Susceptibility to Neutralization by Broadly Neutralizing Antibodies Generally Correlates with Infected Cell Binding for a Panel of Clade B HIV Reactivated from Latent Reservoirs

Although antiretroviral therapies have improved the lives of people who are living with HIV, they do not cure infection. Efforts are being directed towards harnessing the immune system to eliminate the virus that persists, potentially resulting in virus-free remission without medication. HIV-specific antibodies hold promise for such therapies owing to their ability to both prevent the infection of new cells (neutralization) and direct the killing of infected cells. We isolated 36 HIV strains from individuals whose virus was suppressed by medication and tested 14 different antibodies for neutralization of these viruses and for binding to cells infected with the same viruses (critical for engaging natural killer cells). For both neutralization and infected cell binding, we observed variation both between individuals and amongst different viruses within an individual. For most antibodies, neutralization activity correlated with infected cell binding. These data provide guidance on the selection of antibodies for clinical trials. ABSTRACT Efforts to cure human immunodeficiency virus (HIV) infection are obstructed by reservoirs of latently infected CD4+ T cells that can reestablish viremia. HIV-specific broadly neutralizing antibodies (bNAbs), defined by unusually wide neutralization breadths against globally diverse viruses, may contribute to the elimination of these reservoirs by binding to reactivated cells, thus targeting them for immune clearance. However, the relationship between neutralization of reservoir isolates and binding to corresponding infected primary CD4+ T cells has not been determined. Thus, the extent to which neutralization breadths and potencies can be used to infer the corresponding parameters of infected cell binding is currently unknown. We assessed the breadths and potencies of bNAbs against 36 viruses reactivated from peripheral blood CD4+ T cells from antiretroviral (ARV)-treated HIV-infected individuals by using paired neutralization and infected cell binding assays. Single-antibody breadths ranged from 0 to 64% for neutralization (80% inhibitory concentration [IC80] of ≤10 μg/ml) and from 0 to 89% for binding, with two-antibody combinations (results for antibody combinations are theoretical/predicted) reaching levels of 0 to 83% and 50 to 100%, respectively. Infected cell binding correlated with virus neutralization for 10 of 14 antibodies (e.g., for 3BNC117, r = 0.82 and P < 0.0001). Heterogeneity was observed, however, with a lack of significant correlation for 2G12, CAP256.VRC26.25, 2F5, and 4E10. Our results provide guidance on the selection of bNAbs for interventional cure studies, both by providing a direct assessment of intra- and interindividual variabilities in neutralization and infected cell binding in a novel cohort and by defining the relationships between these parameters for a panel of bNAbs. IMPORTANCE Although antiretroviral therapies have improved the lives of people who are living with HIV, they do not cure infection. Efforts are being directed towards harnessing the immune system to eliminate the virus that persists, potentially resulting in virus-free remission without medication. HIV-specific antibodies hold promise for such therapies owing to their ability to both prevent the infection of new cells (neutralization) and direct the killing of infected cells. We isolated 36 HIV strains from individuals whose virus was suppressed by medication and tested 14 different antibodies for neutralization of these viruses and for binding to cells infected with the same viruses (critical for engaging natural killer cells). For both neutralization and infected cell binding, we observed variation both between individuals and amongst different viruses within an individual. For most antibodies, neutralization activity correlated with infected cell binding. These data provide guidance on the selection of antibodies for clinical trials.

and amongst different viruses within an individual. For most antibodies, neutralization activity correlated with infected cell binding. These data provide guidance on the selection of antibodies for clinical trials.
KEYWORDS broadly neutralizing antibody, bNAb, correlation, antibody-dependent cell cytotoxicity, ADCC, human immunodeficiency virus, infected cell binding, neutralization M odern antiretroviral (ARV) drug regimens effectively suppress human immunodeficiency virus (HIV) replication but are unable to cure infection. Interruption of ARV therapy thus results in rapid viral rebound and disease progression. A critical aspect of HIV persistence in the context of ARV therapy is the establishment of latent infection in long-lived resting memory CD4 ϩ T cells (1)(2)(3). Evidence from in vitro latency models supports the hypothesis that these reservoirs can be eliminated by combining latency reversal agents (LRAs), which induce the expression of viral antigens, with enhanced immune effectors, in a paradigm referred to as "kick and kill" or "shock and kill" (4-7). One strategy for harnessing immune effectors for this purpose is to target reactivated infected cells with HIV-specific antibodies, resulting in the engagement of natural killer (NK) cells, monocytes, and granulocytes, which eliminate infected cells through antibody-dependent cell-mediated cytotoxicity (ADCC) and/or antibodydependent cell-mediated phagocytosis (ADCP) (8)(9)(10). For this purpose, it will be crucial for the HIV-specific antibodies to bind to Env proteins expressed on the surfaces of the reactivated latently infected cells. The present study focuses on correlating the susceptibilities to neutralization of viral isolates reactivated from patient CD4 ϩ T cells by a panel of HIV-specific broadly neutralizing antibodies (bNAbs) with the capacity of these bNAbs to bind to Env expressed by the reactivated latently infected cells, thereby providing guidance on the selection of bNAbs to optimally support the clinical translation of kick-and-kill strategies.
The antigenic variability of the HIV envelope protein poses a substantial challenge to the development of both vaccines and immunotherapeutics (11)(12)(13). The past 10 years have seen the identification of a growing number of bNAbs, defined as such based on their activity against globally diverse HIV isolates (14-23; reviewed in references 24 to 27). Recent clinical trials established that passive infusion with bNAbs during chronic HIV infection can temporarily suppress virus replication in individuals whose virus does not escape (28)(29)(30) and can modestly delay viral rebound during antiretroviral treatment interruption (31,32). Additionally, passive immunization with bNAbs has attracted interest as a means of supplying the immune effector component of kick-and-kill HIV eradication strategies (given that virus has typically escaped from autologous antibody responses). This has led to the initiation of additional preclinical trials, as well as pilot clinical studies, aimed at testing the abilities of combinations of bNAbs and LRAs to reduce or eliminate latent HIV reservoirs (e.g., ClinicalTrials.gov trials NCT03041012 and NCT02850016).
