Env exceptionalism – Why are HIV-1 Env glycoproteins atypical immunogens?

Introduction

The recent abandonment, due to futility, of the latest HIV-1 vaccine efficacy trial, HVTN-702, increases the emphasis on immunization strategies for inducing broadly neutralizing Abs (bNAbs) (Cohen, 2020). The HIV-1 Envelope glycoproteins (Env) are the basis of many vaccine development strategies, particularly those for eliciting bNAbs. That statement has been true for around 35 years, and yet we still lack Env-based immunogens that elicit Abs with the breadth and potency sufficient to counter global HIV-1 diversity (Escolano et al., 2017; Kwong and Mascola, 2018). Our collective failure is certainly not for want of effort. Sophisticated and, at times, very well-funded immunogen design programs harness knowledge of Env and bNAb structure, and of the ontogeny of bNAbs and their co-evolution with Env in vivo (Bonsignori et al., 2017; McGuire, 2019; Sok and Burton, 2018; Ward and Wilson, 2017).

Many factors render bNAb induction difficult, including the multiple defenses HIV-1 has evolved to counter humoral immunity. For example, relatively conserved Env regions that would bind Abs are well protected by a dense array of glycans, while other vulnerable sites become exposed only transiently when receptor-triggered conformational changes drive Env-mediated fusion of the virus and host cell membranes (Crispin et al, 2018; Klasse, 2012, West et al., 2014). The tolerance of Env to sequence changes without loss of function facilitates rapid escape from bNAbs that emerge in or are administered to HIV-1-infected people clinically (Escolano et al., 2017; Kwong and Mascola, 2018; Schommers et al., 2020; Sok and Burton, 2018; West et al., 2014). During infection, bNAbs evolve in a minority of individuals, usually only after extensive and prolonged somatic hypermutation (SHM) of precursor Abs (Abbott et al., 2018; Kepler and Wiehe, 2017; Klein et al., 2013a). But are there other dimensions to the problems of eliciting bNAbs by Env-based immunogens? Here, we argue that HIV-1 Env is an inherently poor immunogen because it has exceptional properties. We discuss research strategies that could help overcome these hurdles to a more effective HIV-1-vaccine design.

HIV-1 Env structure and function

The native, functional Env complex is a trimeric spike structure, and typically only ~15 are present on the HIV-1 virion surface (Klein and Bjorkman, 2010). The spikes interact with cell-surface receptors (CD4 and then CCR5 or CXCR4) in a process that drives virus-cell membrane fusion in the first stage of the replication cycle (Klasse, 2012; Weiss, 2013). By binding to a sufficient number of spikes before or, more rarely, during the fusion process, NAbs protect target cells from HIV-1 infection (Klasse, 2007, 2014; Regoes and Magnus, 2015). The native trimer comprises three non-covalently associated gp120 and gp41 subunits: gp120 engages the entry receptors; gp41 anchors the spike into the virus membrane and contains the fusion peptide, which merges the viral and cell membranes (Julien et al., 2013; Lee et al., 2016; Lyumkis et al., 2013; Pancera et al., 2014). Almost half the mass of the trimer consists of N-linked glycans; there are ~90 individual glycan sites per trimer, most of them on gp120. Recombinant gp140 mimics of the viral spike (native-like trimers such as the SOSIP design), engineered for increased stability and rendered soluble by eliminating the membrane-anchoring domain of gp41, are widely used in immunogen-design programs (for example: Dey et al., 2018; Escolano et al., 2017; Sanders and Moore, 2017; Sok and Burton, 2018). Some earlier constructs such as gp120 monomers and non-native gp140 pseudo-trimers are, however, still being evaluated clinically (Bekker et al., 2020; Excler and Kim, 2019). The points made below apply to most of these different Env-immunogen designs (Figure 1). The various forms of Env, including membrane-anchored SOSIP trimers, generally induce similar binding Ab titers in various species. But native-like SOSIP trimers elicit the strongest autologous neutralization titers when administered as soluble proteins or DNA vectors in rabbits and macaques or on virus-like particles in guinea pigs (Aldon et al., 2018; Crooks et al., 2007; Pauthner et al., 2017; Sanders and Moore, 2017; Sanders et al., 2015).

It is certainly possible to raise high titers of Abs when HIV-1 Env is co-administered with a strong adjuvant. These Abs tend to recognize simple, peptidic epitopes and they lack useful, and often any, neutralizing activity. It is also plausible that inducing such non-neutralizing Abs (non-NAbs) or narrowly acting NAbs distracts the immune system from responding to the more complex, less immunogenic, but crucial epitopes for bNAbs (Havenar-Daughton et al., 2018; Haynes et al., 2012; Klein et al., 2013b; Ringe et al., 2019). A key aspect of bNAb epitopes is that they often comprise both peptidic and glycan components; all the known ones on gp120 include, or even consist entirely of, glycans (Crispin et al., 2018; Ward and Wilson, 2017). Since glycans are largely seen as self by the immune system, their immunogenicity requires the breaking of tolerance (Kelsoe and Haynes, 2018). In short, the high titers of non-NAbs or NAbs that act only against highly sensitive (Tier-1) viruses, which are common endpoints in animal studies and clinical trials, are not useful for assessing Env immunogenicity. Even for non-NAb or ELISA-based antigen-binding Ab endpoints, however, the evidence suggests that HIV-1 Env is a relatively poor immunogen, which does not induce sustained Ab titers (Fouts et al., 2015; Fouts et al., 2003; Klasse et al., 2012; Lewis, 2010; Lewis et al., 2014; Robb et al., 2012). Thus, Ab responses to many vaccines persist in humans for decades, often at protective levels (Amanna and Slifka, 2010). In contrast, anti-Env half-lives are orders of magnitude shorter and are, of course, generally non-protective (Lewis, 2010; Lewis et al., 2014). Why is there such a difference?

