Research Article

A generalized HIV vaccine design strategy for priming of broadly neutralizing antibody responses

See allHide authors and affiliations

Science  06 Dec 2019:
Vol. 366, Issue 6470, eaax4380
DOI: 10.1126/science.aax4380

Engineering better bnAbs

A highly effective HIV vaccine has been the goal of vaccinologists for nearly 35 years. A successful vaccine would need to induce broadly neutralizing antibodies (bnAbs) that are capable of neutralizing multiple HIV strains (see the Perspective by Agazio and Torres). Steichen et al. report a strategy in which the first vaccine shot can lead to immune responses that generate desired bnAbs. By combining knowledge of human antibody repertoires and structure to guide design, they validated candidate immunogens through functional preclinical testing. Saunders et al. designed immunogens with differences in binding strength for bnAb precursors, which enabled selection of rare mutations after immunization. The immunogens promoted bnAb precursor maturation in humanized mice and macaques.

Science, this issue p. eaax4380, p. eaay7199; see also p. 1197

Structured Abstract

INTRODUCTION

HIV newly infects 1.8 million people each year, making development of an HIV vaccine a global health priority. Nearly all licensed vaccines protect by inducing antibodies, but highly variable pathogens such as HIV and influenza virus have eluded traditional vaccine strategies. The discoveries of broadly neutralizing antibodies (bnAbs) that bind to conserved epitopes on the surface proteins of these viruses have inspired vaccine design strategies to induce bnAbs. Antibodies are produced by B cells, and highly effective antibodies like bnAbs acquire affinity-enhancing mutations when a bnAb-precursor B cell mutates and matures from the original naïve B cell (or germline) state. Among several new vaccine strategies, germline-targeting vaccine design aims to induce bnAbs by first stimulating bnAb-precursor B cells and then shepherding B cell affinity maturation with a series of rationally designed boosting immunogens. A key rationale for this strategy is that germline-reverted forms of bnAbs—precursors with all recognizable amino acid mutations reverted to germline—typically have no detectable affinity for HIV envelope (Env). Thus, for a vaccine to initiate bnAb induction, a germline-targeting priming immunogen with appreciable affinity for bnAb precursors must be engineered.

RATIONALE

Most HIV bnAbs (and most antibodies to any pathogen) bind to their target by using their heavy chain complementarity-determining region 3 (HCDR3) as a major binding determinant. Hence, an optimal HIV vaccine that induces multiple bnAbs, and a general solution to germline-targeting vaccine design that could be applied broadly to other pathogens, will need to work with HCDR3-dependent antibodies. However, the need to design germline-targeting immunogens to initiate HCDR3-dependent bnAb responses faces major technical challenges. Although each B cell expresses a single unique antibody, different B cells produce diverse antibodies encoded by different combinations of antibody genes, with the greatest antibody genetic diversity encoded in the HCDR3 portion of the molecule. The exceptional diversity in the human B cell repertoire makes any single HCDR3 sequence an impractical vaccine target. Rather, a pool of precursors sharing a set of bnAb-associated genetic features must be identified and targeted. Thus, owing to the enormous diversity of human antibodies, a germline-targeting immunogen should have affinity for diverse bnAb precursors in order to succeed in diverse vaccine recipients.

RESULTS

Herein we report a solution to the above challenges. Using the strongly HCDR3-dependent bnAb BG18 that binds a conserved site on HIV Env as a high-value target and a proof of principle, we demonstrate a method to identify pools of potential bnAb precursors in an ultradeep human antibody sequence database, guided by key genetic features that enable bnAb structural recognition of the antigen. We then use a representative set of those potential bnAb precursors as design targets to guide our engineering of HIV Env immunogens that bind to diverse potential bnAb precursors. Lastly, we provide critical preclinical validation of immunogen design by assessing these immunogens for (i) their ability to select rare potential bnAb-precursor naïve B cells from the blood of healthy human donors, (ii) their modes of binding to bnAb precursors, and (iii) their capacity to prime rare bnAb-precursor B cells with physiologically relevant affinities in a mouse model.

CONCLUSION

Overall, we demonstrate a new approach to defining diverse precursors for a target antibody and designing vaccine immunogens that take advantage of that information. The approach lays out a generalizable pathway for the development and preclinical validation of germline-targeting immunogens to stimulate precursors for HCDR3-dependent antibodies.

