Vaccine-elicited primate antibodies use a distinct approach to the HIV-1 primary receptor binding site informing vaccine redesign

Significance The development of broadly neutralizing antibodies (bNAbs) to HIV-1 is often thought to be a key component of a successful vaccine. A common target of bNAbs is the conserved CD4 binding site (CD4bs) on the HIV envelope glycoprotein (Env) trimeric spike. Although CD4bs-directed bNAbs have been isolated from infected individuals, elicitation of such bNAbs by Env vaccination has proven difficult. To help understand the limitations of current immunogens, we structurally characterized two vaccine-elicited, CD4bs-directed non-bNAbs from primates. We demonstrate that these vaccine-elicited Abs attempt a vertical approach to the CD4bs, thereby clashing with the variable region of the trimeric spike cap, whereas CD4bs-directed bNAbs adopt angles of approach that avoid such clashes. This analysis can inform future vaccine redesign. HIV-1 neutralization requires Ab accessibility to the functional envelope glycoprotein (Env) spike. We recently reported the isolation of previously unidentified vaccine-elicited, CD4 binding site (CD4bs)-directed mAbs from rhesus macaques immunized with soluble Env trimers, indicating that this region is immunogenic in the context of subunit vaccination. To elucidate the interaction of the trimer-elicited mAbs with gp120 and their insufficient interaction with the HIV-1 primary isolate spike, we crystallized the Fab fragments of two mAbs, GE136 and GE148. Alanine scanning of their complementarity-determining regions, coupled with epitope scanning of their epitopes on gp120, revealed putative contact residues at the Ab/gp120 interface. Docking of the GE136 and GE148 Fabs to gp120, coupled with EM reconstructions of these nonbroadly neutralizing mAbs (non-bNAbs) binding to gp120 monomers and EM modeling to well-ordered trimers, suggested Ab approach to the CD4bs by a vertical angle of access relative to the more lateral mode of interaction used by the CD4bs-directed bNAbs VRC01 and PGV04. Fitting the structures into the available cryo-EM native spike density indicated clashes between these two vaccine-elicited mAbs and the topside variable region spike cap, whereas the bNAbs duck under this quaternary shield to access the CD4bs effectively on primary HIV isolates. These results provide a structural basis for the limited neutralizing breadth observed by current vaccine-induced, CD4bs-directed Abs and highlight the need for better ordered trimer immunogens. The analysis presented here therefore provides valuable information to guide HIV-1 vaccine immunogen redesign.

HIV-1 neutralization requires Ab accessibility to the functional envelope glycoprotein (Env) spike. We recently reported the isolation of previously unidentified vaccine-elicited, CD4 binding site (CD4bs)directed mAbs from rhesus macaques immunized with soluble Env trimers, indicating that this region is immunogenic in the context of subunit vaccination. To elucidate the interaction of the trimerelicited mAbs with gp120 and their insufficient interaction with the HIV-1 primary isolate spike, we crystallized the Fab fragments of two mAbs, GE136 and GE148. Alanine scanning of their complementarity-determining regions, coupled with epitope scanning of their epitopes on gp120, revealed putative contact residues at the Ab/gp120 interface. Docking of the GE136 and GE148 Fabs to gp120, coupled with EM reconstructions of these nonbroadly neutralizing mAbs (non-bNAbs) binding to gp120 monomers and EM modeling to well-ordered trimers, suggested Ab approach to the CD4bs by a vertical angle of access relative to the more lateral mode of interaction used by the CD4bs-directed bNAbs VRC01 and PGV04. Fitting the structures into the available cryo-EM native spike density indicated clashes between these two vaccine-elicited mAbs and the topside variable region spike cap, whereas the bNAbs duck under this quaternary shield to access the CD4bs effectively on primary HIV isolates. These results provide a structural basis for the limited neutralizing breadth observed by current vaccineinduced, CD4bs-directed Abs and highlight the need for better ordered trimer immunogens. The analysis presented here therefore provides valuable information to guide HIV-1 vaccine immunogen redesign. NHP | neutralizing antibodies | mode of recognition E ntry of HIV-1 into susceptible primate CD4 + chemokine receptor 5 (CCR5) + target cells is mediated by the trimeric surface envelope glycoproteins (Envs). The exterior Env, gp120, binds to the primary receptor, CD4, and to the coreceptor, CCR5. The conserved CD4 binding site (CD4bs) on gp120 is surrounded by highly variable regions and glycan shielding (1). The CD4bs is a major target for neutralizing Abs (2), and a plethora of potent and broadly neutralizing Abs (bNAbs) to this region, elicited during chronic HIV-1 infection, were isolated recently (3)(4)(5)(6)(7). In addition to the obvious rationale that the primary virus receptor binding site is a desired region to target for vaccine-induced B-cell responses, the identification of potent CD4bs-directed bNAbs, such as VRC01, strengthens the concept that targeting the conserved CD4bs is a desirable approach to accomplish broad neutralization (8)(9)(10). Although the use of Env-based vaccines to elicit bNAbs to the CD4bs has been unsuccessful so far, CD4bs-directed Abs of more limited neutralization breadth have been elicited (11)(12)(13). Recently, a panel of such Abs was isolated from single-cell-sorted memory B cells from nonhuman primates (NHPs) inoculated with soluble HIV-1 Env trimers (gp140-F, foldon trimers) (11,14). Alanine (Ala) scanning of gp120 and subsequent binding studies revealed that the epitopes of these vaccine-elicited mAbs partially overlap with the CD4bs-directed bNAb VRC01 but more closely align with the human infection-elicited non-bNAb, F105 (15).
