Structural Insights on the Role of Antibodies in HIV-1 Vaccine and Therapy

Abstract

Despite 30 years of effort, there is no effective vaccine for HIV-1. However, antibodies can prevent HIV-1 infection in humanized mice and macaques when passively transferred. New single-cell-based methods have uncovered many broad and potent donor-derived antibodies, and structural studies have revealed the molecular bases for their activities. The new data suggest why such antibodies are difficult to elicit and inform HIV-1 vaccine development efforts. In addition to protecting against infection, the newly identified antibodies can suppress active infections in mice and macaques, suggesting they could be valuable additions to anti-HIV-1 therapies and to strategies to eradicate HIV-1 infection.

HIV-1 Env as an Escape Artist

The HIV-1 infection cycle begins with host cell recognition and entry mediated by the HIV-1 envelope spike (Env) on the surface of the virion (Figure 1). Each Env is composed of a trimer of identical gp160 protein subunits that are cleaved posttranslationally to yield two associated glycoproteins, gp120 and gp41: gp120 carries the recognition sites for the host receptor (CD4) and coreceptor (primarily CCR5 and CXCR4), and gp41 mediates fusion between the virus and host cell membranes.

Figure 1. Epitopes of Broadly Neutralizing Antibodies on HIV-1 Env.

Electron microscopy structure of Env trimer (Liu et al., 2008) with approximate locations of epitopes highlighted on surface representation (each epitope is shown once per trimer): CD4-binding site Ab epitope (red), V3 loop/Asn332 Ab epitope (glycan attached to Asn332 as space filling model) (blue), V1/V2 loop/Asn160 Ab epitope (glycan attached to Asn160 as space filling model) (green), MPER epitope (yellow). N-linked glycans are shown as gray sticks and were added to all potential N-linked glycosylation sites present in the coordinates for BG505 SOSIP Env (PDB 4NCO) using the GlyProt server.

During acute infection with HIV-1, serum viral loads peak approximately 3 weeks after transmission and then fall 1–2 logs to a set point determined by the host immune system (Cohen et al., 2011; Daar et al., 1991). T cells are primarily responsible for this partial control of viral replication in the early phases of HIV-1 infection (Goonetilleke et al., 2009; Koup et al., 1994). In rare individuals who carry specific HLA alleles such as HLA-B*5701, T cell responses can even reduce viremia to undetectable levels for prolonged periods of time (Migueles et al., 2000).

In contrast, antibodies appear not to contribute significantly to the control of HIV-1 during natural infection. Anti-Env antibodies can be detected several weeks after infection. The initial antibody response is directed against gp41 and has little effect on viral dynamics (Cooper et al., 1987; McMichael et al., 2010; Tomaras et al., 2008). Gp120-directed antibodies with auto-logous neutralizing activity develop 4–14 weeks after infection and exert considerable selective pressure that shapes the emergence of Env variants (Bar et al., 2012; Gray et al., 2007; Mikell et al., 2011; Wei et al., 2003). However, HIV-1 mutates at a sufficiently high rate and produces enough diversity in the viral population that the viral swarm in any infected person appears to contain resistant variants to any developing antibody (Wei et al., 2003). This extraordinary potential for HIV-1 escape from antibodies parallels its ability to escape from antiretroviral drugs (Davey et al., 1993; Goldberg et al., 2012; Richman et al., 1994). The result is a continuing race between newly developing antibodies and the rapidly mutating virus, which ultimately leads a small percentage of HIV-1-infected individuals to produce antibodies that can neutralize a broad range of different viral strains (Doria-Rose et al., 2010; Mikell et al., 2011; Moore et al., 2012; Simek et al., 2009).

Structural and biophysical studies have revealed a number of features of Env that enable HIV-1 to evade the human antibody response. Remarkably common in the set of resistant variants are those that add or remove potential N-linked glycosylation sites (PNGS) (Sagar et al., 2006; van Gils et al., 2011; Wei et al., 2003). The many glycans decorating the surface of Env form a “glycan shield” that reduces access to protein epitopes. These glycans have the same chemical structures found on host glycoproteins, and individually they are therefore indistinguishable from the host, impeding development of uniquely glycanspecific anti-HIV antibodies.

Immune evasion also results from conformational masking of key conserved functional sites on HIV-1 Env (Kwong et al., 2002). For example, the coreceptor binding site is not fully exposed until after the CD4-binding site on gp120 is engaged by CD4. Similarly the fusion machinery in the membrane-proximal external region (MPER) of gp41 is not exposed until after the coreceptor binding site is occupied and the virus begins the fusion process (Frey et al., 2008). Thus, conformational masking diminishes the effects of antibodies that target these two relatively conserved regions by restricting access.

Consistent with an important role for glycosylation and conformational masking in HIV-1 escape from neutralizing antibodies, HIV-2, which has less glycosylation in the gp120 V4 loop and less conformational masking than HIV-1, elicits broadly neutralizing antibodies (bNAbs) far more frequently than HIV-1 (de Silva et al., 2012; Kong et al., 2012; Ozkaya Sahin et al., 2012).

In addition to the challenges posed by the glycan shield and conformational masking, the potency of anti-Env antibodies is reduced by their limited ability to bind with avidity to the relatively small number of spikes on the viral surface (Klein and Bjorkman, 2010; Mouquet et al., 2010). While most bacterial and viral pathogens present dense arrays of antigens, HIV-1 displays only ~14 Env trimers per virion (Chertova et al., 2002; Liu et al., 2008; Zhu et al., 2006). The paucity of surface spikes makes it difficult to achieve bivalent antibody binding to two adjacent spikes because the average distance between them usually exceeds the distance that can be spanned by the two Fab arms of an antibody. Furthermore, the Env trimer geometry generally prevents bivalent antibody binding to epitopes on adjacent monomers in the same trimer (Figure 1). These factors suggest that anti-HIV-1 antibodies only rarely bind bivalently, an idea supported by the relatively small differences in potency between the IgG and Fab forms of the antibodies that have been tested (Klein and Bjorkman, 2010).

A major advance in understanding the quaternary structure of trimeric Env was achieved by crystallographic and cryo-electron microscopy (cryo-EM) studies of soluble, cleaved forms of the Env trimer (Bartesaghi et al., 2013; Julien et al., 2013a; Lyumkis et al., 2013). The gp120-gp41 protein for two of these studies was produced using a stabilized form of the HIV-1 strain BG505 trimer that had antigenic properties corresponding to native Env (Sanders et al., 2013). The 4.7Å crystal structure (in complex with a V3 stem-specific bNAb PGT122) and the 5.8Å cryo-EM reconstruction (fully glycosylated trimer in complex with the CD4-binding site antibody VRC-PG04) revealed the intricate arrangement of the V1/V2 and V3 loops at the apex of the trimer as well as the locations of gp41 helices HR1 (forming a three-helix bundle along the trimer axis) and HR2 (at the outer edge of the trimer base). These structures will be useful for immunogen design by revealing the full extent and context of the antibody targets.

bNAbs against HIV-1

Despite the ability of HIV-1 to evade the antibody response, a small fraction of infected individuals eventually develop high titers of broadly neutralizing serologic activity (Doria-Rose et al., 2009; Gray et al., 2011; Mikell et al., 2011; Sather et al., 2009). Initial attempts to characterize and map these responses were mostly limited to serum adsorption studies on a polyclonal level (Binley et al., 2008; Li et al., 2007), and only a few potent and broadly neutralizing monoclonal antibodies were described until 2009. These antibodies targeted the CD4-binding site, the MPER, a cluster of high mannose glycans, and the variable loops, and, although not very potent, the antibodies neutralized multiple lab-adapted and/or primary isolates from several HIV-1 sub-types establishing the principle that antibodies to HIV-1 can have neutralizing breadth (Burton et al., 1994; Gorny et al., 1992; Muster et al., 1993; Scott et al., 1990; Tilley et al., 1991; Trkola et al., 1996; Zwick et al., 2001).