General strategy for germline-targeting vaccine design.

Four key steps are defined here for design and validation of germline-targeting immunogens: structural and genetic understanding of target antibody-antigen interaction, identification of diverse antibody precursors, design of an immunogen to bind diverse precursors, and preclinical immunogen validation by isolating human B cell binders, assessing structural interaction with precursors, and stimulating responses in transgenic mice.

Abstract

Vaccine induction of broadly neutralizing antibodies (bnAbs) to HIV remains a major challenge. Germline-targeting immunogens hold promise for initiating the induction of certain bnAb classes; yet for most bnAbs, a strong dependence on antibody heavy chain complementarity-determining region 3 (HCDR3) is a major barrier. Exploiting ultradeep human antibody sequencing data, we identified a diverse set of potential antibody precursors for a bnAb with dominant HCDR3 contacts. We then developed HIV envelope trimer–based immunogens that primed responses from rare bnAb-precursor B cells in a mouse model and bound a range of potential bnAb-precursor human naïve B cells in ex vivo screens. Our repertoire-guided germline-targeting approach provides a framework for priming the induction of many HIV bnAbs and could be applied to most HCDR3-dominant antibodies from other pathogens.

HIV infects 1.8 million new people each year, making development of an HIV vaccine a global health priority (1). Nearly all licensed vaccines protect by inducing antibodies, but highly antigenically variable pathogens such as HIV and influenza virus have eluded traditional vaccine strategies (2, 3). The discoveries of broadly neutralizing antibodies (bnAbs) that bind to relatively conserved epitopes on viral surface proteins have inspired new vaccine design strategies (4, 5).

Antibodies are produced by B cells and acquire affinity-enhancing mutations when the B cell mutates and matures from the original naïve (or germline) state. Germline-targeting HIV vaccine design aims to induce bnAbs by first priming bnAb-precursor B cells and then shepherding B cell affinity maturation with a series of rationally designed boosting immunogens. A key rationale for this strategy is that germline-reverted forms of bnAbs—precursors with all recognizable amino acid mutations reverted to germline—typically have no detectable affinity for HIV envelope (Env) proteins. Thus, for a vaccine to initiate bnAb induction, a germline-targeting priming immunogen with appreciable affinity for bnAb precursors must be engineered. Most HIV bnAbs (and most antibodies to any pathogen) bind to their target by using their heavy chain complementarity-determining region 3 (HCDR3) as a major binding determinant. Hence, an optimal HIV vaccine that induces multiple bnAbs to different HIV Env sites, and a general solution to germline-targeting vaccine design that could be applied broadly to other pathogens, will need to work with HCDR3-dependent antibodies. Many advances have been made in developing germline-targeting immunogens to prime precursors for one particular class of bnAbs (i.e., VRC01-class bnAbs) (615), and at least one such immunogen has entered human clinical testing (16). However, VRC01-class bnAbs represent a specialized case in which non-HCDR3 features are the main determinants of antibody specificity and affinity (615).

The need to design germline-targeting immunogens to initiate HCDR3-dependent bnAb responses brings new challenges. Although each B cell expresses a single unique antibody, different B cells produce diverse antibodies encoded by different combinations of antibody genes, with additional variation at junctions between genes, and the greatest antibody diversity is encoded in the HCDR3 portion of the molecule. The exceptional diversity in the human B cell repertoire makes any single bnAb-precursor HCDR3 sequence an impractical vaccine target. Rather, a pool of precursors sharing a set of bnAb-associated genetic features must be identified and targeted. Thus, owing to the antibody diversity in humans, a germline-targeting immunogen should have affinity for diverse bnAb precursors in order to succeed in diverse vaccine recipients.

Strategy for immunogen design and testing

We report a potential solution to the above challenges. We selected the bnAb BG18 (17, 18) as a test case for a high-value vaccine design target, because BG18 is the most potent bnAb directed to the Asn332 (N332) supersite, one of the major bnAb sites on HIV Env, and BG18 lacks insertions or deletions (indels) and therefore may be easier to induce than other bnAbs that require indels (see the supplementary materials) (19). Using the strongly HCDR3-dependent bnAb BG18 (17, 18), we demonstrate a method to identify pools of bnAb potential precursors and use them as design targets to engineer HIV Env trimer immunogens that bind diverse bnAb potential precursors. We then provide preclinical validation by assessing these immunogens for (i) their ability to select rare bnAb potential precursor naïve B cells from the blood of HIV-seronegative human donors, (ii) their modes of binding to bnAb precursors, and (iii) their capacity to prime rare bnAb naïve precursors with human physiological affinities in a mouse model (fig. S1).