To elucidate better the limitations of current vaccine-elicited, CD4bs-directed mAbs to access the functional Env spike, we crystallized the Fabs of two members of this class, GE148 and GE136. Using the high-resolution structures of the unliganded mAbs, we defined properties of epitope recognition using a systematic analysis of the Ab-antigen interaction by paratope Ala scans and mapping of their putative gp120-interactive regions by both binding and virus neutralization assays. Using this information and available gp120 core structures in concert with ClusPro (16) docking, we obtained relatively high-resolution models of the NHP mAb/gp120 complexes. To validate this analysis, we examined additional substitutions at predicted contact residues and observed increased neutralization potencies for both mAbs. The results indicated that the vaccine-elicited mAbs primarily used their heavy chain (HC) complementarity-determining region 3 (HCDR3) and light chain (LC) complementaritydetermining region 1 (LCDR1) to interact with their cognate Significance The development of broadly neutralizing antibodies (bNAbs) to HIV-1 is often thought to be a key component of a successful vaccine. A common target of bNAbs is the conserved CD4 binding site (CD4bs) on the HIV envelope glycoprotein (Env) trimeric spike. Although CD4bs-directed bNAbs have been isolated from infected individuals, elicitation of such bNAbs by Env vaccination has proven difficult. To help understand the limitations of current immunogens, we structurally characterized two vaccine-elicited, CD4bs-directed non-bNAbs from primates. We demonstrate that these vaccine-elicited Abs attempt a vertical approach to the CD4bs, thereby clashing with the variable region of the trimeric spike cap, whereas CD4bsdirected bNAbs adopt angles of approach that avoid such clashes. This analysis can inform future vaccine redesign. epitopes in the gp120 CD4bs. In silico docking and subsequent superimposition to available gp120 trimer models from electron tomography data of the native spike (17) revealed that the Ab likely approaches at a vertical angle to gain access to the CD4bs. This mode of approach was confirmed by negative-stain EM reconstructions of GE136 and GE148 in complex with gp120, and subsequent fitting to Env trimer reconstructions showed that access to the CD4bs in properly folded Env trimers that mimic the native Env spike was not achievable.
Thus, the results presented here indicate that these two vaccine-elicited, CD4bs-directed mAbs cannot access the functional spike of primary HIV-1 isolates from a vertical angle of approach, where accessibility to the CD4bs is limited by steric clashes with the overlying variable (V1/V2/V3) "loop" region cap (V-region cap) on the membrane-distal region of the native spike. This mode of binding is in contrast to the more lateral angles of approach used by bNAbs, such as VRC01 (4,18). These results provide a structural explanation in the context of the functional HIV-1 spike for the limited neutralization breadth displayed by CD4bs-directed Abs elicited by the current foldon trimers (11). EM studies of the foldon trimers confirmed that they have an open configuration compared with native Env spikes, likely allowing exposure of immunogenic CD4bs elements that are not exposed in the functional spike. These insights are invaluable for informed second-generation iterative immunogen redesign to enhance vaccine-induced B-cell responses against circulating HIV-1 variants.

Results
Crystal Structures of GE136 and GE148. To define the properties of CD4bs-directed Abs elicited by YU2 gp140-F trimers in NHPs, we analyzed the interactions of GE136 and GE148 with several forms of Env. A summary of the gene use, HCDR3 length, binding, and neutralization capacity of GE136 and GE148 is shown in Fig. 1A. Although these mAbs bind with high affinity to monomeric gp120, their neutralization breadth is largely restricted to neutralization-sensitive tier 1 viruses, with limited activity against tier 2 viruses in the TZM-bl neutralization assay (19). However, they neutralized two clade B tier 2 isolates, RHPA and SC22, in the more sensitive A3R5 neutralization assay (20) (Fig. S1).
To obtain a deeper understanding of the properties of vaccineinduced, CD4bs-directed Abs, we crystallized the GE136 and GE148 Fab fragments using high-throughput crystallization screening and iterative optimization (21). We obtained crystals of both Fabs that diffracted to a resolution of 2.0 Å and 1.8 Å, respectively. The structural data are summarized in Table 1. As expected, the overall structures of the GE136 [Protein Data Bank (PDB) ID code 4KTD] and GE148 (PDB ID code 4KTE) Fabs reveal that the six CDRs of both Fabs form a discontinuous surface that comprises their binding sites (Fig. 1B, Top). When examined from a "head-on" perspective of the Ab binding sites (paratope), each mAb displayed relatively prominent and protruding centrally positioned HCDR3s (Fig. 1B, Middle). Analysis with University of California, San Francisco Chimera software revealed the hydrophobic character of the molecular surface of each mAb, demonstrating that the Ab HCDR3 loops were highly hydrophobic at their surface-exposed "tips" (Fig. 1B, Bottom),  indicating that hydrophobic interactions might participate significantly in epitope recognition. This finding was consistent with analysis of additional vaccine-elicited, CD4bs-directed mAbs, which also displayed hydrophobic HCDR3s (Fig. S2A). Using the structure coordinates of the human CD4bs-directed mAbs F105 and b13 (PDB ID codes 1U6A and 3IDX) (22), we analyzed the molecular surface of these non-bNAbs and found that they too possessed hydrophobic HCDR3 tips (Fig. 1C). In contrast, b12 (PDB ID code 2NY7), which is structurally related to b13 but is markedly more broadly neutralizing (22), lacked a hydrophobic HCDR3, as did the recently described CD4bsdirected bNAb CH103 (PDB ID code 4JAM) (6), which uses its HCDR3 to bind gp120 (Fig. 1D). The bNAbs VRC01 (PDB ID code 3NGB) and PGV04 (PDB ID code 3SE9) (Fig. 1D) or VRC23 (PDB ID code 4J6R), VRC03 (PDB ID code 3SE8), and VRC06 (PDB ID code 4JB9) (Fig. S2B), which are less relevant for this comparison because they do not use their HCDR3s for recognition of the gp120 CD4bs, also possessed nonhydrophobic HCDR3s (4,23). Electrostatic analyses of GE136 and GE148 were consistent with the hydrophobic nature of their HCDR3s (Fig. S2C).
