The gp120 protein core in its CD4-bound state contains an inner domain, an outer domain (OD), and a four-stranded bridging sheet minidomain, the latter of which is composed of two β-hairpins which extend from the inner domain (β2–β3) and OD (β20–β21), respectively [15,16] (figure 1b). The OD is extensively glycosylated, and previous studies involving antigenic analysis and fitting into the viral spike reveal that the glycan surface covers most of the exposed surface of the spike, making it immunologically silent [17]. Thus, while it would seem straightforward to induce antibodies to the spike proteins that would be able to neutralize the virus, HIV-1 has developed multiple mechanisms of immune evasion involving the molecular structure of the virus, particularly the viral spike, including glycan shielding, variable loop divergence and extensive conformational flexibility, all of which succeed in preventing either the elicitation or the binding of most antibodies.

Thus, structural biology has provided important information about the three-dimensional organization and chemical structure of the HIV-1 glycoproteins. This information and in particular an understanding of atomic-level structure can be used rationally to design proteins that have specific biological properties and functions that would be important in vaccine design, such as incorporation of conserved sequences, which are generally associated with functions that are essential to the virus, and properties of virus proteins, such as the ability to stimulate specific protective immune responses.

This concept is being applied to HIV-1 to provide a structural definition of the functional viral spike (figure 1a), which is used by the virus to enter host cells and is the target of all known virus-directed neutralizing antibodies. The ability to conduct atomic-level analysis of the spike facilitates immunogen designs that stabilize and help to present potential sites of neutralization more optimally to the immune system. Unfortunately, the same virus strategies noted above that allow the virus spike to evade an effective immune response also hinder structural analysis, and the entire HIV-1 spike has resisted and continues to resist atomic-level characterization.

Nevertheless, structural analysis and molecular modelling have facilitated the understanding of the antibody response against HIV. This knowledge includes (i) the capability to dissect the types of antibodies in sera and to ascertain what regions of the HIV Env are targeted [20], (ii) the ability to isolate neutralizing antibodies from individual B-cells [21,22], and (iii) the capacity to determine the atomic-level structure of neutralizing antibodies bound to the HIV Env [23–25]. We have used this knowledge of the structure of the HIV Env to design protein probes that expose various regions of the HIV Env in specific conformations (figure 2). The concept underlying these probes is that one can use molecular biological tools and structural and functional knowledge to design and generate a molecule that preserves the antigenic structure of the viral surface that one wishes to study, but does not contain irrelevant antigenic regions of HIV-1. These kinds of probes can then be used to evaluate the regions of the HIV Env that are targeted by serum neutralizing antibodies. A key virus functional region, the CD4-binding site of gp120, has been studied in this way. Since CD4 is the primary cellular receptor for HIV, antibodies that bind to the CD4-binding site can block HIV infection of CD4⁺ T cells and thus function as neutralizing antibodies. To define the neutralizing antibodies to the CD4-binding site further, a specific protein probe was designed such that it exposed the CD4-binding site of gp120, while other regions of HIV were altered to be unrecognizable to HIV antibodies, e.g. by substitution with simian immunodeficiency virus (SIV) homologues or other non-HIV residues (figure 2). This epitope-specific probe, along with a knockout mutant was used to identify B-cells making antibodies to the CD4-binding site [25,26]. Following the isolation of these B-cells by flow cytometry, one can then use PCR to amplify the genes encoding the antibody heavy and light chain variable regions (VH and VL) and subsequently express the full IgG monoclonal antibody in tissue culture (figure 3). Having these new informatics-designed monoclonal antibodies in hand, it is possible to verify and to study in detail their ability to neutralize HIV-l. Using this methodological approach, we recently isolated three CD4-binding site neutralizing monoclonal antibodies designated VRC01, VRC02 and VRC03 [26]. Of all antibodies isolated with this method, the percentage of monoclonal antibodies that neutralize HIV-1 ranges from 25 to 75 per cent, while previous attempts to isolate such antibodies with gp140 probes were not successful [21]. The crystal structure of VRC01 bound to HIV gp120 has provided an atomic-level footprint showing the precise region of HIV gp120 that is vulnerable to neutralizing antibodies [25]. This structural information can be further used to make new vaccine immunogens that are designed to induce the immune system to generate antibodies similar to VRC01. It is important to recognize that the same probe used to isolate monoclonal antibodies will not necessarily serve as a vaccine to elicit them. For example, the RSC3 probe does not elicit CD4-binding site neutralizing antibodies and only serves as a starting point to design such immunogens. This observation underscores the important difference between antigenicity and immunogenicity.

