Structural basis of clade-specific HIV-1 neutralization by humanized anti-V3 monoclonal antibody KD-247

Abstract

Humanized monoclonal antibody KD-247 targets the Gly³¹²-Pro³¹³-Gly³¹⁴-Arg³¹⁵ arch of the third hypervariable (V3) loop of the HIV-1 surface glycoprotein. It potently neutralizes many HIV-1 clade B isolates, but not of other clades. To understand the molecular basis of this specificity, we solved a high-resolution (1.55 Å) crystal structure of the KD-247 antigen binding fragment and examined the potential interactions with various V3 loop targets. Unlike most antibodies, KD-247 appears to interact with its target primarily through light chain residues. Several of these interactions involve Arg³¹⁵ of the V3 loop. To evaluate the role of light chain residues in the recognition of the V3 loop, we generated 20 variants of KD-247 single-chain variable fragments with mutations in the antigen-binding site. Purified proteins were assessed for V3 loop binding using AlphaScreen technology and for HIV-1 neutralization. Our data revealed that recognition of the clade-specificity defining residue Arg³¹⁵ of the V3 loop is based on a network of interactions that involve Tyr^L32, Tyr^L92, and Asn^L27d that directly interact with Arg³¹⁵, thus elucidating the molecular interactions of KD-247 with its V3 loop target.—Kirby, K. A., Ong, Y. T., Hachiya, A., Laughlin, T. G., Chiang, L. A., Pan, Y., Moran, J. L., Marchand, B., Singh, K., Gallazzi, F., Quinn, T. P., Yoshimura, K., Murakami, T., Matsushita, S., Sarafianos, S. G. Structural basis of clade-specific HIV-1 neutralization by humanized anti-V3 monoclonal antibody KD-247.

Although the availability of highly active antiretroviral therapy (HAART) has significantly reduced the rate of HIV-1-related deaths (1), the development of a vaccine against multiple HIV-1 clades remains a challenge. The extensive glycosylation of the envelope glycoprotein (Env) (2) and the shedding of Env from the virus surface (3) are two of several mechanisms by which HIV-1 escapes the host immune system. Although most of the neutralizing antibodies elicited after HIV-1 infection are clade specific (4), some are broadly neutralizing, including b12 (5), 2G12 (6), 4E10 (7), 2F5 (8), PG9 and PG16 (9), VRC01 and VRC02 (10), PGT121-145 (11), NIH45-46 (12), and 10E8 (13).

HIV-1 Env comprises the noncovalently associated gp120 and gp41, which are proteolytic cleavage products of the gp160 precursor (14). The third hypervariable (V3) loop of gp120 that interacts with the CCR5 or CXCR4 coreceptor during HIV-1 entry (15) is one of the most immunodominant regions of HIV-1 (16–18). V3 loops of different HIV-1 strains and clades are highly variable in sequence and structural conformation (19). Anti-V3 antibodies elicited from animal immunization studies are generally clade-specific and show little or no cross-reactivity (19). Nevertheless, several anti-V3 antibodies, such as 447-52D, 2219, F425-B4e8, 2557, and 3074, have been reported to show cross-reactivity against a panel of HIV-1 isolates from various clades (20–25). The recent discovery of broadly neutralizing V3-targeting PGT121 and PGT128 antibodies continues to shed light on V3 loop immunogen design (26).

KD-247 is a humanized version of the C25 murine monoclonal antibody (mAb), which was isolated from the sequential immunization of mice with clade B HIV-1 V3 loop peptides (27). KD-247 can neutralize a broad spectrum of CCR5- and CXCR4-tropic viruses and HIV-1 quasi-species from patient plasma and peripheral blood mononuclear cells (27, 28). Passive transfer of KD-247 in simian/human immunodeficiency virus–infected monkeys protected the animals against CD4⁺ T cell loss and increase of virus load (29, 30). This suggests that KD-247 can serve as a potential immunotherapy component in treating HIV-1-infected patients. Therefore, a structural understanding of the KD-247–V3 interactions should help us design strategies for expanding the clade specificity of this antibody.

The minimum V3 sequence required for KD-247 binding was mapped to Ile³⁰⁹-Gly³¹²-Pro³¹³-Gly³¹⁴-Arg³¹⁵ (IGPGR) at the V3 arch, and Arg³¹⁵ is crucial for the interaction with KD-247 (27). We hypothesized that the interactions of KD-247 with residue 315 of the V3 loop strongly affect the clade specificity of KD-247, which can efficiently neutralize clade B viruses with Arg³¹⁵ at the V3 arch, but not viruses from other clades that have a Gln³¹⁵ (e.g., Gly³¹²-Pro³¹³-Gly³¹⁴-Gln³¹⁵ in most non–clade B viruses).

To understand the molecular basis of KD-247 clade specificity, we have solved the crystal structure of its unliganded antigen binding fragment (Fab) and used it in molecular modeling studies with V3 peptides to obtain insights into possible binding interactions between the Fab and the target V3 loop. The proposed interactions were validated by site-specific mutagenesis of single-chain variable fragment (scFv) KD-247 variants, peptide binding assays, and cell-based HIV-1 neutralization assays.

