Interaction of the gp120 V1V2 loop with a neighboring gp120 unit shields the HIV envelope trimer against cross-neutralizing antibodies

Past investigations on the influence of the V1V2 loop on steering neutralization activity have been restricted to few, mostly highly neutralization-sensitive isolates (HxB2, LAI, SF162, R3A, and YU-2; Wyatt et al., 1993, 1995; Cao et al., 1997; Stamatatos and Cheng-Mayer, 1998; Stamatatos et al., 1998; Sullivan et al., 1998; Saunders et al., 2005; Laakso et al., 2007; Bontjer et al., 2009) and thus left a definite verdict on the potency of the V1V2 loop shielding effect against primary viruses pending. In this study, we generated a panel of six infection-competent V1V2 loop–deleted Env’s of tier 1 (NL4-3 and SF162), tier 2 (JR-FL*, RHPA, and REJO), and tier 3 (ZA110.c10.14) subtype B isolates by inserting the linker sequence Asp-Ala-Gly (DAG) as described previously (Fig. S1 A and Table S1; Bontjer et al., 2009). WT and matching V1V2-deleted viruses were then compared for sensitivity to neutralization by plasma samples from 83 individuals with chronic HIV infection from different genetic subtypes (Fig. 1 A and Tables S3 and S4). In the absence of the shielding effect of the V1V2, potent cross-neutralization activity was detected in all probed patient plasma samples with some neutralization titers exceeding those of WT viruses by more than four orders of magnitude. Of note, the neutralization-sensitive, CXCR4-using, T cell line–adapted virus NL4-3 portrayed the least increase (6.9-fold) in susceptibility to neutralization upon removal of its V1V2 domain. Although absolute neutralization titers against V1V2-deleted SF162 matched those of the other four R5 viruses, the fold increase in neutralization sensitivity of SF162 upon V1V2 deletion was also lower (40-fold) as SF162 WT was efficiently inhibited by the patient plasma samples. Plasma neutralization titers against the V1V2-deleted isolates were tightly correlated (Table S5), suggesting that common antibody types were elicited across subtypes that target conserved, highly neutralization-sensitive epitopes usually shielded by the V1V2 domain. Interestingly, although gp120 binding and neutralization efficacies correlated (Table S5), only modest increases in binding efficacy of V1V2-deleted gp120 were evident (Fig. 1 B and Table S4). The latter strongly supports the notion that the V1V2 domain exerts its full shielding effect only in the context of the intact viral spike and not on monomeric gp120. Nevertheless, a comparison of binding and neutralizing antibodies in the context of polyclonal plasma has to be interpreted with caution as neutralizing antibodies may quantitatively represent only a small fraction of the overall binding antibodies in patient sera.

To understand which neutralizing antibody epitopes are protected by the V1V2, we probed the shielding effect of the V1V2 domain for mAbs targeting diverse epitopes within gp120 and gp41 by comparing their ability to neutralize WT and V1V2-deleted viruses or to bind to respective monomeric gp120 proteins (Fig. 2 and Tables S1 and S2). Of all the epitope specificities probed, V1V2 shielding had the most dramatic effect on V3 loop antibodies with up to 240,000-fold higher neutralization titers against a V1V2-deleted virus. Masking by the V1V2 of both V3 loop and CD4bs was dependent on the trimeric state of the Env, as binding efficacies to WT and mutant gp120 proteins were essentially identical (Fig. 2). gp41-specific reagents were included as controls as direct V1V2 shielding of these epitopes is unlikely. Accordingly, only modest effects on their inhibitory capacity were evident, confirming that V1V2 deletion did not inflict gross changes in entry efficacy or trimer stability that would impair further analysis using these mutants. Neutralization potency of CD4i-specific mAbs increased in comparison only moderately, and no or only modest effects were evident for the glycan-specific mAb 2G12. Interestingly, CD4bs-directed inhibitors were differentially affected by V1V2 shielding. Depending on the isolate probed, CD4bs mAbs b12 and VRC01 either neutralized with identical or decreased activity in the absence of the V1V2, highlighting that these antibodies have found a unique way to bypass or even take advantage of the V1V2 domain. All other CD4bs agents probed (tetrameric CD4-IgG2, mAbs b6 and 1F7) greatly improved neutralization activity upon V1V2 deletion for those isolates that were not inhibited as efficiently as WT, indicating a substantial influence of the V1V2 domain on access to their respective epitopes. In sum, our data suggest that the ability of b12 and VRC01 to bypass or take advantage of the V1V2 in binding to the trimeric spike likely accounts for their broad and potent activity. We next compared binding of mAbs to monomeric WT and V1V2-deleted gp120 proteins to obtain initial insight whether the V1V2 masking of their respective epitopes occurs in the context of the trimer or monomer (Fig. 2). Notably, only CD4i mAbs improved their binding capacity to the V1V2-deleted gp120 proteins, suggesting that in the absence of CD4 triggering, access to CD4i epitopes on monomeric gp120 is restricted by the V1V2 loop. Alternatively, the intrinsic equilibrium between gp120 conformations that do not present the CD4i domain versus such that do may favorably be shifted toward the latter in the absence of the V1V2 loop.

Accessibility to all other probed epitopes on monomeric gp120 was not influenced by the V1V2 loop, and binding of mAbs to WT and V1V2-deleted monomeric gp120 occurred at comparable levels. Of particular note, this was also true for inhibitors targeting the CD4bs and the V3 loop, which were neutralized with a several magnitude higher potency in the absence of the V1V2 loop. Thus, effective shielding of these domains by the V1V2 loop entirely depends on the organization of the Env proteins in the native, trimeric spike.