Beginning in 2009, with the introduction of single-cell antibody cloning techniques to characterize the anti-HIV antibody response (Scheid et al., 2009), dozens of broad and potent monoclonal antibodies have been cloned from infected individuals. The best of these are two to three orders of magnitude more potent than the previously characterized antibodies and can neutralize up to 90% of the HIV-1 isolates tested. Occasionally, individual monoclonal antibodies can account for the activity in serum (Walker et al., 2011a; Wu et al., 2010), but breadth and potency are most frequently attributed to a combination of unrelated antibodies that target different sites of vulnerability on the Env spike (Georgiev et al., 2013; Klein et al., 2012a; Moore et al., 2012; Sather et al., 2012; Scheid et al., 2009; Scheid et al., 2011; Wibmer et al., 2013). The targets of the new antibodies can be divided into the following major groups: (1) CD4-binding site, (2) N-linked glycan-containing epitopes, and (3) the membrane proximal external region (MPER). In addition, there are yet to be defined conformational epitopes, e.g., antibody 3BC176 (Klein et al., 2012a) (Figure 1).

CD4-Binding Site Antibodies

Since the CD4-binding site on gp120 is the site of initial host receptor interaction it cannot be completely masked by conformational changes or the glycan shield, and therefore it is potentially vulnerable. Moreover, it must be conserved between strains to maintain its function. It is not an easy target, however, because the site is a recessed pocket that accepts the terminal immunoglobulin-like domain of CD4 (Kwong et al., 1998). Fitting into the CD4-binding pocket is a challenge for the larger antigen-binding unit of an antibody, which is composed of paired immunoglobulin domains (variable heavy and variable light, V_H and V_L).Nevertheless, the human immune system has found at least two distinct ways to produce antibodies that can access this site. One way mimics CD4 itself, by positioning the antibody V_H domain to bind gp120 using primarily the complementarity determining region (CDR) H2 loop and framework residues (Zhou et al., 2010) and engaging the CD4-binding loop on gp120 with backbone atoms in the C” strand of V_H (Diskin et al., 2011; Wu et al., 2011; Zhou et al., 2010) (Figure 2A). The antibodies in this class are CD4 mimetics because the V_H domain occupies the same position as the terminal immunoglobulin-like domain of CD4 and both use the C” strand to interact with the CD4-binding loop of gp120 (Zhou et al., 2010). The other mode of CD4-binding site recognition is exemplified by antibody CH103 (Liao et al., 2013). This mode of recognition does not mimic CD4 but uses the antibody's CDR H3 loop to reach the CD4-binding loop (Figure 2B).

Figure 2. bNAb Recognition Modes for the CD4-Binding Site.

(A) CD4-mimetic: NIH45-46 (magenta) (Diskin et al., 2011) versus CD4 (yellow). (B) CDR H3 loop-based: CH103 (purple) (Liao et al., 2013) versus CD4 (yellow). Structures of the complexes are shown as Cα traces aligned on gp120. The top views display gp120 in a surface representation with the CD4 binding-loop highlighted (salmon). The bottom views display closeup views showing the Cα trace of gp120 (green).

There are two groups of CD4-mimetic bNAbs; those that derive from the IgH VH1-2 germline gene segment (typified by VRC01 (Wu et al., 2010)), and those that derive from the VH1-46 gene segment (Scheid et al., 2011). The number of independently developed antibodies in the VH1-2 group is remarkable—antibodies binding gp120 in essentially the same manner have been isolated from seven different infected individuals (Georgiev et al., 2013; Scheid et al., 2011; Wu et al., 2010; Wu et al., 2011; Zhou et al., 2013). Antibodies belonging to this group share heavy chain signature residues including Trp50, Asn58, Arg71, and Trp100B (West et al., 2012). Inspection of human V_H gene sequences and usage frequencies reveals that VH1-2 is by far the most common germline gene segment that encodes Trp50, Asn58, and Arg71 (Trp 100B is in CDR H3 and not derived from the V_H gene segment). The dependence on antibody structural regions that are less likely to be shaped by recombination and somatic hypermutation may help to explain the restricted usage of V_H genes observed for this class (Diskin et al., 2011; Scheid et al., 2011; Zhou et al., 2013). A number of different kappa and lambda light chain gene segments can be paired with VH1-2 to produce CD4-mimetic antibodies, but some commonalities of the light chains are notable. These are due in part to the requirement for the light chain to avoid steric clashes with gp120. First, the length of CDR L3 is restricted to five amino acids (which occurs in only about 1% of human light chains) (West et al., 2012). Second, the CDR L1 loop appears to require either a relatively short loop or the presence of glycine residues to confer sufficient flexibility to avoid unfavorable interaction with gp120 (Diskin et al., 2011; Zhou et al., 2013) (Figure 2A).

Unlike the VH1-2 group of CD4-mimetic antibodies, the VH1-46 antibodies do not have short CDR L3 loops and lack Trp50, Asn58, and Trp100B (Scheid et al., 2011). The known representatives of the VH1-46 class (isolated from two individuals) are not as potent as the VH1-2 class (Chuang et al., 2013; Scheid et al., 2011). The VH1-2 and VH1-46 gene segments are closely related, suggesting that common features of these germline genes are particularly compatible with the CD4-mimetic binding mode (Kwong et al., 2013).

While the VH1-2 and VH1-46 classes of CD4-binding site antibodies mimic CD4 by using framework residues and CDR H2, another group of CD4-binding site antibodies uses long CDRs to reach into the recessed CD4-binding site. For this group, the main interface with gp120 is formed by the CDR H3 loop (Figure 2B). The first of these antibodies to be characterized was b12, which is a phage display-derived antibody in which the heavy chain was paired with a noncognate light chain (Burton et al., 1994). Antibody CH103, which was cloned by single-cell methods from an infected individual (Liao et al., 2013), also recognizes the CD4-binding site loop using CDR H3. This antibody is considerably broader and more potent than b12; it recognizes a surface that overlaps well with the site of CD4 contact. Less is known about other naturally arising non-CD4-mimetic bNAbs including HJ16 (Corti et al., 2010; Pietzsch et al., 2010). To date, the antibodies targeting the CD4-binding site by non-CD4-mimetic mechanisms are less broad and potent than the CD4 mimetic antibodies, but fewer of this class of antibodies have been cloned. The non-CD4-mimetic Abs do not seem to have a strong preferential germline V gene usage, consistent with binding primarily dependent on their CDR H3 sequence.