Precursor frequency analysis

Crystal structures of BG18 bound to HIV Env trimers indicated a BG18 binding mode in which the HCDR3 engages the conserved Gly-Asp-Ile-Arg (GDIR) motif at the base of the V3 loop like the bnAb PGT121 while the HCDR1 contacts the relatively conserved N332 glycan and the light chain (LC) straddles the V1 loop of gp120, unlike PGT121 (18). This binding mode was corroborated by (i) structural modeling (fig. S2, A to D); (ii) a 4.4-Å resolution cryo–electron microscopy (cryo-EM) structure of BG18 bound to an HIV Env trimer (Fig. 1A, fig. S3, and table S2); (iii) mutagenesis studies (fig. S2, E to F); and (iv) structural model-guided design of a minimally mutated BG18 bnAb (minBG18) that retained ~67% of the neutralization breadth of BG18 with only 11% amino acid mutations in the variable (V) gene regions of immunoglobulin heavy and light chains (VH and VL) compared with ~30% for BG18 (fig. S4). The successful design of minBG18 provided an additional rationale for BG18 vaccine targeting, namely that the high mutation level in BG18 itself is not required to achieve substantial neutralization breadth and potency in a BG18-like response. The elucidation of the BG18 binding mode by these studies enabled structure-guided immunogen design.

Fig. 1 Engineering germline-targeting trimers for an HCDR3-dependent bnAb.

(A) Cryo-EM structure of BG18 (HC, purple; LC, cyan) bound to the BG505 MD39 Env trimer (gray, with N332 and N392 glycans shown as green sticks), and conserved residues near the base of V3 (Gly324, Asp325, Ile326, Arg327, Gln328, Ala329, His330, Thr415, Leu416, and Pro417 colored red). (B) Cryo-EM structure of BG18 iGL0 in complex with the N332-GT2 Env trimer with MD64-stabilizing mutations (23). The coloring is the same as in (A). (C) Schematic of the directed evolution process to design N332-GT1, -GT2, and -GT5. (D) N332-GT binding affinities (Kd) for BG18 iGL0 to BG18 iGL2 (red), BG18 iGL1 with alternate germline VL (blue open symbols) or VH genes (blue filled symbols), and BG18 iGL containing NGS-derived HCDR3s (pre1 to pre15) (black). MD39 is the reference Env trimer with no germline-targeting mutations. Pre8 was found to be highly polyreactive and was not included in the analysis. Solid black, blue, and red lines indicate the geomean Kds for NGS-derived precursors, alternate VH and VL precursors, and inferred germline precursors, respectively. The dashed line indicates the limit of detection.

To assess the extent to which BG18-like precursor HCDR3s are present in the general population, we used a bioinformatics approach to search a custom next-generation sequencing (NGS) dataset of 1.1 × 109 sequences of human B cell receptor (BCR) heavy chains (HCs) from 14 healthy, HIV seronegative donors [8.58 × 108 sequences from four donors were obtained in this work (20) and were combined with 2.55 × 108 sequences from 10 donors from (21)]. Informed by our structural model for the BG18-Env interaction, we searched for BG18-like HCDR3 sequences with the same length as BG18, the same D gene in the same reading frame and position within the HCDR3, and the same heavy chain joining region (JH) gene, allowing for diverse V-D and D-J junctions. Such BG18-like HCDR3 sequences were identified in all 14 donors (fig. S5), encouraging us to proceed with vaccine design. We further hypothesized that a range of BG18-like bnAbs utilizing alternate VH or VL genes could potentially interact with Env in a similar HCDR3-dependent binding mode. This hypothesis was subsequently supported by our ability to engineer BG18-like bnAbs utilizing three alternate VL genes (VL3-1, VL3-21, and VL2-8) and two alternate VH genes (VH4-59 and VH4-61) (fig. S6). Identification of diverse BG18-like precursor HCDR3s from NGS data, and construction of BG18-like bnAbs with alternate VH or VL genes, led us to target a broad range of BG18-like precursors in the germline-targeting design process.

