A Broadly Neutralizing Antibody Targets the Dynamic HIV Envelope Trimer Apex via a Long, Rigidified, and Anionic β-Hairpin Structure

1Numerous antibodies that target and neutralize a broad range of different human immunodeficiency virus (HIV) isolates have been found in chronically infected HIV donors. Some of these bnAbs inhibit HIV Env with remarkable breadth and potency by recognizing conserved “supersites” of vulnerability. One of these epitope clusters is located at the trimer apex, consisting of the variable loops 1 and 2 (V1/V2) that hold together the gp120 subunits of the trimer through inter-protomer interactions.

2True to its name, the V1/V2 region varies greatly in sequence and length. All HIV isolates nevertheless retain two notable features in this region. The V2 contains N-linked glycosylation sites at positions N160 and N156 (or the less common compensatory position N173), and a cluster of positively charged amino acids around the trimer 3-fold symmetry axis. In this manner, the trimer apex forms an immunogenic, structurally conserved motif consisting of an electropositive hole surrounded by N-linked glycans. Examples of patient-derived bnAbs that belong to this class include PG9, PG16, CH01-CH04, the CAP256-VRC26 lineage, PGT141-145, and PGDM1400-1412. PGDM1400 (83% breadth, 0.003 μg/mL median IC50) and CAP256-VRC26.25 (57% breadth, 0.001 μg/mL median IC50), in particular, are remarkably potent.

3Partial descriptions of paratope-epitope interactions have been obtained using epitope scaffolds with PG9, PG16, and the CH01-CH04 apex bnAbs. Hybrid-modeling approaches employing low-resolution negative-stain EM and X-ray structures of scaffolds indicate these bnAbs bind at or near the trimer 3-fold axis with a binding stoichiometry of one antigen-binding fragment (Fab) per trimer. This binding mode results in a symmetry mismatch, unique to this class of antibodies, and glycan heterogeneity makes them difficult targets for structural studies. All characterized apex bnAbs, except for some CAP256-VRC26 lineage antibodies, depend on glycans at N160 and N156/N173, and often fail to bind viruses produced in the presence of α-mannosidase-I inhibitor kifunensine (Kif) that results in homogeneous oligomannose glycans with 8-9 mannose (Man) residues. The structural basis of Env recognition for the PGT145-class of antibodies is highly sought after because its quaternary specificity is now widely exploited to detect and isolate properly formed Env trimers, including under GMP conditions for human vaccine trials. Using cryo-electron microscopy (cryoEM), we determined the structure of PGT145 Fab in complex with the soluble, recombinant Env trimer, BG505 SOSIP.664 to elucidate key molecular interactions at the Env apex. Our structural and biochemical analyses revealed that PGT145-class bnAbs utilize their CDR loops, especially HCDR2 to stabilize a long anti-parallel β-hairpin HCDR3. This structural rigidity allows the antibody to penetrate through the tightly packed N160 glycan shield network, to recognize the electropositive sink generated by the protein elements at the core of the trimer apex. Therefore, despite nearly all epitope contacts coming from the HCDR3, additional maturation of the remaining CDR loops influences the HCDR3 and is crucial for generating a potent PGT145-like antibody.

4Apex bnAbs discovered so far can be grouped according to their heavy chain (HC) complementarity determining region (CDR) 3 topology: (1) PG9-like with—or predicted to have—a hammerhead motif; or (2) PGT145-like with a long, anti-parallel β-hairpin. Here, we solved X-ray structures of unliganded PGT143 and PGT144 Fabs, and they too exhibit the β-hairpin HCDR3 motif as expected (Figure 1A, Table 1, Figure S1). The elongated HCDR3 conformation of this PGT145-class bnAbs results in a paratope that projects a long distance away from the surface of the Fab and enables epitope recognition at the C3 axis of the trimer apex via a long-range interaction (Figures 1B and 1C, S2A). To define the molecular interactions of an apex antibody, we generated the structures of BG505 SOSIP.664-3BNC117-PGT145 and BG505 SOSIP.664-3BNC117 by single particle cryoEM at global resolutions of ∼4.3 Å and ∼4.4 Å, respectively (Figures 1B, S2B–S2G). The PGT145 HCDR3 inserted perpendicularly to the V1/V2 β sheets between the triad of N160 glycans into a pocket formed at the trimer apex, to simultaneously engage all three gp120s (Figures 1B–1D). This quaternary epitope is only present in the correctly folded pre-fusion Env conformation.

