Boosting with AIDSVAX B/E Enhances Env Constant Region 1 and 2 Antibody-Dependent Cellular Cytotoxicity Breadth and Potency

Over one million people become infected with HIV-1 each year, making the development of an efficacious HIV-1 vaccine an important unmet medical need. The RV144 human HIV-1 vaccine regimen is the only HIV-1 clinical trial to date to demonstrate vaccine efficacy. An area of focus has been on identifying ways by which to improve upon RV144 vaccine efficacy. The RV305 HIV-1 vaccine regimen was a follow-up boost of RV144 vaccine recipients that occurred 6 to 8 years after the conclusion of RV144. Our study focused on the effect of delayed boosting in humans on the vaccine-induced Env constant region 1 and 2 (C1C2)-specific antibody repertoire. It was found that boosting with an HIV-1 Env vaccine increased C1C2-specific antibody-dependent cellular cytotoxicity potency and breadth.

antibodies that used VH1 family genes (12). A majority of the MAbs assayed in this study bound epitopes distinct from A32 but in close enough proximity to be sterically cross-blocked by A32 (Fig. 1B).
Boosting increased C1C2-specific ADCC breadth and potency. RV305 C1C2specific MAbs and a subset of RV144 C1C2-specific MAbs were next assessed for ADCC against CEM.NKR CCR5 cells infected with one of seven HIV-1 infectious molecular clones (IMCs), representing three different clades (HIV-1 AE.CM235, B.WITO, C.TV-1. C.MW965, C.1086C, C.DU151, and C.DU422) ( Table 2). These IMCs were chosen because they represent clusters of IMCs with different sensitivity to ADCC (unpublished data). To ascertain ADCC breadth and potency, MAbs were ranked using an ADCC score (see Materials and Methods) analogous to calculating MAb neutralization breadth. Apart from the RV144-derived A32 blockable MAb CH38, which was naturally an IgA MAb but tested here as a recombinant IgG1 MAb, ADCC scores for 16/19 RV305 MAbs ranked higher than those of the RV144 MAbs (Tables 3 and 4).
Boosting of RV144 vaccine recipients with AIDSVAX B/E in the RV305 trial increased ADCC breadth and potency of the RV144-derived C1C2-specific, DH677 clonal lineage. The C1C2-specific DH677 memory B cell clonal lineage next was used to study affinity maturation and ontogeny of AIDSVAX B/E-induced ADCC responses. B cell clonal lineage member DH677.1 was isolated from PBMCs collected from a vaccine recipient 2 weeks after the last boost in the original RV144 trial (ALVAC plus AIDSVAX B/E). DH677.2, DH677.3, and DH677.4 clonal lineage members were isolated from PBMCs collected from the same vaccine recipient 2 weeks after the second boost with AIDSVAX B/E alone given in the RV305 clinical trial (RV305 group II). Thus, this B cell clonal lineage belongs to a long-lived memory B cell pool started by the RV144 vaccine regimen and boosted many years later with the RV305 vaccine regimen (Fig. 2). DH677.1, DH677.2, DH677.3, and DH677.4 MAb sequences were used to infer with  Cloanalyst (13) three intermediate ancestors (IA), IA1, IA2, and IA3, and an unmutated common ancestor (UCA) for the DH677 clonal lineage. The DH677 clonal lineage then was assayed by surface plasmon resonance for binding to the AIDSVAX B/E proteins, AE.A244gp120Δ11 and B.MNgp120Δ11, as well as full-length AE.A244gp120. The DH677 UCA did not bind to B.MNgp120Δ11 and had minimal binding to the full-length AE.A244gp120, and this binding was enhanced with AE.A244gp120Δ11 (Fig. 2). The RV305 boosts more than doubled the V H chain gene mutation frequency from 1.04% (DH677.1; RV144) up to 4.51% (DH677.4; RV305), which resulted in a 100-fold increase in apparent affinity ( (Fig. 2). The ontogeny of vaccine-induced ADCC was studied by assaying the DH677 clonal lineage against a panel of IMC-infected CEM.NKR CCR5 cells (AE.CM235, B.WITO, C.TV-1, C.MW965, C.1086C, C.DU151, and C.DU422) ( Table 2). The RV144 prime-boost immunization regimen minimally increased ADCC breadth and potency (DH677 UCA ADCC score, Ϫ2.32; DH677.1 ADCC score, Ϫ2.20 [see Materials and Methods]). Conversely, RV305 boosting substantially increased ADCC breadth and potency (DH677.3 ADCC score, 4.56) (Fig. 2). These data indicate that the RV144 prime-boost regimen was insufficient to fully affinity mature this C1C2-specific B cell clonal lineage. Rather, RV305 trial boosting of this particular RV144 vaccine recipient profoundly enhanced DH677 lineage ADCC breadth and potency.
