Introduction

Hepatitis B is a viral infection that affects the liver and can lead to both acute and chronic diseases1. The World Health Organization (WHO) estimates that in 2022, ~254 million people were living with chronic hepatitis B infection, with 1.2 million new cases occurring each year. In the same year, hepatitis B was responsible for an estimated 1.1 million deaths, primarily due to cirrhosis and hepatocellular carcinoma (primary liver cancer).

HBV is encased in a viral envelope that contains HBsAg as its only protein component2. There are three isoforms of HBsAg — large (L-HBsAg), medium (M-HBsAg), and small (S-HBsAg) surface antigens — resulting from alternative translational starting sites3. L-HBsAg is the longest ORF of HBsAg and contains PreS1, PreS2, and S-HBsAg, while the M-HBsAg contains PreS2 and S-HBsAg3. S-HBsAg is a transmembrane protein with a cytosolic loop (CYL) and an extracellular hydrophilic antigenic loop (AGL) or the “a”-determinant4, which contains the B cell epitope recognized by the host adaptive humoral immunity.

The process of HBV attachment begins with low-affinity interactions between the AGL of HBsAg on the virus envelope and heparan sulfate proteoglycans on the surface of hepatocytes5. Following this initial attachment, high-affinity interactions occur between the PreS1 region of L-HBsAg and its specific receptor, sodium taurocholate cotransporting polypeptide (NTCP), located on the plasma membrane of hepatocytes6. Notably, the Hepatitis D Virus also uses HBsAg as its envelope protein and shares the same entry pathway as HBV7. Studies have demonstrated that antibodies targeting either the AGL or the PreS1 region exhibit neutralizing activity8. For several decades, polyclonal hepatitis B immunoglobulins (HBIG) have been effective in post-exposure prophylaxis (PEP) against HBV infection, particularly in neonates born to HBsAg carrier mothers and in liver transplant patients with HBV infection9,10. Various neutralizing antibodies (NAbs) targeting AGL are currently being evaluated in clinical trials for their potential use in PEP or the treatment of chronic Hepatitis B8. The crucial role of AGL in HBV infection is evident from the fact that hepatitis B can be effectively prevented through vaccines containing the S-HBsAg protein, which exposes AGL as the major epitope8. Therefore, the AGL on the S-HBsAg is the primary target of broad NAbs11 and is also the site where escape mutations often occur11,12, such as the well-known G145R mutation13.

HBsAg dimer is the basic building block of the viral envelope. The core structure of HBsAg dimer only recently emerged from structures determined from the spherical subviral particles (SVP)14,15,16. Particularly, our studies showed that the CYL of each HBsAg has a zinc finger motif, and the HBsAg dimer interface involves interactions between H1, CYL, and H2 of two protomers16. However, the structure of AGL is missing in these studies due to its poor local resolution. Up to today, there is only one structure of a short peptide of AGL in complex with an NAb (H015) available11. Therefore, the structure of the essential AGL region and how it is recognized by NAbs remain largely enigmatic. Here, we combine biochemical assays and structural determination, uncovering the structural polymorphism of the AGL and the binding mechanism of HBV NAbs.

Results

Diverse NAbs recognize distinct forms of HBsAg

To study the interactions between HBsAg and NAbs, we gathered sequences of high-affinity HBV NAbs that bind to the AGL from the literature and identified NAbH00611, NAbH01511, NAbHBC (the parental antibody of VIR-3434)17, and NAbGC110218 for our investigations. To obtain the HBsAg protein for biochemical characterization, we appended a signal peptide to the N-terminus of a GFP-tagged M-HBsAg construct. This modified protein demonstrated a strong interaction with FabHBC, indicating that its AGL was correctly folded (Supplementary Fig. S1a). To reduce the aggregation of HBsAg, we introduced alanine mutations at three cysteines (C76A, C90A, C221A) in M-HBsAg. These cysteines are not conserved in other HBV-related viruses and do not affect the surface expression of HBsAg16. The resulting M-HBsAg-3CA construct also binds strongly to FabHBC (Supplementary Fig. S1a), confirming that HBsAg-3CA maintains a correctly folded AGL and is suitable for further biochemical and structural characterization.

To investigate whether these NAbs bind to the same epitope on HBsAg, we performed a competitive ELISA (Fig. 1a) and found that NAbHBC and NAbH015 belong to one group (Group A), while NAbH006 and NAbGC1102 are categorized into two other groups (Group B and Group C, respectively). NAbs within each group exhibit competitive binding behavior, whereas NAbs from different groups are non-competitive. This suggests that NAbs in different groups may recognize distinct epitopes on the same type of HBsAg or markedly different types of HBsAg.