Three primary factors argue for the prioritization of bNAbs over other types of HIV-specific antibodies for clinical trials aimed at reducing latent reservoirs through a kick-and-kill mechanism. First, there is extensive clinical experience with and safety data on several bNAbs from their use in passive infusion trials, facilitating their advancement into combination studies with LRAs. Second, the ability to exert the dual activities of neutralizing free virus and mediating ADCC would be favorable for an antibody therapeutic. Third, the antigenic diversity of HIV, both within a given individual's latent reservoir and at a population level, poses a challenge to the development of curative therapeutics, motivating the prioritization of Abs with broad reactivity. With respect to the latter point, while it stands to reason that an Ab with broad neutralizing activity is likely to exert a similar breadth of infected cell binding, this cannot be assumed to be the case. Infected cell binding is a prerequisite for and correlates closely with ADCC activity (9,(33)(34)(35). The conformations of Env on free virions that must be targeted to achieve neutralization may differ from those on infected cells that must be bound to trigger ADCC. For example, binding of Env on an infected cell to CD4 on the same cell (i.e., in cis) may both partially occlude the CD4 binding site (CD4bs) and induce gp120 shedding while exposing CD4-induced (CD4i) epitopes and gp41 stumps (36), thus antigenically changing the protein on a cell compared to that on the virion. Although CD4i antibodies commonly arise during infection (37) and have the potential to mediate ADCC against liganded versions of the Env protein, the addition of soluble CD4 (sCD4) mimetics has been necessary to increase the sensitivity of infected cells to ADCC by these antibodies (38,39). Furthermore, the possibility exists that viral diversity may differentially affect cell surface Env versus virion-associated functional Env trimers, potentially in unexpected ways. Thus, broadly neutralizing antibodies present the possibility of infusing multifunctional antibodies that target genetically diverse viruses on epitopes that do not require CD4 binding for epitope exposure; however, broad neutralizing activity may not equate to broad infected cell binding. Note that the bNAbs tested in this study all share the same IgG1 Fc domain, differing only in their Fab fragments. The present study thus focuses on providing guidance with respect to the selection of the antigen-binding Fab fragments of Abs for use in cure strategies. To maximize potency, these Fab fragments may ultimately need to be combined with Fc domains that are designed to maximally engage ADCC effectors (40).
A limited number of studies have thus far assessed the breadths of infected cell binding and/or ADCC activity by bNAbs in relation to neutralizing activity, and these have reported somewhat conflicting results. In testing 8 viral isolates reactivated from the latent reservoirs of ARV-treated individuals, Bruel et al. reported that a panel of bNAbs (including 3BNC117) could eliminate HIV-infected cells by mediating ADCC (9) and that their breadth of virus recognition was higher than that with nonneutralizing antibodies (33). In contrast, Mujib et al. reported a lack of infected cell binding and ADCC activity by 3BNC117 against a multiclade panel of HIV (41), suggesting a lack of correspondence with its breadth of neutralizing activity (16). Although this relationship has been explored indirectly, to our knowledge, only one study has directly compared infected cell binding or ADCC of bNAbs versus neutralizing activity across different viral isolates. This study showed a correlation between these functions but was limited to the use of two viral isolates of HIV (NL4-3 and JR-FL) and to simian-human immunodeficiency virus (SHIV) AD8-EO (35). We therefore perceived a need to define the relationship between neutralization and infected cell binding of clinically relevant bNAbs to HIV produced by reactivated latently infected CD4 ϩ T cells.
In the present study, we assessed virus neutralization and infected primary CD4 ϩ T cell binding of bNAbs, in parallel, against a panel of 36 viruses that were reactivated from the latent reservoirs of 8 ARV-treated individuals by quantitative viral outgrowth assays (QVOA) (42) (Fig. 1). We defined the intra-and interpatient breadths and potencies of both neutralization and infected cell binding activity of these bNAbs against reactivated reservoir viruses from a geographically localized population of clade B-infected individuals. For all bNAbs that demonstrated appreciable neutralizing activity, this correlated closely with infected cell binding. This represents the most comprehensive study, to date, using a large panel of bNAbs which target a range of different epitopes but share the same IgG1 Fc domain against a panel of ex vivo reactivated reservoir viruses to quantify both neutralization and binding to infected cells.
(This article was submitted to an online preprint archive [43].)

Virus neutralization profiles of bNAbs and bNAb combinations against reactivated reservoir viruses.