Design and antigenic analysis of immunogens

Germline-targeting immunogen design was carried out using a directed evolution method for engineering trimers on the surface of mammalian cells (22, 23). We considered that it would be important to overcome the limitations of using only inferred-germline (iGL) antibodies (BG18 iGL0 to iGL2, fig. S5B) for the directed evolution of a germline-targeting immunogen with strong HCDR3 contacts. We reasoned that the germline-targeting design process directed to only iGL antibodies may fail to produce immunogens with appreciable affinity for diverse naïve precursors. iGL antibodies contain bnAb HCDR3 junctions that have been selected and most likely somatically mutated for high-affinity Env binding during bnAb affinity maturation. Therefore, such iGL antibodies may have features not present in the human antibody sequence repertoire. Furthermore, iGL antibodies likely underrepresent the diversity of potential precursors. We therefore designed a set of 15 BG18-like precursor antibodies that use BG18 germline-reverted genes but contain naïve human BG18-like HCDR3s with diverse junction regions identified in our search of NGS data described above (fig. S5B). On the basis of our finding that BG18-like bnAbs can utilize alternate VH and VL genes, we produced 10 additional BG18-like precursor antibodies with alternate VH or VL (fig. S7). This gave us 28 potential BG18-like precursors that could be used as selection reagents for directed evolution and multitarget optimization of Env trimer immunogens capable of binding and priming a broad range of BG18-like precursor B cells.

Seven Env mammalian cell-surface display libraries, encoding amino acid variation within and around the BG18 epitope, were screened iteratively (20). At each stage, selection antibodies were used to isolate the highest affinity clones from the library, and the best mutations were incorporated into the next-generation Env immunogen. The first library was based on a previously described immunogen, 11mutB (22), that had weak but detectable affinity for BG18 iGL2, the first selection antibody used (Fig. 1C, fig. S8, table S1, and supplementary text). In the early iterations, libraries were screened against the least challenging selection antibodies (e.g., BG18 iGL), whereas in later stages, the libraries were screened against more difficult antibody targets (e.g., NGS-derived and alternate VH or VL precursors) (Fig. 1C, figs. S5 and S7, and table S1). This directed evolution design process resulted in a series of germline-targeting Env trimers with increasing affinity for BG18 precursors (N332-GT1, -GT2, and -GT5; Fig. 1, C and D, fig. S8, and table S1). The N332-GT5 trimer bound with a dissociation constant (Kd) of ~2 pM to BG18 iGL1, which represented a ~14 million–fold improvement over the initial protein design, 11mutB. More importantly, whereas the 14 NGS-derived precursors tested had undetectable affinity to the initial protein design (and undetectable affinity for native HIV Env trimer MD39, Fig. 1D), the design process resulted in 11 of 14 acquiring affinity to N332-GT2 (geomean Kd = 519 nM, Fig. 1D) and 12 of 14 binding to N332-GT5 (geomean Kd = 234 nM, Fig. 1D). [One of the 15 NGS-derived precursors was found to be highly polyreactive and was therefore not included in our surface plasmon resonance (SPR) analyses]. Additionally, although only 3 of 10 alternate VH or VL precursor antibodies bound the starting protein design with low affinity (Kd > 10 μM) and none bound native HIV Env trimer (Fig. 1D), all 10 bound to N332-GT2 and N332-GT5 trimers, with robust affinities (geomean Kd = 11 nM and 572 pM, respectively, Fig. 1D). A Kd ≤ 1 μM may represent an affinity benchmark for generating robust germinal center (GC) responses from rare B cell precursors in the presence of polyclonal competitors in vivo (24), and 20 of 27 potential bnAb precursors bound to the N332-GT5 Env trimer with affinities of Kd ≤ 1 μM (Fig. 1D). Thus, the design process succeeded in extensively improving the immunogen binding properties to potential bnAb precursors with diverse HCDR3s and a variety of VH and VL genes.

