Bispecific Antibodies Targeting Different Epitopes on the HIV-1 Envelope Exhibit Broad and Potent Neutralization

ABSTRACT

The potency and breadth of the recently isolated neutralizing human monoclonal antibodies to HIV-1 have stimulated interest in their use to prevent or to treat HIV-1 infection. Due to the antigenically diverse nature of the HIV-1 envelope (Env), no single antibody is highly active against all viral strains. While the physical combination of two broadly neutralizing antibodies (bNAbs) can improve coverage against the majority of viruses, the clinical-grade manufacturing and testing of two independent antibody products are time and resource intensive. In this study, we constructed bispecific immunoglobulins (IgGs) composed of independent antigen-binding fragments with a common Fc region. We developed four different bispecific IgG variants that included antibodies targeting four major sites of HIV-1 neutralization. We show that these bispecific IgGs display features of both antibody specificities and, in some cases, display improved coverage over the individual parental antibodies. All four bispecific IgGs neutralized 94% to 97% of antigenically diverse viruses in a panel of 206 HIV-1 strains. Among the bispecific IgGs tested, VRC07 × PG9-16 displayed the most favorable neutralization profile. It was superior in breadth to either of the individual antibodies, neutralizing 97% of viruses with a median 50% inhibitory concentration (IC₅₀) of 0.055 μg/ml. This bispecific IgG also demonstrated in vivo pharmacokinetic parameters comparable to those of the parental bNAbs when administered to rhesus macaques. These results suggest that IgG-based bispecific antibodies are promising candidates for the prevention and treatment of HIV-1 infection in humans.

IMPORTANCE To prevent or treat HIV-1 infection, antibodies must potently neutralize nearly all strains of HIV-1. Thus, the physical combination of two or more antibodies may be needed to broaden neutralization coverage and diminish the possibility of viral resistance. A bispecific antibody that has two different antibody binding arms could potentially display neutralization characteristics better than those of any single parental antibody. Here we show that bispecific antibodies contain the binding specificities of the two parental antibodies and that a single bispecific antibody can neutralize 97% of viral strains with a high overall potency. These findings support the use of bispecific antibodies for the prevention or treatment of HIV-1 infection.

INTRODUCTION

The neutralizing antibody response to human immunodeficiency virus type 1 (HIV-1) is directed initially to the infecting viral strain but generally broadens over time to recognize diverse isolates. The recognition that some HIV-1-infected individuals generate highly potent and broadly reactive neutralizing antibodies (bNAbs) led to the eventual isolation and characterization of numerous HIV-1 neutralizing monoclonal antibodies (MAbs) (1, 2). Characterization and structural analysis of these bNAbs have revealed the specific neutralization binding regions on the HIV-1 envelope glycoprotein (Env). We now appreciate at least five regions of vulnerability on the HIV-1 Env trimer: the CD4 binding site (CD4bs), a glycan-dependent site near the variable loop 3 (V3) region (V3-glycan), a variable-region (V1V2) glycan-dependent site on the trimer apex, a membrane-proximal external region (MPER) of gp41, and a region at the interface of gp120 and gp41 (3). Our understanding of the structural mode of recognition by many bNAbs, together with the structure of the native trimer (4–6), is providing new insights relevant to vaccine design. In addition, studies of the immune pathways leading to the development of these bNAbs are providing new approaches for immunization (7–9).

Despite these advances, current vaccine immunogens elicit antibodies with limited neutralization breadth (7, 10, 11), and it will likely take years of improved vaccine designs and iterative clinical trials to developed more effective vaccines. This challenge has led to an interest in the use of bNAbs as part of an overall strategy to prevent new HIV-1 infections (12). Passive immunization in humans has proven highly effective for infections with many viruses, including hepatitis A, hepatitis B, rabies, and respiratory syncytial viruses (13), and passive administration of bNAbs to HIV-1 Env can completely prevent infection of macaques in simian-human immunodeficiency virus (SHIV) infection models (14–16). More recently, bNAbs have been tested for treatment of HIV-1 and SHIV infection in the mouse and nonhuman primate (NHP) models, respectively, with initial promising results (17–19). Notably, in these in vivo therapeutic models, combinations of two or more bNAbs appear to be more effective than a single antibody. To date, two HIV-1 bNAbs that target the CD4 binding site on the HIV-1 Env, VRC01 and 3BNC117, have demonstrated safety in phase I clinical trials and the ability to transiently lower the plasma viral load (20–22). bNAbs to other neutralization epitopes are also being considered for development but have not yet entered phase I trials. While human clinical trials will be needed to assess the potential of bNAbs for prevention or therapy, it is likely that both potency and breadth of neutralization will play roles in overall efficacy. In this regard, the marked antigenic diversity of HIV-1 remains a major obstacle. No single MAb can neutralize the vast majority of virus strains with high potency. Some MAbs are particularly broad in coverage, while others are particularly potent. Thus, combinations of two or more bNAbs have been considered to optimize neutralization potency and breadth and, in the case of treatment, to minimize the potential for emergence of viral resistance (18, 19, 23).