MATERIALS AND METHODS

Fab production and purification

KD-247 was obtained from the Chemo-Sero-Therapeutic Research Institute (27). Fab was prepared by digesting KD-247 (34°C, 7 h) with 0.2 mg of papain agarose (Sigma-Aldrich, St. Louis, MO, USA) per milligram of antibody at 2 mg/ml in sodium acetate pH 5.5, 50 mM l-cysteine and 1 mM EDTA. The reaction was stopped by removing the papain agarose using a 0.22 µm filter. Digested Fab was purified with a HiTrap SP HP 5 ml column (GE Healthcare, Piscataway, NJ, USA) using sodium acetate pH 5.5 as the binding buffer and sodium acetate pH 5.5, 1 M NaCl, as the elution buffer.

Crystallization and data collection

KD-247 Fab crystals were grown in sitting drop trays. Drops containing 1 μl Fab (10 mg/ml) and 1 μl well solution were allowed to equilibrate with 0.5 ml 1.9 M ammonium sulfate/0.05 M sodium acetate pH 4.4 at 21°C. Octahedral-shaped crystals appeared after 3 d and cryoprotected with 20% glycerol. Data were processed to 1.55 Å using d*TREK (31), and indexed in P2₁2₁2₁ (a = 61.1 Å, b = 69.2 Å, and c = 111.8 Å) with one Fab per asymmetric unit. The Matthews coefficient (32) was 2.5 ų/Da (solvent content ∼51%).

Structure determination and refinement

The structure was determined by molecular replacement MOLREP (33). The Fab variable and constant domains of 1T3F from the Protein Data Bank (PDB) were treated as separate search models. After initial rigid-body and restrained refinement in Phenix (34), R_work dropped to 0.3377, with an R_free of 0.3560. Simulated annealing was used to remove model bias. An initial model was built using ARP/wARP (35) with refinement using Refmac (36). Several cycles of model building and refinement were carried out using Coot (37) and Phenix (Table 1). Final atomic coordinates and structure factors have been deposited (PDB ID: 3NTC).

TABLE 1.

Data collection and refinement statistics

Data collection	Value
Wavelength (Å)	1.07
Resolution (Å)	1.55 (1.61–1.55)a
Space group	P212121
Cell dimensions
a (Å)	61.1
b (Å)	69.2
c (Å)	111.8
Observed reflections	448,730
Unique reflections	68,709
Redundancy	6.5 (4.4)
Completeness (%)	99.0 (92.0)
Rsymb	0.068 (0.595)
Avg I/σ	10.2 (1.6)
Refinement statistics for all reflections >0.0 σ F
Resolution (Å)	19.75–1.55
No. of reflections (working)	68,481
No. of reflections (test)	2,748
Rworkc	0.1901
Rfreed	0.2099
No. of Fab atoms	3365
No. of water molecules	575
No. of solvent molecules	30
Overall B value (Å2)
Fab	26.28
Solvents	39.54
Wilson B value (Å2)	24.16
Ramachandran plot (%)e
Favored	98.2
Allowed	1.8
Disallowed	0.0
RMSD bond length (Å)	0.004
RMSD angle (°)	0.992

^aValues in parentheses are for the outer resolution shell. ^bR_sym = Σ_hkl |I – < I >| / Σ_hkl |I|. ^cR_cryst = Σ_hkl |F_obs – F_calc| / Σ_hkl |F_obs|. ^dR_free = R_cryst, except 4% of the data excluded from the refinement. ^eEvaluated by MolProbity (48).

Superposition analysis

The coordinates of several Fab–V3 peptide complexes were downloaded from the PDB: 1ACY, 1AI1, 1F58, 1GGI, 1NAK, 1Q1J, 2B0S, 2QSC, and 3MLW. These complexes were chosen specifically because all have V3 peptides based on the HIV-1_MN sequence, which is efficiently neutralized by KD-247. The Fab portions were aligned with the KD-247 Fab in Coot using the light chain for alignment. Upon each alignment, the position of the V3 peptide with respect to the KD-247 complementarity determining region (CDR) was visually inspected. The V3 peptide that fit best in the KD-247 binding pocket was from 2QSC (RP142 V3). The V3 peptide from the aligned 2QSC coordinates was removed and loaded with KD-247 into SYBYL (7.3.5; Tripos, St. Louis, MO, USA) and taken through a slight minimization procedure to reduce minor steric interactions.

Modeling of G314E and R315K KD-247-resistant V3 peptides with KD-247

Models of the G314E and R315K V3 peptides were generated by performing a simple mutation of the aligned and minimized RP142 peptide used in the superposition analysis at the 314 and 315 positions. All possible rotamers of Glu³¹⁴ and Lys³¹⁵ demonstrated steric clashes with KD-247 CDR residues.

Preparation of KD-247 scFv variants

Single amino acid substitutions of Asn^L27d, Tyr^L32, and Tyr^L92 in the background pET28a3c-KD247 scFv (38) were generated by site-directed mutagenesis and verified by DNA sequencing. scFv variants were expressed in BL21(DE3) Escherichia coli and purified as previously described (38). scFv in the inclusion bodies was denatured and refolded before purification on HisTrap and HiPrep 26/60 Sephacryl S200 HR columns (GE Healthcare, Piscataway, NJ, USA). The secondary structure of the refolded scFv was examined using far-UV circular dichroism (CD) spectroscopy as previously described (38). Data were collected on a J-815 CD Spectrometer (JASCO, Easton, MD, USA) at 0.2 mg/ml from 190 to 240 nm. CD spectra were plotted using GraphPad Prism 5 (GraphPad Software Inc., La Jolla, CA, USA). CD spectra were analyzed by the SELCON3 (39, 40) and K2D3 (41) methods using the DichroWeb online analysis software (42, 43). Reference data sets 4 and 7 (range 190–240 nm) were used in the analyses (44).

