Characterisation of plasmablast-derived HBsAg-specific antibody and its structural basis for binding to native HBsAg dimer

Abstract

Background

Plasmablast-derived HBV surface antigen (HBsAg)-specific monoclonal antibody (mAb) and structural basis for binding to native HBsAg are poorly known.

Objective

We aimed to identify plasmablast-derived HBsAg-specific mAbs, evaluate their antiviral activities and resolve their structure for binding to native HBsAg.

Design

A previously vaccinated volunteer was enrolled in this study, who was boosted with a dose of recombinant hepatitis B vaccine and donated the blood sample. Activated plasmablasts were sorted from fresh peripheral blood mononuclear cells and mAbs were expressed. Their gene features, cross-genotypic binding activities and antiviral functions in vitro and in vivo were comprehensively analysed. The cryo-electron microscopy (cryo-EM) was used to determine the structure of representative mAb bound to the native HBsAg.

Results

In this study, we cloned a series of HBsAg-specific mAbs directly from clonally expanded plasmablasts from a vaccinated individual. Most of the mAbs displayed cross-reactivities of binding to different genotype HBsAg proteins and antiviral functions such as neutralisation and antibody-dependent cellular phagocytosis. These human anti-HBsAg mAbs, especially SY-4-class and SY-23-class, could be good candidates for antibody drugs. The cryo-EM structure of SY-23 bound to the dimeric HBsAg was determined, revealing its binding mechanism and unprecedented structural detail of the major antigenic loop (AGL) of HBsAg.

Conclusion

Overall, our work has uncovered the diverse gene features and varied anti-HBV activities of plasmablast-derived mAbs, providing a series of antibody drug candidates and the long-sought-after atomic model of AGL has paved the way for a wholistic characterisation of the AGL’s dynamic conformation during HBV infection and immune response.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• With the stimulation of virus infection, those vaccine-induced pre-existing HBV surface antigen (HBsAg)-specific memory B cells (MBCs) rapidly differentiate into plasmablasts and then directly secrete monoclonal antibodies (mAbs) against the potential HBV infection.

• Not all HBsAg-specific MBCs could successfully differentiate to plasmablasts.

• Most of the currently available HBsAg-specific mAbs were isolated from MBCs rather than activated plasmablasts.

WHAT THIS STUDY ADDS

• A total of 92 plasmablast-derived HBsAg-specific mAbs are identified from a previously vaccinated individual who received a booster dose of hepatitis B vaccine.

• These mAbs exhibit the diverse germline gene usage, complementarity determining region 3 length and clonal expansion level, as well as varied neutralisation and antibody-dependent cellular phagocytosis activity.

• Cryo-electron microscopy of HBsAg-dimer bound to a potent mAb (SY-23) provides the long-sought-after structural insights into HBV recognition by neutralising antibodies.

• With the help of SY-23-Fab binding, the atomic model of HBsAg antigenic loop (AGL) is successfully built.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Our study fills a gap in the characterisation of plasmablast-derived HBsAg-specific mAbs.

• With the help of antibody binding, our study obtains the long-sought-after structural information for HBsAg dimer itself, AGL conformation and actual antigen–antibody interactions.

Introduction

Hepatitis B caused by the HBV infection is still a major global health problem.¹ Chronic HBV infection puts a high risk of developing liver cirrhosis and hepatocellular carcinoma.^{2 3} There were 254 million people living with chronic HBV infection in 2022, along with 1.2 million new infections each year (https://www.who.int/news-room/fact-sheets/detail/hepatitis-b). Safe and effective vaccines can protect infants against the HBV infection, especially when received as soon as possible after birth.⁴ HBV surface antigen (HBsAg) is the main component of hepatitis B vaccine, which can induce a protective antibody response to HBV.⁵ Although circulating HBsAg-specific antibody (HBsAb) titres decrease over time, the immune memory lasts for many years after vaccination.⁶ On re-exposure to the same stimuli, HBsAg-specific memory B cells (MBCs) could rapidly differentiate into plasmablasts, followed by the robust clonal expansion and secreting antibodies within about 7 days.^7–9 Therefore, these HBsAg-specific monoclonal antibodies (mAbs) derived from proliferative plasmablasts should be the forefront of antibody fighting against the potential pathogen infection including HBV,^9–15 which better reflect the antiviral ability of HBV vaccinees. Because not all HBsAg-specific MBCs could be effectively activated, such as atypical MBCs, which highly express inhibitory receptors and are defective to differentiation,^{16 17} indicating the complexity and diversity of the recalling antibody response. One major question that needs to be answered is what specific features of plasmablast-derived anti-HBsAg mAbs during the antibody recalling phase will facilitate the development of HBV vaccine boosting strategy. Currently, a series of research studies mainly focus on isolating anti-HBsAg mAbs from total specific MBCs and characterising their biological functions in vitro and in vivo, aiming to develop therapeutic antibody drugs for the chronic HBV infection.^18–24 Although several methodology studies have reported that some human anti-HBsAg mAbs could be isolated from plasmablasts,^{7 8 25} the detailed features and functions of this category of mAbs are still poorly understood. Little is known about how these mAbs bind HBsAg or their antiviral abilities in vitro and in vivo.

Regarding antibody binding to HBsAg, the major immunogenic region in HBsAg is the major antigenic loop (AGL) after the second transmembrane helix (TH2).²⁶ The AGL contains the main immunogenic determinants of HBsAg and also plays crucial roles in virus infectivity and assembly.^27–29 A striking feature of the AGL is its high number of cysteine residues which form both intramolecular and intermolecular disulfide bonds to give rise to specific conformation of this region. Despite being crucial for HBV infection and the resulting antibody response, the exact disulfide bond linkages of the AGL are poorly defined. While recent cryo-electron microscopy (cryo-EM) studies of subviral particles have started to unravel the domain arrangement and near-atomic resolution structure of some regions of HBsAg,^30–32 the structures of the AGL and antibody-bound HBsAg remain elusive, seriously hampering our understanding of how mAbs bind and neutralise HBV.

In this study, we isolated 92 HBsAg-specific mAbs from peripheral plasmablasts of a previously vaccinated individual who received a booster dose of hepatitis B vaccine to make a comprehensive analysis of antibody genetic feature, clonal expansion, binding capacity, neutralising activity, antibody-dependent cellular phagocytosis (ADCP) and cross-reactivity to different genotypic HBsAg. Two representative mAbs, SY-4 and SY-23, displayed potent abilities of decreasing the circulating HBsAg, HBV DNA and HBV pregenomic RNA (pgRNA) in the HBV carrier mice model. Here, we determined the cryo-EM structure of a representative potent mAb (SY-23) bound to the dimeric HBsAg extracted from yeast-derived HBsAg particles, which represents an important advance in this field. Unprecedented structural details of AGL of HBsAg and its conformational epitope binding by SY-23 were clearly presented. Some intermolecular disulfide bonds and hydrophobic interactions contributed substantially to the formation of dimeric HBsAg. Furthermore, our optimised workflow of structure elucidation of the fragment of antigen binding (Fab)-bound HBsAg will accelerate mechanistic studies of HBV neutralisation by antibodies whether recognising conformational or linear epitopes. These findings deeply strengthen our understanding of genetic and functional features of the recalling anti-HBsAg mAbs and how they bind to the AGL of native HBsAg.

Materials and methods

Study approval and human samples

This volunteer had provided the written informed consent for the sample collection and subsequent analysis. Peripheral blood mononuclear cells (PBMCs) and plasma from this volunteer who received a booster dose of yeast-derived genotype A recombinant hepatitis B vaccine (Shenzhen Kangtai Biological Products) were stored at the Biobank of Shenzhen Third People’s Hospital.