The development of a monoclonal antibody for clinical use requires extensive preclinical data, stable cell line development, clinical-grade manufacturing, and phase I safety testing. Thus, the potential to make multiple HIV-1 bNAbs may be limited by time and resources. This has led to the consideration of bispecific antibody formats that can incorporate two different antigen specificities into one molecule. For targeting HIV-1 infection, bispecific antibodies have been developed in multiple formats that combine either receptor-targeting antibodies or domains with anti-HIV-1 antibodies or linking the variable domains of two anti-HIV-1 antibodies (24–27). These antibodies demonstrate increased potency and breadth against HIV-1, although the ability to develop them for clinical use may be limited due to their use of nontraditional IgG formats.

In this study, we used the CrossMab format (28) to make bispecific IgGs that were based on four parental HIV-1 bNAbs, one each to the CD4bs, V3-glycan, V1V2-glycan, and MPER regions of Env. Based on these bNAbs, we chose combinations that were predicted to improve overall neutralization coverage and thus constructed four different bispecific IgGs, each with a different combination of two antigen binding fragments. We demonstrate that each arm of the bispecific IgG is functional and that the optimal combination of VRC07 × PG9-16 displayed neutralization potency and breadth characteristics that were better than those of any single parental bNAb and similar to those of the physical combination of the two parental IgGs.

MATERIALS AND METHODS

Antibodies.

Bispecific antibodies in the CrossMab CH1-CL format were designed as described previously (28). Of the three CrossMab formats, we chose the CH1-CL version due to its low by-product profile. The CH1-CL swap was performed on either VRC07 or 10E8 and associated with hole mutations (T366S, L368A, Y349C, and Y407V) in the CH3 region. The partner arm was left unswapped but contained knob mutations (T366W and S354C) in the CH3 region. All antibodies had a kappa light chain on the swapped arm and a lambda light chain on the other. Plasmids encoding the heavy and light chains were made by gene synthesis (GenScript, Piscataway, NJ). Antibodies were made by transient cotransfection of four plasmids—2 heavy chains and 2 light chains—in Expi293F cells (as per the manufacturer's instructions). Plasmid ratios were optimized for each antibody to obtain the greatest purity levels with the least amount of heavy-chain dimers and half-antibodies. Supernatants were harvested 6 days posttransfection, filtered, and purified by protein A affinity chromatography (GE Healthcare). Elutes were neutralized in 1 M Tris (pH 8.0), dialyzed into phosphate-buffered saline (PBS), and concentrated. Using this method, we routinely obtained greater than 90% of the fully assembled bispecific IgG. Antibodies run on a Superose 6 column showed a single peak without aggregates. Parental IgGs and antigen binding fragments (Fab) were made by cotransfection of heavy- and light-chain-encoding plasmids in Expi293F cells and purified from supernatants as described above, with the exception that either anti-human kappa or lambda beads (GE Healthcare) were used for Fab.

Protein purification.

His-tagged versions of proteins RSC3 (29), 3AGJ-MPER-green fluorescent protein (GFP) (30), and 1VH8_CAP256SU (31) were transiently expressed in Expi293F or GnTI^−/− cells and purified on nickel-nitrilotriacetic acid (Ni-NTA) columns (GE Healthcare), followed by gel filtration on a Superdex 200 16/60 column (GE Healthcare). 1VH8_CAP256SU is a trimeric protein that recapitulates the quaternary alignment of the three V1V2 domains present in the native HIV-1 Env spike. BG505.SOSIP.6R.664.T332N.D368R was purified as described previously (32). The D368R mutation was introduced to knock down VRC07 binding.

Biolayer interferometry.

