Structural characterization of a highly-potent V3-glycan broadly neutralizing antibody bound to natively-glycosylated HIV-1 envelope

Broadly neutralizing antibodies (bNAbs) isolated from HIV-1-infected individuals inform HIV-1 vaccine design efforts. Developing bNAbs with increased efficacy requires understanding how antibodies interact with the native oligomannose and complex-type N-glycan shield that hides most protein epitopes on HIV-1 envelope (Env). Here we present crystal structures, including a 3.8-Å X-ray free electron laser dataset, of natively glycosylated Env trimers complexed with BG18, the most potent V3/N332gp120 glycan-targeting bNAb reported to date. Our structures show conserved contacts mediated by common D gene-encoded residues with the N332gp120 glycan and the gp120 GDIR peptide motif, but a distinct Env-binding orientation relative to PGT121/10-1074 bNAbs. BG18’s binding orientation provides additional contacts with N392gp120 and N386gp120 glycans near the V3-loop base and engages protein components of the V1-loop. The BG18-natively-glycosylated Env structures facilitate understanding of bNAb–glycan interactions critical for using V3/N332gp120 bNAbs therapeutically and targeting their epitope for immunogen design. Structure determination of glycosylated HIV-1 envelope (Env) trimers complexed with broadly neutralizing antibodies (bNAbs) promotes a better understanding of bNAb epitopes. Here the authors present the structures of natively-glycosylated Env in complex with the highly-potent bNAb BG18, which is of interest for HIV-1 vaccine development.

I n the ongoing fight against the HIV-1 pandemic, the discovery and characterization of broadly neutralizing antibodies (bNAbs) against HIV-1 envelope (Env) fuel new efforts in treatment and management of infection. Next-generation bNAbs protected against and reduced viral loads in humanized mouse 1,2 and non-human primate 3,4 models of infection and exhibited anti-viral activity in human trials [5][6][7][8][9] . Therefore, a vaccine that elicits such antibodies is likely to be efficacious. Despite their promise, unusual properties of HIV-1 bNAbs-such as a high degree of somatic hypermutations 10 , long heavy chain complementarity determining region 3 (CDRH3) loops 11 , and/or short light chain complementarity determining region 3 (CDRL3) loops 12 -have made it difficult to elicit bNAbs by immunization. In addition, innate features of the viral Env spike that interfere with broad-based immunity include the diversity of Env sequences that arise by mutation 13 , a low Env density on the virion surface that interferes with bivalent antibody binding 14,15 , and host glycans that shield most of the Env surface 16 .
The glycan shield comprises~50% of the mass of HIV-1 Env, a (gp120-gp41) 3 trimer, and consists of N-glycans attached to an average of 30 ± 3 potential N-linked glycosylation sites (PNGSs) per gp120-gp41 protomer 17 . The carbohydrates decorating the surface of Env reduce access to protein epitopes and are generally non-immunogenic because they are assembled by host cell machinery 16 . Although under-processed oligomannose N-glycans cover parts of HIV-1 Env such as the N332 gp120 glycan region of gp120, processed complex-type N-glycans predominate in other regions of Env 18 and protect the CD4 binding site (CD4bs) and the variable loop 3 (V3-loop) of gp120 19 . The production of soluble native-like Env trimers (SOSIPs) 20 has enabled structure determinations of glycosylated Env-bNAb complexes and a better understanding of bNAb epitopes 11,[21][22][23][24] . However, structural information pertaining to bNAb recognition of highly glycosylated HIV-1 Env trimers in the context of native glycan shields has been difficult to obtain due to chemical and conformational heterogeneity of N-glycans that usually precludes crystallization required for an X-ray structure determination. Thus with one exception 25 , all monomeric and trimeric Env crystal structures were solved using glycoproteins produced with exclusively high mannose-only forms of N-glycans, which were usually enzymatically trimmed after being complexed with antibody Fabs 22,26-37 . A single-particle cryo-electron microscopy (cryo-EM) structure of a natively glycosylated HIV-1 Env showed complex-type N-glycans at the gp120-gp41 interface, but many of the remaining glycans were not ordered in the EM map 38,39 .