Study approval and animal experiment

This study was approved by the Shenzhen Third People’s Hospital Laboratory Animal Ethics Committee (approval number: 2023–002). The adeno-associated virus (AAV)-HBV mouse model was established using recombinant AAV serotype 8 vectors carrying the HBV-D genome (Packgene). The viral titre is 1×10¹³ GC/mL. For viral delivery, each mouse received a tail vein injection of 4×10¹¹ GC diluted in 40 µL of viral suspension mixed with 160 µL phosphate-buffered saline (PBS) (total injection volume: 200 µL). Four weeks after virus administration, antibody treatment was initiated. Mice received 400 µg of either SY-4 or SY-23 mAbs (experimental groups), or 400 µg of P2C-1F11 (an anti-SARS-CoV-2 mAb, irrelevant antibody control) via tail vein injection. Serum samples were collected on Day 0, 1 and 2 post treatment. Liver tissues were collected on Day 2 post treatment. All experimental protocols adhered to the institutional animal care guidelines.

Serological analysis of HBsAg and HBeAg

The blood samples were centrifuged at 8000 rpm for 10 mins to isolate serum. Prior to analysis, serum samples were diluted at 1: 100 with the PBS buffer. Quantitative detection of serum HBsAg and Hepatitis B e Antigen (HBeAg) was performed using a chemiluminescence immunoassay platform (innoDx), according to the manufacturer’s instructions.

Measurement of serum alanine aminotransferase

Serum alanine aminotransferase (ALT) activity was determined using a commercially available colourimetric assay based on the Reitman-Frankel method from Nanjing Jiancheng Bioengineering Institute (NJJC Bio). Reactions were performed in 96-well microplates. ALT catalyses the conversion of alanine and α-ketoglutarate to pyruvate, which subsequently reacts with 2,4-dinitrophenylhydrazine to form a coloured hydrazone under alkaline conditions. Absorbance was measured at 505 nm using a microplate reader (Varioskan LUX, Thermo Fisher Scientific), and ALT activity was calculated from a pyruvate standard curve and expressed as units per litre (U/L) according to the manufacturer’s instructions.

HBV pregenomic RNA detection

Serum samples were collected and stored at −80°C until analysis. Viral RNA was extracted using the manufacturer’s recommended procedure. HBV pgRNA was quantified using the commercial HBV pgRNA Detection Kit (Shengxiang) according to the manufacturer’s instructions. Briefly, extracted RNA was mixed with the supplied one-step reverse-transcription-PCR (RT-PCR) master mix, specific primers and probe, and an internal control. Amplification was performed on an ABI 7500 series real-time PCR instrument under the following thermal cycling conditions: reverse transcription and initial denaturation at 95°C for 1 min, followed by 45 cycles of 95°C for 15 s and 60°C for 30 s. The pgRNA concentration was determined by comparison with a standard curve provided in the kit and reported in copies/mL. All samples and standards were run in technical duplicate (or triplicate), and each run included the negative and positive controls supplied with the kit.

For the detection of liver pgRNA and GAPDH, approximately 10–30 mg of liver tissue was collected and total RNA was extracted using a commercial total RNA extraction kit (TIANGEN). The obtained RNA was then used as the template for HBV pgRNA analysis following the procedure described above. For GAPDH quantification by quantitative PCR, 1.5 µL each of 10 pmol/µL forward and reverse primers (Forward: AGGTCGGTGTGAACGGATTTG; Reverse: GGGGTCGTTGATGGCAACA) were used with the FastKing One-Step gDNA Removal cDNA First Strand Synthesis Mix (TIANGEN) for reverse transcription, followed by amplification using the SuperReal PreMix (Enhanced with SYBR Green, TIANGEN). Relative levels of HBV pgRNA were normalised to GAPDH expression using the 2^−ΔCt (ΔCt=Ct_pgRNA − Ct_GAPDH).

HBV DNA detection

Viral DNA was extracted from serum samples using the commercial HBV Nucleic Acid Detection Kit (DaAn Gene), according to the manufacturer’s instructions. Briefly, nucleic acids were purified using the provided extraction solution and an internal control was co-extracted with each sample to monitor the process. Quantitative real-time PCR was subsequently performed on compatible instruments (ABI Prism 7500) using target-specific primers and fluorescent probes. The assay has a lower limit of detection of 30 IU/mL and a quantitative range of 100 to 5.0×10⁸ IU/mL. Results were interpreted based on the amplification curve and cycle threshold values in the Fluorescein Amidite (FAM) channel, with validation of nucleic acid extraction confirmed by signal in the VIC channel (internal control). All experiments included negative and positive controls that passed established quality criteria.

Tissue preparation and H&E staining

Liver tissues were collected from C57BL/6 mice infected with AAV-HBV and treated with mAbs. Tissues were harvested 2 days post-treatment. Age-matched and sex-matched wild-type (WT) C57BL/6 mice were used as negative controls. All harvested tissues were immediately fixed in 4% neutral buffered formalin for 24 hours, followed by routine dehydration, clearing and paraffin embedding. Automated H&E staining was performed using the SAKURA Tissue-Tek Prisma Plus Automated Slide Stainer at the Department of Pathology, Shenzhen Third People’s Hospital. Consecutive sections were cut at a thickness of 4 µm. The staining protocol on the Tissue-Tek Prisma Plus was executed using a commercial H&E staining kit (ShenZhen Beiant Medical Technology).

Isolation of monoclonal antibodies from single plasmablasts by FACS

Fresh PBMCs were stained with an antibody cocktail consisting of CD19-PE-Cy7, CD20-APC-H7, CD3-APC-H7, CD27-PE and CD38-APC (all from BD Biosciences) to sort the proliferative plasmablasts. A LIVE/DEAD Fixable Dead Cell Stain Kit (Invitrogen) was used to exclude dead cells. Flow cytometric data were acquired on an Aria II flow cytometer (BD Biosciences) and analysed using FlowJo software (Tree Star). A total of 192 enriched single plasmablasts were sorted into the 96-well PCR plate containing cell lysis buffer and then snap-frozen on the dry ice and stored at −80°C. RT-PCR and nested PCR were performed to amplify IgG antibody variable genes as previously described.^33–37 A total of 136 paired heavy and light chain variable genes were successfully amplified. After sequencing (Sangon Biotech), antibody variable genes were analysed by the IMGT/V-QUEST programme (www.imgt.org/IMGT_vquest/vquest).^{35 36 38–40} Antibody variable genes were synthesised and cloned into the expression vectors containing full-length human heavy (IgG1) and light (kappa or lambda) chains by Sangon Biotech or Sino Biological, respectively. Paired heavy-chain and light-chain plasmids were cotransfected into HEK293T or HEK293F cells to express mAbs, which were purified from the culture supernatants using protein A column (Senhui Microsphere Technology). Antigen-specific ELISA was also used to detect the binding capacity of transfected culture supernatants to yeast-derived HBsAg. Due to the strong clonal expansion, some mAbs had identical amino acid sequences. Finally, we obtained a total of 92 HBsAg-specific mAbs with 55 distinct sequences.

Clonal expansion analysis of 92 mAbs

For antibody clonotype analysis, mAbs were clustered into a clone with same usage of IGHV/IGK(L)V and IGHJ/IGK(L)J germline gene, same HCDR3 length and with at least 80% of amino-acid identity of HCDR3 sequence. The number of mAbs within a clone was defined as the cluster size. Single-lineage clustering was visualised using the R package ggraph (https://cran.r-project.org/web/packages/ggraph/index.html).⁴¹

Indirect ELISA and competition ELISA

Recombinant HBsAg proteins of yeast-derived genotype A (Shenzhen Kangtai Biological Products) or CHO-derived genotype C (Shenzhen Immuth Company) were coated into 96-well plates at 4°C overnight. The plates were washed with phosphate-buffered saline (PBS) containing 0.05% Tween 20 (PBST) buffer and blocked with 5% skim milk and 2% bovine albumin in PBS at room temperature (RT) for 1 hour. Diluted mAbs at the concentration of 10 µg/mL were added to the wells, and then incubated at 37°C for 1 hour. The plates were washed, and horseradish peroxidase (HRP)-conjugated goat anti-human IgG antibodies (ZSGB-BIO) were added and then incubated at 37°C for 1 hour. Finally, the TMB substrate (Sangon Biotech) was added to the wells and incubated at RT for 20 mins, and the reaction was stopped with 2 M H₂SO₄. The readout was detected at the wavelength of 450 nm.

