A potent human monoclonal antibody with pan-neutralizing activities directly dislocates S trimer of SARS-CoV-2 through binding both up and down forms of RBD

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has caused a global pandemic of novel coronavirus disease (COVID-19). The neutralizing monoclonal antibodies (mAbs) targeting the receptor-binding domain (RBD) of SARS-CoV-2 are among the most promising strategies to prevent and treat COVID-19. However, SARS-CoV-2 variants of concern (VOCs) profoundly reduced the efficacies of most of mAbs and vaccines approved for clinical use. Herein, we demonstrated mAb 35B5 efficiently neutralizes both wild-type (WT) SARS-CoV-2 and VOCs, including B.1.617.2 (delta) variant, in vitro and in vivo. Cryo-electron microscopy (cryo-EM) revealed that 35B5 neutralizes SARS-CoV-2 by targeting a unique epitope that avoids the prevailing mutation sites on RBD identified in circulating VOCs, providing the molecular basis for its pan-neutralizing efficacy. The 35B5-binding epitope could also be exploited for the rational design of a universal SARS-CoV-2 vaccine.


INTRODUCTION
The rapid spread of the COVID-19 pandemic has prompted the unprecedented development of anti-SARS-CoV-2 medical countermeasures, among which the most highlighted ones are neutralizing mAbs and vaccines. Indeed, several neutralizing mAbs have been approved under an Emergency Use Authorization for early therapy of COVID-19, including the first COVID-19-authorized mAb bamlanivimab, 1 the REGN-COV2 cocktail (casirivimab and imdevimab), 2 the combined use of bamlanivimab and etesevimab, 3 regdanvimab, 4 and sotrovimab. 5 Additionally, effective vaccines have been developed and globally used, 6,7 including inactivated vaccines, recombinant protein vaccines, adenovirus-based vaccines, and mRNA vaccines.
Though these neutralizing mAbs and vaccines have been developed to contain COVID-19 in the past 2 years, a major concern is the emergence of more transmissible and/or more immune evasive SARS-CoV-2 VOCs, which are antigenically distinct and become dominant in the COVID-19 prevalence over time. 8,9 Indeed, the D614G variant became prevalent in the early phase of the pandemic and was associated with a higher transmission rate. 10 As the thriving pandemic continued, a rapid accumulation of mutations was observed in SARS-CoV-2 and thus seeded the simultaneous appearance of a plethora of VOCs, which include but not limited to B.1.1.7 (UK; alpha variant), 11 B.1.351 (SA; beta variant), 12 P.1 (Brazil; gamma variant), 13 and B.1.617.2 (India; delta variant). 14 In the RBD of SARS-CoV-2 spike protein, B.1.1.7 harbors a N501Y mutation and thus acquires enhanced binding of RBD to the human receptor ACE2. 9,11 Along with the N501Y mutation, B.1.351 and P.1 develop additional K417N/T and E484K mutations. 12,13 Meanwhile, B.1.617.2 carries E484Q/L452R mutations. 14 These mutations contribute to the immune escape of SARS-CoV-2 VOCs against many mAbs, 9,[15][16][17] including those already approved for clinical use (casirivimab, bamlanivimab, regdanvimab). These mutant VOCs also undermine humoral immune response elicited by the WT SARS-CoV-2 infection or vaccines targeting the WT SARS-CoV-2 protein sequence. [16][17][18][19][20][21] Thus, highly potent and broadly neutralizing mAbs targeting multiple SARS-CoV-2 VOCs are urgently needed for emergency use.
In this study, we identified a neutralizing mAb 35B5 that broadly and potently neutralizes WHO-stated SARS-CoV-2 VOCs both in vitro and in vivo. Further cryo-electron microscopy analyses revealed that 35B5 binds to the RBD domain through a unique epitope and disrupts the spike trimer. Collectively, our findings on 35B5 discriminate it from previously identified neutralizing mAbs and highlight its potential application in the prevention and treatment of SARS-CoV-2 VOCs as well as the design of a universal vaccine against SARS-CoV-2 VOCs.

Isolation and characteristics of mAbs 35B5 and 32C7
To discover potent broadly neutralizing mAbs against circulating SARS-CoV-2 VOCs, we adapted a pipeline to rapidly isolate and characterize mAbs ( Supplementary Fig. 1a). Given the vigorous SARS-CoV-2-specific memory B-cell response in individuals recovering from severe COVID-19 illness, 22,23 cryopreserved PBMCs from these convalescent patients with WT SARS-CoV-2 infection were stained for memory B-cell markers (CD19, CD20, and IgG), avidin-tagged biotinylated SARS-CoV-2 RBD antigen bait, and other immune cell lineage markers (CD3, CD14, and CD56). As expected, we found SARS-CoV-2 RBD-specific memory B cells only enriched in PBMCs of convalescent COVID-19 patients, but not healthy donors ( Supplementary Fig. 1b). Each individual of SARS-CoV-2 RBD-specific memory B cells was further sorted to clone heavy-and light-chain pairs for mAb production. Two mAbs potentially with superior neutralization activity, one VH3-9/VK2-28 mAb named as 35B5, characterized by a heavy-chain CDR3 region of 23 residues and a light-chain CDR3 region of 9 residues and another VH3-30/VK4-1 mAb named as 32C7, featured by heavychain CDR3 of 16 residues and light-chain CDR3 of 9 residues ( Supplementary Fig. 1c, d), were cloned, expressed, and analyzed.