N-Linked Glycan-Containing Epitopes

Although the glycan shield of the HIV-1 Env is generally effective in blocking antibody access to potential protein epitopes, bNAbs can avert this mode of escape by including the glycans themselves as part of their target (McLellan et al., 2011; Mouquet et al., 2012a; Pancera et al., 2013a; Walker et al., 2011a; Walker et al., 2009). Bacterial pathogen-specific anti-carbohydrate antibodies commonly recognize pathogen-derived glycans that are distinct from host carbohydrates. In contrast, the specific recognition of host-derived glycans on the HIV-1 spike is a more challenging task because potentially harmful anti-self antibodies are generally removed from the repertoire during B cell development (Wardemann et al., 2003). HIV-1 antibodies whose epitopes include N-glycans solve this problem because they do not recognize isolated self-glycans. Instead they recognize combinations of glycans and nearby protein features (V1/V2- and V3-glycan-dependent antibodies), and/or they bind to two or more glycans with a precise geometrical relationship that does not normally exist on host glycoproteins (2G12). In contrast to the common recognition mode of the VH1-2 class of CD4-binding site antibodies, each of the recently isolated bNAbs recognizing N-linked glycan-containing epitopes have solved this recognition puzzle in a distinct way.

Asn332 Glycan-V3

The largest and most diverse group of bNAbs recognizing N-linked glycan-containing epitopes targets a region adjacent to the highly conserved PNGS at Asn332, which includes the base of the gp120 V3 loop and several nearby glycans. Whereas antibodies to the CD4-binding site must approach by a restricted angle, bNAbs targeting the region centered on Asn332 approach the site from several different angles (Kong et al., 2013; Pancera et al., 2013b) (Julien et al., 2013a; Lyumkis et al., 2013) (Figure 3). The bNAbs that recognize this region include 2G12 (Scanlan et al., 2002; Trkola et al., 1996), PGT128 (Pejchal et al., 2011; Walker et al., 2011a), PGT121/10-1074 (Mouquet et al., 2012a; Walker et al., 2011a), and PGT135 (Kong et al., 2013; Walker et al., 2011a).

Figure 3. Multiple Approach Angles for Antibodies Targeting the N-Linked Glycan Attached to Asn332.

Surface overlay of aligned complexes. Structures shown: gp140, gray (Liu et al., 2008); Asn332 glycan, yellow; PGT128, green (Pejchal et al., 2011); 2G12, magenta (Calarese et al., 2003); and PGT135, blue (Kong et al., 2013).

2G12 recognizes a cluster of high-mannose glycans attached to Asn332, Asn295, and Asn339 using a unique domain-swapped architecture that combines both Fab arms into a single, large antigen-binding (Fab)₂ unit (Calarese et al., 2003). Domain-swapped dimers of 2G12 show higher neutralization potency and breadth (West et al., 2009) through binding of two (Fab)₂ units that are flexibly linked to two Fc regions in the 2G12 dimer (Wu et al., 2013). PGT128 recognizes an epitope that includes N-glycans attached to Asn332 and Asn301 in combination with amino acid residues near the C-terminal end of the V3 loop (Pejchal et al., 2011). PGT128 recognizes and penetrates the glycan shield using a long CDR H3 loop and a CDR H2 loop with a six amino acid insertion. PGT121, PGT122, and 10-1074 are clonal relatives that represent a separate group of bNAbs recognizing the glycan attached to Asn332 in combination with amino acid residues at the base of the V3 loop and a glycan attached to V1/V2 (Mouquet et al., 2012a) (Julien et al., 2013a; Julien et al., 2013c; Lyumkis et al., 2013). These antibodies are the most potent among the Asn332-dependent antibodies, but they are not as broad as the anti-CD4-binding site antibodies because they lack activity against many clade A and C strains (Mouquet et al., 2012a; Walker et al., 2011a).

Structural and biochemical studies of 10-1074 and PGT121 provided the first characterization of HIV-1 epitopes involving complex-type N-glycans and suggested that some glycanspecific HIV-1 antibodies can exhibit promiscuous recognition of N-glycans through binding to the pentasaccharide core common to both high-mannose and complex-type N-glycans (Mouquet et al., 2012a). This type of glycan recognition may be an adaptation to the varying proportions of high-mannose- and complex-type N-glycans found on HIV-1 Env within a single strain of virus at any particular PNGS. Further definition of the epitope of the PGT121/10-1074 clonal family has been provided by the crystal structure of the Env trimer bound to the PGT121/ 10-1074 clonal relative PGT122 (Julien et al., 2013a). As expected, the glycan attached to Asn332 is the central feature of the epitope, contacted by the long CDR H3 loop and by CDR L2. The glycan attached to Asn137 in the V1 loop of BG505 gp120 occupies the groove previously observed to bind complex glycans (Julien et al., 2013c; Mouquet et al., 2012a). The glycan attached to Asn301, which has a significant effect on PGT122 neutralization, is contacted by the light chain framework region 3 (Julien et al., 2013a). Precisely how this group of antibodies neutralizes HIV-1 is not known, but it has been proposed that they allosterically prevent CD4 binding (Julien et al., 2013c). Another antibody that depends on Asn332-linked carbohydrate, PGT135, recognizes glycans attached to Asn332, Asn392, and Asn386 using long CDR H1 and H3 loops to reach through the glycan shield, contacting glycans and peptide residues belonging to two adjacent b strands connected to the V3 and V4 loops (Kong et al., 2013).

Asn160 Glycan-V1/V2

A second group of glycan-binding bNAbs recognizes the glycan attached to Asn160 in conjunction with nearby peptide residues. This epitope is conformation-dependent, and it is recognized by antibodies PG9 and PG16 (Walker et al., 2009), CH01-CH04 (Bonsignori et al., 2011), and PGT141-PGT145 (Walker et al., 2011a). PG9 has a hammerhead-like 28-residue CDR H3 loop that binds to glycans attached to Asn160 and Asn156 and one b strand of the V2 domain (McLellan et al., 2011). PG9 appears to contact two gp120 monomers at the apex of the HIV-1 trimer (Julien et al., 2013b). One might speculate that the binding of this class of antibodies to the Env trimer interferes with its orderly disassembly during the fusion process. Another possibility that is supported by binding studies using Env trimers expressed on virus-like particles is that a single PG9- or PG16-binding event induces a coordinated change to a trimer state incapable of CD4 binding (Löving et al., 2013). PG16 fails to completely neutralize some sensitive viral isolates even at very high antibody concentrations (Walker et al., 2009) probably because it has an absolute requirement for high-mannose N-glycans and therefore cannot neutralize virions bearing heterogeneous glycans at position Asn160.