The V3/N332 gp120 class of HIV-1 bNAbs, exemplified by the PGT121/10-1074 family 40,41 , evolved to interact with both glycan and protein components on HIV-1 Env to effect the neutralization of a majority of HIV-1 strains 31,[42][43][44] . These bNAbs possess long CDRH3s that interact with the N332 gp120 glycan and penetrate the glycan shield to contact the conserved 324 GD/ NIR 327 peptide motif at the base of the gp120 V3-loop 25,32,42,45 . bNAbs against the V3/N332 gp120 site isolated from several HIV-1-infected donors can adopt different Env-binding orientations to engage the conserved epitope 29,44 and display a wide array of interactions with surrounding glycans, including glycans at positions N301 gp120 (the PGT128 bNAb), N137 gp120 /N156 gp120 / N301 gp120 (PGT121/10-1074 family), and N386 gp120 /N392 gp120 (PGT135). While under-processed N-glycans presenting as a oligomannose patch predominate in the V3/N332 gp120 epitope, the recent structure of 10-1074 bound to a natively glycosylated BG505 SOSIP trimer included complex-type N-glycans at positions N156 gp120 and N301 gp120 25 , and sialic acid-bearing complex-type glycans at the N156 gp120 position were shown to be critical for maturation of the CAP256 V2 apex bNAb lineage 46 . Thus, providing structural information of bNAbs bound to Env trimers that include both complex-type and high mannose glycans will facilitate developing strategies for improving bNAb breadth and for design of high-affinity germline-binding immunogens.
BG18 exhibits the highest potency among the V3/N332 gp120 bNAbs described to date 47 . Isolated from an elite controller infected with clade B HIV-1, BG18 displays a similar breadth of coverage (~64%) across a panel of HIV-1 strains to the PGT121/ 10-1074 family bNAbs (Supplementary Figure 1), but BG18 neutralizes these strains with a geometric mean IC 50 value of 0.03 µg/mL, a higher potency than 10-1074 and other bNAbs in human clinical trials 8 . The structure of unliganded BG18 Fab 47 exhibited a cleft between the CDRH2 and CDRH3 loops and a long CDRH3 loop that forms a two-stranded β-sheet as previously observed for PGT121 and 10-1074 Fab structures solved in the absence of HIV-1 Env 40 . However, comparison of the BG18 Fab structure (PDB 5UD9 (https://doi.org/10.2210/ pdb5UD9/pdb)) with structures of unbound PGT121 and 10-1074 Fabs (4FQ1 (https://doi.org/10.2210/pdb4FQ1/pdb) and 4FQ2 (https://doi.org/10.2210/pdb4FQ2/pdb)) demonstrated structural differences including: (1) a displaced, shorter, and more compact CDRH3 stabilized by a network of hydrogen bonds, (2) differences in variable light (V L ) domain orientation relative to the variable heavy (V H ) domain, and (3) a second cleft in the antigen-binding site between CDRH3 and CDRL1/CDRL3. In addition, single-particle electron microscopy showed that BG18 exhibited a different orientation compared with PGT121/10-1074 and other V3/N332 gp120 bNAbs for binding HIV-1 Env 47 .
To better understand the molecular mechanism underlying BG18's interactions with Env glycans and increased potency compared with other V3/N332 gp120 bNAbs, we determined the crystal structures of natively glycosylated clade A (BG505) and clade B (B41) Envs in complex with BG18 Fab. We used the increased brightness of an X-ray free electron laser 48 (XFEL) to circumvent the limitations of crystal size and improve the resolution to 3.8 Å for our BG18-BG505-35O22 complex. We found that BG18 binds the V3/N332 gp120 glycan site in a distinct manner relative to PGT121-like bNAbs, primarily through rearrangements in its V H and V L domains. Analysis of BG18 interactions bound to a natively glycosylated Env showed engagement with oligomannose glycans near the base of the V3-loop and the conserved GDIR peptide motif. Moreover, BG18's CDRL1 and CDRL2 formed part of a cleft within 8 Å of gp120's variable loop 1 (V1-loop), increasing BG18's protein surface contact with Env. These structures provide valuable information for understanding the promiscuity of V3/N332 gp120 glycan-directed bNAbs that will facilitate current efforts to evaluate them for therapeutic use in HIV-1-infected humans 8 and to target their epitope for immunogen design 49,50 .
The BG18-BG505-35O22, BG18-B41-35O22, and BG18-BG505-IOMA structures each comprised an Env trimer bound to three BG18 Fabs and three 35O22 or IOMA Fabs (Fig. 1a-c), with ordered electron density corresponding to native glycans at the Fab interfaces ( Fig. 1d and Supplementary Figure 3a). The orientations of BG18 Fab with respect to gp120 were preserved across the three crystal structures and a low-resolution EM structure of a BG18-BG505 complex 47 Figure 3b-d), indicating that the orientation was conserved across different Env strains and was not an artifact of crystal packing (Supplementary Figure 3e,f). In addition, BG18 did not alter binding modes at the gp41/gp120 interface and CD4bs by 35O22 or IOMA, respectively, as these Fabs adopted similar conformations as previously observed on trimeric Envs 25,27 .

