Targeting the HIV-1 Spike and Coreceptor with Bi- and Trispecific Antibodies for Single-Component Broad Inhibition of Entry

Design of bi- and trispecific molecules targeting distinct HIV Env neutralizing determinants.

We selected four major neutralizing epitopes located at different regions of the HIV Env trimer for proof-of-principle design of bi- and trispecific antibody-based inhibitors in combinations or formats not fully tested previously. The well-described cross-conserved neutralizing sites themselves, from the base of the trimer to the V-region top cap, included the membrane-proximal external region (MPER), the CD4 binding site (CD4bs), the N332 glycan “supersite” (base of V3), and the V2 trimer apex (Fig. 1). We used the prototypic bNAb sequences derived from the equally well-described bNAbs 10E8 (MPER), N6 (CD4bs), PGT121 (N332 supersite), and PGDM1400 (V2 apex).

To begin, we designed a bispecific molecule to simultaneously target the CD4bs and the MPER on Env. We engineered sequences encoding the heavy (H) and light (L) chain variable regions of the CD4bs-directed bNAb N6 and the MPER-directed bNAb 10E8 in the form of scFvs, each containing a short H-to-L linker. Each scFv was connected by a second linker sequence, (G4S)₃, to generate a bispecific antibody-like molecule (Fig. 2, top panel). We selected these two bNAbs because of their outstanding neutralization breadth as IgGs (5, 6), anticipating broad coverage. Initial screening of the 10E8fv-N6fv bispecific against a 12-virus global indicator panel, which contained several highly resistant tier 3 isolates, resulted in neutralization of 10 of the 12 viruses tested (data not shown). Based upon this neutralization profile, we assessed the 10E8fv-N6fv bispecific molecule against a modified Seaman panel (49) which included 109 viruses, and it neutralized 100% of the viruses tested with reasonable potency (Fig. 2, bottom panel).

Bolstered by the neutralization breadth of this initial bispecific design, we shifted to a trispecific design to potentially increase potency by using a configuration more amenable to restoration of Fc-mediated effector function. Accordingly, we generated trispecific molecules designed as two chains, designated A and B, which interacted via their respective CH1 and CL domains, linking two scFvs on each respective A and B component. The scFvs were again connected with (G4S)₃ linkers (Fig. 3A). The initial design combined the antigen binding sites derived from the H and L chains of three bNAbs, namely, 10E8, PGT121, and PGDM1400. Again, these bNAbs were selected to potentially neutralize most known HIV isolates, based upon their individual neutralization profiles (6, 9, 50). We first designed 10E8 as the Fab, PGT121 as the scFv A arm, and PGDM1400 as the scFv B arm. To determine empirically if other configurations would enhance cross-neutralizing capacity in a head-to-head manner, other, “shuffled” combinations of the individual V region gene segments were designed, with the goal of maintaining trispecific recognition. Therefore, we did not generate combinations that would result in repetition of one of the three bNAbs in the three potential binding sites (Fab, scFv A arm, or scFv B arm). This process resulted in a total of eight different 10E8Fab-PGT121fv-PGDM1400fv trispecific molecules with the three bNAb-based specificities in different orientations, generating versions V1 to V8 (Fig. 3A, bottom section). We first confirmed the functional binding of the trispecifics to well-ordered native flexibly linked (NFL) Env trimers by enzyme-linked immunosorbent assay (ELISA). Shown in Fig. 3B is a representative example with 10E8Fab-PGT121fv-PGDM1400fv.V8 compared to the parental IgGs. The monovalent trispecific molecule bound to the trimer in a manner similar that of PGDM1400 IgG, which binds one quaternary epitope per trimer. Increased binding by PGT121 IgG, which has three binding sites per trimer, was observed as expected.

