Structural basis for diverse N-glycan recognition by HIV-1–neutralizing V1–V2–directed antibody PG16

Abstract

HIV-1 uses a diverse N-linked-glycan shield to evade recognition by antibody. Select human antibodies, such as the clonally related PG9 and PG16, recognize glycopeptide epitopes in the HIV-1 V1–V2 region and penetrate this shield, but their ability to accommodate diverse glycans is unclear. Here we report the structure of antibody PG16 bound to a scaffolded V1–V2, showing an epitope comprising both high mannose–type and complex-type N-linked glycans. We combined structure, NMR and mutagenesis analyses to characterize glycan recognition by PG9 and PG16. Three PG16-specific residues, arginine, serine and histidine (RSH), were critical for binding sialic acid on complex-type glycans, and introduction of these residues into PG9 produced a chimeric antibody with enhanced HIV-1 neutralization. Although HIV-1–glycan diversity facilitates evasion, antibody somatic diversity can overcome this and can provide clues to guide the design of modified antibodies with enhanced neutralization.

The HIV-1 envelope spike (Env) is the primary target of HIV-1–neutralizing antibodies and is heavily glycosylated, especially in its gp120 component, with N-linked glycans contributing approximately half the spike mass and covering most of the spike surface (reviewed in refs. 1,2). Despite the prominent coverage of Env surface by N-linked glycan, sera and antibodies from HIV-1–infected individuals generally show minimal glycan-dependent reactivity^3–5. The low frequency of glycan-reactive antibodies has been attributed to issues of cross-reactivity in antibody recognition of N-linked glycan on HIV-1 Env and of N-linked glycan on host or self proteins. Indeed, the antigenic structure of HIV-1 gp120 displays a ‘silent face’ that corresponds to a dense cluster of N-linked glycans^6,7, which is infrequently recognized by the host immune system.

The 2G12 antibody⁸, which recognizes a cluster of high mannose–type glycans on HIV-1 gp120 (refs. 9,10), provided an early notable exception to this general lack of N-glycan reactivity^3,11, and in recent years, a number of other N-glycan–reactive HIV-1–neutralizing antibodies have been isolated from the sera of HIV-1–infected donors^12,13. Characterization of these antibodies is ongoing, but all appear to recognize either an array of N-linked glycans in a multivalent manner (2G12)^9,10,14–17 or a combination of N-linked glycan and envelope polypeptide (PG9, PGT128)^18,19 (Supplementary Table 1). Such multicomponent recognition provides a means to reduce the affinity of antibody for individual N-linked glycans to a tolerable level, thereby overcoming issues related to self-reactivity^17,18,20.

A common theme with many of these glycan-reactive antibodies is a requirement for high mannose–type N-linked glycans. Characterization of monomeric HIV-1 gp120 indicated substantial glycan diversity^21–23, with complex-type N-linked glycans present at one-third to one-half of the N-linked sites on gp120. The high density of glycan on the assembled viral spike, however, appears to inhibit glycan processing, and high mannose–type N-linked glycans predominate^24–29. The percentage of high mannose–type glycans on functional viral spikes appears to depend on several factors including host cell and viral strain^24,25,30, but a substantial diversity of high-mannose types as well as complex types may be present^24,31. Further, this diversity may have a role in viral infectivity^32,33, cell-mediated viral transmission³⁴, regulation of spike conformation³¹ and immune evasion^7,35,36.

Does glycan variation, such as that between high mannose–type and complex-type glycans, allow for HIV-1 escape from the newly identified glycan-reactive antibodies? Or do these antibodies have mechanisms to cope with glycan diversity? Recent analysis of PGT121 indicated an ability to recognize complex-type N-linked glycans³⁷, but the absence of a PGT121–gp120 structure has made it difficult to understand the context of this recognition. To address these questions, we extended our characterization of broadly neutralizing antibodies that target the V1–V2 region of gp120 and require a high mannose-type N-linked glycan at residue 160_gp120 for HIV-1 neutralization¹³. (For clarity, we add the macromolecule as a subscript when referring to specific residues.) This category of broadly neutralizing antibodies includes three sets of somatically related antibodies: PG9 and PG16 from donor IAVI 24, PGT141–145 from donor IAVI 84 and CH01–CH04 from donor CHAVI 0219. These individually neutralize 70–80%¹³, 40–80%³⁸ and 40–50%⁵, respectively, of circulating HIV-1 isolates. An even higher level of breadth is achieved when somatic variants are combined: for example, the combined neutralization of PG9 and PG16 reaches ∼90% of circulating HIV-1 isolates¹⁸. Among these V1–V2–directed antibodies, the structure of PG9 in complex with the V1–V2 domain of gp120 was solved and revealed cooperative recognition by PG9 of strand C of V1–V2 and two N-linked glycans attached at residue 160_gp120 (N-glycan 160) and either residue 156_gp120 (in most HIV-1 strains) or residue 173_gp120 (in specific strains such as ZM109) (N-glycan 156 or 173)¹⁸. The recognition of N-glycan 160 appeared to be specific to a subset of high mannose–type glycans (such as those with five mannose and two N-acetylglucosamine residues (Man₅GlcNAc₂))¹³, but the recognition of N-glycan 156 or 173 did not appear to be necessarily constrained to high-mannose types.

Here we report the crystal structure of the antigen-binding fragment (Fab) of the PG16 antibody in complex with the V1–V2 domain of HIV-1 gp120 from strain ZM109 along with NMR characterization of glycan recognition by PG16, PG9 and various mutated forms in which we specifically knocked out antibody recognition for either N-glycan 160 or N-glycan 156 or 173. We created a PG9-PG16 chimera with enhanced recognition of complex-type N-linked glycans that retained the clean autoreactivity profile of the parent PG9 and PG16 antibodies. Overall, these analyses describe glycan binding specificities and atomic-level details of the PG16 epitope and reveal how HIV-1 envelope-spike variability can be collectively overcome by somatic mechanisms of clonal antibody diversification.