(Supplementary
Superimposition of the BG18 V H -V L coordinates in the BG18-BG505-35O22 structure with V H -V L in unbound BG18 Fab (PDB 5UD9 (https://doi.org/10.2210/pdb5UD9/pdb)) resulted in a 1.3-Å root mean square deviation (rmsd) (240 Cα atoms), demonstrating that BG18 CDR loops (except for CDRL2, which was disordered in unbound BG18 47 ) did not undergo large conformational rearrangements upon binding Env, and thereby maintained the previously observed clefts (Fig. 2a, b). Most notably, the conformation and location of CDRH3 in unbound BG18, which differs from CDRH3s in unbound PGT121 and 10-1074 40,47 , were preserved in the Env-bound BG18 structure (Fig. 2c, d). Additionally, interactions with gp120 and glycans at the base of the V3-loop resulted in stabilization of CDRL2.
BG18 adopts a distinct Env-binding orientation. A lowresolution Env-bound BG18 structure derived by negative stain EM showed an orientation for BG18 distinct from the orientations of 10-1074 and other V3/N332 gp120 bNAbs 47 . Despite crystallization contacts involving Fabs (Supplementary Figure 3e, f), BG18 maintained this distinct orientation in our crystal structures compared with orientations in structures of HIV-1 Env trimers with Fabs from 10-1074, PGT122, a PGT121 intermediate, and PGT124 25,27,31,50 (Fig. 3 and Supplementary Figure 4). To analyze these differences, we calculated the rotation and translation for V H -V L domains of Env-bound BG18 Fab when compared to Env-bound 10-1074 Fab structures. We found that the mature BG18 V H -V L domains differed by~78˚relative to the orientations of 10-1074 Fab V H -V L domains (Fig. 3d, e), which contrasts with the~5˚difference between Env-bound PGT122 and 10-1074 Fab orientations. Notwithstanding its different orientation, BG18, 10-1074, and PGT121-like bNAbs share a common mode of interaction with the N332 gp120 glycan  Fig. 1 Crystal structures of natively glycosylated HIV-1 Env trimers complexed with BG18. a Cartoon representation of the 3.8 Å structure of BG505 Env (gp120, light gray; gp41, dark gray) in complex with BG18 (blue) and 35O22 (orange) Fabs. Ordered, native high-mannose glycans (cyan) are represented as sticks, and complex-type glycans (salmon) are shown as spheres. b 4.9 Å structure of B41 Env bound to BG18 (blue) and 35O22 (orange). Glycans shown as cyan spheres. c 6.7 Å structure of BG505 Env bound to BG18 (blue) and IOMA (green) Fabs. Glycans were not modeled due to the limited resolution. d Close-up of the BG18 binding site (V H in dark blue, V L in light blue superimposed on the final 2F o −F calc electron density map contoured at 1.2σ) on the surface of gp120 (gray) from the BG18-BG505-35O22 structure. Ordered glycans (cyan) near the BG18 binding site are represented as sticks through their CDRH3 loops (Fig. 3c). This interaction includes a structural motif with a consensus R-I-Y-G-V/I-I sequence (BG18 residue numbers 101-106 and 10-1074/PGT122 residue numbers 100-100E; Fig. 2d) encoded by the same antibody D3-3 gene segment, likely explaining the nearly identical N332 gp120 glycan recognition. Given that CDRH3 is a primary determinant of the interactions of V3/N332 gp120 bNAbs with Env trimer 29,32,40,44 , its displacement in both unbound and Env-bound BG18 (Fig. 2c) rationalizes its orientation of Env binding relative to other V3/ N332 gp120 bNAbs. Consistent with these observations, BG18's footprint at the V3/N332 gp120 epitope differs from PGT121/10-1074-like bNAbs, such that its interactions with Env glycans and protein components are mediated by different CDR loops relative to the CDR loops used by 10-1074 (Fig. 4a, b and Supplementary Figure 5).

N-linked glycan interpretation and interactions with BG18.