V3 peptide binding assay

Biotinylated clade B V3 peptide was synthesized at the Structural Biology Core (University of Missouri, Columbia, MO, USA). The sequence (Biotin-GCRKRIHIGPGRAFYTC) was derived from MN V3 loop sequence (304–309, 312–319 based on HXB2 numbering). Peptide binding assays were performed in 96-well ½ area white plates (Perkin Elmer, Waltham, MA, USA) using a 40 μl reaction volume containing donor and acceptor beads (final concentration 20 μg/ml), 1× phosphate-buffered saline (PBS) pH 7.4, 0.01% Tween-20, and 0.1 mg/ml bovine serum albumin. A total of 200 nM N-terminal 6× histidine (His₆)-tagged scFv variants (50 nM final) were incubated with 400 nM biotinylated V3 peptide (100 nM final) for 1 h at room temperature. A total of 80 μg/ml of nickel chelated acceptor beads (Perkin Elmer, Waltham, MA, USA) were added to the wells and allowed to incubate for another hour in the dark before adding 80 μg/ml of streptavidin-coated donor beads (Perkin Elmer, Waltham, MA, USA). After 30 min incubation in the dark, the plates were analyzed using an EnSpire Plate Reader (Perkin Elmer, Waltham, MA, USA). Parental KD-247 scFv (wild-type, WT) was the positive control that gives high signal counts because of its strong binding to the clade B V3 peptide. A reaction containing only the peptide but no scFv served as the negative control. The signals of all the scFv variants were compared to the WT scFv signal. Statistical analyses were performed using 2-tailed 1-sample t test and Wilcoxon signed rank test at 95% confidence.

HIV-1 neutralization assay

Maraviroc, TZM-bl cells (from Dr. John C. Kappes, Dr. Xiaoyun Wu, and Tranzyme Inc., Durham, NC, USA), pSG3ΔEnv (from Drs. John C. Kappes and Xiaoyun Wu), pWT/BaL plasmid (Dr. Bryan Cullen), and H9/HTLV-III_MN NIH 1984 (Dr. Robert Gallo) were obtained through the U.S. NIH AIDS Reagent Program. The plasmid for expression of JR-FL Env (pCXN-JR-FL-Env) was from Dr. Shuzo Matsushita. TZM-bl cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% heat-inactivated fetal bovine serum (FBS), 100 units/ml penicillin, and 100 µg/ml streptomycin. The 293T cells were maintained in DMEM supplemented with 10% FBS, and penicillin/streptomycin. H9/HTLV-III_MN NIH 1984 was passaged in RPMI 1640 supplemented with l-glutamine and 10% FBS.

Replication-competent HIV-1_BaL and replication-deficient pseudotyped HIV-1 were produced by 293T transfection with pWT/BaL or pSG3ΔEnv and pCXN-JR-FL-Env plasmids (38). Replication-competent HIV-1_MN was obtained from supernatant of H9/HTLV-III_MN NIH 1984 culture (45, 46). Virus titers were determined using the previously described 50% tissue culture infectious dose (TCID₅₀) assay (47).The ability of KD-247 scFv variants to neutralize JR-FL Env pseudotyped HIV-1 was first evaluated. A total of 100 TCID₅₀ pseudotyped virus was preincubated with scFv variants (5 µM final concentration) (37°C) for 1 h before infecting TZM-bl cells (preseeded 1 × 10⁴ cells/well). A total of 10 nM maraviroc (targets CCR5) was used as a positive control. Luciferase activity of infected TZM-bl cells was determined at 48 h postinfection using Bright-Glo Reagent (Promega, Madison, WI, USA). The relative infectivity was determined as ratio of the relative light units in the presence of scFv to virus control (PBS treated). The percentage of neutralization was calculated as 100 × (1 − relative infectivity). scFv variants that showed more than 50% neutralization were further studied as described above using HIV-1_BaL or HIV-1_MN in the presence of scFv at various concentrations to determine 50% neutralization concentration (EC₅₀). Data from at least three independent experiments were plotted using nonlinear regression equations in GraphPad Prism 5 software to obtain EC₅₀ values.

RESULTS

KD-247 Fab crystal structure

The KD-247 Fab structure was determined to 1.55 Å, the highest resolution reported for any apo humanized antibody (Fig. 1A). It was refined to an R-factor of 19% and an R_free of 21%. More than 98% of the KD-247 Fab residues had main chain torsion angles in the energetically favored regions of the Ramachandran plot (48), with no residues in disallowed regions (49) (Table 1). Using RBOW, the elbow angle of the KD-247 Fab was determined to be 127°, which falls within the range commonly observed for Fabs with κ light chains (50).

Figure 1.

