Study approval, convalescent patients and blood samples

The study was approved by the Research Ethics Committee of Shenzhen Third People’s Hospital, China (approval no.: 2020-084). The research was conducted in strict accordance with the rules and regulations of the Chinse government for the protection of human subjects. All participants had provided written, informed consents for sample collection and subsequent analysis. Detailed information of participants in this study is provided in Supplementary Table 1. The study enrolled a total of nine patients aged 32–73 yr and recovered from infection with wild-type SARS-CoV-2 in January 2020. Of these, three (P2, P5 and P10) once developed severe pneumonia whereas the remaining six (P43, P75, P104, P140, P186 and P195) only had mild symptoms during hospitalization at Shenzhen Third People’s Hospital. P2 and P5 donated their blood samples twice and the remaining patient once during a 16–111-d recovery period post symptom onset. According to the policies at that time, local patients with COVID-19 were given free treatments and follow-up visits. The collected blood samples were separated into plasma and PBMCs through Ficoll-Hypaque density gradient centrifugation. The plasma samples were heat-inactivated at 56 °C for 1 h and stored at −80 °C, whereas the PBMCs were maintained in freezing media and stored in liquid nitrogen until use.

Production of pseudoviruses and neutralizing assay

The wild-type pseudovirus used throughout the analysis was the wild-type strain (GenBank: MN908947.3) or had a D614G mutation (D614G). The alpha variant (Pango lineage B.1.1.7, GISAID: EPI_ISL_601443) included a total of nine reported mutations in the S protein (del69-70, del144, N501Y, A570D, D614G, P681H, T716I, S982A and D1118H). The beta variant (Pango lineage B.1.351, GISAID: EPI_ISL_700450) included ten identified mutations in the spike (L18F, D80A, D215G, del242-244, S305T, K417N, E484K, N501Y, D614G and A701V). The gamma variant (Pango lineage P.1, GISAID: EPI_ISL_792681) had 12 reported mutations in the spike (L18F, T20N, P26S, D138Y, R190S, K417T, E484K, N501Y, D614G, H655Y, T1027I and V1176F). The delta variant (Pango lineage B.1.617.2, GISAID: EPI_ISL_1534938) included ten reported mutations in the spike (T19R, G142D, del156-157, R158G, A222V, L452R, T478K, D614G, P681R and D950N). The delta plus variant (Pango lineage AY.x, GISAID: EPI_ISL_ 3019629) had one more mutation, K417N, than the delta variant. The epsilon variant (Pango lineage B.1.429, GISAID: EPI_ISL_2922315) included S13I, W152C, L452R and D614G in the spike. The kappa variant (Pango lineage B.1.617.1, GISAID: EPI_ISL_1384866) included T95I, G142D, E154L, L452R, E484Q, D614G, P681R and N1071H in the spike. The mu variant (Pango lineage B.1.621, GISAID: EPI_ISL_3987640) included T95I, Y144T, Y145S, ins146N, R346K, E484K, N501Y, D614G, P681H and D950N in the spike. The eta variant (Pango lineage B.1.525, GISAID: EPI_ISL_2885901) included Q52R, A67V, del69-70, del144, E484K, D614G, Q677H and F888L in the spike. The iota variant (Pango lineage B.1.526, GISAID: EPI_ISL_2922249) included L5F, T95I, D253G, E484K, D614G and A701V in the spike. The omicron BA.1 variant (Pango lineage BA.1, GISAID: EPI_ISL_6752027) was constructed with 34 mutations in the spike (A67V, del69-70, T95I, G142D, del143-145, del211, L212I, ins214EPE, G339D, S371L, S373P, S375F, K417N, N440K, G446S, S477N, T478K, E484A, Q493R, G496S, Q498R, N501Y, Y505H, T547K, D614G, H655Y, N679K, P681H, N764K, D796Y, N856K, Q954H, N969K and L981F). The omicron BA.2 variant (Pango lineage BA.2, GISAID: EPI_ISL_8515362) was constructed with 29 mutations in the spike (T19I, del24-26, A27S, G142D, V213G, G339D, S371F, S373P, S375F, T376A, D405N, R408S, K417N, N440K, S477N, T478K, E484A, Q493R, Q498R, N501Y, Y505H, D614G, H655Y, N679K, P681H, N764K, D796Y, N969K and Q954H). BA.2.12.1 spike was constructed based on BA.2 with additional L452Q and S704. BA.2.75 spike was constructed based on BA.2 with additional W152R, F157L, I210V, G257S, D339H, G446S, N460K and Q493R (reversion). The omicron BA.3 variant (Pango lineage BA.3, GISAID: EPI_ISL_7740765) was constructed with 30 mutations in the spike (A67V, del69-70, T95I, G142D, del143-145, del211, L212I, G339D, S371F, S373P, S375F, D405N, K417N, N440K, G446S, S477N, T478K, E484A, Q493R, Q498R, N501Y, Y505H, D614G, H655Y, N679K, P681H, N764K, D796Y, Q954H and N969K). The omicron BA.4 variant (Pango lineage BA.4, GISAID: EPI_ISL_12559461) was constructed with 32 mutations in the spike (T19I, del24-26, A27S, del69-70, G142D, V213G, G339D, S371F, S373P, S375F, T376A, D405N, R408S, K417N, N440K, G446S, L452R, S477N, T478K, E484A, F486V, Q498R, N501Y, Y505H, D614G, H655Y, N679K, P681H, N764K, D796Y, Q954H and N969K). BA.4 and BA.5 shared the same amino acid sequence in the spike. The full-length genes of spike variants were synthesized by Genwiz and verified by sequencing.