Our data highlight a remarkably potent effect of the V1V2 shielding on V3 loop–specific mAbs that was comparable in magnitude with the increase in neutralization of V1V2-deleted viruses by patient plasma (Fig. 1 A). It was thus prudent to probe whether V3 loop–specific antibodies are the dominant category of cross-neutralizing antibodies elicited in HIV infection that are counteracted by the V1V2 loop. To this end, we compared the neutralization activity of subtype B plasma samples against V1V2-deleted viruses in the presence or absence of competing peptide encompassing the tip of their cognate V3 loop (Fig. 3, A and B; and Table S1). The outcome of these experiments was astonishingly clear. Cross-neutralization activity against the V1V2-deleted R5 viruses ΔSF162, ΔJR-FL*, and Δ10.14 diminished by several orders of magnitude in the presence of the competing V3 loop peptide in the majority of the probed plasma samples. Thus, V3 loop–reactive antibodies are commonly elicited in HIV infection and are counteracted by the V1V2 domain. Of note, a >90% proportion of the plasma neutralization activity was caused by V3 loop–directed antibodies in 26, 24, and 14 of the 28 probed plasma samples for ΔSF162, ΔJR-FL*, and Δ10.14, respectively (Fig. 3 C). In sharp contrast to the R5 isolates, the NL4-3 V3 loop peptide in large failed to outcompete NL4-3–directed cross-neutralizing plasma activity. Neutralization of NL4-3ΔV1V2 virus in the absence of V3 loop competition was significantly lower compared with all other probed ΔV1V2 viruses (one-way analysis of variance post-test, Tukey’s multiple comparison). However, the fraction of non-V3 loop–reactive antibodies against NL4-3 was comparable with those of the other probed viruses (no significant difference, same test), underlining the fact that V3 loop antibodies do not (or only marginally) contribute to NL4-3 cross-neutralization activity commonly elicited in vivo. The latter likely also accounts for the comparatively low plasma cross-neutralization activity against V1V2-deleted NL4-3 (Fig. 1 A). Whether this reflects a lack of cross-reactivity with specifically the V3 loop of NL4-3 or whether X4 V3 loop–reactive antibodies are generally less frequently elicited in vivo needs to be determined. Humoral immune responses that cross-react with subtype B/R5 V3 loops are commonly elicited in infections of other subtype as JR-FL V3 loop–specific peptide potently competed off JR-FL* cross-neutralizing antibodies in subtypes A, C, AE, and AG plasma (Fig. 3 D). Of note, we found that plasma neutralization activity against WT SF162 is in part mediated by V3 loop–specific antibodies (Fig. S1, B and C). Yet, loss of V1V2 shielding increases V3 loop mAb sensitivity of SF162 substantially (Fig. 3), indicating that V1V2 shielding is, albeit at a lower level, also effective in the context of tier 1 isolates.

Unraveling the molecular basis of the shielding effect conferred by the V1V2 loop in the context of the Env spike would provide much needed knowledge on the loop interaction within trimers and overall quaternary trimer architecture not yet available from crystal structure and cryoelectron microscopy analysis. In principle, two mechanisms can be envisaged and have been suggested (Kwong et al., 2000; Chen et al., 2005; Zhu et al., 2006; Liu et al., 2008a; White et al., 2010; Wu et al., 2010): the V1V2 domain can either protect domains in the gp120 subunit it originates from (intrasubunit, self protection) or alternatively act by shielding domains in an adjacent gp120 unit within the trimeric spike (intersubunit, neighboring protection; Fig. 4). Self protection by the V1V2 loops would thereby depend on a substantially altered conformation of the gp120 subunit in the context of the monomer as opposed to the trimer, as only the latter is protected from neutralization by the V1V2 loop (Figs. 1, 2, and 4).

In this study, we devised a mathematical model and an experimental strategy that allows us to distinguish between neighboring and self protection modes of the V1V2 domain. Our chief interest was to define how the V1V2 creates protection for the V3 loop. The basic principle of our approach is to probe the V1V2 shielding effect on the V3 loop in the context of heterogenous Env trimers that are composed of two different types of Env proteins (Fig. 5 A). One WT Env protein (sensitive, harboring V1V2) contains the epitope of the neutralizing V3 loop antibodies 19b and 1-79. The other Env does not contain the epitope for these V3 loop mAbs and was in addition V1V2 deleted (resistant, ΔV1V2). By cotransfecting altering ratios of the two Env-expressing plasmids, cell populations with varying compositions of Env trimers can be generated that consequently differ in the fraction of Env proteins sensitive to V3 loop mAb binding. Depending on whether self or neighboring protection occurs, distinct binding scenarios will be observed (Fig. 5 A).

To differentiate between the two alternative protection scenarios, we devised a mathematical model that predicts the mean number of antibodies bound to a trimer as a function of the fraction of Env proteins carrying the V3 loop mAb–sensitive epitope (Fig. 5 B). Because of the random nature of trimer formation, spikes that contain zero, one, two, or three mAb-sensitive epitopes will form. However, binding of mAb to sensitive V3 loops will only occur on heterotrimers where V1V2 shielding is not effective and will thus depend on whether V1V2 mediates neighboring or self protection (Fig. 5 A). Accordingly, our mathematical model predicts distinct V3 loop mAb–binding patterns in cell populations expressing different compositions of heterotrimers (Fig. 5 B). If the V1V2 loop masks the V3 loop mAb–binding site of the Env protein it originates from, no mAb binding should be observed (Fig. 5 B, green line). In contrast, if the V1V2 loop masks an antibody-binding site of a neighboring Env protein, binding to several trimer configurations can occur (Fig. 5 A), and our model predicts a hump-shaped curve (Fig. 5 B, blue line). Thus, using our mathematical model, we predicted distinct, unmistakably discernable binding reactivity patterns for the self and neighboring protection scenarios.