Fig. 1: Classification of NAbs targeting AGL of HBsAg.
figure 1

a Heat map illustrating the competitive binding behaviors of four neutralizing antibodies (NAbs) for HBsAg: NAbHBC, NAbH006, NAbH015, and NAbGC1102. The binding of HA-tagged antibodies in the presence of non-tagged competitors, as determined by ELISA, is normalized to their binding without competitors. Numbers are represented with background colors, where darker shades indicate higher values and less competition. Data are presented as mean, with n = 3 technical replicates. The experiment was conducted twice with consistent results. b Strep-tag pull-down experiments were conducted to assess the binding compatibilities of NAbs across different groups. HA-tagged Fabs containing Strep tags were bound to Streptactin resin as bait. GFP-tagged M-HBsAg was added to mediate the binding between compatible NAbs. Myc-tagged Fabs or scFv served as prey. Proteins were detected using western blotting (for HA or Myc tags) or in-gel fluorescence imaging for GFP (M-HBsAg). c Binding of NAbs to HBsAg and its G145R mutant was assessed using ELISA. GFP-PreS1 was used as the negative control to assess the non-specific binding. Data are shown as mean ± standard deviations, n = 3 technical replicates. The experiment was performed twice with consistent results.

Full size image

To further distinguish these possibilities, we conducted ternary pull-down experiments using FabHBC, FabH006, and FabGC1102 as representatives for Groups A, B, and C, respectively. We employed FabHBC with strep and HA tag to pull down Myc-tagged scFvH006 or FabGC1102 in the presence of GFP-M-HBsAg. The results demonstrated that NAbHBC and NAbGC1102 can bind to HBsAg simultaneously, while NAbHBC and NAbH006 cannot. Additional pull-down experiments with either strep-tagged FabH006 or FabGC1102 confirmed that NAbGC1102 and NAbH006 can also bind to HBsAg simultaneously, while NAbHBC and NAbH006 cannot bind to HBsAg at the same time (Fig. 1b). These compelling biochemical results support a model in which FabHBC and FabH006 bind to distinct types of HBsAg (Type A and Type B, respectively), while FabGC1102 binds to both Type A and Type B. These unexpected findings suggest that HBsAg of Types A and B may exhibit different structures.

Additionally, we found that NAbs in Group A (FabHBC and FabH015) and Group B (FabH006) almost lost their binding to G145R escape mutant (Fig. 1c). In contrast, the NAb in Group C (FabGC1102) retains robust binding to HBsAg despite the G145R mutation (Fig. 1c). This agrees with previous results on the neutralizing effects of these NAbs against escape mutations11,17,18 and suggests that NAbs in Group C bind to a unique epitope on HBsAg.

AGLType A and AGLType B show distinct structures with different disulfide linkages

To understand the structures of two types of HBsAg and how they are recognized by different groups of antibodies, we determined the structure of HBsAgType A in complex with FabHBC and HBsAgType B in complex with Fab H006 at resolutions of 3.25 Å and 3.13 Å respectively (Fig. 2; Supplementary Figs. S1b–f, S2–S5, and Table S1). In both cases, we found that the Fabs bind to HBsAg dimer in 1:1 stoichiometry (Fig. 2a–d).

Fig. 2: The structures of HBsAgType A and HBsAgType B.
figure 2

a Side view of the cryo-EM map of the HBsAg–FabHBC complex. HBsAg-A, HBsAg-B, FabHBC-heavy chain variable region (VH), and FabHBC-light chain variable region (VL) are colored in green, purpleblue, pink, and orange, respectively. The sizes of the HBsAg and AGL are labeled. The detergent micelle and unresolved flexible region are shown in semi-transparent gray. The approximate boundaries of the phospholipid bilayer are indicated as thick gray lines. b A 90° rotated top view of a. c Side view of the cryo-EM map of the HBsAg–FabH006 complex. HBsAg-A subunit and HBsAg-B subunits are colored the same as in (a), while FabH006-VH and FabH006-VL are colored in steel blue and light blue, respectively. The approximate boundaries of the outer leaflet of the viral envelope are indicated as thick gray lines. d A 90°-rotated top view of c. e Top view of the AGLType A in cartoon representation. AGL-A and AGL-B are in the same colors as in (a). β-strands and loops are indicated. f 2D topology diagram of AGLType A. β-strands and loops are indicated. Flexible regions that are not resolved are denoted with dashed lines. g Top view of the AGLType B in cartoon representation. AGL-A and AGL-B are in the same colors as in (c). h 2D topology diagram of AGLType B. i Side view of the AGLType A in cartoon representation. Cysteine pairs forming disulfide bonds are indicated, and the disulfide bonds are highlighted in yellow. j Side view of the AGLType B in cartoon representation. k Schematic diagram of AGLType A. Cysteines are indicated by yellow circles with their residue numbers. The intramolecular disulfide bonds are labeled in blue, and the intermolecular disulfide bonds in orange. Flexible regions that are not resolved are denoted with dashed lines. l Schematic diagram of AGLType B.