Results

Structure of PG16 in complex withV1–V2 of HIV-1 gp120

Our interest in diverse glycan recognition by antibody PG16 was spurred by the observation that, when expressed in HEK293F or in HEK293S GnTI⁻ cells, scaffolded V1–V2s that bound well to PG9 generally showed weak (HEK293F) or no binding (GnTI⁻) to PG16 (Supplementary Fig. 1a). However, when expressed in HEK293F cells treated before transfection with the α-mannosidase II inhibitor swainsonine (which is predicted to induce hybrid-type glycans with a Man₅GlcNAc₂ core), improved binding of scaffolded V1–V2s to PG16 could be observed (Supplementary Fig. 1a)³⁹. To understand how PG16 antibody recognizes HIV-1 gp 120, we determined the structure of PG16 bound to the scaffolded V1–V2 produced in HEK293F cells treated with swainsonine. We obtained the PG16–V1–V2 complex, following procedures that had succeeded with PG16's somatic relative PG9 (ref 18). In particular, immobilization of PG16 IgG onto Protein A–agarose resin allowed us to select appropriately folded scaffolded V1–V2s and then perform on-column protease cleavage of IgG to Fab¹⁸ to obtain homogeneous complexes of PG16 bound to scaffolded V1–V2s (Supplementary Fig. 1b,c). We note that the yield of scaffolded V1–V2 bound to PG16 Fab was low (∼2%), and this probably reflects selection of scaffolded V1–V2 containing suitable glycosylation for PG16 recognition. First, we found that lectins that can discriminate between various oligomannosides, and complex-type glycans containing α2–3 or α2–6 sialic acid were able to bind the unselected scaffolded V1–V2s. This indicates that multiple glycoforms are present before PG16 selection (Supplementary Fig. 2a,b and Supplementary Note). Second, MS analyses and glycopeptide mapping of the scaffolded V1–V2 prior to purification with PG16 showed diverse N-linked glycans: 12 glycoforms at Asn160_gp120 including 20% Man₅GlcNAc₂ and 30% hybrid-type glycans; and 13 glycoforms at Asn173_gp120 including 56% hybrid glycan (Supplementary Fig. 2c–e).

To increase our chances of crystallization, we tested several crystallization variants, including one comprising the 1FD6 scaffolded V1–V2 region from the ZM109 isolate, in which N-glycosylation sites at positions 160 and 173 were preserved, but all of the four other potential N-glycosylation sites on V1–V2 were mutated to alanine. We obtained cocrystals of Fab PG16 in complex with 1FD6 scaffolded ZM109 V1–V2 that diffracted to 2.5-Å resolution (Table 1).

Table 1. Data collection and refinement statistics.

	1FD6ZM109–PG16
Data collection
Space group	C2221
Cell dimensions
a, b, c (Å)	81.1, 207.6, 87.6
Resolution (Å)	50–2.43 (2.52–2.43)a
Rsym	17.4 (48.6)
I / σI	5.3 (1.35)
Completeness (%)	91.0 (61.4)
Redundancy	5.5 (3.2)
Refinement
Resolution (Å)	45.1–2.44
Unique reflections	25,294
Rwork / Rfree	0.201 / 0.229
No. atoms	8,630
Protein	8,094
Ligand/ion	388
Water	148
B factors (Å2)
Protein	65.2
Ligand/ion	72.1
Water	43.9
r.m.s. deviations
Bond lengths (Å)	0.002
Bond angles (°)	0.723

^aValues in parentheses are for highest-resolution shell. The data set was collected from a single crystal.

Whereas the structure of PG16 bound to the scaffolded V1–V2 domain of HIV-1 strain ZM109 resembled the previously reported PG9 complex of the same scaffold¹⁸, the presence of a hybrid-type glycan at Asn173_gp120 of ZM109 V1–V2 was newly observed (Fig. 1). As with PG9, the protruding complementarity-determining region (CDR) H3 of PG16 reached through the glycan shield to contact both glycan and polypeptide, and the V1–V2 domain formed a four-stranded β-sheet displaying a Greek-key motif possessing two disulfide linkages between strands A and B and between A and D (Fig. 1). Superposition of scaffolded V1–V2s complexed to PG9 or PG16 showed a Cα r.m.s. deviation of 1.3 Å, and variable domains’ superpositions between PG9 and PG16 resulted in a Cα r.m.s. deviation of 0.6 Å.

Figure 1.

Cocrystal structure of human antibody PG16 and scaffolded-V1–V2 of HIV-1 gp120 with a high mannose–type N-linked glycan at residue 160_gp120 and a hybrid-type N-linked glycan at residue 173_gp120. The Fab of antibody PG16 (tan for heavy chain and light blue for light chain) along with V1–V2 (green) and 1FD6 scaffold (gray) are shown in ribbon representation. N-linked glycans (N) at residues 160 and 173 of V1–V2 are displayed in stick representation, with oxygen atoms in red and carbon atoms colored according to saccharide-subunit standard coloring (blue, N-acetylglucosamine; green, mannose; yellow, galactose; magenta, sialic acid). The hybrid-type glycan at residue 173 is highlighted with a partially transparent surface colored the same as underlying atoms.