Real-time association and dissociation of 35B5 and 32C7 binding to the RBD of SARS-CoV-2 virus were monitored using the surface plasmon resonance (SPR)-based optical assay. We found that 35B5 and 32C7 exhibited different binding affinities in binding to SARS-CoV-2 RBD. Notably, 35B5 exhibited fast-on/slowoff kinetics with an equilibrium dissociation constant (K D ) of 2.19 × 10 −12 M and its binding affinity for SARS-CoV-2 RBD is close to the detection limit at sub-nM level (Fig. 1a); whereas 32C7 showed slow-on/fast-off kinetics with a K D of 1.09 × 10 −8 M (Fig. 1b). Consistently, these binding modes were further evidenced by enzyme-linked immunosorbent assay (ELISA) for SARS-CoV-2 RBD, with a EC 50 value of 0.0183 μg/ml for 35B5 and a EC 50 value of 0.1038 μg/ml for 32C7 ( Supplementary Fig. 1e).
Potent neutralization capacity of mAbs 35B5 to SARS-CoV-2 virus in vitro and in vivo We next investigated the neutralizing capacity of 35B5 and 32C7 against authentic WT SARS-CoV-2 infection in Vero E6 cells. Remarkably, we found that both 35B5 and 32C7 neutralized authentic SARS-CoV-2 virus in the low nanomolar range, with observed IC 50 values of 1.55 ng/ml for 35B5 (Fig. 1c) and 8.60 ng/ ml for 32C7 (Fig. 1d). We further assessed in vivo protection efficacy of 35B5 and 32C7 in the human ACE2 (hACE2)-expressing transgenic mouse model (ICR background) that is sensitized to SARS-CoV-2 infection. 24 The hACE2-humanized mice were treated intraperitoneally with a single dose of 35B5 or 32C7 with 20 mg/kg at 6 h after intranasal infection with 4 × 10 4 PFU of SARS-CoV-2. As control, infected mice of the mock group were administrated with an equal volume of PBS (Fig. 1e). At day 5 post infection, the viral loads in the lungs of the mock group surged to~10 4 RNA copies/ ml (Fig. 1f). By contrast, 35B5 and 32C7 treatment remarkably reduced the viral titers, with~10 2 RNA copies/ml resulting 100fold reduction and~10 3 RNA copies/ml resulting tenfold reduction, respectively (Fig. 1f). In addition, we also determined whether 35B5 and 32C7 treatment ameliorated the pathological lung damage in the hACE2 mice infected with SARS-CoV-2. The transcripts of cytokines and chemokines (e.g., Ccl2, Cxcl1, Il1b, Il6, and Tnf), indicative of tissue inflammation, were greatly reduced in both 35B5-and 32C7-treated groups when compared to those observed in the mock group (Fig. 1g). Parallelly, the SARS-CoV-2 virus caused interstitial pneumonia characterized by inflammatory cell infiltration, alveolar septal thickening and distinctive vascular system injury in the mock group, but not in 35B5 or 32C7 treatment groups; additionally, less pronounced lesions of alveolar epithelial cells or focal hemorrhage were observed in the lung sections from mice that received 35B5 or 32C7 treatment than in those from mice of mock group (Fig. 1h). Thus, these results together indicate the potential therapeutic role for these two mAbs, in particular 35B5, in treating COVID-19.
35B5 exhibits ultrapotent and broad neutralization activity to SARS-CoV-2 VOCs in vitro and in vivo We then sought to determine the neutralizing capacity of 35B5 and 32C7 mAbs against SARS-CoV-2 variants. Neutralizing activity of these two mAbs was assessed against authentic SARS-CoV-2 viruses, including WT, D614G, B.  25 were infected with 4 × 10 4 PFU of SARS-CoV-2 D614G or B.1.351 or B.1.617.2 and then treated with a single dose of 35B5 with 30 mg/kg or PBS at 4 h after infection. Viral loads in the lungs of all infected mice were measured at day 5 post infection (Fig. 2e). Indeed, we found that 35B5 treatment resulted in a~10-fold,~100fold and~100-fold reduction of viral titers in the lungs of mice infected with D614G, B.1.351, and B.1.617.2, respectively (Fig. 2f). Moreover, tissue inflammation and interstitial pneumonia caused by SARS-CoV-2 D614G/B.1.351/B.1.617.2 infection were largely ameliorated in 35B5-treated groups compared to that in mock groups ( Fig.  2g-j). Together, these results suggest that 35B5 potently neutralizes a variety of SARS-CoV-2 variants as a broadly neutralizing mAb.