Full size image

In the map of HBsAgType A in complex with FabHBC, we could resolve the density of a portion of the H1, H2, AGL, and H3 regions of HBsAg (Fig. 2a, b; Supplementary Fig. S2), but not other regions, suggesting that they have considerable flexibility in detergent micelles. The partially disordered transmembrane region is likely due to the loss of contacts between HBsAg dimers during detergent solubilization, as observed in previous studies of the detergent-solubilized spherical SVP14. Further masking out the transmembrane helices and detergent micelle improved the resolution of the AGL-Fab region to 3.20 Å (Supplementary Fig. S2 and Table S1). The structure of HBsAgType A shows that the anti-parallel β1 and β2 strands of two HBsAg subunits pack together to form a central four-strand β-sheet core of AGL (Fig. 2e, f). The long H2-β1 loop and β1-β2 loop are exposed to the solvent and contribute to the majority of the surface properties of the AGL region (Fig. 2e). The AGL region of each HBsAg subunit contains eight cysteines, and thus the AGL of HBsAg dimer contains 16 cysteines in total (Fig. 2i, k). To our surprise, although the 16 cysteines in the AGLType A form six intramolecular disulfide bonds and two intermolecular disulfide bonds (Fig. 2i, k), the cysteine pairs that form these disulfide bonds are completely different between the two HBsAg subunits in HBsAgType A (Fig. 2i, k). For example, C107 pairs with C137 in subunit A, while C107 pairs with C138 in subunit B (Fig. 2i, k; Supplementary Fig. S3a). To our knowledge, such a drastically asymmetric disulfide linkage has not been reported in any homodimeric protein previously. The asymmetric disulfide bonds confer structural asymmetry to the AGLType A. Close inspection reveals that the symmetry breakage occurs in the AGL, starting around C107 where the first disulfide bond is formed, and ending around C149 where the last disulfide bond is formed (Supplementary Fig. S3d, e).

In the map of HBsAgType B in complex with FabH006 (Fig. 2c, d; Supplementary Figs. S4, S5 and Table S1), one AGL protomer bound to FabH006 showed better density than the other, although some residues on AGL remained unresolved due to their flexibility (Fig. 2c, d, g, h). Compared to AGLType A, AGLType B adopts a distinct conformation, which also features a structural core formed by four anti-parallel β-strands at its center, surrounded by loop-like structures composed of connecting residues (Fig. 2g, h). We found that the 16 cysteines in AGL form five intramolecular disulfide bonds and two intermolecular disulfide bonds, with the side chains of C124 and C137 remaining flexible (Fig. 2j, l). Notably, during model building, we observed connecting densities between two adjacent disulfide bonds (between the C139–C149 bonds of each protomer), suggesting a degree of heterogeneity in disulfide linkages (Supplementary Fig. S5). The disulfide linkages in AGLType B are dramatically different from those in AGLType A (Fig. 2k, l).

Mechanism of AGLType A and AGLType B recognition by NAbs

The structure of HBsAgType A in complex with FabHBC shows that FabHBC binds to one side of AGL through an extended interface (Fig. 3a). Both the heavy chain and the light chain of FabHBC contribute to the binding to a structural epitope on the surface of AGL, which is shaped by the H2-β1 loop and the β1-β2 loop on subunit A and the H2-β1 loop on subunit B (Fig. 3a). Particularly, the H2-β1 loop of subunit B is recognized by CDRH1 and CDRH3 of FabHBC and the H2-β1 loop of subunit A is bound by CDRL1 of FabHBC (Fig. 3b, c). The β1-β2 loop on subunit A binds to CDRH2 and CDRL3 of FabHBC (Fig. 3b, c). Detailed polar interactions between HBsAg and FabHBC are listed in Supplementary Table S2.

Fig. 3: Interactions between AGL and different Fabs.
figure 3

a Side view of AGLType A and FabHBC-Fv domains in a semi-transparent surface representation, along with the atomic model in the same color as in Fig. 2a. The viral envelope membrane is indicated with a thick gray line. Interfaces are highlighted with dashed boxes. b Close-up view of the interface between FabHBC-VH and HBsAg boxed in (a). Key residues for the binding between FabHBC-VH and HBsAg are shown as sticks. c Close-up view of the interface between FabHBC-VL and HBsAg boxed in (a). Key residues for the binding between FabHBC-VL and HBsAg are shown as sticks. d Side view of AGLType B and FabH006-Fv domains in a semi-transparent surface representation, along with the atomic model in the same color as in Fig. 2c. The viral envelope membrane is indicated with a thick gray line. Interfaces are highlighted with dashed boxes. e–g Close-up views of the interfaces between FabH006-VH and HBsAg boxed in (d). Key residues for the binding between FabH006-VH and HBsAg are shown as sticks. h Close-up views of the interfaces between FabH006-VL and HBsAg boxed in (d). Key residues for the binding between FabH006-VL and HBsAg are shown as sticks.