To better assess the functionality of these eight related 10E8Fab-PGT121fv-PGDM1400fv trispecifics, a panel of 28 relatively resistant clinical isolates was selected for analysis, along with the parental IgG controls. The viruses that we selected generally displayed resistance or susceptibility to each of the three bNAbs in assessments of independent neutralization capacity. For example, since the viral strain CAAN5342.A3 is highly sensitive to PGT121 but relatively resistant to 10E8 and PGDM1400, it can be used as an indicator of the PGT121fv functionality/specificity of the trispecific molecule. Similarly, 6540.v4.c1 can serve as an indicator of PGDM1400 activity, and BJOX010000.06.2 can be an indicator of 10E8 activity. Given the high potency of the parental bNAbs, we used a relatively low starting concentration (1 μg/ml) for this initial neutralization panel to better evaluate independent neutralization capacity and to rank order the different trispecific molecules. As shown in Fig. 3C, the parental bNAb neutralization patterns were generally consistent with previously reported specificities and potencies (9, 51). With this panel, we observed that if a bNAb potently neutralized a given virus (i.e., at concentrations of <0.009 μg/ml), then most 10E8Fab-PGT121fv-PGDM1400fv trispecific variants also neutralized such viruses. For example, T250-4 is neutralized potently by both PGT121 and PGDM1400, and we anticipated that most of the trispecifics would also neutralize this virus similarly, which was borne out by the data. Occasionally, a lower potency than that of the parental bNAbs was detected, especially if the parental bNAb(s) also neutralized the virus with low potency (e.g., ZM109) or if the neutralization was mostly driven by PGT121. Since the trispecific design now renders the binding monovalent, this was not entirely unexpected.

While the trispecifics were less potent than the parental IgGs in some cases, there was a marked improvement in breadth. The individual parental antibodies 10E8, PGT121, and PGDM1400 neutralized 43%, 61%, and 50% of the viruses of the 28-virus panel, respectively, in this assay format. When the three bNAbs were combined to a total IgG concentration of 1 μg/ml, 100% of the viruses tested were neutralized (Fig. 3C, last column), indicating that this combination of bNAbs in the proper trispecific configuration should display broad neutralization capacity in a single inhibitory molecule. Exhibiting additivity relative to the potencies of individual IgGs, the trispecific 10E8Fab-PGT121fv-PGDM1400fv versions V3 and V5 each neutralized 24 of 28 (86%) viruses in this resistant panel. Neutralization breadth was slightly diminished if both of the scFv VHs were at the C termini (V2 and V4). Version V8 neutralized 23 of 28 (82%) viruses tested, with a lower median 50% inhibitory concentration (IC₅₀) (0.081 μg/ml) than those of the other trispecific variants. We therefore selected 10E8Fab-PGT121fv-PGDM1400fv.V8 for screening against the larger, 109-virus panel of diverse clinical isolates. To better determine the breadth against the larger panel of viruses, we used the less stringent starting concentration of 40 μg/ml. The 10E8Fab-PGT121fv-PGDM1400fv.V8 trispecific was 99% effective against this panel, displaying markedly increased breadth and potency (IC₅₀ = 0.0135 μg/ml) compared to those of the parental bNAbs (Fig. 4).

Modified Env-targeting trispecific designs to restore Fc elements and to increase solubility.

Next, we evaluated whether we could further improve the potency of the 10E8Fab-PGT121fv-PGDM1400fv.V8 trispecific design. We constructed three variants with distinct modifications. First, we focused on restoring the Fc region for potential Fc receptor (FcR) interaction. Since the Fc region has been shown to enhance protective efficacy in vivo (52) and Fc-mediated interaction with the neonatal (n) FcRn contributes to a longer in vivo half-life (53), reconstitution of these effector elements will be valuable for future in vivo testing. Accordingly, we incorporated the CH2-CH3 domains of the IgG H chain between the Fab and the scFvs to reconstitute FcR interaction. The 10E8Fab-PGT121fv-PGDM1400fv.V8Fc trispecific retained a comparable neutralizing capacity and, in some cases, demonstrated enhanced neutralization potency against isolates not readily neutralized by the original V8 variant (Fig. 5A). For the second modification, we utilized a longer linker (LL), (G4S)₆ instead of (G4S)₃, between the 10E8 Fab component and the two scFvs. The trispecific molecule, 10E8Fab-PGT121fv-PGDM1400fv.V8LL, demonstrated equivalent or occasionally enhanced neutralization compared to that with the original V8 design against a 12-virus panel (Fig. 5B). Third, we incorporated structure-based substitutions that improve 10E8 IgG solubility (54), and potentially function, generating the 10E8Fab-PGT121fv-PGDM1400fv.V8.4DS trispecific. The 4DS variant showed improved neutralization potency compared to that with the original V8 design against a small virus test panel (Fig. 5C). Consistent with the neutralization data, both the V8LL and V8.4DS variants bound to well-ordered NFL trimers comparably as assessed by ELISA (Fig. 5D).