Bispecificity of bispecific antibodies was confirmed on an Octet RED384 system (Fortebio, Menlo Park, CA) in a sandwich format. Streptavidin biosensors (Fortebio) were loaded with biotinylated RSC3 (10 μg/ml for 60 s), dipped in the antibody (25 μg/ml for 120 s), and then probed with a ligand against the second arm of the antibody (either 3AGJ-GFP at 100 μg/ml, BG505.SOSIP.6R.664.T332N.D368R at 300 μg/ml, or 1VH8_CAP256SU at 50 μg/ml for 120 s). For 10E8 × PG916RSH, AR2G tips were loaded with 20 μg/ml of 1VH8_CAP256SU in 10 mM sodium acetate (pH 6) for 300 s and dipped in 50 μg/ml of antibody followed by 100 μg/ml of 3AGJ-GFP. Activation and quenching steps were performed as per the manufacturer's instructions. Baselines were established before and after the loading step. All assays were performed in 1× kinetics buffer. Pilot experiments confirmed the specificity of each ligand to the parental monoclonal antibody.

Antibody affinities were measured on an Octet RED384 system. Anti-human Fc biosensors (Fortebio) were loaded with either the parental or bispecific antibodies and dipped in various concentrations of the ligand (RSC3 for VRC07, 3AGJ-MPER-GFP for 10E8, and BG505.SOSIP for PGT121). Fab fragments were loaded on an AR2G sensor in 10 mM sodium acetate (pH 6.0). For PG9-16-containing antibodies, 1VH8_CAP256SU (10 mM sodium acetate, pH 6) was loaded on AR2G biosensors and dipped in various concentrations of PG9-16-Fab or MAbs. Assay variables, like loading concentration, time, and ligand concentration, were optimized for each antibody-ligand pair to obtain an R² of >0.99 and χ² of <3. For all antibody-ligand pairs, biosensors were hydrated in buffer and loaded, and a baseline was established. Association was performed in ligand in 1× kinetics buffer (Fortebio) for 100 to 400 s, followed by dissociation in 1× kinetics buffer for 100 to 600 s. Buffer for binding with 3AGJ-MPER-GFP was supplemented with 0.1% Tween 20 to limit nonspecific binding to the biosensor. All steps were performed at 30°C and 1,000 rpm. Binding curves were fit to a 1:1 Langmuir binding model on Octet data analysis software version 8.1.0.53.

Autoreactivity.

Antibodies were tested for autoreactivity on two platforms: anti-nuclear antibodies by staining on HEp2 cells (Zeus International) and anticardiolipin enzyme-linked immunosorbent assay (ELISA; Inova Diagnostics) as described earlier (14).

In vitro neutralization.

Antibody neutralization against a panel of 206 HIV-1 Env pseudoviruses was tested in a single-round TZM-bl neutralization assay as described previously (14, 23). A 384-well automated system was used to perform the assays. Single bispecific IgGs and parental antibodies were assayed at a starting concentration of 25 μg/ml. For physical combination of two antibodies, we used 12.5 μg/ml as the starting concentration for each antibody to give a final concentration of 25 μg/ml. Potency-breadth (PB) curves were generated using GraphPad Prism version 6.0c.

Animals.

Naive Macaca mulatta animals of Indian origin were used. Each group had equal numbers of male and female animals. All animal experiments were reviewed and approved by the Animal Care and Use Committee of the vaccine Research Center, NIAID, NIH. All animals were housed and cared for in accordance with local state, federal, and institute policies in an American Association for Accreditation of Laboratory Animal Care-accredited facility either at the National Institutes of Health or at Bioqual Inc. USA.

Pharmacokinetics.

Indian-origin rhesus macaques were administered low-endotoxin antibody preparations (<1 endotoxin unit [EU]/mg) intravenously at 10 mg of Ab/kg of body weight. Whole-blood samples were collected prior to injection and at multiple time points until week 4 postadministration. Antibody concentrations in serum were measured by ELISA as described previously (14), with the exception that plates were coated with 200 ng/well of ZM109F.PB4 gp120 (Immune Technology Corp). Pharmacokinetic parameters were calculated with WinNonlin software using a two-compartment model, and statistical significance was tested by one-way analysis of variance (ANOVA) in GraphPad Prism version 6.0c.

RESULTS

Construction and characterization of bispecific antibodies.