Complex structure of mAb 35B5 with the spike protein To investigate the structural basis of the superior and broad neutralizing activity of 35B5 against the SARS-CoV-2 variants, we employed cryo-EM approaches to determine the complex Fig. 1 35B5 and 32C7 protect against authentic SARS-CoV-2 virus. a, b Affinity analysis of 35B5 (a) and 32C7 (b) binding to immobilized SARS-CoV-2 RBD by using SPR. The experimental data were calculated using a 1:1 binding model. c, d In vitro neutralizing activity of 35B5 (c) and 32C7 (d) against authentic SARS-CoV-2. The mixture of SARS-CoV-2 and serially diluted 35B5 or 32C7 were added to Vero E6 cells. After 48 h, IC 50 values were calculated by fitting the viral RNA copies from serially diluted mAb to a sigmoidal dose-response curve. e Schematic diagram of 35B5 and 32C7 treatment in vivo. Six hours after infection with 4 × 10 4 PFU SARS-CoV-2, the hACE2 mice received a single dose of 35B5 or 32C7 with 20 mg/kg or no mAb treatment (mock). At day 5 post infection, lung tissues were collected for viral burden assessment, cytokine/ chemokine assay and histological analysis. f Viral titers in the lungs were measured by qRT-PCR and presented as RNA copies per milliliter of lung abrasive fluid. g Gene expressions of cytokines and chemokines in the lungs were determined by qPT-PCR. h Histopathological analysis of lung tissues. The data are representative of at least two independent experiments. *P < 0.05, **P < 0.01 and ****P < 0.0001. ns not significant. Error bars in (f) and (g) indicate SD structure of the Fab region of 35B5 with the spike protein of SARS-CoV-2. Incubation of 35B5 Fab with the ectodomain of the spike protein S-2P, 26 a stabilized spike mutant, severely caused the dissociation of the trimer and disrupted its structure quickly in vitro ( Supplementary Fig. 2). 35B5 Fab even disrupted the ectodomain trimer structure of the spike S-HexaPro protein (S-6P), 27 a more stable spike variant containing four additional proline substitutions (F817P, A892P, A899P, and A942P) from S-2P ( Supplementary Fig. 2), suggesting that 35B5 harbors the potent dissociation activity toward the spike protein. Nonetheless, a few of the S-6P particles were still found to maintain the triangular architecture 27,28 after 35B5 Fab treatment for 3 min, which thereby allowed us to carry out cryo-EM analyses of the 35B5 Fab-S-6P complex. We successfully determined the structures of the 35B5 Fab-S-6P complex in three conformational states to the resolutions of 3.7, 3.4, and 3.6 Å, respectively (Supplementary Fig. 3 and Supplementary Table 1).
In the 35B5 Fab-S-6P complex structure of State 1, two RBD domains of the S-6P trimer are in the standard "up" conformations 26 and are bound by 35B5 Fabs (Fig. 3a). The other RBD domain is in the "down" conformation 26 as that was found in the Fab-free spike trimers. In the 35B5 Fab-S-6P complex structure of State 2, each of the three RBD domains was in the "up" conformation and bound by a 35B5 Fab (Fig. 3a). In the 35B5 Fab-S-6P complex structure of State 3, although all the three RBD domains were bound by 35B5 Fab, only one RBD domain maintained the "up" conformation. The other two RBD domains were in unprecedented conformations, which we for the first time named as "releasing" conformations ( Fig. 3a). Compared to the "up" RBDs, the two "releasing" RBD domains move out by 6.4 and 23.0 Å, respectively (Fig. 4f). The two "releasing" RBD domains generated large gaps with the adjacent NTD domains in the S-6P trimers (Fig. 3a), suggesting that the spike protein is undergoing structural dissociation.
35B5 targets a unique epitope for pan-neutralizing activity The interactions of 35B5 Fab with the "up" RBDs are identical to those of 35B5 Fab with the "releasing" RBDs in the three states of the 35B5 Fab-S-6P complex. The interface covers a largely buried area of~1029 Å 2 (Fig. 3b, c). The epitope in RBD for 35B5 is composed of 30 interacting residues including R346, F347, N354, A352, K444, Y449, N450, R466, I468, T470, N481, and F490, which form extensive hydrophilic interactions with 35B5 Fab in the structure ( Fig. 3b-f). The corresponding paratope in 35B5 Fab consists of two heavy-chain complementarity determining regions (CDRH2 and CDRH3) and the heavy-chain frameworks (FRH1 and FRH3). The epitope of 35B5 on RBD is distinct from those of the previously identified four classes of neutralizing antibodies to RBD. 29 The 35B5 Fab-binding surface on RBD is located at the site opposite to the receptor ACE2-binding surface, [29][30][31][32] which is targeted by the class 1 antibodies, suggesting that 35B5 doesn't directly block the receptor recognition for neutralization. Although the epitope of 35B5 Fab involves some regions of the epitopes of the classes 2 and 3 of antibodies (Fig. 4a, b and Supplementary  Fig. 4a), the major 35B5-interacting residues, including the SARS-CoV-2-specific residues N354, T470, and N481 (Fig. 3b), which are not conserved in the spike proteins of SARS-CoV and MERS-CoV, are outside of the epitopes of the classes 2 and 3 of antibodies. Therefore, 35B5 targets a distinctive epitope to specifically neutralize SARS-CoV-2.