For competition ELISA, four representative competitor mAbs (SY-35 of C2, SY-11 of C4, SY-22 of C7 and SY-57 of singlet) coupled with HRP (Abcam) were mixed with 55 diluted testing mAbs at the concentration of 10 µg/mL, added into the plates and then incubated at 37°C for 1 hour. The following steps were the same as the standard ELISA mentioned above. VRC01 (an HIV-1-specific mAb) was used here as the negative control, meaning no competition with HBsAg-specific mAbs. The percentage of competition was calculated by the formula: (1 − OD₄₅₀ of tested mAb / OD₄₅₀ of VRC01 control) × 100%.

Production and infection of HBV live virus

Genotype D HBV viral stocks for the infection assay were obtained from the culture medium of HepAD38 cells. HBV particles were concentrated by ultracentrifugation. The viral titre was determined by quantitative PCR, as previously described.⁴²

For HBV infection, HepG2-Na⁺-taurocholate cotransporting polypeptide (NTCP) cells were seeded and cultured in Dulbecco’s Modified Eagle Medium medium supplemented with 10% fetal bovine serum and 1% Penicillin-Streptomycin. After reaching confluency, the medium was replaced with fresh culture medium containing 2.5% Dimethyl Sulfoxide (DMSO). Subsequently, cells were infected with HBV at a multiplicity of infection of 200 in the presence of 5% PEG 8000 for 24 hours. Cells were then washed three times with PBS and further cultivated with DMSO-containing fresh culture media.

Neutralisation assay

For determining the neutralising activity of mAbs, serially diluted mAbs were preincubated with the HBV inoculum for 1 hour at 37°C. The mixture was added to the HepG2-NTCP cells and changed after 24 hours. On Day 7 post-infection, the supernatant was collected, and the HBeAg level was measured by ELISA using Diagnostic Kit (KHB). Inhibition ratio of HBeAg secretion was calculated and the curves were plotted by non-linear regression (log (inhibitor) vs response-variable slope).

Antibody-dependent cellular phagocytosis

Prepared THP-1 cells were resuspended to achieve a single-cell suspension, counted and seeded into a 24-well plate at a density of 2×10⁵ cells per well in 500 µL of medium, followed by incubation at 37°C in a CO₂ incubator for 24 hours. The test antibody was diluted to a concentration of 20 µg/mL using medium and mixed with an equal volume of CHO-derived HBsAg protein solution (1 µg/mL), followed by incubation at 37°C for 1 hour. After incubation, 250 µL of the culture medium was aspirated from each well, and 250 µL of the antibody–antigen mixture was added to the cells. The plate was then incubated at 37°C in a CO₂ incubator for 2 hours. Following the incubation, the supernatant was removed, and the cells were washed once with sterile PBS. The cells were lysed by adding 200 µL of cell lysis buffer (Western and IP Cell Lysis Buffer, Beyotime) to each well, followed by vortexing for 1 min and then incubation at 4°C for at least 15 mins. Subsequently, 200 µL of PBS was added to each well to normalise the lysate volume. For ELISA analysis of the HBsAg level, 100 µL of diluted lysate was measured by the HBsAg ELISA Kit (Beijing Wantai Biotech). The readout was detected at the wavelength of 450 nm.

Binding capacity of mAbs to cell-expressed HBsAg by flow cytometry

HEK293T cells were transiently transfected with HBsAg expressing plasmids of genotypes A to J (HBV-A, GenBank: X02763.1; HBV-B, GenBank: AB073846.1; HBV-C, GenBank: FJ386685.1; HBV-D, GenBank: MN645906.1; HBV-E, GenBank: X75657.1; HBV-F, GenBank: AB036910.1; HBV-G, GenBank: AF160501.1; HBV-H, GenBank: AY090454.1; HBV-I, GenBank: AF241411.1; HBV-J, GenBank: AB486012.1). The flow cytometry assay was performed as previously described.⁴³ Briefly, these transfected cells were fixed and permeabilised, and then incubated with 5 µg/mL mAbs or polyclonal antibodies (pAbs), followed by the staining of Alexa Fluor 647-conjugated secondary antibody (Thermo Fisher Scientific). All the staining was 30 mins at 4°C. Living cells were classified using the LIVE/DEAD Fixable Violet Dead Cell Stain kit (Thermo Fisher Scientific). Data were acquired by FACSymphony A3 (BD Biosciences) and analysed by FlowJo software V.10.9 (BD Biosciences).

Construction of site-directed expressing plasmids

Mutations were constructed by the site-directed mutagenesis based on the above genotype A HBsAg, SY-23 heavy chain and SY-23 light chain expressing plasmids using the Mut Express II Fast Mutagenesis Kit V2 (Vazyme Biotech).

Surface plasmon resonance

The binding affinity of mAbs to CHO-derived HBsAg protein was performed through Biacore 8K system (GE Healthcare). The HBsAg protein was covalently immobilised on a single flow cell of the CM5 sensor chip channel using 10 mM sodium acetate buffer (pH 5.0), achieving a final response unit of approximately 1000. The other flow cell remained unmodified and was blocked as a control. Subsequently, mAbs were serially diluted with 10×HBS EP buffer (10 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.05% Tween-20) and injected into the chip at a flow rate of 30 µL/min for a 60 s. The binding kinetics was analysed using Biacore Evaluation software (GE Healthcare) with a 1:1 binding model to calculate equilibrium constant (K_D), association constant (K_on) and dissociation constant (K_off). Each experiment was performed in triplicate, and the results are presented as mean±SD.

Preparation of fragment of antigen binding of mAbs

Fab-form mAbs were digested from purified IgG-form mAbs. IgGs were diluted to 1 mg/mL and incubated with L-Cysteine hydrochloride (final concentration 20 mM, Sigma-Aldrich), EDTA (final concentration 20 mM, Invitrogen) and papine (final concentration 1.25 µg/mL, Sigma-Aldrich) at 37°C for 12–14 hour. The reaction was stopped by adding 0.5 M Iodoacetamide (Sigma-Aldrich). Fabs were purified by using protein A affinity chromatography to separate the fragment crystallisables and redundant undigested IgGs. Unless otherwise specified, the description of mAbs indicates its IgG form.

Cryo-EM sample preparation and data collection

The HBsAg protein was dissolved from yeast-derived HBsAg particles.⁴⁴ The purified Fab and HBsAg particles were mixed with a molar ratio of 1:1.5 and incubated at RT for 30 mins. Subsequently, the mixture was solubilised at RT for at least 7 hours with 1% n-dodecyl-β-D-maltoside. The sample was centrifuged at 17 000 g for 10 min, and the supernatant was further purified by size-exclusion chromatography (SEC) using a Superose 6 Increase 10/300 GL column (GE Healthcare) in a buffer containing 25 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.01% GDN. The selected SEC fractions were collected and subjected to negative-stain EM analysis using a Talos 120C microscope (Thermo Fisher) with a pixel size of 1.52 Å. Automated particle picking from 3× or 4× binned images and two-dimensional classification were performed with SAMUEL and SamViewer.⁴⁵ The initial three-dimensional models were generated with two-dimensional class averages by SPIDER 3D projection matching refinement (samrefine.py). According to the negative-stain EM images, only the SEC fractions eluted at around 15∼16 mL contained relatively homogeneous HBsAg complex containing two Fabs. Such fractions were pooled and concentrated to ∼12 mg/mL for cryo-EM grid preparation. 3.5 µL of purified complex protein was applied to a freshly glow-discharged Quantifoil holey carbon grid (R1.2/1.3, 300 mesh, Au) (Quantifiol, Jena, Germany). Grids were blotted for 5 s with 100% humidity and plunge-frozen in liquid ethane using Vitrobot (Thermo Fisher Scientific).