Comparative immunogenicity studies

The few immunogenicity studies that directly compare Ab (and sometimes cellular) responses to HIV-1 Env and a different antigen can be highly informative, although there are often interpretative caveats. Do the two immunogens have comparable physical properties? For example, can soluble HIV-1 Env be fairly compared with a Hepatitis B Surface Antigen (HBsAg) multivalent nanoparticle (NP)? Are de novo and possible recall responses being unreasonably treated as equivalent? Are the immunogen doses optimal? Are mice appropriate for HIV-1 Env immunogenicity studies? Nonetheless, in most reports that we have identified, HIV-1 Env is substantially less immunogenic than its comparators, and the anti-Env Ab responses often have atypical properties.

We are aware of only a single comparative human study, a Phase 1 trial with HIV-1 gp120 and the rabies virus G-protein, both immunogens delivered either mucosally or systemically by a canarypox vector-prime followed by soluble protein boosting. The resulting IgG and IgA responses to gp120 were less frequent than to the rabies-G and canarypox-vector proteins in the systemic-delivery groups, although with no significant differences after mucosal immunization (Wright et al., 2004). The paper contained no information on the Ab titers or decay rates to the various antigens, a gap that could perhaps be filled retroactively if samples are still in storage. A much earlier experiment in baboons compared HIV-1 gp120 monomers with the particulate HBsAg vaccine in Alum adjuvant. Initially, HBsAg induced much stronger Ab responses than gp120 and although the titers to the two immunogens did eventually converge, anti-gp120 Abs then declined more rapidly than anti-HBsAg Abs (Anderson et al., 1989). Thus, gp120 may elicit adequate memory B-cell and short-lived plasma-cell responses, but fewer long-lived plasma-cells. Although the particulate presentation of HBsAg complicates interpretation, that study provided an early clue that gp120 might be a poor immunogen compared with those the vaccine field had previously encountered successfully (Anderson et al., 1989). In contrast, when adjuvanted HIV-1 non-native gp140s and influenza HA1 proteins were given to different legs of the same rhesus macaques, ELISA binding Ab titers and decay kinetics were similar. The authors concluded that the transience of both Ab responses resulted from inherent property of soluble protein immunogens (Sundling et al., 2013). The seasonal influenza vaccine, which is based on inactivated virions, also induces short-lived responses in humans with significantly reduced protection 91–180 days after immunization (Young et al., 2018). Nonetheless, a vector-prime, inactivated virion-boost influenza vaccine regimen can elicit autologous ID₉₀ NAb titers > 1 × 10⁴ in mice, and can even broaden the response in animals, including macaques, by partly directing it to cross-reactive epitopes on the HA trimer stem and away from the immunodominant strain-specific epitopes on its head (Wei et al., 2010). These findings suggest that HA immunogenicity is intrinsically stronger and more amenable to improvement than HIV-1 Env. Indirect comparisons support this supposition: HIV-1 Env SOSIP trimer immunogens induce transient autologous Tier-2 NAb ID₅₀ titers in the range 10²–10³ when administered as a bolus in a strong adjuvant (3M052 in polylactic glycolic acid, PLGA, nanoparticles) and somewhat higher, ~ 10³, when given gradually by an osmotic-pump (Pauthner et al., 2017). These peak titers are often below the threshold for protection against autologous SHIV challenge (~ 3 × 10²), and even when initially above that mark, they drop below it within weeks (Pauthner et al., 2019).