Crystal structure of the potent ADCC-mediating MAb DH677.3. We next determined the crystal structure of the antigen binding fragment (Fab) of the highestranking RV305 ADCC MAb, DH677.3 (Table 4), alone and in complex with clade AE gp120 93TH057 core e plus the CD4-mimetic M48-U1 ( Fig. 3 and Table 5). DH677.3 Fab-gp120 93TH057 core e -M48U1 complex (Fig. 4) showed that, similar to other cluster A MAbs, DH677.3 approaches gp120 at the face that is buried in the native Env trimer (3)(4)(5) and binds the C1C2 region exclusively within the gp120 inner domain. The gp120 residues involved in DH677.3 binding map to the base of the 7-stranded ␤-sandwich (residues 82, 84, 86 to 87, 222 to 224, 244 to 246, and 491 to 492) and extend into the mobile layers 1 (residues 53, 60, and 70 to 80) and 2 (residues 218 to 221). By docking at the layer 1/2/␤-sandwich junction, the Fab buried surface area (BSA) utilizes 248 Å 2 of the ␤-sandwich, 542 Å of layer 1, and 135 Å 2 of layer 2 ( Fig. 4 and Table 5). The majority of contacts providing specificity involve a network of hydrogen bonds and a salt bridge (Fig. 4, inset) contributed by the antibody heavy-chain and gp120 side-chain atoms of layer 1 (␣ turn connecting the ␤1-␤0 strands, D 78 and N 80 ) and the 7-stranded ␤-sandwich (strand ␤7, Q 246 ). The contacts provided by the light chain are less specific and consist of hydrogen bonds to the gp120 main-chain atoms and hydrophobic contacts within a hydrophobic cleft formed at the layer 1/2/␤-sandwich junction ( Fig.  4B and C). Overall, DH677.3 utilizes all six of its complementary determining regions (CDRs) and relies approximately equally on both heavy chain and light chain with a total BSA of 973 Å 2 , 498 Å 2 for the light chain, and 475 Å 2 for the heavy chain ( Fig. 5 and Table 5). Interestingly, 25 of 29 gp120 contact residues are conserved in Ͼ80% of sequences in the HIV Sequence Database Compendium (https://www.hiv.lanl.gov/ content/sequence/HIV/COMPENDIUM/compendium.html), with 15 of 29 being effectively invariant (Ͼ99% conserved) (Fig. 4B).
Comparison of the DH677.3 mode of binding and epitope footprint to cluster A prototype MAbs. Antigen complex structures of A32 and N12-i3 (C11-like) (3, 6), MAbs isolated from HIV-1-infected individuals, confirm that DH677.3 recognized a unique epitope between the A32 and C11 antibody-binding sites involving Env epitope elements of both (Fig. 6). While the A32 MAb epitope consists exclusively of gp120 mobile layers 1 and 2 (76% and 24% of gp120 BSA, respectively; Fig. 6 and Table 5), DH677.3 relies less on layers 1 and 2 (53% and 14% of gp120 BSA, respectively) and effectively utilizes the gp120 7-stranded ␤-sandwich (24% of gp120 BSA) ( Fig. 6 and Table 5). The ability to recognize the 7-stranded ␤-sandwich renders DH677.3 similar to the C11-like antibody N12-i3, which almost exclusively depends on the ␤-sandwich for binding (94% of its total gp120 BSA; Fig. 6 and Table 5). Interestingly, N12-i3 and other C11-like MAbs require the N terminus of gp120 for binding and recognize a unique gp120 conformation formed by docking of the gp120 N terminus as an 8th strand to the ␤-sandwich to form an 8-stranded-␤-sandwich structure (6). The DH677.3 complex crystals were obtained with gp120 93TH057 core e , which lacks the N terminus (11-aa deletion), and therefore a direct judgment, based on structure, of whether or not the 8th strand is involved in binding was not possible. However, we were able to model the N/C termini-gp120 93TH057 core e from the N12-i3 Fab complex structure (PDB code 5W4L) to the DH677.3 Fab-gp120 93TH05 7 core e -M48U1 complex without any steric clashes (Fig. 6A, inset). Both the conformation and orientation of CDR H1 and H2 of DH677.3 allowed easy access to the 8-stranded ␤-sandwich structure and enabled contacts to the 8th strand. These data indicated that DH677.3 is capable of accommodating both the 7-and 8-stranded-␤-sandwich conformations of gp120 with effective contacts to the 8th strand. Thus, the vaccine-induced C1C2-specific MAb DH677.3 has a unique binding angle to the C1C2 region compared to the infection-induced C1C2specific MAbs C11 and A32. DH677 lineage MAbs mediate ADCC against CD4-downmodulated HIV-1infected cells. During natural infection the HIV-1 accessory protein Nef downregulates CD4 expression on the surface of virus-infected cells (14,15). Cell surface-expressed CD4 facilitates the exposure of CD4i Env epitopes, like C1C2, by binding to coexpressed cell surface Env (16). The analyses of ADCC breadth was performed using target cells infected with IMCs containing the Renilla luciferase (LucR) reporter gene, which restricts Nef expression, leading to incomplete CD4 downregulation (17). Nevertheless, Vpu expression can compensate for Nef function and induce CD4 downregulation during the 72-h incubation of the target cells before assays were performed. To exclude any possible impact of this technical aspect of IMCs with LucR on our ADCC results, full-length IMCs (n ϭ 7) that do not contain a report gene were used to evaluate ADCC of the affinity-matured RV305 C1C2-specific MAbs DH677.3 and DH677.4 ( Fig. 2) and A32 (2). Since clade CRF01_AE possesses a histidine at Env HXB2 position 375 that influences sensitivity to CD4i antibody binding and ADCC (18,19), only clade B and clade C HIV-1 IMCs were used ( Table 2).