The SARS-CoV-2 VOCs contains several prevailing mutations on RBD, including N501Y (B.  (Fig. 5a). However, in the 35B5 Fab-S-6P complex structures, the residues N501, K417, L452, S477, T478, and F490 are not involved in the 35B5-RBD interactions. Only the residue E484 is located at the edge of the 35B5-RBD interface but not contacted by 35B5 Fab (Figs. 4c and 5b). Substitution of E484 by a lysine residue doesn't generate severe structural collision with 35B5 Fab (Fig. 5c). It has been found that L452R mutation is of significant adaptive value to the B.1.617.2 variant (delta). However, L452 has the distance of more than 4.5 Å and 5.1 Å, respectively, to the residues T69 and Y60 of 35B5 Fab in the 35B5 Fab-S-6P complex structure ( Fig. 5e), suggesting that the residue is not contacted by 35B5 Fab. Substitution for L452 by an arginine does not spatially affect the 35B5 Fab-RBD contacts. Instead, the mutation generates direct interactions with the residues Y60 and T69 of 35B5 Fab within 3 Å and likely forms two hydrogen bonds (Fig. 5e), suggesting that this mutation does not affect the binding affinity of 35B5 to RBD. Consistently, 35B5 has the comparable superpotent neutralization efficacy to the B.1.617.2 variant as that to the wild-type virus (Fig. 2). Recently, it was found that the C.37 variant (lambda) contains the mutations L452Q and F490S. 33 However, the mutations L452Q and F490S would not structurally affect the interactions of 35B5 Fab with RBD in the 35B5 Fab-S-6P complex structures ( Fig. 5f, g), indicating that 35B5 might also exhibit potent neutralizing efficacy to the C.37 variant. Thus, the unique epitope of 35B5 on RBD subtly avoids the prevailing mutation sites, which provides the molecular basis for the potent panneutralizing efficacy of 35B5 to the SARS-CoV-2 VOCs.
Neutralization mechanism of mAb 35B5 To investigate the neutralization mechanism of 35B5, we analyzed the "down" RBD domain in the State 1 35B5 Fab-S-6P complex. In the density map of the State 1 complex, there were some residual densities closed to the "down" RBD domain. We further carried out local refinements on the "down" RBD. The local refinements generated a 4.8-Å density map and revealed that there is a 35B5 Fab contacting the edge of the epitope on the "down" RBD. The low resolution of the local refinement map suggests that the interaction between the 35B5 Fab and the "down" RBD is highly dynamic ( Supplementary Fig. 4b-e). Structural modeling after the local refinements revealed that in contrast to the neutralizing mAbs BD-368 27,28 and C002 29 recognizing the epitopes that are fully exposed in both "down" and "up" RBDs, 35B5 Fab utilizes the CDRH regions to interact with the putative residues E340, T345, R346, F347, R346, K444, Y449, and N450 at the exposed edge of the epitope on the "down" RBD, indicating that this RBD domain is being initially recognized by 35B5 ( Fig. 4d and Supplementary  Figs. 4e and 7a, c). Structural superimposition of the 35B5 Fab-"up" RBD model with the "down" RBD in the 35B5 Fab-S-6P complex of State 1 or those in the Fab-free spike trimers 27,34 reveals that the β-sheet and the linking loop of FRH1 of 35B5 Fab have severe structural clashes with the adjacent NTD domain of the spike protein (Fig. 4e), suggesting that upon the high-affinity binding of 35B5 Fab onto the "down" RBD, the spatial collisions between  f Viral titers in the lungs were measured by qRT-PCR and presented as RNA copies per gram of lung tissue. g-i Gene expressions of cytokines and chemokines in the lungs were determined by qPT-PCR. j Histopathological analysis of lung tissues. ND not detected. Data are representative of two independent experiments. Error bars in a-d, f-i indicate SD The S trimer is represented as surface. 35B5 Fab is shown in cartoon and colored in purple. The "down", "up", and "releasing" RBD domains are colored in blue, yellow, and red, respectively. The NTD domain of the S trimer is colored in deep gray. The SD1, SD2, and S2 domains are colored in light gray. The gap caused by 35B5 Fab between the "releasing" RBD and NTD domains is highlighted with dashed lines. b Interactions of 35B5 Fab with "up" RBD. The interacting residues within 4 Å in RBD are colored in cyan. The RBD-interacting regions CDRH2, CDRH3, FRH1 and FRH3 of 35B5 Fab are colored in yellow, blue, orange, and red, respectively. c The 35B5 epitope on RBD. The epitope residues are labeled as indicated. d-f Detailed interactions of the CDR (d) and FR regions (e, f) of 35B5 Fab with RBD. Hydrogen-bond and saltbridge interactions are shown as blue and orange dashed lines, respectively 35B5 Fab and the NTD domain potentially exert repulsion force onto the NTD domain to induce the conformational conversion of RBD from "down" to "up" conformation. When RBD is in the "up" state, the interactions between S1 and the adjacent spike promoter in the trimer are largely reduced, leading to the loose and unstable packing of the spike protein. Once all the 3 RBDs in the spike trimer are in "up" state, the spike trimeric particle tends to dissociate, which induces further outward movement of the "up" RBDs generated the "releasing" conformations in the State 3 35B5 Fab-S-6P complex. Thus, the 35B5 Fab-S-6P complex structures in these three states suggest that neutralization of SARS-CoV-2 by 35B5 is likely carried out in four sequential steps (Fig. 4g): 35B5 firstly binds to the exposed edge of the epitope to recognizes the "down" RBDs of the spike protein; subsequently, binding of 35B5 imposes structural clashes on the NTD domain to drive the conformational changes of RBDs from "down" to "up"; next, the all "up" conformations of the RBD domains destabilizes the structure of the spike trimer and induces the outward movement and releasing of RBDs; finally, the released RBD domains cause the dissociation of the spike trimer.