Pseudoviruses were generated by cotransfecting HEK 293T cells (ATCC) with human immunodeficiency virus backbones expressing firefly luciferase (pNL4-3-R-E-luciferase) and pcDNA3.1 vector encoding either wild-type or variant S proteins51,52. Viral supernatant was collected 48 h or 72 h later, centrifuged to remove cell lysis and stored at −80 °C until use. Viral infectious titers were measured by luciferase activity in the HeLa-hACE2 cells using Bright-Glo Luciferase Assay Vector System (Promega). Berthold Centro LB 960 was used for measuring luciferase activity. Neutralization assays were performed by incubating pseudoviruses with serial dilutions of heat-inactivated plasma or purified mAbs at 37 °C for 1 h. Approximately 1.5 × 104 per well of HeLa-hACE2 cells were then added in duplicate to the above virus–antibody mixture. At 48 h later, the half-maximal inhibitory dilution of plasma (ID50) or concentration of the mAbs (IC50) was determined by luciferase activity using GraphPad Prism 8.3 (GraphPad Software). HeLa-hACE2 cells were kindly provided by Q. Ding at the Center for Infectious Research of Tsinghua University.

Plasma and antibody binding analyzed by ELISA

The recombinant S, S1, RBD and S2 proteins derived from the wild-type SARS-CoV-2, and S1 proteins or S proteins of SARS-CoV-1 and MERS-CoV (Sino Biological), were diluted to final concentrations of 0.5 or 2 μg ml−1, coated onto 96-well plates and incubated overnight at 4 °C. The plates were washed with PBS-T (PBS containing 0.05% Tween 20) and blocked with blocking buffer (PBS containing 5% skim milk and 2% BSA) at room temperature for 1 h. Serially diluted plasma samples or mAbs were added to the plates and incubated at 37 °C for 1 h. After extensive washing, the plates were then incubated with secondary anti-human IgG labeled with HRP (1:5,000 dilution) (ZSGB-BIO) at 37 °C for 30 min or 1 h before incubation with TMB substrate (Kinghawk) at room temperature for 5 min or 20 min. Optical density was measured by a spectrophotometer at 450 nm.