Cryo-EM data were collected at liquid nitrogen temperature on a Tatan Krios transmission electron microscope (Thermo Fisher Scientific), operated at 300 kV and equipped with K3 direct electron detector (Gatan). The movies were recorded in super-resolution counting mode using K3 with a calibrated pixel size of 0.855 Å on the specimen level and 0.428 Å for super-resolution images. The dose rate was set to be 25 electrons per physical pixel per second. The total exposure time of each movie was 1.46 s, leading to a total accumulated dose of 50 electrons per Ų, fractionated into 32 frames. All movies were recorded with a defocus ranging from 1.0 to 2.0 µm. The detailed data collection parameters are listed in online supplemental table S1.

Cryo-EM data processing

Motion correction and dose weighting were performed using MotionCor2.⁴⁶ Defocus values were calculated using CTFFIND4.⁴⁷ Two-dimensional classification of selected particle images was carried out by ‘sambatch2d.py’, which uses SPIDER operations to run ten cycles of correspondence analysis, k-means classification and multireference alignment. Initial three-dimensional models were generated with two-dimensional class averages by SPIDER 3D projection matching refinement (samrefine.py). Three-dimensional classification and refinement were carried out in RELION-5.0.⁴⁸ After sorting the 4× binned particle images, particle stacks were generated for further three-dimensional classifications. Three-dimensional classification was carried out with K=6 for 35 cycles in RELION V.5.0 and good classes were selected for the next round of three-dimensional classification after keeping all particles assigned in those selected classes within the final five cycles. After selecting particles from the final round of three-dimensional classification, we performed a further three rounds of three-dimensional classification with original, unbinned particles using C2 symmetry, which was followed by auto-refinement. Map sharpening was performed using ‘relion_postprocess’ and yielded an overall resolution of 3.9 Å. Detailed image processing parameters related to cryo-EM image processing are summarised in online supplemental table S1.

Model building, refinement and display

The initial model of the HBsAg-SY-23 complex was generated by separately predicting the structures of HBsAg and SY-23 Fab using AlphaFold2. HBsAg was predicted as a dimer, while the SY-23 Fab was predicted independently. The predicted models were then assembled by docking into the cryo-EM maps using UCSF Chimera.^{49 50} Manual model building and adjustment were then performed using Coot.⁵¹ The structures were refined by real-space refinement using PHENIX,⁵² and the model was improved by several rounds of manual adjustment in Coot and refinement using phenix.real_space_refine. All structure figures were prepared in PyMOL (https://pymol.org/2/), UCSF Chimaera and ChimeraX.⁵³ The model refinement statistics are summarised in online supplemental table S1.

Cryo-electron tomography data collection and processing

Optimised grids were loaded into a Titan Krios 300 kV transmission electron microscope (Thermo Fisher Scientific) equipped with a Gatan imaging filter (GIF) Quantum LS and a Gatan K3 detector. The beam was aligned in nanoprobe mode, and the GIF slit width was set to 20 eV. The tomography data collection was performed by SerialEM V.3.9,⁵⁴ with a dose-symmetric scheme.⁵⁵ Tilt series were collected at a calibrated pixel size of 1.1 Å or 1.5 Å, with a total dose of ∼100 e/ Ų applied over entire series collected between ±60° with 3° tilt increments. In total, 23 tilt series and 30 tilt series were collected for CHO-derived HBsAg particles alone and SY-23-Fab-bound particles, respectively. Tilt series movies were recorded in dose-fraction mode, with a defocus range from −3 to −5 µm.

Frames in each movie were aligned for drift correction with the GPU-accelerated software MotionCor2.⁴⁶ Fiducial-less tilt series alignment was performed by AreTomo,⁵⁶ and tomograms were reconstructed by IMOD.⁵⁷ For better visualisation, the reconstructed tomograms were binned to a pixel size of 9 Å, which were then denoised and corrected for the missing wedge artefact with IsoNet⁵⁸ using defocus values estimated with CTFFIND4.⁴⁷ 51 HBsAg particles and 51 SY-23-Fab-bound particles were manually picked, and the diameter of each particle was measured in IMOD. Segmentation was performed using UCSF ChimeraX.⁵³ Due to the minimisation of missing wedge artefacts and improved contrast, individual spikes on particles were identified without ambiguity. 37 HBsAg particles and 37 SY-23-Fab-bound particles with well-resolved spikes were included for interactive spike counting.

Results

Isolation of 92 HBsAg-specific mAbs from plasmablasts

A previously vaccinated volunteer was boosted with a dose of yeast-derived recombinant hepatitis B vaccine containing 20 µg of genotype A HBsAg protein, whose plasma antibody titres against HBsAg were markedly increased from 3.1 IU/mL to 14.6 IU/mL after 7 days of vaccination, suggesting a robust and effective recalled antibody response. In this process, reactivated HBsAg-specific MBCs differentiated to abundant plasmablasts. After this rapid clonal expansion, most plasmablasts in peripheral blood might be HBsAg-specific. Therefore, in this study, single activated plasmablasts gated by CD19⁺CD20⁻CD3⁻CD27^hiCD38^hi were sorted from fresh PBMCs collected on Day 7 post vaccination (figure 1A). A total of 92 HBsAg-specific IgG mAbs were isolated using previously established methods.^{35 39 59 60} To characterise the genetic features of these anti-HBsAg mAbs, we analysed their variable genes of heavy and light chains using the IMGT/V-QUEST programme. As shown in figure 1B, this booster recalled a wide range of antibody response. A total of 11, 8 and 4 germline genes of heavy, kappa and lambda chains were used, respectively, some of which were overrepresented, such as IGHV4-39 for the heavy chain, IGKV2-28 for the kappa chain and IGLV7-46 for the lambda chain. A preferred pairing between IGHV4-39 and IGLV7-46 was identified, reaching as high as 51% (47/92) among these mAbs, followed by IGHV3-15/IGKV2-28 (15%), IGHV3-30/IGKV3-20 (10%) and IGHV3-23/IGLV2-8 (7%) (figure 1C). Then, we further performed a clonal expansion analysis of these 92 mAbs. As shown in figure 1D,E, 9 distinct clonotypes were observed and occupied the most part as high as 87% in 92 mAbs, and the rest of 12 were singlets. Two IGHV4-39/IGLV7-46-related clonotypes, named C1 and C2, were dominant at 40% (37/92) and 11% (10/92), respectively, followed by IGHV3-30/IGKV3-20 (C3: 10%), IGHV3-15/IGKV2-28 (C4: 8%), IGHV3-23/IGLV2-8 (C5: 5%), etc. These results indicated that 92 human-derived HBsAg-specific IgG mAbs were successfully identified from ongoing clonal expansion plasmablasts of a vaccinated volunteer.

Figure 1. Isolation, germline gene usage and clonal expansion analysis of 92 HBsAg-specific mAbs from plasmablasts in a vaccinated volunteer. (A) Gating strategy for isolating the enriched plasmablasts (CD19⁺CD20⁻CD3⁻CD27^hiCD38^hi) from fresh PBMCs collected on Day 7 post administration with a booster dose of yeast-derived hepatitis B vaccine in a vaccinated volunteer. (B) Germline heavy and light gene usage and pairing among 92 HBsAg-specific mAbs, presented in the chord diagram. Each of the paired germline heavy and light chains is linked by arcs, the sizes of which are proportional to the total mAbs. (C) Germline heavy and light gene usage and pairing among 92 HBsAg-specific mAbs, presented in the heat map. The number and colour represent the counts of mAbs with paired heavy and light chains. (D, E) Clonal expansion analysis of 92 HBsAg-specific mAbs, presented in the graph (D) and in detail (E). Each point represents one mAb. Interconnecting lines indicate the relationship among mAbs sharing same V and J gene at both heavy and light chains, same HCDR3 length and ≥80% amino acid identity of HCDR3. FSC-A, forward scatter area; FSC-W, forward scatter width; HBsAg, HBV surface antigen; mAbs, monoclonal antibodies; PBMCs, peripheral blood mononuclear cells; SSC-A, side scatter area.