We next probed our model experimentally and assessed mAb 19b binding to 11 different cell populations expressing varying compositions of heterotrimers composed of Env’s NL4-3ΔV1V2 (lacking the V3 loop mAb epitope) and JR-FL* (harboring the V3 loop mAb epitope; Fig. 5 C). The data provided a strikingly clear pattern: 19b binding followed a hump-shaped curve, suggesting that neighboring protection is the mode by which the V1V2 domain shields the V3 loop. The same pattern of reactivity was observed for a second V3 loop mAb, 1-79. Of note, our approach relies on a random formation of heterotrimers. As the experiment depicted in Fig. 5 C necessitates two genetically divergent Env proteins to associate into trimers, we repeated these experiments with two alternate Env pairings that only differed in their V3 loop sequence and the presence and absence of the V1V2 loop (Fig. 5, D and E). In both instances, we again observed a hump-shaped reactivity pattern for mAbs 19b (Fig. 5, D and E) and 1-79 (Fig. 5 E), supporting the notion that both genetically divergent and closely related Env subunits form heterotrimers.

We further verified the V1V2 protection modus by assessing an alternate trimer scenario (setup 2) in which the 19b epitope is exposed on the V1V2-deleted Env protein, whereas the WT Env is resistant to 19b (Fig. 6, A–C). In this scenario, our model predicts reactivity patterns that are less divergent as binding in both settings is possible. If the self protection scenario holds true, our model predicts a straight line, and if the neighboring protection holds true, we predict a convex curve. Experimental probing retrieved a convex-shaped reactivity pattern, again suggesting neighboring protection (Fig. 6 C).

Although our data carved out neighboring protection as the mode by which the V1V2 domain protects the V3 loop, we sought to further ascertain that our experimental approach is valid and is indeed capable of distinguishing between neighboring and self protection. Our approach depends on two key parameters: for one that heterotrimers form and, second, that heterotrimers consisting of WT and V1V2-deleted subunits adopt a conformation that resembles the WT-enabling V1V2 shielding to occur. Should the trimer assemble in a more open formation when WT and V1V2-deleted Env’s are mixed, V1V2 shielding could be lost, allowing mAbs to access the V3 loop. To verify that heterotrimers form and have the appropriate conformation, we performed a control experiment (Fig. 7). Here, two identical Env’s with a sensitive V3 loop that only differed in the presence/absence of the V1V2 loop were paired. When heterotrimers adopt the appropriate conformation and V1V2 shielding is intact, then irrespective of whether neighboring or self protection occurs, a linear increase in V3 loop mAb–accessible sites should be observed in this setup (Fig. 7, blue and green straight lines). In contrast, if heterotrimers assemble in a formation that jeopardizes V1V2 shielding, the V3 loop would be accessible in all heterotrimers, and the reactivity pattern in this case would dramatically shift to the left (Fig. 7 B, gray curve). Experimental probing revealed a straight line reactivity pattern, confirming that heterotrimers with intact shielding must form. These results verify that our model assumptions are correct and that neighboring and self protection can be dissected as described in setups 1 and 2 (Fig. 7).

To exclude the possibility that nonequal Env expression confounds our findings, gp120 expression levels on the various cell populations were monitored in all shielding experiments by assessing reactivities of the V3 loop mAb 19b and the carbohydrate-specific mAb 2G12 to control for V3 epitope and total gp120 Env expression, respectively (Fig. S2). Although Env expression varied to some extent between experiments and depended on the Env pairs studied, sufficient expression allowing assessment of mAb binding by FACS was achieved throughout. Fluctuations in Env expression levels are however unavoidable (Fig. S2; comparison of 2G12 binding to 100% sensitive and 100% resistant Env). As the observed differences were relatively small, we chose not to normalize the data by correcting for Env expression, as actual differences in Env expression and influences of differential mAb affinity for the heterologous Env pairs are difficult to tease apart. Importantly, variation in total Env expression levels was in all instances associated with a slight overexpression of one of the Env’s, never with a decrease for cell populations that contain higher fractions of heterotrimers. From this, we conclude that the generated heterotrimers were stable and not genuinely prone to spontaneous gp120 shedding. The latter was important to verify as V1V2 shielding would not be effective on partially shed trimers and would report false positive signals in our system. We found that JR-FL*WT was expressed at slightly higher levels than NL4-3ΔV1V2 in most experiments, which corresponded with a modest rightward skew of the hump-shaped binding curves (Fig. 5 C). In an alternate pairing, JR-FL*ΔV1V2^V3NL4-3 was expressed at higher levels than JR-FL*WT, which is consistent with a modest leftward skew of the binding pattern observed in Fig. 5 E. Thus, nonequal expression levels may in part cause slight, off-center skewing of the hump-shaped reaction patterns.

In a series of additional modeling approaches, we verified the influence of further parameters on our shielding assay to rule out the possibility that other confounding factors exist that affect our analysis and conclusions. In a first approach, we probed whether nonrandom trimer formation (i.e., preferential homotrimer formation) could affect our experimental approach. However, mathematical modeling of the potential influence of homotrimers on the hump-shaped reactivity pattern confirmed that, regardless of the frequency by which homotrimers are generated, no shift in the reactivity pattern can occur as mAb binding in this scenario completely depends on heterotrimer formation (Fig. S3 A). In this sense, our experimental approach provides also a formal proof that heterotrimer formation occurs.

Although we found that skewing of the hump-shaped reaction pattern cannot be explained by preferential homotrimer formation, this could affect the alternate shielding scenario (setup 2) to some extent and needs to be factored in when solely this protection scenario is used for assessments (Fig. 6 and Fig. S3 B). Of note, the influence of homotrimer formation cannot be assessed in the protection scenario studied in setup 3 (Fig. 7 and Fig. S3 C).