Isolation of SWT-specific single B cells by FACS

SARS-CoV-2 SWT-specific B cells were sorted as previously described53,54. In brief, PBMCs from convalescent individuals were collected and incubated with an antibody and recombinant SWT trimer cocktail for identification of SWT-specific B cells. The cocktail consisted of CD19-PE-Cy7 (1:50 dilution), CD3-Pacific Blue (1:50 dilution), CD8-Pacific Blue (1:25 dilution), CD14-Pacific Blue (1:50 dilution), CD27-APC-H7 (1:25 dilution), IgG-FITC (1:12.5 dilution) (or IgM−PerCP-Cy5.5 (1:50 dilution), IgD−PE-CF594 (1:25 dilution)) (BD Biosciences) and recombinant SWT-Strep or SWT-His purified in our laboratory. Three consecutive staining steps were conducted. The first was using a LIVE/DEAD Fixable Dead Cell Stain Kit (Invitrogen) to exclude the dead cells. The second was mixing with an antibody and recombinant SWT trimer cocktail to identify SWT-specific B cells. The third was to target the recombinant SWT trimer captured on the surface of B cells by either Streptavidin-APC (eBioscience) or anti-his-APC/PE antibodies (1:25 dilution) (Abcam). The stained cells were thoroughly washed and resuspended in PBS before being strained through a 70-μm cell mesh (BD Biosciences). SARS-CoV-2 SWT-specific single B cells were gated as either CD19+CD3−CD8−CD14−CD27+IgG+SWT+ or CD19+CD3−CD8−CD14−IgM−IgD−SWT+ and sorted (BD Aria II) into the 96-well PCR plates containing 20 μl of lysis buffer (5 μl of 5 × first strand buffer, 0.5 μl of RNaseOUT, 1.25 μl of 0.1 M dithiothreitol (Invitrogen) and 0.0625 μl of Igepal (Sigma) per well). Plates were then snap-frozen on dry ice and stored at −80 °C until the reverse transcription reaction.

Single B cell PCR, cloning and expression of mAbs

The IgG heavy and light chain variable genes were amplified by nested PCR and cloned into linear expression cassettes to produce full IgG1 antibodies as previously described54,55. Specifically, all second round PCR primers containing tag sequences were used to produce the linear immunoglobulin expression cassettes by overlapping PCR. Meanwhile, the variable genes of heavy and light chain were sequenced, synthesized and then cloned into the backbone of antibody expression vectors containing the constant regions of human IgG1 by GenScript56. Overlapping PCR products of paired heavy and light chain expression cassettes were cotransfected into the HEK 293T cells (ATCC) to produce antibodies for binding analysis. Large quantities of mAbs were produced by transient transfection of 293F cells (Life Technologies) with equal amounts of paired heavy and light chain plasmids. Antibodies in the culture supernatant were purified by affinity chromatography using Protein A bead columns (GE Healthcare) and their concentrations were determined by a NanoDrop2000 (Thermo Scientific).

Gene family usage and recombination analysis of mAbs

The IMGT/V-QUEST program (http://www.imgt.org/IMGT_vquest/vquest) was used to analyze the germline gene, degree of somatic hypermutation, framework region and the loop lengths of complementarity determining region 3 (CDR3) of each antibody. Chord diagrams showing the germline gene usages of paired heavy and light chain were analyzed and presented by the R package circlize v.0.4.14. The width of the linking arc is proportional to the number of antibodies identified.

Epitope mapping by competition surface plasmon resonance

For epitope mapping, two different mAbs were sequentially injected and monitored for binding activity to determine whether the two mAbs recognized separate or closely situated epitopes. To determine competition with the human ACE2, antibodies (1 μM) were injected onto the RBD-immobilized CM5 chip until the binding steady-state was reached. ACE2 (2 μM) was then injected for 60 s. Blocking efficacy was determined by comparison of response units with and without previous antibody incubation. Biacore 8K Control Software v.3.0.12.15655 was used for binding competition studies.

Crystallization and data collection

The SARS-CoV-2 RBD and the Fab fragment were mixed at a molar ratio of 1:1.2, incubated at 4 °C for 2 h and further purified by gel-filtration chromatography. The purified complex concentrated to approximately 10–15 mg ml−1 in HBS buffer (10 mM HEPES, pH 7.2, 150 mM NaCl) was used for crystallization. The screening trials were performed at 18 °C using the sitting-drop vapor-diffusion method by mixing 0.2 μl of protein with 0.2 μl of reservoir solution. Crystals of P5S-2B10 Fab and RBD complex were successfully obtained in 0.2 M magnesium sulfate heptahydrate, 17% w/v PEG 3350, whereas P5-1H1 Fab and RBD complex was obtained in 2% v/v Tacsimate pH 5.0, 0.1 M sodium citrate tribasic dihydrate pH 5.4 and 13% w/v PEG 3350. P2S-2E9 Fab and RBD complex was obtained in 15% v/v 2-Propanol, 0.1 M sodium citrate tribasic dihydrate pH 4.8, 11% w/v PEG 10000, and P5S-3B11 Fab and RBD complex in 0.05 M citric acid pH 4.4, 0.05 M BIS-TRIS propane, 16% w/v PEG 3350. All crystals were collected, soaked briefly in mother liquid with 20% glycerol and flash-frozen in liquid nitrogen. Diffraction data were collected at a wavelength of 0.987 Å on the BL18U1 beam line of the Shanghai Synchrotron Research Facility and processed by HKL2000. The data processing statistics are listed in Extended Data Fig. 5.