The SARS-CoV-2 VOCs contain the most prevalent mutation D614G in the spike protein, which enhances infectivity by inserting a disordered loop (residues 620-640, the "630 loop") between the SD1/CTD1 and NTD domains within a protomer to prevent premature dissociation of the G614 trimer. 35 The residue D614 is not located in the epitope of 35B5 (Fig. 3b), thereby having no effects on its binding to the RBD domain. Although the D614G showed more up-RBDs, near 80% D614G particles contain at least one "down" RBD, 36 suggesting that RBDs have an equilibrium on the transition between up and down states. 35B5 The "down", "up", and "releasing" RBD domains are colored in blue, yellow, and red, respectively. The conformational changes of RBDs are indicated with arrows. g Schematic diagram of the neutralizing mechanism of 35B5 against SARS-CoV-2. The "down" and "up" RBDs in the tight-closed or loose-closed S trimers are colored in blue and yellow, respectively. The conformation changes and transmission of RBDs from the "down" to "up", finally to "releasing" states are shown as red arrows binding on the "up" RBD likely affects the dynamics of the RBD conformational transition. Moreover, upon binding to the "down" RBD domain, the capability of 35B5 to exert the repulsion force onto the NTD domain can probably counteract the enhancement effects of the 630 loop in the stability of the spike protein. Therefore, the unique epitope of 35B5 avoiding the prevailing mutation sites on RBD and the repulsion force of 35B5 exerting onto NTD during initial recognition renders its super-potent panneutralizing efficacy to the SARS-CoV-2 VOCs.
Cryo-EM structure of mAb 32C7 In contrast to 35B5, mAb 32C7 could not efficiently neutralize the SARS-CoV-2 VOCs. We also solved the cryo-EM structure of the Fab region of mAb 32C7 (hereafter named as 32C7 Fab) in complex with S-6P at a resolution of 2.8 Å (Supplementary Fig. 5 and Supplementary Table 1). The complex structure only contains one 32C7 Fab molecule, which is bound to a "down" RBD domain in the spike protein ( Supplementary Fig. 6a-f). The 32C7 Fab-RBD interface only covers a buried area of~935 Å 2 , which is much less than that of 35B5 Fab (Fig.  3c, Supplementary Figs. 6g and 7b, d). In the S-6P-32C7 Fab complex structure, 32C7 Fab does not structurally clash with the NTD domain. Instead, 32C7 Fab interacts with the glycan moiety on the residue N165 of the NTD domain ( Supplementary Fig. 6b). The binding surface of 32C7 Fab on RBD overlaps with the epitopes of the class 3 antibodies (Supplementary Fig. 6c), suggesting that 32C7 belongs to the classic class 3 family of neutralizing antibodies, which neutralizes SARS-CoV-2 via the Fc-mediated effector mechanism. 32 Therefore, the neutralization of 32C7 to SARS-CoV-2 is likely carried out via the same mechanism. The mutations in the D614G and B.1.351 variants likely induce the conformational changes of the spike protein, which decrease the neutralizing activities of 32C7 to VOCs via the Fcmediated effector mechanism.

Fig. 5
Interactions of 35B5 Fab with RBD. a Mutations in the RBD of reported SARS-CoV-2 variants of concern or interest. b-g The residues E484 (b) and L452 (d) are not involved in the interactions of RBD with 35B5 Fab. E484 is more than 4.2 Å away from 35B5 R84, which forms cation-π interaction with RBD F490. The E484K mutation in RBD does not interact with 35B5 R84 through molecular modeling (c). The L452Q modeling suggests none interactions formed at this site with 35B5 (f). In structural modeling, the L452R mutation (e) and F490S mutation (g) in RBD generate potential additional interactions with 35B5 Fab. Potential hydrogen bonds are shown as dashed lines
Previous works indicated that the ACE-binding surface is partially exposed on the "down" RBDs in the tight-or looseclosed spike trimer. 27,34 The epitopes of some neutralizing mAbs are also partially exposed on the "down" RBDs. 29,[39][40][41][42] The 35B5 Fab-S-6P complex structures in the three states we determined to provide for the first time the direct structural evidence for the possibility that the ACE2 or mAbs can approach the partially exposed surface or epitope residues for initial recognition and fulfill the conformational transformation of RBD. In previously identified RBD-targeted mAbs, almost all class 1 mAbs interact extensively with the residues K417 and N501. Most class 2 and class 3 mAbs contact E484, and most class 3 mAbs interact with L452. 9 35B5 does not directly bind to these prevailing mutant sites. In contrast to the Class 3 mAb S309 32 and the Class 4 mAb CR3022, 43 which target the residues conserved in the spike proteins of SARS-CoV, SARS-CoV-2, and MERS-CoV and therefore have cross-species neutralizing activities, 35B5 is a SARS-CoV-2specific antibody with super-potent pan-neutralizing activities to SARS-CoV-2 VOCs.