Besides intact trimers, partially shed spikes composed of gp120 dimers or monomers or solely gp41 stumps are expressed on infected or Env-transfected cells and virions (Moore et al., 1990; Moore et al., 2006; Ruprecht et al., 2011). In the context of monomers or dimers, V3 loops will not be, or only imperfectly, shielded (Fig. 2). As V3 loop mAb binding to monomers and dimers could influence our readout, we modeled the potential influence of partially shed spikes in our shielding analysis. We found that the presence of monomers and dimers induces a small rightward skew of the hump-shaped binding pattern (Fig. S3 D), which could add to the observed skewing in some of our experiments (Fig. 5 C). Monomer and dimer binding could also potentially influence the binding pattern in the alternate shielding scenarios (Figs. 6 and 7), although to an even smaller extent (Fig. S3, E and F).

In summary, we conclude that our experimental approach to uncover the mechanistic basis of V1V2 loop shielding is very robust. Small deviations from the in silico–derived reactivity patterns can occur when shielding is probed in vitro. However, these differences are marginal and are accounted for by differences in Env expression levels and the presence of monomeric and dimeric gp120 on the transfected cells.

Advances in recent years have highlighted the significant impact of the V1V2 domain on steering the sensitivity to antibody neutralization (Cao et al., 1997; Chackerian et al., 1997; Fox et al., 1997; Morikita et al., 1997; Stamatatos and Cheng-Mayer, 1998; Ly and Stamatatos, 2000; Losman et al., 2001; Johnson et al., 2003; Cole et al., 2004; Pinter et al., 2004; Pugach et al., 2004; Krachmarov et al., 2005, 2006; Saunders et al., 2005; Sagar et al., 2006; Shibata et al., 2007; Ching et al., 2008). Collectively, these studies point toward very effective epitope masking by the V1V2 loop. The observation that currently circulating HIV strains have evolved their V1V2 domain to gain in neutralization resistance (Bunnik et al., 2010) supports the notion that the V1V2 loop provides an essential barrier for neutralizing antibodies elicited in natural infection. We reasoned that the protective effect of the V1V2 must be dominant to allow for such a focused evolution. Indeed, as we show in this study, the V1V2 domain has a tremendous capacity to protect against the humoral immune response elicited in vivo. The effect was particularly pronounced for more neutralization-resistant, primary tier 2 and 3 isolates, whereas neutralization-sensitive strains, as used in previous assessments of the V1V2 shielding effect (Morikita et al., 1997; Ly and Stamatatos, 2000; Losman et al., 2001; Pinter et al., 2004; Pugach et al., 2004; Krachmarov et al., 2005, 2006; Ching et al., 2008; Bontjer et al., 2009), underappreciated its magnitude. Unmasking of V1V2-protected domains unraveled a potent neutralization activity ubiquitously elicited in chronic infection across genetic subtypes. Most intriguingly, we found that this neutralizing activity was to a large extent conferred by V3 loop–specific antibodies, which cross-reacted with genetically divergent R5 V3 loop domains but not an X4 V3 loop. As the investigated chronic infection cohort did not include late stage infection, the lack of X4 cross-reactivity may reflect the predominance of R5 strain infections at the explored disease stage. mAb studies supported the tremendous impact of the V1V2 on shielding the V3 loop (Ly and Stamatatos, 2000; Losman et al., 2001; Krachmarov et al., 2005, 2006). However, our results also highlight shielding effects of the V1V2 on other neutralization-sensitive domains, further supporting its role as a broadly active barrier for neutralization in the context of the native trimer.

To date, the molecular basis of the V1V2 shielding capacity remains uncertain, largely as high-resolution structural information of the loop orientation in the context of intact HIV trimers is not available. Recent cryo-ET studies suggest that V1V2 and V3 loops may interact in the apex of trimeric Env spikes (White et al., 2010; Hu et al., 2011). In this study, we report on a structure–function analysis based on a specifically devised mathematical model and an experimental strategy to probe the theoretical assumptions. By studying V3 loop accessibility in the context of heterogeneous Env trimers composed of gp120 units with or without the V1V2 domain, we were able to demonstrate that the V1V2 loop protects the V3 loop of a neighboring gp120 in the context of the native trimer. Thus, the HIV-1 Env trimer adopts a quaternary structure that effectively shields highly cross-neutralization–sensitive domains by enforcing intersubunit contact through the V1V2 domain.

Further analysis will be needed to define whether and how increases in length and glycosylation within the V1V2 steer the efficacy of the intersubunit interaction and thus potentially improve shielding. However, a meaningful dissection of the contribution of sequence changes, glycosylation patterns, and length variation on V1V2 functionality will depend on the availability of crystal structure data of the V1V2 domain, as naturally occurring and artificially induced structural rearrangements upon sequence variation must be carefully evaluated in this type of analysis. The latter holds particularly true when the contribution of glycosylation is evaluated as removal of the specific potential N-linked glycosylation site necessitates changes in the amino acid backbone, which by themselves may cause structural rearrangements. In our current study, we explored the V1V2 shielding modus in the context of a neutralization-sensitive, tier 1 Env, NL4-3, and a relatively neutralization-resistant, tier 2 Env, JR-FL. Although V1V2 domains of these Env’s differ in their degree of glycosylation, both were capable of protecting the adjacent gp120 molecule’s V3 loop in the context of the trimer. The latter supports the notion that the V1V2 domain has an intrinsic capacity to confer protection throughout all disease stages and is, albeit at a lower level, also active in the setting of tier 1 viruses (Fig. S2).