Asn234 and Asn276 Glycans

The third and smallest class of glycan-dependent anti-HIV-1 antibodies is represented by 8ANC195, which is one of only two clonal relatives belonging to this class of antibodies to date (Scheid et al., 2011). Neutralization by this antibody depends on glycans attached to positions Asn234 and Asn276 (Chuang et al., 2013; West et al., 2013) in a region adjacent to the CD4-binding site. Binding and neutralization experiments indicate that 8ANC195 recognizes either high-mannose N-glycans or the core pentasaccharide common to both complex-type and high-mannose N-glycans. Although the 8ANC195 epitope appears to be different from previously-characterized epitopes (Chuang et al., 2013; West et al., 2013), the structure of liganded 8ANC195 has not yet been reported, and therefore its precise mode of binding, its protein contact residues if any, and its mode of interference with viral entry are not known.

MPER Epitopes

The bNAbs 4E10, 2F5, Z13, 10E8, and m66.6 recognize linear epitopes in the MPER of gp41 (Huang et al., 2012; Nelson et al., 2007; Zhu et al., 2011; Zwick et al., 2005). The MPER contains conserved residues that presumably play a critical role in the fusion process. The transient exposure of this region (Alam et al., 2009; Frey et al., 2008; Sattentau et al., 1995), steric factors (Klein et al., 2009), hydrophobicity (Haynes et al., 2005), and host mimicry (Finton et al., 2013; Haynes et al., 2005; Yang et al., 2013) appear to limit the frequency with which bNAbs to this region are elicited. Moreover, the partially buried MPER epitopes tend to elicit hydrophobic and polyreactive antibody paratopes (Haynes et al., 2005; Sun et al., 2008), which are usually selected against during B cell development (Chen et al., 2013; Doyle-Cooper et al., 2013; Mouquet and Nussenzweig, 2012; Verkoczy et al., 2010; Wardemann et al., 2003). Finally, antibodies against linear MPER epitopes appear to be selected against because of potential autoreactivity against MPER sequences that are also found in human proteins; human kynureninase contains the 2F5 epitope (ELDKWA), and the 4E10 recognizes a motif in splicing factor 3b subunit 3 (SF3B3) (Yang et al., 2013) and an inositol triphosphate receptor (Finton et al., 2013). Not all B cells expressing such antibodies are deleted, however, as demonstrated by the ability of MPER peptide liposomes to trigger expansion and activation of residual anergic B cells in a mouse model (Verkoczy et al., 2013).

A Distinct Conformational Epitope

The clonal bNAb variants 3BC176 and 3BC315 appear to recognize a distinct conformational epitope on the spike trimer, but the location of their binding site on the HIV-1 Env remains to be determined (Klein et al., 2012a).

Unusual Features of Anti-HIV-1 bNAbs

HIV-1 infection in humanized mice and SHIV infection in macaques can be prevented by passive transfer of bNAbs (Klein et al., 2013b; Kramer et al., 2007; Mascola and Montefiori, 2010; Watkins et al., 2011). Although it is therefore widely believed that a vaccine that elicits such antibodies would be effective in protecting against HIV-1 infection, extensive experimentation in animal models and clinical trials in humans have not elicited these antibodies by vaccination (McCoy and Weiss, 2013). One key question in this area of biology is why not?

Molecular characterization of the newly available bNAbs and their interactions with antigen has provided some clues to understanding why these antibodies are difficult to elicit. Antibody variable domains can be divided into structurally diverse CDRs that usually contact the antigen and relatively invariant framework regions (FWRs) (Figure 4). The heavy chain CDR H3 is usually the most important of the six CDRs. It is highly variable because it is assembled by random and imprecise joining of the variable (V), diversity (D), and joining (J) region gene segments of the antibody gene during V(D)J recombination.

Figure 4. Antibody Structure Showing FWR and CDR Regions.

(A) Side + top view of ribbon diagram with color-coded FWR and CDR regions. (B) Side + top view of surface diagram with color-coded FWR and CDR regions. HC, dark blue, LC, light blue; CDRHs, magenta; CDRLs, teal. (C) Map of the FWR and CDR regions of the heavy and light variable domains.

CDR Loops

Many bNAbs have unusually long or short CDR loops. The length of CDR3 regions of the heavy and light chain variable domains is normally determined by N and P nucleotide addition and by resection of DNA ends during V(D)J recombination (Fugmann et al., 2000; Jung et al., 2006). Antibodies with very long CDR H3 loops are frequently deleted at the first tolerance checkpoint (Meffre et al., 2001). In humans the average CDR H3 is 16 residues (Tiller et al., 2007), but in PGT145 the CDR H3 is 33 residues (CDR lengths based on the IMGT definition)(Lefranc, 2011; Walker et al., 2011a). Other HIV-1 bNAbs with long CDR H3 loops include PG9/PG16 (30); CH01-CH04, PGT121, 10-1074 (26); 2F5 (24); 10E8, 8ANC195 (22); PGT127, PGT128, HJ16 (21); 4E10, b12, and PGT135 (20). In some cases such as PG9/PG16, the long CDR H3 loops permit penetration of the glycan shield, whereas in others such as the anti-MPER antibody 2F5, the CDR H3 binds its linear epitope and extends a hydrophobic patch into the viral membrane (Ofek et al., 2010). Long CDR3s are not obligate features of bNAbs against HIV-1, however. For example, the group of CD4-mimetic bNAbs derived from VH1-2, which include some of the broadest and most potent bNAbs, have unusually short CDR L3 loops of only five amino acids (West et al., 2012).

Somatic Mutation

The one feature shared by all anti-HIV-1 bNAbs characterized to date irrespective of their target site on the HIV-1 spike is a high degree of somatic hypermutation. Most affinity-matured human V_H domains carry 15–20 somatic mutations at the nucleotide level (Tiller et al., 2007), while anti-HIV-1 antibodies are generally more extensively mutated (Scheid et al., 2009), and members of the bNAb subset of anti-HIV-1 antibodies carry 40–100 V_H gene mutations (Klein et al., 2013a; Mouquet et al., 2012a; Scheid et al., 2011; Walker et al., 2011a; Walker et al., 2009; Wu et al., 2010; Wu et al., 2011; Xiao et al., 2009). These mutations are essential for the binding and neutralizing activity of the bNAbs since reverting the antibodies to their germline coding sequence results in complete loss of neutralizing activity (Hoot et al., 2013; Klein et al., 2013a; Liao et al., 2013; Scheid et al., 2011; Zhou et al., 2010).