As these full-length IMCs did not contain a reporter gene, we used an infected cell elimination assay, which measures the reduction of live p24 ϩ , p24 ϩ CD4 ϩ , and/or p24 ϩ CD4 Ϫ cell populations in the presence of effector cells (Fig. 7). When evaluating elimination of total p24 ϩ cells, no significant difference (P Ͼ 0.05 by Wilcoxon rank sum test) in specific killing was noted among the three MAbs (Fig. 8A). However, when infected cells were separated into p24 ϩ CD4 ϩ (Fig. 8B) and p24 ϩ CD4 Ϫ (Fig. 8C), it was found that the RV305-boosted DH677.3 MAb mediated ADCC against 4 out of 7 HIV-1 IMCs, whereas DH677.4 and A32 MAb mediated ADCC against two or none of these IMCs, respectively. Although DH677.3 percent specific killing against these IMCs was low (mean, 6%; range, 0 to 24%), the level was significantly higher (P ϭ 0.03 by Wilcoxon rank sum test) for mediating ADCC against p24 ϩ CD4 Ϫ -infected cells (Fig. 8C) than that of A32. This is likely related to the unique DH677 clonal lineage epitope, which may be more frequently exposed on Env conformations on the surface of IMC-infected cells even in the context of CD4 downmodulation.

DISCUSSION
In this study, it was found that late boosting of RV144 vaccinees increased C1C2specific MAb V H ϩ V L chain gene mutation frequency and increased clonal lineagespecific ADCC breadth and potency ( Fig. 1 and 2 and Table 4). We analyzed the somatic hypermutation frequency of MAbs obtained after boosting and did not observe a clear correlation between somatic hypermutation frequency and ADCC breadth or potency. Boosting likely affinity matures antibody clonal lineages that are capable of acquiring broad and potent ADCC, as well as antibody clonal lineages incapable of acquiring broad and potent ADCC. For example, DH689 clonal lineage members DH689.1 and DH689.2 were both ϳ10% mutated (Table 1), but ADCC breadth and potency was worse than that of most of the RV144 MAbs assayed (Table 4). Conversely, boosting increased somatic hypermutation along with ADCC breadth and potency within the DH677 clonal lineage (Fig. 2). Thus, clonal lineages with fine epitope specificities capable of being matured to increased ADCC breadth and potency greatly benefited from the two AIDSVAX B/E boosts given in the RV305 clinical trial. It should be noted that in VAX003 and VAX004 clinical trials ADCC responses peaked at 3 to 4 immunizations and declined after 5 to 7 immunizations (20). Collectively these data indicate that while the RV144 clinical trial was underboosted, repetitive subsequent boosting beyond RV305 does not necessarily lead to continuously better functional antibody outcomes.
Improving vaccine-induced NNAb effector function will also require more detailed immunological studies on the timing and frequency of boosting. In the VAX003 (ClinicalTrials registration no. NCT00002441) and VAX004 (ClinicalTrials registration no. NCT00002441) trials, frequent protein immunizations skewed Env-specific antibody subclass usage from the highly functional IgG3 to IgG4 (21)(22)(23). The RV305 boosts that were studied here occurred several years (6 to 8 years) after the final RV144 boost, unlike previous HIV-1 vaccine trials. Whether the boosting interval can be shortened without skewing antibody subclass usage is not known, but it is possible that boosting with long rest intervals (Ն1 to 2 years) will be necessary.