Although our study does not provide data on the physical structure of the V1V2 domain per se, it offers new information on the quaternary organization of the V1V2 loop in the context of the Env trimer. The results of our analysis fit well with currently available structural data of gp120 and proposed models of the quaternary geometry of gp120 subunits within the trimeric spike (Kwong et al., 1998, 2000; Chen et al., 2005; Huang et al., 2005; Zanetti et al., 2006; Zhu et al., 2006; Liu et al., 2008a; Chen et al., 2009; Diskin et al., 2010; Finzi et al., 2010; Kong et al., 2010; Pancera et al., 2010; White et al., 2010; Xiang et al., 2010; Hu et al., 2011). A potential interaction of the V1V2 domain with the adjacent monomer’s V3 loop has been considered for long (Kwong et al., 2000; Chen et al., 2005) but until to date could not be confirmed in the absence of the definition of a high-resolution trimer structure. Of note, results from recent cryo-ET studies of HIV Env spikes appear compatible with an interaction of V1V2 and V3 loops of neighboring subunits on the tip of the spike, but these analyses yet lack in resolution to unambiguously confirm such an interaction (Liu et al., 2008a; White et al., 2010; Hu et al., 2011). Our study confirms the proximity of the V1V2 and V3 loops of neighboring gp120 subunits within the trimeric spike and thus aids in validating the current models of Env trimer structure. By formally defining the orientation of the V1V2 loop toward the neighboring subunit in the trimer, our data close an important gap in the understanding of the structural architecture of the trimeric spike and provide a molecular basis on how the V1V2 masks neutralization-sensitive domains.

Antibodies used in this study were made available by J. Robinson (Tulane University Medical Center, New Orleans, LA), M. Nussenzweig (The Rockefeller University, New York, NY), D. Burton (The Scripps Research Institute, La Jolla, CA), H. Katinger (Polymun Scientific, Vienna, Austria), S. Zolla-Pazner (New York University School of Medicine, New York, NY), J. Mascola (Vaccine Research Center, National Institutes of Health, Bethesda, MD), and the AIDS Research and Reference Reagent Program (Division of AIDS, National Institute of Allergy and Infectious Diseases, National Institutes of Health). The precise origin and epitope specificities of mAbs used in this study are listed in Table S2. Plasma samples were obtained from individuals with chronic HIV-1 infection (>6 mo infected) enrolled in (a) the Swiss HIV Cohort Study (Schoeni-Affolter et al., 2010), (b) the Zurich Primary HIV Infection Study (Rieder et al., 2010), and (c) the Swiss Spanish Treatment Interruption trial (Trkola et al., 2004). All experiments were approved by the ethics committee of the University Hospital Zurich, and written informed consent was obtained from all individuals. Detailed patient demographics are listed in Table S3. Peptides were provided by J. Robinson and T. Riedel (University of Zurich, Zurich, Switzerland).

293-T and TZM-bl cells were obtained from the American Type Culture Collection and the National Institutes of Health AIDS repository, respectively, and cultured as described previously (Rusert et al., 2009).

All HIV Env genes used in this study have been previously described (Table S1) with the exception of the Env ZA110.C.10.14, which was previously isolated in our laboratory from a rapidly progressing individual during chronic infection (Table S1).

An overview of all mutant Env genes generated in the frame of this study is provided in Table S1. For the construction of V1V2-deleted viruses, the natural restriction sites DraIII (upstream) and StuI (downstream) of the V1V2 were used to substitute the original V1V2 domain with a 3-aa linker sequence DAG as described previously (Fig. S1 A; Bontjer et al., 2009). Env’s were cloned into the expression vector pcDNA 3.1. DraIII and StuI restriction sites in pcDNA 3.1 were removed before construction of the V1V2 mutant viruses. The integrity of the derived mutant Env’s was verified by sequencing, and functionality was ascertained in a pseudotype virus infection assay. All Env mutants proved infection competent in pseudotype infection upon V1V2 deletion/DAG insertion. As expected, infectivity upon V1V2 deletion decreased to some extent compared with the corresponding WT isolates. However, for all Env’s described here, this reduction was <10-fold, and infectivity was in all cases sufficient for inhibitor experiments. The only exception was the Env of JR-FL as this protein did not tolerate the substitution by its cognate V1V2 linker sequence. However, we were able to restore its infectivity by introducing a D197N mutation that generates an additional potential N-glycosylation site downstream of the V1V2. We designated this virus as ΔV1V2JR-FL* and constructed a corresponding WT JR-FL derivative harboring the D197N mutation (designated JR-FL*). This modification proved to have no influence on the neutralization sensitivity of the virus as we found that both JR-FL and JR-FL* had identical sensitivity to neutralization by mAbs and patient plasma.

To generate V3 loop–swapped NL4-3 and JR-FL* Env’s, an overlap extension PCR was used. The NL4-3 and JR-FL* Env sequences upstream of the V3 loop (NL4-3 primers, 5′-GAGACAGCGACGAAGAGCTCCTC-3′ and 5′-GGGTCTTGTACAATTAATTTCTACAG-3′; JR-FL* primers, 5′-CTCAAGACAGTCAGACTCATCTTGC-3′ and 5′-GTATTGTTGT­TGGGTCTTGTA­CAATTAATTTC-3′), the individual V3 loops (NL4-3 primers, 5′-GTACAAGACCCAACAACAATACAAG-3′ and 5′-GTTACAATGTGCTTGTCTCATATTTCC-3′; JR-FL* primers, 5′-CTGTAGAAATTAATTGTACAAGACCC-3′ and 5′-GCTCTACTAATGTTACAATGTGC-3′), and the NL4-3 and JR-FL* Env sequences downstream of the V3 loop (NL4-3 primers, 5′-GCACATTGTAACATTAGTAGAGC-3′ and 5′-TTTTGACCACTTGCCACCCATCTTATAGC-3′; JR-FL* primers, 5′-TATGAGACAA­GCACATTGTAA­CATTAGTAGAGC-3′ and 5′-ACTTTTTGACCACTTGCCACCC-3′) were amplified in separate PCR reactions. Subsequently, the NL4-3 and JR-FL* Env upstream fragments were mixed with the desired V3 loop fragment and subjected to a first overlap extension PCR followed by purification of the fusion products by gel extraction. The purified fusion products were subjected to a second overlap extension PCR with the Env downstream fragments that correspond to the upstream fragment yielding the desired chimeric Env gene. The final fusion products were gel purified and cloned into pcDNA3.1. The sequence of the derived mutant Env’s was verified, and epitope accessibility was probed by FACS. Env JR-FL*ΔV1V2_V3^NL4-3 proved not infectious, whereas NL4-3WT_JR-FL^V3 was infectious in the context of pseudovirus infection.