5The trimer-specific PGT145 is one of the few antibodies with which we could previously extract cleaved wild-type JR-FL E168K EnvΔCT trimers from cell membranes (Figure S2A). PGT145 also increases the thermal stability of BG505 SOSIP.664 upon binding and is predicted to neutralize HIV by stabilizing the prefusion conformation of the Env trimer. However, surface plasmon resonance (SPR) showed improved binding of the Env-receptor mimic CD4-IgG2, when PGT145 Fab had pre-bound the trimer (Figures 1E, S3A). A subtle opening of the trimer apex induced by PGT145 HCDR3 (Figure 1F) might improve access to the CD4-binding site, albeit shown only in vitro with soluble components. In vivo, PGT145 IgG binding at the Env apex would however provide a steric block to CD4 binding on the host cell membrane.

6In the BG505-3BNC117 complex, we also observed a subtle opening of the trimer apex (Figure 1F) that appeared to be mediated by HC framework region 3 (HFR3) of 3BNC117. 3BNC117 and its clonal relative, 3BNC60, have a 71dWDFD74 insertion in HFR3, which is critical for neutralizing activity. In 3BNC117, the HFR3 interacts with the N-acetylglucosamine (GlcNAc) core of the N197 glycan with H71a, and residue W71d interacted with V3 R308 of the adjacent gp120 (Figure 1G). Env residue 308 is typically Arg or His (Arg: 32%, His: 39%), both of which provide favorable interactions with tryptophan. Thus, while PGT145 recognizes the closed conformation of the Env trimer, the trimer apex is very subtly open in the PGT145 bound conformation. Induced allosteric effects, such as 3BNC117 binding at the CD4 binding site, that lead to an increase in spacing between the N160 glycan triad, likely make the apical binding site more accessible to PGT145.

7The glycoforms to which apex antibodies bind have been a matter of debate. Studies have shown that apex antibodies such as PG16 bind sialylated hybrid glycans with much higher affinity than oligomannose glycans. Similarly, PG9 showed increased binding affinity, neutralization potency, and maximum percentage neutralization (MPN) in the presence of complex/hybrid glycans (Figure 2A). PGT145-family antibodies have also been shown to preferentially bind to wild-type viruses and trimers compared to Kif treated Envs. However, SPR analysis of PGT145 Fab binding to BG505 SOSIP.664 trimers demonstrated that PGT145 has the highest affinity for glucose N-acetyltransferase 1 (GnT1)-deficient 293S-produced trimers that only have oligomannose Man5-9 glycans, although the larger Man8-9 glycans in the +Kif condition are somewhat inhibitory (Figures 2B, S3B and S3C). Further, a 31-pseudovirus neutralization panel showed a ∼3.5-fold improved median IC50 of PGT145 in 293S compared to wild-type glycan producing 293T cell produced pseudoviruses (Figure 2A), suggesting that, while PGT145 can accommodate glycan heterogeneity, there is a preference for oligomannose forms.

8Previously, glycan arrays failed to detect PGT145 binding across a wide range of glycans. Here, we employed a dendrimer format of glycan array slides that creates a higher density of glycans. PGT145 and PGDM1400 bound the C-isomer of a Man8 glycan (M8C) (Figures S3D–S3F), but not the Man8 B-isomer (M8B) indicating that the terminal Man(α1-2) residue of the D3 arm may sterically clash with PGT145-family Fabs. These data suggested that not only the correct glycoform, but also close spacing of glycans is essential for PGT145 binding.