Modeling of glycans was achieved in our structures by interpreting electron density at PNGSs using 2F o −F c maps calculated with model phases and in composite annealed omit maps to reduce model bias 56 . In the BG18-BG505-35O22 structures, 17 Nlinked glycans were modeled into ordered electron density, forming glycan arrays that extended~22 Å from the Env surface ( Fig. 1a and Supplementary Figure 6). N-glycans near the BG18 and 35O22 interfaces were modeled as mostly oligomannose (Man 5-9 GlcNAc 2 ) in our structures to avoid over-interpretation of the electron density, except for complex-type glycans at positions N88 gp120 and N156 gp120 , which were assigned as complex-type based on density for a core fucose in our 3.8 Å XFEL structure (Supplementary Figure 7). Although complex-type N-glycans were modeled at positions N160 gp120 , N276 gp120 , N301 gp120 , and N392 gp120 in a previous natively glycosylated 10-1074-Env-IOMA structure 25 and observed at positions N160 gp120 , N197 gp120 , and N276 gp120 by mass spectrometry 18,52,57 , our datasets did not show characteristic densities for complex glycans at these positions. Modeling of oligomannose N-linked glycans into ordered electron density was also possible in the 4.9 Å BG18-B41-35O22 structure, and for the N332 gp120 glycan in the 6.7 Å BG18-BG505-IOMA structure (Fig. 1b,c and Supplementary Figure 6). Overlay of BG505 and B41 Env structures showed conservation of the glycan shield surrounding the BG18 binding site (Supplementary Figure 6). Despite resolution limitations that necessitated glycan modeling as predominantly oligomannose at PNGSs shown to attach complex glycoforms, the use of natively glycosylated Env trimers in the crystallization complexes ensured accuracy of bNAb binding orientations and interfaces with Env trimer.
Molecular details of BG18-Env interactions. BG18 conserves interactions with the gp120 GDIR peptide motif at the base of the V3-loop that are seen in other V3/N332 gp120 bNAbs (Fig. 5a, b and Supplementary Figure 5). However, rotation of BG18's light chain relative to other V3/N332 gp120 bNAbs places only CDRL2 in close proximity to the gp120 GDIR peptide motif, compared to engagement of GDIR by multiple CDR loops and framework a d c regions observed in 10-1074/PGT121-like bNAb recognition (Fig. 5b, c) 25,29,31 . These differences in light chain interactions reflect the germline origins of the BG18 and PGT121/10-1074 light chains, which derive from different VL gene segments (Fig. 5d). However, in an example of convergence toward a common chemical binding mechanism, light chain residues involved in GDIR recognition by BG18 CDRL2 derive from hypermutation from the germline LV3-25*02 gene segment, whereas serines in the CDRL3s of PGT121 and 10-1074 derive from J regions chosen during V-J joining. B41 Env harbors a GNIR sequence instead of GDIR, allowing us to analyze the D325N gp120 substitution in our BG18-B41-35O22 structure. Side chain placement was not possible due to low resolution (4.9 Å), but the BG18 interaction with B41 resembled its interaction with BG505 in the BG18-BG505-35O22 structure ( Supplementary Figure 9), suggesting that BG18 recognizes the GNIR motif analogously to how it recognizes GDIR. Consistent with this result, analysis of BG18 neutralization of HIV-1 isolates containing GNIR motifs showed only a 2-fold loss in potency, by contrast to PGT121 or PGT122, which showed 6-fold and~44-fold losses, respectively (Table 2). However, while BG18 S53 LC potentially engages N325 gp120 ( Supplementary  Figure 8b), the loss of BG18 Q54 LC contacts resulted in the disorder of CDRL2 residues 54-60 in our BG18-B41 structure.

Discussion
Structures of bNAbs complexed with HIV-1 Env trimers have helped elucidate the molecular correlates for anti-HIV-1 antibody breadth and potency. Here we report four crystal structures of the highly potent V3/N332 gp120 bNAb BG18 bound to natively glycosylated clade A and clade B Envs (Fig. 1a-c). Our structures of clade A BG505 and clade B B41 HIV-1 strains represent only the second example of fully and natively glycosylated Env crystal structures, with the first being BG505 complexed with another V3/N332 gp120 bNAb, 10-1074, and with the CD4bs bNAb IOMA 25 . Given the crucial role the Env glycan shield plays in HIV-1's immune evasion strategies 13,58 , the prevalence of bNAbs that interact with complex glycans 18,25,58 , and the importance of complex glycans in bNAb maturation 46 , solving structures containing both complex and high-mannose glycans provides a more complete picture of bNAb recognition of Env epitopes. The newly identified BG18-Env-35O22 crystal lattice system packs solely through Fab interactions (Table 1 and Supplementary  Figure 3e), similar to crystals of previously described Env-bNAb complexes 25,27 , thus providing an additional system to study HIV-1 Env diversity. Moreover, the improvement in resolution from 4.1 Å using