Crystal structure of KD-247 Fab. (A) Stereo view of the Fab including CDR regions (L1, cyan; L2, green; L3, blue; H1, red; H2, orange; H3, purple). Light and heavy chains are shown in light and dark gray. (B) 3Fo–2Fc electron density map of light chain residues (σ = 2.0). All structural images were generated using PyMOL (http://www.pymol.org/).

KD-247 is the first humanized anti-V3 mAb to be structurally characterized. The heavy and light chains were numbered using the Kabat numbering system (51, 52). The heavy and light chains are designated with an “H” and an “L”, respectively. The CDR loops L1, L2, L3, H1, and H2 belong to canonical classes κ-3, 1, κ-1, 1, and 1 (53). The CDR H3 loop exhibits a kinked base conformation with the observation of the conserved hydrogen bond between the ring nitrogen of Trp^H103 and the carbonyl oxygen of Met^H100A; there is no salt bridge between Arg^H94 and Asp^H101 (54). Excellent electron density was continuous almost throughout the entire molecule (Fig. 1B), allowing assignment of partial occupancy and alternate conformations. Slightly disordered loops were observed for residues 42–43, 64–66, and 101–102 in the heavy chain but were included in the model.

The KD-247 Fab CDR L1 loop is very similar to the L1 loop from other Fabs, including the Fab of nonneutralizing HIV-1 antibody 13H11 (PDB ID: 3MNV; light chain RMSD compared to KD-247 light chain [RMSD_L] determined by PDBeFold = 0.62 Å), the Fab of a mAb that neutralizes human rhinovirus serotype 2 (PDB ID: 1BBD; RMSD_L = 0.73 Å), an antitumor CH2-domain-deleted humanized antibody (PDB ID: 1ZA6; RMSD_L = 0.55 Å), the Fab of human germ-line antibody 1-69/B3 (PDB ID: 3QOT; RMSD_L = 0.44 Å), and the Fab of Mus musculus germ-line antibody S25-2 (PDB ID: 1Q9K; RMSD_L = 0.65 Å) (Fig. 2A). The CDR L1 loop is considerably longer in the KD-247 Fab (17 residues long) than what is observed for other anti-V3 loop antibodies, including murine antibody 83.1 (PDB ID: 1NAK) and human antibodies F425-B4e8 (2QSC), 2219 (2B0S), and 1006-15D (3MLW) (Fig. 2B). The CDR L1 loop of anti-V3 loop murine antibody 83.1 (PDB ID: 1NAK) is 16 residues long but is bent in the crystal structure to avoid crystal packing clashes with residues from the variable heavy chain of a symmetry-related molecule (55).

Figure 2.

Structural comparison of KD-247 with other Fabs and potential interactions with a V3 peptide. (A) Similarities of the KD-247 CDR L1 loop (cyan) to other Fabs (PDB ID: 3MNV, red; 1BBD, orange; 1ZA6, green; 3QOT, magenta; 1Q9K, blue). The RMSD between the light chain of all Fabs and KD-247 is <0.75 Å. (B) The KD-247 CDR L1 loop (cyan) is unique from other anti-V3 mAbs (PDB ID: 1NAK, yellow; 2QSC, gray; 2B0S, pink; 3MLW, brown). (C) Stereo view of KD-247 (surface representation, colors as in Fig. 1) with the RP142 peptide (white cartoon/yellow sticks) modeled in the binding pocket. (D) Interactions between Arg³¹⁵ of the RP142 peptide, Tyr^L32 and Asn^L27d of CDR L1 (cyan) and Tyr^L92 of CDR L3 (blue; dashed lines are H bonds). Tyr^L32 is stabilized by an H-bond network that includes Asp^L28 and Asn^L27d of CDR L1.

We pursued cocrystallization of the KD-247 Fab in complex with various V3 peptides of different sequences and lengths, which resulted in many beautiful crystals but poor X-ray diffraction or internal lattice problems that could not be resolved. The crystal packing of the apo Fab does not allow room for the peptide to bind through soaking experiments. Thus, we constructed a molecular model of the KD-247 Fab in complex with the V3 loop in order to understand the molecular details of HIV-1 clade B V3 loop recognition by KD-247.

Superposition of the V3 loop

In our model of the KD-247 Fab crystal structure in complex with a clade B V3 Gly³¹²-Pro³¹³-Gly³¹⁴-Arg³¹⁵ (GPGR)-containing peptide, the V3 residues are designated with a “P” and numbered according to the HXB2 sequence (56). We aligned previously determined Fab-V3 complexes to KD-247 Fab. The V3 peptides of many Fab-V3 complexes that aligned with the KD-247 Fab did not fit well after superposition with the KD-247 binding pocket. One exception was the V3 peptide from 2QSC, derived from MN and called RP142 (Y^P301NKRKRIHI^P309 G^P312PGRAFYTTKNIIGC^P326) (57, 58).