9In the cryoEM map, the primary interacting N160 glycan appeared as oligomannose, consistent with our glycan array analysis. None of the asymmetric N160 glycans revealed density corresponding to core fucosylation (Figures S4A and S4B), often found in complex glycans. The total buried surface area between PGT145 and the modeled N160 glycans was ∼555 Å2 (glycan 1: 384 Å2, glycan 2: 131 Å2, glycan 3: 40 Å2). PGT145 most prominently interacted with a single N160 glycan (N160glycan1), less extensively with a second N160 glycan (N160glycan2), and likely not at all with the third N160 glycan (N160glycan3) (Figures 3A–3D, S4A–S4C). The density resolved for N160glycan1 is approximately the size of Man6 (Figure S4A) and its core GlcNAc sugars bound the peptide backbone of the HCDR3 descending strand, while the D1 arm packed between HCDR2 and HCDR3, making polar contacts with highly conserved H52aHCDR2 (Figure 3B). Y100fHCDR3—a tyrosine known to be sulfated—orients R100a HCDR3, and the sulfate might also interact with N160glycan1 Asn, while the benzene ring face of Y100f HCDR3 is packed against T162 (Figure 2B). Although the D2 and D3 arms of N160glycan1 are not fully resolved, we infer from glycan array data that the terminal D3 arm would clash with the LC of PGT145 in its orientation in the structure (Figures 3A and S3E). Together, the data predicted that much of the glycan recognition is driven by interactions made with the GlcNAc stalk, with steric accommodation for a Man6, Man7, Man8, or even a hybrid glycan at N160glycan1. In contrast, a complex glycan with its bulkier Gal(β1-4)GlcNAc(β1-2/4/6) branching units would likely clash with PGT145. Thus, the mode of N160 glycan recognition differed from that of the PG9-like antibodies, in which the N160 glycan branches bury into a pocket formed by the hammerhead-shaped HCDR3 and the LCDRs (Figure S4D). The GlcNAc base of N160glycan2 interacted with Y100cHCDR3 and to some extent with LCDR1, although not well resolved in our structure (Figure 3C). LCDR1 is the most variable CDR segment among the PGT145-family variants (Figure S1B), suggesting that the PGT145-family antibodies might accommodate N160glycan2 in different ways. N160glycan3 projected away from the antibody likely because it is sterically inhibitory to PGT145 binding in its native conformation (Figures 3D, S4C). Consequently, the N160 glycan is both essential (on one, probably two protomers), as well as inhibitory (on the third protomer) to binding of PGT145, presenting a unique challenge to epitope recognition of an asymmetric antibody molecule at a symmetry axis.

10In contrast to hammerhead HCDR3-type apex antibodies, which are dependent on the N156 glycan and bind complex glycoforms on glycan arrays, none of the N156 glycans directly contacted PGT145 (Figures S4D and S4E). Ablation of the N156 glycosylation site by an N156K substitution abrogated neutralization by PGT145 (Figure 3E). We therefore looked for an indirect effect that could be ascribed to differential processing of N160 glycans in N156 glycan-lacking Env variants expressed in different cell-lines. SPR showed negligible binding of PGT145 Fab to all of the N156D substitution variants (Figure S3B), and inspection of the N156D mutant by negative stain EM revealed that this mutation enriched for non-native-like trimers (Figure S4F). Thus, the reduced binding of PGT145 to N156 glycan knockouts most likely arose from disruption of the conformational epitope through the loss of the inter-V2-strand interaction between Y173 and N156 core GlcNAc. Residue 173 is a Tyr, His, or Phe in >86% of Envs, all of which are suitable for stabilizing the N156 glycan. Indirect glycan effects on the PGT145 epitope might also result from variability in the length and number of glycosylation sites on V1 and V2 loops, as such viral attributes have been noted to increase with disease progression and correlate with reduced neutralization sensitivity. The most variable segment of V2, which is distal to the 3-fold axis, is flexible and disordered in the majority of high-resolution SOSIP.664 structures. The density corresponding to the projection of the V2 loop in BG505, however, suggested that its glycans would be close to the bound PGT145 Fab, and the N-linked glycosylation sites at N185e and N185h could restrict accessibility to the epitope (Figure S4G). Removal of the N185h glycan site in particular, led to a ∼1.5-fold increase in the association constant of PGT145 Fab, and an increase in the stoichiometry (Sm) from 0.87 to 0.96 (Figures 2B, S3B and S3C), confirming that outer-V2 region glycans can partially shield the apex bnAb epitope.