Ala Scanning of GE148 CDRs Reveals Affinity-Related Effects. Next, we sought to define the specific interactions of GE148 and GE136 with their epitopes in the gp120 CD4bs. We used selected forms of Env, both in soluble formats and in the context of the functional virus spike, to determine the interactions of the vaccine-elicited mAbs with Env. To begin, we used our available unliganded Fab structures as the basis to interrogate interactions at the CD4bs. An extensive Ala scan of the GE148 HCDRs and LCDRs was performed (Fig. 2) to decipher any specific and critical contacts of GE148 with gp120 Env. Delineation of the framework and CDR regions is shown in Fig. S2D. A panel of the GE148 IgG HCDR Ala mutants was used initially to assess recognition of monomeric YU2 gp120 by ELISA compared with the unmodified WT GE148 IgG ( Fig. 2A, left columns). Somewhat surprisingly, we found little impact of most Ala substitutions on mAb recognition except for a single residue on the C-terminal flank of the HCDR3 loop at position K100e (Kabat numbering). Because the immunogen that elicited GE148 was a YU2-based trimer, we reasoned that perhaps the relative affinity (avidity) to the homologous Env was too high to detect single-residue effects of Ala substitution on GE148 recognition. Accordingly, we performed a similar scan on an alternate Env target of a different strain, the HXBc2 core variant, V3S (24). Again, we detected reduced binding at residue K100e ( Fig. 2A), as well as a weaker effect at E100f. Because GE148 binds the cysteine-stabilized HXBc2-based core, 2CC (24), but with 100-fold lower affinity compared with the isogenic nonstabilized core (11), V3S, we performed a "low-affinity" Ala scan and found several additional residues that reduced recognition of 2CC by the GE148 Ala variants. The strongest effect again centered on residue K100e, but Ala substitutions of flanking residues F100g and E100f also displayed greatly reduced binding. Several other residues in the HCDR3 and HCDR2 regions also affected binding. To confirm the specificity of these effects in a format assessing direct affinity rather than ELISA-based avidity, we measured the binding kinetics of the Abs to gp120 by biolayer light interferometry (BLI; I, integrated intensity; σI, estimated SD of that intensity. *R sym = (Σ hkl Σ i jI i (hkl) − <I(hkl)>)/Σ hkl Σκ i (hkl), where I i (hkl) is the intensity of the ith measurement of reflection (hkl) and <I(hkl)> is the average intensity. † R meas = (Σ hkl (sqrt(N hkl /(N hkl − 1))Σ i jI i (hkl) − <I(hkl)>)/Σ hkl Σk i (hkl), where I i (hkl) is the intensity of the ith measurement of reflection (hkl) and <I(hkl)> is the average intensity. ‡ R cryst = (Σ hkl jFo − Fcj/Σ hkl Fo), where Fo and Fc are the observed and calculated structure factors, respectively. § R free is calculated as for R work , but from a randomly selected subset of the data (5%) that were excluded from the refinement (49). ¶ Chen et al. (50). Fig. 2A). Here, we found effects with much larger magnitude that were again focused around the tip of HCDR3. The kinetics revealed that the decreased affinity to gp120 was largely due to the rapid off-rate of the mutant mAbs ( Fig. 2B and Fig. S3).
To assess the impact of these substitutions regarding neutralizing capacity, we performed virus neutralization comparing WT GE148 with the HC Ala mutants ( Fig. 2A, right columns). GE148 IgG neutralizes HXBc2 relatively potently (11) and, in this instance, with an IC 50 of 0.17 μg/mL. Similar to the broader range of residues affected in the low-affinity 2CC ELISA scan, neutralization of the HXBc2 virus was negatively influenced by multiple Ala substitutions in HCDR3. We then performed an additional scan using the SF162 virus, which GE148 neutralizes (11) but with a slightly lower IC 50 potency of 0.62 μg/mL. Again, we detected several Ala substitution residues clustered at the tip of HCDR3 that decreased SF162 neutralizing capacity in a pattern similar to the 2CC or HXBc2 scan. Analysis in the context of neutralization of the MN virus produced similar results. The GE148 LCDRs (Fig. 2C) were similarly scanned, and, again, more effects with respect to the low-affinity 2CC scan were detected compared with full-length or core gp120 scan by ELISA, roughly paralleling the affinity scan of gp120 by BLI. The binding kinetics confirmed that the decreased affinity was largely due to more rapid off-rates (Fig. S3). Regarding neutralization, we saw more disperse effects in the GE148 LC compared with the HC, and these effects were of greatest magnitude on the C-terminal side of LCDR1, at the base of LCDR2, and in the N-terminal region of LCDR3. These data suggest that when non-bNAbs, such as GE148, are able to access their epitopes on the functional virus spike of tier 1 viruses, they achieve effective virus neutralization and that such a productive interaction with the Env spike is largely determined by mAb off-rate.
