Restricting HIV-1 pathways for escape using rationally designed anti–HIV-1 antibodies

We previously postulated that neutralization of an NIH45-46–resistant virus by a chimera of the NIH45-46 heavy chain (HC) paired with the VRC01 light chain (LC) was achieved via additional contacts that Tyr28_VRC01(LC), but not Ser28_NIH45-46(LC), makes with an N-linked glycan attached to Asn276_gp120 (Diskin et al., 2011) that is predicted to be present in 94.7% of 4279 Env sequences in the Los Alamos HIV Sequence Database (www.hiv.lanl.gov/). In an effort to increase the breadth of NIH45-46^G54W, we introduced an S28Y substitution into the NIH45-46 LC. We expressed NIH45-46^{G54W(HC), S28Y(LC)} (hereafter referred to as 45-46m2; Fig. 1 A) and evaluated it using in vitro neutralization assays against a cross-clade panel of 118 primary HIV-1 isolates including transmitted founder viruses (Tables S1 and S2). Against this panel, 45-46m2 was as potent as NIH45-46^G54W (geometric mean IC₅₀ values of 0.028 and 0.030 µg/ml for 45-46m2 and NIH45-46^G54W, respectively) but exhibited increased breadth (Fig. 1 B), neutralizing up to 96% of strains. Of 28 strains that were either poorly neutralized (IC₅₀ ≥ 1.0 µg/ml) or NIH45-46 resistant (IC₅₀ > 50 µg/ml), 45-46m2 had a geometric mean IC₅₀ of 0.35 µg/ml compared with geometric means of 1.84 µg/ml (NIH45-46^G54W) and 14.51 µg/ml (NIH45-46; calculated with IC₅₀s > 50 µg/ml input as 50 µg/ml; Fig. 1 C).

We next compared the neutralizing activity of NIH45-46^G54W and 45-46m2 using viruses from an elite neutralizer, patient VC10042, who developed unusually broad and potent cross-neutralizing antibody responses but whose circulating viruses evolved to escape their action; thus, with few exceptions, the viral clones isolated from VC10042 were highly resistant to known bNAbs, especially CD4bs antibodies (Sather et al., 2012). Only one viral clone isolated from VC10042 was susceptible to VRC01, but the majority (8 of 12) were moderately susceptible to NIH-45-46^G54W (Sather et al., 2012). In contrast, NIH45-46m2 neutralized all (12 of 12) of the viral clones with up to 10-fold greater potency relative to NIH45-46^G54W when evaluated against NIH45-46^G54W-sensitive clones (Fig. 2).

To explore the mechanism underlying the increased breadth of 45-46m2, we solved the 2.8 Å crystal structure of the 45-46m2 Fab bound to 93TH057 gp120 (Fig. 3 A and Table S3) and compared it with the structure of NIH45-46 bound to the same gp120 (Diskin et al., 2011). The total buried surface area of the 45-46m2/gp120 complex (3214 Ų) was larger than the buried surface area calculated for NIH45-46/gp120 (2636 Ų) or NIH45-46^G54W/gp120 (2869 Ų; Fig. 3 B). Consistent with this observation, surface plasmon resonance (SPR) binding assays (Fig. 4 A) demonstrated that 45-46m2 Fab exhibited a higher affinity and slower dissociation rate than NIH45-46^G54W Fab or NIH45-46 Fab for binding 93TH057 gp120 (Fig. 4 B). The increased buried surface area for the 45-46m2–gp120 complex largely resulted from the predicted (Diskin et al., 2011) penetration of Trp54_45-46m2 into the Phe43_CD4 pocket of gp120 (Fig. 3 C). Local rearrangements in the surrounding residues of the gp120 Phe43_CD4 pocket occurred; Gly473_gp120 was shifted by ∼4Å (Cα-Cα distance) toward Trp54_45-46m2 due to a hydrogen bond between its carbonyl oxygen and the ε-nitrogen of Trp54_45-46m2, and Ile371_gp120 adopted a different rotamer to accommodate Trp54_45-46m2 (Fig. 3 C). The Gly–Trp substitution at position 54 allowed the tryptophan indole side chain to bury an additional 233 Ų (144 Ų on 45-46m2 and 89 Ų on gp120) of surface area in the Fab–gp120 complex, representing a favorable energy change of complex formation of approximately −3.5 kcal/mol (assuming −15 cal/mol•Å² of buried hydrophobic surface [Vallone et al., 1998]). Although the Fab–gp120 protein complex was treated with a deglycosidase before crystallization, some N-linked glycans remained intact; thus, we found extended electron density for six monosaccharides of a glycan attached to Asn276_gp120 (Fig. 3 D). The 45-46m2 LC made extensive contacts with the Asn276_gp120–attached glycan, and Tyr28_45-46m2(LC) was a central component in this interaction through contacts with the first two N-acetylglucosamines and the subsequent mannose (Fig. 3 E). Arg64_45-46m2(LC), Trp65_45-46m2(LC), and Tyr89_45-46m2(LC) also contributed to the contact surface with the glycan, which consisted of 554 Ų buried surface area (280 Ų on 45-46m2 and 274 Ų on the glycan). Interactions with the Asn276_gp120–attached glycan by an aromatic residue in position 65_LC rationalize the observation that Trp65_LC/Phe65_LC are characteristic residues selected for during affinity maturation in the LCs of PVL antibodies (West et al., 2012).