PG16–V1–V2 interactions

Several types of interactions arbitrate PG16 recognition of HIV-1 V1–V2. These can be classified into sequence-independent, electrostatic and glycan-mediated interactions (Fig. 2). Sequence-independent interactions are prominent in the PG16–V1–V2 complex, and the protruding CDR H3 of PG16 aligns its most distal β-strand (residues 100E_HC to 100H_HC in Kabat nomenclature⁴⁰, with HC and LC subscripts for heavy chain and light chain, respectively) in parallel orientation with strand C of V1–V2 to form an intermolecular β-sheet with four hydrogen bonds (Fig. 2b). These sequence-independent interactions are similar to those observed previously in the PG9–V1–V2 complex¹⁸. Electrostatic interactions were also observed between PG16 and V1–V2, most prominently a double salt bridge between Arg168_gp120 and both the side chain of Asp100L_HC and the sulfate of the tyrosine-sulfated 100H_HC (Tys100H_HC). A salt bridge between Asp100I_HC and Lys171_gp120 was also observed (Fig. 2c). Electrostatic interactions between PG16 and gp120 were less extensive than interactions between PG9 and gp120. In the PG9–V1–V2 complex, two additional electrostatic interactions, between Tys100G_HC and Lys169_gp120 and between Tyr100K_HC and Asn173_gp120, were observed; with PG16, Tyr100G_HC is not sulfated, and residue 100K_HC is an asparagine.

Figure 2.

PG16 and PG9 interactions with V1–V2 domain of HIV-1 gp120. Cocrystal structures of PG16 and PG9 bound to V1–V2 allow the relative strength of interactions to be compared. (a) PG16 and PG9 CDR H3 sequences based on Kabat numbering. Sequence differences are shown in cyan; sulfated tyrosines are underlined. (b,c) Sequence-independent, electrostatic main chain (b) and side chain (c) interactions of PG16 (left) and PG9 (right) with V1–V2 domain from strain ZM109. Ribbon representation of V1–V2 (green) and PG16 (tan) or PG9 (yellow) CDR H3s. Cyan, sequence-variable region of CDR H3s, as in a. (d,e) Sequence-independent glycan-mediated Asn160_gp120 glycan (Man₅GlcNAc₂) (d) and Asn173_gp120 glycan (complex-type) (e) interactions of PG16 (left) and PG9 (right) with V1–V2 domain from strain ZM109. Ribbon representation of V1–V2 (green), PG16-heavy chain (tan), PG16-light chain (light blue), PG9-heavy chain (yellow) and PG9-light chain (blue). Complex-type glycan and Man₅GlcNAc₂ are displayed as sticks and are color coded as in Figure 1. Antibody residues involved in hydrogen bonds to glycans are shown in sticks. Saccharide subunits are colored as in Figure 1. Intermolecular hydrogen bonds and electrostatic interactions are displayed by black (<3.5 Å) or gray (>3.5 Å). The number of hydrogen bonds (<3.5 Å) and the buried surface area for each interface are shown.

The most prominent interactions occur between PG16 and N- glycans 160 and 173 (Fig. 2d,e). Extensive interactions between PG16 and an N-glycan present at Asn160_gp120 are observed. From the visualized glycan electron density (Supplementary Fig. 2f), the N-glycan 160 appears to be Man₅GlcNAc₂, though it is formally possible that it could be a glycan that shares the Man₅GlcNAc₂ substructure but is disordered at residues outside the Man₅GlcNAc₂ core. Interactions between this glycan and PG16 bury a total surface area of approximately 1,100 Ų and form nine intermolecular hydrogen bonds (<3.5 Å; Supplementary Table 2). Eight of these nine hydrogen bonds are formed between the terminal mannose moieties and the side chains of His100R_HC, Asn100P_HC, Asp31_LC, Ser32_LC and Asp50_LC. One hydrogen bond is present between the reducing N-acetyl glucosamine of Man₅GlcNAc₂ and the side chain of His100B_HC in the PG16 CDR H3 (Fig. 2d). Comparison of the interactions of PG16 and PG9 with high-mannose glycans at position 160 showed that the major differences lie in two residues at positions 100_HC and 100B_HC in the CDR H3 (Supplementary Table 2). In PG9, Asp100_HC and Arg100B_HC contribute one and three hydrogen bonds, respectively, with the protein-proximal N-acetyl glucosamine unit of Man₅GlcNAc₂ 160_gp120, whereas Ile100_HC and His100B_HC in PG16 contribute zero and one hydrogen bond, respectively (Fig. 2d and Supplementary Table 2).

PG16 also interacts extensively with an N-linked hybrid-type glycan present at Asn173_gp120, burying a total surface area of 800 Ų at the antibody-glycan interface and contributing six hydrogen bonds (<3.5 Å; Supplementary Table 2). In particular, interactions between PG16 and the terminal sialic acid moiety of the hybrid-type glycan account for more than half of the buried surface area at the interface and contribute all the intermolecular hydrogen bonds. PG16 Arg94_LC and His95A_LC make three hydrogen bonds with the terminal sialic acid, and Lys57_HC and His59_HC contribute three. Arg94_LC also has one hydrogen bond (3.9 Å) with the preceding galactose moiety of the glycan, which also has some stacking interactions with Trp64_HC (Fig. 2e and Supplementary Table 2). This confirmed our previous paratope analysis of PG16 that revealed that mutation of Trp64_HC to arginine affected neutralization¹⁸. Unlike Asn160_gp120-linked Man₅GlcNAc₂, in which six of seven saccharide rings contribute to the buried surface of the interface, the hybrid-type glycan at Asn173_gp120 makes contacts with the antibody through only four of the total ten saccharide rings. Of note, in the structure of PG9 bound to scaffolded V1–V2 ZM109, density could only be seen for the protein-proximal N-acetylglucosamine at position 173, with only one hydrogen bond with PG9. However, the structure of PG9 bound to scaffolded V1–V2 CAP45 showed multiple interactions with Man₅GlcNAc₂ at Asn156_gp120, the equivalent spatial position to Asn173_gp120. We also observed binding of PG9 to scaffolded V1–V2 grown in HEK293F cells in the presence of swainsonine (Supplementary Fig. 1a), and this indicates that PG9 may also bind hybrid-type glycan present in some sequences. The PG16 heavy-chain residues that interact with the sialic acid are equivalent in PG9 with His59_HC and Trp64_HC, both obtained by somatic mutation^13,41. However, the light-chain residues that interact with sialic acid and galactose in PG16 differ in PG9 (Thr94_LC and Arg95A_LC in PG16 compared to Arg94_LC and His95A_lc in PG9). These appear to preclude ideal interactions with the hybrid-type glycan, and this suggests that affinity maturation of a few light-chain residues may allow for recognition of hybrid-type and complex-type glycans (Fig. 3c).

