A Combination of Human Broadly Neutralizing Antibodies against Hepatitis B Virus HBsAg with Distinct Epitopes Suppresses Escape Mutations

Serologic Responses Against HBV

To select individuals with outstanding antibody responses to HBsAg, we performed ELISA assays on serum obtained from 159 volunteers. These included 15 uninfected and unvaccinated controls (HBsAg⁻, anti-HBs⁻, anti-HBc⁻), 20 infected and spontaneously recovered (HBsAg⁻, anti-HBs^+/−, anti-HBc⁺), and 124 vaccinated (HBsAg⁻, anti-HBs^+/−, anti-HBc⁻) volunteers. These individuals displayed a broad spectrum of anti-HBs titers (x-axis in Figure 1A and Figure S1B; Table S1). To determine their neutralizing activity, we tested their ability to block HBV infection in sodium taurocholate co-transporting polypeptide (NTCP)-overexpressing HepG2 cells (Michailidis et al., 2017; Yan et al., 2012) (y-axis in Figure 1A and Figure S1B and S1C; Table S1). Sera or antibodies purified from individuals with high levels of neutralizing activity were then compared across a wide range of dilutions (Figure 1B and 1C). Although anti-HBs ELISA titers positively correlated with neutralizing activity (r_s=0.492, p<0.001, Spearman’s rank correlation), there were notable exceptions as exemplified by volunteers #99 and #49, whose sera failed to neutralize HBV despite high anti-HBs ELISA titers (Figure 1A). Thus, ELISA titers against HBsAg are not entirely predictive of neutralizing activity in vitro.

The HBV surface protein, HBsAg can be subdivided into PreS1-, PreS2- and S-regions (Figure 1D). To determine which of these regions is the dominant target of the neutralizing response in the selected top neutralizers, we used S-protein to block neutralizing activity in vitro. As expected, the neutralizing activity in volunteers that received the HBV vaccine, which is composed of S-protein, was completely blocked by S-protein (black lines in Figure 1E). The same was true for the spontaneously recovered individuals in our cohort despite a reported ability of this population to produce anti-PreS1 or anti-PreS2 antibodies (Coursaget et al., 1988; Li et al., 2017; Sankhyan et al., 2016) (red lines in Figure 1E). These results suggest that the neutralizing antibody response in the selected individuals is directed primarily against the S-protein irrespective of immunization or infection.

Human Monoclonal Antibodies to HBV

To characterize the antibodies responsible for neutralizing activity in the selected individuals, we purified S-protein binding class-switched memory B cells (Escolano et al., 2019; Scheid et al., 2009a). Unexposed naïve controls and vaccinated individuals with low anti-HBs ELISA titers showed background levels of S-protein specific memory B cells (Figure 2A and S2A). In contrast, individuals with high neutralizing activity displayed a distinct population of S-antigen binding B cells constituting 0.03–0.07% of the IgG⁺ memory compartment (CD19-MicroBeads⁺ CD20-PECy7⁺ IgG-Bv421⁺ S-protein-PE⁺ S-protein-APC⁺ ovalbumin-Alexa Fluor 488⁻) (Figure 2A and S2A). Consistent with the findings in elite HIV-1 neutralizers (Rouers et al., 2017), the fraction of S-protein specific cells was directly correlated to the neutralization titer of the individual (r_s=0.699, p=0.0145, Spearman’s rank correlation) (Figure 2B).

Immunoglobulin heavy (IGH) and light (IGL or IGK) chain genes were amplified from single memory B cells by PCR (Robbiani et al., 2017; Scheid et al., 2009b; von Boehmer et al., 2016). Overall, we obtained 244 paired heavy and light chain variable regions from S-protein-binding memory B cells from eight volunteers with high anti-HBs ELISA titers (Figure S2B and S2C; Table S2). Expanded clones composed of cells producing antibodies encoded by the same Ig variable gene segments with closely related CDR3s were found in each of the top neutralizers #146, #60 and #13 (Figure 2C). Moreover, IGHV3–30/IGLV3–21 was present in #146 and #60; IGHV3–33/IGLV3–21 in #146 and #13; and IGHV3–23/IGLV3–21 in #146, #60 and #13. The variable diversity and joining (V(D)J) region of these antibodies was approximately 80% identical at the amino acid level (Figure 2D). Antibodies with related Ig heavy and light chains were also identified between volunteer #55 (HBV infected but recovered) and vaccinated individuals (Figure 2C and S2B). We conclude that top HBV neutralizers produce clones of antigen-binding B cells that express related Ig heavy and light chains.