To test the ability of bNAbs to neutralize reservoir viruses, we obtained a panel of 14 bNAbs that are currently being developed for clinical use in humans and categorized these by their targeted epitopes (see Materials and Methods). We measured the neutralizing activities of these bNAbs against 36 viral isolates that had been reactivated from the latent reservoirs of 8 individuals by use of limiting dilution quantitative viral outgrowth assays (QVOA) ( Fig. 1 and 2A). The V3-glycanspecific bNAbs PGT121 and 10-1074, and the V1V2-specific bNAb PG9, exhibited potent but relatively narrow activity, exhibiting detectable neutralization (50% inhibitory concentration [IC 50 ] of Ͻ50 g/ml) of 53 to 69% of viruses, with geometric mean IC 50 s ranging from 0.3 to 0.6 g/ml (Fig. 2B). In contrast, the CD4 binding site (CD4bs)specific antibodies VRC01, VRC07-523, N6, and 3BNC117, as well as the MPER-targeting antibody 10E8, exhibited broad activity, with detectable neutralization of 77 to 100% of viruses (IC 50 Ͻ 50 g/ml), but with substantially higher IC 50 values (geometric mean IC 50 s of 2.1 to 8.9 g/ml) (Fig. 2B). These trends parallel the results in previous reports on pseudovirus assays, which also observed that CD4bs antibodies and 10E8 were generally much broader but less potent than V3-glycan and V1V2 apex antibodies (44,45). In the current experiment, CAP256.VRC26.25 neutralized only 9 of 36 reactivated reservoir viruses (26%) with a detectable IC 50 (IC 50 Ͻ 50 g/ml) (Fig. 2B). Because CAP256.VRC26.25 has been reported to preferentially neutralize subtype C viruses and the QVOA viral isolates tested here are all subtype B (Table 1), the low neutralization breadth we observed is compatible with published data (23). 4E10 and 2F5 are known to be less broad and potent than more recently published antibodies, so their lack of breadth against these viruses was expected. One exception to the general agreement between our data and those from published pseudovirus panels was for 2G12, which, although not broadly neutralizing against genetically diverse viruses, has been shown to potently neutralize subtype B viruses in published pseudovirus panels (20,46), but we observed only weak neutralization in our assays, with only two viruses reaching 80% neutralization (Fig. 2).
We frequently observed high degrees of similarity in neutralization sensitivities within an individual's viral quasispecies, consistent with genetic relatedness. For exam-

FIG 1
Schematic for paired assessment of virus neutralization and infected cell binding with reactivated reservoir viruses. Quantitative viral outgrowth assays (QVOA) were performed using CD4 ϩ T cells from ARV-treated study participants. Virus was isolated from HIV p24 ϩ wells at a dilution where Ͻ50% of wells were positive. A portion of the supernatant from each of these wells was used directly to assess virus neutralization using a TZM-bl cell assay. Another portion was used to infect activated primary CD4 ϩ T cells. Binding of bNAbs to these infected cells was assessed by flow cytometry, with costaining of CD3, CD4, and HIV Gag to identify infected cells.
ple, the five viral isolates from patient CIRC0196 were all sensitive to neutralization by CD4bs and MPER antibodies but resistant to V3-glycan and V1V2 antibodies ( Fig. 2B and C), but exceptions to this were common. For example, of the four QVOA viruses from patient OM5346, two (viruses 2 and 4) were highly sensitive to V1V2 antibodies (PG9, CAP256.VRC26.25, and PGDM1400) and resistant to V3-glycan antibodies (PGT121, 10-1074, and 2G12), whereas virus 3 exhibited the opposite sensitivity profile (   the viral isolates tested from a given participant (IC 80 Յ 10 g/ml) (Fig. 2C, bold cyan values). Given the limitations observed above in the breadths of coverage and potential escape of any single bNAb, it is likely that any clinical intervention would require combinations of multiple bNAbs to be effective. We therefore calculated the summed breadths of all combinations of two of the bNAbs tested in this study. To clarify, the results for antibody combinations are theoretical/predicted and not based on experiments performed with antibody combinations. We determined breadths of coverage by using an IC 80 of Յ10 g/ml as the cutoff for the geometric mean sensitivity of the quasispecies, based on our previous demonstration that this concentration correlated with a reduction in viremia in bNAb-treated clinical trial subjects (30). The combination of N6 with 10-1074 showed the greatest breadth of coverage, at 83% (IC 80 Յ 10 g/ml) (Fig. 2D), followed by the combination of VRC07-523 and 10-1074, which displayed an IC 80 of Յ10 g/ml for 81% of the reservoir virus isolates. The following antibody combinations displayed IC 80 s of Յ10 g/ml for 78% of the reservoir virus isolates: N6 and PGT121, VRC07-523 and PGT121, 3BNC117 and 10-1074, 3BNC117 and PG9, and 10E8v4-V5R-100cF and 10-1074. Thus, two antibody combinations were able to provide broad neutralization coverage of reactivated reservoir viruses (IC 80 Յ 10 g/ml) for this geographically discrete clade B-infected population. Infected cell binding profiles of bNAbs and bNAb combinations against reactivated reservoir viruses. We next measured the binding of bNAbs to the surfaces of primary CD4 ϩ T cells infected with the same reservoir virus isolates that had been assessed for neutralization. Activated CD4 ϩ T cells from HIV-uninfected donors were infected with reactivated reservoir viruses and stained with unconjugated bNAbs, followed by an Alexa Fluor 647-conjugated anti-human IgG secondary antibody. These samples were also stained with HIV Gag to identify infected cells. We used the median fluorescence intensity (MFI) ratio to quantify specific bNAb binding activity toward infected cells [MFI ratio ϭ (MFI of bNAb staining in HIV Gag ϩ cells)/(MFI of bNAb staining in HIV Gag Ϫ cells)] (Fig. 3A). Since we had already established the geometric mean IC 80 neutralization values for each virus, we opted to test infected cell binding at the following two concentrations for each antibody: (i) 5 g/ml, selected based on titration experiments (data not shown); and (ii) the geometric mean neutralization concentration (IC 80 ) (values are indicated below the table in Fig. 2C). For the latter, this meant that some antibodies were tested at Ͼ5 g/ml (e.g., 4E10 at 49.2 g/ml), while other antibodies were tested at substantially lower concentrations (e.g., PGT121 at 0.6 g/ml) (geometric mean IC 80 s are given below the heat map in Fig. 2C). This approach thus seeks to normalize intrinsic differences in avidity between different bNAbs.