Somatic mutations are introduced into antibody genes by the enzyme activation-induced cytidine deaminase (AID), which is expressed in B cells undergoing clonal expansion in the germinal center (Victora and Nussenzweig, 2012). The same enzyme also produces insertions and deletions, which are rare in most antibodies but common among bNAbs (Klein et al., 2013a; Victora and Nussenzweig, 2012). Since mutation is random, it can lead to autore-activity, decreased affinity, or have no effect. AID is preferentially targeted to the CDRs and not to the FWRs because the latter have fewer sequences that serve as AID hotspots. In addition, the nucleotide code in the framework regions has evolved to be maximally tolerant of mutational base changes without affecting the amino acid sequence (Reynaud et al., 1995; Wagner et al., 1995). This is an essential protective feature of the mutational mechanism because mutations that alter the FWRs can destroy the antibody fold, whereas mutations in the CDRs are more easily toler ated since they are in loops that do not affect the structural underpinnings of the antibody domain (Figure 4). Thus, while FWR mutations occur in mature antibodies, these nucleotide alterations generally do not alter antigen binding (Jones et al., 1986). For example, the neutralizing activity of broadly neutralizing anti-influenza antibody FI6 was not affected by reverting framework mutations (Corti et al., 2011). Indeed the relatively constant nature of the FWR explains why mouse monoclonal antibodies can be humanized for use as therapeutics in humans by creating chimeric molecules bearing mouse CDRs and human FWRs (Jones et al., 1986).

Consistent with these observations, reverting the FWRs to germline in anti-HIV-1 antibodies with limited breadth and potency had little effect on their activity (Klein et al., 2013a). In contrast, FWR mutations were critical for anti-HIV-1 bNAb neutralization and binding. Reversion of the FWR to germline eliminated binding and neutralizing activity for nearly all of the bNAbs tested even when FWR contact residues were spared and only noncontact residues were reverted (Klein et al., 2013a).

Why are FWR mutations essential for HIV-1 bNAb breadth and potency? In some cases, FWR residues contribute to bNAb binding directly by contacting the antigen. For example in the CD4-binding site antibodies VRC01, NIH45-46, 12A12, and 12A21, FWR residue 74 is mutated from Ser to Tyr, and the larger aromatic residue forms hydrogen bonds with main chain atoms of the gp120 bridging sheet. In another CD4-binding site antibody, 3BNC117, somatic hypermutation introduced a four amino acid insertion in the FWR3 that is essential for neutralization of certain viral strains and might bridge the CD4-binding site epitope with V1/V2 residues (Klein et al., 2013a; Scheid et al., 2011; Zhou et al., 2013).

In bNAbs recognizing N-linked glycan-containing epitopes, FWR mutations can play at least two roles; in 2G12 they contribute to the assembly of the contact surface between the domain-swapped V_H domains that form its unusual tripartite antigen-binding site (Calarese et al., 2003). In PGT135, as in CD4-mimetic antibodies, FWR mutations also contribute the direct contacts between antigen and antibody; for example, light chain FWR residue 63 is mutated from Ser to Val, which allows van der Waals contacts with mannose residues on the glycan attached to gp120 residue Asn392 (Kong et al., 2013). Finally, FWR mutations can also contribute to breadth and potency by increasing the overall flexibility of the antibody. For example, a FWR mutation in 3BNC60 introduces a Pro that disrupts the main chain hydrogen bonding between V_H domain strands C’ and C” (Klein et al., 2013a). Reversion of this mutation greatly reduced the neutralization potency of 3BNC60 against more than a third of strains tested, indicating that the conformational flexibility provided by this FWR mutation was often critical. In conclusion, FWR mutations contribute to the activity of bNAbs in a number of different ways, and they are a consistent feature of the most broad and potent anti-HIV-1 antibodies irrespective of their Env target sites.

Polyreactivity

HIV-1 antibodies are frequently polyreactive, which means that they bind to numerous non-HIV-1 antigens, although with relatively low affinities. Highly polyreactive antibodies are removed from the B cell repertoire during development (Wardemann et al., 2003), but anti-HIV-1 antibodies appear to be an exception to this rule (Haynes et al., 2005; Mouquet and Nussenzweig, 2012). One potential explanation for this unexpected finding is that the low density of Env spikes and Env trimer geometry generally prevent bivalent binding (Klein and Bjorkman, 2010), and polyreactivity is selected as a means of providing a second binding interaction for anti-HIV antibodies (Mouquet et al., 2010). For example, lipid binding by bNAbs 2F5 and 4E10 has been proposed to increase the local concentration of these anti-MPER antibodies in the vicinity of their gp41 epitopes (Alam et al., 2009). In a proof-of-concept experiment that supports this idea, bivalent anti-HIV reagents with one Fab unit targeting gp120 neutralizing epitopes and the other targeting a non-neutralizing gp41 epitope showed enhanced anti-HIV potency (Mouquet et al., 2012b).

Vaccination to Produce Anti-HIV-1 Neutralizing Antibodies

An important unresolved issue is why only certain infected individuals produce these bNAb responses to HIV-1. It is possible that the ability to produce bNAbs is determined by host genetic factors, but there are as yet no studies that support this hypothesis (Euler et al., 2013). An alternative possibility, first suggested by studies in macaques, is that the initial envelope sequence that the immune system encounters is an important determinant of whether or not neutralization breadth develops (Shingai et al., 2012; Walker et al., 2011b). Two SHIVs that differ only in their Env sequences elicit distinct neutralizing antibody responses in infected macaques. SHIV_AD8 elicited neutralizing breadth in most infected animals, while only 1 out of 36 macaques infected with SHIV_DH12 developed breadth (Shingai et al., 2012).

The idea that the Env sequence, or the antigen, is one of the factors that favors development of bNAbs is also supported by recent longitudinal observations on a superinfected patient who developed neutralizing breadth (Doria-Rose et al., 2014). Although two viruses were present in that individual, bNAbs developed in response to only one of the two viruses, and therefore one virus was better than the other in eliciting the response in this individual. Moreover, it appears that patients that are superinfected have a higher probability of developing neutralizing breadth (Cortez et al., 2012; Powell et al., 2010) possibly because their immune systems are exposed to a larger number of envelope sequences and therefore have a higher likelihood of encountering one that can initiate and then shepherd the appropriate response.

Given the remaining uncertainties about how bNAbs develop in infected individuals, two approaches to vaccination have been proposed: (1) rational immunogen design based on the structure of bNAbs and their ligands (Burton et al., 2012; Sattentau and McMichael, 2010; Zolla-Pazner, 2004); (2) reproduction of the naturally-occurring pathway to bNAb development by replicating what happens in individuals who exhibit serologic breadth and potency (Haynes et al., 2012; Liao et al., 2013; Scheid et al., 2009; Wibmer et al., 2013).

As reviewed above, many of the important molecular details of the interactions between bNAbs and Env have been described. However, the first group of HIV-1 immunogens designed based on these interactions failed to elicit bNAbs in experimental animals (Burton et al., 2012; Kwong and Mascola, 2012). An unappreciated problem at the time of their design was that most of the unmutated germline precursors of the bNAbs do not bind to HIV-1 spike proteins from most viral isolates (Hoot et al., 2013; Klein et al., 2013a; Scheid et al., 2011; Zhou et al., 2010). Therefore, B cells producing bNAb precursors could not be recruited into immune responses by these immunogens. Although MPER peptides and a gp41 fusion intermediate peptide mimic bind to the germline precursor of 2F5 with modest affinity (Alam et al., 2011), the B cells expressing these antibodies are cross-reactive with autologous antigens and likely deleted during development (Finton et al., 2013; Haynes et al., 2005; Wardemann et al., 2003; Yang et al., 2013).