Ab responses to Env tend to be qualitatively atypical. When DNA vaccines encoding either HIV-1 gp120 or influenza HA were administered to mice, the Ab responses to the two proteins differed IgG-subclass and cytokine profiles. Thus, the mice responded to HA with Th1-associated high IgG2a/IgG1 ratios but to gp120 with Th2-associated low IgG2a/IgG1 ratios. This finding was attributed to the gp120-specific induction of IL-10 (Daly et al., 2005), although IL-10 has since emerged as distinct from Th2-associated cytokines, being instead both the inducer and product of Type 1 regulatory (Tr1) cells (Martinez-Sanchez et al., 2018). In other mouse experiments, Addavax-adjuvanted pseudo-trimeric HIV-1 gp120 constructs derived from three strains were compared with particulate HBsAg (Yu et al., 2016; Yuan et al., 2018). The major differences between the immunogens were that, compared with HBsAg, gp120 induced weak germinal center reactions, slow Ab recall responses, poor B-cell memory and low endpoint titers for all IgG subclasses, but strong MHC class II expression on B cells and high frequencies of Tfh-helper cells. Programmed death (PD)−1+ T cells were also more frequent in the gp120-immunized mice, which was proposed to contribute to the slow Ab-recall response. The IgG2a/IgG1 ratios for gp120 were low, but significantly higher than for HBsAg (Yu et al., 2016; Yuan et al., 2018). Also in mice, soluble HIV-1 gp140 behaved unusually in that it induced much lower IgG2a/IgG1 ratios than HIV-1 Gag, Influenza-HA and RSV-F proteins (Hess et al., 2019). Furthermore, the IgG subclass-skewing effect of HIV-1 Env was imparted to the anti-Gag IgG response against an Env-Gag fusion protein; the anti-Gag IgG2a/IgG1 ratio then resembled that of the response to Env but not to Gag alone. The unusual IgG isotype profile was attributed to an interaction between HIV-1 Env and mannose C-type lectin receptors (MCLRs) (Hess et al., 2019). Along similar thematic lines, when mice received monomeric HIV-1 gp120 or Gag proteins in the presence or absence of the mannose-binding protein griffithsin, IgG1-dominated anti-gp120 responses were strongly increased by griffithsin but anti-Gag IgG1 responses were not. The formation of gp120-griffithsin complexes probably accounted for an increase in the anti-griffithsin response when gp120, but not Gag, was a co-immunogen (Banerjee et al., 2012).

The above differences in murine IgG subclass profiles reflect Th1 vs. Th2 polarization that may not have any bearing on how the primate immune system responds to HIV-1 Env (Gor et al., 2003; Martinez-Sanchez et al., 2018). They are, however, indicative of unusual, Env-dependent variations in the interplay between innate and adaptive immunity that could be relevant across species. Nonetheless, any problems with HIV-1 Env immunogenicity in primates are not rooted in IgG subclass profiles per se. The primate immune response to Env in both virus-infected and Env-immunized subjects is generally strongly dominated by IgG1, which is not associated with Th1 vs. Th2 polarization in humans (Banerjee et al., 2010; Klasse and Blomberg, 1987; Wahren et al., 1988). Most bNAbs are also of the IgG1 subclass (https://www.hiv.lanl.gov/content/immunology). Indeed, a study of bNAb development using plasma from >4,000 HIV-1-infected patients found that an IgG1-dominated response to HIV-1 antigens and IgG1 Ab binding to the BG505 SOSIP trimer were the two best predictors of HIV-1 neutralization breadth (Kadelka et al., 2018).

Unusually modest benefits to nanoparticle presentation of HIV-1 Env

One method to improve the immunogenicity of vaccine antigens in general is to present them on NPs of, typically, 25–50 nm diameter (Figure 2). Thus, multivalent antigen display may drive stronger, longer-lasting Ab responses through efficient cross-linking of B-cell receptors (BCR) and improved antigen trafficking, endocytic uptake and thereby presentation (Brinkkemper and Sliepen, 2019; Tokatlian et al., 2019). The licensed HPV and HBsAg vaccines involve NPs, as do impressive programs to create influenza and RSV vaccines (Darricarrere et al., 2018; Hsia et al., 2016; Kanekiyo et al., 2019; Marcandalli et al., 2019). Multiple studies in animals show that NP presentation substantially improves the quantity and quality of Ab responses, compared with the delivery of the same antigens as soluble proteins (Brinkkemper and Sliepen, 2019). Thus, NP display of RSV antigens enhanced NAb titers over 10-fold (Marcandalli et al., 2019). Ongoing or new studies may show even greater increases. NP presentation also allows the creation of multivalent, antigenically mosaic immunogens, which can improve NAb breadth by increasing the avidity of interactions specifically with the most cross-reactive BCRs, as shown for influenza HA (Kanekiyo et al., 2019).

There are, then, substantial grounds to pursue the NP approach with HIV-1 Env. The greatest benefits of NP presentation may be for immunogens with such low affinities for BCRs that they are completely unable to activate B-cells when delivered as soluble antigens (i.e., K_D values in the μM range). The avidity benefits of NP presentation could bring the functional K_D value into a range allowing B-cell activation. HIV-1 Env trimers with complete glycan shields might be one example of such an inert immunogen (Wagh et al., 2018). However, only modest, if any, immunogenic improvements occurred when early-generation HIV-1 Env was tested on NPs (Bale et al., 2017; Ingale et al., 2016; Martinez-Murillo et al., 2017; Tokatlian et al., 2018). Design limitations such as NP instability in vivo and poor epitope display on non-native Env may have compromised these studies. It is now evident that presenting more modern native-like trimers on NPs of various designs also confers at most modest (i.e., ~3-fold) increases in NAb titers, and does so only when the immunogenic epitopes are located close to the apex of the trimer (Brinkkemper and Sliepen, 2019; Brouwer et al., 2019; Brouwer and Sanders, 2019; Ringe et al., 2020; Tokatlian et al., 2019). The relevance of epitope location may reflect varied NAb epitope accessibility on the NP surface and occlusion by mannose binding lectins (MBL), as discussed further below (Brouwer et al., 2019; Ringe et al., 2020).