Design of a cMAb to target two sites on CCR5 with potentially enhanced avidity.

Besides directly targeting the virus, we reasoned that it might be valuable to improve the potency of a CCR5-directed antibody, PRO 140 (55), to potentially increase the avidity for the predominant HIV coreceptor. To do so, besides targeting the extracellular loop 2 (ECL2)-directed epitope recognized by PRO 140, we generated a bispecific IgG by combining the PRO 140 specificity with that of RoAb13, which recognizes the CCR5 Nt region (56) (Fig. 6A). The bispecific RoAb13-PRO 140 antibody was constructed based on the CrossMAb (cMAb) design. In this design, arms of the IgG were encoded using either RoAb13- or PRO 140-derived nucleotide sequences. We used the published “knob-in-hole” (KIH) approach, in which the antibody is engineered to possess mutated residues in the CH3 domain, allowing association of only the desired heavy and light chains via a complementary “knob” and “hole.” As described previously, a domain exchange of the CH1 and CL regions was introduced to ensure proper association of a given light chain with the correct heavy chain (57) (Fig. 6B; see Materials and Methods). Expression (and formation) of the cMAb was confirmed by SDS-PAGE analysis and compared to that of the parental IgGs (Fig. 6B, right panel).

To confirm independent binding of the humanized RoAb13 component to the CCR5 N terminus and of humanized PRO 140 to ECL2 independently, we performed a cross-competition assay using murine antibodies specific for these sites. Control RoAb13 (Nt) and PRO 140 were expressed as human IgG1 antibodies. The competing mouse antibodies specific to the same epitopes, 45523 (ECL2) and CTC8 (Nt), were obtained commercially (Fig. 6C). To establish a cross-competition assay, samples containing equivalent numbers of cells were first incubated with a human IgG (e.g., PRO 140 IgG), followed by incubation with the competing mouse antibody for the same site (e.g., 45523). A noncompeting mouse antibody was used as a negative control. Fluor-labeled secondary antibodies were used to detect binding of the primary antibodies as assessed by fluorescence-activated cell sorter (FACS) analysis. When we performed this assay using the cMAb, greatly diminished signals from the competing mouse antibodies were detected (Fig. 6D), indicating that the cMAb could bind to both epitopes effectively. Since we also detected comparable binding of both PRO 140 and CTC8 (or RoAb13 and 45523) in the same sample, these data suggest that the Nt and ECL2 epitopes are sufficiently far apart to not interfere with their respective ligands. Individually, PRO 140 and RoAb13 are quite broad, as they neutralized 97% and 92% of the viruses, with mean values of 0.323 μg/ml and 0.177 μg/ml, respectively. When we tested the PRO 140-RoAb13 cMAb, all viruses in the global panel were neutralized (data not shown). When neutralization was assessed against the larger virus panel, the cMAb neutralized 99% of the viruses, with a mean potency of 0.272 μg/ml (Fig. 7). In the rare case where both PRO 140 and RoAb13 could not neutralize a given virus, such as WEAU, the cMAb was also ineffective. As expected, if either one of the antibodies could effectively neutralize a given virus, then the cMAb could as well.

Bispecifics and trispecifics cross-targeting both HIV Env and the CCR5 coreceptor.

In our third overall approach, epitopes on HIV Env and CCR5 were simultaneously targeted using antibody binding regions designed as either bispecific or trispecific antibody-like molecules. We initially designed two bispecific molecules, with the rationale that their respective epitopes are likely to be proximal during receptor-coreceptor interaction during the initial stages of the HIV entry process. Accordingly, a bispecific molecule using PGDM1400 (V2 apex) and PRO 140 (CCR5) was constructed as tandem scFvs (PGDM1400fv-PRO 140fv) (Fig. 8). This bispecific displayed considerable breadth of neutralization and was more effective against clade C viruses. Next, we sought to incorporate a CD4-induced (CD4i) specificity to assess the impact on neutralization capacity. To do so, we selected, m36.4, a small, single-VH-domain inhibitor that was originally found by screening a phage library to identify small domains with capacities similar to that of the CD4i MAb X5 (58). By itself, m36.4 neutralized only 3 of 12 viruses in the global panel (data not shown), but we reasoned that colocalization of this CD4i-directed variable domain with the CCR5 coreceptor might be effective, since they are proximal during viral entry. Accordingly, we paired the m36.4 VH domain with the PRO 140 scFv to generate a bispecific molecule cross-targeting Env CD4i residues and the ECL2 region of CCR5. The m36.4-PRO 140fv bispecific neutralized 100% of the viruses, with an impressive median IC₅₀ of 0.0131 μg/ml (Fig. 8). Among the bispecific designs, m36.4-PRO 140fv was the most potent inhibitor, even after adjustment on a molar basis for the slightly lower molecular mass.