New approaches to structure-based design have succeeded in engineering immunogens that bind to the unmutated VRC01 class precursors (Jardine et al., 2013; McGuire et al., 2013). In addition, chemically synthesized oligomannose V1/V2 glycopeptides have been developed that bind to the germline precursors of PG9 and CH01 (Alam et al., 2013). In theory, immunization with members of this new generation of antigens should expand the pool of B cells that carry the bNAb precursors. However, recruiting these B cells into the immune response may not to be sufficient to produce breadth or potency, which only appear to develop with extensive somatic mutation and selection in the germinal center (Victora and Nussenzweig, 2012).

Typical high-affinity human antibodies that develop in response to pathogens carry only 10–15 somatic mutations (Tiller et al., 2007). By first principles, immunization with an HIV-1 protein, including a rationally-designed immunogen, should also produce high-affinity antibodies. But repeated boosting with the immunogen will not significantly alter affinity or result in additional accumulation of somatic mutations because affinity-based selection in the germinal center selection is physically limited by two invariant parameters: diffusion and the rate of antigen internalization (Batista and Neuberger, 1998; Foote and Milstein, 1991). Affinity is a ratio of the onand off-rates of the antigen-antibody complex. The on-rate is diffusion limited. The off-rate is limited by the rate of antibody internalization by the B cell because once the antigen is inside the cell it will be degraded irrespective of affinity for the membrane-bound antibody on the cell surface. Moreover, memory B cells with high-affinity antigen receptors are typically shunted away from the germinal center fate to become antibody-producing plasma cells (Victora and Nussenzweig, 2012).

So how do anti-HIV-1 bNAbs accumulate 40–100 mutations in their heavy chain variable genes? One possibility is that high rates of mutation reflect the ongoing arms race between the virus and the antibody system that involves repetitive cycles of germinal center maturation (Klein et al., 2013b; Liao et al., 2013; Scheid et al., 2009; Scheid et al., 2011; Wibmer et al., 2013) (Figure 5A). According to this model, the initial production of a strain-specific neutralizing antibody results in selection for antibody-resistant viral variants that are no longer efficiently neutralized and are only recognized by the antibody with low affinity. The low-affinity interactions between the strain-specific antibody and the newly resistant viral variant would allow B cell recruitment to an additional round of affinity maturation and mutation in the germinal center leading to the development of an antibody that neutralizes the initial escape variant and selects for a second generation of antibody-resistant variants (Klein et al., 2013a; Klein et al., 2013b). Iterative cycles of viral escape and antibody maturation would eventually produce highly mutated antibodies that recognize a broad spectrum of HIV-1 viruses. An additional possibility is that escape mutations in Env render previously shielded sites open to antibody recognition and the recruitment of new antibodies into the anti-HIV response (Wibmer et al., 2013).

Figure 5. Arms Race between Antibodies and HIV.

(A) Interplay between evolving antibody response and Env trimer. Antibodies acquire a growing number of somatic mutations, shown as colored bars in the antigen binding region of the antibody. The mutations arise in germinal centers and are responsible for increasing breadth and potency. Rapid escape by mutation in Env is shown as a color change in the trimer. (B) Successive emergence of broadly neutralizing antibodies (Wibmer et al., 2013). Following an initial strain-specific antibody response, a V2-specific bNAb emerges (red wave). Escape from the initial bNAb produces a mutant virus that elicits a CD4-binding site bNAb (blue wave). This response is followed by a third wave against an unknown epitope, shown as a green wave.

Both the germinal center recycling and the epitope exposure due to escape models are supported by recent retrospective studies that traced the parallel evolution of HIV-1 antibody responses (Liao et al., 2013; Moore et al., 2012; Wibmer et al., 2013). Liao and colleagues traced the development of an anti-CD4-binding site antibody in a single individual (Liao et al., 2013). As might be predicted, the B cell clone that was initially expanded in response to infection produced an antibody that bound to and neutralized the transmitted founder virus but failed to neutralize rapidly emerging variants. However, ongoing somatic hypermutation of the original B cell clone produced bNAbs that neutralized the emerging HIV-1 variants and gained breadth as a result. Similar results have been reported for the development of a bNAb targeting the V1/V2 region (Doria-Rose et al., 2014). Reproducing the development of such antibodies might require a new type of vaccination scheme involving the use of specific Env proteins to recruit B cells expressing the germline antibody precursor and subsequent immunizations with antigens representing antibody escape variants.

Further insights into the development of bNAbs have been provided by studies by Moore and colleagues (Moore et al., 2012; Wibmer et al., 2013) (Figure 5B). These studies reveal how viral escape from early strain-specific neutralizing antibodies may drive the development of bNAbs through sequential recognition of multiple epitopes. In the first study, viral escape from a strain-specific antibody induces the appearance of a new epitope (containing an N-linked glycan attached to Asn332) that is subsequently recognized by a bNAb (Moore et al., 2012). A fuller picture of the interplay of antibody development and viral escape was revealed by characterizing the sequential and transient appearance in one individual of three distinct bNAb specificities over the first 4.5 years of infection (Wibmer et al., 2013). After escape from a strain-specific antibody targeting an unusual V2 epitope, a series of bNAbs developed targeting V2, the CD4-binding site, and an uncharacterized quaternary epitope. A common feature of these events was that escape from one antibody specificity exposed new targets for recognition (Figure 5B).

Speculative General Scenario for the Development of bNAbs and Vaccination

HIV-1 Env has evolved such that potent bNAbs are rarely, if ever, generated from the human antibody repertoire by the initial exposure to the transmitted viral strain. Many strain-specific antibodies develop that the virus readily escapes. However, the presence of these strain-specific antibodies in serum has the effect of reducing the Env sequence space available for escape from new antibodies. Perhaps a year, or in some cases more time, is required before this effect becomes significant. Then, when a new antibody develops for which the virus has more limited avenues for escape, an incipient bNAb is able to “chase” the mutating virus, undergoing germinal center maturation and developing breadth. This process contrasts with the early strain-specific antibodies that are not able to overcome escape mutations. Ultimately, in the face of a potent bNAb, the virus must completely escape the bNAb's chase by mutating to a less-fit form that may subsequently restore its fitness by secondary mutations.

An HIV-1 vaccine may have to mimic this sequential approach starting with immunogens designed to engage the appropriate B cell precursors followed by iterative boosts with naturally occurring Envs associated with the development of bNAbs. Given the nature of the physiologic process that leads to production of bNAbs, it is highly unlikely that repeated vaccination with a single monomorphic antigen could elicit a broadly active vaccine.