Env-pseudotyped viruses were prepared by cotransfection of 293-T cells with the plasmids encoding the respective Env genes and the luciferase reporter HIV vector pNLluc-AM as described previously (Rusert et al., 2009).

The neutralization activity of mAbs and patient plasma was evaluated on TZM-bl cells as described previously (Rusert et al., 2009). Virus input was chosen to yield virus infectivity corresponding to 5,000–20,000 RLU (relative light units) in the absence of inhibitors. The antibody concentrations or reciprocal plasma titers causing 50% reduction in viral infectivity (inhibitory concentration IC₅₀ or neutralization titer NT₅₀) were calculated by fitting pooled data from two to three independent experiments to sigmoid dose–response curves (variable slope) using Prism software (GraphPad Software). If 50% inhibition was not achieved at the highest or lowest drug or plasma concentration, a greater than or less than value was recorded.

Neutralization assays were performed in the presence of linear peptides compromising the aa 301–324 (HxB2) of the corresponding V3 loop: JR-FL–V3, NNTRKSIHIGPGRAFYTTGEIIG; NL4-3–V3, NNTRKSIRIQRGPGRAFVTIGKIG; SF162-V3, NNTRKSITIGPGRAFYATGDIIG; and c10.14-V3, NNTRRSIHIGPGKAFYTGGIIG. We used a fixed concentration of 10 µg/ml of the V3 peptides in all experiments. This dose of V3 peptide was sufficient to completely compete off neutralization activity of the V3 loop–specific mAb 19b against JR-FL*ΔV1V2 at mAb concentrations of 50 µg/ml. None of the probed peptides directly influenced the infectivity of the probed R5 and X4 viruses, and murine leukemia virus Env–pseudotyped virus was used as control. Patient plasma samples were preincubated with the corresponding V3 loop peptides for 1 h. Then virus was added, and the mixture was preincubated for a further 1 h before infection of TZM-bl cells. The final, lowest dilution of the plasma sample probed was 1:200. 50% inhibitory concentration and neutralization activity in the presence and absence of competing V3 loop peptide were determined.

Sequences encoding gp120 WT and gp120ΔV1V2 proteins were derived from full-length Env genes and cloned into the expression vector CMV-R (provided by P. Kwong [Vaccine Research Center, National Institutes of Health]). Protein was generated upon transfection of FreeStyle 293 cells (Invitrogen). In brief, 1–10⁶/ml cells in a total volume of 100 ml of 293 FreeStyle medium (Invitrogen) were transfected with 160 µg DNA and 320 µg polyethyleneimine. gp120-containing supernatant was harvested on days 3 and 5 and used for binding experiments by ELISA. To this end, gp120 proteins were immobilized on ELISA plates coated with sheep antibody D7324 directed to the C5 domain of gp120 (Aalto). Plasma IgG titers and mAb to gp120 were determined by ELISA as described previously (Trkola et al., 2004). Midpoint titers were defined by fitting dose-dependent curves with a variable slope by Prism software. Each individual sample was controlled for unspecific binding activity (binding in absence of gp120), and data was corrected for unspecific binding before 50% binding titers or concentrations were derived. Curve fitting was constrained at 0% but not at 100% binding because plasma and mAbs show comparable but not equal saturation levels. In rare cases, anti-gp120 antibody titers in plasma were too low to reach binding saturation. To assess binding efficacy in these cases, maximal binding was set as the mean level of maximal binding observed in the 83 plasma samples, and curve fitting was performed using this mean level as 100% binding. To calculate the 50% binding titers, curves were fitted through the pooled data from two to three independent assays.

We developed a mathematical model with the aim to dissect experimentally whether neighboring or self protection is the mode by which the V1V2 loop protects against neutralization. Our theoretical and experimental approach is based on studying heterogeneous trimers composed of two different Env proteins. These Env proteins must differ in two regions. The first region concerns the V1V2 loop and the second region an antibody epitope (in our case the epitope of the V3 loop–specific antibodies 19b and 1-79), which is presumably masked by the V1V2 loop. We engineered Env proteins in which the V1V2 loop is either present or deleted, and 19b and 1-79 V3 loop mAb epitopes are present or defective (naturally or engineered). Two experimental setups with two different Env proteins out of the total eight possible combinations with these four Env variations can be used to distinguish between neighboring and self protection: (1) one Env protein contains the V1V2 loop and is sensitive to antibody binding, whereas the other Env protein is V1V2 deleted and antibody binding resistant; (2) one Env protein contains the V1V2 loop and is resistant to antibody binding, whereas the other Env protein is V1V2 deleted and antibody binding sensitive.