We also generated three trispecifics in this Env-CCR5 cotargeting class, revealing that the PRO 140Fab-PGDM1400fv-PGT121fv molecule neutralized 97% of viruses tested (Fig. 8). Since several of the Env apex-specific bNAbs contain sulfated tyrosines, we wanted to ensure maximal posttranslational sulfation by coexpression of the tyrosyl protein sulfotransferase enzyme (TPST-1). Sulfated Ys can enhance bNAb neutralizing potency (e.g., for PG9, PGT145, PGDM1400, and CAP256 [9, 53, 54, 59]), whereas the lack of Y sulfation has been reported to decrease potency (60). Cotransfection of a TPST-1-expressing plasmid did not greatly alter the neutralization capacity (Fig. 8), suggesting that there was adequate posttranslational modification of these trispecifics when they were expressed from the 293 producer cells. Another trispecific molecule in this cross-targeting series, 10E8Fab-PGDM1400fv-PRO 140fv, was the most potent trispecific, neutralizing 98% of the viruses with a relatively high potency (0.011 μg/ml) (Fig. 8). TPST-1 coexpression had a minimal but slightly enhanced effect on the potency (0.0071 μg/ml) of this trispecific. In addition, the trispecific RoAb13Fab-PGDM1400fv-PGT121fv combination neutralized 94% of the viruses tested, with moderate potencies (Fig. 8). Other trispecifics, using RoAb13Fab in combination with PRO 140fv-PGDM400fv or with PGT121fv-PRO 140fv, neutralized most of the viruses in the smaller global panel but were not exceptionally potent (data not shown).

IgG and bi- and trispecific construct cloning.

10E8, PGT121, PGDM1400, N6, PRO 140, and RoAb13 were cloned as human IgG1 antibodies in cytomegalovirus (CMV)-driven expression plasmids (66). Bispecific constructs were essentially two scFvs connected by a G4S linker sequence of two or three repeats. The initial eight trispecific design variations of 10E8Fab-PGT121fv-PGDM1400fv were generated by Creative Biolabs, and they all contained C-terminal His tags for ease of purification. Subsequently, we cloned all the constructs into the pCMV-1 expression vector, between the HindIII and EcoRI restriction sites. Trispecific molecules were of a Fab design. For both the N and C termini, the sequences for two scFvs were cloned in frame with sequences encoding connecting G4S linkers. Four “heavy” and four “light” constructs in multiple orientations were generated to express eight different trispecifics by transfecting one heavy and one light construct into 293F cells at a 1:1 plasmid DNA ratio. In the trispecific coding sequences, within the portion for the linker connecting the Fab part and the scFv, is a BamHI site which allows other bNAb scFvs to be exchanged. Variations of the original trispecific version 8 (V8) design included using the variable regions of the more soluble version of 10E8, named V8.4DS (54); using a longer linker (V8LL), with (G4S)₆ between the Fab and the scFv regions; and incorporating the Fc region of the IgG (V8Fc) between the Fab and scFv regions. Designs including the PGDM1400 scFv were also expressed with and without the addition of a plasmid encoding the enzyme TPST-1 to enhance posttranslational sulfation of tyrosines.

Expression and purification of bi- and trispecific molecules and IgGs.

The bi- and trispecific molecules were produced by transient transfection of expression plasmids into FreeStyle 293F suspension cells (Invitrogen) and purified from cell supernatants by use of a His tag purification resin (Roche), using chelation chemistry directed toward the C-terminal His tags. Following chelation on a nickel affinity column, the molecules were eluted and buffer exchanged into phosphate-buffered saline (PBS), pH 7.4, using either 30-kDa-cutoff Amicon Ultra-4 centrifugal filter units (Millipore) or Slide-A-Lyzer G2 dialysis cassettes with 20-kDa-cutoff membranes (Thermo Fisher Scientific). Parental IgGs were purified by passing supernatants collected after transient transfection through 1-ml affinity columns of protein A Sepharose beads (GE Healthcare). Each column was washed with PBS, and the column-bound IgG was eluted with 100 mM glycine-HCl buffer, pH 2.7, neutralized by addition of 2 M Tris-HCl, pH 8.5, and dialyzed extensively against PBS, pH 7.4.

cMAb design and cloning.