Modeling of GE148 and gp120 Interactions. Further insight into the binding interaction between GE148 and Env was gained through computer modeling. The GE148 Fab crystal structure was docked onto the gp120 structure derived from the F105-bound core gp120 (PDB ID code 3HI1; hereafter referred to as "core gp120") using ClusPro 2.0. The best-fit model was then obtained using the available data we had from the Ala scan of the GE148 paratope ( Fig. 2) and epitope on gp120 (11). In Fig. 3A (Left), the residues in the GE148 CDRs that had the greatest impact on neutralization of the three viruses tested when substituted to Ala are highlighted in red. The structure of core gp120 is presented in Fig. 3A (Right), with the CD4 loop highlighted in yellow, where the Ala substitutions that had the greatest effects on GE148 recognition were located (11), including residues D368 and E370, which are critical recognition determinants of many CD4bs mAbs. As demonstrated in Fig. 3B, the best-fit model shows a docking orientation that is consistent with proximal interactions of the GE148 CDRs with the gp120 CD4 loop. We also performed a similar docking analysis of GE148 but used the gp120 derived from the b13-bound conformation (PDB ID code 3IDX) instead, which resulted in a convergent solution that closely matched the F105-bound model (Fig. S4).
Interestingly, the model of GE148 docked to the F105-bound conformation of gp120 (Fig. 3C, Left) predicts that the K100e residue on the C-terminal side of the HCDR3 loop interacts with D368 and E370 of the gp120 CD4 loop, potentially involving hydrogen bonding (Fig. 3C, Right). Such an interaction is not surprising, because the published GE Abs were selected as CD4bs-directed B cells in the original flow cytometric single-cell sorting approach based upon a D368R substitution. In addition, the modeling suggests interaction of F100b with the hydrophobic phenylalanine (Phe)-43 cavity on the gp120 molecular surface. The fitting of both of these residues is reminiscent of the CD4 interaction with gp120 defined by crystallography, in which the R59 of CD4 interacts with D368 and the F43 of CD4 interacts with the so-named "Phe-43 cavity" (25). Similar structural interactions to that of CD4 have been described in previous studies for several CD4bs-directed mAbs (22).
To confirm the accuracy of the model, we performed sitedirected mutagenesis of residues predicted to interact between the interface of the paratope of GE148 and the epitope of gp120. We substituted K100eE in the GE148 HCDR3 and observed that this nonconservative change completely eliminated neutralization of HIV-1, consistent with the modeling (Fig. 3D). Similarly, altering F100bW, which is predicted to create a clash in the Phe-43 cavity of gp120, also reduced GE148 neutralization capacity. Next, we attempted to increase neutralization potency by model-guided "gain-of-function" substitutions to increase hydrophobic interactions between GE148 HCDR2 and HCDR3 with hydrophobic elements of gp120. As predicted by the model, enhanced neutralization capacity was observed for changes at S53I/W, V99I, and T100cW (Fig. 3D). These data strongly suggest that the GE148/gp120 complex model provides structural information at a relatively high level of resolution that can then be exploited to both decrease and enhance the functional capacity of the mAb.
Ala Scan, Modeling of GE136, and Validation with F105. We next performed a similar modeling analysis using the Ala scan data of GE136 (Fig. 4A) in conjunction with the published epitope scan of the footprint of this CD4bs-directed mAb on gp120 (11) to dock GE136 with core gp120 (Fig. 4B). A convergent solution was also seen when the alternative gp120 core (PDB ID code 3IDX) was used (Fig. S4). The model revealed potential interactions of GE136 K100e with D368 of gp120, a hydrogen bond between GE136 V100d and E370 of gp120, and a hydrophobic interaction of GE136 L100b with the gp120 Phe-43 cavity (Fig.  4C). Consistent with the modeling, substitution of these residues to Ala (Fig. 4A) and K100eE (Fig. 4D) resulted in a loss of neutralization capacity. To confirm the accuracy of the model further, we designed substitutions to increase hydrophobic interactions between the GE136 HCDR2 and HCDR3 with gp120 at residues S54, L99, and L100b as predicted by the model, most of which did result in gains in GE136 neutralization potency (Fig. 4D). Although substitutions with Phe at L99 and L100b increased GE136 neutralization potency, substitutions with tryptophan (Trp) did not. This may be due to the bulkiness of the Trp indole group compared with Phe benzyl group, which may not fit in the gp120 hydrophobic pocket due to steric hindrance.
To validate the modeling methodology using an existing structure of an Ab/Env ligand complex, we independently performed the docking and modeling by the same methods and constraints with the human CD4bs-directed mAb F105, using the Ala scan of the paratope (Fig. S5A) and epitope on gp120 (11) to obtain the best-fit model, which is remarkably close to the crystal structure in terms of the contacts between the Ab CDRs and gp120 (Fig. S5B). We also compared the ClusPro model of the human CD4bs mAb b13 in complex with gp120 to the b13/ gp120 crystal structure, with similar results (Fig. S5C).
EM Analysis of GE136 and GE148 with gp120. To assess the interactions of GE148 and GE136 with gp120 by direct structural analysis further, we performed EM studies of these Abs interacting with gp120. Initially, we generated equimolar mixtures of both GE136 and GE148 Fabs in complex with monomeric full-length gp120. Following negative staining, EM images were generated and analyzed. Reference-free 2D classification images were selected from the primary dataset to generate 3D reconstructions of these Fabs in complex with gp120 ( Fig. S6 A and B). Fitting of the Fab crystal structures of GE136 or GE148 with the core gp120 was performed, which generated good fits with the density. Fitting of the GE136/gp120 and GE148/gp120 into these densities was consistent with the ClusPro modeling of the Abs docked to gp120.