We made four bispecific IgGs based on antibodies targeting four major sites of neutralization on the HIV-1 trimer: VRC07 for the CD4-binding site, PGT121 for the V3-glycan supersite, PG9-16-RSH for the V1V2 apex, and 10E8 for the MPER (Fig. 1A). Note that PG9-16-RSH (henceforth referred to as PG9-16) is a chimeric antibody based on PG9, but it includes three specific contact residues from the closely related PG16 to enhance neutralization potency, as previously described (33). We chose four combinations of these bNAbs that are predicted to give >95% neutralization coverage based on the characteristics of the parental bNAbs: VRC07 × 10E8, VRC07 × PGT121, VRC07 × PG9-16, and 10E8 × PG9-16 (Fig. 1B). The bispecific IgGs were constructed using the CrossMab CH1-CL format with knob-hole mutations in the CH3 domain to confer heavy-chain heterodimerization and swapping of the CH1 and light-chain constant (CL) domains on one arm of the antibody to limit light-chain mispairing (28). We optimized transfection ratios of the two different heavy chains and used a 2-fold excess of each light chain to its cognate heavy chain to obtain >90% fully assembled IgG. This approach reduced the generation of side products such as half-antibodies (one heavy chain and one light chain) and heavy-chain dimers (IgG lacking light chains), enabling the isolation of IgG in a single step of protein A chromatography with a predominant single peak on the size exclusion chromatogram (Fig. 1C and D). Under reducing SDS-PAGE conditions, the two different heavy chains were discernible in all antibodies and were of equal intensities. The glycosylated PG9-16 light chain of a higher molecular weight could be identified in the corresponding two bispecific IgGs and was of a quantity similar to that of the other light chain (Fig. 1C).

FIG 1.

Bispecific antibodies. (A) Structure of HIV-1 BG505 SOSIP.664 Env trimer (PDB code 4TVP) showing the contact sites of the antibodies used in this study. Due to the lack of a crystal structure of PGT121, the footprint of the closely related antibody PGT122 is shown. A portion of the PGT122 binding site on the second protomer can be seen. The third binding site of PGT122 and the two other binding sites of VRC07 are on the reverse face of the trimer. (B) Schematic of the CrossMab antibody configuration. (C) Nonreduced and reduced SDS-PAGE analysis of bispecific IgGs made in the CrossMab format. HC, heavy chain; LC, light chain. (D) Analytical size exclusion profiles of the bispecific IgGs.

Binding characteristics of the bispecific antibodies.

We assessed the bispecificity of the IgGs by using biolayer interferometry with a unique ligand for each of the parental IgGs (Fig. 2A). The first ligand was loaded on the biosensor, followed sequentially by binding with either the parental or bispecific IgG and then the second ligand. All bispecific IgGs bound both ligands simultaneously, whereas the parental antibodies could not bind the second ligand. In each case, the binding of the second ligand was of somewhat lower magnitude than that of the first, likely due to steric hindrance introduced by immobilization of the first ligand and orientation of binding of the antibody that limits accessibility to the second antigen-binding site of the bispecific antibody. Nonetheless, these data confirmed that both arms of the bispecific antibodies were active.

FIG 2.

Binding characteristics of anti-Env bispecific IgGs. (A) Simultaneous ligand binding to both arms of the bispecific IgG is demonstrated by a sandwich assay using biolayer interferometry. Octet biosensors were loaded with the ligand against one arm of the bispecific IgG, either by biotinylation or by amine coupling, and then probed sequentially with the bispecific IgG and the ligand against second arm. As controls, parental IgGs were used in place of the bispecific IgG. (B) Kinetic characterization of bispecific and parental IgGs was assessed using biolayer interferometry (Octet RED384). Kinetic constants with standard error and fit parameters are summarized.

We also assessed affinity of each antibody specificity against a specific ligand and compared the bispecific IgG to the parental IgG and Fab (Fig. 2B; see also Fig. S1 in the supplemental material). The CD4bs scaffold (RSC3) and MPER scaffold (3AGJ-MPER-GFP) ligands are monomeric and were therefore kept in solution while the antibodies were immobilized on anti-human Fc sensors. We noted that due to the difference in number of ligand-binding arms, the parental VRC07 and 10E8 IgGs bound roughly 2-fold more ligand than the bispecific IgG. Despite the difference in the amount of ligand binding, the affinity of binding of the bispecific was not affected, as shown by the kinetics of binding of the respective VRC07 and 10E8 MAbs and Fabs. The ligands for the V3-glycan bNAbs (BG505.SOSIP.T332N.D368R) and V1/V2 bNAbs (1VH8_CAP256SU) are trimeric and were therefore immobilized on the biosensor and probed with IgG or Fab in solution. With the BG505.SOSIP.D368R trimer, we did not observe measurable dissociation and therefore performed the assay with the same trimer in solution. In this format, the IgG and Fab of PGT121 showed kinetics similar to those of the bispecific VRC07 × PGT121, with a marginal improvement for the bispecific IgG due to background binding from the VRC07 arm. When the V1V2 scaffold 1VH8_CAP256SU was immobilized, the bispecific affinities were similar to that of the Fab, while the PG9-16 IgG had a much slower off-rate, presumably due to rebinding from the second arm of the IgG or bivalent binding. Overall, each of the four bispecific IgGs retained the affinity of binding against a specific ligand, maintaining both on- and off-rates, compared to the relevant Fab.

