Cryo-EM structure of a fully glycosylated soluble cleaved HIV-1 Env trimer

Specimen preparation, EM data acquisition, and image processing of SOSIP trimers

BG505 SOSIP.664 gp140 trimers were produced in HEK 293T cells and, hence, have a typical human cell glycosylation profile. The Env trimer is relatively small by EM standards (~425 kDa, of which almost half is glycan) and lacks features that facilitate high-resolution image processing (21). We therefore adopted a recently described cryo-EM feature-enhancement strategy (22), by adding PGV04 Fabs as fiducial markers for computational alignment of the trimer. We recorded the EM data on a direct electron detector, which improves the signal compared to conventional methods and enables correction for beam-induced motion and specimen drift (23). New image processing algorithms, similar to those that have recently provided near-atomic resolution characterization of select macromolecular complexes (24, 25), were used in the analysis. Together, these cryo-EM technical advances, combined with design and production of a stable soluble Env trimer, have enabled the reconstruction of the SOSIP.664:PGV04 complex to 5.8 Å resolution (Fig. 1 and fig. S1). The reconstructed electron potential map provided sufficient detail for modeling most of gp120, including the variable loops and the heptad repeat 1 (HR1) and HR2 components of gp41 (Fig. 1 and fig. S1). The EM reconstruction was validated by the appearance of the Fab and gp120 densities that were in excellent agreement with the previously determined structures, by several recently described quantitative metrics for EM (fig. S2) (21, 26, 27) and also by an independently obtained X-ray structure of the same trimer (but from HEK 293S GnT−/− cells and hence with a simpler glycan profile) in complex with the PGT122 bnAb at a similar resolution (28). The EM map presented here is significantly improved in resolution and in new features compared to previous trimer reconstructions; it also revealed additional density that is consistent with N-linked glycans on both gp120 and gp41 (fig. S4) (29).

Structural arrangement of gp120 and variable loops V1, V2, and V3

The gp120 core crystal structure in complex with PGV04 (PDB 3SE9, (30)) was docked into the EM map and further refined (see supplementary online material (31)). The crystal structure of a scaffolded V1/V2 protein (PDB 3U2S (9)) could also be fitted into density at the trimer apex without significant rearrangements (Fig. 2A and fig. S5A). Densities corresponding to GlcNAc residues of the glycans at the N156 and N160 sites (HXB2 numbering) within V1/V2 are apparent at the trimer apex (Fig. 2C and figs. S4 and S5A), and are consistent with the arrangement of the V1/V2 N156 and N160 glycans predicted by a low-resolution model of the same trimers in complex with bnAb PG9 (14). The V1 loop is in a slightly different conformation than in scaffolded ZM109 V1/V2 (9). The V2 loop could not be built in its entirety but was localized above the CD4 binding site (fig. S5A–B) in a position to restrict access to the CD4bs (discussed below). The V3 loop is situated directly beneath the V1/V2 hairpin with the tip pointing toward the trimer axis and interacting with the V2 base from the adjacent protomer, in a different configuration than previously observed (Fig. 2B and figs. S5A and C). The N197 glycan at the V2 base sits at the inter-protomer interface (Fig. 2C and fig. S5A) and may shield V3 epitopes on the trimer; various V3 antibodies have higher affinities for monomeric than trimeric Env, implicating the existence of such steric constraints (32, 33). Overall, the gp120 subunits have a compact globular configuration and are assembled into a trimer via a limited set of interactions at the apex that involve the V1/V2 and V3 loop structures (Figs. 1 and 2C). This arrangement exposes several basic residues at the trimer apex that are important targets for bnAbs such as PG9, PG16, PGT145 and CH01, which have acquired negatively charged sulfated tyrosine moieties in their paratopes (9) (Fig. 2D).

Structural arrangement of gp41

Gp41 forms a pedestal at the base of the trimer (Fig. 3, A and B). Each protomer is characterized by two long, prominent helices. The first (HR1) forms a three-helix bundle with the neighboring protomers in the trimer core; the second (HR2) wraps around the outer periphery of the trimer base, angling downward and with its C-terminus proximal to the viral membrane (Fig. 3, A and B). The three-helix bundle is formed by the C-terminal half of HR1 and resembles the arrangement in the crystal structure of post-fusion gp41 (34), and also in the proposed open, intermediate state seen by single-particle EM (35) (Fig. 3, C and D). To assess whether the three-helix bundle is a feature induced by PGV04 binding, we calculated an independent, 12.7 Å reconstruction of the unliganded trimer (fig. S6). The three-helix bundle was present in this unliganded structure; hence, it is likely that the gp41 conformation that we describe here represents the ground state, and not an activated intermediate, as previously suggested (35). Just above the three-helix bundle and below the trimer apex, we observe a small opening that is continuous with the exterior (fig. S7). The large hole that is a previously reported feature of the trimer core is likely a result of lower resolution reconstructions (> 20 Å) (36), as illustrated in fig. S3A.