Characteristics and functions of 92 HBsAg-specific mAbs

Due to the above-mentioned clonal expansion, some mAbs shared identical amino acid sequences. Therefore, 55 mAbs with distinct sequences were expressed and purified, and then comprehensively measured for their biological activities. As shown in figure 2, almost all mAbs displayed the cross-reactivity of binding to both yeast-derived genotype A and CHO-derived genotype C HBsAg proteins, except for SY-10 and SY-21. Only nine mAbs could not inhibit the HepAD38-derived genotype D HBV live virus entry into targeted cells at the highest concentration of 50 µg/mL in vitro, particularly those singlet mAbs (online supplemental figure S1). A similar result was found in another functional assay. A majority of mAbs (91%, 50/55) displayed effective ADCP activities, mediating the intake of phagocytes to HBsAg particles in vitro. Almost all clonally expanded mAbs had presentable neutralisation and ADCP activities except for several mAbs in C6 and C9 clones, suggesting that these HBsAg-specific mAbs derived from proliferative plasmablasts were functional in directly fighting against the HBV infection.

Figure 2. Gene family analysis, binding capacity, neutralisation and ADCP of 55 plasmablast-derived HBsAg-specific mAbs. The germline gene usage (IGHV, IGKV and IGLV), the length of CDR3 and the proportion of SHMs were estimated using the IMGT programme. Binding capacity to yeast-derived genotype A and CHO-derived genotype C HBsAg proteins was measured by ELISA at the antibody concentration of 10 µg/mL. Neutralisation against the HepAD38-derived genotype D HBV was measured by the inhibition assay of live virus infection in vitro, whose cut-off value of IC₅₀ was set as 50 µg/mL. ADCP of mAbs mediating the phagocytes taking in CHO-derived genotype C HBsAg particles was measured by ELISA at the antibody concentration of 10 µg/mL. ‘*’ indicates the representative mAb of each clonal cluster and ‘count’ means the number of mAbs sharing the identical amino acid sequence of heavy and light chains. All results are calculated from at least two independent experiments. ADCP, antibody-dependent cellular phagocytosis; CDR3, complementarity determining region 3; HBsAg, HBV surface antigen; IC₅₀, inhibitory concentration; IMGT, the ImMunoGeneTics information system; mAbs, monoclonal antibodies; SHMs, somatic hypermutations.

In each clonotype, mAbs had the maximum genetic similarity sharing the same V and J germline genes and the length of complementarity determining region 3, as well as comparable somatic hypermutations, contributing to their consistent biological functions (figure 2). The neutralising activities of two dominated IGHV4-39/IGLV7-46-related clonotypes were not the most potent, whose geometric means of 50% inhibitory concentration (IC₅₀) were 0.715 µg/mL and 0.421 µg/mL, respectively. A dozen of mAbs from C3 and C8 clonotypes displayed potent neutralisation against the HBV infection with IC₅₀s of 0.047 µg/mL and 0.023 µg/mL, respectively, despite different gene usages. Based on the divergence across the entire genome, HBV could be classified into nine major genotypes (A to I) and one putative genotype (J).⁶¹ The amino acid sequence identity of HBsAg among different genotypes was 94.20% (figure 3A). To further assess the cross-reactivity of these mAbs, 9 mAbs from clones and 12 mAbs from singlets were selected as representatives to measure their binding capacities to different genotype HBsAg proteins transiently transfected in HEK293T cells by flow cytometry analysis (figure 3B,C, and online supplemental figure S2). The commercial pAbs could effectively recognise these HBsAg proteins from A to J genotypes, suggesting that all HBsAg proteins were successfully expressed in HEK293T cells. SY-51, SY-10, SY-52 and SY-21 could not bind to any genotype HBsAg, which might be not suitable in the flow cytometry assay. The binding of SY-54 and SY-50 was obviously decreased to the genotype B and genotype G HBsAg proteins. Overall, the rest of mAbs bound to all tested genotype HBsAg proteins with comparable abilities, suggesting their excellent cross-genotypic activities. These results indicated that these human HBsAg-specific mAbs, especially SY-4-class and SY-23-class, could be good candidates for antibody drugs.

Figure 3. Cross-reactivity of 21 human mAbs to different HBsAg from genotypes A to J. (A) The amino acid sequence alignment of HBsAg from genotypes A to J was performed by Clustal W and created by ESPript V.3.0. (B) The cross-reactivity of mAbs binding to different genotypes of HEK293T-derived HBsAg proteins was measured by flow cytometry. Experiments are independently performed at least twice, and one representative result is shown. (C) The percentage of HBsAg-specific cells was shown by the heat map. FSC-A, forward scatter area; HBsAg, HBV surface antigen; mAbs, monoclonal antibodies; pAbs, HBsAg-specific polyclonal antibodies.

Binding affinity and potential treatment of representative mAbs

The binding affinity is an important index for evaluating the potential of antibody drugs.⁶² CHO-derived HBsAg proteins from genotype C were immobilised to the sensor chip and their binding kinetics for six clonally expanded mAbs with distinct germline gene usages were determined by the surface plasmon resonance (SPR). As shown in online supplemental figures S3, 6 representative mAbs bound to HBsAg proteins with high affinities ranging from 0.24 nM to 35.23 nM of dissociation constants (K_Ds). Among these mAbs, SY-4 and SY-23 exhibited potent neutralising activities in vitro (figure 2).

To assess their potential treatment in vivo, we generated the HBV-persistent C57BL/6 mice based on the transduction of a recombinant genotype D HBV-encoding AAV. A total of 12 HBV carrier mice with a certain level of serum HBsAg (approximately 10³ IU/mL to 10⁴ IU/mL) were administrated with a single dose of SY-4, SY-23 or irrelevant control mAbs at 20 mg/kg by the tail vein injection, respectively. A series of serum samples were collected at 0, 1 and 2 days postinjection (dpi) and monitored the level of circulating HBsAg, HBV DNA, HBV pgRNA and HBeAg. As shown in online supplemental figure S4A, both SY-4 and SY-23 rapidly led to a marked drop of 2.8 log₁₀ of HBsAg at 1 dpi, whereas the irrelevant mAb showed no effect. Correspondingly, HBV DNA levels markedly decreased on Day 1 following the treatment of SY-4 (2.4 log₁₀) and SY-23 (2.8 log₁₀), although there was a slight fluctuation (0.6 log₁₀ reduction) in the control group (online supplemental figure S4B). A similar trend was observed for HBV pgRNA levels, which declined by 2.7 log₁₀ and 3.2 log₁₀ in the SY-4 and SY-23 groups, respectively (online supplemental figure S4C). By contrast, HBeAg levels remained relatively stable in all groups (online supplemental figure S4D). As the decline of passive SY-4 and SY-23 and the persistent production of HBV, serum HBsAg, HBV DNA and HBV pgRNA levels partially rebounded from 2 dpi (online supplemental figure S4A–C). There were no significant changes in serum ALT levels in all groups throughout the experimental process (online supplemental figure S4E), suggesting no obvious hepatocellular injury in this AAV-HBV mice model. These results indicated the potent antiviral activity of SY-4 and SY-23 in vitro translated into strong treatment in vivo, especially contributing to the rapid clearance of circulating HBsAg, HBV DNA and HBV pgRNA.