In order to establish breadth, we defined binding as an MFI ratio of Ͼ2. In general, with the exception of VRC01, CD4bs Abs exhibited superior breadths of infected cell binding, covering 83 to 89% of reservoir isolates at the neutralization concentration (IC 80 ) ( Fig. 3C and D). The binding potencies of CD4bs Abs were relatively modest, however, with most exhibiting MFI ratios between 2 and 4 ( Fig. 3B and C). The V3-glycan antibodies PGT121, 2G12, and 10-1074 exhibited more limited breadths (42 to 75%) than those of CD4bs antibodies but showed substantially higher levels of specific binding to cells infected with susceptible viruses, with many MFI ratios exceeding 5. Sensitivity/resistance profiles were generally related for different viral isolates from the same individual, e.g., 10-1074 bound strongly to all isolates from 5/8 participants ( Fig. 3B) but exhibited a lack of binding to all viruses from CIRC0196 (at both concentrations [ Fig. 3B and C]). Intrapatient variability was observed, however; for example, 1 of 5 viruses from patient OM5162 exhibited high sensitivity to 10-1074, and the remaining 4 exhibited resistance. With the exception of CAP256.VRC26.25 (which is predominately clade C specific [23]), the V1/V2 bNAbs showed potent binding activity, particularly in the case of PG9, which at the IC 80 showed high levels of specific binding to 16 of 36 reservoir viruses, with MFI ratios of Ͼ4 (Fig. 3C). Infected cell binding of MPER-specific antibodies varied. 10E8v4-V5R-100cF (a version of 10E8 optimized for increased solubility and potency [47]) at 5 g/ml bound to 30 of 35 isolates, with high-level binding (MFI ratios of Ͼ4) observed for 13 of these. However, 10E8 and 10E8v4-V5R-100cF also showed substantial binding to uninfected bystanders (Gag Ϫ    population) (see Fig. 3A, right panel, for representative staining). In contrast, the MPER-specific bNAbs 2F5 and 4E10 exhibited generally narrow and weak binding of reservoir viral isolates ( Fig. 3B and C). Note that virus 1 from patient OM5162 showed a highly distinct bNAb binding profile compared to those of other isolates from the same individual: it was bound strongly by antibodies VRC07-523, 3BNC117, N6, PGT121, . Both correlations were tested with 2 reservoir viruses combined with 9 bNAbs and the A32 antibody. Each virus-bNAb combination is indicated by a symbol, and each color represents one effector cell type. Red symbols, haNK cells; green symbols, primary NK cells from the PBMCs of an HIV-negative donor (allogeneic). Correlation coefficients (r) and statistical significance (P) were calculated using Spearman's rank order correlation.  bNAb Binding and Neutralization of HIV from Reservoirs Journal of Virology 10-1074, and PGDM1400, whereas other autologous viral isolates were bound weakly, if at all, by these bNAbs. Similarly, viruses from OM5346 showed intraindividual diversity in binding to V1/V2-specific bNAbs, e.g., PGDM1400 and PG9 bound robustly to viruses 2 and 4 (with MFI ratios of Ͼ6), while no binding was observed for viruses 3 and 5 (Fig.  3B). Our data indicate both intra-and interindividual variabilities in binding to cells infected with reservoir viral isolates, highlighting the limitations of using any single antibody in a therapeutic. Achieving broad coverage of viral reservoir isolates in a population is likely to require combinations of at least two bNAbs. To assess this in the current population, we calculated the binding coverage of all possible two-antibody combinations by using the binding data obtained with the neutralization concentration (IC 80 ) (MFI ratio of Ͼ2) (Fig. 3D). Again, results for antibody combinations are theoretical/predicted and not based on experiments performed with antibody combinations. All CD4bs (excluding VRC01) antibodies combined with 2G12 or V1/V2 antibodies or MPER antibodies (except for 4E10) reached Ն92% coverage. Notably, the combination of 2G12 with VRC07-523, N6, 10E8, or 10E8v4-V5R-100cF reached 100% coverage; however, as previously mentioned, 10E8v4-V5R-100cF showed a high level of bystander binding in our in vitro assays. The 3BNC117-2G12 and VRC07-523-PG9 combinations reached 97% coverage, thus representing promising combinations for targeting reactivated clade B reservoir viruses (Fig. 3D).
With respect to the effects of the different concentrations of antibodies tested on binding, 10E8v4-V5R-100cF exhibited generally more favorable binding profiles (MFI ratios) at 5 g/ml, due to a reduction in the background binding compared to that observed at its IC 80 of 9.3 g/ml. In contrast, 10-1074 showed a lack of background binding even at 5 g/ml and thus displayed favorable binding profiles at this higher concentration compared to those at its IC 80 (0.7 g/ml) ( Fig. 3B and C).