To identify the gp41 C-terminus, we calculated an ~8 Å reconstruction of a similar trimer but with the last 14 amino acids deleted (designated SOSIP.650) (Fig. 3E). Difference maps showed that a segment corresponding to ~3.5 helical turns, or ~14 amino acids, was absent from the SOSIP.650 reconstruction (Fig. 3E and fig. S8). Additionally, our data indicate that residues around 650 in HR2 interact strongly with the bottom of HR1 (fig. S9) (37).

The N-terminal half of HR1, including a short helix, and the fusion peptide proximal region (FPPR) and fusion peptide (FP) emanate from the top of the internal coiled coil around the three-fold axis and turn down toward the trimer base (Fig. 3A). The internal region at the base of the trimer contains additional density that could not be assigned unambiguously, but likely corresponds to the intervening residues between HR1 and HR2, as well as the gp120 C1 and C5 regions (Fig. 3, A and B). This interpretation is consistent with the known inter-subunit interactions between gp120 C1/C5 and gp41 residues ~589–610 (18). This element of gp41 contains the internal disulfide bonded loop and also the engineered Cys substitutions that covalently link gp120 residue-501 to gp41 residue-605 via a disulfide bond in SOSIP gp140s (18). The relatively hydrophobic gp120 N- and C-termini and nearby gp41 residues are all sequestered towards the middle of the trimer. We have putatively assigned the hydrophobic FP to this solvent-inaccessible region (Fig. 3B); there are similarities here to influenza hemagglutinin, another type I fusion protein (28, 38). Localization at the gp120/gp41 interface would allow the FP to be released in response to CD4 and co-receptor-induced conformational changes within gp120 and gp41 that drive membrane fusion.

PGV04 binding site on the Env trimer

Subtle differences in the Fab PGV04 interaction with gp120 are apparent when the monomeric and trimeric forms of core gp120 are aligned, which may be explained by trimer-specific contacts (Fig. 4A). On the glycosylated trimer, the Fab light chain interacts extensively with glycan N276 (Fig. 4B and fig. S17). The same glycan prevents the VRC01 germline antibody from binding to gp120 monomers (39), so it may be a general impediment to this bNAb class. The glycosylated trimer EM structure also contains additional density for a region corresponding to the V2 loop and/or glycans (Fig. 4B) that likely play a role in recognition of the CD4bs.

Many SOSIP.664 trimer particles had <3 PGV04 Fabs bound, despite the Fab being in molar excess of trimer during sample preparation. We used template-based 3D sorting of single particle images to generate trimer populations with 3, 2, 1 and 0 Fabs bound. Each of the subsets was then independently refined to generate maps with resolutions of 5.8 Å, 7.9 Å, 16.9 Å and 19.1 Å, respectively (figs. S10–S15). The 7.9 Å, 2-Fab reconstruction (Fig. 4C) was sufficient to interpret subtle structural differences created by the presence or absence of a bound Fab. Specifically, PGV04 binding induced positive density corresponding to the putative V2 loop and/or glycans N363/386 (Fig. 4D). Additional density corresponding to glycan N301 from the neighboring gp120 protomer and N276 from the same gp120 protomer was also seen to interact with PGV04 (fig. S16C) (40). One interpretation is that PGV04 binding stabilizes both the glycans and the flexible loops in the trimer structure.

Solution isothermal titration calorimetry (ITC) measurements also revealed that PGV04 bound BG505 SOSIP.664 trimers in a sub-stoichiometric manner (<2), consistent with the EM results (Fig. 4E). The EM map does not reveal an obvious structural impediment to stoichiometric PGV04 binding (i.e., 3 per trimer). One hypothesis for why some trimers do not bind the bnAb or do so at a low stoichiometry is that incomplete or differential glycan processing creates some micro-heterogeneity. When we measured PGV04 binding to 293S cell-produced trimers that bear only high mannose sugars, it was again sub-stoichiometric (i.e., <2). However, when the latter trimers were enzymatically deglycosylated, the number of bound Fabs increased from <2 to ~2.5 and the average Kd improved to 73nM from 135nM; the implication is that specific glycans of variable composition and size can indeed interfere with antibody binding to the CD4bs. The unfavorable entropic contribution of the binding signal was also significantly lower for the deglycosylated trimers (Fig. 4F). The entropic cost of locking normally flexible glycans in place may be yet another viral defense against antibodies, potentially one that restricts the stimulation of bnAb germline B-cell receptors (39, 41). Eliminating such glycans may help the design of immunogens intended to induce CD4bs antibodies (39, 41).

