Ontogeny‐based immunogens for the induction of V2‐directed HIV broadly neutralizing antibodies

Summary The development of a preventative HIV vaccine able to elicit broadly neutralizing antibodies (bNAbs) remains a major challenge. Antibodies that recognize the V2 region at the apex of the HIV envelope trimer are among the most common bNAb specificities during chronic infection and many exhibit remarkable breadth and potency. Understanding the developmental pathway of these antibodies has provided insights into their precursors, and the viral strains that engage them, as well as defined how such antibodies mature to acquire breadth. V2‐apex bNAbs are derived from rare precursors with long anionic CDR H3s that are often deleted in the B cell repertoire. However, longitudinal studies suggest that once engaged, these precursors contain many of the structural elements required for neutralization, and can rapidly acquire breadth through moderate levels of somatic hypermutation in response to emerging viral variants. These commonalities in the precursors and mechanism of neutralization have enabled the identification of viral strains that show enhanced reactivity for V2 precursors from multiple donors, and may form the basis of germline targeting approaches. In parallel, new structural insights into the HIV trimer, the target of these quaternary antibodies, has created invaluable new opportunities for ontogeny‐based immunogens designed to select for rare V2‐bNAb precursors, and drive them toward breadth.

bNAbs to prevent HIV infection. Passive immunization of bNAbs isolated from infected donors has long been known to protect non-human primates from infection [reviewed in (1)]. Indeed, a recent study showed that a single injection of bNAbs protected animals against repeated exposure for up to 23 weeks. 2 Furthermore, studies of HIV infected donors have shown that the human immune system has the capacity to make such bNAbs. These findings, along with the failure of traditional vaccine strategies, has led the field to consider next-generation vaccine regimens which are based on detailed studies of the ontogeny of bNAbs during infection, a strategy referred to as the B cell lineage approach. 3 Here, we describe recent virological, immunological, and structural studies supporting and informing this approach, specifically focusing on bNAbs that target the V2 region at the apex of the envelope trimer.

| WHY IS VAN ATTRACTIVE BNAB VACCINE TARGET?
The HIV-1 envelope (Env) glycoprotein complex, which consists of a heterotrimer of three molecules of gp120 and three molecules of gp41, is responsible for mediating viral entry into host cells, and is the sole target of neutralizing antibodies. The first and second variable regions (V1V2) of gp120 are located at the apex of the envelope trimer, and are highly variable in terms of sequence, glycosylation and length, largely due to mutations and insertions in two regions, in the middle of V1 and toward the C-terminal end of V2 ( Figure 1). In contrast, semi-conserved regions exist, particularly in the V2 region which is the focus of this review, in strands B and C, including highly conserved glycans at positions 156 and 160 and other fairly conserved residues such as those 166 and 169, which will be described in further detail below. The V1V2 domain is an important contributor to viral entry and neutralization resistance of HIV isolates [reviewed in (4)].