The AIDSVAX B/E protein used for boosting in the RV144 and RV305 HIV-1 vaccine a Details are based on the DH677.3-gp120 93TH057 core e -M48U1, A32 Fab-ID2 93TH057 , and N12-i3 Fab-gp120 93TH057 core e ϩN/C-M48U1 structures, as calculated by the EBI PISA server (http://www.ebi.ac.uk/msd-srv/prot_int/cgi-bin/piserver). The two copies in the asymmetric unit of the DH677.3, A32, and N12-i3 complexes are averaged. (Continued on next page) trial contained an N-terminal 11-amino-acid deletion. Previously it was shown that this modification enhanced exposure of the C1C2 region and V2 loop (10). Here, we show that this modification disrupts C11-like MAb binding ( Fig. 1) but does create a germ line-targeting immunogen for DH677-like B cell lineages (Fig. 2). Ligand crystal structure analysis found that DH677.3 recognized a unique C1C2 epitope that involves epitope footprints of cluster A MAb A32 and N12-i3 (C11-like), as well as new elements of the inner domain layer 1 and the 7-stranded ␤-sandwich (Fig. 6). The DH677.3 epitope is positioned midway between the A32 and N12-i3 binding sites, with most residues being highly conserved. Interestingly, DH677.3 binds at the edge of the gp120 inner domain 7-stranded ␤-sandwich and with layers 1 and 2. This binding mode allows it to bind a gp120 conformation emblematic of the late stages of HIV entry recognized by C11 and C11-like MAbs (6). Most likely this allows DH677.3 to recognize a broader range of Env targets, emerging in both early (when the A32 epitope becomes available) and late (when the C11 epitope becomes available) stages of the viral entry process. Identification of stage 2A of the HIV-1 Env expressed on the surface of infected cells in the presence of the CD4 molecule or CD4 mimetics reiterate the importance of targeting these epitopes by vaccine-induced responses as detected in our assays (24). In addition, a model of DH677.3 in complex with gp120 antigen bound to CD4 of a target/infected cell confirms that the recognition site and angle of approach position the DH677.3 IgG for easy access for effector cell recognition and Fc-effector complex formation (Fig. 6A). Interestingly, we recently characterized JR4, a MAb isolated from a nonhuman primate that, like DH677.3, recognizes an epitope that includes elements of both the A32 and C11 binding sites (5). JR4 uses its CDR H1 to contact layer 2 residues of the A32 epitope region and its CDR H3 to reach the residues of the 7-stranded ␤-sandwich. However, in contrast to DH677.3, which largely relies on recognizing the 7-stranded ␤-sandwich, access of JR4 to this region is limited and involves only a few residue contacts. In this regard, JR4 is more like A32, with main anchoring contacts to layer 2 that limit its reach to the 7-stranded ␤-sandwich. DH677.3 misses these A32 layer 2 residues and shifts its epitope footprint more toward the 7-stranded ␤-sandwich, placing it midway between the two epitope regions.
ADCC-mediating antibodies have been shown to reduce mother-to-child HIV-1 DH677.3 heavy-and light-chain contact residues. MAb side-chain (ϩ) and main-chain (Ϫ) contact residues, colored green for hydrophobic, blue for hydrophilic, and black for both, as determined by a 5-Å cutoff value over the corresponding sequence. CDRs are colored as described for Fig. 1, and buried surface residues as determined by PISA are shaded.

FIG 4 Legend (Continued)
CDR L2, brown; CDRL3, blue. A blow-up view shows the network of hydrogen (H) bonds formed at the Fab-gp120 interface. H bonds contributed by side-chain and main-chain atoms of gp120 residues are colored magenta and blue, respectively. (B) Fab buried surface area (BSA) and gp120 residues forming DH677.3 epitope are shaded in blue according to BSA (antibody) and percent conservation of gp120 residues (Env). gp120 main-chain (blue) and side-chain (red) hydrogen bonds (H) and salt bridges (S) are shown above the residue. (C) The DH677.3 Fab-gp120 93TH057 core e interface. CDRs are shown as ribbons (left) and balls and sticks of residues contributing the binding (right) over the gp120 core. The molecular surface of gp120 is colored as described for panel A (left) and by electrostatic potential (right).