Full size image

The structure of HBsAgType B in complex with FabH006 shows that FabH006 binds to the top of AGLType B on one side (Fig. 3d), primarily interacting with one subunit of HBsAg. Both the heavy and light chains of FabH006 contribute to AGL recognition (Fig. 3e, h). The H2-β1 loop on AGL is bound by CDRL1, CDRH2, and CDRH3 of FabH006 (Fig. 3e, h). The β1-β2 loop binds to CDRH1 and CDRH3 of FabH006 (Fig. 3e, h). Detailed polar interactions between HBsAg and FabH006 are listed in Supplementary Table S3.

Structural alignment of AGLType B with AGLType A using the central four β-strands revealed that the structural epitopes recognized by NAbH006 and NAbHBC are strikingly distinct (Supplementary Fig. S6). Notably, due to the different disulfide bond linkages, the structural epitope of NAbHBC on AGLType A does not exist in AGLType B, and vice versa (Supplementary Fig. S6), which agrees with the high cross-group specificity of NAbs in Groups A and B (Fig. 1).

NAbGC1102 in Group C binds to a unique epitope shared by HBsAgType A and HBsAgType B

NAbGC1102, a Group C NAb, can bind to both HBsAgType A and HBsAgType B, indicating that it recognizes a common epitope shared by both types of HBsAg proteins. To elucidate the underlying mechanism, we determined the structure of HBsAg in complex with both FabHBC and FabGC1102 to the resolution of 3.09 Å (Fig. 4; Supplementary Figs. S1g–h, S7, and Table S1). The structure reveals that one FabGC1102 binds to the opposite side of FabHBC, engaging the lower region of one HBsAg Type A subunit (Fig. 4a–d). The epitope recognized by FabGC1102 includes the H2-β1 loop, β2-H3 loop, and H3 of HBsAg subunit A (Fig. 4e, f). β2-H3 loop and H3 are bound by CDRH3 and CDRH1, while H2-β1 loop interacts with CDRH1, CDRH3, and CDRL2 (Fig. 4e, f). Detailed polar interactions between HBsAg and FabGC1102 are listed in Supplementary Table S4.

Fig. 4: The structure of AGLType A in complex with FabGC1102 and FabHBC.
figure 4

a Side view of the cryo-EM map of the HBsAg–FabGC1102–FabHBC complex. FabGC1102-VH and FabGC1102-VL are colored in purple and light purple, respectively, and others are colored the same in Fig. 2a. The approximate boundaries of the outer leaflet of the viral envelope are indicated as dashed lines. b A 180°-rotated side view of a. c Atomic model of the HBsAg-FabGC1102-FabHBC complex is shown as cartoons in the same colors as in (b). Locations of interactions between FabGC1102 and HBsAg are indicated with dashed boxes. d A 90°-rotated top view of a. e Close-up view of the interface between FabGC1102-VH and HBsAg boxed in (c). Key residues for the binding between FabGC1102-VH and HBsAg are shown as sticks. f Close-up view of the interface between FabGC1102 and HBsAg H2-β1 loop boxed in (c). Key residues for the binding between FabGC1102 and HBsAg H2-β1 loop are shown as sticks.

Full size image

Further structural modeling suggests that the epitope recognized by FabGC1102 is also present in HBsAgType B (Supplementary Fig. S7h, i). Additionally, both FabGC1102 and FabH006 can bind to HBsAgType B simultaneously without sterical clashes, consistent with our biochemical results (Supplementary Fig. S7h, i).