To further evaluate the hepatic inflammatory status and explore the change of intrahepatic HBV, liver tissues were collected and performed the H&E staining assay. As shown in online supplemental figure S5A–D, liver H&E staining at 2 dpi revealed well-preserved liver architecture and cellular morphology in all antibody treatment groups (SY-4, SY-23 and irrelevant control), compared with the C57BL/6 mice without AAV-HBV challenge (negative control). No obvious histopathological abnormalities, such as inflammatory infiltration, hepatocyte ballooning, necrosis or disorganisation of the hepatic lobular structure, were observed in any antibody-treated mice, indicating the absence of detectable immune-mediated liver injury. In addition, all mice exhibited normal behavioural activity at the administered dose level of mAbs (20 mg/kg per mouse), which was safe during the animal experiments. The relative levels of intrahepatic HBV pgRNA were also detected at 2 dpi. The intrahepatic HBV pgRNA levels in the SY-4-treated and SY-23-treated mice were slightly higher than those in the irrelevant mAb-treated group (online supplemental figure S5E), which was consistent with the preinjection serum HBV pgRNA levels (online supplemental figure S5F). These results indicated that HBsAb injection via tail vein did not affect the transcription of intrahepatic HBV pgRNA in this AAV-HBV mice model.

Structural basis of SY-23 binding to HBsAg

For recognition epitope mapping, SY-35 (C2), SY-11 (C4), SY-22 (C7) and SY-57 (singlet) were coupled with HRP, and their competitiveness for binding to HBsAg was measured with all 55 mAbs using competition ELISA. As shown in online supplemental figure S6, several clusters of HBsAg-specific mAbs were observed according to the results of clustering analysis. Notably, SY-4-class and SY-23-class mAbs with most potent neutralisation belonged to the same cluster, revealing an important antiviral target on the HBsAg. To understand the structural basis for HBV neutralisation by these mAbs, SY-23 was first selected as a representative for cryo-EM study of its binding to HBsAg proteins.

For obtaining a sufficient amount of HBsAg protein to form a complex with the Fab of SY-23, we initially solubilised the yeast-derived HBsAg particles⁴⁴ in detergents but failed to generate HBsAg proteins in any defined oligomer state, likely due to strong association between neighbouring HBsAg proteins on particles. After extensive experimentation, we developed a workflow in which HBsAg particles were bound with SY-23-Fab before detergent solubilisation and purification with SEC (online supplemental figure S7A–D). This strategy enabled us to isolate two complexes of HBsAg bound to two or four SY-23-Fabs, as visualised in negative-stain EM images (online supplemental figure S7E). The HBsAg complex with two SY-23-Fabs was more homogeneous and subjected to single-particle cryo-EM analysis (online supplemental figure S8).

Meanwhile, we also purified HBsAg-SY-4-Fab complex using the same method as for HBsAg-SY-23-Fab complex and preformed negative-stain EM analysis (online supplemental figure S9). Similar to HBsAg-SY-23, HBsAg-SY-4 displayed a characteristic Y-shape, with two SY-4-Fabs on either side, suggesting that SY-4 also binds to the AGL region on the top of the HBsAg dimer. However, despite much effort, we failed to obtain a high-quality sample of dimeric HBsAg-SY-4-Fab complex that is homogeneous enough for cryo-EM structure determination.

The final cryo-EM map of HBsAg bound to SY-23-Fab was determined at 3.9 Å resolution (figure 4A,B, online supplemental figure S8F and table S1). The overall architecture of the complex exhibits a characteristic Y-shape, with one HBsAg-dimer at the centre and two SY-23-Fabs on either side. In our cryo-EM map, the central part of HBsAg (residues 73–170), containing TH2 and the entire AGL involved in SY-23-Fab binding, is well resolved, enabling reliable atomic model building (online supplemental figure S10). Except TH2, the transmembrane regions of HBsAg are not visible, suggesting their flexibility in this HBsAg-SY-23-Fab complex, but the protein sequences in the membrane nevertheless contribute to the large micelle density (figure 4A). TH2 contributes to dimer formation through hydrophobic interactions of F85, L88 and I92, including the aromatic ring stacking between two F85 residues (online supplemental figure S10A). Compared with previously reported structures,^{31 32} TH2 in our structure is positioned closer to the dimer interface, likely an effect from SY-23-Fab induced stabilisation (online supplemental figure S11A).

Figure 4. Cryo-EM structure of SY-23-Fab bound HBsAg-dimer. (A) Cryo-EM map of yeast-derived HBsAg-dimer bound to two SY-23-Fabs filtered to 3.9 Å resolution, viewed from the side (upper panel) and the top (lower panel). Two HBsAg proteins are in blue and light orange. One SY-23-Fab is coloured in green (heavy chain) and light green (light chain), and the other Fab is coloured in pink (heavy chain) and magenta (light chain). The weaker densities displayed at lower contour levels are shown as outline. (B) Side view of the HBsAg-SY-23-Fab structure in ribbon representation, coloured as in A. The Fabs are shown with only their variable regions. The selected areas indicated with dashed boxes are shown in G and I. (C) The arrangement of secondary structures in the HBsAg regions which were resolved in the cryo-EM map of the HBsAg-SY-23-Fab complex. (D) Atomic model of the HBsAg regions as in C. The selected areas indicated with dashed boxes are shown in E and F. (E) Superimposition of EM densities and atomic models of β1 (green) and residue 101–110 of AGL (cyan). (F) Superimposition of EM densities and atomic models of EH1 (red). (G) Hydrophobic interactions between the two AGLs. The amino acids from EH1 are labelled in red, and those from the loop region labelled in green. (H) Intermolecular disulfide bond formed by C139 from two HBsAg molecules, intramolecular disulfide bond formed by C107-C138 in one HBsAg molecule. (I) Interaction details between EH1 and β2 of HBsAg (blue) and HCDR1 and HCDR2 of SY-23-Fab (pink). AGL, antigenic loop; cryo-EM, cryo-electron microscopy; Fab, fragment of antigen binding; HBsAg, HBV surface antigen; TH2, second transmembrane helix.

The AGL of HBsAg plays crucial roles in HBsAg oligomerisation and antibody recognition, but the structural details of AGL and its interaction with antibody have long been missing. In the HBsAg complex with SY-23-Fab, the AGLs at the centre were stabilised by antibody binding from both sides, resulting in high-quality EM density, which is in sharp contrast to the previously reported cryo-EM maps of HBsAg (online supplemental figure S11B).^{31 32} Our cryo-EM structure provides unprecedented structural insights into how AGLs contribute to the formation of HBsAg-dimer and critical antigen epitopes. The AGL (residues 100–169) consists of two β-sheets (β1, 137–140; β2, 148–151), one α-helix (EH1, embedded helix, 157–169) and three loops connecting them (figure 4C–F). Right above the membrane, AGL dimer contact forms as a network of hydrophobic interactions involving the loop sequence after TH2 (M103, L104, P105, V106 and L109) and EH1 (F158, F161, L162 and W165) (figure 4E–G). The abundance of disulfide bonds is a characteristic feature of AGL. Our cryo-EM map clearly resolved an intermolecular disulfide bond between the C139 residues on β1 which covalently links two HBsAg proteins at the dimer interface (figure 4H and online supplemental figure S10A), as well as an intramolecular disulfide bond formed between C107 and C138 (figure 4E,H, and online supplemental figure S10A). Notably, these three disulfide bonds appear to lock the two β1 strands in position and stabilise the overall AGL architecture. These observations differ from the prevailing assumptions that C107 forms intermolecular disulfide bonds and C139 and C147 form an intramolecular disulfide bond.³¹

Our cryo-EM structure reveals that SY-23-Fab binds to HBsAg with the heavy chain primarily interacting with the β2 and EH1 regions in the AGL of HBsAg (figure 4I). The binding interface involves the residues in HCDR1 (R30, S31, Y32) and HCDR2 (I54, F55) from SY-23 and β2 (P151) and EH1 (K160, W163, E164) from AGL. S31 and Y32 in HCDR1 form hydrogen bonds with K160 and P151 in AGL, respectively. The binding is stabilised by a salt bridge between R30 in HCDR1 and E164 in EH1, as well as the hydrophobic contact contributed by I54 and F55 in HCDR2 and W163 in EH1. All residues in the interactions between HBsAg and SY-23-Fab are summarised in online supplemental table S2.