We are unaware of comparative experiments with HIV-1 Env NPs and those based on other viral antigens. Indirect comparisons are inevitably inexact, but the available data do suggest that NP presentation of HIV-1 Env is substantially less beneficial than in other NP-vaccine systems (Brinkkemper and Sliepen, 2019). Two unusual structure-function facets of HIV-1 Env merit discussion: The capacity of the gp120 subunits to bind to the cell surface CD4 and transduce signals that may affect immune cell function; and the extremely high content of mannose-rich glycans on the surface of Env. Neither feature is shared by other vaccine-relevant antigens, but both may influence the induction and maturation of anti-Env Abs in general or indeed of bNAbs.

Do CD4 binding and Env conformational changes influence immunogenicity?

The ability of gp120 to signal by ligating CD4 in vitro has long been observed in multiple cell systems. Generally, when amounts of gp120 that occur in infection are added to cultures of CD4+ cells, the consequences are almost invariably inimical to normal cell function (Klasse and Moore, 2004; Schwartz et al., 1994; Weinhold et al., 1989; Weissman et al., 1997). The purity of the gp120 proteins, particularly commercial products, is sometimes questionable, and any LPS or other contaminants could interfere with the cell function. Nonetheless, gp120 binding to CD4 in vitro can affect the normal properties of various cells important to immune responses in vivo (e.g., CD4+ T-helper cells, dendritic cells, monocytes and macrophages). One example is that HIV-1 gp120 monomers, in vitro, impaired the maturation of human plasmacytoid dendritic cells and hence the activation of co-cultured B cells through cell surface interactions with both CD4 and MCLRs (Chung et al., 2012). In another report, gp120 was found to suppress B-cell proliferation by binding to α4β7-integrin receptors on the B-cell surface and thereby increasing the expression of inhibitory cytokines and receptors (Jelicic et al., 2013).

It is dubious whether such events play much role in the pathogenesis of HIV-1 infection because gp120 concentrations in vivo are both low and vastly exceeded by the quantities of circulating anti-gp120 Abs that block gp120-CD4 binding (Callahan and Norcross, 1989; Klasse and Moore, 2004; Moore et al., 1994). However, as gp120 vaccines are administered to naïve volunteers in amounts up to 500 μg/dose, the local concentrations at or near injection sites (e.g., in draining lymph nodes) will be high, at least in the short term. Suppressive signals might be transduced during the critical early phases of an immune response, and CD4 binding could also reduce the concentration of free Env immunogen available for eliciting immune responses. Similar considerations apply to native-like trimer immunogens, although the latter are usually engineered to bind CD4 with lower affinity than gp120 monomers (Chuang et al., 2017; de Taeye et al., 2015; Joyce et al., 2017; Torrents de la Pena et al., 2017; Zhang et al., 2018).

Are Ab responses to Env compromised by the binding of the immunogen to CD4 on immune system cells in vivo? A definitive answer still remains to be obtained. This topic can only be addressed in species that express CD4 capable of binding HIV-1 Env with high affinity, which do not include rabbits, guinea pigs or mice. A mutation, W427S, in the CD4 binding site (CD4bs) of gp120 was reported to enhance effector memory Tfh-cell responses, reduce non-specific Tfh-cell responses, and increase the gp120-specific Ab titers in wild-type BALB/c mice (Yu et al., 2014). As HIV-1 gp120 does not bind to murine CD4, the explanation for these observations is obscure. Unmodified HIV-1 Env also binds inefficiently to macaque CD4, limiting the power of any conclusions drawn so far from this species. In a macaque study involving non-native HIV-1 gp140 proteins with standard and disabled CD4-binding capacities, there were similar binding-Ab and non-NAb responses to the two immunogens (Sundling et al., 2013). However, the standard gp140 did not incorporate sequence changes that were later shown to be required for high affinity HIV-1 Env binding to macaque CD4 (Li et al., 2016). A more definitive macaque experiment could now be performed to compare the immunogenicity of CD4-optimized and CD4-disabled native-like trimers, or even gp120 monomers with the same sequence changes. Alternatively, SIV Env variants could be used.

In a complementary strategy, a CD4-optimized or -disabled Env mutant could be injected, alone or together with a different protein (e.g., HIV-1 Gag), into macaque lymphoid tissue or delivered more conventionally. The goal would be to assess whether CD4-binding or another property of gp120 interferes with the development or maintenance of the immune response to gp120 or the co-administered immunogen. Does CD4 engagement in lymphoid tissues trigger apoptosis or otherwise disrupt the cytokine milieu within germinal centers? Are memory responses affected? Is there an impact on the rate of decline of post-peak Ab titers? The answers are achievable and could strongly improve immunogen design.

Even in mice, where high affinity CD4 binding by Env does not occur, there is evidence that Env interferes with the immunogenicity of co-immunized antigens. For example, in mice given DNA plasmids expressing HIV-1 Gag protein with or without gp120, Ab and cellular responses to Gag were suppressed by gp120. Curiously, the inhibitory effect of HIV-1 gp120 was not mimicked by the corresponding glycoproteins from SIV or EIAV, or by influenza HA (Toapanta et al., 2007).