We next examined the interactions between 45-46m2 and gp120 residues within a gp120 consensus sequence identified as involved in resistance to NIH45-46^G54W (West et al., 2012). In most resistant HIV-1 strains, uncommon amino acids replace highly conserved residues in the gp120 D and V5 loops, including Asn279_gp120/Asp279_gp120, Asn280_gp120 (loop D), Gly458_gp120, or Gly459_gp120 (base of loop V5; hereafter referred to as the N/DNGG consensus). The N/DNGG consensus appears in 96.4, 96.3, 96.4, and 93.0% of 2,869 unique Env sequences from the Los Alamos Database for positions 279_gp120, 280_gp120, 458_gp120, and 459_gp120, respectively. Furthermore, YU-2 escape mutant viruses isolated from HIV-1_YU-2–infected mice that were treated with NIH45-46^G54W carried mutations within the N/DNGG motif (Klein et al., 2012), selectively mutating the N/DNGG motif in YU-2 pseudoviruses reduced or completely prevented neutralization by NIH45-46^G54W (West et al., 2012), and reverting the nonconsensus residues of the resistant T278-50 strain to N/DNGG (A279N/E459G) made it sensitive to NIH45-46^G54W (IC₅₀ = 0.07 µg/ml; unpublished data). Analysis of the 45-46m2–gp120 complex structure suggests that most changes to the N/DNGG motif (Fig. 5 A) would cause steric constraints for binding of PVL antibodies (Fig. 5, B–D); specifically, substituted residues in the N/DNGG motif are likely to clash with Trp102_45-46m2(HC) and Trp47_45-46m2(HC) (Fig. 5, B and D), leading to suboptimal or completely abrogated binding. Other residues that may be affected by N/DNGG motif alterations include Tyr100_45-46m2(HC) and Tyr89_45-46m2(LC) (Fig. 5 E).

To examine the effects of mutations within the N/DNGG motif on neutralization by 45-46m2, we used modified YU-2 pseudoviruses each including a substitution in a N/DNGG consensus residue (Fig. 6 A). We also generated a YU-2 strain (YU-2^R456W) with a mutation in Arg456_gp120, a highly conserved (95%) residue mutated in five NIH45-46^G54W–resistant strains, which is not directly contacted by PVL antibodies but that may influence the conformation of the gp120 V5 loop region (West et al., 2012). With the exception of the Arg456_gp120 mutant, the positions substituted were based upon mutations found in YU-2 escape mutants generated in YU-2–infected humanized mice in response to treatment with NIH45-46^G54W (Klein et al., 2012). Unlike NIH45-46^G54W, which lost activity against most of these mutants (West et al., 2012), 45-46m2 neutralized all to at least some degree and potently neutralized YU-2^R456W, YU-2^G458D, and YU-2^G459D, which partially resisted NIH45-46^G54W neutralization (Fig. 6 A). Alterations in residues 279_gp120 and 280_gp120, including an A281T mutation that creates a potential N-linked glycosylation site at Asn279_gp120, were less tolerated by 45-46m2 and resulted in significant reductions in neutralization potencies (Fig. 6 A). However, the ability of 45-46m2 to neutralize mutant YU-2 strains that carried changes in the V5 loop (YU-2^G458D and YU-2^G459D), one of which (YU-2^G458D) was only weakly sensitive to neutralization by NIH45-46^G54W, indicates that the favorable interaction of 45-46m2 with the Asn276_gp120-linked glycan can effectively compensate for the loss of an optimal interaction in the vicinity of V5_gp120 even though the 45-46m2 substitution makes no direct contacts to this region. Nonetheless, when evaluated against natural HIV-1 isolates, NIHG45-46^G54W–resistant strains that carried a nonconsensus N/DNGG motif were also resistant to 45-46m2 (Fig. 1 C, blue highlighted strains).

To further improve the potency of 45-46m2 against escape mutants in general and against nonconsensus N/DNGG motifs in particular, we created 24 antibody mutants designed to reduce steric clashes between 45-46m2 and substituted residues in the gp120 N/DNGG motif (Fig. 1 A). The neutralization potencies of the new mutant antibodies were evaluated against the panel of YU-2 mutants (Figs. 6 and 7 A). Modifying critical somatically mutated residues in PVL antibodies (Trp102_45-46m2(HC), Tyr100_45-46m2(HC), and Tyr89_45-46m2(LC); West et al., 2012) to create mutants 45-46m4, m5, m6, m16, m17, m18, m20, m21, m22, m23, m29, m30, m31, m32, m34, m35, and m36 did not improve the neutralization profiles of the antibodies (Fig. 6). However, replacing the highly conserved Trp47_45-46m2(HC) (a germline residue) with different smaller amino acids resulted in antibodies capable of neutralizing YU-2^N279K, YU-2^N280D, and YU-2^N280Y (Fig. 7, A and B). These mutants, 45-46m7, 45-46m25, and 45-46m28 (45-46m2 + HC mutations W47V, W47I, and W47T, respectively), effectively neutralized all YU-2 mutants with the exception of YU-2^A281T, which included a newly introduced potential N-linked glycosylation site at Asn279_gp120.