transmission (25)(26)(27) and slow virus disease progression (27)(28)(29), and in RV144 they were correlated with reduced risk of infection in vaccine recipients with lower anti-Env plasma IgA responses (7). Synergy between the RV144 C1C2 and V2 MAbs suggest a role for the C1C2 plasma responses that could not be directly identified by the correlates of the protection study. That DH677.3 was better than A32 at mediating ADCC against HIV-1 clade B and C CD4 downmodulated cells (Fig. 8) makes this MAb an attractive candidate for targeting HIV-1-infected cells in vivo in the setting of HIV-1 infection. We have previously shown that the C1C2 MAb A32, when formulated as a bispecific antibody, can potently opsonize and kill HIV-1-infected CD4 ϩ T cells (30).  ) is plotted on the gp120 surface with layers colored, as described for Fig. 1, with the A32 and N12-i2 epitope footprints shown in black. (C) DH677.3, A32, and N12-i3 gp120 contact residues are mapped onto the gp120 sequence. Side-chain (ϩ) and main-chain (Ϫ) contact residues are colored green for hydrophobic, blue for hydrophilic, and black for both, as determined by a 5-Å cutoff value over the corresponding sequence. Buried surface residues, as determined by PISA, are shaded. The DH677.3 epitope footprint overlays with the epitopes of both A32 and N12-i3.
Whether the DH677.3-type MAbs are superior to A32 for targeting virus-infected cells remains to be determined. In summary, our data demonstrate that if the RV144 vaccine trial had been boosted, ADCC-mediating C1C2-specific antibodies would have undergone affinity maturation for both ADCC potency and breadth of recognition of HIV-1-infected CD4 ϩ T cells. Antigen-specific single-cell sorting. All PBMCs used in this study were collected 2 weeks after the second boost in RV305. A total of 1 ϫ 10 7 PBMC per vaccine recipient were stained with AE.A244gp120Δ11 fluorescently labeled proteins and a human B cell flow cytometry panel. Viable antigen-specific B cells (AqVd Ϫ CD14 Ϫ CD16 Ϫ CD3 Ϫ CD19 ϩ IgD Ϫ ) were single-cell sorted with a BD FACSAria II SORP (BD Biosciences, Mountain View, CA) into 96-well PCR plates and stored at -80°C for RT-PCR.

MATERIALS AND METHODS
Single-cell reverse transcriptase PCR. Single B cell cDNA was generated with random hexamers using SSIII. The antibody heavy-and light-chain variable regions were PCR amplified using AmpliTaq360 master mix (Applied Biosystems). PCR products were purified (Qiagen, Valencia, CA) and sequenced by Genewiz. Gene rearrangements, clonal relatedness, unmutated common ancestors, and intermediate ancestor inferences were made using Cloanalyst (13). The DH677 clonal lineage tree was generated using FigTree.
MAb production. PCR-amplified heavy-and light-chain gene sequences were transiently expressed as previously described (32). Ig-containing cell culture supernatants were used for ELISA binding assays. For large-scale expression, V H and V L chain genes were synthesized (V H chain in the IgG1 4A backbone) and transformed into DH5␣ cells (GeneScript, Piscataway, NJ). Plasmids were expressed in Luria broth and purified (Qiagen, Valencia, CA), and Expi293 cells were transfected using ExpiFectamine (Life Technologies, Carlsbad, CA) by following the manufacturer's protocol. After 5 days of incubation at 37°C 5% CO 2 , the Ig-containing medium was concentrated and purified with protein A beads, and the antibody buffer was exchanged into PBS.
Antibody binding and blocking assays. Direct ELISAs were performed as previously described (32). Briefly, 384-well microplates were coated overnight with 30 ng/well of protein. Antibodies were diluted and added for 1 h. Binding was detected with anti-IgG-horseradish peroxidase (HRP) (Rockland) and developed with SureBlue reserve TMB one component (KPL). Plates were read on a plate reader (Molecular Devices) at 450 nm. A32-blocking assays were performed by adding the RV305 antibodies followed by biotinylated A32 and detection with streptavidin HRP.
Neutralization assays. TZM-bl neutralization assays were performed as previously described (33). No neutralization was detected for the MAbs assayed in this study.
Infection of CEM.NKR CCR5 cell line with HIV-1 IMCs. CEM.NKR CCR5 cells were infected with HIV-1 IMCs as previously described (36). Briefly, IMCs were titrated in order to achieve maximum expression within 48 to 72 h postinfection, as determined by detection of luciferase activity and intracellular p24 expression. IMC infections were performed by incubation of the optimal dilution of virus with CEM. NKR CCR5 cells for 0.5 h at 37°C and 5% CO 2 in the presence of DEAE-dextran (7.5 g/ml). The cells were subsequently resuspended at 0.5 ϫ 10 6 /ml and cultured for 48 to 72 h in complete medium containing 7.5 g/ml DEAE-dextran. For each ADCC assay, we monitored the frequency of infected target cells by intracellular p24 staining. Assays performed using infected target cells were considered reliable if cell viability was Ն60% and the percentage of viable p24 ϩ target cells on assay day was Ն20%.