The RP142 peptide fits nicely into the binding pocket, with only minor obstructions. The side chains of Ile^P309 and Tyr^P318 and the main chain of Phe^P317 in V3 were in close contact (<2.0 Å) with the Trp^H33, Asn^H58, and Tyr^L94 side chains of KD-247 Fab. After minimization, the positions of the important GPGR residues were virtually unchanged. The primary interactions between the RP142 peptide and the KD-247 Fab involve the V3 arch of the loop and the CDR L1 and L3 regions (Fig. 2C). Previous studies have shown that the characteristic observed interactions between anti-V3 Fabs with the V3 peptides occur in the long extended CDR H3 (59–61). In our case and in the 2QSC structure, Arg³¹⁵ of the V3 loop interacts with light chain Fab residues. The most interesting interaction occurs between Arg³¹⁵ of the RP142 peptide and KD-247 Tyr^L32 and Tyr^L92 (Fig. 2D). Arg³¹⁵ appears to be stabilized by van der Waals interactions with the side chains of the two tyrosines and additionally by a hydrogen bond with the phenoxy group of Tyr^L92. The arginine is also further stabilized by a hydrogen bond with Asn^L27d. An intricate hydrogen bond network between CDR L1 loop residues Tyr^L32, Asp^L28, and Asn^L27d helps to stabilize Tyr^L32 and Asn^L27d in positions to interact with Arg³¹⁵ (Fig. 2D).

G314E and R315K KD-247-resistant V3 peptides with KD-247

Two mutations in the V3 arch confer KD-247 resistance. HIV-1_JR-FL with G314E was neutralized 16-fold less efficiently by KD-247 than HIV-1_JR-FL without this mutation (62). An HIV-1_BaL variant containing a potential N-linked glycosylation site (PNGS) insertion in V2 and an R315K mutation in the V3 arch provided a very high resistance phenotype to KD-247 (63). The long Glu chain in G314E V3 causes steric clashes with KD-247 CDR L1, L2, and H3 (data not shown). Similarly, R315K also creates steric clashes with CDR L1 and L3 of KD-247.

KD-247 scFv variants

On the basis of our KD-247 structure-guided model, we designed KD-247 variants in a smaller size 30 kDa scFv antibody format. scFv variants with single or double Ala substitutions at Asn^L27d, Tyr^L32, and/or Tyr^L92 were generated to disrupt the proposed interactions. To understand the proposed interactions of Tyr^L32 and Tyr^L92 with Arg³¹⁵, we generated scFv variants with Phe substitutions at these positions. Additional substitutions at Asn^L27d, Tyr^L32, and Tyr^L92 were generated to further understand the effects of side chains on the KD-247 interactions with the GPGR V3 arch. We proposed that substitution with polar side chain of Asn or Gln at Asn^L27d, Tyr^L32, and Tyr^L92 will maintain affinity for Arg³¹⁵. Substitution with the long basic side chain of Arg and Lys or the short acidic side chain of Asp and Glu at Asn^L27d, Tyr^L32, and Tyr^L92 of KD-247 may provide additional insights into the interactions with the GPGR V3 arch.

All scFv variants were expressed and purified by a stepwise refolding process (38) (Supplemental Fig. 1A) that overcomes the previously reported challenge of obtaining anti-V3 loop scFvs (64). Despite the low recovery of refolded monomeric scFvs, we were able to obtain sufficient quantities for the subsequent analyses. A correctly folded scFv should retain an immunoglobulin-like structure with each variable domain consisting of nine antiparallel β-sheets (65). We previously used CD spectroscopy to confirm efficient refolding of WT KD-247 scFv (38). Here, we used far-UV CD spectroscopy to analyze the folding of mutant KD-247 scFv variants (Supplemental Fig. 1B–E). Protein folding can be assessed from the CD ellipticity value (y axis intercept): unordered protein structure increases with decreasing ellipticity value at short wavelength (200 nm) (66). Overall, the scFvs were better folded in the presence of a 16mer HIV-1_MN-derived V3 peptide (Supplemental Fig. 1B–E; cf. right vs. left panels). Folding was further confirmed by analyzing the CD data using the SELCON3 (39, 40) and K2D3 (41) methods in the DichroWeb online analysis software (data not shown) (42, 43). Generally, the quantification of the CD data by SELCON3 and K2D3 confirm that the secondary structure of the Asn^L27d scFv variants had a comparable β-sheet content to that of WT scFv (Supplemental Fig. 1B). Of the Tyr^L92 variants, results from the K2D3 method demonstrated that Phe^L92, Gln^L92, and Arg^L92 showed the most comparable β-sheet content to WT scFv, which is again consistent with the CD spectra (Supplemental Fig. 1B). The results from the SELCON3 and K2D3 methods showed that Tyr^L32 variants and the double mutants Y32A/Y92A and N27dA/Y32A exhibited lower β-sheet contents compared with WT scFv, which demonstrates poor folding in agreement with the CD spectra (Supplemental Fig. 1D, E). This observation indicated that the Tyr^L32 aromatic ring is required for proper conformation of KD-247 when purified in vitro. We selected scFv variants that showed a proper β-sheet profile for further investigation of the KD-247 scFv-V3 loop interactions.

scFv–V3 loop interactions

To compare binding of KD-247 scFv variants to GPGR V3 loop peptides, we performed a protein–protein interaction assay using the AlphaScreen technology (67). The streptavidin-coated donor beads and the nickel-chelate acceptor beads bind to the biotinylated V3 peptides and the purified His₆-tagged scFvs, respectively. A favorable interaction between the V3 loop peptide and the scFv variant brings the 2 types of fluorophore-coated beads into proximity, and an amplified light signal is generated upon excitation (Fig. 3A). Our results showed that similar to WT scFv, several scFv variants, including N27dD, N27dE, N27dK, N27dR, Y92A, Y92F, Y92R, and Y32F, interact with the clade B V3 loop peptide, and the signals measured were significantly different from the no-scFv negative control (Fig. 3B).