Infected cell binding correlates with elimination by ADCC. Our primary interest in assessing infected cell binding is to predict the ability of a bNAb to direct ADCC against these cells. Infected cell binding is a prerequisite for ADCC, and multiple studies have indicated that in cases where antibody Fc domains are matched (all bNAbs tested here share the same IgG1), levels of binding correlate with ADCC activity (9,(33)(34)(35). To confirm this relationship under our experimental conditions, we performed paired infected cell binding and ADCC assays by using two reservoir virus isolates (OM5334 virus 7 and OM5162 virus 1) in combination with 9 bNAbs. The following two types of NK cells were tested in parallel as effectors: (i) haNK cells (NantKwest), a derivative of the NK-92 cell line (48) that has been enhanced for ADCC by expressing the highaffinity (ha) huCD16 V158 Fc␥RIIIa receptor, as well as engineered to express IL-2 (49); and (ii) freshly isolated NK cells from the peripheral blood of an HIV-uninfected donor. Note that NK cells were allogeneic to target cells, which may have resulted in some nonself recognition; however, it was previously demonstrated that NK cells can exert robust ADCC activity against HIV-infected cells in an allogeneic system (50). Binding assays were performed in parallel with ADCC assays under the same conditions, i.e., 10 g/ml over a total of 7 h at 37°C. For both haNK cells and primary NK cells, we observed moderate levels of NK cell-mediated elimination of HIV-infected cells in the absence of bNAbs, likely due in part to HIV-mediated downregulation of HLA molecules "missing self" (Fig. 3E and F) (51,52). To scrutinize the ability to detect specific ADCC in our assay, we took the approach of comparing wild-type (wt) 10-1074 with Fc mutants designed to either enhance or abrogate ADCC activity (GASDALIE and GRLR mutants, respectively). With haNK or primary NK cells as effectors, we observed potent elimination of infected cells with the GASDALIE mutant, moderate elimination with wt 10-1074, and no elimination with the GRLR mutant (see Fig. S1 in the supplemental material). As expected, we observed additional elimination of infected cells with the addition of bNAbs and significant direct correlations between total levels of elimination of HIV-infected cells (for haNK cells, r ϭ 0.69 and P Ͻ 0.001; for primary NK cells, r ϭ 0.65 and P Ͻ 0.001) as well as ADCC-specific elimination of infected cells, determined as follows: % ADCC ϭ [(% Gag ϩ cells under with-NK no-bNAb conditions) Ϫ (% Gag ϩ cells under with-NK with-bNAb conditions)]/(% Gag ϩ cells under with-NK no-bNAb conditions) (for haNK cells, r ϭ 0.73 and P Ͻ 0.0001; for primary NK cells, r ϭ 0.65 and P Ͻ 0.001) (Fig. 3E and F). Thus, our results are consistent with previous studies in indicating that infected cell binding is moderately predictive of ADCC activity for bNAbs with matched Fc domains.
Virus neutralization correlates with infected cell binding for most bNAbs. The breadths and potencies of neutralizing activity of bNAbs against diverse HIV isolates have been studied extensively (14)(15)(16)(17)(18)(19)(20)(21)(22)(23). In contrast, relatively few studies have assessed breadths and potencies of infected cell binding, which is an important prerequisite for ADCC (9,(33)(34)(35). Efforts to harness bNAbs to direct ADCC against infected cells would therefore benefit from an understanding of the degree to which infected cell binding can be inferred from neutralizing activity against a given virus. Our paired binding and neutralization data sets allowed us to assess this by using a number of analytic approaches in regard to both concentrations of bNAbs used for binding.
We first tested for correlations between the virus neutralization (IC 80 ) and the level of binding (MFI ratio) to all HIV Gag ϩ cells at 5 g/ml bNAb. When all antibodies were considered together, we observed a significant, direct correlation between virus neutralization and infected cell binding (P Ͻ 0.0001; Spearman's r ϭ 0.56) (Fig. 4A). For each of the bNAbs that showed appreciable neutralizing activity (VRC01, VRC07, 3BNC117, N6, PGT121, 10-1074, PGDM1400, PG9, 10E8, and 10E8v4-V5R-100cF), we observed significant direct correlations between neutralizing activity and infected cell binding (Fig. 4B). The antibodies 2F5 and CAP256.VRC26.25 showed little in the way of either neutralization or binding, precluding the possibility of detecting a relationship between these factors. 2G12 and, to a lesser extent, 4E10 were notable outliers, as they showed appreciable binding capacity for many of the viruses in this panel but very little corresponding neutralizing activity. This lack of potent neutralization activity is inconsistent with data from pseudovirus assays but in agreement with previous data obtained using virus produced from T cells, suggesting that 2G12 sensitivity is particularly tied to the source of virus (53)(54)(55). We also tested for correlations between the virus neutralization (IC 80 ) and binding of antibodies tested at the concentration of IC 80 and observed results similar to those observed with 5 g/ml bNAb, with a significant correlation between binding and neutralization for all antibodies considered together (P Ͻ 0.0001; Spearman's r ϭ 0.49) (Fig. 5A) and significant correlations for all individual bNAbs, except for (i) 2F5 and CAP256-VRC26. 25, which showed little activity by either measure, and (ii) 2G12 and 4E10, which again showed binding without substantial neutralization (Fig. 5B). The infection of a cell by HIV results in a progressive and almost complete loss of surface CD4 expression through the concerted actions of Nef, Vpu, and Env (56)(57)(58)(59)(60). Thus, in short-term in vitro infections of activated CD4 ϩ T cells, Gag ϩ CD4 Ϫ cells represent a later stage of infection than that represented by their Gag ϩ CD4 ϩ counterparts (which have not yet downregulated CD4). We further sought to control for any influence of the stage of cellular infection on the observed relationships (the approach is described and the results shown in Fig. S2 and S3). We observed similar patterns of significant correlations between virus neutralization and binding to cells in a late stage of infection (Gag ϩ CD4 Ϫ ) when bNAbs were tested at either 5 g/ml (Fig. S4) or the neutralization concentration (IC 80 ) (Fig. S5). Thus, we observed that, for most bNAbs, virus neutralization and infected cell binding were correlated and that these relationships were robust enough to be detected with or without controlling for the avidity of a given bNAb (by using various bNAb concentrations) or for infection dynamics (by considering total versus late infected cells). Interestingly, no such relationships were observed for 2G12 and 4E10, which were unique in displaying minimal neutralizing activity despite detectable infected cell binding.