11Peptide contacts between PGT145 and Env (total buried surface area ∼729 Å2) are largely restricted to interactions between the tip of HCDR3 with the C-strand of V1/V2 and the C1 region near the base of V1 including K121 (Figure 3F). K121 from each protomer interacted with sulfated tyrosine Y100i, while R166 from each protomer interacted with various electronegative moieties in HCDR3 of PGT145 (Figure 3F). Finally, Y100mHCDR3 stacked up against the hydrocarbon segment of K169, which probably helps engage the N160glycan2 via a network of interactions along with E100o (Figure 3G).

12To confirm our structural observations, we conducted alanine-scanning neutralization assays on BG505 pseudovirus mutants (Figures 3E, S5A). Neutralization was reduced when the N160 or N156 glycan sites were knocked out (N156K, N160K, N160A, T162A), or direct peptide contact residues were substituted (K121A, R166A). Point mutations in V1/V2 that caused BG505 SOSIP.664 trimers to adopt aberrant trimer forms (M161A, L165A, D167A) when observed by EM, selectively decreased neutralization potencies of PGT145-family bnAbs versus hammerhead-group, PG9 and CH01 (Figures 3E, S5). Neutralization by CAP256-VRC26.09 was affected mostly by the same Env mutations that affected the PGT145 class. Thus, despite CAP256-VRC26.09 having a HCDR3 structure that more closely resembles the hammerhead-class, it is dependent on proper quaternary apex formation like PGT145, and engages all three protomers upon binding. Overall, our results further classified PGT145 as a distinct class of apex bnAb, compared to PG9, and that it is an N160 glycan-dependent antibody that relies on simultaneous recognition of peptide contacts from the apex central residues of all three gp120 V1V2 loops, and is indirectly affected by the glycan at N156.

13Apex-bnAbs recognize the conserved electropositive V2, particularly strand C. We looked for additional residues near the central electropositive layer that could add to the overall local charge. K117, which is 97% conserved, is deeply buried under the trimer apex and is likely important for PGT145 recognition. Additionally, 84% of clade B Envs have an arginine or lysine at V3 position 315, instead of glutamine commonly found in other clades, effectively increasing the positive charge potential at the center of the trimer apex (Figure S6). Residue 315 is located between the electropositive layer created by strand C and K121, which is part of the PGT145 epitope, and thus the additional electropositive layer might interact favorably with the electronegative HCDR3 (Figure S6E). However, PGDM1400- and CAP256-VRC26-lineage bnAbs have lower clade B breadth. Therefore, this extra positive residue likely influences trimer apex dynamics through electrostatic repulsion, as often observed in clade B SOSIP trimers. An increased tendency to “breathe” and open up at the apex would both alter the apex epitope but also enhance accessibility to the CD4bs, consistent with clade B Env infected-individuals preferentially generating CD4bs bnAb responses, but reduced apex responses.

14To determine whether there is any correlation between PGT145 IC50 and electropositive charges near the PGT145 epitope, we analyzed sequence variability at select positions in relation to neutralization using a linear regression model (Figure 4A). The goal of the model was to weigh the contribution of each residue to the escape phenotype defined by two different IC50 cut-off values, by considering the amino acid variation in each strain and their respective neutralization IC50 values. This weighting was reflected in the magnitude of the regression coefficient. The model predicts that the larger the magnitude of this coefficient, the more the residue influences resistance. In this analysis, residues that contact PGT145 and are important for neutralization, such as K121, R166, and T162, are also well conserved, having an exact match conservation score (relative to BG505) of ∼70% or higher (Figure 4A). This analysis also revealed that K171 is important even though it does not directly contact PGT145. K171 seemed to interact with the base of the N160 glycan and/or restrict the movement range of N156 and N160 glycans to indirectly affect PGT145 binding (Figures 3G and 4B). A subtler effect on PGT145 neutralization was observed among hydrophobic residues involved in stabilization of the trimer apex (Figures 4A, 4C, and 4D). Particular V1/V2 residues have neutralization regression coefficients that were more prominent with an IC50 > 1 μg/mL cut-off, such as L125, I309, L175, and I326, which also tend to be more conserved. In binding, the anionic HCDR3 of PGT145 stabilizes the trimer by offsetting the net charge at the apex, especially when Env contains additional cationic residues. Mechanistically, Env requires CD4 triggering to transition into an open state leading to exposure of the co-receptor binding site. The proximal localization of positively charged residues results in repulsion between the gp120 protomers, contributing to meta-stability and priming Env for CD4 induced conformational changes. Too much repulsion between protomers would result in a constitutively open trimer susceptible to degradation and therefore render the virus defunct. Thus, we propose that a fine balance between tight hydrophobic packing and localization of electropositive charge is required for viral fitness, which PGT145 exploits for epitope targeting (Figure 4E).