To explore the effects of the A281T_gp120 mutation that abrogated neutralization of the 45-46m antibodies, we used a previously described in vitro assay (Sather et al., 2012) to compare the relative fitness of YU-2 mutants that either included (YU-2^A281T) or did not include (YU-2 WT, YU-2^N279K, and YU-2^N280D) a potential N-linked glycosylation site (Asn279_gp120–X–Ser/Thr281_gp120; Fig. 8 A). Although infectious, YU-2^A281T exhibited a disrupted replication profile relative to the other viruses (Fig. 8 B), suggesting a fitness cost associated with an Asn279_gp120–attached glycan. Consistent with this suggestion, only three strains in the Los Alamos Database carry a potential N-linked glycosylation site at Asn279_gp120, all with low predicted glycosylation potentials (Fig. 8 A and below). In addition, unlike the curves for other YU-2 variants, the in vitro neutralization curves for YU-2^A281T saturated at ∼50%, suggesting the existence of heterogeneous viral populations resulting from incomplete incorporation of N-linked glycan at Asn279_gp120 (Fig. 7 B and Fig. 8 C).

Available structural information about CD4 binding to gp120 can be used to rationalize why the viral replication profiles of the YU-2^A281T mutant were retarded compared with YU-2 WT (Fig. 8 B). The N-linked glycan attached to Asn279_gp120 that was introduced by the A281T_gp120 substitution is predicted to exert a fitness cost for HIV-1 by partially blocking the CD4bs. Indeed if there was not a fitness cost associated with having a glycan at residue 279, one would not expect the highly correlated amino acid distribution at sites 279 and 281 observed in the Los Alamos database. Specifically, residue 279_gp120 is usually Asn (51%) or Asp (46%), and this preference is not clade-specific. Residue 281_gp120 is Thr in ∼11% (n = 314) of the sequences. However, Thr occurs in only one strain (CH080183_e_p1) that includes Asn at 279_gp120, a distribution which has a <1 in 10¹⁰⁰ chance of occurring randomly (Fisher Exact test). Furthermore the middle residue of the potential N-linked glycosylation sequence in this strain is proline, which would prevent attachment of an N-glycan to Asn279_gp120. Additionally, of the 39 strains with Ser281_gp120, only three have Asn279_gp120 (CY122, 99CMA121, and U14842). An analysis of the glycosylation potential of these three strains (Fig. 8 A) using the NetNGlyc 1.0 Server indicates no glycosylation potential (strains CY122 and U14842) or a very low potential (strain 99CMA121) for Asn279_gp120. The YU-2^A281S mutant has a higher glycosylation potential at Asn279_gp120 compared with the 99CMA121 strain (0.5402 vs. 0.5188; Fig. 8 A), but unlike YU-2^A281T, the YU-2^A281S mutant was neutralized well (Fig. 8 C), suggesting that an N-glycan was not incorporated at Asn279_gp120 despite the N-X-S motif.

Considering the close contacts that 45-46m2 makes with Asn279gp₁₂₀ (Fig. 5 C), an N-linked glycan attached to Asn279_gp120 would most likely prevent 45-46m2 binding to gp120s on spike trimers, resulting in resistance to neutralization. The saturation of the YU-2^A281T neutralization curves at values <100% (Fig. 7 B and Fig. 8 C) is consistent with both a sensitive and a resistant population of virions, suggesting only partial incorporation of N-linked glycan attached to Asn279_gp120. Incomplete processing at the level of individual gp120 protomers would likely give rise to heterogeneously glycosylated trimeric Env spikes, i.e., trimers that were fully, partially, or not glycosylated at residue 279. At the population level, the sensitivity of YU-2^A281T viruses to 45-46m2-like antibodies would vary according to the Env composition of each virus, thereby giving rise to both resistant and sensitive virions within the set of YU-2^A281T viruses.

We next evaluated the breadth and potencies of selected antibodies alone and in combination using the 118 strain cross-clade virus panel (Tables S1 and S2). 45-46m7, 45-46m25, and 45-46m28 effectively neutralized YU-2 N/DNGG consensus variants, but these antibodies and a 45-46m2/45-46m7 combination did not neutralize consensus variant strains that were resistant to 45-46m2 (Tables S1, S2, and S4). These results suggest that changing Trp47_45-46m2(HC) to a smaller amino acid can only partially alleviate steric constraints associated with PVL antibody binding to an N/DNGG consensus variant. Thus, effective neutralization by 45-46m7, 45-46m25, and 45-46m28 of escape mutants that use non-N/DNGG consensus residues is likely only when the parent viral strain is sensitive to a PVL antibody. However, given the broad neutralization profiles of parental PVL antibodies (West et al., 2012), strains resistant to parental PVLs and to 45-46m7, 45-46m25, and 45-46m28 as a result of changes in the gp120 N/DNGG consensus sequence are likely to be rare. Addition of 10–1074, a more potent clonal variant of PGT121 (Walker et al., 2011) that recognizes a carbohydrate-dependent epitope associated with the gp120 V3 loop (Mouquet et al., 2012), or PG9, a carbohydrate-dependent bNAb recognizing a V1/V2 epitope (Walker et al., 2009), into the mixture resulted in neutralization of almost all resistant strains (Table S4).