Neutralization.

We next tested the neutralization potency and coverage of the parental and bispecific IgGs using a panel of 206 Env-pseudoviruses. As expected, all four bispecific IgGs demonstrated broad coverage, neutralizing between 94 and 97% of viruses at <25 μg/ml, with median potencies between 0.055 and 0.270 μg/ml (Fig. 3A). Comparisons of each bispecific IgG to its constituent parental IgGs showed several patterns. Parental VRC07 had a breadth of 93%, with a median 50% inhibitory concentration (IC₅₀) of 0.143 μg/ml. Each of the VRC07 bispecific antibodies had slightly better coverage (94 to 97%) than VRC07 and in two cases, VRC07 × PGT121 and VRC07 × PG9-16, the median potencies (0.091 and 0.055 μg/ml, respectively) were better than that of VRC07. These median potencies were not better than those of parental PGT121 (0.014 μg/ml) and PG9-16 (0.040 μg/ml), but it is important to note that the potency values of PGT121 and PG9-16 are based on a smaller set of neutralization sensitive viruses due to lower neutralization coverage (n = 65 and 79, respectively). This finding is represented in potency-breadth (PB) curves where the VRC07 × PGT121 bispecific curve matched the overall breadth of coverage of parental VRC07 IgG at 25 μg/ml and was closer to that of VRC07 at higher concentrations and closer to that of PGT121 IgG at lower concentrations (Fig. 3B). Among the bispecific IgGs, VRC07 × PG9-16 displayed the most favorable characteristics compared to its parental IgGs, as evidenced by its PB curve showing coverage as good as, or better than, that of either parental IgG at all concentrations. A comparison of all four bispecific IgGs confirms that VRC07 × PG9-16 displayed the most favorable potency-breadth profile (Fig. 3C). Interestingly, the extraordinary breadth of 10E8 (99%) could not be fully reproduced in the bispecific format, with the VRC07 × 10E8 bispecific antibody missing 5 viruses (97% coverage) and 10E8 × PG9-16 missing 8 viruses (96% coverage). Notably, for these 13 viruses, 10E8 IgG had a modest (0.3 to 1.0 μg/ml) or weak (>1.0 μg/ml) potency, which was lost in the bispecific format (see Fig. S2 in the supplemental material). Despite this overall slight loss of breadth, the 10E8-containing bispecific IgGs covered more viruses than parental 10E8 IgG at the lower antibody concentrations (e.g., <1.0 μg/ml) (Fig. 3A and B).

FIG 3.

Neutralization breadth-potency of the bispecific IgGs. (A) Neutralization breadth of the parental and bispecific IgGs was tested against a panel of 206 viral strains. The antibodies were tested at a starting concentration of 25 μg/ml with serial dilutions. Breadths based on IC₅₀s and IC₈₀s are summarized. Potency is shown as medians and geometric mean values calculated against sensitive viruses. (B) Potency-breadth curves comparing the four bispecific IgGs to their parental IgGs. (C) Potency-breadth curves of the four bispecific IgGs.

Next, we compared the potency of the bispecific antibodies with the parental antibodies in two ways: first by comparing each bispecific IgG with its parental IgG on a per-virus basis (Fig. 4) and second by splitting the virus panel into 4 categories (dually sensitive, monosensitive to one or the other parental IgG, and dually resistant) based on the parental IgGs and assessing bispecific antibody performance in each category at the population level (Fig. 5). To facilitate comparison of antibody potencies against each virus, the data for each bispecific antibody are shown compared to that of each parental antibody (left and right sides of Fig. 4). Viruses were ordered by the neutralization potency of one of the parental antibodies (black dots in graphs) and overlaid with the potency of the bispecific antibody (red, blue, green, or purple symbols) and the second parental antibody (gray symbols) against the same virus. When graphed in this way, it is possible to compare the bispecific antibody to each parental IgG (Fig. 4). The ideal bispecific antibody would retain the full potency of each parental IgG, but this outcome is not expected because neutralization by a bispecific IgG is mediated by a single arm, compared to potential bivalent reactivity of the parental IgG. The resultant gain or loss in potency of the bispecific IgG in comparison to the parent IgG was variable across the different bispecific IgGs (2.4- to 27-fold gain and 1.3- to 14-fold loss).