Studies of transmission pairs suggest that in many cases infection is mediated by viruses with compact V1V2 regions, which subsequently become longer during the course of infection, suggesting a complex interplay between infectivity and the need for neutralization resistance through V1V2 sequence changes, elongation and glycosylation. The role of the V1V2 region in immune evasion is emphasized by the extreme neutralization sensitivity of V1V2-deleted viruses. [5][6][7][8][9] F I G U R E 1 Global alignment and trimeric location of V1V2 within the envelope trimer. (A) Logogram illustrating the amino acid profile of the 2015 version of the LANL premade alignment for Group M V1V2 envelope sequences. The size of each amino acid in the profile indicates prevalence of that amino acid in global sequences. The V1V2 sequence of HXB2 is placed below the profile. Beta strands and strand-connecting loops (SCL A-B and SCL C-D ) are illustrated by arrows and dashed lines respectively. Modified from (58) (B) Side view ribbon representation of the HIV envelope trimer, highlighting the location of the V1V2 domain at the trimer apex. The five anti-parallel beta strands (beta-barrel) are shown in purple, while stick and surface representations of the N156 and N160 glycans are shown in cyan and blue respectively. Key amino acid residues at position 166 and 169 are shown as spheres in green and orange respectively. The approximate position of the viral membrane is indicated. (C) Ribbon representation of the trimer showing the view from the angle of approach by V2-apex bNAbs. PDB ID: 4TVP. N-linked glycans were modeled using http://www.glycosciences.de/modeling/glyprot/php/main.php The V1V2 region is itself a frequent target of neutralizing antibodies that drive viral escape mutations within this region. [10][11][12][13][14][15] In some cases, multiple unrelated B cell lineages target this region, highlighting the immunogenicity of V1V2 during infection. 10 The high variability in this region results in the majority of these autologous neutralizing responses being strain-specific and easy for the virus to evade, providing limited insights for HIV vaccine design. However, within this region, the semi-conserved elements of V2 may also be the target of bNAbs able to recognize diverse circulating viruses, and these are the basis of this review. Thus, while the V1V2 region has long been recognized as a target for antibodies, it was only in 2009 with the isolation of the bNAbs PG9/PG16 that its relevance for vaccine design was truly appreciated.
While almost all infected people develop antibodies to the HIV envelope which have some cross-neutralizing activity, 16 18 In addition, several cohort studies have shown that a major contributor to the development of bNAbs is high levels of antigenic stimulation, in the form of high viral loads and duration of infection, 19,20,28 though there are occasional reports of viral controllers who nonetheless mount bNAb responses. 29 BNAbs normally emerge only several years after infection, and frequently have unusual features that are not favored by the immune system, including autoreactivity, very long or short CDRs depending on the class of antibodies and high levels of somatic hypermutation (SHM), in some cases >30%. 1 The association with duration of infection and the high levels of SHM of many HIV bNAbs suggest a long co-evolutionary pathway, requiring variation in both the virus and the antibody. Indeed, two recent studies suggest that bNAb lineages evolve as quickly as HIV, particularly during the early stages of their development, though these rates later decline. 30,31 This suggests that even during infection in the context of high levels of viral replication and antigenic stimulation, the development of bNAbs is a difficult pathway. These challenges in developing bNAbs are obviously even greater in the context of vaccination, where antigenic stimulation is normally limited to two or three immunogen exposures, and SHM is normally restricted to about 6%. 32 Despite these impediments to bNAb development, much of the surface of the HIV trimer is now known to be vulnerable to bNAbs, with many antibody epitopes including the glycans that provide half of the molecular weight of the HIV envelope, and were initially assumed to function solely as a shield for underlying epitopes. 33,34 These findings come from the isolation of dozens of broadly neutralizing monoclonal antibodies, 35,36 largely due to the optimization of relatively new technologies for human antibody isolation including antigen-specific sorting (most recently using trimeric antigens which are particularly important for the quaternary V2-apex bNAbs 37,38 ), B cell culture with micro-neutralization assays for screening of individual wells, and multiplex RT-PCR for amplifying immunoglobulin genes from single cells [reviewed in (35,39,40)]. Characterization of these new bNAbs has enabled the field to define several conserved viral epitopes including the V2-apex site that is the focus of this review, as well as the N332 glycan supersite, the membrane proximal external region (MPER), the CD4 binding site (CD4bs), and the gp120-gp41 interface, recently shown to include the fusion peptide. 33,41 Although many vaccine strategies aim to elicit a polyclonal response that would ideally target multiple conserved epitopes, several approaches are target-specific.
The use of strategies such as minimal epitopes, and have thus resulted in a concerted effort to understand the breadth of antibodies to specific epitopes such as the V2-apex.