Figure 3.

PG16 interactions with hybrid-type N-linked glycan at residue 156_gp120 or 173_gp120. (a) Paratope-epitope interface highlighting N-glycan 156_gp120 or 173_gp120. PG16 is shown as a gray surface with the N-glycan 156_gp120 or 173_gp120 glycan-contact surface colored light blue. The contribution of each component of V1–V2 to the total buried surface is indicated. Saccharide subunits are colored as in Figure 1. NAG, N-acetylglucosamine; Man, mannose; Gal, galactose; Sia, sialic acid. (b) Ligplot showing hydrogen bonds (<3.5 Å) and hydrophobic interactions of hybrid-type glycan with PG16 heavy (tan) and light (light blue) chains. (c) Schematics of hybrid-type glycan showing the contribution of each saccharide ring to the PG16-glycan interfacial buried surface and the number of intermolecular hydrogen bonds (<3.5 Å) made by each ring. (d) Top, complex-type carbohydrate shown schematically with each pyranose ring labeled A–G and sialic acid labeled as Sia. Bottom, STD-NMR (STD) and reference-NMR (Ref) spectra of 2 mM complex-type carbohydrate in the presence of 20 μM Fab PG16; # denotes residual Tris; asterisk (in H6′G*) indicates overlapping signals with H5D and H5F. The STD enhancement for H6′G was estimated to be equivalent to its nonoverlapped geminal ¹H H6G.

It is notable that we only observed an α2–6–linked sialic acid on the hybrid-type glycan in the crystallized structure. Studies have shown that glycoproteins produced in HEK293 cells contain primarily α2–6 sialic acids, although α2–3 sialic acids are also present^42–45. To confirm the presence of the sialic acid linkage observed in the crystal structure, we treated the scaffold 1FD6 V1–V2 ZM109 A4 with various neuraminidases that have different specificities for α2–3 and α2–6 sialic acids and evaluated binding to PG16 and PG9 by enzyme-linked immunosorbent assay (ELISA; Supplementary Fig. 2b and Supplementary Note). When scaffolded V1–V2 was treated with neuraminidase or sialidase A (which preferentially cleaves α2–6 sialic acids over α2–3), binding of PG16 was no longer observed, but when treated with α2–3 neuraminidase (which is highly selective for α2–3 sialic acids), binding to PG16 was retained. PG9 was less affected by neuraminidase treatment, although a preference of α2–6 sialosides was observed. These results indicate that an α2–6 sialic acid in the complex glycan is needed for PG16 binding.

NMR analysis of PG16 and PG9 recognition of glycan

To determine glycan recognition and binding-site selectivity for PG16 and PG9, we examined glycan binding by each antibody and by four antibody mutants designed to specifically knock out glycan recognition at individual sites. These included a PG16 W64_HCR mutant that binds only N-glycan 160 (PG16₁₆₀), a PG16 N100P_HCR mutant that binds only N-glycan 173 (PG16₁₇₃), a PG9 S55_HCR mutant that binds only N-glycan 160 (PG9₁₆₀) and a PG9 N100P_HCR mutant that binds only N-glycan 173 (PG9₁₇₃). By using saturation transfer difference (STD) NMR, we screened these antibodies for qualitative binding to commercially available glycans and relevant glycan fragments (Supplementary Fig. 3a), including biantennary complex type (a surrogate for hybrid type) and Man₅GlcNAc₂, and when binding was sufficiently strong we measured binding affinities. The NMR spectra showed that PG16 bound only two of the glycans tested, namely biantennary complex type (K_d = 1.26 ± 0.48 mM) and Man₅GlcNAc₂ (Fig. 3d). (We observed binding of Man₅GlcNAc₂, but the signal intensities were too weak for us to measure the K_d. This is probably because of the absence of hydrogen bonds that are present in PG9, especially at 100_HC and 100B_HC where aspartate-to-isoleucine and arginine-to-histidine variations occur in PG9 and PG16, respectively.) The biantennary complex–type glycan contains two terminal branches that are identical to the complex-type branch present in the hybrid-type glycan that makes most of the contacts with PG16, as shown in the crystal structure (Fig. 3). Hybrid-type glycans are not commercially available and therefore were not tested for binding. The STD spectrum of biantennary complex–type glycan in the presence of PG16 showed very selective STD enhancements for protons of sialic acid and galactose; this suggests proximity of these residues to PG16 in the bound state (Fig. 3). These data are in agreement with the crystal structure, which showed that the major interactions between hybrid-type glycan and PG16 are mediated through the terminal sialic acid and preceding galactose ring. To confirm the role of sialic acid, we tested binding of PG16 to an asialocomplex-type glycan lacking the terminal sialic acid by STD NMR under identical conditions. In addition to the absence of the sialic acid signals, the signal intensities over the entire difference spectrum were markedly lower than those for the biantennary complex–type glycan. This indicates substantially weaker binding of PG16 to the asialocomplex–type glycan (Supplementary Fig. 3b, overlay of two STDs). Previously we reported that PG9 binds Man₅GlcNAc₂ with a K_d = 1.6 ± 0.91 mM by STD NMR¹⁸, and although we observed binding of PG9 to complex-type glycans by NMR, the affinity was too weak to be measured.