FIG 4.

Potencies of the bispecific IgGs compared to those of the parental IgGs. Each dot on the graph represents a virus. On the x axis, viruses are ordered 1 to 206 based on the sensitivity to the parental IgG (black) and overlaid with the potency of the bispecific IgG (red, blue, green, or purple symbols) and the second parental IgG (gray symbols) against the same virus. Dots below the black dots indicate increased potency. Number of viruses and median IC₅₀ change of the bispecific antibody compared to those for each parent IgG are shown for viruses with decreased (above) or increased (below) susceptibility to the bispecific IgG.

FIG 5.

Potency comparison of bispecific and combination IgGs. (A) Viruses were classified as sensitive (S) or resistant (R) based on the parental antibodies. Within each category, the median neutralization IC₅₀ (μg/ml) was calculated for each parental or bispecific IgG and the physical combination of parental IgGs. Parental and bispecific IgGs were started at 25 μg/ml, whereas the physical combination consisted of 12.5 μg/ml of each parental IgG (total 25 μg/ml). Fold difference in potency for the bispecific IgGs and combination was calculated as median IC₅₀ of parental IgG divided by the median IC₅₀ of the bispecific (or combination) IgG. Red values (values greater than 1.0) show improved potency over the parental IgG. (B) Median IC₅₀ (nanomolar) of parental Fab and bispecific IgG is compared to that of parental IgG among monosensitive viruses. Numbers in parentheses indicate fold change of Fab or bispecific IgG compared to parental IgG.

We then analyzed the performance of the bispecific IgG against viruses classified as sensitive to both or only one of the arms of the IgG and compared it to that of the parental IgG alone or in combination (Fig. 5A; see also Fig. S8 in the supplemental material). For dually sensitive viruses, the median IC₅₀ of the bispecific IgG was generally intermediate to that of the two parental antibodies, matching the more potent of the two parents within 2-fold. The remarkable exception was VRC07 × PG9-16, which was 6.9- and 2.2-fold more potent than VRC07 and PG9-16, respectively, against dually sensitive viruses. Against the monosensitive viruses (i.e., viruses sensitive to only one of the parental bNAbs), the bispecific IgGs were functional, with IC₅₀s ranging between 0.202 and 7.570 μg/ml. However, compared to the parental IgG, a 2.4- to 18.4-fold reduction in potency was observed (Fig. 5A, second-to-last column). Similar results were observed when the IC₈₀s were compared (see Fig. S8). Against monosensitive viruses, the bispecific IgG has only one active arm capable of binding the HIV-1 Env and would essentially behave like a Fab. We therefore compared the neutralization profiles of the IgGs and Fab fragments with the bispecific VRC07 × PG9-16 against a subset of viruses (Fig. 5B). This comparison demonstrated that for monosensitive viruses, the bispecific IgG generally recapitulated the activity of the Fab fragment of the parent IgG. We also assessed the neutralization of physical IgG combinations, assayed at the same final concentration as the single bispecific IgG and therefore with equivalent number of antigen-binding arms. For dually sensitive viruses, the physical combination was usually more potent than the bispecific IgG, with the exception again of VRC07 × PG9-16, which was 1.4-fold more potent than the combination of PG9-16 and VRC07 parental IgGs, respectively (Fig. 5A, last column). When tested against monosensitive viruses, the physical combinations were more potent than the bispecific IgG. The physical combinations were 0.4- to 1.3-fold less active than the better of the individual parental IgGs. This result was expected, as the physical combination of two IgGs contained a 2-fold-lower concentration of each IgG than the single parental IgG.

Pharmacokinetics of bispecific IgG in vivo.

To evaluate potential clinical use, we investigated the half-life and overall pharmacokinetics of the best bispecific IgG VRC07 × PG9-16 in Indian-origin rhesus macaques and compared it to those of the parental IgGs, VRC07 and PG9-16. All IgGs contained the LS mutation in the CH3 domain, which has been shown previously to improve antibody persistence (34), and were infused intravenously at 10 mg/kg. The serum half-life of VRC07 × PG9-16-LS IgG was ∼10 days, similar to those of both parental VRC07-LS and PG9-16-LS IgGs (Fig. 6A). Similar to the parental IgG, the bispecific IgG persisted in circulation for 4 weeks at >10 μg/ml and also retained neutralization activity in the serum (see Fig. S9 in the supplemental material). Also, like the parental IgGs, the bispecific IgGs did not show detectable poly- or autoreactivity as assessed by HEp-2 cell ANA staining and anticardiolipin ELISA (see Fig. S3).