The schematic in Fig. 5 A shows how many antibodies could bind to the different trimers depending on the mechanism of protection in the experimental setup 1, and Fig. 6 A shows the number of trimers binding to the different trimers in setup 2. As a control experiment, we also studied an experimental setup (setup 3) in which two homologue antibody binding–sensitive Env proteins were mixed that only differed in the presence and absence of the V1V2 loop. This setup does not allow us to distinguish between neighboring and self protection but instead allows us to validate the integrity of heterotrimer formation. Fig. 7 A illustrates how many antibodies can bind to the different trimer types that can form in this setup.

In our experiments, we seek to determine the accessibility of the V3 loop on trimers by determining the binding efficacy of the given V3 loop mAbs by deriving the mean fluorescent intensity (MFI) of fluorophore-labeled mAb binding by FACS. The MFI is proportional to the number of antibodies binding to the Env-expressing cells. To predict the MFI, we need to average the number of bound antibodies over all trimer types in a cell population. This requires us to factor in the frequency of each trimer type, which depends on the fraction of different Env proteins. Let us define f_bs as the fraction of Env proteins with a binding-sensitive epitope (independent of whether it is protected by a V1V2 loop). Under the assumption that the different Env proteins trimerize perfectly randomly, the probability that a trimer consists of i Env proteins with an antibody binding–sensitive epitope and 3 − i antibody-resistant Env proteins is

For setup 3, all Env proteins are binding sensitive. For this setup, we introduce the parameter f_Δ that denotes the fraction of Env proteins without V1V2. The frequencies of each trimer type are given by the same formula as above with f_Δ instead of f_bs.

The number of antibodies binding to an average trimer can be calculated by summing over the products of the number of antibodies binding to a specific trimer (determined in Figs. 5 A, 6 A, and 7 A) and the probability that this trimer is formed. Table I lists the number of antibodies binding to an average trimer for each experimental setup and mode of protection.

To be able to compare our mathematical prediction with the observed MFI, we normalize the MFI and the expressions in Table I such that its maximum over the various levels of f_bs is 1. Thus, we obtained mathematical expressions that predict the behavior of the normalized MFI as a function of the fraction of antibody-sensitive Env proteins, f_bs. These predictions are shown in Figs. 5 B, 6 B, and 7 B.

Setup 3 does not allow us to distinguish between neighboring and self protection as identical reactivity patterns are predicted for both instances. Instead, setup 3 allows us to derive information on whether or not the formed heterotrimers adopt an appropriate conformation that enables V1V2 shielding to occur. Should the formed heterotrimers disable V1V2 shielding, the V3 loop would become accessible. In this case, MFI reactivity patterns would change substantially and follow the function

In the segregation model, we assume that the formation of homotrimers is more likely than that of heterotrimers. To assess the potential impact of Env protein segregation, we introduced a segregation parameter, 0 ≤ ξ ≤ 1, into our basic model. ξ = 0 corresponds to no segregation, and ξ = 1 corresponds to complete segregation. Table II lists the number of antibodies that bind, on average, to a trimer with segregation.

The probabilities to form a trimer with i binding–sensitive Env proteins, P_i, appearing in Table II are defined according to the segregation model in Magnus and Regoes (2010) as

For setup 3, f_bs must be replaced by f_Δ to obtain the probabilities P_i. As the V3 loop is only accessible on heterotrimers in the shielding scenario studied in Fig. 5 (setup 1; hump-shaped curve), homotrimer formation, regardless of its extent, is not predicted to influence the normalized MFI. In the alternate shielding scenario (setup 2; Fig. 6), the V3 loop can be accessed on heterotrimers, and their influence, although small, must be considered (Fig. S3 B).

Additionally to intact trimeric Env spikes, partially shed Env spikes displaying gp120 monomers and dimers will be expressed on the cell surface of Env-transfected cells. V1V2 shielding on these partially shed spikes will not be effective, and the V3 loop will thus be accessible. With what frequency monomers and dimers form is not known. Assay variations (e.g., transfection and expression efficiencies) can be envisaged to play a role. The fraction of Env monomers and dimers may also differ depending on the Env proteins involved. The stability of homo- and heterotrimers may differ, rendering one more prone to spontaneous gp120 shedding than the other. The latter would cause fluctuations in the expression levels of monomers and dimers but also the total Env content of the cell as a substantial proportion of the Env proteins will be released and thus lost for analysis. In the event that heterotrimers dissociate at a higher frequency than homotrimers, total Env expression levels should be lower at those Env ratios where the highest heterotrimer formation is to be expected (f_bs = 0.5). We have controlled for Env expression in each individual experiment by assessing the capacity of mAb 2G12 to bind to cell surface–expressed Env’s. This mAb recognizes monomeric, dimeric, and trimeric gp120 and is largely unaffected by the presence of V1V2 and the mutations in V3 studied here. Based on 2G12 reactivity, we were able to exclude the possibility that the studied heterotrimers are inherently instable and promote spontaneous shedding and thus monomeric and dimeric gp120 formation. Env concentration was comparable at all Env ratios studied. Should a heterotrimer be instable, total Env content as measured by 2G12 would be lower in cell populations that contain a higher heterotrimer content (e.g., f_bs = 0.5).

However, as some degree of spontaneous shedding will occur regardless of what specific Env is studied, we had to evaluate the effect of antibody binding to gp120 monomers and dimers in our assay system. To assess the potential impact of monomers and dimers on our prediction of the normalized MFI, we assumed that the total number of gp120 monomers and dimers expressed on a transfected cell is the same for all investigated cell populations. For setup 1 and 2, we assumed that one antibody binds to a binding-sensitive Env monomer, one antibody binds to a heterodimer (consisting of one V3 loop–sensitive and one resistant Env), and two antibodies bind to a homodimer consisting of two binding sensitive Env’s. In the case of setup 3, two antibodies are assumed to bind to heterotrimers. Monomers are expressed at a rate a, and dimers of either type are expressed at a rate b. Under these assumptions, we simply have to add the terms af_bs for antibodies binding to monomers and a term for the binding to dimers, which, for setup 1 and 2, is b(2f_bs(1 − f_bs) + 2f_bs²), or, for setup 3, b(4f_bs(1 − f_bs) + 2f_bs²), to the expressions in Table I. Upon simplification, we derived the following formulas for setup 1 and 2 under the consideration of monomer/dimer binding (Table III).