Quaternary nature of the CD4 binding site

The CD4bs is relatively highly conserved and immunogenic in the context of HIV-1 infection. Env subunit vaccines, such as monomeric gp120 and uncleaved gp140s, have failed to elicit bnAbs against this site, while often inducing non- or poorly-neutralizing Abs that recognize overlapping epitopes (1). One implication is that the structure and immunogenicity of the CD4bs are adversely affected when presented in a non-native context; thus, if the site assumes the CD4-bound state (42), the outcome may be the predominant elicitation of antibodies that are non-neutralizing because they cannot bind to native trimers. The EM structure that we describe delineates the narrow range of approach angles available to CD4bs bnAbs when engaging the trimer, and shows how epitope complexity makes it difficult to target this site effectively via non-native immunogens. Amongst new information are the following points:

The CD4-induced subdomain known as the bridging sheet is presented on the SOSIP.664 gp140 trimers differently from how it appears on various versions of monomeric gp120 (42). Specifically, the strand arrangement at the base of V1/V2 is reversed on the trimer, such that V2 forms an adjacent/parallel interaction with β-20 instead of the anti-parallel β -interaction between the V1 base and β-20 found in the gp120 core (Fig. 5A and fig. S18). Thus, quaternary interactions mean that the topological relationship between the base of V2, which includes the N197 glycan, and the CD4bs is not the same on trimeric and monomeric gp120.

The neighboring protomer also plays a significant role in restricting access to the CD4bs. When the heavy chain of PGV04 or the related VRC01 bnAb binds the trimer, it is within 5 Å of a loop (residues 61–62) that precedes a short α-helix (α-0) in C1 of a neighboring gp120 protomer (Fig. 5B); likewise, when CD4 (PDB 4JM2) is docked into the CD4bs on the trimer, it also contacts an adjacent protomer (Fig. 5B). The α-0 helix is directly downstream from His66, a residue implicated in regulating the sampling of the CD4-bound conformation (43). Thus, the first stage of CD4 binding may require the trimer to flex to an extent, but once initial contact is made, the trimer is further destabilized, and co-receptor binding and fusion can proceed.

Docking CD4bs antibodies with a range of neutralization capabilities into the EM trimer model suggests that the crystal structures of monomeric gp120-Ab complexes do not completely define this epitope cluster. Additional, quaternary structure-dependent contacts are important for CD4bs Fabs to recognize trimeric Env, with the V1/V2 loop structure influencing how they bind and, hence, neutralize (Fig. 5C). Previous mutagenesis data are now more readily interpretable; for example, we can now explain why the light chain CDR1 and CDR3 residues are essential for neutralization by the b12 bnAb (44); the trimer structure shows they are positioned to interact with V1/V2 (fig. S19). The non-neutralizing Fabs b13 (3IDX) and F105 (3HI1) (6) clash with V1/V2 (same protomer) and V3 (adjacent protomer), impeding their ability to bind the trimer (Fig 5C). These results are consistent with both BG505 virus neutralization and antibody-binding studies using the corresponding SOSIP.664 gp140 trimers (19).

Immunoglobulin framework insertions and mutations are important for broad and potent neutralization (5). Our structure shows that framework region 3 in the heavy chain (HFR3) of CD4bs antibodies 3BNC117 (also 3BNC60), VRC03 and VRC06 interact with basic residues on an adjacent protomer (Fig 5, D and E). HFR3 insertions contain acidic residues that point toward conserved basic residues near the V3 tip of the adjacent gp120 protomer (Fig. 5E). A properly folded trimer positions these basic residues in an appropriate quaternary orientation for interacting with CD4bs bnAbs. The preference of VRC06 for cleaved over uncleaved trimers (45) is also consistent with previous observations that cleavage is critical for Env to adopt a native-like conformation (20).

Affinity maturation events in the evolution of the CH103 CD4bs bNAb involve the early appearance of acidic residues in the HFR3 loop that then persist (46). However, HFR3 makes no contact with gp120 in the crystal structure of the monomeric gp120-CH103 complex (PDB 4JAN) (46). The trimer structure now shows that HFR3 of CH103 is in close proximity to the basic residues on V3 from the adjacent gp120 (Fig. 5E). Thus, the quaternary context of the CH103 epitope helps explain why somatic hypermutation within HFR3 is critical, and why the acidic residues are conserved once they appear.

The features of the CD4bs outlined above are likely to be relevant to other bnAbs that are yet to be identified. Thus, the V3 interactions and the evolution of framework mutations essential for neutralization breadth and potency should now be taken into consideration. The quaternary constraints on these epitopes will affect how potent CD4bs bnAbs, such as VRC01, are induced and then undergo the affinity maturation process.

Summary

The 5.8 Å resolution cryo-EM structure presented here is very similar to the x-ray structure of the same SOSIP.664 gp140 trimers in complex with bnAb PGT122 (28). Both structures are, however, very different from a 6 Å structure recently reported for a detergent-solubilized, full-length but uncleaved trimer (47). Differences in the design of the Env trimers might contribute to some of these structural inconsistencies. Alternatively, the cryo-EM methodology used to derive the full-length, uncleaved trimer structure has been criticized and the controversy is as yet unresolved (48). Considered in isolation, our soluble SOSIP.664 gp140 trimer structures are highly consistent with known Env structure-function relationships, while revealing new information on the pre-fusion form of the trimer. The structures explain how bnAbs recognize their epitopes and why quaternary constraints prevent some non-nAbs from binding the trimer. Overall, the new information is a step towards understanding trimeric Env at atomic-level resolution. The trimer structure should also guide improvements to Env-based vaccines.

Supplementary Material