Antibodies to the V2-apex may be especially attractive from a vaccine perspective because they are among the most prevalent of broadly neutralizing responses, with mapping studies in several cohorts of infected donors showing that these account for up to a third of bNAb responses. 17,18,[42][43][44] This indicates that the immune system of many individuals is amenable to the development of these specificities. Furthermore, monoclonal antibodies (mAbs) to this epitope have been isolated from several donors (PG9/PG16 from donor IAVI24, the CH01-04 lineage from donor CH0219, the PGT145/PGDM1400 lineage from IAVI84, and the CAP256-VRC26 lineage from CAP256) enabling a comprehensive understanding of their features and neutralization capacity 37,38,[45][46][47] (Table 1 and Figure 2). These mAbs all have structural properties in common, such as an unusually long CDR H3 loop that is generally highly anionic. This negative CDR H3 charge is provided both by an abundance of aspartate and glutamate residues, and by sulfation of tyrosines, and facilitates recognition of the positively charged V2 epitope through a shared mode of recognition, described below ( Figure 2). These mAbs often have substantial breadth, neutralizing 70%-85% of viruses, 38,46,48 and in some cases remarkable neutralization potency in the nanomolar range. 37,38,46,47 The need to understand future vaccine coverage, and more immediately the potential utility of mAbs in passive immunization, has led to studies using large panels of transmitted/founder viruses (the infecting virus that both active or passive immunization strategies will need to block) and assessed which combinations of bNAbs may provide optimal potential clinical benefit. 48,49 In one such study, a V2-apex antibody, CAP256-VRC26.25 was included in both the best 2-mAb and 3-mAb combinations, highlighting the potential of this specificity in passive immunization, but also in vaccine-elicited responses. Similarly, a combination of V2-directed PGDM1400 with PGT121 (which targets the N332 supersite) has been shown to achieve extraordinary breadth and potency, neutralizing 98% of that particular panel of viruses at a median IC 50 of 0.007 μg/mL. 38 Although the breadth of some individual CD4bs mAbs and MPER antibodies exceeds those of V2-directed broad antibodies, their high levels of SHM and (in the latter case) lipid reactivity and autoreactivity may suggest difficulties in their elicitation by vaccination. In contrast, V2-apex antibodies may acquire breadth with moderate levels of SHM, often as little as 15% divergent from their unmutated common ancestor (UCA) 47 (Table 1).
Interest in the V2 region has further increased since the findings of the RV144 vaccine trial, a canary-pox prime and gp120 protein boost regimen that was tested in Thailand in 2009, and which showed moderate efficacy of 31%. 50 A subsequent immune correlates analysis showed that vaccine-induced IgG against V1V2 was inversely correlated with risk of HIV acquisition, 51 a finding supported by a "sieve analysis" that showed immune pressure in V2 in viruses infecting vaccine recipients 52 and by several follow-up studies [reviewed in (53)].
However, although such antibodies were able to mediate antibodydependent cellular cytotoxicity and virus capture, there is no evidence that the low level protection observed during the RV144 vaccine trial was mediated through neutralization. Nonetheless, RV144 has spawned a wide array of immunologic studies, and follow-up vaccine trials such as those currently underway in sub-Saharan Africa in the HVTN 702 trial to test HIV-1 subtype C tailored equivalents. These trials will undoubtedly continue to reveal insights into the role of nonneutralizing antibodies to V1V2 and under-appreciated effector functions of antibodies in HIV vaccine efficacy.