Discussion

The structures of the AGL region of HBsAg reveal a remarkable and unexpected structural polymorphism. Our studies identified three distinct groups of NAbs, closely aligning with findings from a comprehensive study involving 144 vaccinated or HBV-infected individuals11. By mapping our results to theirs using the common NAbs H015 and H006, we found that Groups A and B are the largest NAb groups identified in humans11. This analysis suggests that Type A and B HBsAg may represent the predominant forms of HBsAg on the viral surface, while the existence of other forms remains elusive. Additionally, the cysteines forming the disulfide bonds are highly conserved in the surface proteins of HBV-related viruses (Supplementary Fig. S8). This conservation leads us to speculate that these viral surface proteins may exhibit similar structural polymorphisms to those observed in HBsAg. Our structures also elucidate how Types A and B HBsAg are recognized by their respective NAbs. Notably, numerous naturally occurring HBV escape mutations have been documented, including the well-known G145R mutation13. Many of these mutations occur within the epitopes of NAbHBC and NAbH00613 (Supplementary Figs. S9, S10). However, only one mild mutation, T126S, is found within the epitope of NAbGC1102 (Supplementary Figs. S9, S10), likely because that NAbGC1102 binds to the extracellular lower region of HBsAg. This unique binding mode correlates with its broad neutralizing activity against several common escape mutations18, whereas NAbHBC and NAbH006 are more sensitive to specific escape mutations like G145R11,17. Furthermore, these structures provide insights into how the human humoral response generates diverse antibodies that recognize distinct forms of HBsAg, as well as different epitopes on the same form (Supplementary Fig. S11). These mechanistic insights lay the foundation for developing combination therapies that utilize multiple NAbs targeting different epitopes on distinct types of HBsAg for the treatment of HBV infection.

Materials and methods

Cell culture

Sf9 insect cells (Thermo Fisher Scientific) were cultured in SIM SF (Sino Biological) at 27 °C. FreeStyle 293 F suspension cells (Thermo Fisher Scientific) were cultured in FreeStyle 293 medium (Thermo Fisher Scientific) supplemented with 1% fetal bovine serum (FBS, VisTech), 67 μg mL–1 penicillin (Macklin), and 139 μg mL–1 streptomycin (Macklin) at 37 °C with 6% CO2 and 70% humidity. The cell lines were routinely checked to be negative for mycoplasma contamination but have not been authenticated.

Constructs

Superfolder-GFP-tagged M-HBsAg was generated by fusing the coding sequence of the GFP superfolder before the preS2 region of M-HBsAg (serotype ayw, genotype D3), with a PreScission Protease cleavage sequence between them. A rat FSHβ signal peptide sequence was inserted at the N-terminus of the GFP. The ORF was cloned into a modified BacMam vector19. Point mutations on M-HBsAg were introduced using Quikchange PCR.

The coding sequences of the variable heavy (VH) and light chains (VL) of Fabs obtained from literature and patents were synthesized and cloned into a pBMCL1-based plasmid20 and fused with the human CH1 region and human kappa CL region, respectively. The 3× HA-tagged Fab was generated by introducing a 3× HA tag at the C-terminus of the light chain. The coding sequence of the anti-human kappa nanobody21 was synthesized and cloned into pET26b, with a tandem 6× His-twin Strep-FLAG tag at the C-terminus.

For FabH006, V11 and P118 on the VH, P158 on CH1, P39 and A79 on VL, and E166 and S172 on the kappa CL region were mutated to cysteine. In addition, a cysteine was inserted between E159 and P160 on the CH1 region. These eight cysteines were expected to generate four disulfide bonds (4DS) between the variable and constant regions of the Fab fragment to stabilize its conformation22.

For FabGC1102, L11 and T124 on the VH, P163 on CH1, P40 and P80 on VL, and E165 and S171 on the kappa CL region were mutated to cysteine. In addition, a cysteine was inserted between E164 and P165 on the kappa CL region. These eight cysteines were expected to generate four disulfide bonds between the variable and constant regions of the Fab fragment to stabilize its conformation.

Expression and purification of FabHBC, FabGC1102, and FabH006

FabHBC, FabGC1102, and FabH006 were purified similarly. Briefly, FreeStyle 293F cells at a density of around 2.5 × 106 cells mL–1 were co-transfected with constructs containing the heavy chain and the light chain of Fab at a ratio of 1:1 using polyethyleneimine (PEI-MAX, Polysciences). 10 mM sodium butyrate was added 12 h after transfection and the cells were incubated at 37 °C for 108 h. Cell pellets were removed by centrifugation at 3993× g for 20 min at 4 °C (Beckman JA8.1), and the supernatant was subjected to Ni-NTA affinity chromatography. The column was washed tandemly with W buffer 1 (50 mM Tris (pH 7.5), 500 mM NaCl, and 10 mM imidazole), W buffer 2 (20 mM Tris (pH 7.5), 150 mM NaCl, and 20 mM imidazole), and W buffer 3 (20 mM Tris (pH 7.5), 50 mM NaCl and 30 mM imidazole). Fab was eluted with elution buffer containing 300 mM imidazole (pH 8.0) and 25 mM NaCl. The pH of the eluate was adjusted to 6.0 with 10% glacial acetic acid and SP-A buffer (20 mM MES, pH 6.0) was added into the eluate until the conductance was under 5000 μS cm–1. The Fab was loaded onto an SP HP 5 mL column (Cytiva) and eluted with SP-B (20 mM MES, pH 6.0, 1 M NaCl) in a linear gradient using the AKTA pure system (GE Healthcare). Fractions containing Fabs were collected and supplemented with 10% glycerol, flash frozen, and stored at –80 °C.