Immunogen structural analysis

A cryo-EM–derived structure of BG18 iGL0 complexed with the N332-GT2 trimer at ~3.9-Å resolution (Fig. 1B and table S2) showed that BG18 iGL0 HCDR3 made a similar interaction to the base of the V3 as the HCDR3 of mature BG18 bound to the native-like trimer MD39 (Fig. 1A). Most of the additional interactions of BG18 iGL0 complexed with N332-GT2 arise from V1 mutations in N332-GT2 that occupy a groove in the LC and also contact HCDR3 (figs. S9 and S10). HCDR3 dominates the interaction in the BG18 iGL0 complex with N332-GT2, accounting for 64% of the total buried surface area. In the mature BG18 complex with the MD39 Env trimer, HCDR3 maintains the same key interactions and contributes 35% of the total buried area as the antibody makes substantially increased contacts to glycans N332, N392, and N137 (table S3). Overall, cryo-EM structures showed that N332-GT2 binds to BG18 iGL0 with a similar HCDR3-dependent, V1-straddling binding mode as the BG505 MD39 Env trimer does with BG18.

Immunogenicity testing in a mouse model with rare bnAb precursors

To test the immunogenicity of the N332-GT2 Env trimer, we used a BG18gH knock-in mouse engineered with a CRISPR-Cas9 rapid targeting strategy, in which ~30% of B cells express the BG18 iGL2 HC variable region and mouse constant region paired with mouse LCs (25). The N332-GT2 Env trimer (but not MD39) bound to 12 ± 1% of naïve B cells in this mouse compared with 0.06 ± 0.01% in wild-type (WT) (C57BL/6) mice, demonstrating N332-GT2 specificity for BG18gH naïve B cells (Fig. 2, A and B). Antigen-specific single–B cell sorting and BCR sequencing demonstrated that the N332-GT2–specific naïve BG18gH B cells carry a variety of mouse LCs paired with BG18gH (Fig. 2C). Furthermore, N332-GT2 had similar affinities for naïve BG18gH B cell Fabs (geomean Kd of 582 nM) as for NGS-derived human BG18-like precursors (geomean Kd of 519 nM), showing the physiological relevance of the BG18-like precursor affinities in this mouse model.

Fig. 2 Immunization of BG18gH B cell adoptive transfer recipient mice with N332-GT2 Env NPs.

(A) Gating strategy to identify epitope-specific (N332-GT2++/N332-GT2-KO) B cells in BG18gH and WT mice. (B) Frequency of epitope-specific B cells in nonimmunized BG18gH and WT mice. Each symbol represents a different mouse. Bars indicate mean ± SD from experiments in three mice in each model. (C) Distribution of VH and VL genes in epitope-specific naïve B cells in nonimmunized BG18gH mice. (D) Frequency of GC B cells (left) or CD45.2+ GC B cells (right) in four immunization conditions. Each symbol represents a different mouse. Error bars indicate mean ± SD from experiments in the following number of mice in each condition: BG18gH (GT2), n = 6; WT (GT2), n = 5; BG18gh (MD39), n = 3; and WT (MD39), n = 3. (E) Frequency of CD45.2+ (left) or CD45.1+ (right) epitope-specific B cells in four immunization conditions. Each symbol represents a different mouse. Error bars indicate mean ± SD from experiments in the following number of mice in each condition: BG18gH (GT2), n = 6; WT (GT2), n = 5; BG18gH (MD39), n = 3; and WT (MD39), n = 3. (F) Serum ELISA 50% equilibrium dilution (ED50) values for N332-GT2 and N332-GT2-KO at day 14 after immunization for four immunization conditions. Each symbol represents a different mouse. Error bars indicate geometric mean and geometric SD from experiments in the following number of mice in each condition: BG18gH (GT2), n = 5; WT (GT2), n = 5; BG18gH (MD39), n = 3; and WT (MD39), n = 3. Student’s t test was used. Not significant (ns) P > 0.05; *P < 0.05; **P < 0.01. Data in (A) to (F) are from one of three representative experiments with three or more animals in each group. (G) Distribution of VH and VL genes in epitope-specific GC (CD38lowCD95+) B cells 8 and 42 days after immunization of BG18gH B cell adoptive transfer recipient mice. (H) SPR dissociation constants for N332-GT2 trimer binding to epitope-specific Fabs derived from naïve B cells in nonimmunized BG18gH mice and GC B cells 8 and 42 days after immunization of BG18gH B cell adoptive transfer recipient mice. Each symbol corresponds to a different Fab and represents one or two measurements. Error bars indicate geometric mean and geometric SD. (I) Phylogenetic trees of BCR HCs isolated from epitope-specific CD45.2+ B cells 8 and 42 days after immunization with N332-GT2 NPs. Tree scale indicates the number of substitutions per site. (J) SPR dissociation constants for the five highest affinity naïve Fabs from (H) binding to the V1 loop-modified BG505 trimer (BG505_V1mod) and for nine of the high-affinity day 42 Fabs from (H) and five inferred-germline variants of the high-affinity day 42 Fabs (Day42.iGL) binding to V1 loop-modified trimers from BG505 and three other HIV isolates (SF162P3, AC10, and AD8), as well as a BG505 trimer with a less modified V1 loop (BG505_7mut), a native-like trimer (BG505_MD39), and an epitope-KO trimer (N332-GT2_KO). Each symbol corresponds to a different Fab and represents one or two measurements. Error bars indicate geometric mean and geometric SD. The dashed line indicates the limit of detection. (K) Neutralization potency (IC50) against native (BG505 T332N) and V1 loop-modified (BG505-V1mod) pseudoviruses for the BG18 bnAb, the five highest affinity naïve Fabs from (H), five inferred-germline variants of the high-affinity day 42 Fabs (d42.iGL), and five high-affinity day 42 Fabs (d42). Each IC50 is an average from two measurements. ND indicates not determined.

