Minimally Mutated HIV-1 Broadly Neutralizing Antibodies to Guide Reductionist Vaccine Design

Abstract

An optimal HIV vaccine should induce broadly neutralizing antibodies (bnAbs) that neutralize diverse viral strains and subtypes. However, potent bnAbs develop in only a small fraction of HIV-infected individuals, all contain rare features such as extensive mutation, insertions, deletions, and/or long complementarity-determining regions, and some are polyreactive, casting doubt on whether bnAbs to HIV can be reliably induced by vaccination. We engineered two potent VRC01-class bnAbs that minimized rare features. According to a quantitative features frequency analysis, the set of features for one of these minimally mutated bnAbs compared favorably with all 68 HIV bnAbs analyzed and was similar to antibodies elicited by common vaccines. This same minimally mutated bnAb lacked polyreactivity in four different assays. We then divided the minimal mutations into spatial clusters and dissected the epitope components interacting with those clusters, by mutational and crystallographic analyses coupled with neutralization assays. Finally, by synthesizing available data, we developed a working-concept boosting strategy to select the mutation clusters in a logical order following a germline-targeting prime. We have thus developed potent HIV bnAbs that may be more tractable vaccine goals compared to existing bnAbs, and we have proposed a strategy to elicit them. This reductionist approach to vaccine design, guided by antibody and antigen structure, could be applied to design candidate vaccines for other HIV bnAbs or protective Abs against other pathogens.

Fig 1. Antibody features frequency analysis.

(A) Log₁₀ antibody features frequencies plotted for HIV bnAbs of different classes (left y-axis), the distribution of Log₁₀ antibody features frequencies plotted for a set of 388 "normal" human memory (and plasmablast) antibodies isolated by B cell sorting from human memory B cells [56](and this study), influenza infection [55], HPV vaccination [54](and this study), anthrax vaccination (this study), tetanus toxoid vaccination [52], and HIV RV144 glycoprotein vaccination [51, 53] (gray histogram, right y-axis), and the distribution of Log₁₀ antibody features frequencies plotted for a set of 300,000 antibody sequences generated by Monte Carlo ("mc") via the AFF method (black line histogram, right y-axis). Potent HIV bnAbs (mean or median IC₅₀ < 0.5 μg/mL [27]) are shown with solid blue symbols, while less potent HIV bnAbs (mean or median 0.5 ≤ IC₅₀ < 5.0 μg/mL) are shown with open blue symbols. HIV bnAbs previously engineered with reduced mutations are indicated (VRC01-5fH6fL [50], the 10E8 variant 2fH10fL [50], and the PGT124 variant 32H3L [40]). The shape of the distributions for "normal" and "mc" memory antibodies reflects the smearing of the germline distributions shown in (B) to lower features frequencies due to the effects of mutations, insertions and deletions. The slightly increased smearing of the "normal" compared to the "mc" memory distribution stems from the slightly higher mutation frequencies in the "normal" Abs, which are likely due to the fact that all the "normal" Abs except those from Tiller et al. [56] were affinity-selected either by antigen-specific B cell sorting [–54] or by direct affinity measurements on recombinant antibodies after cloning from plasmablast B cells [55]); hence, the "mc" features frequency distribution is probably a better representation of the human memory repertoire. (B) Antibody features frequencies for germline versions of the antibodies in (A). The shape of the germline distribution curve (for GL-Normal or GL-mc) reflects both the great diversity of the human antibody repertoire and combinatorial statistics. The minimum in the distribution at high features frequency (log(f) = -7) is due to germline antibodies composed of the most common V_HD_HJ_H, V_HV_L, and V_LJ_L combinations and having the most common CDR-H3 and CDR-L3 lengths; such Abs have the highest features frequencies (of ~10⁻⁷), but there are relatively few such combinations, so they are created infrequently. The peak of the germline distribution at features frequency of ~10⁻¹⁰ is due to antibodies that utilize somewhat less frequent but not rare individual components; as there are a very large number of such combinations, these are created frequently. The tail in the distribution at low features frequency is due to germline antibodies composed of the least common V_HD_HJ_H, V_HV_L, and V_LJ_L combinations and the use of rare H-CDR3 and/or L-CDR3 loop lengths. Potent HIV bnAbs with the lowest germline features frequencies either had long H-CDR3 loops (V2/Apex and PGT151) or short L-CDR3 loops combined with less frequent V_L or J_H chains (VRC13 and some members of the VRC01-class), and all but two HIV bnAbs (CAP256-VRC26.08, with a rare H-CDR3 length of 39, and VRC13) had germline features frequencies greater than 10⁻¹⁴.

Fig 2. Limited somatic hypermutation of VRC01-like antibodies is sufficient to afford broad and potent neutralization.