The anti-CCR5 bispecific cMAb utilizing the knob-in-hole (KIH) and light chain domain crossover formats was constructed as previously reported (57). Briefly, four plasmids were required to express the cMAb. The first-component plasmid encoded the PRO 140 heavy chain with the VH region followed by the light chain constant region, the hinge region, and CH2-CH3 (Fc portion). The CH3 domain on the PRO 140 side had the “hole” (T366S/L368A/Y407V) engineered into it. The second-component plasmid encoded the PRO 140 light chain with the VL region followed by the CH1 domain. The third component was the RoAb13 heavy chain, in which the “knob” (T366W) was engineered into its CH3 domain. The fourth plasmid encoded the RoAb13 light chain. The plasmids were transiently transfected in equal amounts into 293F cells, and the cMAb was purified from the supernatant by use of a protein A Sepharose affinity column as described above. Individual mutations required to generate cMAb component plasmids were created using a QuikChange II site-directed mutagenesis kit (Agilent). Plasmids that required domain crossovers were constructed by gene synthesis of fragments (GeneArt) and seamless cloning (Invitrogen) into expression vectors according to the manufacturer's instructions. Note that the PRO 140 antibody used clinically is an IgG4 to limit activation of complement and cell lysis, which is an additional modification that may be used for in vivo testing of the cMAb in preclinical models or the clinic.

Flow cytometric cross-competition assay.

Cf2Th cells overexpressing CCR5, obtained from the NIH AIDS repository, were used to show that both arms of the cMAb are active by binding to both the Nt and ECL2 epitopes on CCR5. The human antibodies RoAb13 (Nt) and PRO 140 (ECL2) and the cMAb (Nt and ECL2) were incubated on ice for 15 min, followed by fixation with 1% paraformaldehyde (PFA) and washing, and then the corresponding competing antibody or the noncompeting mouse antibody was added to different tubes to evaluate if the mouse antibody could bind the epitope potentially occupied by the corresponding human antibody. The human and mouse antibodies were detected with fluor-labeled secondary antibodies as assessed by FACS analysis. The mouse antibodies 45523 (ECL2) and CTC8 (Nt) were purchased from R&D Systems.

ELISA.

ELISA plates were coated with the trispecific molecules or IgGs in PBS, pH 7.4, at 4°C overnight. Plates were blocked with 5% nonfat dry milk in PBS-T (PBS with 0.2% Tween 20) at room temperature (RT) for 2 h. Wells were then incubated with serial dilutions of biotinylated JRFL NFL TD CC+ trimers at RT for 90 min, followed by washing in PBS-T and subsequent incubation with streptavidin-horseradish peroxidase (HRP) (Sigma) to detect binding of the trimer. Plates were washed in PBS-T and developed using 3,3′,5,5′-tetramethylbenzidine chromogenic substrate solution (Life Technologies).

Virus neutralization assay.

Virus neutralization was assessed with a single-cycle assay using huCD4⁺ CCR5⁺ TZM-bl target cells and HIV-1 pseudoviruses as described previously (49). To determine the antibody concentration that resulted in a 50% reduction in entry (IC₅₀; determined in relative luciferase units), which reflects neutralization of HIV, we performed serial dilutions of the parental antibodies or the bi- and trispecific molecules and fitted the neutralization dose-response curves by nonlinear regression. The sources of the Env-encoding plasmids derived from the HIV-1 isolates used in the neutralization assays were described previously (67). We used a modified Seaman panel containing 109 of the 117 viruses, as our clones for THRO4156.18 (virus 8 in the panel), Ce1086_B2 (virus 38), Ce0682_E4 (virus 42), Q259.d2.17 (virus 65), Q842.d12 (virus 66), 9004SS_A3_4 (virus 71), 231966.c02 (virus 105), and 191821_E6_1 (virus 106) did not express.