| STRUCTURAL INSIGHTS INTO V2-APEX ANTIBODY MEDIATED NEUTRALIZATION
The failure of empirical vaccine approaches for HIV has led to an increased dependence on more structure-based approaches to immunogen design. Studies of HIV envelope structural biology have provided key insights into the HIV-1 envelope and broadly neutralizing antibodies, including V2-directed antibodies, that are able to circumvent its considerable defenses. A crystal structure of the HIV-1 envelope gp120 core was solved in 1998, and provided the first atomic-level details of the HIV-1 Env. 54 This core construct, however, lacked the V1V2 domain (residues 126-196, HXB2 numbering), which resisted crystallization for another 13 years. A critical breakthrough in determining the structure of the conformationally variable and highly glycosylated V1V2 domain came through the isolation of glycan-dependent broadly neutralizing antibodies, such as mAb PG9/PG16 that bound V1V2 only in its native form. 46 These antibodies led, in 2011, to the first crystal structures of the V1V2 domain, which were achieved using scaffolded constructs, with the V1V2 region engrafted onto a heterologous protein, bound to PG9 55 ( Figure 3A). Importantly, the recently discovered V1V2-directed bNAbs enabled an "on-column" method, where mAbs were coupled to the column and used to specifically purify properly folded and glycosylated V1V2 scaffolds. Viral strains ZM109 and CAP45, both clade C viruses, were solved at 1.8 Å and 2.2 Å respectively. The structures not only revealed atomic-level details regarding the neutralization mechanism of PG9 but also provided the first glimpse at the two variable domains that cap the trimeric viral envelope. V1V2 was shown to comprise a four stranded beta-sheet with disordered loops between strands A and B and strands C and D, and later structures would reveal an additional strand not seen in the initial scaffolded structures due to dimerization of the V1V2 domains. [56][57][58] Although, as described above, V1V2 had long been recognized as a highly variable region, the sequence elements comprising the strands of the domain are more conserved than the outer loop regions ( Figure 1A). The unusually long CDR H3 engaged strand C of V2 through main-chain interactions forming a parallel β-strand along the edge of the V1V2 sheet. 59  side-chain interactions with this positively charged region buried under the glycan shield. Although the C-strand of V2 is more conserved than the loop regions, the sequence variations that do exist could be better tolerated due to the CDR H3 main-chain interactions with the peptide backbone, which is largely sequence-independent, providing a mechanism for breadth among antibodies targeting this site.
PG9-bound V1V2 revealed how bNAbs could target this region not only by penetrating the glycan shield, but by using the dense array of glycans, typically regarded as a viral defense, as part of their epitope. PG9 neutralization is dependent upon glycans at residues N156 (or N173 in some strains) and N160, with the latter conserved glycan being critical for recognition ( Figure 3A). The scaffolded proteins in the crystal structures were grown in HEK293S GnTI − cells, which lack N-acetylglucosaminyltransferase I activity, trapping the glycosylation pathway at an early stage. On-column purification further selected scaffolds with glycosylation profiles favorable to PG9 binding, Man5GlcNAc2 (Man5) glycans at N156 and N160. ZM109 falls among a minority of HIV-1 strains that do not have a glycan at N156 but instead contains a glycan at residue N173, which is in close threedimensional proximity to N156 in the structure and serves as a viable substitute. The co-crystal structure of PG16 with a scaffolded V1V2 domain from ZM109, grown in cells treated with the α-mannosidase II inhibitor swainsonine (predicted to induce hybrid-type glycans), highlighted a difference in glycan specificity for this clonal relative of PG9.