Luciferase ADCC assay. ADCC activity was determined by a luciferase-based assay as previously described (8,37). Briefly, CEM.NKR CCR5 cells (NIH AIDS Reagent Program, Division of AIDS, NIAID, NIH, from Alexandra Trkola) (38) were used as targets after infection with the HIV-1 IMCs. PBMCs obtained from an HIV-seronegative donor with the heterozygous 158F/V and 131H/R genotypes for Fc␥R3A and Fc␥R2A (39,40), respectively, were used as a source of effector cells and were used at an effector-totarget ratio of 30:1. Recombinant MAbs were tested across a range of concentrations using 5-fold serial dilutions starting at 50 g/ml. The effector cells, target cells, and Ab dilutions were plated in opaque 96-well half area plates and were incubated for 6 h at 37°C in 5% CO 2 . The final read-out was the luminescence intensity (relative light units, or RLU) generated by the presence of residual intact target cells that have not been lysed by the effector population in the presence of ADCC-mediating MAb (ViviRen substrate; Promega, Madison, WI). The percentage of specific killing was calculated using the formula percent specific killing ϭ [(number of RLU of target and effector well Ϫ number of RLU of test well)/number of RLU of target and effector well] ϫ 100. In this analysis, the RLU of the target plus effector wells represents spontaneous lysis in the absence of any source of Ab. The ADCC endpoint concentration (EC), defined as the lowest concentration of MAb capable of mediating ADCC in our in vitro assay, was calculated by interpolation of the MAb concentration that intersected the positive cutoff of 15% specific killing. The RSV-specific MAb palivizumab was used as a negative control.
ADCC score. Antibodies were tested across a range of concentrations using 5-fold serial dilutions starting at 50 g/ml. Since the dilution curves are not monotonic due to prozone effect of MAbs, nonparametric area under the curve (AUC) was calculated using the trapezoidal rule, with activity of less than 15% set to 0%. In this study, we used principal component analysis (PCA) to compute an ADCC score that explains both potency and breadth of the MAbs. This method of using PCA to calculate breadth score for neutralizing antibodies has been used in a previous study (41). Similarly, in 2011 a study published in Atherosclerosis aimed to study the effect of various risk factors on increase in carotid and femoral intima-media thickness used PCA to calculate a cumulative risk score (42).
PCA is the most commonly used method to reduce the dimensionality of the data set (43). It uses eigenvector decomposition of the correlation matrix of the variables, where each variable is represented by an HIV-1 IMC in our study. Most of the shared variance of the correlations of ADCC AUC of HIV-1 IMCs is explained by the first principal component (PC1) (41). Ideally, one would want to explain 70% of the variance, but this should not be at the expense of adding principal components with an eigenvalue of less than 1 (44). Eigenvalue is a measure of variance in the data along that principal component (PC), and a larger value would mean that corresponding PC explains a larger amount of variance in the data. The rationale behind this methodology of calculating ADCC score is that MAbs will not target all seven HIV-1 IMCs equally, hence, calculation of breadth score for MAbs needs to account for this variation.
In this study, a panel of 7 HIV-1 IMCs was tested, which implies that our data set has seven dimensions. ADCC activity was measured as AUC. In our analysis, PC1 and PC2 have eigenvalues above 1 and together account for 80.57% variance (Table 6). Scores obtained from the first principal component can be interpreted as a weighted average of the 7 HIV-1 IMCs that would account for both potency as well as breadth of the MAbs (44). Higher PC1 score would mean that MAb has higher breadth as well as potency for ADCC activity. To calculate the ADCC score, the standardized AUC value for each MAb is first calculated for each HIV-1 IMC and then multiplied by factor loading of the corresponding HIV-1 IMCs. Lastly, these products are added together. Factor loadings are the correlation coefficients between PCs and HIV-1 IMCs (42). The ADCC score obtained from PC1 is a weighted average of the standardized AUC, where factor loadings obtained from PC1 are used as weights to calculate the weighted average. Standardized AUC values imply zero mean and unit standard deviation. The AUC values below the value of mean AUC will result in negative PC1 scores. Infection of primary cells with HIV-1 IMCs. IMCs encoding the full-length transmitted/founder sequence of seven individuals infected with either subtype B or C viruses from the CHAVI acute infection cohort (CH77, CH264, CH0470, CH042, CH185, CH162, and CH236) were constructed as previously described (45,46) and used to infect primary CD4 ϩ cells. To infect cells, cryopreserved PBMCs were thawed and stimulated in R20 medium (RPMI medium [Invitrogen] with 20% fetal bovine serum [Gemini Bioproducts], 2 mM L-glutamine [Invitrogen], 50 U/ml penicillin [Invitrogen], and 50 g/ml gentamicin [Invitrogen]) supplemented with interleukin-2 (IL-2) (30 U/ml; Proleukin), anti-CD3 (25 ng/ml clone OKT-3; Invitrogen), and anti-CD28 (25 ng/ml; BD Biosciences) antibodies for 72 h at 37°C in 5% CO 2 . CD8 cells were depleted from the PBMCs using CD8 microbeads (Miltenyi Biotec, Germany) according to the manufacturer's instructions, and 1.5 ϫ 10 6 cells were infected using 1 ml virus supernatant by spinoculation (1,125 ϫ g) for 2 h at 20°C. After spinoculation, 2 ml of R20 supplemented with IL-2 was added to each infection, and infections were left for 72 h. Infected cells were used if viability was Ͼ70% and more than 5% of cells were p24 ϩ .