Antibodies can drive HIV-1 mutation or even control viral replication in humanized mice (Klein et al., 2012), offering the opportunity to examine HIV-1 escape mutations that arise in response to treatment with selected bNAbs. We compared escape mutations in HIV-1_YU-2 that arose in response to a 45-46m2/45-46m7 combination versus monotherapy with NIH45-46^G54W. Treatment with 45-46m2/45-46m7 resulted in a significant initial drop in viremia by 7 d (Fig. 9, A and B, P = 0.0057). Although viremia rebounded to pretreatment levels after 21 d in seven of eight mice, the Env sequences isolated from the 45-46m2/45-46m7–treated viremic mice revealed striking differences compared with viruses isolated after escape from NIH45-46^G54W monotherapy (Klein et al., 2012; Fig. 9 C, Fig. 10 A, and Table S5 A). Notably, mutations in the GG portion of the N/DNGG consensus sequence (Gly458_gp120 and Gly459_gp120), which resulted in resistance to NIH45-46^G54W (Fig. 7 A) and that were isolated after NIH45-46^G54W monotherapy (Klein et al., 2012), were absent (Fig. 9 C; Fig. 10 A; Table S5 A). Although effective against potential mutations in the V5 region (residues 458_gp120 and 459_gp120), the combination of 45-46m2 and 45-46m7 did not eliminate mutations in loop D (residues 279_gp120 and 280_gp120). This may indicate that the antibody concentrations reached in vivo were not sufficient. Consistent with this suggestion, the in vitro IC₅₀ values for 45-46m2 and 45-46m7 against loop D variants were >0.1 µg/ml whereas the IC₅₀ values for V5 variants were <0.01 µg/ml (Fig. 7 A).

The predominant escape mutant found in viruses isolated from the 45-46m2/45-46m7–treated mice was A281T_gp120, a substitution which introduces a potential N-linked glycosylation site at Asn279_gp120 and results in a less fit virus (Fig. 8 B). In the context of an Asn279_gp120-linked glycan, compensatory mutations to remove the potential N-linked glycosylation site at Asn276_gp120 were selected (Fig. 9 C, Fig. 10 A, and Table S5 A). Specifically, attachment of an N-linked glycan to Asn276_gp120 was prevented by altering the asparagine (N276D and N276S) or the final residue (T278A) in the Asn-X-Ser/Thr potential N-linked glycosylation sequence motif. We speculate that a glycan attached to Asn279_gp120 in a gp120 lacking Asn276_gp120-attached glycan could be pushed toward the empty space created by elimination of the Asn276_gp120 glycan to facilitate binding to CD4. Thus, eliminating the glycan at Asn276_gp120 could compensate for the otherwise unfavorable addition of a glycan to Asn279_gp120. The suggestion that mutations to remove an Asn276_gp120-linked glycan are compensatory mutations required when an Asn279_gp120-linked glycan is introduced rather than escape mutations on their own is consistent with potent neutralization of N276S and T278A mutants of YU-2 by NIH45-46^G54W, 45-46m2, and 45-46m7 (Fig. 8 C) and the emergence of N276S and T278A mutations only when A281T was present (Fig. 10 A and Table S5 A).

When HIV-1_YU2–infected mice were treated with a combination of 45–46-m2, 45-46m7, and 10–1074 (Mouquet et al., 2012), we observed control of viremia in all animals that lived beyond 20 d after the start of treatment (Fig. 9, A and B). With regards to the animal that died before this time, gp120 sequences just before its death did not harbor mutations that would indicate escape from either 10–1074 or 45-46m2/m7 (Fig. 10 B and Table S5 B). Although some mice had detectable viral loads during treatment, we did not find any known escape mutations in viruses isolated during treatment for the bNAbs used in the treatment mix (Fig. 10 B and Table S5 B). Thus the combination of 45-46m2 and 45-46m7 effectively reduced the available pathways for escape, and the 45-46m2, 45-46m7, and 10–1074 combination potently treated HIV-1_YU-2–infected mice.

Antibody-based therapies using combinations of bNAbs are a promising approach for treating or preventing HIV-1 infection (Klein et al., 2012), with the best antibody regimen still to be devised. The successful use of structure-based design to improve NIH45-46 demonstrates that bNAbs, as isolated from HIV-1–infected donors (Scheid et al., 2011), are not necessarily optimal in neutralization breadth or potency. Thus, we were able to create NIH45-46^G54W, which exhibited an overall 10-fold increased potency over its parental bNAb by changing only a single amino acid (Diskin et al., 2011). Here, we used a combination of structure-based rational design and bioinformatics analyses of sequence features of HIV-1 resistance to construct further improved versions of NIH45-46^G54W, resulting in new bNAbs with improved breadth that restrict potential pathways for HIV-1 escape by mutation.