15While HCDR3 comprises the majority of the PGT145 paratope, the other CDR loops play important roles in neutralization. For example, reversion of HCDR2 in PGT143 and PGT145 to their predicted germline sequences abrogates neutralization despite few direct contacts with Env (Figure 3). Structural examination revealed that HCDR3 makes extensive intra- and inter-CDR contacts in PGT145-family Fabs. Within the HCDR3, Y101 and D100b, and E100o and Y100m form electrostatic pairing interactions. W100p packs against R100a, which interacts with Y100f to orient the tyrosine side chain. Eliminating these pairwise stabilizing interactions reduced PGT145 neutralization (Figures 5A–5C, 6), indicating that structural stabilization of the long HCDR3 via tertiary and quaternary contacts is necessary for neutralizing activity.

16We found a 52aH-D/E-X-D/Q53 motif in HCDR2 that appeared to stabilize the HCDR3 stalk (Figures 5B, 6, S1A). PGT143-PGT145 and PGDM1400 all contained extensive electrostatic interactions between the HCDR2 and 3, such as the interaction between R99HCDR3 with D53HCDR2 (Q53 in PGDM1403-6) and D100rHCDR3. The R99A paratope substitution had the second largest increase in PGT145 IC50 against all three pseudoviruses tested (Figure 5C). The D100rAHCDR3 or D53AHCDR2 substitutions also reduced potency, although not as severely as a single R99A substitution, which simultaneously ablates two electrostatic interactions. K97HCDR3 formed electrostatic interactions with E52bHCDR2 and D33HCDR1, where H52aHCDR2, a residue that contacted the N160glycan1 also played a role in orienting the side chain of E52bHCDR2. K97 is conserved across all PGT145-family mAbs except for PGDM1411, a weak neutralizer that has K98 in place of K97 and R99, thereby incompletely substituting for two basic residues. Further, introduction of the 4- (52aH-E-G-D53) or 6-residue HCDR2 (52S-H-E-G-D-K54) stabilization motif back into inferred-germline (iGL; germine reversion with the exception of HCDR3) PGT145 partially rescued the neutralizing activity of PGT145 iGL variants (Figures 5D and 5E). LCDR1 exhibited the greatest variability in CDR length and sequence (Figure S1B). Nevertheless, a semi-conserved Y32LCDR1 (9 of 18 MAbs) when mutated to alanine in PGT145 resulted in a large drop in neutralization (Figure 5C). In our four Fab structures, Y32LCDR1 hydrogen bonded with the π-nitrogen of conserved H98HCDR3 (Figures 5B and 5C). The remaining nine PGDM14XX MAbs had an F32LCDR1-F98HCDR3 interaction pair instead that plausibly stabilizes the HCDR3 by π-stacking.

17Based on the SHM patterns, B cell affinity selection during the Ab maturation process led to four distinct PGT145 Ab sub-lineages (Figures S1A and S1B), reminiscent of a fork, which contains a “shaft” consisting of mutations that accumulated early to give rise to the most recent common ancestor (MRCA) Ab. These mutations are shared across all the Ab lineage members, and the four “prongs” represent the sub-lineages. Conservation of the 52aH-D/E-X-D/Q53 motif suggests that certain HCDR2 mutations arose in the MRCA Ab and were critical for stabilizing a given HCDR3, then further diversified as the Ab sub-lineages diverged. Most of the prong mutations occurred in the C-term half of HCDR2, HCDR3, and LCDR1, likely due to strong viral selection pressure. The insertions in LCDR1 likely correspond to different N160glycan2 recognition strategies.