We next analyzed binding of mutant PG9 and PG16 antibodies to Man₅GlcNAc₂ and biantennary complex–type glycans by using STD NMR. These mutant antibodies could bind only one site, N-glycan 160 or N-glycan 173. We used Man₉GlcNAc₂ as a control and did not observe binding for any of the antibodies. PG16₁₇₃ (mutated to bind N-glycan 173 only) bound complex-type glycan with a K_d of 1.19 + 0.45 mM, whereas we did not observe binding to Man₅GlcNAc₂; PG16₁₆₀ (mutated to bind only N-glycan 160) bound negligibly to Man₅GlcNAc₂ as well as to biantennary complex-type glycan (Fig. 4 and Supplementary Fig. 3c). PG9160 (mutated to bind only N-glycan 160) bound Man₅GlcNAc₂ with a K_d = 1.16 + 0.40 mM and did not show binding to biantennary complex-type glycan; PG9₁₇₃ (mutated to bind only N-glycan 173) showed very weak binding to biantennary complex-type glycan and to Man₅GlcNAc₂. However, the STD signal intensities were too weak to integrate accurately (Fig. 4 and Supplementary Fig. 3d). Nonetheless, qualitative comparison of the two STD spectra obtained under identical conditions showed that PG9₁₇₃ binds to biantennary complex-type glycan better than to Man₅GlcNAc₂ at this site (Supplementary Fig. 3d). PG16, therefore, binds preferentially to a complex-type glycan at Asn173gp₁₂₀. Unexpectedly, binding of PG16 to Man₅GlcNAc₂ at Asn160gp₁₂0 was negligible in the NMR spectra even though Man₅GlcNAc₂ linked to Asn160 of V1-V2 contributes seven intermolecular hydrogen bonds and over 1,100 Ų of buried surface area at the PG16-V1-V2 interface. N-glycan 160 is also critical for neutralization¹³. The weaker affinity of PG16 for Man₅GlcNAc₂ at Asn160_gp120 compared to that of PG9 can be attributed to the lack of sufficient interactions between PG16 and the protein-proximal saccharide moiety of the glycan. The large buried surface area at the interface between Man₅GlcNAc₂ at Asn160_gp120 and PG16, as revealed in the crystal structure, however, suggests that PG16 recognizes Man₅GlcNAc₂ with a reasonable affinity only in the glycan-plus-peptide context. PG9, in contrast, binds preferentially to Man₅GlcNAc₂ at Asn160_gp120 but binds very weakly to complex-type glycan at Asn173_gp120.

Figure 4.

PG16 binds complex-type glycans more tightly than Man₅GlcNAc₂, whereas PG9 prefers Man₅GlcNAc₂. (a,b) STD NMR analysis of glycan affinities. Complex type–glycan affinity for PG16₁₇₃, a PG16 mutant (Asn100P_HCArg) that binds only glycan 173_gp120 (a) and Man₅ affinity for PG9₁₆₀, a PG9 mutant (Ser55_HCArg) that binds only glycan 160_gp120 (b) are shown. (Fig. 8 shows wild-type affinities.) STD enhancements (0.5-s saturation time) as a function of carbohydrate concentration are shown as a series of stacked plots. K_d values were obtained from the initial slope of the best-fitted Langmuir binding curves (insets) for the STD amplification factor, A_STD, as a function of glycan concentration by keeping antibody concentration constant (20 μM). Asterisk denotes region integrated to obtain A_STD values by the equation A_STD = (I₀ – I_SAT) I₀ – 1 ([L_t]/[P]), where L_t and P are the total ligand and protein concentrations, respectively, and I₀ and I_SAT are the integrals of signals in the reference and difference spectra, respectively.

A chimeric antibody, PG9-PG16-RSH

The structure revealed that recognition of complex-type N-linked glycan by PG16 involves three residues in the light chain (Arg94_LC, Ser95_LC and His95A_lc), two of which were obtained through somatic hypermutation from germ line (from Ser94_LC and Thr95A_LC; Figs. 2 and 5). In PG9, the equivalent residues are Thr94_LC, Arg95_LC and Arg95A_LC. We created a chimeric antibody (PG9-16-RSH) that combines PG9's recognition of the V1–V2 side chain and the high-mannose N-glycan 160 with PG16's recognition of hybrid-type or complex-type glycans by substituting Arg94, Ser95 and His95A from the PG16 light chain into the light chain of PG9. Subsequently, the binding affinities of the chimeric antibody PG9-16-RSH with Man₅GlcNAc₂ or biantennary complex–type glycan were determined through single ligand–titration STD NMR experiments. The K_d values obtained were 1.57 ± 0.41 mM and 1.34 ± 0.35 mM for Man₅GlcNAc₂ and biantennary complex–type glycan, respectively (Fig. 5 and Supplementary Fig. 3e). When compared to the affinity of PG9 for Man₅GlcNAc₂ (1.60 ± 0.91 mM) and the affinity of PG16 for biantennary complex–type glycan (1.26 ± 0.48 mM), these results suggest that the chimeric antibody preserves the glycan specificity of the individual antibodies when their corresponding recognition elements are combined in a single antibody.

Figure 5.

Design and glycan recognition of PG9-16-RSH antibody. (a) Structure-based design. Ribbon representation of light chains interacting with hybrid-type N-glycan 173 are shown for initial germline recombinant of PG9-PG16 (IGLV2-14*01) (dark green and cyan, upper left), mature PG16 (light blue, lower left), mature PG9 (yellow, upper right) and PG9-16-RSH (yellow and light blue, lower right). Key light-chain residues at residues 94_LC, 95_LC and 95A_LC are labeled and colored cyan in the germline, yellow in the mature PG9 and light blue in the mature PG16, where they are arginine, serine and histidine (RSH), respectively. Saccharide subunits are colored as in Figure 1, and modeled glycan interactions are shown in semitransparent sticks. (b) Glycan recognition. STD-NMR (STD) and reference-NMR (Ref) spectra of 2 mM complex-type carbohydrate (right) and Man₅GlcNAc₂ (left) in the presence of 20 μM PG9-16-RSH antibody. K_d values (obtained as shown in Supplementary Fig. 3e from the initial slope of the best-fitted Langmuir binding curves for A_STD as a function of glycan concentration) are shown.