CD4-binding will also occlude bNAb epitopes overlapping the CD4bs. Again, the impact can only be determined by comparative studies in species with high CD4-Env affinity. A different but related question is whether reducing CD4-binding or CD4-induced conformational changes qualitatively improves the immunogenicity of Env by preventing responses to previously cryptic and potentially distractive non-NAb epitopes. Multiple SOSIP trimer variants have been designed and tested to explore this hypothesis, but although some reductions in non-NAb responses have been observed, they have not been accompanied by major increases in autologous or heterologous NAb titers (Chuang et al., 2017; de Taeye et al., 2015; Henderson et al., 2020; Joyce et al., 2017; Kulp et al., 2017; Ringe et al., 2017; Torrents de la Pena et al. 2017; Zhang et al., 2018). Additional research in this area is ongoing, including attempts to better suppress the immunogenicity of non-NAb neo-epitopes at the base of soluble trimers (Kulp et al., 2017; Ringe et al., 2020).

One of the mechanisms by which HIV-1 Env reduces the antigenicity of the CD4bs is entropic or conformational masking (Kwong et al., 2002). Whether the conformational flexibility inherent in HIV-1 Env glycoproteins also affects their immunogenicity by, for example, masking potential bNAb epitopes has been an open question for many years. This general topic was indirectly revisited recently by the use of biophysical assays to study conformational transitions within SOSIP trimers in vitro (Lu et al., 2019; Stadtmueller et al., 2018). However, different spectroscopy techniques, DEER (double electron-electron resonance) and smFRET (single-molecule fluorescence resonance energy transfer), yielded discordant outcomes (Lu et al., 2019; Stadtmueller et al., 2018). The interpretation of the data obtained by smFRET, and hence the conclusions drawn in one of the studies, have also been questioned (Lu et al., 2019; Moore, 2019; Pan et al., 2020). Furthermore, it is unknown to what extent conformational transitions that take place over the microsecond to millisecond range affect the induction of Ab responses in vivo in the hours to days after immunization (Moore, 2019). An answer might be obtained if a native-like trimer immunogen could be designed to adopt only one, optimal conformation, as judged by multiple and reliable in vitro assays. No such immunogen has yet emerged.

Glycans mould the response to both B-cell and T-helper cell epitopes

Compared with other virus-derived immunogens, HIV-1 Env has an exceptionally high content of glycans. As noted, there can be over 30 individual glycan moieties on each protomer of a trimeric spike or recombinant mimic. In comparison, an influenza HA protomer typically contains about 3 glycans while RSV F has 5 or 6 (Figure 1). The glycan shield on HIV-1 Env plays a thoroughly studied and well-understood role in resistance to, and escape from, NAbs; equally fundamental is the inclusion of glycan moieties in all known bNAb epitopes on gp120 and some on gp41 (Crispin et al., 2018; Ward and Wilson, 2017; Zhou et al., 2017). In short, effective bNAb responses must accommodate the presence of glycans.

Less well explored is how glycans affect T-helper cell epitopes (THCEs) (Wolfert and Boons, 2013). Extensive SHM, driven by T-cell help, is critical to the emergence of many bNAbs. Much more research in this area seems justified, particularly using modern trimeric Env immunogens, which were not available when most of the experiments reviewed below were performed 10–20 years ago. Moreover, knowledge of Env structure is much more detailed and sophisticated now than then, perhaps allowing the structure and function of glycans and THCEs to be better understood than in most of the studies outlined below.

Although some THCEs are present in HIV-1 Env sequences, their availability for presentation by MHC class II (MHC-II) to T-helper cells may be highly restricted. Thus, glycans close to proteolytic sites can sterically block the access of antigen-processing proteases in the lysosomal compartment and hence prevent the liberation of 10–25–residue, THCE-containing Env peptides that would otherwise be loaded into the MHC-II peptide binding cleft. But not all N-and O-linked glycosides will survive lysosomal degradation, and some resulting glycol-peptides can be presented by MHC-II. Nonetheless, there are so many bulky glycans on HIV-1 Env that too much carbohydrate may survive processing to allow peptide binding into the MHC-II cleft (Purcell et al., 2008; Werdelin et al., 2002).