One of the new bNAbs, 45-46m2, neutralizes 96–97% of HIV-1 strains including VC10042 isolates (Sather et al., 2012) and is equivalent or better on its own than a combination of VRC01 and PG9 (Doria-Rose et al., 2012). Remarkably, 77% of tested viruses were extremely sensitive to 45-46m2, having IC₅₀ values at or below 0.1 µg/ml, as compared with 27% of viruses neutralized by VRC01 at this IC₅₀ cutoff (Fig. 1 B). The crystal structure of 45-46m2 bound to the clade A/E 93TH057 gp120 revealed that Tyr28_45-46m2(LC) makes a favorable interaction with the Asn276_gp120-attached glycan, rationalizing the observed increase in efficacy compared with NIH45-46^G54W. However, although the Asn276 _gp120 glycan is highly conserved (a potential N-linked glycosylation site at Asn276_gp120 is found in 94.7% of 4279 sequences in the Los Alamos database), it is not required for the formation of infectious particles; thus, virions evolving in the presence of 45-46m2 might include mutations that result in the loss of this glycan. Because 45-46m2 is not solely dependent on this interaction for neutralization, the loss of the Asn276_gp120 glycan would not necessarily lead to resistance to 45-46m2. Indeed, when the Asn276_gp120-attached glycan was eliminated from YU-2 gp120, either by mutating the asparagine (N276S) or the threonine (T278A) in the Asn-X-Ser/Thr glycosylation motif, 45-46m2 still potently neutralized YU-2 (Fig. 8 D), likely due to other favorable interactions with gp120 of 45-46m2 and its parent antibody NIH45-46^G54W, which lead to potent neutralization of YU-2 by the NIH45-46^G54W. Strains that were resistant to NIH45-46^G54W but sensitive to 45-46m2, such as T250-4 (Tables S1 and S2), are likely to have a higher dependency on the interaction with the Asn276_gp120-linked glycan.

Although 45-46m2 showed improved breadth compared with NIH45-46^G54W and other bNAbs, it was still unable to neutralize natural HIV-1 isolates that included substitutions in the N/DNGG consensus sequence within the gp120 D and V5 loops. This consensus feature was identified by noticing that uncommon amino acids replaced highly conserved (≥93%) residues in the gp120 D (Asn279_gp120/Asp279_gp120 and Asn280_gp120) and V5 (Gly458_gp120 and Gly459_gp120) loops of most HIV-1 strains that are resistant to NIH45-46^G54W (Diskin et al., 2011; West et al., 2012) or arose in YU-2 viruses during in vivo therapy experiments under selective pressure from NIH45-46^G54W (Fig. 9 C; Klein et al., 2012). By replacing a highly conserved germline residue, Trp47_45-46m2(HC), with different smaller amino acids, we created mutants such as 45-46m7 that neutralized YU-2 pseudoviruses with substitutions in the N/DNGG motif more effectively than 45-46m2 (Fig. 7 A). However, both 45-46m2 and 45-46m7 were ineffective against YU-2^A281T, a mutant which introduced a potential N-linked glycosylation site at Asn279_gp120. Interestingly, YU-2^A281T exhibited a disrupted replication profile compared with wild-type YU2 and other YU-2 mutants (Fig. 8 B), suggesting a fitness cost associated with an Asn279_gp120-attached glycan. Decreased fitness of HIV-1 strains with Asn279_gp120-attached glycan is consistent with analyses of gp120 sequences in the Los Alamos database indicating that Asn279_gp120-attached glycan is rare or nonexistent in natural HIV-1 strains. Thus, in vivo treatment with 45-46m7 and 45-46m2 may drive HIV-1 into a mutation pathway that results in progressively less fit virions.

To assess mutations in HIV-1 that evolve in response to the new bNAbs, we conducted in vivo therapy experiments in HIV-1_YU-2–infected humanized mice (Fig. 9). Viral escape mutants arising in response to a combination of 45-46m2 and 45-46m7 were compared with previous results for HIV-1_YU-2–infected mice treated with NIH45-46^G54W alone (Klein et al., 2012). Env sequences isolated from 45-46m2/45-46m7–treated viremic mice did not include mutations in the GG portion of the N/DNGG consensus sequence, whereas viruses including these mutations were isolated after NIH45-46^G54W monotherapy (Fig. 9 C, Fig. 10 A, and Table S5 A). Thus, as designed, the new bNAbs successfully restricted the number of pathways that led to escape from their parental bNAb. Indeed, escape from the 45-46m2/45-46m7 combination was mostly limited to substitutions involving introduction of a potential N-linked glycosylation site at Asn279_gp120 and substitutions to compensate for the reduced fitness of virions with an Asn279_gp120-associated glycan (Fig. 9 C). Specifically, the most frequent mutation after 45-46m2/45-46m7 treatment was A281T, a substitution which introduces an Asn279_gp120-attached N-linked glycan, and compensatory mutations to remove the glycan at Asn276_gp120 in order to accommodate the otherwise unfavorable addition of carbohydrate at Asn279_gp120.

Although we were able to restrict potential pathways for HIV-1 escape from bNAbs, these results illustrate the great difficulty of identifying a bNAb or pair of closely related bNAbs targeting the same epitope that can eliminate all escape pathways. Successful antibody-based therapy strategies may therefore require combinations of bNAbs targeting different epitopes. Previous results demonstrated viremic control in HIV-1_YU-2–infected humanized mice using a mix of five bNAbs targeting four different epitopes on the viral spike (Klein et al., 2012). Here, we showed that three antibodies, 45-46m2, 45-46m7, and 10–1074 (Mouquet et al., 2012), targeting only two epitopes, can control viremia in vivo (Fig. 9, A and B); thus, this combination of bNAbs was able to suppress both the parental viral YU-2 viral strain and N/DNGG consensus-based escape mutants.