To evaluate whether the improved glycan binding by the chimeric antibody would lead to improved HIV-1 neutralization properties, we performed neutralization assays in triplicate with 21 different HIV-1 strains. We observed a roughly three-fold improvement in neutralization for the chimeric antibody PG9-16-RSH compared to PG9 (50% inhibitory concentration, IC₅₀; Fig. 6a). However, we observed strain-to-strain variation: PG9-16-RSH showed improved potency compared to PG9 on 17 viruses, with a median fold improvement of 4.6; no improvement on two viruses; and decreased potency on two viruses. Notably, the more substantial effects were observed with strains that were already neutralized potently by PG9 (Fig. 6b and Supplementary Table 3): for the eight strains with PG9 IC₅₀ < 0.05, the median improvement in IC₅₀ for PG9-16-RSH as compared to PG9 was five-fold; three of the four strains with PG9 IC₅₀ > 5 showed a modest decrease in potency. Overall, PG9-16-RSH showed no increase in breadth relative to PG9. We also tested neutralization by PG9-16-RSH, PG9 and PG16 in a large panel (195 viruses described previously)¹⁸ (Supplementary Table 3) and generated breadth/potency curves (Fig. 6c). We generated an optimal PG9 and PG16 curve by taking the lower IC₅₀ between PG9 and PG16 for each HIV-1 isolate (MIN(PG9-IC₅₀,PG16-IC₅₀)) and comparing it to the PG9-16-RSH breadth/potency curve (Fig. 6c). Notably, the PG9-16-RSH curve and the (MIN(PG9-IC₅₀,PG16-IC₅₀)) curve almost completely superposed (Fig. 6c).

Figure 6.

Neutralization activity of PG9, PG16 and PG9-16-RSH. (a) Antibody neutralization is shown against a cross-clade panel of 21 pseudoviruses: red, IC₅₀ < 0.05 μg ml⁻¹; orange, 0.05 < IC₅₀ < 0.5 μg ml⁻¹; yellow, 0.5 < IC₅₀ < 5 μg ml⁻¹; green, 5 < IC₅₀ < 50 μg ml⁻¹. The neutralization of PG9-16-RSH relative to PG9 is shown in the far-right column. (b) Correlation between PG9-16-RSH improvement as a function of PG9 potency. PG9-16-RSH is approximately five-fold more potent on viral isolates that are potently neutralized by PG9. Correlation on a 195-virus panel is shown in Supplementary Table 3. (c) Breadth/potency curve obtained from a panel of 195 HIV-1 strains. The fraction of HIV-1 strains neutralized (vertical axis) is shown graphed versus the neutralization potency of PG9, PG16, PG9-16-RSH and MIN(PG9-IC₅₀,PG16-IC₅₀) (horizontal axis), where MIN(PG9-IC₅₀,PG16-IC₅₀) refers to the lower IC₅₀ of either PG9 or PG16 for each HIV-1 isolate.

Because the enhanced glycan recognition of PG9-PG16-RSH may lead to reactivity with glycans on self proteins, we tested the chimera along with the parent PG9 and PG16 for autoreactivity (Supplementary Fig. 4). PG9-16-RSH retained the clean autoreactivity profile of the parent PG9 and PG16 antibodies (Supplementary Fig. 4). Overall, our results demonstrate how the combination of optimal features of recognition between somatically related antibodies can lead to enhanced neutralization of HIV-1.

Role of sialic acid in HIV-1 neutralization by PG9-16-RSH

In the preceding sections, structural and STD NMR analyses show that PG16 binds a complex-type or hybrid-type N-glycan through its terminal sialic acid and preceding galactose moiety. To further investigate the role of the terminal sialic acid on complex-type glycan on the viral spike, we performed HIV-1 neutralization assays with and without treating viruses with neuraminidase, an enzyme that cleaves the glycosidic linkage between sialic acid and galactose. We chose three HIV-1 strains (AC10, Q769.h5 and ZM176), for which the PG9-16-RSH chimera showed at least a five-fold improvement in potency over PG9. PG16 and PG9-16-RSH were less potent against neuraminidase-treated viruses than against untreated viruses, exhibiting 2-30 times higher IC₅₀ values (Fig. 7). PG9 neutralization was less affected by the neuraminidase treatment (two times higher IC₅₀). This decreased neutralization of neuraminidase-treated viruses by PG16 and PG9-16-RSH confirms the role of hybrid-type or complex-type glycan at Asn156_gp120 or Asn173_gp120 and the terminal sialic acid moiety in the recognition of gp120 by PG16. PG16 and PG9-16-RSH are more potent when the sialic acid on hybrid-or complex-type N-linked glycans are present on the viral spike, and this raises the question of which glycan types are on the viral spike. Of note, the neuraminidase-treated viruses were still neutralized by PG16 and PG9-16-RSH (at a lower potency), and this indicates that other interactions, including those with V1–V2 residues and with Man₅GlcNAc₂ at N-glycan 160 and weak interactions with the galactose moiety of N-glycan 173, were sufficient for neutralization. Indeed, a few viruses do not possess a glycan at position 156 or 173 but are neutralized by both PG9 and PG16, and this indicates that this glycan is not necessary for effective neutralization by these antibodies. Nevertheless, the reduced potency after neuraminidase treatment suggests that interaction with sialic acid is an important contributor to overall binding strength.

Figure 7.

Effect of sialic acid removal by neuraminidase treatment on PG9, PG16 and PG9-16-RSH antibody-mediated neutralization of HIV-1. Neuraminidase treatment and HIV-1 neutralization by PG9, PG16 and PG9-16-RSH are shown. Three pseudoviruses, AC10, Q769.h5 and ZM176, were either treated or not treated with neuraminidase, which removes sialic acid from complex-type glycans. Potency ratios of wild type (WT)/neuraminidase treated are shown for PG9, PG16, PG9-16-RSH, PGT128 and VRC01. Asterisk denotes inability of PGT128 to neutralize Q769.h5 strain.