(A) The crystal structure of VRC01 (heavy chain in black, light chain in gray) in complex with core gp120 (PDB ID 3NGB, green surface with CD4bs highlighted in yellow) with mutations from the inferred germline variable genes highlighted as pink spheres. (B) The same structure as in (A) showing only the mutations retained in MinVRC01 following directed evolution. Mutations are grouped together into patches distinguished by color. (C) Alignment of VRC01 and MinVRC01, 12A21, and Min12A21 with their germline variable genes, highlighting mutations from germline for MinVRC01. Mutations are colored according to scheme in (B). (D) Neutralization of a 80-virus cross-clade panel by VRC01, MinVRC01, 12A21, and Min12A21. Values for median neutralization IC₅₀ (μg/mL) and percent breadth are displayed by clade and colored according to the legend. (E) Summary of neutralization breadth and potency of VRC01, MinVRC01, 12A21 and Min12A21.

Fig 3. Polyspecificity analysis of MinVRC01 and Min12A21.

(A) HEp-2 cell staining assay, for which 4E10 is a positive control while Humira, 10E8, and undiluted human serum "Negative control" are negative controls. (B) Polyspecificity reagent (PSR) binding assay measuring binding to preparations of solubilized membrane proteins and cytosolic proteins from CHO cells. Here, 4E10 and 2F5 are positive controls, while Humira, 10E8, PG9 and 40 HPV-vaccine induced human mAbs are negative controls. Error bars reflect mean ± standard deviation with N = 40.

Fig 4. Reversions of MinVRC01 patches to germline reveal importance of affinity maturation in the light chain to accommodate the N276 glycan on Env.

(A) Neutralization on a 16-pseudovirus cross-clade panel was performed for each reverted antibody variant. Neutralization values are presented in μg/mL and colored by most potent (red) to least potent (green). (B) Neutralization of MinVRC01 light chain reversions against a 6-pseudovirus panel of wild-type (WT) virus, viruses produced with kifunensine, which yields virions with Man₉GlcNAc₂ glycans on Env, viruses produced in GnTI^-/- cells, which yields virions with Man_5-9GlcNAc₂ glycans on Env, and viruses with the glycan site at N276 on Env removed by alanine substitution (N276A). (C) VRC01-class bnAbs were produced as mature heavy chain and germline light chain chimeras and tested on a 13-pseudovirus panel with (wild-type, WT) and without (N276A) the glycan site at position 276 of Env. Values are percent breadth and median neutralization IC₅₀ (μg/mL); all measurements were performed in duplicate. (D) Neutralization of N276A viruses by variants of GL-VRC01 and GL-VRC01-HC/MinVRC01-LC in which either the H-CDR1, H-CDR2 or H-FW3 mutation patch was restored. All glycan-modified viruses were tested against non-neutralizing and weakly neutralizing antibodies b6, b12 and F105, as well as CD4-IgG2 (S6 Table). These agents showed no neutralization or no increase of neutralization over WT virus, confirming that the N276 interaction is specific to the VRC01 light chain and does not cause a global increase in neutralization sensitivity.

Fig 5. Structural definition of the full VRC01-class bnAb epitope in the context of the Env trimer.

(A) Crystal structure of eOD-N276Kif containing a Man₉GlcNAc₂ glycan at N276 in complex with VRC01, with the critical MinVRC01 mutations from germline shown as spheres and colored as in Fig 2B. A 2mFo-DFc simulated annealing composite omit map displayed at a contour level of 1.0 σ for the N276 glycan is shown in a blue mesh. (B) Close-up view of the interactions of the N276 glycan with the light chain. Mutations identified in MinVRC01 are highlighted as spheres and the side chains are shown as sticks. (C) Crystal structure of NIH45-46 scFv in complex with the BG505 SOSIP Env trimer (gp120, green surface; gp41, yellow surface) and PGT122 Fab (omitted for clarity). A blue mesh 2mFo-DFc composite omit map is displayed at a contour level of 1.2 σ for glycans that are part of or surround the CD4bs epitope. Numbers in parentheses below each glycan name indicate the fold decrease in VRC01 IC₅₀ (i.e. increase in potency) upon substituting that glycan with an alanine. (D) Close-up view of the interaction between the NIH45-46 H-CDR2 and H-FW3 patches with the gp120 bridging sheet (green) in a pre-fusion, pre-CD4 bound conformation and with the N197 glycan. (E) Buried surface area analysis of VRC01-class bnAbs on eOD, gp120 and partially deglycosylated BG505 SOSIP reveal that the majority of the epitope components are contained on eOD, but important components are also present on the inner domain and bridging sheet of gp120 on the same protomer (gp120) as well as loops and a glycan from the adjacent protomer (quaternary). P2 refers to the adjacent protomer. The trimer restricts the antibody angle of approach to the CD4bs due to quaternary packing constraints and glycan fencing.

Fig 6. A proposed reductionist vaccine strategy to induce VRC01-class bnAbs.

This working-concept strategy has four main objectives for achieving affinity maturation by using a sequence of at least four types of immunogens. Incremental, step-wise progress can be assessed by using neutralization assays against panels of mutant viruses, as well as analysis of antibody sequences from antigen-specific B cells. Thus, using this strategy, each step can be optimized individually. One strategy is indicated; variant strategies, for example incorporating cocktails of Env, can be readily envisaged. "SOSIP" is one type of native-like trimer but other types of native-like trimers could be used.