conventional crystals to 3.8 Å resolution by exposing smaller crystal volumes using an XFEL (Supplementary Figure 2) offers the potential to examine natively glycosylated Env trimer-Fab structures to higher resolutions. The demonstration that two HIV-1 Env trimers (BG505 and B41) can be crystallized in different crystal packing lattices without converting their   4 Glycan interactions with V3/N332 gp120 bNAbs in structures including natively glycosylated Env. a, b Comparison of the orientations on gp120 (gray surface and cartoon) of the CDR loops from (a) BG18 (blue, ribbon) and (b) 10-1074 (magenta, ribbon) demonstrating that the BG18 variable domains are rotated clockwise about CDRH3 relative to the 10-1074 variable domains. The distinct BG18 orientation on gp120 resulted in contacts with N156 gp120 , N386 gp120 , and N392 gp120 glycans (cyan, sticks) in proximity to the V3-base. Red dotted outline: Differences in the N332 gp120 glycan conformation on BG505 Env bound to BG18 (a) or to 10-1074 (b). c Close-up of BG18 interaction with the N332 gp120 glycan showing CDRH3, CDRH1 and CDRL2 loops at the glycan interface. d Overlay of BG18-BG505 and 10-1074-BG505 (PDB 5T3Z (http://dx.doi.org10.2210/pdb5T3Z/pdb)) structures showed that the BG18-bound N332 gp120 glycan conformation would clash (yellow star) with light chain CDR loops of 10-1074 and other PGT121-like bNAbs that display nearly identical binding modes. e Surface representation of BG18 interactions with the N392 gp120 and N386 gp120 glycans showing the N392 gp120 glycan buried (~800 Å 2 total BSA) inside cleft 2 located between the CDRH3 and CDRL1/3 loops. The N386 gp120 glycan associates weakly with light chain BG18 (45 Å 2 total BSA), but forms branch-branch interactions with the N392 gp120 glycan. c, e Electron density contoured at 1σ from 2F obs −F calc composite annealed omit maps calculated with phases from models with glycan coordinates omitted to reduce potential phase bias (gray mesh) NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03632-y ARTICLE NATURE COMMUNICATIONS | (2018) 9:1251 | DOI: 10.1038/s41467-018-03632-y | www.nature.com/naturecommunications glycans to exclusively high-mannose forms provides an impetus for further crystallization efforts using natively glycosylated HIV-1 Envs. Resulting crystal structures can be compared to natively glycosylated Env structures determined by cryo-EM 38 , a method that does not require crystallization and can therefore be used for heterogeneously glycosylated samples.
Our structures were consistent with previous evidence that BG18 binds with a distinct orientation compared to the prototype PGT121/10-1074 bNAbs in the V3/N332 gp120 glycan-targeting family 47 and showed extensive interactions with both protein and glycan components of gp120 (Fig. 1d). BG18's CDRH3 interactions with GDIR and the N332 gp120 glycan are conserved with other V3/N332 gp120 bNAbs (Supplementary Figure 5), Table 2 Glycan and sequence preference for V3/N332 gp120 targeting bNAbs in the presence of N332 gp120 glycan , and gp120 (gray; residues 318-330). Electron density contoured at 1σ from 2F obs −F calc final refined electron density map with −100 Å 2 B-sharpening is shown (gray mesh). b BG18 interactions with the gp120 GDIR motif (G324g p120 -D325g p120 -I326g p120 -R327g p120 ) at the base of the V3-loop. In common with the PGT121-like bNAbs, CDRH3 E111 HC forms a potential salt bridge with R327 gp120 , while CDRH3 Y104 HC stacks against D325 gp120. In addition, CDRL2 residues S53 LC and Q54 LC engage D325 gp120 forming potential H-bond interactions. c Comparison of BG18 (light blue) and 10-1074 (pink; PDB 5T3Z (http://dx.doi.org10.2210/pdb5T3Z/pdb)) interactions with D325 gp120 in the GDIR motif. Engagement of the carboxylate group of D325 gp120 is achieved by residues in CDRL2 (BG18) or serine residues in CDRs L1 and L3 (10-1074). Potential H-bonds represented as dashed lines. d Alignment of sequences of inferred germline and mature light chains of BG18 and 10-1074. The LV3-21*02 germline V gene segment is used for both 10-1074 and PGT121-like antibodies 40 . Red asterisks indicate residues involved in D325 gp120 recognition in the GDIR motif as shown in c serving as the main driver for epitope recognition (Figs 3c and  5b). Strikingly, the D3-3 gene for BG18/10-1074/PGT121 bNAbs that encodes a CDRH3 consensus structural motif responsible for N332 gp120 glycan interactions plays a role analogous to the VH1-2 gene for VRC01-like bNAbs 24,59,60 for epitope targeting. This common feature, which unites the PGT121 and BG18 bNAb families, suggests that it provides the key interaction in the initial binding of their unmutated ancestor precursors to Env. Interestingly, this common interaction occurs despite different orientations for the rest of the V H -V L domains (Fig. 4a, b). For example, unlike other V3/N332 gp120 bNAbs, BG18's light chain CDR loops straddle the V1-loop of gp120 (Fig. 