DISCUSSION
The primary conclusion of the current study is that the ability of a given bNAb to neutralize clinical viral isolates is a strong correlate of its ability to bind to cell surface Env on primary CD4 ϩ T cells infected with the same virus. Furthermore, on comparison across a large panel of bNAbs, relative levels of infected cell binding and virus Correlations for bNAbs tested independently. Each virus-bNAb combination is indicated by a symbol, and each color represents one study participant. Correlations were analyzed by determining the Spearman correlation coefficient (r), with statistical significance (P) highlighted in red.
bNAb Binding and Neutralization of HIV from Reservoirs Journal of Virology neutralization continued to correlate; for example, 10-1074 showed both high-level infected cell binding and potent neutralization compared to those of VRC01. Thus, we conclude that-with respect to the Fab component of Abs sharing the same Fc-the selection of Abs based on broad and potent neutralizing activity is very likely to also select those that are suitable for infected cell clearance. Note that the reciprocal was not always true, with 2G12 exhibiting reasonably potent and broad infected cell binding contrasted by a general lack of neutralization of the reservoir-derived primary isolate viruses. Also, while PG9 and PGT121 showed overall significant correlations between neutralization and binding, there were several reservoir viruses for each which showed infected cell binding without virus neutralization. This is most readily apparent in the graphs shown in Fig. 4B. Though it was less striking, the MPER-specific bNAbs 2F5 and 4E10 also exhibited appreciable infected cell binding (similar in breadth and magnitude to that of VRC01), but with minimal neutralizing activity. We propose that the differences based on the directionality of this relationship may be related to the differential antigen conformational requirements of these two functions. For a bNAb to neutralize virus, it must bind functional Env trimers present on the surfaces of cells producing infectious virus. In contrast, an antibody that also binds to nonfunctional envelope proteins, such as gp41 stumps (36) or envelope monomers, may bind to infected cells to a greater degree than that to which they mediate neutralization (if they neutralize at all). From this perspective, the results with PG9 may be of particular interest for future study. PG9 is generally thought to be specific for a quaternary epitope found on functional trimers and thus may be expected to show a strict correlation between binding and neutralization. However, it has also been observed that PG9 can bind to some gp120/gp140 monomers from HIV clones such as BG505 (61,62). In the case of BG505, PG9 exhibits even higher-avidity binding to monomeric gp120 than to trimeric gp140, suggesting that binding may be critically different from that on the bona fide epitope on the native Env trimer (62,63). Thus, one speculative explanation for the outlier viral isolates which showed infected cell binding by PG9 without corresponding neutralization is that these clones may be "BG505-like," with preferential binding for monomeric gp120. We plan to follow-up with more in-depth studies of such outliers for both PG9 and PGT121 to determine if these may yield novel insights into the binding characteristics of these bNAbs for diverse Envs. Overall, the results of the current study demonstrate that virus neutralization is a strong predictor of infected cell binding but that the reciprocal relationship does not always hold. This is true both at the level of certain antibodies, such as 2G12, and at the level of specific virus isolates, such as outliers for PG9.
While it may be intuitive that virus neutralization correlates with infected cell binding, we do not feel that this could have been assumed to be the case without experimental evidence. The conformation of Envs may be affected by differences between the cell surface and virion environments, and this variability could differentially impact viral isolates. For example, interactions in cis between CD4 and Env on the surfaces of infected cells have been shown to induce gp120 shedding and to expose gp41 stumps. This has been reported to enhance infected cell binding by gp41-specific Abs while diminishing binding by gp120-specific Abs (36). Such an effect might differentially impact different viruses; for example, Horwitz et al. reported that the R456K mutation on YU2 gp120 decreased gp120 shedding, which led to less bystander (Gag Ϫ CD4 ϩ ) binding (64). However, despite any such differences between the virion and cell surface environments, the ability to neutralize virus was significantly correlated with infected cell binding, and these relationships held whether we considered all infected cells (Gag ϩ ) or only late infected cells (Gag ϩ CD4 Ϫ ).
To investigate factors that may predict the efficacy of bNAb treatment for contributing to HIV cure, we felt it important to study the properties of bNAbs against viruses derived from reactivated latent reservoirs. By combining a QVOA approach with isolation of virus from dilutions of CD4 ϩ T cells from different antiretroviral therapy (ART)-suppressed patients for whom Ͻ50% of wells were p24 ϩ , we were able to isolate viruses that were likely clonal to test bNAb binding and neutralization profiles (Fig. 1) and to assess both intra-and interpatient variability. We observed a considerable level of heterogeneity even within a given individual, such that in the majority of cases any single bNAb failed to provide universal coverage of an individual's reservoir isolates. However, combinations of two antibodies (results are theoretical/predicted) provided broad coverage both within and across individuals, reaching up to 100% coverage as assessed by binding. Note that as our study population was derived from a single site (Toronto, Canada), from a clinical perspective this assessment of breadth is representative of what might be expected in a single-site study in a North American clade B-infected cohort. We propose that the method presented here may be applied to different populations as a means of prioritizing antibody combinations for a given regional population of patients and personalizing individual HIV cure strategies as ART drug resistance is used to guide ART therapy. Clinical use of QVOA will likely be limited by its expense, cell number requirements, and protracted timeline (14 days) for results. However, a notable opportunity is present in the fact that infectious clonal autologous reservoir viruses are generated as a by-product of the primary measurement. The pairing of quantitative and qualitative assessments of the HIV reservoir in this way was previously termed Q 2 VOA (65). While we show here the feasibility of directly assessing both neutralization and infected cell binding from QVOA wells, our data also support the hypothesis that-for most antibodies-infected cell binding can reliably be inferred from potent neutralization. Also, substantial efforts are under way to develop sequence-based computational models to predict the susceptibility of HIV to neutralization by bNAbs (66). Our results support the hypothesis that, at least for the bNAbs tested in our study, it can reasonably be inferred that an antibody that is predicted to neutralize a given virus can also be predicted to bind to a corresponding infected cell. This provides a rationale for applying such computational models to the screening of potential clinical trial participants for studies aimed at targeting HIV reservoirs through the targeting of infected cells.