18The overall matured HCDR3 β-motif in the PGT145 family is similar to bovine antibodies, which form an anti-parallel β sheet “stalk” that support a globular domain at the turn of the β sheet, thought to recognize antigen. In humans, a cross-reactive influenza antibody called C05 uses a HCDR3 β-hairpin to bind the hemagglutinin receptor binding site and was shown to contain potential HCDR3 stabilizing mutations in HCDR1 and HCDR2; although its HCDR3 reflects the hammerhead shape of PG9, rather than the long β sheet stalks in bovine antibodies and PGT145. Therefore the structural and sequence analysis reveals that the evolution of these types of antibodies is driven by somatic hypermutation in which substitutions stabilize an extended HCDR, thereby reducing the entropy for binding antigen.

19HIV Env is remarkable in its ability to substantially alter its sequence and accommodate a high density of surface glycans and yet still fold and retain its conserved receptor binding and fusion functions. Here we described the structural basis of recognition and neutralization by a class of trimer apex-targeting bnAbs that exploit the intrinsic conservation in the trimer apex that enables Env to undergo receptor-induced conformational changes to attain its fusion active configuration. In the trimer ground state, the V1/V2 regions of the three protomers interact at the trimer apex to hold the gp120 subunits together and mask the co-receptor binding site. These interactions, however, cannot be overly stabilizing, as the V1/V2 region must remodel to enable the V3 base to bind the negatively charged N terminus of the CCR5/CXCR4 co-receptor for viral fusion to proceed. We hypothesize that Env achieves this meta-stable balance by enriching the apex in basic amino acids, a subset of which are also important for co-receptor engagement. The PGT145-family bnAbs then exploit this cationic apex through extended, unusually stabilized HCDR3 loops that contain acidic residues and sulfated tyrosines distributed along the length of the β-hairpin loop, with most charge localized to the apical tip. The HCDR3 stalk is structurally conserved across the PGT145 family yet primarily interacts with a single glycan at N160. The stalk, stabilized by tertiary and quaternary interactions with the other CDR loops, serves as a platform to insert the somewhat variable tip of the β-hairpin into the glycan-shielded trimer apex. The 3-fold axis at the trimer apex comprises a large deep pocket that is permissive to a variety of negatively charged residues at the apical tip of the β-hairpin. This type of Env recognition is unique to this class of antibodies as most bnAb families, such as the VRC01-class, PGT121-124 family, or even PG9/PG16, retain a few key specific peptide interactions that are conserved. Overall, our observations demonstrate that the potency of PGT145 is highly dependent on the CDR-stabilized HCDR3 structure and the acidic charge localization in the HCDR3 β-turn. We hypothesize that apex-directed bnAbs like PGT145 require less SHM to achieve breadth and potency because of the less-specific nature of binding to the apex; that is, the specific molecular determinants of binding are less stringent than for other bnAb classes. The steep angle of approach made possible by the long rigidified β-hairpin of HCDR3 also minimizes potential steric block by the hypervariable V1 and V2 loops. Viral fitness relies on preserving a fine balance between the closed and open conformations of the meta-stable, charged apical region of the Env trimer. By targeting these conserved properties, rather than sparsely conserved residues on the surface of Env, the PGT145-family of bnAbs have solved the HIV neutralization puzzle in a particularly creative fashion.

20Our structural description of Env in complex with an apex targeting bnAb highlights the importance of an intact trimer apex for eliciting an appropriate immune response to this region. As germline targeting strategies gain more popularity the apex will likely require a trimeric immunogen that has the correct disposition of glycans and positive charges rather than a monomeric scaffold. Most rational vaccine design approaches involve engagement of appropriate germline precursors and biasing SHM to evolve specific molecular contacts with the antigen. The trimer apex presents a different challenge. Here, several mutations that do not contact the antigen are critical for bnAb evolution. One strategy for analyzing germline antibodies capable of evolving these types of responses would be to search for homologous HCDR3-stabilizing residues that reside in the other CDRs. These mutations can also act as early guideposts for evaluating if an immune response is going down the right path.

21Untagged or C-term His6-tagged BG505 SOSIP.664 trimers were expressed in HEK293S or 293F cells and affinity purified using a 2G12 IgG cross-linked Sepharose column as described previously. The affinity-purified trimers were size exclusion purified using a HiLoad 26/600 Superdex 200 pg column in 20 mM Tris pH 7.4, 150 mM NaCl (1× TBS), unless stated otherwise. The JR-FL EnvΔCT-PGT145 Fab complex was purified as described using PGT145 as the pull-down reagent. Antibody IgGs and Fabs were expressed in 293F cells and purified as previously described (Supplemental Experimental Procedures).