To generate a mouse model with rare bnAb precursor B cells, we carried out adoptive transfer experiments in which 5000 CD45.2 BG18gH B cells were transferred to CD45.1 WT mice on day −1, establishing a frequency of approximately seven GT2++/KO BG18gH CD45.2 B cells per million CD45.1 B cells by day 0 (fig. S11) (KO indicates knockout). Control transfers were 50,000 CD45.2 WT B cells. Previously, we constructed ferritin nanoparticles (NPs) that displayed up to eight copies of MD39 native-like trimers (26), and mouse immunization studies showed that such NPs were superior to MD39 trimers in trafficking to follicular dendritic cell networks, concentrating in GCs, and eliciting immunoglobulin G (IgG) responses (27). We therefore engineered ferritin NPs displaying N332-GT2 trimers (fig. S12). Recipient mice were immunized at day 0 with either N332-GT2-NPs or control NPs displaying MD39 trimers lacking GT mutations, for a total of four immunization conditions (BG18gH or WT B cells transferred, N332-GT2- or MD39-NPs immunized). Splenocytes were analyzed by flow cytometry at day 8 (Fig. 2, D and E, and fig. S13). GC B cells (CD38lowCD95+) were detected in all four immunization conditions, but CD45.2 GC B cells were detected only in the case of N332-GT2-NP immunization of BG18gH B cell recipients, demonstrating that N332-GT2-NPs activated rare BG18gH B cells in vivo but MD39-NPs did not (Fig. 2D). N332-GT2-NPs induced CD45.2 GC B cells that bound to N332-GT2 but not to N332-GT2-KO (Fig. 2E) and were thus epitope-specific, consistent with a BG18-like response. By contrast, the same NPs induced considerably weaker epitope-specific responses among host CD45.1 GC B cells (Fig. 2E). In day 14 serum-binding analyses, N332-GT2-NPs induced strong epitope-specific IgG responses in BG18gH B cell recipients and 15-fold weaker epitope-specific responses in WT B cell recipients (Fig. 2F), qualitatively consistent with the day 8 GC data. This demonstrated that activation of rare BG18gH precursor B cells led to potent serum antibody responses and also showed that WT B cells responded to the BG18 epitope on N332-GT2. By contrast, MD39-NPs induced negligible BG18 epitope–specific serum responses in either BG18gH or WT B cell recipients (Fig. 2F). Together, these results demonstrated that N332-GT2-NPs elicited GC and antibody responses from rare BG18gH B cells.