Breadth of Reactivity

Twenty representative antibodies from 5 individuals, designated as H001 to H020, were selected for expression and further testing (Figure S2B). All 20 antibodies showed reactivity to the S-protein antigen used for B cell selection (HBsAg adr CHO) by ELISA with 50% effective concentration (EC₅₀) values ranging from 18–350 ng/ml (Figure 3A). These antibodies carried somatic mutations that enhanced antigen binding as determined by reversion to the inferred unmutated common ancestor (UCA) (Figure 3B). Thus, affinity maturation was essential for their high binding activity.

Four major serotypes of HBV exist as defined by a constant “a” determinant and two variable and mutually exclusive determinants “d/y” and “w/r” (Bancroft et al., 1972; Le Bouvier, 1971) with a highly statistically significant association between serotypes and genotypes (Kramvis et al., 2008; Norder et al., 2004). To determine whether our antibodies cross-react to different HBsAg serotypes, we performed ELISAs with 5 additional HBsAg antigens: yeast-expressed serotype “adr”, “adw”, and “ayw”, as well as “ad” and “ay” antigen purified from human blood (Figure 3C). Many of the antibodies tested displayed broad cross-reactivity and EC₅₀ values lower than libivirumab, a human anti-HBs monoclonal antibody that was isolated from HBV-immunized humanized mice and then tested clinically (Eren et al., 2000; Eren et al., 1998; Galun et al., 2002). These antibodies were not polyreactive or autoreactive when tested in polyreactivity ELISA and HEp-2 immunofluorescence assays respectively (Figure S3A and S3B). We conclude that the antibodies tested are broadly cross-reactive with different HBV serotypes.

Antigenic Epitopes on S-protein

To determine whether the selected antibodies bind to overlapping or non-overlapping epitopes, we performed competition ELISA assays, in which the S-protein was pre-incubated with a selected antibody followed by a second biotinylated antibody. Antibodies that showed weak levels of binding in ELISA (H002, H012, H013, H014, H018) were excluded. As expected, all of the antibodies tested blocked the binding of the autologous biotinylated monoclonal (yellow boxes in Figure 4A), while control human anti-HIV antibody 10–1074 failed to block any of the anti-HBs antibodies. The competition ELISA identified three mutually exclusive groups of monoclonal antibodies, suggesting that there are at least three dominant non-overlapping antigenic sites on HBsAg (red box for Group-I, blue box for Group-II, and H017 in Group-III, Figure 4A). Each of the individuals that had 2 or more antibodies tested in the competition ELISA expressed monoclonal antibodies that targeted 2 of the 3 non-overlapping epitopes (Figure 4A and S2B).

To define these epitopes more precisely, we produced a series of alanine mutants spanning most of the predicted extracellular domain of the S-protein with the exception of cysteines, alanines, and amino acid residues critical for S-protein production (Salisse and Sureau, 2009) (Figure 1D). ELISA assays with the representative antibodies from each antibody group and the mutant proteins revealed a series of binding patterns partially corresponding to the three groups defined in the competition assays (Figure 4B and S4). For example, mutations I110A and T148A interfered with binding by Group-I antibodies exemplified by H004, H006, H019, and H020, but had little measurable effect on Group-II antibodies exemplified by H007, H015, and H016 or Group-III antibody H017 (Figure 4B and S4).

However alanine scanning suggested that some residues such as D144 and G145 are critical for binding of monoclonals in both Group-I and Group-II despite their inability to compete with each other for binding to the native antigen (Figure 4B and S4). We speculate that D144A and G145A mutations alter the overall structure of HBsAg thereby interfering with binding of antibodies that normally target independent sites on the protein.

In addition to alanine scanning, we also produced 44 common naturally occurring escape variants found in chronically infected individuals (Hsu et al., 2015; Ijaz et al., 2012; Ma and Wang, 2012; Salpini et al., 2015). Whereas alanine scanning showed that some of the antibodies in Group-I and -II were resistant to G145A, the corresponding naturally occurring mutations at the same position, G145E and G145R, revealed decreased binding by most antibodies (Figure 4C). Among the antibodies tested, H017 and H019, in Groups-I and -III respectively, showed the greatest resistance to G145 mutations and the greatest breadth and complementarity (Figure 4C). We conclude that human anti-HBs monoclonals obtained from the selected individuals recognize distinct epitopes on HBsAg, most of which appear to be non-linear conformational epitopes spanning different regions of the protein.