Immunotherapy with Anti-HIV-1 Antibodies

The idea that antibodies could be used to treat infectious diseases was first proposed by Kitasato Shibasaburo and Emil von Behring over 100 years ago. In the 20th century antibiotics rapidly replaced serum therapy for infectious diseases, and today antibodies are in clinical use primarily for inflammatory diseases and cancer.

Like many other infections, HIV-1 can be treated with conventional small-molecule drugs. Combinations of antiretroviral drugs that target different aspects of the viral life cycle are particularly effective in controlling viremia (Grant and Zolopa, 2010; Olender et al., 2012; Simon et al., 2006; Vergidis et al., 2009). Therapy must be continued indefinitely, however, because antiretroviral therapy (ART) fails to eliminate latent viral reservoirs and viremia rebounds as soon as the drugs are discontinued (Hamlyn et al., 2012).

Antibodies differ from ART in a number of important ways that might be useful in therapeutic approaches to HIV-1 and may complement small-molecule-based approaches to HIV-1 treatment. For example, antibodies usually have a half-life of 14–21 days in humans, which means that they may only have to be administered once every 2–4 months. In addition to their direct neutralizing effects, antibody effector functions mediated through the Fc region expedite viral clearance from circulation (Igarashi et al., 1999; Nimmerjahn and Ravetch, 2008). Through antibody-dependent cellular cytotoxicity (ADCC), antibodies can engage diverse effector cells such as NK cells, monocytes, and activated macrophages to kill HIV-1-infected target cells. Finally, some broadly neutralizing antibodies can interfere with cell-to-cell spread of HIV-1 (Malbec et al., 2013).

The idea that antibodies might be useful adjuncts to standard HIV-1 therapy was initially tested over a decade ago in humanized mice and later in clinical trials in infected individuals using combinations of first generation bNAbs (Mehandru et al., 2007; Poignard et al., 1999; Trkola et al., 2005). Humanized mice were infected with HIV-1_JR-CSF or HIV-1_SF162 and treated with high concentrations of a mixture of b12 (a CD4-binding site antibody), 2G12 (a glycan-specific antibody) and 2F5 (an anti-MPER antibody). This mixture failed to control viremia. Indeed, triply resistant viral variants were reported to emerge within one week of initiation of therapy, and it was concluded that antibodies would be unable to control viremia due to rapid escape (Poignard et al., 1999).

Human trials re-enforced the conclusions of the mouse studies (Mehandru et al., 2007; Trkola et al., 2005). Two independent groups of investigators treated individuals infected with antibody-sensitive HIV-1 strains using a combination of the anti-glycan bNAb 2G12 and anti-MPER bNAbs 2F5 and 4E10. In these clinical trials, viremia was initially lowered by conventional ART, and antibodies were added before discontinuing ART (Mehandru et al., 2007; Trkola et al., 2005). Antibody cocktails were unable to maintain viremic control in all but a very small number of individuals. As in the humanized mice, resurgent viremia on antibody therapy was associated with the emergence of antibody-resistant variants. The viruses emerging during combination therapy in humans were only resistant to 2G12 suggesting that this may have been the only antibody in the combination to exert pressure on the virus in humans (Mehandru et al., 2007; Trkola et al., 2005). The results of the human studies reinforced the conclusion that antibody therapy would not be possible for HIV-1 infection due to rapid emergence of resistant viral variants (Mehandru et al., 2007; Poignard et al., 1999; Trkola et al., 2005). An important caveat to interpreting these initial humanized mouse and human therapy experiments is that the antibodies that were used were far less potent than currently available reagents.

The possibility that antibodies might be used in therapy was recently retested using newly available bNAbs in humanized mice infected with HIV-1_YU-2, a difficult to neutralize (tier 2) clade B strain. When used as single agents, some antibodies (3BNC117, 10-1074, PGT128, PG16, NIH45-46^G54W) but not others (10E8, 3BC176) induced a small transient drop in viremia that was associated with the selection of resistant HIV-1_YU-2 variants that carried specific mutations in antibody target sites (Diskin et al., 2013; Horwitz et al., 2013; Klein et al., 2012b; Malbec et al., 2013). For example, anti-CD4-binding site antibody NIH45-46^G54W (Klein et al., 2012b) selected variants carrying mutations that altered the CD4-binding site. One of the notable and unexpected results was that the variants that escaped from one of the antibodies, 10-1074 (directed at the Asn332 glycan at the base of V3), were identical in several independent samples, suggesting that the pathways to escape this class of antibodies while maintaining viral fitness are limited (Klein et al., 2012b).

In contrast to earlier experiments with humanized mice, combined antibody therapy with second generation bNAbs led to sustained viremic suppression to levels below the limit of detection (Klein et al., 2012b). Most importantly, there was no escape during a 120 day period or for as long as antibodies remained at therapeutic concentrations, and in 10%–15% of the mice, viremia failed to rebound even after antibodies were no longer detectable (Diskin et al., 2013; Horwitz et al., 2013; Klein et al., 2012b). Durable control could also be achieved with single antibodies delivered by infusion or by gene therapy using adeno-associated virus vectors if the initial level of viremia was lowered with ART (Horwitz et al., 2013). Thus, the ability of a single antibody to control viremia is directly related to the initial viral load (Horwitz et al., 2013). In addition to controlling viremia, antibody therapy also lowered cell-associated viral RNA and DNA, which are thought to reflect the content of the viral reservoir (Horwitz et al., 2013). These initial experiments in humanized mice established the principle that antibodies can be used to treat HIV-1 and that this type of therapy may impact the overall size of the viral reservoir.

Many of these initial observations in humanized mice were confirmed and extended in macaques infected with SHIV, a chimeric virus that carries a human HIV-1 Env spike and causes an AIDS-like syndrome (Barouch et al., 2013; Shingai et al., 2013). Two independent groups of investigators performed antibody therapy experiments in macaques that were chronically infected for a period of 9 months to 3 years with one of two SHIV strains, SHIV_SF162P3 (Barouch et al., 2013), and SHIV_AD8 (Shingai et al., 2013). Both viruses can cause an AIDS-like disease but some monkeys infected with SF162P3 can resolve their infections spontaneously (Hsu et al., 2003), whereas those infected with AD8 invariably died due to loss of CD4 T cells, lymphoma, and opportunistic infection (Gautam et al., 2012). Macaques infected with SHIV are far larger than humanized mice. They carry a greater viral load that is more diverse, and therefore these animals may represent a greater challenge to antibody therapy. In contrast to humanized mice that do not have an intact immune system, macaques, like humans, have an immune system that actively attempts to control the infection through both T and B cell responses.

As might be expected from the documented effects of antibodies on viral half-life in macaques (Igarashi et al., 1999), both studies showed that antibodies rapidly controlled SHIV_SF162P3 and SHIV_AD8 viremia (Barouch et al., 2013; Shingai et al., 2013). Like in humanized mice (Horwitz et al., 2013; Klein et al., 2012b), antibodies decreased cell-associated RNA, and viremia remained controlled for as long as antibody levels were above the therapeutic range of 5–10 μg per milliliter for up to 56 days after infusion (Barouch et al., 2013). Moreover, in both humanized mice and macaques, the ability of an antibody to control infection was directly related to the level of viremia before initiating therapy (Barouch et al., 2013; Horwitz et al., 2013). Like in humanized mice, in a small fraction of the SHIV_SF162P3- infected macaques viremia failed to rebound even after the antibodies were cleared (Barouch et al., 2013; Klein et al., 2012b).