To constrain the fits within the density further and to derive an orientation of the Abs and core gp120 in the context of the trimeric spike, we added 2G12 to the analysis to provide a distinct landmark on gp120. As a glycan-reactive bNAb, 2G12 binds to at least three high-mannose glycans on the gp120 outer domain surface, and the 2G12 crystal structure in complex with glycan is available (PDB ID code 1OM3) (26,27). The 2G12 Ab is unusual in that it binds via domain exchange, resulting in an IgG with the two Fab regions tightly cross-linked. This domain swap creates a unique multivalent surface compared with a conventional bivalent IgG, limiting 2G12 IgG binding to only one gp120 monomer. Negative-stain, reference-free 2D class averages of 2G12 generated a distinctive double-lobed pattern relative to the GE136 Fab (Fig. 5A, Upper Left), aiding in orientation by providing a distinctive feature for the EM analysis. Using this landmark and the fitting of the mAbs with gp120, we were able to assign individual elements of the 2D classification images of the complex to 2G12, core gp120, and GE136 (Fig. 5A, Upper Right). Manual docking Fig. 3. Docking models of GE148 onto gp120 and confirmatory substitutions. (A, Left) Variable regions of GE148 HC (VH, dark green) and LC (VL, light green) with key residues of the paratope shown in red (i.e., residues when mutated to Ala severely decreased Ab function), with their side chains shown as a stick representation. (A, Right) Ribbon representation of core gp120 (PDB ID code 3HI1) with key features labeled. The amino acids within the gp120 CD4 loop that resulted in decreased Ab binding when altered in the gp120 Ala scan (11) are shown in yellow. (B) ClusPro model of the GE148 and core gp120 complex showing convergence of the Ab paratope CDRs and the gp120 CD4 loop. (C) ClusPro model in the ribbon representation (Left) and a close-up view at the Ab interaction with the CD4bs loop (Right). HCDR3 residue K100e (red) is predicted to form hydrogen bonds (black dashed lines) with D368 and E370 on gp120 (yellow). F100b at the tip of the HCDR3 is positioned in proximity to the gp120 Phe-43 pocket in the CD4bs, resulting in a putative hydrophobic interaction. (D) Confirmatory Ab mutations based on the ClusPro docking model were assayed for neutralization potency against the HIV-1 isolate SF162. K100e was predicted to abrogate Ab neutralization when mutated to E, whereas S53W, V99I, and T100cW were predicted to interact with hydrophobic elements of gp120 (shown in tan on the molecular surface of core gp120). Fold changes compared with WT with enhanced activity (>+threefold) are highlighted in green, and those with decreased activity (>−threefold) are highlighted in red.
of the crystal structures of these proteins (PDB ID codes 1OM3, 3HI1, and 4KTD, respectively) to form a complex was consistent with the EM images and assignments (Fig. 5A, Lower). Using the 2D-classified images, we generated a 3D density volume and docked GE136, core gp120, and 2G12 into this density of the complex (Fig. 5B, Upper). The fitting of the 2G12 Fab and gp120 coordinates into the density was consistent with an interaction of this mAb with the gp120 outer domain glycans that comprise its epitope ( Fig. 5B; highlighted in red on gp120). Similarly, the GE136 Fab was positioned to interact with its cognate epitope on the CD4bs (Fig. 5B; highlighted in yellow). We then positioned the 2G12/gp120/GE136 tricomplex above the published cryo-EM density of the functional spike [Electron Microscopy Data Bank (EMDB) ID 5019] (17), using the core gp120 molecular surface, the visible 2G12 epitope N-glycan, and the CD4 loop as landmarks (Fig. 5B). This analysis provides approximate angles of approach for the spike by both 2G12 and GE136 (Fig. 5B, arrows) and suggests that GE136 would have to approach the CD4bs from the top of the spike, which we examine in greater detail below. We performed a similar EM analysis on GE148 with core gp120 and 2G12 with an analogous outcome (Fig. S6C). The EM 3D reconstructions shown were generated at a resolution of ∼20 Å. Projection matching and Fourier shell correlation for 2G12/gp120/GE136 tricomplex are shown in Fig. S7A.
Angle of Ab Approach to the CD4bs of the Functional Spike. To define the orientations of the vaccine-induced mAbs with the gp120 CD4bs in the context of the functional spike further, we performed a similar analysis as that described above but now using the N332directed bNAb PGT122 (28) Fab instead of the 2G12 IgG in tricomplex with gp120 and the GE136 or GE148 Fab. Because the interaction of PGT122 with the soluble BG505 SOSIP.664 gp140-cleaved trimeric spike mimetic was previously assessed by EM (EMDB ID 5624) (29), and although the non-bNAbs GE136 and GE148 cannot bind to the functional spike of primary isolates such as BG505, we nevertheless were able to approximate the theoretical positions of the PGT122/gp120/GE136 and PGT122/gp120/GE148 tricomplexes in the context of the BG505 SOSIP.664 using the PGT122 Fab density (derived from EMDB ID 5624) as a feature in common (Fig. 5C, Left and Center). Similarly, we docked PGT122/gp120/PGV04 onto the BG505 trimeric spike (Fig. 5C, Right) as a control and for comparison. Because the EM densities of both PGT122/BG505 SOSIP.664 and PGV04/KNH1144 SOSIP.664 (EMDB ID 5706) are available (29,30), we could confirm the accuracy of the tricomplex docking on the trimeric spike. Projection matching and Fourier shell correlations for the tricomplexes are presented in Fig. S7 B-D. This analysis revealed that GE136 and GE148 attempt to access their respective CD4bs epitopes from "the top of the spike" relative to the lateral, "side approach" of PGV04 (Fig. 5C, side view and Fig.  S8A, top view), which may be why GE136 and GE148 cannot bind the SOSIP trimers or cannot neutralize primary HIV-1 isolates.