ELISA

To estimate the binding affinity between HBsAg and Fabs, ELISA was performed using purified M-HBsAg and Fabs. Briefly, 12-well strips (Yunpeng) were coated with anti-GFP nanobody23, then blocked by TBSG-FBS (20 mM Tris (pH 7.5), 150 mM NaCl, 0.0058% GDN, and 2% FBS) for 30 min. 50 nM GFP-tagged M-HBsAg (WT or 3CA) in TBSG (20 mM Tris (pH 7.5), 150 mM NaCl, 0.0058% GDN) was added (100 μL per well) and incubated on ice for 1 h to bind anti-GFP nanobody. An unrelated protein, GFP-tagged anti-ALFA nanobody24 was used as the negative control. Unbounded antigens were washed away by TBSG before the addition of different concentrations of 3× HA-tagged FabHBC (50 μL per well in TBSG-FBS) for 1 h on ice. After washing with TBSG, rabbit anti-HA (3724; CST, diluted 4000 times in TBSG-FBS, 50 μL per well) was added and incubated on ice for 1 h to bind 3× HA-tagged Fab. After washing with TBSG, goat anti-rabbit antibody conjugated with horseradish-peroxidase (HRP) (31460; Thermo Fisher Scientific, diluted 4000 times in TBSG-FBS, 50 μL per well) was added and incubated on ice for 1 h. After extensive washing with TBSG, 100 μL of ELISA developing buffer (51.5 mM Na2HPO4, 24.3 mM citric acid, 0.006% H2O2, and 0.1 mg mL−1 3,3′,5,5′-tetramethylbenzidine (TMB)) was added and incubated at room temperature for 10 min before stopping with 100 μL 2 M H2SO4. ELISA signals were measured by absorbance at 450 nm with an Infinite M Plex plate reader (Tecan).

To assess the binding capability of Fabs to HBsAg G145R mutant, 50 nM GFP-tagged M-HBsAg (3CA or 3CAG145R) and GFP-PreS1 proteins were added to anti-GFP nanobody-coated strips. 3× HA-tagged Fabs (50 μL per well in TBSG-FBS) were added and incubated on ice for 1 h. After washing with TBSG, rabbit anti-HA (3724; CST, diluted 4000 times in TBSG-FBS, 50 μL per well) was added and incubated on ice for 1 h to bind 3× HA-tagged Fab. The following steps were the same as described above.

For competitive ELISA, GFP-M-HBsAg-3CA was immobilized on anti-GFP nanobody-coated strips as described above. The antigen was incubated with 200 nM of different Fabs (50 μL per well, for GC1102, the concentration was 400 nM) for 2 h at 4 °C to block their structural epitope on HBsAg. After washing with TBSG, HA-tagged Fabs (50 μL per well) were added and incubated on ice for 30 min. Empty buffer was used as the negative control. After washing with TBSG, rabbit anti-HA (3724; CST, diluted 4000 times in TBSG-FBS, 50 μL per well) was added and incubated on ice for 1 h to bind HA-tagged Fabs. The following steps were the same as described above.

Expression and purification of anti-Fab nanobody

E. coli. NiCo21 (DE3) was transformed with the plasmid of the anti-Fab nanobody. Overexpression was induced by 200 μM isopropyl-β-d-thiogalactoside (IPTG) when the cell density reached OD600 = 0.6 for 16 h at 16 °C in LB medium. The nanobody was extracted by osmotic shock from the periplasmic space. In brief, the cell pellet was resuspended with hyperosmotic buffer (50 mM Tris (pH 7.5), 20% sucrose, and 0.5 mM EDTA (pH 8.0)), and stirred at 4 °C for 30 min, then centrifugated at 10,000× g for 20 min at 4 °C (Beckman, JA14). The cell pellet was resuspended with hypoosmotic buffer (20 mM Tris (pH 7.5) and 5 mM MgCl2), and stirred at 4 °C for 30 min, then centrifuged at 25,954× g for 30 min at 4 °C (Beckman JA14). The supernatant was subjected to a Ni-NTA column for affinity chromatography and washed with Washing Buffer 1, 2, and 3 in sequence, then eluted with 300 mM imidazole (pH 6.0) and 25 mM NaCl. The nanobody was diluted, loaded onto SP HP 5 mL column (Cytiva), and eluted with SP-B (20 mM MES (pH 6.0), 1 M NaCl) in a linear gradient using the AKTA pure system (GE Healthcare). The peak fractions were pooled for complex assembly.