Discussion

Somatic variation represents a general antibody mechanism to enhance affinity and overcome viral diversity^5,37,46–48. In the case of the PG9 and PG16 antibodies, the combination of PG9 and PG16 neutralizes HIV-1 with better breadth (PG9) and potency (PG16) than for either antibody alone (Fig. 6c). Here we used both crystallographic and NMR techniques to elucidate the structural basis for the neutralization differences between PG9 and PG16 and to create a PG9-16 variant with the ability to neutralize and determine the optimal PG9-PG16 combination.

Can our results with PG9 and PG16 be extended to other antibodies? Our analysis identifies critical contact residues that differ between somatic relatives, and it makes use of somatic variation to design antibodies with enhanced function. Such analysis would probably have identified the important G54W_HC alteration, which increases the potency of antibody NIH45-46 by approximately ten times⁴⁸; Trp54 is at a critical contact between antibody and gp120, and a tryptophan is naturally present in the VRC03 antibody^47,49, a somatic relative of NIH45-46. Meanwhile, analysis of the somatic variation for various gp120 glycan–reactive antibodies shows that a number of somatic variants occur at glycan-interactive residues (Supplementary Fig. 5). Our results suggest that a complex-type N-linked glycan often exists at 173_gp120 or 156_gp120. This may in part explain observations regarding the presence and importance of complex glycans on the viral spike. Differences have been observed in potency between various neutralization assays⁵⁰ as well as in the amount and types of glycans present on viruses from different sources²⁴. In the case of glycopeptide-directed antibodies, such as PG9 and PG16, antibody neutralization assays should be re-evaluated to take into account the in vivo diversity of glycans on the functional viral spike (especially at Asn156 or Asn173), as this may more accurately reflect the breadth and/or potency of these antibodies⁵¹. PG9 and PG16 both require Man₅GlcNAc₂ at position 160_gp120 to neutralize HIV-1 (ref. 13), whereas the N-linked glycan at position 156 or 173, although usually present in most viruses, is not required¹⁸. Overall, we have defined the structural basis for recognition of diverse glycans at the Asn156_gp120 or Asn173_gp120 site by antibodies PG9 and PG16 (Fig. 8).

Figure 8.

Glycan recognition, binding affinities, and detailed interactions of PG16, PG9 and PG9-16-RSH. (a) Recognition and binding affinities for complex-type glycan and Man₅ binding to the individual glycan-binding sites of PG16, PG9 and PG9-16-RSH. Antibodies are depicted schematically with heavy and light chains in tan and light blue (PG16) and yellow and blue (PG9), respectively; antibody sites that bind glycan 160_gp120 are green and those that bind glycan 156_gp120 or 173_gp120 are yellow and magenta, respectively. X denotes sites mutated to remove glycan binding. Glycan schematics are depicted in standard nomenclature and are semitransparent when recognition is poor. K_d values were obtained by fitting the data to a single-site binding model. (^# data did not fit a two-site model, consistent with very weak affinity for glycan at the second binding site; *very weak and negligible binding are described in Supplementary Fig. 3f.) (b) Detailed interactions governing binding of Man₅–Asn160 and complex-type glycan–Asn173 to PG16. Hydrogen bonds between heavy- and light-chain residues and sugar moieties of Man₅ (top) and hybrid-type glycan (bottom) are labeled, and the interfaces that also include hydrophobic interactions are represented by thick pink lines. Top, longer oligomannose branches whose interactions with PG16 are sterically allowed (D1 arm) are shown in pale green and those that would clash (D2, D3 arms) in pale gray. The Asn173_gp120 site (bottom) is optimal for binding hybrid or complex-type glycans (shown) but can also accommodate oligomannosides (transparent rings).

Our results, along with the structure of PGT121 bound to a complex-type glycan³⁷, represent some of the first reports of an HIV-1–neutralizing antibody whose epitope comprises a terminal sialic acid–containing glycan. Sialic acid is atypical in terms of its carbohydrate chemistry: it is a C9 sugar with a C-2 anomeric carbon, a carboxylate at C-1, an amine at C-5 and a glycerol side chain at C-7 to C-9. Accordingly, sialic acid–binding proteins, such as the Siglec family and the selectins, have developed precise ways of recognizing and binding to this unique sugar^52,53. This is best illustrated in the Siglecs, in which the glycerol side chain and N-acetyl group of sialic acid hydrogen-bond to main chain atoms, to precisely mimic an extended antiparallel β-sheet, and a conserved arginine forms hydrogen bonds to the C-1 carboxylate⁵⁴. For PG16, the immune system appears to have evolved a different strategy for sialic acid recognition in which backbone and side chain nitrogens of two histidine residues, one located in the heavy chain and one in the light chain of the antibody, dominate the electrostatic interactions. Although arginine residues are present, direct interactions between their guanidine groups and sialic acid are (somewhat unexpectedly) not observed.

In terms of vaccine design, our data indicate that design efforts for V1–V2–containing immunogens will probably need to take into account specific glycans at two different sites; a gp120-based immunogen targeting this region should probably have a Man₅GlcNAc₂ at residue 160_immunogen and a complex-type glycan at residue 156_immunogen or 173_immunogen. Testing of heterogeneously glycosylated immunogens, in which multiple permutations of glycan type and glycosylation site have been introduced into a given V1–V2 sequence, may provide a means to confirm this hypothesis. In general, it is notable that, whereas variation can be observed at N-glycan 156 or 173 sites, our data support the presence of a Man₅GlcNAc₂ at position 160. An understanding of the factors that lead to conservation at N-glycan 160 and diversity at N-glycan 156 or 173 may assist in understanding the interplay between viral glycan diversity and antibody-mediated recognition and neutralization.