Infected cell elimination assay. HIV-1-infected or mock-infected CD8-depleted PBMCs were used as targets, and autologous cryopreserved PBMCs rested overnight in R10 supplemented with 10 ng/ml IL-15 (Miltenyi Biotec) were used as a source of effector cells. Infected and uninfected target cells were labeled with a fluorescent target cell marker (TFL4; OncoImmunin) and a viability marker (NFL1; OncoImmunin) for 15 min at 37°C, as specified by the manufacturer. The labeling of the target cells with these two markers allowed us to clearly identify only the live viable cells in our gating strategy and exclude artifacts related to the presence of dead cell staining. Cells were washed in R10 and adjusted to a concentration of 0.2 ϫ 10 6 cells/ml. PBMCs were then added to target cells at an effector/target ratio of 30:1 (6 ϫ 10 6 cells/ml). The target-effector cell suspension was plated in V-bottom 96-well plates and cocultured with 10 g/ml of each MAb. Cocultures were incubated for 6 h at 37°C in 5% CO 2 . After the incubation period, cells were washed and stained with anti-CD4-peridinin chlorophyll protein-Cy5.5 (clone OKT4; eBioscience) at a final dilution of 1:40 in the dark for 20 min at room temperature. Cells were then washed, resuspended in 100 l/well Cytofix/Cytoperm (BD Biosciences), incubated in the dark for 20 min at 4°C, washed in 1ϫ Cytoperm wash solution (BD Biosciences), coincubated with anti-p24 antibody (clone KC57-RD1; Beckman Coulter) to a final dilution of 1:100, and incubated in the dark for 25 min at 4°C. Cells were washed three times with Cytoperm wash solution and resuspended in 125 l PBS-1% paraformaldehyde. The samples were acquired within 24 h using a BD Fortessa cytometer. The appropriate compensation beads were used to compensate for the spillover signal for the four fluorophores. Data analysis was performed using FlowJo 9.6.6 software (TreeStar). Mock-infected cells were used to appropriately position live cell p24 ϩ/Ϫ and CD 4 ϩ/Ϫ gates. Specific killing was determined by the reduction in percentage of viable p24 ϩ cells in the presence of MAbs after taking into consideration nonspecific killing and was calculated as [% p24(target ϩ effector cells) Ϫ % p24(targets ϩ effectors ϩ MAb or plasma)]/[% p24(target ϩ effector cells)].
CH65 (an anti-influenza monoclonal antibody, kindly provided by M. A. Moody) was used as a negative control. To remove background signal, the highest value of percent specific killing induced by CH65 was subtracted from the calculated reduction in percentage of p24 ϩ cells, and then negative values were rounded to 0%. As the data were background subtracted, no positivity criteria were applied to the data.
SPR. The binding and kinetic rate measurements of gp120 proteins against RV305 antibodies were obtained by surface plasmon resonance (SPR) using the Biacore 3000 instrument (GE Healthcare). SPR measurements were performed using a CM5 sensor chip with anti-human IgG Fc antibody directly immobilized to a level of 9,000 to 11,000 RU (response units). Antibodies then were captured at 5 l/min for 60 s to a level of 100 to 300 RU. For binding analyses, the gp120 proteins were diluted to approximately 1,000 nM in PBS and injected over the captured antibodies for 3 min at 30 l/min. For kinetics measurements, the gp120 proteins were diluted from 5 to 750 nM and injected using a high-performance kinetics injection for 5 min at 50 l/min. This was followed by a dissociation period of 600 s and surface regeneration with glycine, pH 2.0, for 20 s. Results were analyzed using Biacore BiaEvaluation software (GE Healthcare). Negative-control antibody (Ab82) and blank buffer binding were used for double reference subtraction to account for nonspecific protein binding and signal drift. Subsequent curve-fitting analysis was performed using a 1:1 Langmuir model with a local R max , and the reported rate constants are representative of two measurements.