The humanized mice used in these experiments have defective innate and adaptive immunity. For example, they fail to make endogenous antibody responses to HIV (Klein et al., 2012) and T cell responses are difficult to detect (Baenziger et al., 2006; Manz, 2007). Therefore, HIV-1 infections in these mice may be difficult to control in comparison to macaques or humans with an intact immune system. However, humans carry a higher overall viral load, and are susceptible to infection with a diverse group of HIV-1 strains that could make antibody therapy more challenging than the mouse model. Thus, whether or not antibody therapy can be effective in humans remains to be determined. Nevertheless, the addition of 45-46m2 and 45-46m7 to a mixture of other bNAbs increases the efficacy of antibody-based therapeutics by restricting potential HIV-1 escape pathways and/or selecting mutant virions with reduced fitness.

Modifications of NIH45-46 HC and LC genes were made using QuikChange Lightning (Agilent Technologies) and verified by DNA sequencing (Eton Bioscience). Antibodies were expressed as IgGs using described protocols (Diskin et al., 2010). In brief, secreted IgGs from polyethyleneimine (25kD, linear; Polysciences)-mediated, transiently transfected HEK293-6E cells were captured on protein A or protein G affinity columns (GE Healthcare) and eluted in 100 mM citrate, pH 3.0, and 150 mM sodium chloride. Antibodies subsequently used in neutralization assays were dialyzed into 10 mM citrate, pH 3.0, and 150 mM sodium chloride and adjusted to a concentration of 1 mg/ml. Fab fragments for crystallization and binding assays were obtained by digesting IgGs in 20 mM Tris, pH 8.0, and 150 mM sodium chloride (TBS) with a 1:100 ratio of papain (Sigma-Aldrich) activated in 50 mM phosphate, pH 7.0, 2 mM ethylenediaminetetraacetic acid, and 10 mM cysteine at 37°C until completion of the cleavage (20–60 min, monitored by SDS-PAGE). The Fc was removed by protein A chromatography and Fabs were further purified using Superdex 200 (GE Healthcare) 10/300 Size Exclusion Chromatography (SEC).

The clade A/E 93TH057-derived gp120 core (Zhou et al., 2010; a gp120 construct lacking the V1/V2 and V3 loops) was expressed in insect cells and purified using previously described protocols (Diskin et al., 2011). In brief, supernatants from baculovirus-infected insect cells were collected, buffer exchanged into TBS, and passed through a Ni²⁺-NTA affinity column (GE Healthcare). gp120 was eluted from the column using TBS plus 250 mM imidazole and purified using Superdex 200 16/60 SEC (GE Healthcare) in TBS supplemented with 0.02% (wt/vol) sodium azide.

We used a previously described pseudovirus neutralization assay (Montefiori, 2005) to assess the neutralization potencies of the various antibodies against multiple HIV-1 strains. YU-2 escape mutant pseudoviruses were generated by cotransfecting HEK293T cells with vectors encoding Env and a replication-deficient HIV-1 backbone as previously described (Montefiori, 2005). Neutralization assays were preformed in-house for evaluating antibody mutants against the YU-2 escape mutants (Table S1 and Fig. 6) and by the Collaboration for AIDS Vaccine Discovery (CAVD) core neutralization facility for testing a subset of the antibodies against a large panel of isolates (Tables S2 and S3). Some of the in-house data were derived from neutralization assays that were dispensed automatically by a Freedom EVO (Tecan) liquid handler (IC₅₀ values derived from manual and robotic assays agreed to within two- to fourfold). In all cases, neutralization was monitored by the reduction of a Tat-induced reporter gene (luciferase) in the presence of a three-or fivefold antibody dilution series (each concentration run in duplicate or triplicate) after a single round of pseudovirus infection in TZM-bl cell line (Montefiori, 2005). Antibodies were incubated with 250 viral infectious units at 37°C for 1 h before incubation with the reporter cells (10,000 per well) for 48 h. Luciferase levels were measured from a cell lysate using BrightGlo (Promega) and a Victor3 luminometer (PerkinElmer). Data were fit by Prism (GraphPad Software) using nonlinear regression to find the concentration at which 50% inhibition occurred (IC₅₀ value). For evaluating the neutralization of YU-2 escape mutants, at least two independent experiments were performed. Fig. 6 reports mean IC₅₀ values for the various 45–46 mutants if 0.1 < (IC₅₀1/IC₅₀2) < 10. In cases where the two IC₅₀ values did not agree, additional experiments were performed. The reported IC₅₀ values for NIH45-46^G54W, 45-46m2, and 45-46m7 are means calculated from at least five independent experiments.

Env plasmids for pseudovirus production were provided by B. Hahn (Perelman School of Medicine, University of Pennsylvania Philadelphia, PA), F. McCutchan (The Bill and Melinda Gates Foundation, Seattle, WA), G. Shaw (Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA), D. Montefiori (Duke University Medical Center, Durham, NC), M. Thomson (University of Massachusetts, Amherst, MA), J. Overbaugh (Fred Hutchinson Cancer Research Center, Seattle, WA), R. Swanstrom (University of North Carolina at Chapel Hill, Chapel Hill, NC), L. Morris (Centre for HIV and STIs, National Institute for Communicable Diseases of the National Health Laboratory Services, Johannesburg, South Africa), J. Kim (University College London, London, UK), L. Zhang (School of Medicine, Tsinghua University, Tsinghua, China), D. Ellenberger (Centers for Disease Control and Prevention, Atlanta, GA), and C. Williamson (Institute of Infectious Diseases and Molecular Medicine (IIDMM) and the Division of Medical Virology, University of Cape Town, Cape Town, South Africa).