22All negative stain grids were prepared using 400 Cu mesh carbon coated grids as previously described. PGT145 bound complexes were stained using NanoW for ∼30 s, and unliganded trimers were stained using 2% uranyl formate for 45 s to 1 min. The EnvΔCT-PGT145 Fab complex was imaged on a Tecnai T12 coupled with a Tietz PXL 2k × 2k CCD camera, at a magnification of 52,000× resulting in 2.65 Å/pix images. Images were collected using Leginon. Particles were processed as previously described. Unliganded trimer mutants were imaged on a Tecnai T12 microscope coupled with a Tietz TemCam F416 CMOS detector, at a magnification of 52,000× resulting in 2.05 Å/pix on the specimen plane. Reference-free 2D classes were generated using MSA/MRA to sort the different trimer forms. Because these trimers were SEC purified over a Superdex 200 Increase 10/300 GL column, a large number of the mutants contained a significant portion of particles corresponding to monomers and dimers. Thus monomer and dimer populations were included in the analysis. In the first round of classification, all classes that had more than one particle in the boxed class average were eliminated. From a second round of 2D classification, reference-free 2D class averages were sub-grouped into three populations; (1) “non-native” that includes monomers, dimers, and badly assembled trimers, (2) “native-like closed,” and (3) “native-like total” that includes both the closed (population [2]) and “breathing” (“native-like” but open) trimers (Supplemental Experimental Procedures).

23BG505 SOSIP.664 trimers produced in 293F cells were pre-complexed with 6-molar excess of 3BNC117 Fab overnight at 4°C, and size exclusion purified. To break the pseudo-symmetry of PGT145, PGT145 Fab was pre-complexed with a mouse Fab obtained from a commercial hybridoma cell line ATCC CRL-1757 (1757) that binds the HC of human Fabs (Figures S2A and S2B). To make the BG505 SOSIP.664-3BNC117-PGT145-1757 complex, PGT145 was combined with 1757 Fab (1:2 molar ratio) and size exclusion purified. The PGT145-1757 complex was then incubated with the purified BG505 SOSIP-3BNC117 complex and size exclusion purified. Samples were concentrated and manually frozen in liquid ethane. Data were collected via Leginon on an FEI Titan Krios electron microscope operating at 300 KeV coupled with a K2 Summit direct electron detector camera (Gatan) in counting mode at a magnification of 22,500× resulting in a pixel size of 1.31 Å/pixel, using a total dose of ∼32 e−/Å2. Data were processed using RELION 1.4b1. The BG505 SOSIP.664-3BNC117 complex was refined using 22,625 particles with C3 symmetry imposed, to ∼4.4 Å resolution at a Fourier shell correlation (FSC) cut-off of 0.143. 1757 was found to be specific for the constant region of the Fab HC, near the protein G binding site (Figure S2F) allowing for confidence in the PGT145 Fab orientation. The BG505 SOSIP.664-3BNC117-PGT145-1757 and BG505 SOSIP.664-3BNC117-PGT145 classes from 3D sorting were combined, resulting in a total of 65,060 particles that were refined without imposing symmetry to 4.7 Å resolution (FSC = 0.143). A soft edge mask masking out 1757 Fab and PGT145 Fab constant domains was applied for one additional iteration of focused refinement, resulting in the ∼4.3 Å resolution model (FSC = 0.143) (Figure S2D) (Supplemental Experimental Procedures).

24The crystal structures of BG505 SOSIP.664 (4TVP) and 3BNC117 Fab (4JPV) were used as templates to generate an initial atomic model using the Modeller plug-in in UCSF Chimera. The resulting model was iteratively fixed and refined in Coot and RosettaRelax employing Ramachandran constraints. Final models were chosen based on a combination of the Rosetta energy score, MolProbity and clash scores, and EMRinger score. Glycans were modeled into the finalized protein model as previously described, with all glycans being modeled as oligomannose. The protein structure in the BG505-3BNC117 model was used as an initial model to refine the PGT145-bound structure. The sulfated tyrosines in the PGT145 Fab X-ray structure were replaced with regular tyrosines because Rosetta fails to recognize sulfated tyrosines. The complete BG505-3BNC117-PGT145 complex was refined and followed by glycan modeling as was done for the BG505-3BNC117 complex (Supplemental Experimental Procedures).