Expression of GFP-M-HBsAg-3CA and purification of HBsAg–Fab complex

The BacMam virus for GFP-M-HBsAg-3CA was generated using Sf9 cells25. Freestyle 293 F cells at a density of around 3 × 106 cells mL–1 were infected with a 2.5% volume of P2 virus. 10 mM sodium butyrate was added 12 h after infection and the cells were incubated at 37 °C for 48 h before harvest. The cell pellet was washed with TBS (20 mM Tris pH 7.5, 150 mM NaCl), then lysed by sonification in lysis buffer (TBS supplemented with 1 μg mL–1 aprotinin, 1 μg mL–1 pepstatin, 1 μg mL–1 leupeptin, 1 mM phenyl methane sulfonyl fluoride). Cell debris and unbroken cells were removed by centrifugation at 7500× g for 20 min. Membrane fractions were pelleted by ultracentrifugation at 185,679× g for 1 h at 4 °C (Beckman Ti45) and homogenized in lysis buffer, flash frozen, and stored at –80 °C.

To purify the HBsAg-Fab complex, the cell membrane corresponding to around 1.4 L of cell culture was solubilized with 1% LMNG and 0.1% CHS in lysis buffer for 90 min at 4 °C. The insoluble materials were removed with ultracentrifugation at 147,775× g for 40 min at 4 °C (Beckman 50.2 Ti), and the supernatant was supplied with Fab and anti-Fab nanobody to form the complex. The lysate was loaded onto 4 mL Streptactin Beads 4FF (Smart-Lifesciences) column and washed with 30 mL buffer W (20 mM Tris (pH 7.5), 150 mM NaCl, and 0.0058% GDN), followed by 30 mL buffer W supplemented with 10 mM ATP and 2 mM MgCl2, and 30 mL buffer W again. The target protein was eluted with 5 mM desthiobiotin, 20 mM Tris (pH 8.0), 40 mM Tris (pH 7.5), 50 mM NaCl, and 0.0058% GDN. The eluate was diluted, loaded onto Q HP 1 mL column (Cytiva), and eluted with a linear gradient of Q-B (20 mM Tris (pH 7.5), 1 M NaCl) on AKTA pure system (GE Healthcare). Eluted protein was concentrated using a 100-kDa cutoff concentrator (Millipore) and further purified by Superose 6 increase (Cytiva) running in a buffer containing 20 mM HEPES (pH 7.41), 50 mM NaCl, and 0.0058% GDN. The peak fractions were analyzed by SDS-PAGE, and fractions corresponding to the complex were pooled, concentrated, and used in cryo-EM sample preparation.

For the HBsAg–FabHBC–FabGC1102 complex, the purification was the same as that for HBsAg–FabHBC complex, except that FabGC1102 was added before size-exclusion chromatography. The peak fractions were analyzed by SDS-PAGE, and fractions corresponding to the complex were pooled, concentrated, and used in cryo-EM sample preparation.

Strep-tag pull-down assay

The membrane of cells expressing GFP-M-HBsAg was used in the Strep pull-down assay. Cell membrane containing GFP-M-HBsAg was solubilized and ultracentrifuged as described in HBsAg–Fab complex purification. The supernatant was added with different strep-HA-tagged NAbs (used as bait) and Myc-tagged NAbs (used as prey), then incubated with streptactin resin (SA053100; Smart-Lifesciences) at 4 °C on a rotator for 2 h. The beads were washed with TBSG for four times. Bound proteins were eluted with 50 mM Tris (pH 8.0), 150 mM NaCl, 5 mM desthiobiotin, and 0.0058% GDN.

For western blot, input and pull-down fractions were separated with SDS-PAGE and transferred onto polyvinylidene difluoride membranes. Membranes were blocked using 5% nonfat milk in TBST (20 mM Tris (pH 7.4), 150 mM NaCl, and 0.1% Tween-20) for 1 h at room temperature and incubated with primary antibodies (rabbit anti-HA (3724; CST), rabbit anti-Myc (YN5506; Immunoway), both of antibodies were diluted 5000 times) overnight at 4 °C. Then membranes were incubated with HRP-labeled secondary antibody (31444; Thermo Fisher Scientific, the antibody was diluted 10,000 times) for 1 h at room temperature and developed using High-Sig ECL Western Blotting Substrate (Tanon).