FIG 6.

Bispecific IgG has pharmacokinetic properties similar to those of the parent IgG in rhesus macaques. (A) Endotoxin-free IgGs were administered at 10 mg/kg intravenously, and serum levels were monitored by ELISA against gp120 (ZM109F). Data from 2 or 4 animals are plotted as means ± standard deviations. (B) Pharmacokinetics parameters were calculated using a two-compartment model using WinNonlin software. Numbers are means ± standard errors.

DISCUSSION

Since 2009, a new generation of potent and broadly reactive neutralizing antibodies has been isolated and characterized. These bNAbs target a limited number of regions on the native trimeric Env and neutralize many HIV-1 strains with high potency (3). Depending on the epitope targeted or even specific characteristics of bNAbs to the same epitope, the overall potency or breadth of neutralization may vary. Antibodies to the CD4bs and the MPER are the most broadly reactive, neutralizing 90% to 98% of viruses, but often are not as potent as bNAbs to the V1V2-glycan or V3-glycan epitopes. Interestingly, these more potent glycan-reactive bNAbs are generally not as broadly reactive as bNAbs to the CD4bs or MPER, neutralizing 60% to 80% of viral strains. Thus, no single monoclonal antibody has the optimal characteristics of both high potency and high coverage. In this study, we sought to construct a single bispecific bNAb with the most favorable neutralization characteristics, by including arms from known bNAbs.

Neutralization of HIV-1 by IgG is generally believed to be monovalent; i.e., only one arm of the IgG engages the viral spike. A proposed explanation for this finding is that the spatial distance between epitopes on the HIV-1 spike and the distance between adjacent spikes prevents bivalent binding by IgG. Thus, bNAbs are believed to bind monovalently (see Fig. S4 in the supplemental material) (35). Despite monovalent binding, the presence of two binding arms in an IgG (compared to one in a Fab) may lower the off-rate and improve neutralization efficacy via a binding avidity effect. Thus, HIV-1 neutralization by IgG forms of bNAbs is often better than with the corresponding Fab (35, 36). When binding is truly bivalent, that is, both arms are simultaneously engaged in binding the same antigen molecule, there could be significant improvement in the off-rate and thereby neutralization potency as well. This has been demonstrated in a recent study that obtained bivalent binding by linked Fab molecules which exhibited superior potency over the physical combination of Fabs (24). Other studies have used different formats of bispecific antibodies. A bispecific antibody targeting human CD4 (ibalizumab) and either PG9 or PG-16 showed improved breadth and potency over both ibalizumab and the HIV-1 bNAb (25). Recently, a different format of bispecific antibody that combines bivalent CD4 with a CCR5 mimetic sulfopeptide was described (26). This antibody binds the conserved receptor and coreceptor binding sites of HIV-1 and showed a high level potency and breadth.

Since traditional IgG antibodies may be preferred for clinical use but are constrained by monovalency, we sought to construct and evaluate bispecific antibodies with optimal binding and neutralization characteristics. Of the available bispecific formats that retain natural IgG architecture (28, 37–39), we chose the CrossMab format for its ease of design, lack of autoreactive mutations, pharmacokinetic properties, and demonstrated in vivo efficacy (28, 40). We chose to combine bNAbs with the highest breadth, VRC07 and 10E8, with the more potent bNAbs—PGT121 and PG9-16—to generate VRC07 × PGT121, VRC07 × PG9-16, and 10E8 × PG9-16 bispecific IgGs. We also combined the two broadest antibodies and made the VRC07 × 10E8 bispecific IgG. While other combinations were possible, these four combinations provided a testable set by which to assess binding properties and in vitro neutralization breadth and potency in comparison to the parent Fab and IgG molecules.

The binding kinetics of the bispecific antibodies against specific ligands confirmed that the affinities of the epitope-paratope interactions were not affected by the CrossMab format. To control for differences in affinity versus avidity measurements due to assay orientation effects, we used Fab fragments of the corresponding parent IgGs as controls. Both on- and off-rates of the bispecific IgGs were comparable to those of the parental MAbs and Fabs (except for PG9-16 IgG, for which only apparent K_D could be measured). 10E8-containing bispecific IgGs made in two different formats, either as the switched arm with a kappa light chain in 10E8 × PG9-16 or as the unswitched arm with a lambda light chain in VRC07 × 10E8, had an affinity similar to that of the parental IgG against a specific ligand, 3AGJ-MPER-GFP. Therefore, in each case, the bispecific IgG retained the binding properties of its two parental IgGs.