These preferences suggest that while glycans at residues N156/N173 and N160 may primarily consist of Man5, 60 they are not completely homogeneous and that antibody lineages differentiate to accommodate various glycoforms, as has been observed also for glycantargeting bNAb specificities to other conserved sites. 61 Comparison of V2-apex targeting bNAbs with non-broadly neutralizing antibodies that target this same region has also yielded useful information. Sieve analysis during the RV144 vaccine trial had detected immune signatures at sites 169 (an important part of bNAb epitopes) and residue 181. This was supported by the isolation of two V2directed antibodies, called CH58 and CH59, both isolated from RV144 vaccines. These antibodies, despite apparently targeting conserved elements of V2, did not display broadly neutralizing activity. 62 Co-crystal structures of these vaccine-elicited antibodies with a peptide from the V1V2 domain corresponding to the C-strand revealed that these antibodies bound to helical and loop conformations considerably different from those seen in the scaffolded structures ( Figure 3B). Surface Plasmon Resonance assays with these antibodies and PG9 confirmed that within a population of monomeric gp120s, the C-strand of V1V2 could adopt multiple conformations, each recognized by a different

| ELUCIDATING THE TRIMER STRUCTURE PROVIDES NEW TOOLS FOR IMMUNOGEN DESIGN
Broadly neutralizing antibodies that target the V2 region prefer or require a quaternary epitope, [45][46][47] which has complicated attempts to elicit such antibodies. The establishment of a soluble prefusion closed trimer 63 enabled these antibodies targeting the trimer interface at the apex of the viral spike to be imaged by negative stain electron microscopy (EM). The first structure, a three-dimensional reconstruction of PG9 in complex with an engineered trimeric envelope called BG505 SOSIP.664, provided a clue as to the quaternary nature of these antibodies. 64 PG9 bound asymmetrically to the cap of the spike with a stoichiometry of a single Fab per trimer ( Figure 3C). Using the threefold symmetry of the Env and fitting the previously determined scaffolded complex crystal structure with PG9, the authors suggested that PG9 likely would interact with the N156 (or N173) and N160 glycans of the bound protomer, but would likely also bind the N160 glycan of a neighboring protomer ( Figure 3C) Figure 3D). Both antibodies bound the C-strand through mainchain strand-strand hydrogen bonds, though PG9 also showed more

| TRIGGERING PRECURSORS OF V2-APEX LINEAGES
The long CDR H3s that characterize V2-apex antibodies range from 24-37 amino acids (Kabat numbering). These are in contrast with typical antibodies, which have a mean length of 14-16 amino acids. [70][71][72][73] Antibodies with such long CDR H3s are relatively rare -those with CDR bind the vast majority of strains. Indeed, in CAP256 for which the most accurate V2-apex precursor has been determined, the UCA neutralized only the superinfecting transmitted/founder virus, but not the primary transmitted/founder virus or any other virus from a panel of two hundred. 69 Identifying the virus that triggered the lineage was therefore a major interest, especially given that the CAP256-VRC26 lineage was detected only 20 weeks after superinfection. 47

| INCORPORATING ONTOGENY INTO STRUCTURE-BASED IMMUNOGEN DESIGN
Studies of the ontogeny of V2-apex antibodies have led to increased efforts to incorporate structural and virological findings into targeted immunogens. Many clinical trials based on immunization with monomeric gp120 have failed to elicit bNAbs, including those directed to the V2-apex. 82 The structural explanation for this latter finding is that such gp120 monomers do not form the full V2-apex epitope and can adopt multiple conformations of C strand. 62

| CONCLUSIONS
With the availability of a soluble trimer that mimics that native spike, a great many ideas for immunogen design that were technologically unavailable just a few years ago are now possible. An explosion of crystal and EM structures 56, 57, 64-66 has also provided atomic-level details to further stabilize the molecules. Replacement of the V1V2 domain or design of gp120 chimeras which require an understanding of the atomic-level details of the trimeric interactions are possible now that structures of the trimers are available. Recent advances have also been achieved in structurally defining the glycan shield surrounding the trimer that must be accounted for by V2 bNAbs. Additionally, technology to assemble these trimers onto nanoparticles is still under development but the field is moving forward at an extraordinary pace with these new tools. Open questions still remain regarding the exact mechanism of how specific strains are engaged by V2 bNAb precursors, what the optimal glycan requirements are for eliciting bNAbs and whether vaccine strategies need to incorporate diversity (and if so how much) to drive antibodies toward breadth, as suggested by studies of infection. However, stabilized trimers presenting the appropriate quaternary epitope for stimulating V2 bNAb precursors represent one of the most exciting immunogen strategies, and has only become possible with the through our recent understanding of these antibody ontogenies as well as the emergence of sophisticated soluble trimeric reagents in the past few years.