In Vitro Neutralizing Activity

To determine whether the new monoclonals neutralize HBV in vitro, we performed neutralization assays using HepG2-NTCP cells (Figure 5A and 5B). The 50% inhibitory concentration (IC₅₀) values were calculated based on HBsAg/HBeAg ELISA or immunofluorescence staining for HBcAg expression (Michailidis et al., 2017) (Figure 5C). Neutralizing activity was further verified by in vitro neutralization assays using primary human hepatocytes (Michailidis et al., 2020) (Figure 5C and 5D). Fourteen of the 20 antibodies tested showed neutralizing activity with IC₅₀ values as low as 5 ng/ml (Figure 5C). By comparison, libivirumab had an IC₅₀ of 35 and 128 ng/ml in the neutralization assays based on ELISA and immunofluorescence assays respectively (Figure 5C). Somatic mutations were essential for potent neutralizing activity as illustrated by the reduced activity of the inferred UCAs (Figure S5A and S5B). In addition, optimal activity required bivalent binding since the IC₅₀ values for Fab fragments were 2 orders of magnitude higher than intact antibodies (Figure 5E). Finally, there was no overt synergy when Group-I, -II, and -III antibodies were combined (Figure S5C). We conclude that half of the new monoclonals were significantly more potent than libivirumab including Group-I H004, H005, H006, H008, H009, H019, and H020 and Group-II H007, H015, and H016 (Figure 5C).

Structure of the H015 Antibody/Peptide Complex

H015 differed from other antibodies in that its binding was inhibited by 5 consecutive alanine mutations spanning positions K141-G145 indicating the existence of a linear epitope. This idea was verified by ELISA against a series of overlapping peptides comprising the predicted extracellular domain of S-protein (Figure 6A and S6A). The data showed that H015 binds to KPSDGN, which is near the center of the putative extracellular domain and contains some of the most frequently mutated amino acids during natural infection.

To examine the molecular basis for H015 binding, its Fab fragment was co-crystallized with the target peptide epitope PSSSSTKPSDGNSTS, where all cysteine residues that flank the recognition sequence were substituted with serine to avoid non-physiological cross-linking. The 1.78 Å structure (Figure 6B and S6B) revealed that the peptide is primarily bound to the immunoglobulin heavy chain (Figure 6B and 6C), interacting with residues from CDR1 (R31), CDR2 (W52, F53) and CDR3 (E99, P101, L103, L104) of IgH with only one contact with CDR3 (P95) of IgL. The peptide adopts a three-residue beta hairpin (class 3) of the 3:5 type involving residues K141 through G145 as only one hydrogen bond is seen, between K141 and G145 (Milner-White and Poet, 1986), and they are not part of a beta sheet. The peptide is further stabilized by a salt-bridge formed between K141 and D144 (Figure 6D and S6C). Interestingly, the distance between the Cαs of the two residues (C139 and C147) flanking the recognition residues is 6.4 Å and are poised to form a disulfide bond between C139 and C147 found in the native HBsAg structure (Ito et al., 2010). The H015 Fab appears to stabilize the conformation of the peptide via the Fab-peptide contacts (Figure S6D) including a large binding surface (866 Ų; antibody-antigen buried surface of 600–900 Ų (Braden and Poljak, 1995)) comprised primarily of a single salt-bridge (lysine to aspartate; 0.9± 0.3 Kcal/mol) (White et al., 2013) and five hydrogen bonds (1–2 Kcal/mol/bond) (Sheu et al., 2003). Moreover, the peptide further restricts loop through intra-peptide contacts (Figure S6D) even in the absence of the disulfides.

The residues that form the hairpin are important for anti-HBs antibody recognition as determined by alanine scanning (Figure 4B and S4). In addition, each of these residues has been identified as important for immune recognition during natural infection (Ma and Wang, 2012). G145R, the most common naturally occurring S-protein escape mutation substitutes a large positively charged residue for a small neutral residue (circled residue in Figure 6E) potentially altering the antigenic binding surface. G145 adopts a positive phi angle of 77.9 and by doing so introduces a kink in the beta-strand, a structure that would be disrupted with the substitution to arginine.