The analyses described above indicated that access from the top might be a general property of nonbroadly neutralizing CD4bs-directed mAbs to tier 1 isolates. To examine if binding orientation to the functional Env spike discriminates between  Fig. 4. GE136 Ala scans, modeling, and confirmatory substitutions. (A) Binding and neutralization activity of Ala substitution mutants in GE136 HCDR2 and HCDR3 (Kabat numbering) are shown compared with WT (blue). Decreases in activity compared with WT are highlighted (four-to 10-fold, beige; 10-to 20fold, yellow; >20-fold or >50 μg/mL, red). (B) ClusPro model of GE136 with HC (dark brown) and LC (light brown) in complex with core gp120 (colored as in Fig.  3A). Specific HCDR mutations that decreased Ab function in the Ala scan are shown in red. (C) Magnified view of potential molecular interactions between GE136 and the gp120 CD4bs loop. Potential hydrogen-bonding interactions between residues in GE136 and gp120 are indicated with black dashed lines. GE136 L100b (red), at the tip of the HCDR3, is another important residue that may contribute to the binding energy of the complex by a hydrophobic interaction. (D) GE136 mutants with substitutions in HCDR2 or HCDR3 predicted to enhance or decrease GE136-gp120 interactions were assayed for neutralization potency against the HIV-1 SF162 isolate. Fold changes compared with WT are shown, with enhancements greater than threefold highlighted in green and decreases greater than threefold highlighted in red.
nonbroadly and broadly neutralizing CD4bs-directed Abs, whether elicited by chronic HIV-1 infection or Env vaccination, we used the coordinates of core gp120 fitted into the cryoelectron tomography density of the functional spike (EMDB ID 5019) (17) and superimposed the ClusPro models of the GE136 and GE148 Fab/gp120 dicomplexes onto the gp120 trimer (PDB ID code 3DNN) fitted in the density (Fig. 6A). To assess the relative angles of approach for other CD4bs-directed mAbs, we performed a similar analysis using available coordinates and structures of CD4bs-directed mAbs in complex with the gp120 core ( Fig. 6A and Fig. S8B). As can clearly be seen, the vaccine-or infection-induced non-bNAbs approach their respective epitopes in the CD4bs from the top of the spike. The modeling suggests that the non-bNAbs require a deeper penetration toward the trimer axis, which is evident in the top views of GE136, GE148, and F105, where the non-bNAbs protrude further into the V-region cap area toward the trimer axis, as outlined by the black dashed line (Fig. 6A, Lower). This likely creates clashes (red starbursts in Fig. 6 and Fig. S8) with the V-region cap of neutralization-resistant viruses, thereby preventing mAb binding to the spike and subsequent neutralization of the virus. Probably contributing to these clashes are the N-linked glycans on the surface of the trimeric spike emanating from the V1/V2/V3 loops, which are not visible in the density. Such a route of access to the spike may be achievable on the more neutralizationsensitive viruses that likely possess less tightly packed trimeric spikes but would be hampered against the more tightly packed spikes present on primary isolates. In contrast, the broadly neutralizing CD4bs-directed mAbs approach their epitopes in the functional spike via a more lateral "side access" (18), avoiding clashes with the quaternary topside cap of the functional spike. To clarify such clashes better, interaction of GE148 with the functional spike is compared with the bNAb, VRC01. In Fig. 6B, GE148 (ClusPro model with gp120) and VRC01 are docked into the functional spike (EMDB ID 5019) in a similar fashion as in Fig.  6A. On the right, the spike is rotated 45°from the previous orientation on the y axis and reveals substantial clashes of the GE148 HC and LC with cap elements of the trimeric spike, as highlighted. In contrast, VRC01 approaches its epitope in the CD4bs laterally, bisecting the wedges defined by two adjacent protomer arms of the spike, thus avoiding clashes with either the V-region cap or adjacent protomers. A similar analysis was performed with GE148 and CH103 with comparable results (Fig.  S8B), indicating that although CH103 binds slightly more proximal to the V-region cap of the native spike, access is still achievable as long as the bNAb "splits the wedge" by stacking the HC and LC in a vertical manner to minimize width and clashes as it approaches the CD4bs. Comparison of GE136 access relative to VRC01 yielded an analogous result as GE148 (Fig. S8C).
These results prompted us to investigate the structural conformation of the YU2 gp140-F trimers used to elicit the GE136 and GE148 mAbs by EM. As seen in Fig. 6C, these trimers appear relatively heterogeneous in composition, indicating that they are likely open in their conformation compared with more faithful mimetics of the native spike, such as the BG505 SOSIP trimers (31). Collectively, the modeling and EM studies suggest a distinct difference in terms of angle of approach to the CD4bs between bNAbs and the class of non-bNAbs typified by the two mAbs studied here and suggest a strategy to modify current soluble trimeric immunogens to restrict access of B-cell receptors (BCRs) toward a "side angle" approach to the CD4bs while occluding a vertical approach, perhaps by strengthening interactions in the gp120 trimer association domains. Such an immunogen design strategy might better elicit neutralizing Abs to this recessed and important functionally conserved neutralizing determinant on the HIV-1 spike.

Discussion
Recently, we reported the isolation of previously undescribed Env trimer vaccine-elicited, CD4bs-directed mAbs from rhesus macaques, including GE136 and GE148, which neutralize tier 1 HIV-1 strains relatively potently (11). Similar to the majority of (blue), gp120 (cyan), and GE136 (brown). (Lower) Surface representation of the 2G12 crystal structure (PDB ID code 1OM3, blue) and the ClusPro model of GE136 (brown) bound to core gp120 (cyan) modeled following EM data is shown. The patch of red in gp120 depicts the N332 and N295 glycan sites important for recognition by 2G12. (B) EM density of the gp120/2G12/GE136 complex filled with ribbon representations. GE136 (brown), 2G12 (blue), and gp120 core (cyan) are shown with the CD4bs (yellow) and the glycan sites, N295 and N332 (red). Superimposition of the GE136 tricomplex with the unliganded trimeric gp120 from published electron tomography data of the native BaL HIV viral spike to approximate the angle of Ab access (2G12, blue arrow; GE136, brown arrow) is illustrated. (C) EM density data of the broadly neutralizing antiglycan Ab PGT122 bound to the SOSIP soluble trimer (EMDB ID 5624) are shown in light yellow with the theoretical superimposition of the EM density (gray) of gp120/PGT122/GE136 (Left), gp120/PGT122/GE148 (Center; note that GE136 and GE148 do not bind to the SOSIP trimers; hence, these are theoretical superimpositions), and gp120/PGT122/PGV04 (Right) using the common PGT122 density. A ribbon representation of PGT122 (PDB ID code 4JY5, red) GE136 (brown), GE148 (green), and PGV04 (pink), with the molecular surface of gp120 core (cyan), is shown filling the EM data for identification of density.