Early studies of glycans and THCEs generally involved immunizing mice with gp120 monomers or gp140 pseudo-trimers, experimental systems that complicate extrapolations to the use of more modern native-like trimers in primates. It is also fair to assume that the structural constraints on the presentation of THCEs will be stronger for a fully assembled, covalently stabilized trimer than for a monomeric gp120 subunit (Landry, 2000). Reports from 2001–2002 were among the first to analyze how Env glycosylation and gp120 structure might restrict the presentation of THCEs (Sarkar et al., 2002a, b; Surman et al., 2001). The first study was based on peptide assays of T-helper cell responses in mice immunized with DNA expressing a non-native gp140 (Surman et al., 2001). The identified THCE sequences were then plotted on the only available Env structure, that of a gp120 core (Kwong et al., 1998). Three “T-help hotspots” were identified in the C2, V3 and V4/C4 regions of gp120, and another near gp41-residue 650. It was noted that N-glycans flank most clusters of human and murine THCEs (Surman et al., 2001). In a conceptually related approach, macaques were infected with a live, attenuated SIV vaccine (hence at least some of the Env was fully native). The THCEs were identified by screening 20mer peptides and then plotted on the linear sequence of SIV Env, but without placing the information in a tertiary-structural context. Here, the most common epitopes included gp120 residues 341–360 (C3) and 421–470 (V4–V5), in linear regions with no or few glycans. Other sites were flagged in gp41 (Sarkar et al., 2002a, b). The most detailed of several later papers based on mice immunized with gp120 or gp140, delivered as DNA plasmids or soluble proteins, again depicted the THCE data on the gp120-core crystal structure (Brown et al., 2003).

A hypothesis that local polypeptide instability in model antigens would promote THCE processing and presentation was formulated, supported in silico and then corroborated in vivo in gp120-immunized mice and HIV-1-infected humans. The dominant THCEs were located in the gp120 outer domain (Dai et al., 2001; Landry, 2000; Mirano-Bascos et al., 2008). Furthermore, the influence of glycans on THCE presentation is not always negative: mutation of glycan sites in gp160 can suppress the priming of T cells that recognize glycan-proximate epitopes (Sjolander et al., 1996). Indeed, although eliminating either the N448 or the N230 glycan did not affect endocytic uptake or lysosomal trafficking, it interfered with the presentation of N-terminally located THCEs. Mutating N448 reduced proteolytic processing, but a change at N230 acted by an unidentified mechanism (Li et al., 2009).

Disulfide bonds have a complex influence on THCE presentation (Li et al., 2002). There are typically eight intramolecular disulfide bonds in a monomeric HIV-1 gp120 subunit, and another within gp41 (Checkley et al., 2011). Additional inter- and intra-protomeric disulfide bonds are deliberately engineered into recombinant native-like trimers of various designs, to confer additional stability (Chuang et al., 2017; de Taeye et al., 2015; Joyce et al., 2017; Kwon et al., 2015; Torrents de la Pena et al., 2017; Zhang et al., 2018). A correlation between local polypeptide flexibility and THCE localization suggested that eliminating some disulfide bonds from gp120 monomers would promote antigen processing and THCE presentation (Landry, 2000). But when the 298–331, 378–418 and 378–445 gp120 disulfide bonds were individually knocked out, T-cell helper responses in BALB/c mice were actually reduced, although Ab responses were enhanced. Specifically, the 378–418 mutation redirected the Ab specificity towards the CD4bs (Mirano-Bascos et al., 2010). However, in CBA mice, in which MHC-II have well-defined binding properties, disrupting a single gp120 disulfide-bond redirected responses to cryptic CD4+ T-cell epitopes (Li et al., 2014). Thus, manipulating THCE epitopes in an immunogen as complex as a gp120 monomer can have unpredictable and undesired outcomes. By extension, liberating THCE peptides from native-like trimers by proteolytic cleavage, for MHC-II loading, must be even more complicated. Stabilizing these trimers to optimize the presentation of bNAb epitopes and to reduce the exposure of non-NAb sites may compromise the presentation of THCEs, particularly when additional disulfide bonds are introduced. In short, there could be a direct conflict between optimizing the liberation of THCEs and presenting bNAb epitopes on the same engineered trimer. But inducing HIV-1-specific CD4+ T-helper cells carries a serious risk, as their hyper-susceptibility to HIV-1 infection could be disastrous in the vaccine context (Douek et al., 2002). On this argument, any promotion of endogenous THCEs in Env may be best avoided, even to the extent of eliminating the few existing ones by targeted mutagenesis. Research here may be fruitful.

An alternative approach is to provide exogenous THCEs unrelated to HIV-1 sequences as C-terminal extensions to Env flanked by cathepsin cleavage sites, or intrastructurally as free peptides incorporated within NPs (Elsayed et al., 2018; Lake and Mitchison, 1976). Thus, there were some initial increases in Ab responses when the PADRE THCE was genetically fused to gp120 monomers or, more recently, native-like trimers (Grundner et al., 2004; Ringe et al., 2020). THCE peptides can also be attached to, or incorporated within, NPs (Bale et al., 2016; Damm et al., 2019; Elsayed et al., 2018; Hsia et al., 2016). Various other immune-stimulatory molecules such as BAFF, APRIL and CD40L have been genetically fused to Env proteins but without major immunogenicity benefits (Gupta et al., 2015; Liu et al., 2020; Melchers et al., 2012). Vaccines such as tetanus toxoid, against which most vaccinated people have already developed T-helper cell responses, can also be exploited (Damm et al., 2019; Elsayed et al., 2018). These approaches to improving Env immunogenicity all merit further exploration with the current generation of trimers.