HBsAg can be glycosylated at N146 and this site is also strictly conserved. However, some studies have suggested that this glycosylation site is never fully occupied, resulting in a nearly 1:1 ratio of glycosylated and non-glycosylated isoforms on the surface of viral envelope (Julithe et al., 2014). The glycosylation may be either NAG-NAG-MAN or NAG-(FUC)-NAG-MAN (Hyakumura et al., 2015). We have modeled both fucosylated and non-fucosylated options by grafting a 7mer and 11mer glycan conjugated at N146 of peptide in the presence of the Fab. We found that both glycosylation forms are tolerated at that location with only minimal torsional adaptations without clashes with the Fab, though the fucosylated (branched) glycan required some additional torsional angle changes to the Fab, as well.

Protection and Therapy in Humanized Mice

HBV infection is limited to humans, chimpanzees, tree shrews, and human liver chimeric mice (Sun and Li, 2017). To determine whether our anti-HBs bNAbs prevent infection in vivo we produced human liver chimeric Fah^−/−NODRag1^−/−IL2rg^null (huFNRG) mice (de Jong et al., 2014) and injected them with control or H020 (Group-I) or H007 (Group-II) antibodies before infection with HBV (Figure 7A-7D). These two antibodies were chosen because they bind to non-overlapping sites, and have broad and potent neutralizing activity. Whereas all six control animals in two independent experiments were infected, pre-exposure prophylaxis with either H007 or H020 was fully protective (Figure 7B-7D). We conclude that single anti-HBs bNAbs targeting different epitopes on the major virus surface antigen can prevent infection in vivo.

To determine whether bNAbs can also control established infections, we infused control antibody or bNAb H020 (Group-I) or H007 (Group-II) into huFNRG mice with HBV viral loads of 10⁶-10⁸ copies/ml of serum (Figure 7E-7H and Figure S7A). Fah^−/−NODRag1^−/−IL2rg^null mice are highly immunodeficient and unable to mount adaptive immune responses due to absence of T and B lymphocytes. In addition, the IL2rg^null mutation prevents cytokine signaling through multiple receptors, leading to a deficiency in innate immune function including antibody-dependent cellular cytotoxicity. Thus, elimination of viremia of 10⁶-10⁸ DNA copies/ml in huFNRG mice by antibody therapy alone would not be expected.

Animals that received the control antibodies further increased viremia to as high as $~$10¹¹ DNA copies/ml (Figure 7F). In contrast, the 5 mice that received H020 maintained stable levels of viremia for around 30 days (Figure 7G), after which time 2 mice showed increased viremia (arrow-1/3 in Figure 7G). A similar result was observed in the 5 mice that received H007 (Figure 7H), where only one showed a slight increase viremia at around day 50 (arrow-5 in Figure 7H).

To determine whether the animals that showed increased HBV DNA levels during antibody monotherapy developed escape mutations, we sequenced the viral DNA recovered from mouse blood. All three mice that escaped H020 (Group-I) or H007 (Group-II) monotherapy developed viruses that carried a G145R mutation in the S-protein (arrow-1/3 in Figure 7G, arrow-5 in Figure 7H, Figure 7I, and Figure S7). This mutation represents a major immune escape mutation in humans (Zanetti et al., 1988). Furthermore, mutations at the same position in the S-protein were also identified in mice that maintained low level viremia (arrow-2/4 in Figure 7G, arrow-6/7 in Figure 7H, Figure 7I, and Figure S7), but not in control animals (Figure S7). These results show that anti-HBs bNAb monotherapy leads to the emergence of escape mutations that are consistent with bNAb binding properties in vitro (Figure 4C).

To determine whether a combination of bNAbs targeting 2 separate epitopes would interfere with the emergence of resistant strains, we co-administered H006 + H007 (Group-I and -II, respectively) to 8 HBV-infected huFNRG mice (Figure 7J). H006 (Group-I) was chosen for this purpose because of its resistance to D144A and G145A mutation (Figure 4B). Similar to H007 monotherapy, there was only a slight increase in viremia in animals treated with the H006 + H007 anti-HBs bNAb combination during the 60-day observation period (Figure 7J and S7A). However, sequence analysis revealed that 3 of the mice developed resistance mutations including K122R/G145R, C137Y, and C137Y/D144V (arrow-8/9/10 in Figure 7J, Figure 7I, and Figure S7). These mutations confer loss of binding to both H006 (Group-I) and H007 (Group-II) (Figure 4C). Thus, the combination of 2 anti-HBs bNAbs targeting separate epitopes but susceptible to the same clinical escape variants is not sufficient to inhibit emergence of escape mutations.