Previously, there was no structure for antibody-bound HBsAg, and the structural information was only available for H015-Fab bound to an AGL peptide (PDB: 6VJT).¹⁸ Compared with the binding region of SY-23, the crystal structure of H015-peptide complex indicates a completely different antigenic epitope (online supplemental figure S11C). Notably, the β-hairpin fold peptide seen in the crystal structure displays a similar conformation to the sequence between β1 and β2 of the AGL in our structure (online supplemental figure S11D).

Validation of key amino acid residues at the interface between SY-23 and HBsAg

To further investigate the roles of the amino acid residues of HBsAg involved in SY-23 binding (online supplemental table S2), we first generated eight variants of HBsAg (Q129A, G130A, P151A, P153A, W156Q, K160A, W163A and E164K). These variants fused with the red fluorescent protein (RFP) at the N terminal were transiently expressed in HEK293T cells and measured by the flow cytometry. All HBsAg variants could be detected by the commercial pAbs, and the ratio of HBsAg-positive percentage to RFP-positive percentage among these variants and WT HBsAg showed no obvious difference (online supplemental figure S12A,B). In contrast, the ratios of HBsAg-positive percentage to RFP-positive percentage of K160A and E164K variants were markedly lower than that of the WT group (figure 5A and online supplemental figure S12C). The ratios of other variant groups displayed little or no difference to that of WT. These results demonstrated that single point mutations on K160 and E164 residues of HBsAg could weaken the binding capacity of SY-23.

Figure 5. Verification of critical residues for the interaction between SY-23 and HBsAg. (A) The binding activity of SY-23 to HBsAg (genotype A) and its variants detected by flow cytometry. Eight variants (Q129A, G130A, P151A, P153A, W156Q, K160A, W163A, E164K) and WT genotype A HBsAg fused with RFP at the N terminal were transiently transfected into HEK293T cells and detected by flow cytometry. Upper: HBsAg detected by SY-23. Bottom: RFP detection. (B) ELISA analysis of the binding activities of SY-23 and its variants to HBsAg. 10 single-mutated variants of SY-23 (heavy chain: R30E, S31A, Y32A, V52Q, I54Q, F55Q, R74E, E110K; light chain: Y49A, T56A) and one dual-mutated variant (R30E+Y32A) on the heavy chain were constructed. Then, the binding activities of these variants and WT SY-23 to HBsAg were evaluated by ELISA. Left: the binding activities of antibody to yeast-derived genotype A HBsAg. Right: the binding activities of antibody to CHO-derived genotype C HBsAg. (C) The EC₅₀s of SY-23 and its variants to HBsAg from (B). Upper: the EC₅₀ of antibody to yeast-derived genotype A HBsAg. Bottom: the EC₅₀ of antibody to CHO-derived genotype C HBsAg. Experiments are independently performed at least twice and one representative result is shown. FSC-A, forward scatter area; EC₅₀, 50% of maximal effect concentration; HBsAg, HBV surface antigen; RFP, red fluorescent protein.

To further ascertain the critical residues of SY-23 for HBsAg binding (online supplemental table S2), a series of antibody variants were carried out, including R30E, S31A, Y32A, V52Q, I54Q, F55Q, R74E, E110K mutations on the heavy chain and Y49A, T56A mutations on the light chain. These antibody variants were expressed, purified and checked by variations in 50% of maximal effect concentration (EC₅₀) values. As shown in figure 5B,C, three mutants R30E, Y32A and F55Q showed dramatically lower binding capacities to both yeast-derived genotype A and CHO-derived genotype C HBsAg. I54Q and R74E mutants showed moderately lower binding capacities to both genotype A and C HBsAg, while S31A, V52Q and T56A mutants showed no obvious effect on reducing the antibody binding activity. The EC₅₀ values of R30E, Y32A and F55Q mutants binding to genotype A HBsAg were more than six times higher than that for WT SY-23, meanwhile the EC₅₀ values of these mutants binding to genotype C HBsAg were more than 200 times higher than that for WT SY-23. Furthermore, a dual mutant (R30E and Y32A) was conducted and showed no binding activity to the genotype A and C HBsAg at the highest concentration of 10 µg/mL, which supports the results of single mutations. As mentioned above (figure 4), R30, S31 and Y32 of SY-23 heavy chain form hydrogen bond or salt bridge interactions with HBsAg. Our results demonstrate that the key amino acid residues in HCDR1 of SY-23 make a major contribution to the binding of HBsAg.

Cryo-electron tomography analysis of SY-23 bound HBsAg particles

In an attempt to understand how SY-23-Fab may bind to HBV, we docked our HBsAg-SY-23-Fab complex structure into previously proposed HBsAg particle model using TH2 as the reference,³¹ and found that the model is structurally incompatible with full occupancy of SY-23-Fab on every HBsAg spike, due to pronounced steric clashes (online supplemental figure S13A).

To visualise SY-23-Fab binding to intact HBsAg particles, the CHO-derived HBsAg particles alone and SY-23-Fab-bound particles were subjected to cryo-electron tomography and three-dimensional reconstruction. HBsAg particles display similar sizes and shapes with a small range of variations. With a diameter of ∼240 Å, the observed particles are not perfectly spherical and slightly larger than the theoretical 22 nm D2 particle³¹ (online supplemental figure S13B,C,F). These particles each contained ∼62 HBsAg spikes, and, upon the binding of SY-23, showed ∼43 Fab-bound spikes (online supplemental figure S13G). The SY-23-Fabs on the HBsAg particles displayed a mix of inclined and vertical orientations (online supplemental figure S13D,E). While the inclined orientation is consistent with our cryo-EM structure of SY-23-Fab bound to HBsAg, the vertically oriented Fabs are present in pairs, suggesting that they bind to two neighbouring HBsAg spikes. Interestingly, such parallel Fabs were also observed in the negative-stain EM images of presumable HBsAg tetramer complex with four SY-23-Fabs (online supplemental figure S7E, left). These observations suggest that two Fabs on neighbouring spikes can become perpendicular to the membrane to avoid clash.

Discussion

When this volunteer received a booster dose of hepatitis B vaccine, pre-existing HBsAg-specific MBCs in the body were reactivated and rapidly differentiated into plasmablasts. Along with a robust clonal expansion, a large number of anti-HBsAg plasmablasts existed in the peripheral blood, many of which carried the similar or even identical antibody amino acid sequences. Although without the use of a specific antigen probe, we successfully isolated and cloned a series of HBsAg-specific mAbs from these plasmablasts. We identified at least nine clonal clusters and several singlets of mAbs, and then confirmed that the vast majority of these mAbs possessed effective antiviral functions including neutralisation and ADCP. These pre-existing HBsAg-specific MBCs, activated plasmablasts and secreted mAbs with varied antiviral activities and binding epitopes together formed a necessary line of defence against the potential HBV infection. These findings provided us with a new perspective for evaluating the potential protection of HBV vaccine-induced mAbs during the antibody recalling phase, compared with previous research on isolating HBsAg-specific mAbs from peripheral MBCs.^{18–21 24} Although anti-HBsAg neutralising antibodies do not directly block the preS1 region of HBV binding to cellular receptor, human NTCP, and also do not directly interfere with the intracellular HBsAg dimerisation or particle assembly, they mainly destroy the initial attachment of HBsAg to the heparan sulfate proteoglycans on hepatocytes, which is the first step of HBV entry.^{19 24 63 64}

Previous studies showed that several IGHV3 germline genes, such as IGHV3-23, IGHV3-30 and IGHV3-33, were often used by human anti-HBsAg mAbs,^{8 18 19 65 66} which were also found in our antibody panel. For example, SY-4-class clonally expanded mAbs with potent neutralising activities shared the common immunoglobulin gene usage of IGHV3-30. Besides, another potent SY-23-class mAbs belonged to the IGHV1-69 germline, which was also used by both HBsAg-specific mAbs¹⁹ and other virus-specific mAbs including HIV type-1 (HIV-1), HCV and influenza virus.^{38 67–69} By contrast, a relatively uncommon IGHV4-39-derived clonotype was markedly dominated in the HBsAb repertoire of this individual. This does not rule out the sample selection bias. IGHV4-39-derived clonotype is probably not the unique feature of HBsAg-specific mAbs from plasmablasts. Although there is not obvious enrichment in other studies, a considerable number of MBC-derived HBsAg-specific mAbs use the IGHV4-39 germline gene.^{18 19 65} These public and private clonal clusters contributed to the diversity of B-cell repertoire, creating the potential for recognising and fighting against the HBV infection in different individuals.