25PGT143 and PGT144 Fabs were concentrated to 4-24 mg/mL. Fab samples were screened for crystallization using the 384 conditions of the JCSG Core Suite at both 277 and 293 K using the TSRI/IAVI/JCSG robotic Crystalmation system as described previously. Data collection was performed at cryogenic temperature (100 K) at beamline 23-ID of the Argonne Photon Source (APS), using a beam wavelength of 1.033 Å. The diffraction data were indexed, processed and scaled with HKL-2000 or XDS. Both structures were determined by molecular replacement using Phaser with PGT145 Fab as an initial model (3U1S). Model building and refinement was carried out using Coot-0.7 and Phenix (Supplemental Experimental Procedures).

26SPR analysis of PGT145 Fab binding to His-tagged BG505 SOSIP trimers was analyzed on a Biacore 3000 instrument at 25°C. Glycan knockout mutants were expressed in HEK293F cells unless otherwise indicated. All trimers were immobilized on the chip by His-tag capture, as previously described. To study interference or enhancement of CD4 and PGT145 binding, sequential binding analyses were also performed as stated previously (Supplemental Experimental Procedures).

27For pseudovirus production, we cotransfected HEK293T or 293S cells with an Env encoding and an Env-deficient backbone (pSG3DEnv) plasmids using Fugene 6 (1:2 ratio). Pseudoviruses were harvested 48–72 hr post-transfection, filtered, and titrated for use in neutralization assays. Neutralization was measured in TZM-bl target cells, as described previously (Supplemental Experimental Procedures).

28mAbs were screened on a custom high mannose array, consisting of 9 mannosides and 1 control sialylated N-glycan. The 10 amine-linked glycans were covalently immobilized onto custom NHS-ester dendron functionalized glass microscope slides (G3 and G4, ZBiotech). To assess mAb binding, the antibodies were pre-mixed with the detection antibody (anti-human-IgG R-PE). Following 15 min, the pre-complexed antibodies were applied directly to the slide surface and allowed to incubate for 1 hr and then washed. Washed arrays were dried by centrifugation and then scanned for RPE signal on a confocal microarray scanner (Supplemental Experimental Procedures).

29TZM-bl neutralization assay derived IC50 values of PGT145 from 135 strains was employed for linear regression analysis. Strains with IC50 > 10 μg/mL were classified as escape. Multiple sequence alignment was performed by MUSCLE using default parameters. By comparing the amino acid identity of each residue of interest in each strain to that of BG505, a numeric value was assigned based on an adjusted BLOSUM62 matrix. For each substitution relative to a given amino acid at BG505, the numeric value was computed by subtracting the substitution score from the self-substitution score. No substitution is represented by a value of 0. A negative value would represent the conservativeness of the substitution, with less conservative being more negative. As a result, the amino-acid sequence for each strain was converted to a list of integers. Combining these sequences in the integer representation generated a matrix. A logistic regression model with L1 regularization was then fit to the matrix with the escape phenotype as the targets. Logistic regression was performed using “linear_model.LogisticRegressionCV” in scikit-learn in python. Each residue of interest would be assigned a coefficient. A larger magnitude of coefficient a residue implied more influence it has on the escape phenotype. The absolute value of the coefficient was reported (Supplemental Experimental Procedures).

30J.H.L, R.A., and J.-P.J. designed the project. J.H.L. performed the EM work and structural analysis. J.-P.J. and L.K. performed the crystallization studies. J.H.L. and C.A.C. performed atomic modeling of the EM structures. R.A., C.S., D.S., and M.P. performed the neutralization assays. A.Y. and P.J.K. performed the SPR studies. R.M. performed the glycan array assays. N.C.W. performed the linear regression analysis. T.N. and C.B. provided reagents. J.H.L., R.A., P.J.K., D.R.B., and A.B.W. analyzed the data. R.A. contributed to figure generation. J.H.L., A.B.W., D.R.B., and I.A.W. wrote the manuscript. All authors were asked to comment on the manuscript.