Protective vaccines are considered as keys to terminate the COVID-19 pandemic. However, accumulating evidence suggests that SARS-CoV-2 VOCs compromise neutralization by antibody responses induced by multiple vaccines, [18][19][20][21][44][45][46] including mRNA vaccines (BNT162b2 and mRNA-1273), adenovirus-based vaccines (AZD1222 and JNJ-78436735), a nanoparticle-based vaccine (NVX-CoV2373), and an inactivated protein vaccine (Coronavac), primarily due to RBD mutations. Thus, vaccines aimed at eliciting broadly neutralizing mAbs against SARS-CoV-2 VOCs are urgently needed. In the scenario of coping with diverse HIV-1 variants, an experimental pan-HIV-1 vaccine was designed based on the structurally defined epitope of a HIV-1 neutralizing mAb N123-VRC34.01, which targets a linear peptide of fusion peptide that is across-clade conserved. 47,48 In our study, cryo-EM structure of neutralizing 35B5 to the SARS-CoV-2 spike revealed an epitope that is composed of 30 residues and conserved among all the currently known SARS-CoV-2 VOCs. Importantly, mAb 35B5 was cloned from SARS-CoV-2 RBD-specific B cells obtained from individuals during their early convalescence, which might endow mAb 35B5 with limited somatic hypermutation (SHM). In contrast, a broad neutralizing mAb, ADG-2, was engineered by in vitro affinity optimization 49 and the extensive SHM makes ADG-2 epitope challenging as a target for SARS-CoV-2 vaccines. Thus, we propose that the 35B5 epitope might be a valuable target for the rational design of a universal SARS-CoV-2 vaccine, which is presumably essential for the ultimate termination of COVID-19 pandemic in the future.

Human samples
The COVID-19 patients enrolled in the study were admitted to Guangzhou Eighth People's Hospital (January to March 2020) and provided written informed consent. Blood samples were collected from patients during their convalescence and the time between symptom onset to sample collection was around 20 days. Healthy donors were two adult participants in the study. Blood samples were collected in cell preparation tubes with sodium citrate (BD Bioscience). Then, peripheral blood mononuclear cells (PBMCs) were isolated from blood samples using Ficoll (TBD Science), washed with PBS, suspended in cell freezing medium (90% FBS plus 10% DMSO), frozen in a freezing chamber at −80°C, and then transferred to liquid nitrogen. The study received IRB approvals at Guangzhou Eighth People's Hospital (KE202001134).
IgG VH and VL genes from B cells were PCR-amplified and cloned into vectors as previously described. 50,51 Briefly, plates containing cell lysate were thawed on ice, added with RT-PCR master mix and IgG VH/VL primers per well, and performed with RT-PCR following the one-step RT-PCR kit protocol (Takara, RR057A). Then, RT-PCR products were nested PCR-amplified with primers for IgG VH or IgG VL following the HS DNA polymerase kit protocol (Takara, TAK R010). The IgG VH and IgG VL nested PCR products were purified and next cloned into human IgG1 heavychain and light-chain expression vectors, respectively.
Monoclonal antibody production and purification Monoclonal antibodies were produced by using the ExpiCHO TM Expression System (Thermo Fisher). Briefly, ExpiCHO-S TM cells of 200 ml (6 × 10 6 cells/ml) in 1-L flask were transfected with a master mixture containing 200 μg heavy-chain plasmid, 200 μg lightchain plasmid, and 640 μl ExpiFectamine TM CHO reagent. On the day after transfection, cultured cells were added with 1.2 ml ExpiCHO TM enhancer and 48 ml ExpiCHO TM feed. A second volume of ExpiCHO TM feed was added into cultured cells on day 5 post transfection. Supernatants were harvested on day 12 post transfection for antibody purification using rProtein A Sepharose affinity chromatography (Sigma).

Cells and viruses
Vero E6 cells were obtained from ATCC and maintained in Dulbecco's Modified Essential Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher) and 1% penicillin-streptomycin (Thermo Fisher) at 37°C with 5% CO 2 . The Surface plasmon resonance (SPR) assay SPR experiments were performed using a Biacore ® T100 instrument (GE Healthcare, Uppsala, Sweden). All binding analyses were performed at 25°C using HBS-EP + (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.005% Tween-20) as the running buffer. Experiments were executed following the previously described SPR protocol. 52 Briefly, RBD protein was immobilized on the sensor chip CM5-type surface raising a final immobilization level of~600 Resonance Units (RU). Serial dilutions of 35B5 were injected in concentration from 10 to 0.625 nM. Serial dilutions of 32C7 were injected in concentration from 120 to 7.5 nM. For the competitive binding assays, the first sample flew over the chip at a rate of 20 μl/min for 120 s, then the second sample was injected at the same rate for another 200 s. The response units were recorded at room temperature and analyzed using the same software as mentioned above.

ELISA
The ELISA plates were coated with 50 ng of SARS-CoV-2 RBD protein (Sino Biological, 40592-V08H) in 100 μl PBS per well overnight at 4°C. On the next day, the plates were incubated with blocking buffer (5% FBS + 0.1% Tween-20 in PBS) for 1 h. Then, serially diluted mAbs in 100 μl blocking buffer were added to each well and incubated for 1 h. After washing with PBST (PBS + 0.1% Tween-20), the bound antibodies were incubated with HRPconjugated goat anti-human IgG antibody (Bioss Biotech) for 30 min, followed by washing with PBST and addition of TMB (Beyotime). The ELISA plates were allowed to react for~5 min and then stopped by 1 M H 2 SO 4 stop buffer. The optical density (OD) value was determined at 450 nm. EC 50 values were determined by using Prism 6.0 (GraphPad) software after log transformation of the mAb concentration using sigmoidal dose-response nonlinear regression analysis.