HIV-1 infection-induced CD4bs-directed mAbs, these vaccineelicited mAbs do not neutralize the more resistant tier 2 HIV-1 isolates. To investigate the structural basis for the limited neutralization by these mAbs sharing this CD4bs-directed, nonbroadly neutralizing phenotype, we solved the high-resolution structures of the GE136 and GE148 Fabs and performed a detailed examination of the potential mode of interaction of these mAbs with the Env spike, including Ab paratope and epitope mapping, in silico docking of the mAbs to gp120, and EM structural analysis with gp120. These data provide a clear explanation for the shortcomings of the nonbroadly neutralizing CD4bs-directed Abs elicited by the foldon Env trimer designs analyzed here and may provide structural insight for rational immunogen modification to improve the elicitation of bNAbs against the conserved HIV-1 primary receptor binding site. One limitation of the current analysis is that it directly reveals the angle of approach used by CD4bs mAbs generated by the foldon trimers. However, given the shared inability of many CD4bsdirected Abs elicited by the foldon trimers or other forms of Env to neutralize primary isolates, this congruence suggests that many Env immunogens that do not elicit bNAbs also generate CD4bs-directed mAbs that take similar suboptimal angles of approach to this conserved neutralization determinant. The crystal structures of GE136 and GE148 revealed modestly protruding HCDR3s with hydrophobic tips, a property that is shared by other nonbroad CD4bs-directed mAbs but not by is indicated in degrees relative to a fully horizontal approach, parallel to the viral membrane. As can be seen, the non-bNAbs attempt to access the CD4bs vertically relative to the viral membrane, whereas the bNAbs more closely approximate a horizontal path.
(Upper) Red starburst indicates clashes with the primary isolate spike density V-region cap if a more vertical angle of approach is attempted. The black dashed outline marks protrusion of the mAb into the V-region cap density. Note that in this rendering, docking of the bNAbs into the unliganded spike density, b12 assumes a more vertical orientation relative to the b12-liganded orientation described by Tran et al. (18), which can be attributed to b12-induced conformational changes of the spike, altering the angle of binding relative to VRC01 and the viral membrane. (B, Left) VRC01 and GE148 docked into the spike density (EMDB ID 5019) in the same orientation as in A. (B, Right) VRC01 and GE148 docking rotated ∼45°to reveal substantial clashes of GE148 HC and LC with protomer proximal V cap elements of the trimeric primary isolate spike, as indicated by the red starburst, but few clashes for the bNAb. (C, Left) Representative reference-free 2D class averages of the YU2 gp140-F antigens that were used to inoculate the NHPs. The irregular shapes suggest that these foldon trimers are relatively open in configuration. (C, Right) Although held together by gp41 determinants and the foldon motif, nonneutralizing epitopes (arrows) are exposed and available for Ab recognition. Conversely, a native trimer provides constraints on Ab approach.
bNAbs. Ala scanning of GE136 and GE148 indicated that their HCDR3s harbor critical contacts for binding and neutralization, which is quite different from the bNAbs, such as VRC01 and PGV04, for which the HCDR3s form markedly fewer contacts with gp120 (4, 23, 32) but more in HCDR2 and framework regions (33). Many more effects on binding of GE148 were revealed by the lower affinity interaction of this mAb with the stabilized gp120 core protein, 2CC, suggesting that the effect of single Ala substitutions is more apparent if the mAb-gp120 interaction is of lower affinity. Interestingly, the lower affinity recognition of GE148 for 2CC, and the resulting Ala scan recognition profile of 2CC by ELISA, more closely mapped to the residues on GE148 affected by Ala substitutions in the context of neutralization than did a similar scan on gp120. This result may relate to more inherent flexibility in monomeric gp120 than to the stabilized core and gp120 in the context of the native Env trimer. These data suggest that HIV-1 neutralization may be a "threshold" event requiring mAb access to its epitope in the context of the functional spike, and that, if achieved, the mAb must maintain this interaction long enough to inhibit virus entry. This interpretation is consistent with kinetic binding data with GE148, where decreases in affinities of the Ala substitutions were determined mainly by their off-rate to gp120. The data are also consistent with the general model that if an Ab neutralizes HIV-1, it must be able to bind to the functional viral spike glycoprotein with a reasonable efficiency and occupancy, as previously suggested (34). For HIV-1 CD4bs-directed mAbs, such interactions appear to be "all or none"; that is, if an Ab cannot access its spike due to steric occlusion, quaternary packing, or conformational masking, there is no detectable on-rate and essentially no binding or neutralization. This is opposed to a model in which the CD4bs-directed mAb can access its epitope on the primary isolate spike with a detectable on-rate but cannot induce conformational changes required for high-affinity binding to mediate neutralization. Such a two-step mechanism of binding has been proposed for CD4 interaction with HIV Env (35). The interactions defined from a combination of mutagenesis, crystallography, EM, and modeling suggest that the hydrophobic HCDR3s of these mAbs protrude into the Phe-43 cavity and involve interaction of hydrophobic residues on both mAbs with this cavity. The lack of such hydrophobicity in the HCDR3s of other broadly effective mAbs, b12 and CH103, indicates that this property is associated with non-bNAbs, although their HCDR3s interact in different ways. This analysis suggests that perhaps deeper penetration by the HCDR3s of non-bNAbs into this pocket or neighboring regions not available on primary HIV-1 isolates limits neutralization. The structural models also suggest that a basic residue in the HCDR3s of each mAb interacts with D368 or E370 in the gp120 CD4 binding loop, reminiscent of similar cationic + hydrophobic motifs in F105 and b13 (22). The GE148 mAb LC Ala scan revealed that many putative interactions outside of the HCDR3 were involved in neutralization, suggesting this charge-plus-hydrophobic motif may be necessary, but not sufficient, for broad CD4bs-directed HIV-1 neutralization.