Online Methods

Formation and purification of PG16–V1–V2 scaffold complexes

Expression and purification of 1FD6 V1–V2 scaffolds and PG16 N23Q HRV3C were performed as described previously¹⁸. Briefly, the genes coding for the scaffolded V1–V2 were transiently transfected in HEK293S GnTI⁻ cells and HEK293F in which 10 mg of swainsonine had been added 2 h before transfection. Supernatants were collected after 4–5 days, and the scaffolded V1–V2 proteins were purified over a Ni column, the tags cleaved with HRV3C and the samples run over gel-filtration columns. Complexes were formed as described previously, with a PG16 N23Q HRV3C IgG on-column procedure with 30-fold molar excess of the V1–V2 scaffold¹⁸. The yield of complex obtained was low, ∼2%.

PG16–V1–V2 complex crystallization and data collection

A complex of PG16 bound to 1FD6ZM109 with four N-linked asparagines mutated to alanine (except Asn160 and Asn173) was screened against 576 crystallization conditions by a Cartesian Honeybee crystallization robot. Initial crystals were grown by the vapor diffusion method in sitting drops at 20 °C by mixing 0.2 μl of protein complex with 0.2 μl of reservoir solution (15% (w/v) PEG 3350, 20% (v/v) isopropanol and 0.2 M ammonium citrate, pH 7.5). Crystals suitable for diffraction were manually reproduced in hanging drops by mixing equal volumes of protein complex with reservoir solution (15% (w/v) PEG 3350, 20% (v/v) isopropanol and 0.2 M ammonium citrate, pH 7.5). Single crystals were flash frozen in liquid nitrogen in 25% (w/v) PEG 3350, 34% (v/v) isopropanol, 0.2 M ammonium citrate, pH 7.5, and 15% (v/v) 2R,3R-butanediol. Data to 2.50 Å were collected at a wavelength of 1.00 Å at the SER-CAT beamline ID-22 (Advanced Photon Source, Argonne National Laboratory).

PG16–V1–V2 complex structure determination and refinement

The diffraction data were processed with the HKL2000 suite⁵⁵. The PG9-1FD6ZM109 structure was used as the search model for the PG16-1FD6ZM109 data set. A molecular replacement solution consisting of one complex per asymmetric unit was obtained by Phenix⁵⁶. Coot⁵⁷ and Phenix⁵⁸ were used for model building and refinement, respectively. Final data collection and refinement statistics are presented in Table 1. The Ramachandran plot as determined by MolProbity⁵⁹ shows 95.1% of all residues in favored regions and 99.8% of all residues in allowed regions. The coordinates were deposited in the PDB under PDB ID 4DQO.

NMR experiments

NMR data were acquired on a Bruker Avance 600 MHz NMR equipped with a triple-resonance cryoprobe with z-axis gradients. Topspin 2.1 was used to process and analyze the NMR data. Resonance assignments of the glycans were made by 2D ¹H-¹³C correlation spectra including HSQC, HSQC-TOCY and HSQC-NOESY experiments using standard pulse sequences (Supplementary Table 4). 1D STD NMR spectra were obtained by applying a train of 50-ms Gaussian-shaped radio-frequency pulses separated by 1-ms delays for a total of 0.2-s saturation time with the carrier set at −1 p.p.m. (to selectively irradiate protein resonances), or +40 p.p.m. for reference spectra. Difference spectra were obtained by subtracting the reference spectrum from the saturated spectrum. Each was acquired with 2,048 scans giving an average experiment time of 8 h. During NMR experiments, water and protein resonances were suppressed by a binomial 3-9-19 pulse sequence and a 10-ms T1ρ filter, respectively. Samples were prepared in 20 mM sodium phosphate buffer containing 50 mM sodium chloride at pH 6.8. The equilibrium dissociation constants (K_d) of glycans were measured by fitting ASTD values as a function of glycan concentration using the equation y=B_max/(K_d+x), where x is the ligand concentration and B_max represents the plateau of the curve. Experiments were carried out at 298 K.

Neutralization assays

Neutralization was measured by using single-round-of-infection HIV-1 Env-pseudoviruses and TZM-bl target cells, as described previously^49,60,61. Neutralization curves were fit by nonlinear regression by a five-parameter Hill slope equation as previously described⁶⁰. The 50% and 80% inhibitory concentrations (IC₅₀ and IC₈₀) were reported as the antibody concentrations required to inhibit infection by 50% and 80%, respectively. For the sialic acid experiments, unconcentrated pseudoviruses were either treated or mock treated with 75 mU of neuraminidase (Sigma-Aldrich, cat. no. N8271) at 37 °C for 4 h before addition of the antibodies.

Autoreactivity assays

Reactivity to HIV-1 negative human epithelial (HEp-2) cells was determined by indirect immunofluorescence on slides, with Evans Blue as a counterstain and FITC-conjugated goat anti-human IgG (Zeus Scientific)⁶². Slides were photographed on a Nikon Optiphot fluorescence microscope. Kodachrome slides were taken of each monoclonal antibody binding to HEp-2 cells at a10-s exposure and the slides scanned into digital format (Supplementary Fig. 4). The Luminex AtheNA Multi-Lyte ANA test (Wampole Laboratories) was used to test for monoclonal antibody reactivity to SSA/Ro, SS-B/La, Sm, ribonucleoprotein (RNP), Jo-1, double-stranded DNA, centromere B and histone and was performed per the manufacturer's specifications and as previously described⁶². Monoclonal antibody concentrations assayed were 50, 25, 12.5 and 6.25 μg/ml. Ten milliliters for each concentration were incubated with the Luminex fluorescent beads and the test performed per the manufacturer's specifications. Reactivity with cardiolipin was tested with Luminex technology as well, with the ACL assay (Wampole Laboratories).

ELISAs

For details on the assays used to measure binding of PG9 and PG16 or of lectins to scaffolded V1–V2 (Supplementary Figs. 1 and 2), please see Supplementary Note.

Figures

Structure figures were prepared with PyMol (http://www.pymol.org/).