Another approach to the identification of relevant immunogens seeks to bypass the difficulties of using the entire viral spike and rather focuses on the functionally critical sites that the virus uses for entry. By definition, these sites cannot be altered without hindering spike function. We and others have used this approach to elicit antibodies against the highly conserved site of co-receptor binding [27].¹ Unfortunately, while it is possible to prepare immunogens that will elicit antibodies to this site, because the site is only revealed after the engagement of the viral and target cell membranes, it is effectively inaccessible to protective antibodies that might be elicited by a vaccine [28]. Thus, in choosing potential vaccine immunogens, one must verify that in addition to functional importance and sequence conservation, the site is vulnerable to the neutralizing antibody.

Our investigations have shown that HIV-1 recognition by VRC01 primarily involves regions of the antibody derived from VH and variable K light chain precursors, and that its generation does not appear to be dependent on unusual recombination events. Additionally, VRC01 and its genomic precursors do not appear to be autoreactive. An important characteristic of VRC01 is that the level of affinity maturation is approximately10-fold higher than that of most antibodies in the repertoire. Another important factor is that the putative genomic precursors appear to have low (mM or weaker) affinity for gp120. Thus, a key barrier to eliciting VRC01-like antibodies appears to be the inability of probable genomic precursors of these antibodies to bind the gp120 immunogen. One potential approach to bypassing this barrier might involve creating altered gp120s that are better able to bind to unmutated precursors of VRC01 and thus better drive antibody maturation.

Moreover, the understanding of the interactions of broadly neutralizing antibodies, particularly studies of the b12² and VRC01 antibodies directed to the CD4-binding site of HIV Env, has provided a conceptual basis for the development of four strategies to elicit antibodies with similar specificities. These strategies use biological and structural information to design immunogens that might enhance the elicitation of effective neutralizing antibodies or that might be used as probes for the analysis of complex mixtures of antisera.

The first strategy is the generation of trimeric forms of the HIV-1 Env by including the gp41 trimerization sites in the absence of the transmembrane domain. Trimerization sequences from heterologous proteins, such as the fibritin protein from phage lambda, can be used to further stabilize this form of the protein. Thus, employing this approach, or by using the natural gp160 trimerization motifs, one can generate stable trimers using site-specific mutations to fix the core structure. The variable domains of these proteins can be deleted because they might otherwise divert immune responses to strain-specific determinants.

In a second strategy, stabilized core Env proteins are further modified using structure-based design [26]. The underlying concept is to generate a molecule that preserves the key surface antigenic structures of HIV-1 involved in neutralization, but that eliminates other antigenic regions. In practice, one uses bioinformatics and computer-assisted design to introduce mutations that eliminate HIV residues on the surface of gp120 and replace these residues with those of a virus that shows minimal serologic cross-reactivity with HIV-1, e.g. SIV Env. By a progressive process of modifying the surface of the constrained Env core protein and subsequently covering the modified region with glycans, one can generate a set of molecules that can function both as probes to analyse complex antisera for the presence of broadly neutralizing antibodies, as well as prototype immunogens to elicit antibodies directed to the highly conserved CD4-binding site (figure 2).

A third approach towards eliciting antibodies with desired specificities is to eliminate immunologic determinants that are not relevant to the desired immune response. As an example of this, we have generated a subdomain of the HIV-1 Env which includes the OD that contains the initial CD4 binding loop, but eliminates most other immunologically irrelevant protein sequences. However, a previous study showed that a soluble form of the OD that contains the β15 loop was not able to bind to the b12 antibody with high affinity. One approach to overcoming this problem involves the inclusion of a transmembrane domain [31]. Others use further site-directed mutagenesis based on the b12/Env and the VRC01/Env structures. These types of methods represent ways to stabilize this interaction and also provide additional hydrophilic surfaces that may improve folding or stabilize additional contacts of the VRC01 antibody. Recently, additional mutations in the OD have been generated that preserve high-affinity binding by decreasing the off-rate in binding as demonstrated by surface plasmon resonance spectroscopy studies. The vectors developed by this approach are currently being evaluated for their ability to elicit broadly neutralizing antibodies and for their usefulness in characterizing these complex antisera.

In summary, the use of standard vaccine design approaches for HIV-1 has failed, primarily because the virus uses a range of biological and molecular tricks to evade the development of protective immune responses. We have used knowledge of atomic-level structure and computer-assisted protein design to increase our understanding of both the viral spike and the human immune response. This approach seeks to provide a knowledge base from which relevant hypotheses can be formed and tested. An increasing number of broadly neutralizing antibodies have been identified recently that target the CD4-binding site [32] or other structures in the V1 and V2 or OD regions [22,33]. Structural information about some of these antibodies has been generated [34,35], and further understanding of the molecular basis for their interaction with Env will facilitate HIV vaccine design. Taking advantage of the ability of structural biology to provide detailed atomic-level information required for precise manipulation of molecular structures, this effort has focused on (i) maximizing the application of structural methods of definition (e.g. of the functional viral spike), (ii) using structural techniques to interrogate the natural response to HIV (e.g. the use of epitope-specific probes to identify specific monoclonal antibodies), and (iii) incorporating structural feedback (e.g. of the immunogen and for the elicited response).