45-46m2 Fab was purified by Superdex 200 (GE Healthcare) 10/300 SEC in 100 mM citrate, pH 3.0, 150 mM sodium chloride and combined with an equimolar amount of 93TH057 gp120. After concentration using an Amicon (Millipore) spin column, the complex was incubated with 40,000 units of Endoglycosidase H (NEB) per 2 mg gp120 in the absence of detergents at 37°C for 16 h in the manufacturer’s recommended buffer. The complex was further purified using Superdex200 (GE Healthcare) SEC in TBS and concentrated to OD₂₈₀ = 9.5. Data for the structure determination were collected from rod-like crystals grown in a vapor diffusion sitting drop set at a final volume of 2 µl (1:1 protein/reservoir ratio) with 12% (vol/vol) isopropanol, 10% (wt/vol) polyethylene glycol 10,000 kD, and 0.1 M citrate, pH 5.0, at 20°C. The crystals were briefly soaked at 30% (vol/vol) isopropanol, 5% glycerol, 10% (wt/vol) polyethylene glycol 10,000 kD, and 0.1 M citrate, pH 5.0, before flash cooling using liquid nitrogen.

Data to 2.82 Å resolution were collected from a P2₁2₁2₁ 45-46m2–gp120 complex crystal with similar cell dimensions as the NIH45-46–gp120 crystals (Diskin et al., 2011) at the Stanford Synchrotron Radiation Lightsource (SSRL) beamline 12–2 using a Pilatus 6M (Dectris) detector and 0.9537 Å radiation (Table S5). Data were indexed, integrated, and scaled using XDS (Kabsch, 2010; Table S5). Using Phaser (McCoy et al., 2007) and the NIH45-46–gp120 complex (PDB: 3U7Y) as a search model, we found a molecular replacement solution comprising one 45-46m2 Fab and one gp120 in the asymmetric unit. Several rounds of simulated annealing were performed in initial refinement cycles to minimize model bias. The structure was refined using iterative cycles of refinement using the Phenix crystallography package (Adams et al., 2010) and Coot (Emsley et al., 2010) for manual rebuilding. To facilitate refinement at 2.82 Å, we restrained the model by using the NIH45-46–gp120 structure as a reference and applying secondary structure restraints. The final model (R_free = 23.1%, R_work = 19.3%) consists of 5,998 protein atoms, 242 carbohydrate atoms, and 23 water molecules. 95.63, 4.1, and 0.26% of the residues are in the favored, allowed, and disallowed regions, respectively, of the Ramachandran plot. The first glutamine of the 45-46m2 HC was modeled as 5-pyrrolidone-2-carboxylic acid. Disordered regions that were not modeled include residues 1–2 and 210 (the C terminus) of the 45-46m2 LC, residues 133–136 and 219–221 (the C-terminus) of the 45-46m2 HC, and residues 302–308 (a short linker substituting for the V3 loop), residues 397–408 (a total of 6 residues from V4), and the 6x-His tag of 93TH057 gp120.

Structures were analyzed and figures were prepared using PyMol (Schrödinger). Buried surface areas were was calculated using a 1.4 Å probe using Areaimol as implemented in CCP4i package (Collaborative Computational Project, Number 4, 1994).

SPR data were collected using a Biacore T200 instrument (GE Healthcare). We used primary amine coupling chemistry to immobilize 1000 resonance units (RU) of the Fabs of NIH45-46, NIH45-46^G54W, or 45-46m2 in 10 mM acetate, pH 5.0, at a concentration of 0.2 µM to a CM5 sensor chip as described in the Biacore manual. Flow channel 1 was mock coupled and served as a blank subtraction channel. gp120 protein was injected as a twofold dilution series (500–31.2 nM) at a flow rate of 80 µl/min at 25°C in 20 mM HEPES, pH 7.0, 150 mM sodium chloride, and 0.005% (vol/vol) P20 surfactant, and sensor chips were regenerated using 10 mM glycine, pH 2.5. A 1:1 binding model was fit to the blank-subtracted data using the Biacore analysis software to derive kinetic constants (k_a and k_d; on and off rates) that were subsequently used to calculate affinities (K_D; equilibrium dissociation constant).

Replication experiments were performed as described previously (Neumann et al., 2005; Sather et al., 2012) using wild type YU-2 and three point mutants in gp120 designated as YU-2^N279K, YU-2^N280D, and YU-2^A281T. The entire gp160 portion of each env variant was inserted into the TN6 replication competent viral backbone, and each construct was transfected into 293T cells to produce infectious virions. Stimulated PBMCs were prepared from whole human blood by Ficoll gradient separation, followed by 72 h of stimulation by culturing in complete RPMI containing 2 µg per ml IL-2 and 3 µg/ml phytohemagglutinin (PHA). 15 × 10⁶ stimulated PBMCs were infected for 3 h with viral inoculum containing the equivalent of 12.5 pg p24. After inoculation, the cells were resuspended in fresh complete RPMI/IL-2 media at a density of 3 × 10⁶ cells per ml. At 2–3-d intervals, half of the culture supernatant was harvested and replaced with fresh media. Harvested supernatants were assayed for p24 content by capture ELISA (Zeptometrix). During the culture period, the cultures were monitored to ensure that viability remained >90%.

