0.001 nM, respectively   ( Fig. 4A, bottom). Moreover, 2M-10B11 bound the RBD with K d of 24.3 nM, which was obtained by using heterogeneous ligand model owing to the avidity effects (Fig. 4A, bottom).
To investigate whether these mAbs block the binding of S protein to ACE2, we performed flow cytometry using human embryonic kidney (HEK) 293T cells expressing human ACE2. As expected, only 2M-10B11 among the five mAbs and ACE2-Fc prevented S protein from binding to ACE2. In the presence of 2M-10B11, only 0.52% of cells were double positive for IgG and S protein (Fig. 4B). CR3022, which competes with 2M-10B11, did not block the binding of S to ACE2. The control mAb 1A8, targeting the Marburg glycoprotein, did not interfere with the binding either, and the 5.13% of double positives may be due to the nonspecific binding of 1A8 to S protein. 4A8 also failed to interfere with the binding of the S protein to ACE2.

Cryo-EM structure of the complex between 4A8 and S-ECD
The mAb 4A8 was overexpressed and purified by Protein A resin, and the S-ECD of SARS-CoV-2 was purified through M2 affinity resin and size exclusion chromatography (SEC). 4A8 and S-ECD protein were mixed and incubated at a stoichiometric ratio of~1.2 to 1 for 1 hour and applied to SEC to remove excess proteins ( fig. S3A). The fraction containing the complex was concentrated for cryo-electron microscopy (cryo-EM) sample preparation.
To investigate the interactions between 4A8 and the S protein, we solved the cryo-EM structure of the complex at an overall resolution of 3.1 Å (Fig. 5 and movie S1). Details of cryo-EM sample preparation, data collection and processing, and model building can be found in in the supplementary materials, materials and methods (figs. S3 to S5). The S protein exhibits asymmetric conformations similar to the previously reported structures (21,22), with one of three RBDs in "up" conformation and the other two RBDs in "down" conformation (Fig. 5).

Recognition of the NTD by 4A8
In the S protein-4A8 complex, each trimeric S protein is bound with three solved 4A8 Fabs, each of which interacts with one NTD of the S protein. Despite the different conformations of the three S protein protomers, the interface between 4A8 and each NTD is identical (Fig. 5 and fig. S3I). The map quality at the NTD-4A8 region was improved through focused refinement to a local resolution of 3.3 Å, enabling reliable analysis of the interactions between the NTD and 4A8.
Association with 4A8 appears to stabilize the NTD epitope, which is invisible in the reported S protein structure alone (21,22). Supported by the high resolution of NTD, we were able to build the structural model for five new loops for NTD, designated N1 (residues 14 to 26), N2 (residues 67 to 79), N3 (residues 141 to 156), N4 (residues 177 to 186), and N5 (residues 246 to 260), among which the N3 and N5 loops mediate the interaction with 4A8 ( fig. S5A). Besides, three new glycosylation sites (Asn 17 , Asn 61 , and Asn 149 ) on the NTD are identified in this structure ( fig. S6).
The heavy chain of 4A8 mainly participates in binding to the NTD mainly through three complementarity-determining regions (CDRs), named CDR1 (residues 25 to 32), CDR2 (residues 51 to 58), and CDR3 (residues 100 to 116) ( Fig. 6A and fig. S5B). The interface is constituted by an extensive hydrophilic interaction network, and the buried surface area at the 4A8-NTD interface is 832 Å 2 . Arg 246 on the N5 loop of the NTD represents one docking site, which is stabilized by Trp 258 , simultaneously interacting with Tyr 27 and Glu 31 of 4A8 on CDR1 (Fig. 6B). On the N3 loop of the NTD, Lys 150 and Lys 147 respectively form salt bridges with Glu 54 and Glu 72 of 4A8 (Fig. 6C). Lys 150 is also hydrogen (H)-bonded with 4A8-Tyr 111 , while His 146 forms a H-bond with 4A8-Thr 30 (Fig. 6C). In addition to the hydrophilic interactions, Trp 152 and Tyr 145 on the N3 loop of the NTD also interact with Val 102 , Pro 106 , and Phe 109 on the CDR3 of 4A8 through hydrophobic and/or p-p interactions (Fig. 6D)  the NTD is close to the 4A8-NTD interface, of which N-glycans might participate in the interactions on the interface ( Fig. 6A and fig. S6).

Discussion
There is an urgent need for prophylactic and therapeutic interventions for SARS-CoV-2 infections given the ongoing COVID-19 pandemic. Our work reveals that naturally occurring human SARS-CoV-2 mAbs isolated from the B cells of 10 recovered donors are diverse in gene usage and epitope recognition of S protein. The majority of the isolated mAbs did not recognize the RBD, and all the mAbs that neutralize authentic SARS-CoV-2 failed to inhibit the binding of S protein to ACE2. These unexpected results suggest the presence of other important mechanisms for SARS-CoV-2 neutralization in addition to suppressing the viral interaction with the receptor. The S1-targeting mAb 4A8 does not block the interaction between ACE2 and S protein but exhibits high levels of neutralization against both authentic and pseudotyped SARS-CoV-2 in vitro. Many neutralizing antibodies against SARS-CoV-2 were reported to target the RBD of the S protein and block the binding between RBD and ACE2 (28)(29)(30)(32)(33)(34). Our results show that 4A8 binds to the NTD of S protein with potent neutralizing activity. Previous study has shown that mAb 7D10 could bind to the NTD of S protein of MERS-CoV probably by inhibiting the RBD-DPP4 binding and the prefusion-to-postfusion conformational change of S protein (26). We aligned the crystal structure of 7D10 in complex with the NTD of S protein of MERS-CoV with our complex structure and found that the interfaces between the mAb and the NTDs are partially overlapped ( fig. S7). 7D10 may inhibit the interaction between MERS-CoV and DPP4 through its light chain, which is close to the RBD. In our complex, the light chain of 4A8 is away from the RBD (fig. S7). Therefore, we speculate that 4A8 may neutralize SARS-CoV-2 by restraining the conformational changes of the S protein. Furthermore, sequence alignment of the S proteins from SARS-CoV-2, SARS-CoV, and MERS-CoV revealed varied NTD surface sequences that are respectively recognized by different mAbs (fig. S8).
This work reports a fully human neutralizing mAb recognizing a vulnerable epitope of NTD on S protein of SARS-CoV-2, functioning with a mechanism that is independent of receptor binding inhibition. Combination of 4A8 with RBD-targeting antibodies may avoid the escaping mutations of the virus and serve as promising "cocktail" therapeutics. The information obtained from these studies can be  used for development of the structure-based vaccine design against SARS-CoV-2.