The potencies of neutralization observed in the current study are weaker overall than those that have previously been reported based on pseudovirus assays, most notably for 2G12, which failed to achieve 80% neutralization for all but two viruses. While this is likely due in part to our use of clinical viral isolates, which are generally less sensitive to bNAbs than laboratory-adapted viruses (67,68), we also note the role of virus-producing cells in modulating sensitivity to neutralization. Studies addressing this issue have reported that T cell-derived virus is more resistant to neutralization than pseudovirus generated by transfected 293T cells and, in particular, that replicationcompetent viruses produced by peripheral blood mononuclear cells (PBMCs) are more neutralization resistant than Env-matched pseudoviruses (53)(54)(55). However, there appear to be antibody-specific differences in the level of influence that a producer cell has on sensitivity to neutralization. For example, one study reported that PG9 is not very sensitive to differences in producer cells (69), while large differences in IC 50 have been reported between T cells and pseudoviruses for antibody 2G12 (53,54). These data suggest that producer cells differentially influence the conformations of Env on resulting virions, as well as their densities and glycosylation, or the number of gp120 molecules in the viral membrane. As PG9, in general, preferentially targets well-ordered, closed, trimeric viral spikes, this indicates that an equal number of well-folded spikes exists on virions produced by either cell type, whereas bNAbs such as 2G12 perhaps can bind equally well to misfolded trimers and are therefore more sensitive to increases in the latter. Furthermore, the epitopes of certain antibodies, such as 2G12, include glycans, and producer cells can affect glycosylation patterns of gp120 (69). Thus, in addition to the comparison between neutralization and infected cell binding, the current study contributes a reassessment of bNAb neutralization potency that may be more clinically applicable than data from pseudovirus assays.
In conclusion, our study provides novel insights into the relationship between infected cell binding and virus neutralization that may help to guide immunotherapeutic strategies aimed at either curing infection or enabling durable immune control of viral replication. The degree of intra-and interindividual variation in bNAb sensitivity within even this geographically discrete clade B population reinforces the importance of utilizing combinations of at least two bNAbs in such therapies. Screening reactivated reservoir viruses for sensitivity to bNAbs, at either an individual or population level, can help in the selection of antibody combinations for optimal coverage, e.g., combinations of PG9 and either 3BNC117 or N6 provided potent infected cell binding coverage of 94% and 72 to 78% coverage of neutralization (IC 80 Յ 10 g/ml) of viruses in the current study population. For the bNAbs that exhibited correlations between infected cell binding and neutralization, our study indicates that screening for either one of these factors is sufficient to infer that both functions will be present against reactivated reservoir viruses. Consistent with previous studies, we also confirmed that this infected cell binding-as measured by our assay-correlated with NK cell-mediated ADCC. The correlation was only moderate, however, indicating that binding should not be considered an absolute surrogate for a functional ADCC assay. It will be of interest for future studies to build upon these results with more extensive functional assays (potentially using various Fc domains and/or effector cells, including autologous NK cells). Such future directions may potentially uncover more subtle aspects of the relationship between virus neutralization and the targeting of cell-mediated, Fcdependent functional activities against infected cells, which may lead to the elimination of latent reservoirs.

MATERIALS AND METHODS
Ethics statement. All participants (HIV-infected individuals) were recruited from the Maple Leaf Medical Clinic in Toronto, Canada, through a protocol approved by the University of Toronto Institutional Review Board. Secondary use of the samples from Toronto was approved through the George Washington University Institutional Review Boards. All subjects were adults and gave written informed consent. Clinical data for these participants are given in Table 1.
QVOA. Human CD4 T cells were enriched from peripheral blood mononuclear cells (PBMCs) (Stemcell Technologies), processed via leukapheresis, which were drawn from long-term ARV-treated HIV-infected participants (Table 1). Cells were serially diluted (2 million, 1 million, 0.5 million, 0.2 million, and 0.1 million cells per well) and plated into 24-well plates, with 12 wells for each concentration. Phytohemagglutinin (PHA) and irradiated PBMCs were added to reactivate the infected cells, and MOLT-4 cells were added 24 h later to amplify the viruses. The medium was changed every 3 to 4 days, and a p24 enzyme-linked immunosorbent assay (ELISA) was run on day 14 to measure the amount of virus production.
p24 ELISA. The p24 enzyme-linked immunosorbent assay (ELISA) was performed with kit components obtained from the National Cancer Institute, NIH. In brief, 96-well high-binding microplates (Greiner Bio-One) were coated with capture antibody overnight, followed by blocking with a 1% bovine serum albumin (BSA) solution overnight. Supernatants from QVOA wells were collected and lysed with 1% Triton X buffer for 2 h, followed by transfer to ELISA plates and incubation for 1 h at 37°C. Plates were then washed with PBST buffer (phosphate-buffered saline ϩ 0.1% Tween 20) 6 times and incubated with primary antibody for 1 h at 37°C. After 6 additional washes, a peroxidase-labeled goat anti-rabbit IgG secondary antibody (KPL) was added and incubated for another 1 h at 37°C. After 6 additional washes, TMB substrate (Thermo Fisher) was added and developed for 15 min, and then the reaction was stopped with stop solution (BioLegend). Absorbance was measured with a SpectraMax i3x Multi-Mode microplate reader (Molecular Device) at 450 and 570 nm. The cutoff for positive wells was set to Ͼ2ϫ the average of negative-control values.