The infant macaque model may also lend itself to assessing the various influences on immunogenicity outlined above: bNAbs can develop more rapidly after SIV/SHIV infection of young animals than adults, and with less hypermutation, in a kappa-chain maturation-dependent manner (Simonich et al., 2019; Simonich et al., 2016). Although the frequencies of Tfh cells and germinal-center B cells were greater in SHIV-infected infant macaques than adults, however, autologous Tier 2 NAbs developed with similar kinetics (Nelson et al., 2019). Ab responses in infants are generally weak; although the priming of memory B cells is effective before and at birth, affinity maturation is constricted until after 4–6 months of age (Siegrist and Aspinall, 2009). Hence, the immaturity of the SHM machinery in infants may open other routes to bNAb development that merit more exploration.

Mannose moieties are not just components of shielding glycans and bNAb epitopes

The glycans on HIV-1 Env are unusually rich in terminal mannose residues, even more so for native-like trimers than gp120 monomers (Cao et al., 2018; Crispin et al., 2018). This mannose density arises from steric hindrance, particularly on native-like trimers, to glycan-processing enzymes as Env migrates along the secretory pathway (Checkley et al., 2011). Hence, discrete areas of oligomannose glycans extend over the trimer surface (Crispin et al., 2018). These mannose-rich patches are targets for MBLs and MCLRs, components of the innate immune system contributing to complement activity and the presentation of mannose-rich antigens. It seems plausible that other viral vaccine antigens with far less mannose may not proceed along the routes available to HIV-1 Env.

Although MBLs bind to soluble HIV-1 Env, they do so more avidly when Env is presented on NPs (Ezekowitz et al., 1989; Ringe et al., 2020; Tokatlian et al., 2019). This is relevant in vivo because SOSIP trimer ferritin-NPs are trafficked to follicular dendritic cells and germinal centers in mice via an MBL-dependent antigen trafficking pathway that is much less available to the corresponding soluble trimers (Tokatlian et al., 2019). In vitro, MBL binding to SOSIP trimers on iron oxide-NPs can occlude NAb epitopes that are located away from the trimer apex region (Ringe et al., 2020) (Figure 3). That observation may help explain why presenting Env trimers on NPs is of comparatively limited benefit (Brinkkemper and Sliepen, 2019).

Earlier studies with gp120 also indicated the relevance of interactions between mannose moieties and MCLRs in vitro (Shan et al., 2007) and MBL or MCLR in vivo (Banerjee et al., 2009, 2010, 2012). Thus, depleting mannose moieties on monomeric HIV-1 gp120 skewed the IgG isotype response of immunized mice away from IgG1 and towards IgG2a and IgG3, and increased the overall anti-gp120 titers (Banerjee et al., 2009). As noted above, it has been suggested that MCLRs play a role in the atypical IgG subclass profile induced in mice by HIV-1 gp140s (Hess et al., 2019).

The full consequences of HIV-1 Env interactions with MBL or MCLRs are far from clear, but an atypically high mannose content may be yet another complex influence on the interplay between Env and the humoral immune system that is not shared with most other pathogens. Such complexities may even extend to the adjuvant field. The different antigen-trafficking and -presentation properties of mannose-rich vs. more typical antigens (soluble or NP) could affect the responses to TLR-activators and other adjuvant components. Whether these influences matter in practice can be explored by preparing mannose-depleted Env through enzymatic digestion. Conversely, mannose–enriched glycoproteins can be expressed in mutant producer cells (Crispin et al., 2018). It may also be possible, although challenging, to knock-out MBL binding sites on HIV-1 Env by targeted mutagenesis. MBL-knock-out mice represent another resource, although there are other limitations to this species for the induction of NAbs.

What adjuvants best support the induction of bNAbs?

Adjuvants are crucial for providing supporting stimuli, including T-cell help, to HIV-1 Env. Various adjuvants have been tested with early forms of Env in multiple species, including macaques (Francica et al., 2015). A particularly strong adjuvant effect occurred with cholera toxin, which acts in a Th17-dependent manner, in that it increased responses to gp120 1,000-fold when both molecules were administered to mice as DNA (Bagley et al., 2003). It is of course important to verify that any chosen adjuvant does not compromise the structural integrity of the Env immunogen, particularly for native-like trimers (Ozorowski et al., 2018).