Figure 3.

Interactions of scFv variants with V3 loop. (A) Schematic representation of KD-247 scFv interactions with a V3 peptide using AlphaScreen technology. His-tagged scFvs were allowed to interact with biotinylated V3 peptide. The interaction was detected using nickel (Ni²⁺)-chelated AlphaScreen acceptor beads and streptavidin (SA)-coated AlphaScreen donor beads. (B) Interaction of scFv variants with cyclic clade B V3 peptide. Results are expressed as the means of signal-to-background ratio from 3 independent experiments. Error bars indicate sem. Statistically significant differences compared to WT scFv are represented by an asterisk (*P < 0.05). (C) Neutralization assay of HIV-1 Env pseudotyped virus on TZM-bl cells. A total of 10 nM maraviroc (CCR5 antagonist) or 5 μM KD-247 Fab or His-tagged scFv variants were preincubated with HIV-1_JR-FL Env pseudotyped HIV-1 before infection of TZM-bl. Luminescence was measured 48 h after infection to determine scFv variants that result in >50% neutralization of pseudotyped virus. Results are represented as the average of percentage neutralization relative to virus control from at least three experiments; error bars indicate sem.

In addition to measuring the binding interactions of scFv with cyclic V3 loop peptides, we also determined their ability to neutralize the infectivity of pseudotyped HIV-1 (HIV-1_JR-FL, Table 2) that contains a V3 loop similar to the one used in the modeling studies (HIV-1_MN-derived RP142 peptide). Only WT KD-247 Fab and scFv, as well as N27dD and Y92F scFvs, could neutralize by at least 80% pseudotyped clade B HIV-1 using 5 μM scFv (Fig. 3C). Other scFv variants, which initially showed binding to the clade B GPGR V3 loop peptide (Fig. 3B), neutralized less than 30% of pseudotyped HIV-1 under the same conditions (Fig. 3C).

TABLE 2.

V3 loop sequences of clade B HIV-1 strains

HIV-1 strain	Tropism	V3 loop sequence
MN	X4	CTRPNYNKRKRIHIGPGRAFYTTKNIKGTIRQAHC
BaL	R5	-----N-T--S--------L---GE-I-D------
JR-FL	R5	-----N-T--S------------GE-I-D------

Dashes indicate identical amino acid residue.

We also determined the EC₅₀ of the most potent scFvs in TZM-bl neutralization assays using fully infectious CCR5-tropic HIV-1_BaL and CXCR4-tropic HIV-1_MN (Table 3). Similarly, the refolded WT scFv neutralized the replication-competent HIV-1_MN and HIV-1_BaL at comparable EC₅₀ values (0.7 and 0.6 µM, respectively). Y92F scFv showed similar EC₅₀ values compared to WT scFv (0.8 µM for HIV-1_MN and 0.5 µM for HIV-1_BaL), suggesting that the aromatic interaction with this residue is crucial. A slight increase in the EC₅₀ of N27dD scFv (2.1 µM for HIV-1_MN and 1.2 µM for HIV-1_BaL), indicated that Asn^L27d is better at stabilizing the interaction with Arg³¹⁵. These results provided insights into the interactions of Arg³¹⁵ with side chains at the interface of the KD-247 light chain.

TABLE 3.

Fifty percent neutralization concentration (EC₅₀) (µM) against clade B HIV-1

Assay	HIV-1MN(CXCR4-tropic)	HIV-1BaL(CCR5-tropic)
Maraviroc	>0.1	0.002 ± 0.001
KD-247 Fab	0.2 ± 0.06	0.1 ± 0.02
KD-247 scFv	0.7 ± 0.2	0.6 ± 0.1
N27dD scFv	2.1 ± 0.7	1.2 ± 0.2
Y92F scFv	0.8 ± 0.08	0.5 ± 0.09

Data represent the mean ± sd from the results of three independent experiments.

DISCUSSION

The V3 loop of HIV-1 gp120 is flexible and structurally diverse, as evidenced by the wide range of different conformations of Arg³¹⁵ observed in gp120 and Fab/V3 complexes (aligned GPGR arches shown in Fig. 4A). This diverse loop, and specifically Arg³¹⁵, which correlates with genotypic specificity, is also recognized in a variety of ways by anti-V3 mAbs. Although most mAbs interact with their respective antigen through heavy chain interactions (68), several anti-V3 mAbs interact with Arg³¹⁵ through residues of their light chains. For example, murine Fab 83.1 interacts with Arg³¹⁵ through hydrogen bond between the main and side chains of Thr^L91 (55) (Fig. 4B). Human Fab 2219 recognizes Arg³¹⁵ through a hydrogen bond with Asn^L31 (69) (Fig. 4C). In the human F425-B4e8 Fab/V3 complex, Arg³¹⁵ is sandwiched between residues Tyr^L32 and Asp^L92 and is stabilized by a salt bridge with Asp^L92, as well as hydrogen bonds with the main and side chains of Asp^L92 (58) (Fig. 4D). Additionally, Arg³¹⁵ interacts with the human 1006-15D Fab through hydrogen bonds with the side chains of Asn^L30 and Asp^L93 (60) (Fig. 4E). KD-247 also uses its light chain to recognize Arg³¹⁵, but in what appears to be a different manner than other anti-V3 mAbs: in addition to hydrogen bond interactions with Tyr^L92 and van der Waals interactions with Tyr^L32 and Tyr^L92, Asn^L27d also forms a hydrogen bond with the Arg³¹⁵ side chain, providing additional stability. Asn^L27d and Tyr^L32 are held in place by an elaborate hydrogen bond network also involving Asp^L28 (Fig. 2D). These distinct interactions involve the unique CDR L1 insertion of KD-247 and provide a plausible answer for why Arg³¹⁵ is needed for HIV-1 neutralization by KD-247.