Neutralization assay. Neutralization of QVOA virus samples by bNAbs was measured by infection of TZM-bl cells as described previously (31,70). The p24 protein in each virus sample was quantified by use of an AlphaLISA HIV p24 biotin-free detection kit (Perkin Elmer, Waltham, MA), and input virus was normalized to 5 to 10 ng/ml for the assay. Ten-microliter aliquots of 5-fold serially diluted monoclonal antibodies (MAbs) (from a starting concentration of 50 g/ml) were incubated with 40-l aliquots of replication-competent virus samples in duplicate for 30 min at 37°C in 96-well clear-and flat-bottomed black culture plates (Greiner Bio-One). TZM-bl cells were added to each well at a concentration of 10,000 cells per 20 l in Dulbecco's modified Eagle's medium (DMEM) containing 75 g/ml DEAE-dextran and 1 M indinavir. Cell-only and virus-only controls were included in each plate. Plates were incubated for 24 h at 37°C in a 5% CO 2 incubator, after which the volume of culture medium was adjusted to 200 l by adding complete DMEM containing indinavir. At 48 h postinfection, 100 l was removed from each well, and 100 l of SpectraMax Glo Steady-Luc reporter assay reagent (Molecular Devices, LLC, CA) was added to the cells. After a 10-min incubation at room temperature to allow cell lysis, the luminescence intensity was measured using a SpectraMax i3x multimode detection platform per the manufacturer's instructions. Neutralization curves were calculated by comparing luciferase units to those of the virus-only control after background subtraction and were fit by nonlinear regression using the asymmetric five-parameter logistic equation in GraphPad Prism (Fig. 2A). The 50% and 80% inhibitory concentrations (IC 50 and IC 80 , respectively) were defined as the antibody dilutions that neutralize 50% and 80% of the virus, respectively. bNAb binding assay. All binding assays were tested with the unconjugated bNAbs. CD4 ϩ T cells (which were all CD3 ϩ ) were isolated by use of a human CD4 T cell enrichment kit (Stemcell Technologies) and activated with CD3/28 antibodies (BioLegend) for 48 h. Supernatants collected from QVOA wells (p24 ϩ ; the same viruses as those for the neutralization assay were used) were used for infection by being added to the activated CD4 ϩ T cells, followed by spinoculation for 1 h and 6 days in culture, with a medium change every 3 days. The infection rate was checked on days 3 and 5 postinfection. When most of the infection reached Ͼ5%, bNAb staining was performed. Cells were collected, washed twice with 2% fetal bovine serum (FBS)-PBS, and then aliquoted into 96-well plates (1 million cells per well). Unconjugated bNAbs were added according to the outlined wells by dilution to a final concentration of 5 g/ml or the neutralization concentration (IC 80 ), which was the geometric mean for neutralized virus generated from the neutralization assay, and then incubated at 37°C for 1 h. Without washing, the Alexa Fluor 647-labeled secondary antibody (Southern Biotech) was added and incubated at 4°C for 30 min. After washing once with 2% FBS-PBS, a surface antibody mixture of BV786 -anti-human CD3 (SK7; BD Biosciences), Pacific Blue-anti-human CD4 (RPA-T4; BD Pharmingen), and LIVE/DEAD Aqua (Life Technologies) was added. Thirty minutes later, cells were washed twice and fixed/permeabilized with fixation/permeabilization solution (BD Bioscience). Anti-HIV-1 core antigen antibody (KC57-RD1; Beckman Coulter) was used to stain intracellular HIV-1 Gag protein. After two washes with 1ϫ Perm/Wash buffer, cells were detected by flow cytometry (BD Fortessa X-20), and data analysis was performed with FlowJo v10 (Treestar).
Antibody-mediated NK cell killing assay. ADCC assays were performed with unconjugated bNAbs and one of two types of NK cells: (i) haNK cells (NantKwest), an NK-92 cell line engineered to express the high-affinity (ha) CD16 V158 Fc␥RIIIa receptor, as well as engineered to express IL-2 (49); and (ii) primary NK cells enriched from the PBMCs of an HIV-negative donor (using buffy coat from the Gulf Coast Regional Blood Center) by use of a human NK cell enrichment kit (Stemcell Technologies). To generate target cells, primary CD4 ϩ T cells were enriched from the PBMCs of allogeneic healthy donors and infected with reservoir viruses as described for binding assays (see above). Infections were monitored by flow cytometry, and ADCC assays were performed when target cells were Ͼ5% infected. Both types of NK cells were treated with a 10 nM concentration of an interleukin-15 (IL-15) superagonist complex, ALT-803 (71,72), for 1 h to prime and activate them. Infected cells were collected and washed twice with 2% FBS-PBS. Cells (2 ϫ 10 5 /well) were added to U-bottomed 96-well plates. Unconjugated bNAbs (VRC01, VRC07-523, 3BNC117, N6, PGT121, 2G12, 10-1074, PGDM1400, PG9, A32, or no Ab) were added to a final concentration of 10 g/ml and then incubated at 37°C for 2 h. After this incubation, 4 ϫ 10 5 ALT-803treated NK cells were added to each well to give an effector/target (E:T) ratio of 2:1. bNAb binding assays were performed in parallel with the ADCC assay, under the same conditions, but no NK cells were added. Plates were centrifuged at 100 ϫ g for 30 s to bring target and effector cells into contact with each other and then incubated at 37°C and 5% CO 2 . Cells were mixed by pipetting after 2 h of incubation and then cocultured for an additional 5 h. After a total of 7 h of coculture, cells were washed twice with 2% FBS-PBS and stained with fluorochrome-conjugated antibodies against human IgG, CD3, CD56, and CD4 (all from BioLegend), as well as with LIVE/DEAD Aqua amine-reactive dye (Molecular Probes). Cells were then fixed and permeabilized using a BD Cytofix/Cytoperm kit following the manufacturer's instructions. Statistical analysis. Statistical analyses were performed using Prism 7 (GraphPad). Flow data were analyzed with FlowJo v10. Heat maps were generated with Excel. Comparisons between MFI ratios of Gag ϩ CD4 ϩ and Gag ϩ CD4 Ϫ cells were performed by using the Wilcoxon matched-pair signed-rank test. All correlations were calculated using Spearman's rank order test.