To attempt to block the emergence of escape mutations, we combined H017 + H019 (Group-III and -I, respectively) bNAbs because they displayed complementary sensitivity to commonly occurring natural mutations (Figure 4C). None of 7 mice treated with the combination of showed increased viremia or escape mutations as assessed by sequence analysis (Figure 7K and S7A). Similar effects were also observed in the 9 animals treated with the H016, H017 and H019 (Group-II, -III, -I, respectively) triple antibody combination (Figure 7L and S7A). Moreover, both these combinations dramatically reduced the HBsAg levels in serum (Figure S7C and S7D). Altogether, these findings suggest that control of HBV infection by bNAbs requires a combination of antibodies targeting non-overlapping groups of common escape mutations.

RESOURCE AVAILABILITY

Data and Code Availability

The published article includes all dataset generated or analyzed during this study. Atomic coordinates and experimental structure factors have been deposited in the PDB under accession code 6VJT.

EXPERIMENTAL MODELS AND SUBJECT DETAILS

METHOD DETAILS

HBV Stock

HepDE19 cells (Cai et al., 2012) were cultured in the absence of tetracycline to induce HBV replication. After seven days, supernatant was collected every other day for two weeks and fresh medium was added. After each collection, medium was spun down to remove cell debris, passed through a 0.22 μm filter, and kept at 4°C. Collected medium was concentrated 100-fold via centrifugation using Centricon Plus-70 centrifugal filter devices (Millipore-Sigma, Billerica, MA). Mouse-passaged genotype C HBV virus (Billerbeck et al., 2016) was used for in vivo mouse experiment.

In Vitro HBV Neutralization Assay

In vitro HBV infection was performed as previously described (Michailidis et al., 2017). Briefly, HepG2-NTCP cells were seeded in 96-well collagen-coated plates in DMEM supplemented with 10% FBS and 0.1 mM NEAA. The medium was changed to DMEM with 3% FBS, 0.1 mM NEAA, and 2% DMSO the next day and cultured for an additional 24 hours before infection. The inoculation was in DMEM supplemented with 3% FBS and 0.1 mM NEAA 4% PEG and 2% DMSO. Antibodies or serum samples were incubated with the virus in the inoculation medium for one hour at 37°C before adding to cells. Serum neutralization capacity (y-axis in Figure 1A and S1B) was calculated as the reciprocal of the relative percentage of infected HepG2-NTCP cells immunostained by rabbit anti-HBV core antibody (AUSTRAL Biologicals). For example, if the relative percentage of infected cells were 100% (no serum added or the sera from unexposed naïve control donors), the serum neutralization capacity would be calculated as 1; but if the relative percentage of infected cells were 50% or 10%, the serum neutralization capacity would be 2 or 10. For the blocking neutralization assay, S-protein antigen at different concentration was incubated with purified polyclonal antibodies for one hour at 37°C before incubation with HBV virus. The cells were then spinoculated for one hour by centrifugation at 1,000 g at 37°C. After a 24-hour incubation, supernatant was removed, cells were washed five times with PBS, and 100 μl of fresh DMEM supplemented with 3% FBS, 0.1 mM NEAA, and 2% DMSO. Both supernatant and cells were harvested 7 days after infection for analysis. Neutralization assays in primary human hepatocytes were performed as above using hepatocytes from livers of highly humanized mice that were harvested and seeded on collagen-coated plates in hepatocyte defined medium (Corning) (Michailidis et al., 2020).

QUANTIFICATION AND STATISTICAL ANALYSIS

The detailed information of statistical analysis could be found in the Result and Figure Legends. Correlation was evaluated by Spearman’s rank correlation method (Figure 1A and 2B). Statistical significance was calculated by Dunn’s Kruskal-Wallis multiple comparisons with p values corrected with the Benjamini-Hochberg procedure (Figure S1C). The 50% effective concentration (EC₅₀) values by ELISA assays (Figure 3A and 3C) and 50% inhibitory concentration (IC₅₀) values by neutralization assays (Figure 5C) were calculated by nonlinear regression analysis in PRISM software.