By single-cell sorting and BCR sequencing CD45.2+/N332-GT2++/KO GC B cells from BG18gH recipient mice immunized with N332-GT2-NPs, we obtained HC-LC pairs at days 8 and 42. Of the HCs, 100% were derived from BG18gH, formally proving that these GC responses utilized the knock-in HC (Fig. 2G). In contrast to the wide variety of mouse kappa genes used in LCs of N332-GT2–specific naïve BG18gH B cells, by day 8 the LCs from GC B cells were highly enriched for two mouse kappa genes: Igkv12-46 and Igkv12-44 (Fig. 2G). By day 42, GC BCRs showed substantial somatic hypermutation, diversification, and affinity maturation compared with naïve B cells or day 8 GC BCRs (Fig. 2, H and I, and fig. S14). BG18gH BCR Fab affinities for N332-GT2 trimers increased by a factor of ~6 from day 0 to day 8 (geomean Kds of 582 and 97 nM, respectively; Fig. 2H). BG18gH BCR Fab affinities increased dramatically by a factor of ~900 from day 0 to day 42 (geomean Kd = 640 pM, Fig. 2H). We conclude that N332-GT NPs can induce sustained GC responses and considerable affinity maturation and diversification from rare BG18-like precursors with human physiological affinities (see below), even in the presence of polyclonal competition.

To assess whether the affinity maturation induced by this single priming immunization was on a potential path toward bnAb development, we tested whether day 42 antibodies could bind Env trimers more native-like than the germline-targeting immunogen or neutralize viruses with more native-like Env. The N332-GT2 immunogen has 17 germline-targeting mutations, eight of which are in two highly conserved regions of HIV Env (base of V3 loop around the GDIR motif and β19) and nine of which are in one highly variable region (V1 loop) (fig. S8). Thus, a key question was whether antibodies induced by N332-GT2 could recognize Env trimers with more native sequences lacking mutations in the two conserved regions. We therefore constructed a stabilized BG505 Env trimer that included eight of the N332-GT2 mutations in the V1 loop but was otherwise native-like (BG505-V1mod) (fig. S15 and table S1), and we tested nine of the day 42 Fabs (those with highest affinity for N332-GT2) for their ability to bind this V1-modified Env trimer in SPR. All nine Fabs bound BG505-V1mod, with a geomean Kd of 49 nM (Fig. 2J). By contrast, five naïve Fabs (those with highest affinity for N332-GT2) bound BG505-V1mod ~200-fold more weakly, with a geomean Kd of 10 μM, and inferred-germline variants of the day 42 Fabs (day 42 iGL Fabs) either showed no detectable affinity (four of five tested) or bound weakly (10 μM) (Fig. 2J). Furthermore, three of the day 42 Fabs, but none of the day 42 iGL Fabs, bound to the BG505 “7mut” trimer that is only six mutations away from a native-like Env trimer and was previously shown to be on the path toward development of PGT121-class bnAbs (22, 28) (Fig. 2J, fig. S15, and table S1). The day 42 antibodies did not bind N332-GT2-KO, consistent with BG18-like binding (Fig. 2J). None of the day 42 Fabs had detectable binding to the native-like trimer BG505 MD39, which was not surprising given the 17-mutation difference between the N332-GT2 immunogen and MD39 (Fig. 2J and fig. S8). Neutralization assays with BG505 WT and V1mod HIV pseudoviruses were consistent with our SPR findings: five of six day 42 Fabs tested neutralized V1mod HIV but not WT HIV, and none of the naïve or day 42 iGL Fabs neutralized either virus (Fig. 2K). We conclude that a single N332-GT2-NP priming immunization elicited functional BG18-like antibodies that could bind and neutralize viruses bearing Env that retains HIV-conserved regions and is more native-like than the immunogen.

To assess whether the affinity maturation due to priming conferred a degree of reactivity breadth beyond clade A BG505, we tested whether day 42 Fabs could bind to HIV Env trimers from three different isolates and two additional clades (SF162P3 and AC10, clade B; AD8, clade C), all with the same modified V1 loop as BG505-V1mod (fig. S15 and table S1). All nine day 42 Fabs tested bound to the three Env trimers with highly heterologous sequences, with geomean Kds of 50, 110, and 69 nM for SF162-V1mod, AC10-V1mod, and AD8-V1mod, respectively. By contrast, four of five day 42 iGL Fabs had no detectable affinity for these trimers (Fig. 2J). These data show that priming with N332-GT2 in this mouse model induced antibodies with a substantial degree of breadth in that they can bind with relatively high affinity to diverse stabilized Env trimers that share the same V1 loop.