Cryo-EM sample preparation and data collection

For HBsAg–Fab complexes, purified complexes were concentrated to A280 = 6.7 (for HBsAg–FabHBC complex), A280 = 7.38 (for HBsAg–FabH006 complex), and A280 = 5.2 (for HBsAg–FabHBC–FabGC1102 complex), respectively. Holey carbon grids (Quantifoil Au 300 mesh, R 0.6/1) were glow-discharged by Solaris advanced plasma system (Gatan) for 120 s using 25% O2 and 75% Ar. Aliquots of 3 μL concentrated protein sample were applied on glow-discharged grids and the grids were blotted for 3 s before being plunged into liquid ethane using Vitrobot Mark IV (Thermo Fisher Scientific). Cryo-grids were first screened on a Talos Arctica electron microscope (Thermo Fisher Scientific) operating at 200 kV with a K2 camera (Gatan). The screened grids were subsequently transferred to a Titan Krios electron microscope (Thermo Fisher Scientific) operating at 300 kV with a K3 camera (Gatan) and a GIF Quantum energy filter (Gatan) set to a slit width of 20 eV. For FabHBC, images were automatically collected using EPU (Thermo Fisher Scientific) in super-resolution mode at a nominal magnification of ×105,000, corresponding to a calibrated super-resolution pixel size of 0.417 Å with a preset defocus range from –1.5 to –1.8 μm. Each image was acquired as a 2.57-s movie stack of 32 frames with a dose rate of 20.13 e– Å–2 s–1, resulting in a total dose of about 52 e– Å–2. For FabH006 and FabHBC-FabGC1102 complex, images were automatically collected using EPU (Thermo Fisher Scientific) in super-resolution mode at a nominal magnification of ×81,000, corresponding to a calibrated super-resolution pixel size of 0.5335 Å with a preset defocus range from –1.5 to –1.8 μm. Each image was acquired as a 2.56-s movie stack of 32 frames with a dose rate of 22.36 e– Å–2 s–1, resulting in a total dose of about 50 e– Å–2.

Cryo-EM image analysis

The image processing workflows are illustrated in Supplementary Figures. Super-resolution movie stacks were collected. Motion-correction, two-fold binning, and dose weighting were performed using MotionCor226. Contrast transfer function (CTF) parameters were estimated with cryoSPARC27. Micrographs with ice or ethane contamination and empty carbon were removed manually. Particles were auto-picked using Gautomatch (provided by K. Zhang). All subsequent classification and reconstruction were performed in cryoSPARC27 unless otherwise stated. Reference-free 2D classification was performed to remove contaminants. The resulting particles were subjected to 3D classification using the initial models generated from cryoSPARC.

For datasets of HBsAg–Fab complexes, particles were first cleaned using 2D and 3D classifications. The classes that showed secondary structure features were selected and the resulting particles were subjected to non-uniform refinement28, which resulted in maps with a resolution of 7.63 Å for the HBsAg–FabHBC complex, 5.51 Å for the HBsAg–FabH006 complex, and 5.29 Å for the HBsAg–FabHBC–FabGC1102 complex. To further improve the resolution, seed-facilitated 3D classification29 was performed. The CTF parameters were re-estimated with Patch CTF and particles were re-picked with Template Picker. Several rounds of seed-facilitated 3D classification using good and biased references or references with resolution gradients were performed. The particles were re-picked with Topaz30. Several rounds of seed-facilitated 3D classification were performed. To further clean up the dataset, 3D classification without alignment was performed in RELION 3.131 with resolution gradient references with the mask of Fv+AGL. For the HBsAg–FabHBC complex, the resulting particles were subjected to local non-uniform refinement in cryoSPARC with the mask of Fv+AGL + TM and Fv+AGL, yielding maps with resolutions of 3.25 Å and 3.20 Å, respectively. For the HBsAg–FabH006 complex, the resulting particles were subjected to local non-uniform refinement with the mask of Fv+AGL, yielding a map with a resolution of 3.13 Å. For the HBsAg–FabHBC–FabGC1102 complex, the resulting particles were subjected to local non-uniform refinement with the mask of Fv+AGL, yielding a map with a resolution of 3.09 Å. All of the resolution estimations were based on a Fourier shell correlation of 0.143 cut-off after correcting the masking effect. The local resolution was estimated with cryoSPARC.

Model building

For HBsAg–Fab complexes, the initial models of FabHBC, FabH006, and FabGC1102 were generated using SWISS-MODEL32. The initial models of the Fv region of Fabs were fitted into the maps using UCSF Chimera33 and rebuilt manually in Coot34. The model of HBsAg was manually built according to the maps. Model refinement was performed using phenix.real_space_refine in PHENIX35. The validation statistics are provided in Supplementary Table S1.

Sequence alignments

Germline sequences of antibodies were determined using IgBlast36 and the somatic hypermutation sites on FR1–FR3 of each NAb were determined accordingly. Sequence alignments were performed using ClustalW and illustrated by BioEdit.

Quantification and statistical analysis

Global resolution estimations of cryo-EM density maps are based on the 0.143 Fourier Shell Correlation criterion37. The local resolution was estimated using cryoSPARC. The number of independent experiments (N) and the relevant statistical parameters for each experiment (such as mean or standard deviation) are described in the figure legends. No statistical methods were used to pre-determine sample sizes.