Structure determination and refinement

The structure was determined by the molecular replacement method with PHASER in CCP4 suite 7.1.007 (ref. 57). The search models were the SARS-CoV-2 RBD structure (PDB: 6M0J) and the structures of the variable domains of the heavy and light chains available in the PDB with the highest sequence identities. Subsequent model building and refinement were performed using COOT v.0.9.2 and PHENIX v.1.18.2, respectively58,59. The structural refinement statistics are listed in Extended Data Fig. 5. All structural figures were generated using PyMOL 2.0 and Chimera v.1.15.

Cryo-EM structural determination

Aliquots of complexes (4 μl, in buffer containing 20 mM Tris, pH 8.0, and 150 mM NaCl) of SARS-CoV-2 BA.1 S ectodomains (2 mg ml−1) and P2-1B1 Fab were applied to glow-discharged holey carbon grids (Quantifoil grid, Cu 300 mesh, R1.2/1.3). Fab fragments were mixed with SARS-CoV-2 S trimer at a molar ratio of 1.2:1. The grids were then blotted for 3 s and immediately plunged into liquid ethane using Vitrobot Mark IV (Thermo Fisher Scientific). The cryo-EM data of complexes were collected by the FEI Titan Krios microscope (Thermo Fisher Scientific) at 300 kV with a Gatan K3 Summit direct electron detector (Gatan) at Tsinghua University. In total, 2,628 movies were collected by SerialEM version 4.0.4, with a magnification of 29,000 and defocus range between −1.3 and −1.5 μm. Each movie has a total accumulated exposure of 50 e− Å−2 fractionated in 32 frames of 2.13-s exposure. The stacks were binned twofold, resulting in a pixel size of 0.97 Å per pixel. Motion correction (MotionCor2 v.1.2.6), CTF estimation (GCTF v.1.18) and nontemplated particle picking (Gautomatch v.0.56; http://www.mrc-lmb.cam.ac.uk/kzhang/) were automatically executed using the TsingTitan.py program60,61. Sequential data processing was carried out with cryoSPARC v.3.3.1 (refs. 62,63). The initial models of SARS-CoV-2 BA.1 RBD (PDB: 7WHH) and P2-1B1 Fab were fitted to the map using UCSF Chimera v.1.15 (ref. 64). Manual model rebuilding was carried out with COOT v.0.9.2 and refined with PHENIX v.1.18.2 real-space refinement. The quality of the final model was analyzed by PHENIX v.1.18.2. The validation statistics of the structural models are summarized in Extended Data Fig. 7. All structural figures were generated using PyMOL 2.0 and Chimera v.1.15.

Binding of mAbs to cell-surface-expressed S proteins

The entire procedure was conducted as previously published51,52,65. Specifically, HEK 293T cells were transfected with expression plasmids encoding either wild-type or omicron variant S proteins, and incubated at 37 °C for 36 h. Cells were digested from the plate with trypsin and distributed onto 96-well plates. Cells were washed twice with 200 µl of staining buffer (PBS with 2% heated-inactivated FBS) between each of the following steps. First, cells were stained with each antibody (1 μg ml−1), or ACE2 (1 μg ml−1), or S2-specific mAb (1 μg ml−1) (1:200 dilution) (MP Biomedicals) at 4 °C for 30 min. PE-labeled anti-human IgG Fc (1:40 dilution; BioLegend), anti-his PE secondary antibody (1:200 dilution; Miltenyi) or anti-mouse IgG FITC (1:200 dilution; Thermo Fisher Scientific) was added and incubated at 4 °C for 30 min. After extensive washes, the cells were resuspended and analyzed with BD LSRFortessa (BD Biosciences) and FlowJo 10 software (FlowJo). HEK 293T cells with mock transfection were stained as background control. Fold changes in antibody binding were calculated by the ratio of the total fluorescence intensity (TFI) of omicron over wild-type, normalized by that of S2-specific antibody (nTFI). TFI was calculated by multiplying the mean fluorescence intensity (MFI) and the number of positive cells in the selected gates. For example, the fold change in BA.1 of P5S-2B10 was calculated by the following formula: fold change = (BA.1 of TFI/BA.1 of nTFI)/(wild-type of TFI/wild-type of nTFI). TFI = MFI × subset frequency.