In contrast to humanized mice, in which single antibodies produced only a small transient drop in viremia with rapid emergence of escape variants, a majority of the macaques showed control of viremia for as long as their antibody concentrations remained above 1–5 μg/ml (Barouch et al., 2013; Shingai et al., 2013). In addition, all SHIV_SF162P3-infected animals that were retreated after the initial rebound of viremia responded with a shorter, but significant suppression of the viral load with no emergence of characteristic escape mutations (Barouch et al., 2013). The inability to find previously described escape variants to PGT121 in SF162P3-infected macaques is likely because mutations in the target site for this antibody failed to produce resistant variants in SF162P3 (Barouch et al., 2013). This is likely to be a peculiarity of SHIV_SF162P3 since HIVs carrying similar mutations become resistant to PGT121 and 20%–30% of all HIV-1 strains in large viral panels including many clade A and C viruses are resistant to this antibody (Moore et al., 2012; Mouquet et al., 2012a; Walker et al., 2011a). In the case of SHIV_AD8, resistance failed to emerge in animals during initial treatment with the combination of 10-1074 and 3BNC117 (5 macaques), or 3BNC117 alone (2 macaques), but escape variants to 10-1074 were detected (2 macaques) (Shingai et al., 2013). PGT121 and 10-1074 have nearly indistinguishable levels of activity when tested against large standardized panels of HIV-1 strains in TZM-bl assays (Moore et al., 2012; Mouquet et al., 2012a; Walker et al., 2011a). However, 10-1074 is more active than PGT121 against recently obtained clade B transmitted founder viruses (Mouquet et al., 2012a). Whether these two antibodies more generally differ in their ability to prevent viral escape cannot be determined from the available macaque experiments because they were performed with two different SHIVs. However, the fact that the two antibodies are very similar in their breadth and potency on large viral panels suggests that the observed differences are related to the specific SHIVs being tested and may be anecdotal.

Why SHIV infection in macaques appears to be easier to control than HIV-1 infection in humanized mice and whether the human infection will be similar to the mouse or monkey is not known. The differences between the two species could be due to existing antiviral immune responses in the macaques, differences between HIV-1 and SHIV, or other factors that remain to be determined. SHIV infection in macaques resembles HIV-1 infection in humans in that initial infection is accompanied by high levels of viremia that decrease by 1–2 logs to a set-point that is maintained for long periods of time before the development of AIDS. The set point is established and then maintained by immune pressure that is imposed predominantly by cytotoxic T cells, and perhaps by antibodies that select resistant viral variants (Huang et al., 2010; Simon et al., 2006; Walker and McMichael, 2012). Although the immune response is unable to control HIV-1 infection in macaques, it is possible that it contributes to and enhances the effects of the passively administered antibodies making escape more difficult in macaques than in the humanized mice that lack these responses. A second possibility is that the human Env protein has fewer degrees of freedom to escape from bNAbs and still retain the ability to infect macaque cells (versus human cells). In support of this idea, only a limited number of Envs can be used to produce SHIVs. Thus, it may be more difficult for the human Env to produce a CD4-binding site escape (to an anti-CD4-binding site antibody) that still binds to macaque CD4 than for the same Env to escape the antibody and retain binding to the autologous human CD4. Understanding why it is easier to control viremia in macaques than in humanized mice will be interesting and is likely to impact on therapeutic strategies. But irrespective of the mechanism, the extensive clinical experience with antiretroviral drugs, and the fact that no single antibody covers all HIV-1 variants, suggests that combinations of antibodies will be required to control viremia in humans.

Perspectives for Antibody-Based Vaccines and Therapy

Antibodies can prevent HIV-1 infection in mice and chimpanzees, and SHIV infection in macaques (Balazs et al., 2012; Bar-ouch et al., 2013; Gruell et al., 2013; Prince et al., 1991; Safrit et al., 1993; Shingai et al., 2013), but attempts to produce effective HIV-1 vaccines by eliciting neutralizing antibodies have been disappointing to date (McCoy and Weiss, 2013). However, the vaccine strategies that were tested did not take into account the recent discovery of naturally arising broad and potent anti-HIV neutralizing antibodies. A key observation is that development of broadly neutralizing antibodies requires extensive somatic mutation (Klein et al., 2013a), which is likely to occur as a result of iterative rounds of germinal center selection (Klein et al., 2013a; Klein et al., 2013b). By longitudinal tracking of the coevolving Env antigen and a clone of broadly neutralizing antibodies, Liao and colleagues illustrated how breadth and poly-reactivity may develop (Liao et al., 2013). The germline ancestor of the CH103 lineage bound to the donor's transmitted/founder Env, and reactivity to the evolving virus developed over time by somatic hypermutation, thereby increasing the potency and breadth of the antibody. Similarly, longitudinal studies by Moore and colleagues revealed how viral evolution in response to antibody pressure exposes additional targets that can elicit bNAbs (Moore et al., 2012; Wibmer et al., 2013). This type of naturally occurring antibody evolution arises as a result of exposure to a series of mutant Env variants over time. Incorporating this concept into vaccine strategies would be a significant departure from current vaccines that typically involve immunization and boosting with a monomorphic antigen.

Antibody cloning also revealed that most humans that develop serologic breadth and potency do so by producing antibodies to more than one site of vulnerability. Moreover, it uncovered a series of targets on Env that are vulnerable to broad and potent antibodies. With a growing understanding of the structural underpinnings of antibody-mediated neutralization, it may become possible to design antigens that recruit antibodies targeted specifically to these regions in Env (Jardine et al., 2013).

Despite recent advances, an effective antibody-based HIV-1 vaccine is still on the distant horizon in part because there are as yet no successful preclinical demonstrations of how to elicit broad and potent anti-HIV-1 antibodies. Moreover, any new vaccine concept is very time consuming to test in humans. In contrast, preclinical experiments in humanized mice and macaques strongly support testing antibody therapies for HIV-1 in humans in the near term.

Since ART is less expensive, and easier to administer, than antibody therapy, passive immunotherapy will have to complement existing therapies in unique and clinically significant ways before it can considered for clinical use. In addition, there are a number of important questions that will need to be addressed to evaluate whether there is any role of antibodies as therapeutics in the anti-HIV armamentarium. First and foremost, it will be essential to determine whether single antibodies or combinations of antibodies are safe and can impact the level of circulating viremia and/or cell-associated viral reservoirs in infected humans. Antibody and ART combinations should be tested for the possibility that antibodies can further decrease viremia in ART-treated individuals with detectable viral loads. Finally, because antibodies have the potential to clear infected cells it will be important to begin to think about the role of antibodies in attempts to develop a cure for HIV-1.