Protein preparation and complex crystallization. DH677.3 Fab alone was grown and crystallized at a concentration of ϳ10 mg/ml. The structure was solved by molecular replacement with PDB entry 3QEG in space group P2 1 to a resolution of 2.6 Å. Clade A/E 93TH057 gp120 core e (gp120 93TH057 core e , residues 42 to 492 [Hxbc2 numbering]), lacking the V1, V2, and V3 variable loops and containing an H375S mutation to allow binding of the CD4 mimetic M48U1 (47), was used to obtain crystals of DH677.3 Fab-antigen complex. gp120 93TH057 core e was prepared and purified as described previously (3). Deglycosylated gp120 93TH057 core e was first mixed with CD4 mimetic peptide M48U1 at a molar ratio of 1:1.5 and purified through gel filtration chromatography using a Superdex 200 16/60 column (GE Healthcare, Piscataway, NJ). After concentration, the gp120 93TH057 core e -M48U1 complex was mixed with a 20% molar excess of DH677.3 Fab and again passed through the gel filtration column equilibrated with 5 mM Tris-HCl buffer, pH 7.2, and 100 mM ammonium acetate. The purified complex was concentrated to ϳ10 mg/ml for crystallization experiments. The structure was solved by molecular replacement using the DH677.3 Fab and PDB entry 3TGT as the searching models in space group P1 to a resolution 3.0 Å. The final R factor /R free (%) for the Fab structure is 19.9/26.1, and the final R factor /R free for the complex is 21.4/27.4 ( Table 7). The PDB codes for the deposited structures are 6MFJ and 6MFP, respectively. In each case the asymmetric unit of the crystal contained two almost identical copies of Fab or the Fab-gp120 93TH057 core e complex (Fig. 3).
Crystallization and data collection. Initial crystal screens were done in vapor-diffusion hangingdrop trials using commercially available sparse matrix crystallization screens from Hampton Research (Index), Emerald BioSystems (Precipitant Wizard Screen), and Molecular Dimensions (Proplex and Macrosol Screens). The screens were monitored periodically for protein crystals. Conditions that produced crystals were then further optimized to produce crystals suitable for data collection. DH677.3 Fab crystals were grown from 20% polyethylene glycol (PEG) 3000, 100 mM HEPES, pH 7.5, and 200 mM sodium chloride. DH677.3 complex crystals were grown from 25% PEG 4000 and 100 mM morpholineethane- sulfonic acid, pH 5.5. Crystals were briefly soaked in crystallization solution plus 20% 2-methyl-2,4pentanediol (MPD) before being flash frozen in liquid nitrogen prior to data collection. Data collection and structure solution. Diffraction data were collected at the Stanford Synchrotron Radiation Light Source (SSRL) at beam line BL12-2, equipped with a Dectris Pilatus area detector. All data were processed and reduced with HKL2000 (48). Structures were solved by molecular replacement with Phaser (49) from the CCP4 suite (50). The DH677.3 Fab structure was solved based on the coordinates of the N12-i2 Fab (PDB entry 3QEG), and the DH677.3 complex was then solved with coordinates from the DH677.3 Fab model, gp120 (PDB entry 3TGT), and M48U1 (PDB entry 4JZW). Refinement was carried out with Refmac (51) and/or Phenix (52). Refinement was coupled with manual refitting and rebuilding with COOT (53). Data collection and refinement statistics are shown in Table 7.
Structure validation and analysis. The quality of the final refined models was monitored using the program MolProbity (54). Structural alignments were performed using the program lsqkab from the CCP4 suite (50). The PISA (55) webserver was used to determine contact surfaces and residues. All illustrations were prepared with the PyMol Molecular Graphic suite (http://pymol.org) (DeLano Scientific, San Carlos, CA, USA). Conservation of the DH677.3 epitope was calculated using the HIV Sequence Database Compendium (https://www.hiv.lanl.gov/content/sequence/HIV/COMPENDIUM/compendium.html), comparing gp120 residues to those of clade B Hxbc2. Only unique sequences in the database having an equivalent residue at each position were included in the calculated percentage, representing approximately 32,000 sequences on average.
Statistical methods. For luciferase-based ADCC assay, background correction was performed by subtracting the highest value of percent specific killing induced by CH65 and then rounding off the negative values to zero. In Fig. 5, positivity cutoff criteria were not applied but background correction was performed.
In order to assess if two groups have different responses, pairwise comparisons between groups was conducted using Wilcoxon rank sum test. Statistical analysis was performed using SAS software (SAS Institute, Inc., Cary, NC).
Data availability.