6), increasing surface contact with gp120 protein components. In addition, BG18's interactions with gp120 N-glycans differ from the PGT121-like family in both engagement and ability to retain potent neutralization properties. For instance, despite extensive interactions with glycans at positions N392 gp120 and N386 gp120 (Fig. 4a, e and Supplementary Data 1), BG18 can potently neutralize HIV-1 strains lacking these glycans, by contrast to the more weakly neutralizing V3/N332 gp120 bNAb, PGT135, which relies on N332 gp120 , N392 gp120 , and N386 gp120 glycans for its anti-HIV-1 activity 44 (Table 2). Consistent with this observation, analysis of neutralization by BG18 and PGT121/10-1074 family bNAbs of HIV-1 strains including the N332 gp120 glycan, but with and without glycans known to interact with bNAbs targeting the V3/N332 gp120 epitope, showed that BG18 and 10-1074 retained their anti-HIV-1 potency (mean IC 50 s < 0.6 µg/ mL), whereas PGT121/PGT122 potency was abrogated upon removal of the N301 gp120 glycan (Table 2). Furthermore, PGT122 showed sensitivity to D325N gp120 mutations in the GDIR peptide motif. Taken together, these differences illustrate the divergence of solutions evolved by bNAbs to target this epitope, and BG18's successful strategies to accommodate Env sequence diversity.
Recent studies showed that priming with designed SOSIP Env trimers that bind weakly to the common inferred germline sequence (iGL) of PGT121 and 10-1074 50 could elicit bNAbs resembling PGT121 in PGT121 iGL Ig knock-in mice 49 .  ig. 6 BG18 contacts with gp120 V1-loop. a Surface and cartoon representation of BG18 V H (dark blue) and V L (light blue) loops involved in gp120 V1-loop (gray; residues 128-158, disordered residues depicted as dashed red line) interactions. Electron density contoured at 1σ from a 2F obs −F calc composite annealed omit map is shown (gray mesh). Glycans at position N133 gp120 and N156 gp120 are shown. b Cartoon and stick representation of residue interactions between BG18 V L domain (light blue) and the gp120 V1-loop (gray). Potential H-bonding occurs between T139 gp120 and BG18 T30 LC in the CDRL1 loop. In addition, BG18 W67 LC in FWRL3 stacks against I138 gp120 . BG18 contacts with gp120 positions 138 and 139 are likely specific to Envs with V1 characteristics similar to BG505, since similar conformations were not observed in our BG18-B41 structure. H-bonds and pi-stacking are indicated by black dashed lines. Red asterisk on N137 gp120 indicates a PNGS. c Electrostatic surface potentials with red indicating negative electrostatic potential and blue indicating positive electrostatic potential for BG18 shown with cartoon and stick representation of nearby gp120 elements. BG18 includes a positively charged cleft in the vicinity of the gp120 V1-loop (dashed line indicates disordered region), which may provide increased protein-protein interactions with the gp120 surface in HIV-1 strains harboring charged residues in the V1-loop However, the iGL of BG18 did not bind detectably to the tightestbinding designed priming immunogen 11MUT B (Supplementary Figure 11a), which can be rationalized by one of several differences in BG18 and PGT121/10-1074 recognition of Env. First, mature BG18 CDRL2 residues are involved in GDIR recognition with S53 LC and Q54 LC engaging D325 gp120 of GDIR (Fig. 5b). Comparison of iGL and mature BG18 CDRL2 amino acid sequences showed that four of five residues were mutated ( 50 YKDSE 54 vs. 50 SRSSQ 54 , respectively) (Fig. 5d). Not only does the germline CDRL2 have increased bulk, but also acidic residues flank S53 LC , which interacts with D325 gp120 . In contrast, iGL sequences for CDRL1/3 in PGT121/10-1074 show conservation of serines responsible for interacting with D325 gp120 in GDIR and flanking residues (Fig. 5d). Thus, since the 11MUT B priming immunogen contains no substitutions in GDIR, BG18 iGL likely shows a reduced ability to interact with this region compared with PGT121/10-1074 iGLs. An additional predicted impediment to iGL BG18 binding to 11MUT B is the relative proximity of light chain CDRs to the gp120 V1-loop (Fig. 6). Since the CDRs are the most heavily substituted between iGL and mature sequences, altering the chemical properties of this cleft could negatively affect binding. Moreover, 11MUT B harbors seven mutations in the V1-loop necessary for iGL PGT121 binding 50 , which likely clash with iGL BG18 given its V1-loop interactions (Supplementary Figure 11b). Finally, mature BG18 is heavily substituted compared to iGL sequences (35 and 26 heavy chain and light chain amino acid mutations, respectively). Previous structural studies of a PGT121 intermediate bound to BG505 Env showed that orientations are defined early during maturation and that differences in amino acid composition can alter antibody footprints on gp120 31 . Thus, it is possible that the conformation of iGL BG18 is incompatible with 11MUT B binding.