HIV-1 escape experiments were performed in HIV-1_YU2–infected humanized mice as previously described (Klein et al., 2012). In brief, nonobese diabetic Rag1^−/− IL2Rγ^NULL (NRG) mice (The Jackson Laboratory) were reconstituted with fetal liver–derived hematopoietic stem cells and infected with HIV-1_YU2 (57.5 ng p24). Mice with viral loads >4 × 10³ copies/ml at 14–17 d after infection were included in treatment experiments. Antibody-treated mice were injected subcutaneously with 1.5 mg 45-46m2 and 1.5 mg 45-46m7 every 2 d, and mice receiving 10–1074 were injected with 0.5 mg antibody twice per week. All experiments were performed with authorization from the Institutional Review Board and the IACUC at the Rockefeller University.

Viral load measurements and sequence analysis of HIV-1 gp120 were performed as previously described (Klein et al., 2012). In brief, total RNA was extracted from 100 µl EDTA-plasma using the QiaAmp MinElute Virus Spin kit as per the manufacturer’s protocol. Viral RNA was detected by quantitative reverse-transcriptase PCR using a Stratagene Mx3005P real-time thermal cycler. HIV-specific forward and reverse primer sequences were 5′-GCCTCAATAAAGCTTGCCTTGA-3′ and 5′-GGCGCCACTGCTAGAGATTTT-3′, respectively. An internal probe (5′-AAGTAGTGTGTGCCCGTCTGTTRTKTGACT-3′) contained a 5′ 6-carboxyfluorescein reporter and internal/3′ ZEN-Iowa Black FQ double-quencher (Integrated DNA Technologies, Inc.). The reaction mix was prepared using the TaqMan RNA-to-Ct 1-Step kit (Applied Biosystems). Cycle threshold values were converted to viral loads using an HIV-1(NL4/3-YU-2) viral preparation of known copy number as a standard.

For gp120 sequencing, viral cDNA was generated from extracted viral RNA (described above) using Superscript III Reverse transcription (Invitrogen) and amplified by gp120-specific nested PCR using the Expand Long Template PCR system (Roche). PCR amplicons were gel purified, cloned into pCR4-TOPO (Invitrogen), transformed into One-Shot TOP10 cells (Invitrogen), and sequenced using the insert-flanking primers M13F and M13R. Sequence reads were assembled using Geneious Pro software version 5.5.6 (Biomatters Ltd) and aligned to HIV-1^YU2 gp120 (accession no. M93258). Manual edits to sequence assemblies and alignments were performed in Geneious. gp120 residues were numbered according to HXB2, as determined by the Los Alamos Sequence Locator tool (http://www.hiv.lanl.gov/content/sequence/LOCATE/locate.html).

The supplemental material of this paper includes tables of IC₅₀ values for in vitro neutralization assays against a panel of 118 viruses, a table with crystallographic statistics, and a table listing isolated sequences of gp120s recovered from HIV-1 isolated from infected humanized mice after antibody therapy experiments.

We thank B. Hahn, F. McCutchan, G. Shaw, D. Montefiori, M. Thomson, J. Overbaugh, R. Swanstrom, L. Morris, J. Kim, L. Zhang, D. Ellenberger, and C. Williamson for contributing the HIV-1 envelope plasmids used in our neutralization panel, Alexander Ploss and Charles M Rice for providing mice for therapy experiments, the Caltech Protein Expression Center, and members of the Bjorkman laboratory for critical reading of the manuscript.

R. Diskin, A.P. West, M.C. Nussenzweig, and P.J. Bjorkman have pending patent applications with the U.S. Patent and Trademark Office, patent numbers U.S. CIT-5930-P3 and CIT-6235-P, titled “Improving the Potency and Breadth of Anti-HIV Antibodies–NIH45-46S28Y(LC) with NIH45-46G54W(HC)” and “Overcoming HIV Pathways for Escape using Rationally Designed Anti-HIV Antibodies,” respectively. The reagents are available with a Materials Transfer Agreement. R. Diskin is the Tauro Career Development Chair in Biomedical Research. This work was supported by Collaboration for AIDS Vaccine Discovery (CAVD) grants with support from the Bill and Melinda Gates Foundation (grant 1040753 to P.J. Bjorkman and grant 1032144 to M.S. Seaman); the NIAID of the National Institutes of Health under Award numbers P01AI100148 (P.J. Bjorkman and M.C. Nussenzweig) and UM1AI100663 (M.C. Nussenzweig); NIH grant R01AI081625 (L. Stamatatos)—the content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH; and the Molecular Observatory at Caltech supported by the Gordon and Betty Moore Foundation. Operations at Stanford Synchrotron Radiation Lightsource are supported by the U.S. Department of Energy and NIH. Coordinates and x-ray crystallographic data for the 45-46m2/gp120 complex have been deposited in the PDB under accession code 4JKP.

The authors declare no competing financial interests.

Author contributions: R. Diskin, M.C. Nussenzweig, and P.J. Bjorkman conceived the study, R. Diskin performed structural determinations, analyses, and design, P.M. Marcovecchio and H. Gao produced and purified proteins, T. Lee, M.S. Seaman, D.N. Sather and L. Stamatatos performed neutralization assays, D.N. Sather and L. Stamatatos performed viral replication analysis, F. Klein, A. Halper-Stromberg, and J.A. Horwitz performed in vivo therapy experiments and analyzed escape mutants, A.P. West helped with data analysis, and R. Diskin and P.J. Bjorkman wrote the paper with contributions from all coauthors.