Figure 4.

Structural variability of Arg³¹⁵ and recognition by anti-V3 mAb light chains. (A) Stereo view of “GPGR”-containing V3 loops in the following: 58.2 Fab complexes (PDB ID: 1F58, red; 2F58, green; 3F58, magenta), 83.1 Fab complex (1NAK, yellow); 447-52D (1Q1J, orange); 2219 Fab complex (2B0S, pink); F425-B4e8 Fab complex (2QSC, dark gray); 268-D Fab complex (3GO1, gray); 1006-15D Fab complex (3MLW, brown); 3074 Fab complex (3MLX, blue); 2557 Fab complexes (3MLR, light green; 3MLT, cyan); 2558 Fab complex (3UJI, mauve); 537-10D Fab complex (3GHE, hot pink); the V3 loop from the X5 Ab complex with JR-FL gp120 and CD4 (2B4C, dark green); and the V3 loop from the sulfated-tyrosine 412d Ab complex with YU2 gp120 and CD4 (2QAD, light blue). Arrow indicates the space occupied by Arg³¹⁵ conformations in the various structures. Arg³¹⁵ is also stabilized by interactions with light chain residues in some mAbs, including 83.1 (B), 2219 (C), F425-B4e8 (D), and 1006-15D (E). H bonds and salt bridges are shown as black and red dashed lines.

The interactions between mAbs and the V3 loop are also affected by the use of different CDRs. For example, the human mAb 537-10D recognizes the same IGPGR epitope as KD-247, yet it uses a long insertion in the CDR H3 loop to make nonspecific interactions with the V3 loop in the form of a 4-stranded antiparallel β-sheet (Fig. 5A), in comparison to KD-247, which primarily uses CDR L1 for V3 loop interactions (Fig. 5B, C). Although they recognize the same epitope, 537-10D demonstrates a narrow HIV-1 neutralization profile compared to KD-247 (70). This is likely due in part to the more restricted antigen binding site of 537-10D, in which the Trp^H33, Glu^H95, and Tyr^H100J residues that interact with Arg³¹⁵ of the V3 loop are buried in a deep pocket (∼6Å) that requires a close fit to bind the V3 loop. The CDR region of KD-247 demonstrates a shallow binding pocket that may be able to better tolerate flexibility of the V3 arch (59).

Figure 5.

The 537-10D and KD-247 target the same IGPGR epitope using different modes of V3 binding. (A) 537-10D Fab (heavy and light chains in dark and light gray) in complex with an MN V3 peptide (yellow, PDB ID: 3GHE). The primary binding interactions occur in the CDR H3 loop (purple). (B) KD-247 Fab bound to the RP142 V3 peptide (same colors as in (A)). Most of the binding interactions occur in the CDR L1 loop (cyan). (C) Sequence alignment of the KD-247 and 537-10D variable regions (CDR regions in red). Identical residues are highlighted in blue and similar residues in yellow. Asterisks mark the residues making interactions with Arg³¹⁵ of the RP142 V3 peptide. Sequence alignment was performed by the Sequence Manipulation Suite (http://www.bioinformatics.org/sms2/index.html). CDRs were labeled on the basis of previously published sequences (27, 59).

Resistance mutations at gp120 affect interactions between mAb and V3 loop and help HIV-1 escape neutralizing antibodies. Two mutations in the V3 arch region of gp120 cause resistance to KD-247. KD-247 neutralizes HIV-1_JR-FL with a G314E mutation ∼20-fold less efficiently than WT (62), and also binds 100× weaker than WT to HIV-1_BaL with a V2 PNGS insertion in addition to the V3 R315K mutation (63). Our modeling studies suggest that mutation from Gly to Glu at 314 and from Arg to Lys at 315 creates steric interactions with the binding pocket of KD-247 (data not shown). These steric contacts likely affect KD-247 binding to V3 loops containing these mutations and may explain the reported differences in KD-247 binding and HIV-1 neutralization.