The hepatitis B immune globulin (HBIG) is derived from plasma of HBV vaccine recipients with high levels of HBsAb, which is usually used in newborns from HBsAg-positive mothers, non-immune people exposed to HBV and patients with liver transplantation for HBV-related disease.^{6 70 71} However, the HBIG is a mixture of pAbs with complex components. In recent years, human anti-HBsAg mAbs have been deeply developed, aiming to replace HBIG.^{18–21 23 24} It should be noted that the antiviral potency of mAbs does not directly depend on the capability of clonal expansion mentioned above. Actually, some mismatches between clonal dominance and antiviral activity have been widely reported in HBsAg-specific and other virus-specific mAbs. For instance, Hehle et al previously identified a potent HBsAg-specific neutralising antibody, Bc1.187, deriving from the IGHV3-33 germline gene rather than the dominant clonotype IGHV1-69 in the antibody repertoire.¹⁹ It is usually considered that the binding epitope and affinity of mAbs contribute to their antiviral potencies, which has been systematically analysed in SARS-CoV-2-specific antibodies. Some mAbs competing with cellular receptor often displayed more potent antiviral activities than those not competing with receptor.^72–74 When mAbs recognised the same binding epitopes, mAbs possessing relatively higher affinity exhibited stronger antiviral potency.^{75 76} In this study, two classes of mAbs represented by SY-4 and SY-23 displayed the potent binding affinity, neutralisation, ADCP and cross-genotypic activities in vitro, as well as the strong ability of decreasing the serum HBsAg, HBV DNA and HBV pgRNA in the AAV-HBV mice model, which could be good candidates for the development of therapeutic antibody drugs. Except for neutralisation, ADCP function is also regarded as a kind of protective effector mechanism contributing to the viral clearance in HBV mice model.^{77 78} While both ADCP and antibody-dependent enhancement (ADE) involve the interaction of the antibody Fc domain with some FcγRs on immune cells, their virological outcomes are fundamentally different and determined by the specific virus and the permissiveness of the FcγR-bearing cells.⁷⁹ ADE was first clearly described in the Dengue virus infection,^{80 81} where subneutralising or cross-reactive non-neutralising antibodies could form complexes with virions and exploit activating FcγRs (eg, FcγRIIa) on myeloid cells for internalisation, leading to more enhanced viral infection and pathogenesis.⁸² In contrast, the mechanism for HBV is really distinct. FcγR-expressing cells like macrophages and other leukocytes are not permissive for productive HBV replication. Thus, this kind of FcγR engagement results in antiviral clearance without amplifying the viral lifecycle.⁸³

Based on our cryo-EM structure of SY-23-Fab bound HBsAg-dimer, we have obtained the long-sought-after structural information for AGL-mediated HBsAg dimer contact and the HBsAg-antibody interactions. In recent years, cryo-EM studies of HBsAg particles and HBsAg-dimer have made significant progress,^30–32 revealing the mechanism underlying HBsAg-dimer interaction and assembly into subviral particles. The overall structure of HBV subviral particles exhibits a clear symmetry, forming an icosahedral architecture. The further decomposition of this icosahedral structure involved testing D2 and D4 symmetries, ultimately revealing that the smallest stable repeating unit of subviral particles is HBsAg-dimer. However, due to the flexibility of HBsAg on the particle surface and technical limitations, important structural information regarding AGL and the intramolecular/intermolecular disulfide bonds remains elusive. In this study, with the help of SY-23-Fab binding, our structure offers unprecedented insights into the AGL of HBsAg and reveals how disulfide bonds and hydrophobic interactions contribute to the formation of dimeric HBsAg. Furthermore, from the view of antigen–antibody complex, our cryo-EM study of SY-23-Fab/HBsAg-dimer generated the reliable atomic model of mAb bound to intact HBsAg. The bioactivity analysis based on site-directed mutagenesis further confirmed the precise interaction sites between SY-23-Fab and HBsAg-dimer, revealing its recognition mechanism. Previously, the structure elucidation of HBsAg-specific mAbs relied on co-crystallisation with linear peptides, such as H015,¹⁸ which cannot present a full picture of antibody–antigen binding. When this crystal structure was aligned with our structure based on the AGL peptide (online supplemental figure S11C,D), we found that the H015 epitope lies approximately perpendicular to our identified binding region of SY-23. Furthermore, other regions of HBsAg would generate steric clashes with the H015, suggesting that additional structural elements may be involved in binding H015 in the context of the native HBsAg-dimer. Established in this study, the workflow of using mAb Fab-bound HBsAg-dimer provides an effective approach to determine the structures of HBsAg-specific mAbs recognising the entire conformational epitopes. Unlike SY-23, SY-4 binding to the HBsAg particles prior to the detergent solubilisation seemed to produce much lower amounts of dimeric HBsAg-Fab complex, but more tetrameric complex (online supplemental figure S9B). This may be due to the different binding surfaces of SY-4 and SY-23 that cause varied degrees of HBsAg dissociation. Our current purification method likely depends on the contribution of antibody binding to help isolate HBsAg dimer, and thus is not readily suitable for producing and comparing HBsAg complexes bound to different antibodies. In future work, we will further explore various methods for the HBsAg purification to facilitate structural studies of different HBsAg-Fab complexes.

Due to the limitation of available HBsAg resource and methodology, yeast-derived and CHO-derived HBsAg particles were both used in this study to evaluate various biochemical activities of HBsAg-specific mAbs. The former lacks glycosylation because yeast cells possess a different folding machinery.^{84 85} However, some mammalian cell-derived HBsAg particles contain a potential N-glycosylation site at Asn146,^86–88 which is located at the top of the dimeric HBsAg complex and distant from the binding surface of SY-23 (figure 4I and online supplemental table S2). The glycans on Asn146 are unlikely to interfere with SY-23 binding to CHO-derived HBsAg particles, which is also supported by ELISA and SPR results (figure 2 and online supplemental figure S3).

In conclusion, we performed a comprehensive analysis of 92 HBsAg-specific mAbs derived from proliferative plasmablasts, including gene features, binding capacities and antiviral functions in vitro and in vivo. Some potent protective human mAbs may serve as antibody drug candidates in the future development. We also reported the cryo-EM structure of a potent anti-HBsAg mAb in complex with the HBsAg-dimer, revealing the native conformation of AGL and detailed interaction information in the antigen–antibody interface. These findings will extend our understanding of recalling antibody responses and contribute to the mechanism dissection of mAb and the discovery of anti-HBV drug targets.

Limitations of the study

There are still three aspects of limitations in this study. More genotype HBsAg proteins and HBV viral stocks could be prepared for measuring the binding affinities and neutralising activities of these HBsAg-specific mAbs. Some representative mAbs could be further evaluated for their antiviral activities in more HBV-infected animal models, better supporting their potential clinical application. More cryo-EM studies of other anti-HBsAg mAbs bound to HBsAg could be performed in the future, providing more information on HBsAg dynamics and different interactions between antigen and antibody.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.