Although eliciting BG18-like bNAbs would require a different set of designed immunogens than being used for PGT121/10-1074 bNAbs 49,50 , they might be easier to elicit because of a shorter CDRH3 than PGT121/10-1074 bNAbs. By elucidating the molecular details of BG18's distinct interaction with the V3/ N332 gp120 epitope, the structural information reported here will facilitate future immunogen design efforts.

Methods
Protein expression and purification. Fabs from BG18 (including a N26Q HC substitution 47 ), 35O22, and IOMA IgGs were produced as previously described 37 . Briefly, Fabs were expressed by transiently transfecting HEK293-6E cells (National Research Council of Canada) with vectors encoding the appropriate light chain and C-terminal 6x-His tagged heavy chain genes. Secreted Fabs were purified from cell supernatants using Ni 2+ -NTA affinity chromatography (GE Healthcare), followed by size exclusion chromatography (SEC) with a Superdex200 16/60 column (GE Healthcare). Purified Fabs were concentrated and maintained at 4°C in storage buffer (20 mM Tris pH 8.0, 150 mM NaCl, and 0.02% sodium azide).
Genes encoding BG505 SOSIP.664 gp140 20 or B41 SOSIP.664 gp140 61 trimers were stably expressed in CHO Flp-In TM cells (Invitrogen) as described 51 using cell lines kindly provided by John Moore (Weill Cornell Medical College). Plasmids encoding the BG505 SOSIP.664 gp140 N392 gene variant (see Supplementary  Table 1 for primer sequences) was transiently expressed in HEK293-6E cells (National Research Council of Canada) as previously described 22 . In both cases, secreted SOSIP trimers expressed in the absence of kifunensine were isolated from cellular supernatants using 2G12 immunoaffinity chromatography by covalently coupling 2G12 IgG monomer to an activated-NHS Sepharaose column (GE Healthcare). Trimers were eluted using 3 M MgCl 2 and immediately dialyzed into storage buffer before SEC purification with a Superdex200 16/60 column (GE Healthcare) against the same buffer. Peak fractions pertaining to SOSIP trimers were pooled and repurified over the same column and buffer conditions. Twelve 1.0-mL fractions were stored separately at 4°C.
Crystallization of BG18-Env complexes. Complexes for crystallization were produced by an overnight incubation of SOSIP with BG18 and 35O22 or IOMA Fabs at a 1:1:1 molar ratio, and subsequently concentrated to 5-8 mg/mL by centrifugation with a 30-kDa concentrator (Amicon). Initial matrix crystallization trials were performed at room temperature using the sitting drop vapor diffusion method by mixing equal volumes of protein sample and reservoir using a TTP LabTech Mosquito robot and commercially available screens (Hampton Research and Qiagen). Initial hits were optimized and crystals were obtained for BG18-BG505-35O22 and BG18-B41-35O22 in 0.1 M Tris pH 8.0, 5% Tacismate pH 8.0, and 14% polyethylene glycol (PEG) 3350 at 20°C. BG18-BG505-IOMA crystals were obtained in 0.1 M citric acid pH 3.7, 16% PEG 3350 at 20°C. Crystals were cryo-protected stepwise with reservoir and a final 20% v/v glycerol concentration before being cryopreserved in liquid nitrogen.