Antibody protection in hACE2 transgenic mice

Mouse experiments were performed in a Biosafety Level 3 (BSL-3) facility in accordance with the National University of Singapore (NUS) Institutional Animal Care and Use Committee (protocol no. R20-0504), and the NUS Institutional Biosafety Committee and NUS Medicine BSL-3 Biosafety Committee approved SOPs. As previously described66, 8-week-old female K18-hACE2 transgenic mice (InVivos) were utilized for this study. The mice were housed and acclimatized in a BSL-3 facility for 72 h before the start of the experiment. The housing conditions were 23 ± 2 °C (high/low temperature), 50 ± 10% (high/low humidity) and 12-h light/12-h dark (light cycle). K18-hACE2 transgenic mice were subjected to P2-1B1, P5S-2B10, P5-1H1 or P2S-2E9 (10 mg kg−1) delivered through intraperitoneal injection a day before infection or left untreated. P5S-2B10 and P5-1H1 experiments shared a group of untreated mice. The viral challenge was conducted through intranasal delivery in 25 μl of 1.7 × 103 PFU of the infectious SARS-CoV-2 omicron BA.1 or beta variant. Body weights were measured before infection as baseline and monitored daily throughout the following 14 d. Mice were euthanized when their body weight fell below 80% of their baseline body weight. Some of the mice from each experimental group were killed at 3 d for omicron BA.1 or 4 d for the beta variant post infection, and lung and brain tissues were collected. Each organ was halved for the plaque assay and histology analysis, respectively.

For virus titer determination, supernatants from homogenized tissues were tenfold serially diluted in DMEM supplemented with antibiotic and antimycotic, and added to A549-hACE2 cells (omicron virus) or Vero E6 cells (beta virus) in 12-well plates. The inoculum was removed after 1 h of incubation for virus adsorption. Cells were washed once with PBS before 1.2% MCC-DMEM overlay medium was added to each well. Then cells were incubated at 37 °C, 5% CO2 for 72 h for plaque formation. Cells were fixed in 10% formalin overnight and counterstained with crystal violet. The number of plaques was determined and the virus titers of individual samples were expressed as logarithm of PFU per organ.

For histopathological analyses, lung lobes were fixed in 3.7% formaldehyde solution before removal from BSL-3 containment. Tissues were routinely processed, embedded in paraffin blocks (Leica Surgipath Paraplast), sectioned at 4-μm thickness and stained with hematoxylin and eosin (Thermo Scientific) following standard histological procedures. For immunohistochemistry, the sections were deparaffinized and rehydrated, followed by heat-mediated antigen retrieval, quenching of endogenous peroxidases and protein blocking. Sections were then covered with rabbit anti-SARS-CoV-2 N protein mAb (Abcam, 1:1,000 dilution) for 1 h at room temperature. Subsequently, sections were incubated with rabbit-specific HRP polymer secondary antibody (Abcam, no dilution), visualized using chromogenic substrate DAB solution (Abcam) and counterstained with hematoxylin.

Data reporting

No statistical methods were used to predetermine sample sizes but our sample sizes are similar to those reported in previous publications. The sample size of nine COVID-19 convalescent patients is sufficient for isolating nAbs in the field (PMID: 32454513 and PMID: 32698192). For animal experiments, the number of mice for the in vivo protection assay in each group was 4–6, which is acceptable in the field (PMID: 33657424 and PMID: 33431856). Data distribution was assumed to be normal but this was not formally tested. The experiments for antibody isolation from COVID-19 convalescents were not randomized. Mice in antibody protection experiments were randomly divided into treated and untreated groups. Data collection and analysis were not performed blind to the conditions of the experiments. No animals or data points were excluded.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.