Neutralization assay An infectious SARS-CoV-2 neutralization assay was performed according to previous reports. 53 Vero E6 cells were seeded in 24well culture plates at a concentration of 4 × 10 4 cells per well at 37°C for 24 h. For infection with authentic SARS-CoV-2 at an MOI of 0.005, 200 μl of diluted authentic SARS-CoV-2 and fivefold serially diluted 35B5 or 32C7 (from 5 μg/mL to 0.064 μg/mL) mAbs were mixed in the medium with 2% FBS, and were then added into the Vero E6 cells. The culture supernatant was collected at 48 h post infection for focus forming assay and qRT-PCR. IC 50 values were determined by nonlinear regression using Prism 6.0 (GraphPad).
Focus forming assay (FFA) The virus titer was detected by FFA, which is characterized by its high throughput as compared to the traditional plaque assay. 54 Briefly, Vero E6 cells were seeded in 96-well plates 24 h prior to infection. Virus cultures were serially diluted and used to inoculate Vero E6 cells at 37°C for 1 h, followed by changed with fresh medium containing 1.6% carboxymethylcellulose. After 24 h, Vero E6 cells were fixed with 4% paraformaldehyde and permeabilized with 0.5% Triton X-100. Cells were then incubated with anti-SARS-CoV-2 nucleocapsid protein polyclonal antibody (Sino Biological), followed by an HRP-labeled secondary antibody (Proteintech). To generate a standard curve, the SARS-CoV-2 nucleocapsid (N) gene was cloned into a pcDNA3.1 expression plasmid and in vitro transcribed to obtain RNAs for standards. Indicated copies of N standards were 10-fold serially diluted and proceeded to qRT-PCR utilizing the same one-step SARS-CoV-2 RNA detection kit to obtain standard curves. The reactions were carried out on a QuantStudio 7 Flex System (Applied Biosystems) according to the manufacturer's instruction under the following reaction conditions: 50°C for 15 min, 95°C for 15 min, and 45 cycles of 94°C for 15 s and 55°C for 45 s. The viral RNA copies of each tissue were calculated into copies per ml and presented as log10 scale. The N-specific primers and probes were: N-F (5′-CAGTAGGGGAACTTCTCCTGCT-3′), N-R (5′-CTTT GCTGCTGCTTGACAGA-3′) and N-P (5′-FAM-CTGGCAATGGCGG TGATGCTGC-BHQ1-3′). In each qRT-PCR experiment, both positive control and negative control of simulated RNA virus particles were included to monitor the entire experimental process and ensure the reliability of the test results.

Histopathology
At day 5 post-SARS-CoV-2 infection, hACE2 mice were euthanized, and lungs were collected and fixed in 4% paraformaldehyde buffer. Tissues were embedded with paraffin and sections (3-4 mm) were stained with hematoxylin and eosin (H&E).
Production and purification of S-2P and S-6P proteins The plasmids encoding the ectodomains of the SARS-CoV-2 S-2P and S-6P mutants were kindly provided by Dr. Junyu Xiao.
HEK293F cells were cultured in the SMM 293-TI medium (Sino Biological Inc.) at 37°C with 8% CO 2 . The S-2P and S-6P plasmids were transiently transfected into HEK293F cells using 25-kDa linear polyethyleneimine (PEI) (Polysciences) with the PEI:DNA mass ratio of 3:1 and 1 mg DNA for per liter of culture when the cell density reached 2 × 10 6 cells per mL. At day 4 post transfection, the supernatants of the cell culture were harvested by centrifugation at 10,000×g for 30 min. The secreted S-2P and S-6P proteins were purified using HisPur TM cobalt resins (Thermo Scientific) and StrepTactin resins (IBA). Further purification was carried out using size-exclusion chromatography with a Superose 6 10/300 column (GE Healthcare) in the buffer containing 20 mM HEPES pH 7.2, 150 mM NaCl, and 10% Trehalose. The Fab regions of the 35B5 and 32C7 were obtained after the digestion by papain for 40 min at 37°C in a buffer containing 20 mM HEPES pH 7.2, 150 mM NaCl, 5 mM EDTA, and 5 mM L-cysteine. The obtained Fabs were purified with a Desalting column (GE Healthcare Life Sciences) to remove Lcysteine, and then further purified in a HiTrap Q column (GE Healthcare Life Sciences). The purified Fabs were collected and concentrated to 0.6 mg/mL.

Negative-staining analysis
For negative-staining assays, the S-2P, S-6P, 35B5 Fab, and 32C7 Fab proteins were diluted to 0.02 mg/ml in the buffer of 20 mM HEPES, pH 7.2, and 150 mM NaCl. In all, 2 μL of 35B5 Fab or 32C7 Fab was mixed with 2 μL S-2P or S-6P and incubated on ice for 3 and 10 min, respectively, or at room temperature for 30 min. The samples were loaded in the glow-discharged carbon-coated copper grids and stained with 3% uranyl acetate (UA). The prepared grids were examined using a Tecnai G2 Spirit BioTWIN transmission electron microscope (FEI) operated at 120 kV. Micrographs were recorded and analyzed using Digital Micrograph software with ×120,000 nominal magnification.