The parameter c = a + 2b denotes the fraction of antibodies bound to monomers and dimers. If the neighboring protection scenario holds true, c can be estimated from the normalized MFI in setup 1 at the point f_bs = 1.

293-T cells seeded in 12-well tissue culture plates were transfected with 1.6 µg of the respective HIV Env–encoding plasmids and 0.4 µg of the rev-encoding plasmid pCMV-rev using jetPEI (Polyplus Transfection; Chemie Brunschwig AG) according to the manufacturer’s instruction. To generate virions bearing heterogeneous Env trimers, two different Env-encoding plasmids were transfected. To vary the heterogeneous trimer composition, cells were transfected with different ratios of the two respective Env plasmids. In total, 11 ratios with the respective plasmids’ concentration varying in 10% increments were probed (one Env ranging from 100% to 0%, the second ranging from 0% to 100% in the same cotransfection). The following Env genes were studied: Env’s that carry the epitope of the V3 loop mAbs 19b and 1-79 (JR-FL*WT, JR-FL*ΔV1V2, and NL4-3WT_JR-FL^V3) and Env’s that do not carry the 19b epitope (NL4-3WT, NL4-3ΔV1V2, and JR-FL*ΔV1V2_V3^NL4-3). The following pairs were investigated: (a) JR-FL*WT with NL4-3ΔV1V2 (Fig. 5 C), (b) NL4-3WT_JR-FL^V3 with NL4-3ΔV1V2 (Fig. 5 D), (c) JR-FL*ΔV1V2_V3^NL4-3 with JR-FL*WT (Fig. 5 E), (d) NL4-3WT with JR-FL*ΔV1V2 (Fig. 6 C), and (e) JR-FL*WT with JR-FL*ΔV1V2 (Fig. 7 C).

Cells were detached 48 h after transfection using PBS, 1% FCS, and 10 mM EDTA, and V3 loop accessibility was studied by monitoring the binding efficiency of the V3 loop mAbs 19b or 1-79 to the heterogeneous Env proteins expressed on the cells’ surface. As control for equivalent Env expression of the heterotrimer, binding efficiency of the carbohydrate-specific mAb 2G12, which binds to all probed WT and Env mutants with comparable efficiency, was assessed in all experiments (Fig. S2). Stainings were performed with mAbs 19b and 1-79 and biotinylated mAb 2G12, and bound antibody was detected using Cy5-conjugated F(ab′)₂ goat anti–human IgG (Jackson ImmunoResearch Laboratories, Inc.) and streptavidin-APC (BioLegend), respectively. Viability stain propidium iodide (Invitrogen) was added to all samples 10 min before acquisition, and dead cells were excluded from analysis. Samples were acquired on a CyAn ADP Analyzer (Beckman Counter) and analyzed using FlowJo software (Tree Star). The highest MFI measured in each experiment was set as the maximum fluorescence level (100%), and measured MFI signals were normalized to the percentage of maximum fluorescence (% max).

All 50% neutralization, inhibition, or binding doses were determined by fitting a sigmoid dose–response curve with a variable slope through the pooled data of two to three independent assays. Error bars indicate the 95% confidence interval calculated by Prism software. Fold increases in plasma neutralization and binding activity refer to the difference between median activity of the 83 plasma samples against WT and V1V2-deleted Env’s (Fig. 1). One-way analysis of variance with post-tests Tukey’s multiple comparison and Newman Keuls was used to test whether the mean neutralization activity of the 83 patients’ plasma measured against NL4-3ΔV1V2 is significantly different from the other ΔV1V2 viruses. The same test was used to compare the datasets after V3 peptide competition (Fig. 3 B). To assess potential interdependencies between measured neutralization and binding activities, Pearson correlation parameters for 50% plasma neutralization or binding activities were assessed (Table S5).

Fig. S1 analyzes the shielding capacity of the V1V2 loop. Fig. S2 shows the analysis of mAb binding to viral Env on cell populations expressing varying composition of heterotrimers. Fig. S3 shows predictions for binding patterns in the shielding analysis considering model extensions for preferential homotrimer formation and the presence of monomeric and dimeric gp120 units. Table S1, included as an Excel file, shows an overview and characteristics of WT and chimeric HIV-1 Env’s. Table S2, included as an Excel file, shows the origin and specificity of HIV-1–directed mAbs and entry inhibitors. Table S3, included as an Excel file, provides an overview of plasma samples from individuals with chronic HIV-1 infection used in this study. Table S4, included as an Excel file, depicts columns statistics of 50% plasma neutralization and binding titers reported in Fig. 1. Table S5, included as an Excel file, lists Pearson correlation analysis of 50% plasma neutralization and binding titers.

We thank Herbert Kuster and Therese Uhr for technical assistance and John Robinson and Tina Riedel for providing V3 loop peptides. We thank the patients participating in the clinical studies from which samples for the current study was derived. We gratefully acknowledge the Swiss HIV Cohort Study (SHCS) for access to samples and Jürg Böni for the use of the SHCS drug resistance database.

Support was provided by the Swiss National Science Foundation (grants 310000-120739 and 310030-135527 to A. Trkola, 315200-114148 to R.R. Regoes, and 324730-130865 to H.F. Günthard). A. Trkola is an Elizabeth Glaser Scientist supported by the Elizabeth Glaser Pediatric AIDS Foundation.

The authors declare no competing financial interests.