Structure determination and refinement. Conventional X-ray diffraction data were collected for BG18-Env complexes at the Stanford Synchroton Radiation Lightsource (SSRL) beamline 12-2 on a Pilatus 6M pixel detector (Dectris). Data from a single crystal for each complex were indexed and integrated in XDS 62 , and merged with AIMLESS in the CCP4 software suite 63 . Structures were determined by molecular replacement in PHASER 64 using a single search with coordinates of an a glycosylated gp120-4 protomer (PDB 5T3Z (https://doi.org/10.2210/pdb5T3Z/ pdb)), BG18 Fab (PDB 5UD9 (https://doi.org/10.2210/pdb5UD9/pdb)), and 35O22 Fab (PDB 4TOY (https://doi.org/10.2210/pdb4TOY/pdb)) or IOMA Fab (PDB 5T3Z (https://doi.org/10.2210/pdb5T3Z/pdb)). Models were refined using B-factor refinement in CNS 65 and Phenix 56 , followed by several cycles of manual building with B factor sharpening in Coot 55,66 . Glycans were initially interpreted and modeled using F o −F c maps calculated with model phases contoured at 2σ, followed by 2F o −F c simulated annealing composite omit maps in which modeled glycans were omitted to remove model bias 56 . N-linked glycans identified at individual PNGSs in our crystallographic studies on both BG505.664 and B41.664 SOSIP trimers were generally consistent with the mixture of glycans observed by mass spectroscopy 18,57 and previous crystallographic studies of a natively and fully glycosylated Env trimer 25 . Therefore, modeling of complex-type glycans at positions N88gp120 and N156gp120 were primarily determined by the presence of electron density characteristic of a core fucose, which when modeled, slightly lowered R free values. Additional details of glycan modeling and coordinate refinement were followed as previously described 25,67 . Inclusion of higher resolution data with weak intensities improved refinement behavior and stereochemistry as described 68 .
XFEL diffraction experiments were performed at the MFX endstation of LCLS using a standard goniometer setup 48 , 9.5 keV X-ray pulses with 40 fs duration and a 5-µm beam focus at the crystal interaction point. Diffraction images were recorded on a Rayonix MX325 detector data and were integrated with IOTA 69 using the data reduction algorithms implemented in cctbx.xfel 70 . Of the 627 collected images, 589 contained discernible diffraction; of these, 570 were successfully integrated and yielded correct crystal symmetry and unit cell values. Scaling, post-refinement, and merging was carried out using PRIME 71 . Of the 570 integrated diffraction images, 526 were included in the final merged dataset, which was complete (99.1%) to 3.8 Å, and exhibited good multiplicity (9.2-fold) and reasonable merging statistics (Table 1). Phases were generated by molecular replacement, using our refined 4.1 Å synchrotron structure with glycans omitted as a search model. Superposition and figures were rendered using PyMOL (Version 1.5.0.4 Schrodinger, LLC), and protein electrostatic calculations were achieved using APBS and PDB2PQR webservers 72 . BSAs were determined with PDBePISA using a 1.4-Å probe 73 . Potential hydrogen bonds were assigned using a distance of <3.6 Å and an A-D-H angle of >90°, while the maximum distance allowed for a van der Waals interaction was 4.0 Å. Putative H-bonds, van der Waals assignments, and total BSA should be considered tentative, owing to the relatively low structure resolutions.
Binding experiments. SPR experiments were carried out on a Biacore T100 (Biacore) using a standard single-cycle kinetics method as previously described 22,37 . Briefly, a CM5 chip, primarily amine coupled with Protein A, was used to immobilize 8ANC195 IgG, a gp120-gp41 interface bNAb 37 . Remaining Protein A sites were blocked with 1 µM human Fc. BG505 HIV-1 SOSIP trimers were captured by injecting 10 µM solutions at a flow rate of 30 µL/s for 180 s. Mature BG18 Fab was injected at decreasing concentrations (3-fold dilution series with a starting top concentration of 110 nM) at 30 µL/s for 60 s and allowed to dissociated for 300 s. Kinetic analyses were done after subtraction of reference curves to obtain k a , k d , and K D values for a 1:1 binding model with or without a bulk refractive index change correction as appropriate (Biacore T200 Evaluation software).
An ELISA to evaluate binding of mature and iGL BG18 IgG to 11MUT B SOSIP 50 was performed by coating of High-Bind 96-well plates (Corning #9018) with 50 µL per well of a 2-µg/mL solution of purified 11MUT B in PBS overnight at 4°C. Plates were washed six times with washing buffer (1× PBS with 0.05% Tween 20 (Sigma-Aldrich)) and incubated in blocking buffer (1× PBS with 1% non-fat milk) for 1 h at room temperature (RT). Immediately after blocking, IgGs were added in blocking buffer and incubated for 2 h at RT. Antibodies were assayed at a 5-µg/mL starting dilution and seven additional 3-fold serial dilutions. Plates were washed six times with washing buffer and then incubated with anti-human IgG secondary antibody conjugated to horseradish peroxidase (HRP) (Jackson Laboratories) in washing buffer at a 1:5000 dilution. Plates were developed by addition of the HRP substrate, ABTS (Life Technologies), and absorbance was measured at 405 nm with an ELISA microplate reader (FluoStar Omega, BMG Labtech).
Data availability. Coordinates and structure factors reported in this manuscript have been deposited in the Protein Data Bank with accession codes 6CH7, 6CH8, 6CH9, and 6CHB. Other data are available from the corresponding author upon reasonable request.