Cryo-EM sample preparation and data collection In total, 2 μL of S-6P (1.2 mg/mL) and 2 μL of 35B5 Fab or 32C7 Fab (0.6 mg/mL) were incubated for 3 min at room temperature and then loaded onto the glow-discharged holy-carbon gold grids (Quantifoil, R1.2/1.3). The grids were washed using the buffer containing 20 mM HEPES, pH 7.2, and 150 mM NaCl. The washed grids were blotted using a Mark IV Vitrobot (Thermo Fisher) at 100% humidity and 16°C for 3 s, and submerged in liquid ethane by plunge-freezing. For the S-6P-35B5 Fab complex, micrographs were recorded on a FEI Titan Krios (Thermo Fisher) electron microscope operated at 300 kV. Totally, 3740 movies were recorded on a K3 Summit direct electron detector (Gatan) in the super-resolution mode (0.5475 Å/pixel) at a nominal magnification of 81,000 using a defocus range of 1.2-1.3 μm. A GIF Quantum energy filter (Gatan) with a slit width of 20 eV was used on the detector. The micrographs were dose-fractioned into 32 frames with a total electron exposure of~50 electrons per Å 2 .
For the S-6P-32C7 Fab complex, micrographs were collected on a FEI Titan Krios (Thermo Fisher) operating at 300 kV using the AutoEMation software. 55 Totally, 2528 movies were recorded on a K3 Summit direct electron detector (Gatan) in the super-resolution mode (0.53865 Å/pixel) at a nominal magnification of 81,000 using a defocus range of 1.4-1.8 μm. A GIF Quantum energy filter (Gatan) with a slit width of 20 eV was used on the detector. The micrographs were dose-fractioned into 32 frames with a total electron exposure of~50 electrons per Å 2 .
Cryo-EM image processing Raw movie frames were binned, aligned and averaged into motion-corrected summed images using MotionCor2. 56 The doseweighted images were then imported into cryoSPARC 57 for the following image processing, including CTF estimation, particle picking and extraction, 2D classification, ab initio 3D reconstruction, heterogeneous 3D refinement and nonuniform homogeneous refinement. For the S-6P-35B5 Fab complex, eight representative particle templates were generated in 2D classification of 65,242 particles auto-picked by the blob picker from the first 1000 micrographs. Using these templates, 1,178,527 particles were extracted with a box size of 330 × 330 and classified into 150 classes in 2D classification. Among them, 43 classes that included 818,470 particles were selected for ab initio 3D reconstruction and heterogeneous refinement. Finally, 392,378 particles reconstructed an apparent architecture of the S-6p-35B5 Fab complex and were subjected to two more rounds of ab initio 3D reconstruction and heterogeneous refinement before nonuniform refinement. Then the particles were subjected to global and local CTF refinement for final nonuniform refinement, and generated three abundant populations of the S-6P-35B5 Fab complex and structures. Local refinements of the RBD-35B5 Fab region were performed to improve the interface density in cryoSPARC. Sharpened maps were generated and validated for model building and refinement. Reported resolutions are based on the gold-standard Fourier shell correlation 58,59 of 0.143 criterion.
For the S-6P-32C7 Fab complex, four representative particle templates were generated in 2D classification of 122,851 particles auto-picked by the blob picker from the first 1000 micrographs. Based on these templates, 744,865 particles were extracted with a box size of 380 × 380 and classified into 150 2D classes. Among them, 28 classes including 338,454 particles were selected for ab initio 3D reconstruction and heterogeneous refinement. 125,858 particles that reconstructed an apparent architecture of the S-6P-32C7 Fab complex were subjected to two more rounds of ab initio 3D reconstruction and heterogeneous refinement before nonuniform refinement. Finally, the S-6P-32C7 Fab complex structure that only includes a bound 32C7 Fab was reconstructed from 119,062 particles. Local refinement of the RBD-32C7 Fab region was also performed.
Cryo-EM structure modeling and analysis To build the S-6P-35B5 Fab complex structural model, an "up" RBD-35B5 Fab model was first generated using a Fab structure (PDB ID: 2X7L) and a RBD model from the Spike trimer (PDB ID: 7K8Y) and manually built in Coot 60 in the locally refined map. The obtained RBD-35B5 Fab model was superimposed with the intact Spike trimer structure (PDB ID: 7K8Y) to generate an initial model of the S-6P-35B5 Fab complex model. The "down" RBD-35B5 Fab model was obtained from fitting the structures of 35B5 Fab, the RBD, and NTD domains into the locally refined map of the the S-6P-35B5 Fab complex of State 1 and then validated by Phenix. 61 The model building of the S-6P-32C7 Fab complex was carried out in a similar procedure as that of the S-6P-35B5 Fab complex with a Spike trimer structure (PDB ID: 6XKL) as a template. All model building was performed in Coot. 60 Structural refinement and validation were carried out in Phenix. 61 Structural figures were generated using UCSF ChimeraX version 1.2. 62

Statistics
In the mouse study assessing mAb protection against WT SARS-CoV-2, the comparisons of lung viral titers and lung cytokine/ chemokine mRNA were performed using one-way ANOVA with Tukey's post hoc test by Prism 6.0 (GraphPad). In the mouse study assessing mAb protection against SARS-CoV-2 VOCs, the comparison of lung viral titers was performed using t test (unpaired) by Prism 6.0 (GraphPad).

DATA AVAILABILITY
The data used in the study are available from the corresponding authors upon reasonable request. The cryo-EM maps and atomic coordinates have been deposited to the Electron Microscopy Data Bank (EMDB) and Protein Data Bank (PDB) with accession codes EMD-31033 and PDB 7E9N (State 1 of the 35B5 Fab-S-6P complex), EMD-31444 and PDB 7F46 ("down" RBD/35B5 Fab local refinement of State 1), EMD-