To date, only few adjuvants have been tested for their ability to support the immunogenicity of native-like SOSIP trimers in animals. An ISCOM-based adjuvant, “ISCOMATRIX”, worked well in rabbits and macaques (Pauthner et al., 2017; Sanders et al., 2015). “Adjuplex” and 3M052-PLGA adjuvants are also effective with native-like trimers in macaques, much more so than Alum (Zhou et al., 2017; Arunachalam, 2020). Substantial immunogenicity increases were found when SOSIP trimers were stably attached to Alum particles to allow gradual and sustained release of the antigen in vivo (Moyer et al., 2020). Slow delivery by osmotic pumps was also beneficial, probably by mimicking more conventional delivery at gradually increasing doses (Cirelli et al., 2019; Pauthner et al., 2017; Tam et al., 2016). At least one adjuvant, GLA-LSQ, which was predicted based on its performance with other antigens to be strong, however, fared very poorly in conjunction with soluble and NP-presented SOSIP trimers in rabbits and macaques (Brouwer et al., 2019; Ringe et al., 2020). A human trial of the prototypic native-like trimer, BG505 SOSIP.664, with multiple different adjuvants is now ongoing, in concert with similarly designed experiments in small animals (Clinicaltrials.gov: NCT04177355). Another human trial of the same trimer with the AS01b adjuvant is also underway (NCT03699241), as is one with a similar trimer delivered in Alum (NCT03783130). Animal studies have shown that the kinetics of non-NAb and autologous Tier-2 NAb responses induced by the same SOSIP trimer can differ, the former Abs arising earlier (Klasse et al., 2016; Ringe et al., 2017; Sanders et al., 2015). The above human trials may therefore reveal that eliciting Abs against glycan-affected NAb/bNAb epitopes on native-like trimers requires a different adjuvant from those that work well for the simpler antigens and epitopes present on earlier HIV-1 proteins and other vaccine antigens. Whatever the outcome, the clinical data should be very informative about how to immunize humans (and animals) with native-like trimers. What is the best adjuvant? If autologous Nab responses occur, how do they compare with those associated with protection of macaques immunized with the same trimers and then SHIV-challenged? As noted earlier, for trimer-only immunization the protective ID₅₀ are ~3 × 10² (Pauthner et al., 2019; Arunachalam et al., 2020).

Given the importance of SHM for inducing bNAbs, it will be critical that the adjuvant fully supports this process, including the interactions between dendritic cells and lymphocytes. Dendritic cells are key players in the interface between innate and adaptive immunity, which is central to the actions of adjuvants, not least for boosting Ab responses (Boscardin et al., 2006; Pulendran and Ahmed, 2006). Dermally derived CD14+ dendritic cells can be triggered by TLR ligands to express B cell stimulatory cytokines that, in turn, stimulate naïve B cells. Thus, in vitro, activation of these dendritic cells by a combination of ligands for TLR-3, -4 and -7/8 induced naïve B cells to proliferate and differentiate into CD27+ CD38+ B cells that secrete high levels of IgG and IgA (Matthews et al., 2012; Matthews et al., 2013).

In both mice and humans, a CpG-based, TLR-9-activating adjuvant supported the production of high levels of IgG Abs to a malaria antigen, but a more sophisticated analysis then revealed that SHM of these Abs was unexpectedly weak (Akkaya et al., 2018). All Ab endpoints are not equally useful when assessing immunogenicity. The reported antagonizing of Ab affinity maturation by TLR-9 activation, as in in the above malaria-vaccine study, could be a substantial hindrance to eliciting bNAbs by HIV-1 Env immunogens. Another concern about TLR-9 as an adjuvant target for Env vaccines arises from in vitro studies in which gp120 inhibited the CpG-induced maturation of plasmacytoid dendritic cells in a CD4-and MCLR-dependent manner, and hence their ability to stimulate B cells (Chung et al., 2012). Similarly, inhibitory effects of gp120 were seen in a co-cultures of CD14+ dermal dendritic cells and B cells when TLR-9 was activated by CpG, but not when TLR-3, -4 or -7/8 were activated (Matthews et al., 2012; Matthews et al., 2013).

Conclusion

Despite over 30 years of study in multiple species, including many thousands of immunized humans, several aspects of how Ab responses to HIV-1 Env are generated remain unknown. Comparative immunogenicity experiments can be particularly informative about the exceptional properties of HIV-1 Env, but they are rarely performed and have not always been thoroughly analyzed. Those lacunae are understandable for human trials, given their complexities and expense, but here there is a role for the macaque models that may be predictive for humans.

Env immunogen design has improved over the past decade, largely through high-resolution structures of native-like trimers. These methods will remain necessary for further advances in HIV-1 vaccine development, but they may not be sufficient. It is entirely possible that no single, fixed-composition Env, whether a native-like trimer or not, will ever be able to mimic the complicated bNAb-evolution processes in infected people. The sophisticated techniques now available to study bNAb ontogeny reveal how difficult it is for the humoral immune response to overcome the multiple defenses that have evolved to hinder the antigenicity and immunogenicity of the native trimer (Klein et al., 2013b; Kwong and Mascola 2018; West et al., 2014). Immunological approaches such as, but not limited to, germline-targeting may breach some of these barriers (McGuire, 2019; Medina-Ramirez et al., 2017; Escolano et al, 2017). More attention could also be paid to human genetics. What factors underlie the well-documented variation of at least 100-fold in the magnitude of the Ab responses to Env variants within vaccine cohorts, whether animal or human (Corey et al., 2015; Gilbert et al., 2005)? In appropriately designed comparative immunogenicity studies (see above), would poor responders to HIV-1 Env immunogens also respond weakly to other viral protein immunogens, and vice versa? Can advanced genetic approaches such as systems vaccinology provide answers (Cortese et al., 2020; Pulendran, 2019; Querec et al., 2009)?

In conclusion, we suggest that all Env glycoprotein-based strategies would benefit from an improved knowledge of how these elaborate, and highly unusual, molecular constructs interact with the innate and adaptive immune systems. The immunology of Env vaccine development may turn out to be as critical as structure-guided design to the eventual success of the overall concept.