Recognition of the primary isolate functional spike by the GE136 and GE148 appears limited, as reflected by their modest neutralization capacity. The structural analysis, coupled with the neutralization data, suggests that these vaccine-elicited, CD4bsdirected mAbs adopt a mode of recognition involving an approach to the spike from the trimer apex to attempt access of their respective epitopes in the CD4bs. This mode of binding, reminiscent of that of F105 (22), contrasts to that used by the bNAbs VRC01 and PGV04, which use a side approach, parallel to the viral membrane, to "duck under" the quaternary V-region cap effectively to access their epitopes in the CD4bs efficiently. CD4bsdirected bNAbs, such as b12 and CH103, which are more limited in neutralization breadth relative to VRC01 or PGV04, adopt a slightly more vertical mode of recognition, invoking clashes with the V-region cap on viruses that they cannot neutralize (36).
For both the vaccine-and infection-elicited non-bNAbs analyzed in this study, it appears that steric clashes with the V-region cap limit their neutralizing capacity against the tight quaternary packing of most primary isolates. That the CD4bs-directed non-bNAbs appear relatively commonly elicited by the soluble trimer immunogens used here suggests that these spike mimetics are open in conformation, permitting unfettered top access of B cells recognizing the CD4bs, resulting in the generation of non-bNAbs. Efforts to redesign the soluble foldon trimers to improve the structural integrity of the V-region cap are warranted. Trimers displaying improved quaternary packing, such as soluble SOSIP trimers, may better elicit Abs to the conserved primary receptor binding site by restricting BCR access to a side approach to the recessed CD4bs. However, more open trimers might still have a useful role in priming B-cell responses to the CD4bs that is recessed on more native spikes. In sum, the present studies offer important insight on how vaccine-induced, CD4bs-directed mAbs interact with Env, providing a clear explanation for a long-standing problem in the HIV-1 vaccine field (37)(38)(39) that can now be rationally addressed through improved soluble trimer redesign to accelerate the development of an effective HIV-1 vaccine.

Materials and Methods
Experimental methods used in this study are briefly summarized here. Additional details are presented in SI Materials and Methods.
Fab and Core gp120 Expression and Purification. Similar procedures were used as described in previous studies (11,24), and additional details can be found in SI Materials and Methods.
Crystallization and Structural Determination of the GE136 and GE148 Fabs. Initial crystallization trials in 384 conditions [Joint Center for Structural Genomics (JCSG) core I-IV] were set up using the automated IAVI/JCSG/TSRI CrystalMation robotic system (Rigaku) at the JCSG (www.jcsg.org). Diffraction-quality crystals of the GE136 Fab fragment could be grown in a solution composed of 0.1 M citric acid (pH 3.5) and 1.6 M ammonium sulfate (pH 4.0). For the GE148 Fab fragment, diffraction-quality crystals were obtained in 0.17 M ammonium sulfate, 25.5% (wt/vol) PEG 4000, and 15% (vol/vol) glycerol. Diffraction data for the GE136 and GE148 Fab fragment were collected on beamline (BL) 23-ID-D at the Advanced Photon Source and on BL 11-1 at the Stanford Synchrotron Radiation Lightsource, respectively. The diffraction data were integrated and scaled using the XDS program package in space group P2 1 2 1 2 1 . The initial model of GE148 and GE136 was solved and built by the automated crystal structure determination platform Auto-Rickshaw (40), using the Fab HC and LC from PDB ID code 3G6D in the case of GE148 and the Fab structure of GE148 for GE136. Auto-Rickshaw returned models having about 95% of the model built by ARP/wARP. Final refinements were made by using Phenix refine, applying the TLS (translation/libration/screw-motion) method (41), and using manual rebuilding with Coot (42). Refinement statistics can be found in Table 1.
Ala Scanning, Binding, and Neutralization Assays. Similar procedures were used as described in previous studies (11,19), and additional details can be found in SI Materials and Methods.
Modeling of NHP Ab Structures in Complex with gp120 Core. The ClusPro server 2.0 (16) was used in the Ab mode to produce structural models of the NHP Abs GE148 and GE136 in complex with previously determined crystal structures of core gp120. Published structures of the two non-bNAbs F105 and b13 (PDB ID codes 3HI1 and 3IDX) were used based on the criteria that these mAbs displayed a similar binding and neutralizing profile as the NHP Abs. To input the gp120 core structure in the ClusPro server, a separate PDB file was created for each core gp120 from published structures, removing the human mAbs to expose the CD4bs on the core molecular surface. The high-resolution models of the complexes resulting from pairing each NHP mAb and the two gp120 core structures were obtained based primarily upon the available Ala scan data that we had determined experimentally. To confirm the accuracy of the ClusPro analysis, models of GE148 and GE136 in complex with the gp120 core were used to identify paratope positions that, upon specific