All four of the bispecific IgGs we constructed retained functional binding and neutralization by both arms. They also displayed remarkable coverage on a panel of 206 viral strains, neutralizing 94% to 97% of viruses. The most favorable combination, VRC07 × PG9-16, neutralized 97% of all viruses tested, with a remarkable median potency of 0.055 μg/ml. Thus, this bispecific IgG has a neutralization profile that is superior in overall breadth and potency to that of any single parental bNAb, and the IgG was similar to the physical combination of the two parental IgGs. However, in other cases, the bispecific IgGs often did not retain the full potency of both parental IgGs. The potency of the bispecific IgG was often better than that of the weaker parental IgG but not as good as that of the more potent parental IgG. This feature can be seen in potency-breadth curves where the bispecific IgG curve falls in between the two parental curves (Fig. 3B). As an example, VRC07 × PGT121 matched the breadth of VRC07 at 25 μg/ml but had somewhat less coverage than VRC07 at between 0.2 and 25 μg/ml. At concentrations lower than 0.2 μg/ml, its coverage was better than that of VRC07 and more similar to that of PGT121. Similarly, at <1 μg/ml, 10E8 × PG9-16 matched the breadth of PG9-16, but at a higher concentration, it was unable to match the breadth of 10E8. Thus, not all bNAbs are equally amenable to function in a bispecific molecular format. In this regard, VRC07 × PG9-16 was an exception in having a PB curve that was equal to or above those of both parental IgGs (i.e., better coverage), accounting for its excellent overall potency and breadth.

Of note, the epitopes targeted by the antibody arms in the bispecific IgGs cannot be bridged by the Fab arms of an IgG molecule, thereby not facilitating bivalent binding (see Fig. S5 and S6 in the supplemental material). Thus, for any given virus, the bispecific IgG could be similar to, but generally not more potent than, the most potent parental IgG. However, since binding of either arm can contribute to virus neutralization, there is an increase in breadth of neutralization compared to those of the individual parental IgGs. Curiously, in a small subset of viruses, the bispecific IgG was at least 3-fold more potent than both parental IgGs and either matched or surpassed the potency of the combination antibodies (see Fig. S7). The mechanism for this observation is uncertain. When potency was analyzed as a function of virus sensitivity to the individual parental IgG, several interesting observations emerged. Among dually sensitive viruses, the bispecific IgG displayed only a small loss in potency. The exception was VRC07 × PG9-16, which showed a small gain in potency. The loss in number of binding arms of PG9-16 in the bispecific format is perhaps compensated for by requiring only one IgG molecule to be bound per trimer (41). This is supported by the similarity in the neutralization potency of the bispecific IgG and the physical combination of VRC07 and PG9-16 IgGs. Also, among dually sensitive viruses, the relationship between the potency of the bispecific IgG and that of the parental IgGs was variable, depending on the identity of the parental IgG (Fig. 4). VRC07 × 10E8 was mostly intermediate between VRC07 and 10E8 IgGs. However, for the other bispecific IgGs, there was a tendency to reproduce the characteristics of either PGT121 or PG9-16 IgGs. It is possible that accessibility, glycan recognition, and surface rather than internal location of contact residues for PGT121 and PG9-16 aid in this process. Of note, PG9-16 in the bispecific format largely retained the neutralization potency of the parental IgG, a characteristic that may be unique to the V1V2 apex antibodies.

In this study, we constructed bispecific IgGs containing the antigen binding arms of two different bNAbs. The ability of bispecific IgGs to recognize either of two neutralization epitopes led to improved neutralization coverage and may have advantages with regard to preventing neutralization escape. Bispecific IgGs that combined the broadly neutralizing CD4bs bNAb VRC07 with a potent glycan-dependent bNAb demonstrated the best combination of breadth and potency. Specifically, the VRC07 × PG9-16 bispecific IgG neutralized more viruses (97%) than either parental bNAb, with a potency that was similar to that of the physical combination of the two parental bNAbs. Thus, a single bispecific antibody can potentially substitute for a combination of two parental bNAbs. Also, the in vivo pharmacokinetic properties of the bispecific IgG were similar to those of the parental IgGs in rhesus macaques. Together, these data provide support for the clinical use of bispecific IgGs in the prophylaxis and treatment of HIV-1 infection.