Our proposed interactions were validated by the generated scFv variants and the results of our binding and neutralization assays. The data suggest that KD-247 uses an elaborate network of interactions that are based on the long insertion in CDR L1 and involve residues Asn^L27d, Asp^L28, Tyr^L32, and Tyr^L92 (Fig. 2D). Importantly, one of these residues (Tyr^L32) appears to be important for proper scFv folding, as almost all of the mutants at this position were inactive in neutralization assays (Fig. 3C) and poorly folded (CD spectra in Supplemental Fig. 1D). Phe^L32 was the only mutant at position 32 that demonstrated proper scFv folding. Although variant Phe^L32 demonstrated some binding activity to the V3 loop peptide (Fig. 3B), it did not neutralize pseudotyped clade B HIV-1 (Fig. 3C). Hence, an aromatic residue at position 32 in the light chain is important for proper scFv folding but not necessarily for neutralization, where it seems that the phenoxy group is critical. Changes at position 27d do not affect scFv folding (Supplemental Fig. 1B) and appear to either have limited effect (with the exception of Ala^L27d and Gln^L27d) or even enhance binding to the V3 loop (Fig. 3B). The loss of V3 loop binding and neutralization ability of the Ala^L27d KD-247 scFv variant is consistent with the proposed role of Asn^L27 in interacting with Arg³¹⁵. However, the only variant that was able to efficiently neutralize pseudotyped clade B HIV-1 was Asp^L27d (Fig. 3C). It is likely that the Glu^L27d and Gln^L27d variants are not as efficient in neutralization of pseudotyped virus because of their longer side chains (as compared to Asp^L27d and Asn^L27d). Similarly, the even longer side chains of Arg^L27d and Lys^L27d are likely to negatively affect interactions with Arg³¹⁵ because of steric constraints. The ability of the N27dD variant to neutralize clade B HIV-1 suggests that KD-247 may be substituted with Asp^L27d to retain activity against clade B HIV. This substitution would still be able to participate in the intricate network of hydrogen bond interactions proposed in our model for recognition of Arg³¹⁵. Although many of the scFv variants at position 92 of the light chain demonstrated proper folding (Supplemental Fig. 1C), Phe^L92 demonstrated enhanced binding to the clade B V3 loop peptide while Ala^L92 and Arg^L92 had similar binding compared to the WT scFv (Fig. 3B), but only Phe^L92 showed effective neutralization of clade B HIV-1 Env pseudotyped virus (Fig. 3C). These results establish that the aromatic ring at Tyr^L92 is essential for clade B V3 loop recognition. Overall, these observations highlight the importance of Asn^L27d, Tyr^L32, and Tyr^L92 in neutralizing clade B HIV-1 containing a GPGR V3 loop arch. Importantly, these results reveal that binding of scFv variants to clade B V3 loop peptides does not necessarily correlate with efficient HIV-1 clade B neutralization. It is possible that other factors may affect binding of the antibody in the context of the full Env glycoprotein that do not factor in to V3 peptide binding. The recent crystal structure of the native Env trimer by Julien et al. (71) revealed that the gp120 subunits are stabilized by β-hairpin interactions of the V3 loop and the V1/V2 strands B and C near the top of the trimer. The arch of the V3 loop is hidden by an N-acetyl glucosamine from the Asn¹⁹⁷ glycan at the C-terminal region of V2 strand D from a neighboring protomer. This glycan blocks access to the V3 arch and may affect mAb binding.

Our findings are in agreement with previous work highlighting the dramatic flexibility of V3 loop and underscore the challenges in engineering antibodies that will be useful for treatment and vaccine design. Antibody engineering based on the interactions reported in our and other structures may lead to enhanced interactions with both Arg and Gln residues at position 315 of the V3 loop. For example, we expect that mutations of KD-247 residue Asn^L27d to longer polar residues (N27dQ, N27dK, and N27dR) should maintain the hydrogen bond interaction with clade B Arg³¹⁵ and also be able to hydrogen bond with the shorter non–clade B Gln³¹⁵. Additionally, mutations of KD-247 residue Asp^L28 to long polar residues (D28K, D28R, and D28E) should reach and interact with both clade B Arg³¹⁵ and non–clade B Gln³¹⁵. Similarly, the Y32R mutant should maintain van der Waals interactions with clade B Arg³¹⁵ and also make hydrogen bond interactions with non–clade B Gln³¹⁵. Combinations of the above mutations in the 27d, 28, 32, and 92 positions of the light chain of KD-247 may also help to maintain interactions with Arg³¹⁵ and improve interactions with Gln³¹⁵. Additional approaches may include mutations of other KD-247 residues that do not interact directly with residue 315 of the V3 loop. Such interactions are observed in the structures of the 2557 and 3074 Fabs (22, 60) in complex with the V3 loop. Finally, extended CDR H3 loops may be designed to make nonspecific main chain interactions with the V3 loop, as is the case with 447-52D (61) and 537-10D (59), which form three-stranded and four-stranded antiparallel β-sheets with the V3 loop target.

Although the propensity of V3 loop to serve as an immunogen to elicit broadly neutralizing antibodies against HIV-1 of multiple clades has been challenging, the extensive studies performed by the Zolla-Pazner, Gorny, Wilson, and Kong groups suggest that this strategy may be achievable (22, 25, 55, 58–61, 69, 70, 72).

Glossary

CD

circular dichroism

CDR

complementarity determining region

Env

envelope glycoprotein

Fab

antigen-binding fragment

GPGR

Gly³¹²-Pro³¹³-Gly³¹⁴-Arg³¹⁵

mAb

monoclonal antibody

scFv

single-chain variable fragment

V3 loop

third hypervariable loop