Immunogen reactivity with naïve human B cells

A critical test of the germline-targeting design process was to determine if the N332-GT Env trimers could bind rare bnAb precursor human naïve B cells (29). To our knowledge, this is a human immunogen design benchmark that has only been met previously by the germline-targeting immunogen eOD-GT8 that targets VRC01-class bnAb precursors (9, 15). Attempts to isolate PGT121-related bnAb precursors using 11mutB-related trimers did not succeed (supplementary text and fig. S16), consistent with our hypothesis that germline-targeting design using only iGL antibodies would be unsuccessful because of an inability to accommodate the natural sequence diversity among bnAb precursors in human B cell repertoires. To probe human naïve B cell reactivity to N332-GT Env trimers, we used N332-GT1 and N332-GT2 as sorting reagents and either BG505-MD39 Env (containing a native N332 epitope) or N332-GT2-KO Env (an epitope knockout) as negative sorting probes (Fig. 3A). About 16 million naïve B cells from six donors were probed with N332-GT1, and 62 million naïve B cells from 10 donors with N332-GT2, after accounting for polymerase chain reaction (PCR) and sorting efficiencies (table S4). All donors for ex vivo B cell sorting were distinct from the 14 NGS donors mentioned above (20). N332 glycan supersite epitope–specific naïve B cells [termed high-mannose patch clones (HMP) here] were isolated at a frequency of ~0.001% (Fig. 3B and fig. S17). These epitope-specific B cells were enriched for long HCDR3s (Fig. 3C). The B cells were also highly enriched for VL3-25 and VL3-1 LCs (Fig. 3D), which corresponded to the BG18 VL and a VL that we showed could be used by BG18-like precursors and bnAbs (Fig. 1D and fig. S6). We expressed and purified Fabs from 46 HMP naïve B cell clones (table S5) for further examination of the biochemical properties and specificities of the naïve Abs. Twenty-three HMP Fabs bound to N332-GT1 and/or N332-GT2 by SPR and did not bind detectably to the MD39 native-like trimer, demonstrating proper N332 glycan supersite epitope specificity (Fig. 3, E and F). These SPR-validated epitope-specific clones were highly enriched for VL3-25, VL3-1, or the closely related VL3-10 (Fig. 3E). Thus, the protein design strategy resulted in Env trimers that could successfully bind human naïve B cells with BG18-like LCs.

Fig. 3 Naïve human B cells sorted with N332-GT Env trimers.

(A) Gating strategy for N332-GT epitope–specific sorting of naïve human B cells. (B) Frequency of epitope-specific B cells among IgG-negative B cells. Each symbol represents a different human subject. Error bars indicate geometric mean and geometric mean SD from the following number of independent subjects: N332-GT1, n = 9; N332-GT2, n = 11; and N332-GT5, n = 4. (C) HCDR3 length distribution from epitope-specific sorted cells compared with control B cells. (D) Frequency of VL3-25 or VL3-1 LCs from epitope-specific sorted cells relative to control B cells. Significance of differences from control was evaluated by a chi-square test. *P = 0.01; **P = 0.005; ****P = 0.0001. (E) SPR-derived binding specificities for 46 HMP Fabs corresponding to epitope-specific naïve human B cells isolated by N332-GT1 or N332-GT2 (top), with LC V gene usage for nonbinding Fabs (bottom left) and for N332 epitope–specific Fabs (bottom right). (F) SPR dissociation constants for HMP epitope–specific Fabs isolated with N332-GT1 and N332-GT2 Env trimers. The dashed line indicates the limit of detection.

All N332-supersite bnAbs identified to date require a long HCDR3 [≥20 amino acids (aa)], owing to the structural requirements for the bnAb HCDR3 to reach the HIV Env protein surface at the base of V3 while avoiding V1 loop glycans (19). Although epitope-specific human B cells with HCDR3 lengths <20 aa were isolated using N332-GT1 or N332-GT2, only four of eight such clones tested by SPR were confirmed to be epitope specific, and their binding was weak (Kd >10 μM) (Fig. 3F and table S6). We considered that such B cells with HCDR3s <20 aa are probably unable to develop into N332-supersite bnAbs, and thus we did not study those clones further. Numerous epitope-specific naïve B cell clones with HCDR3s ≥20 aa were isolated with N332-GT1 and N332-GT2 probes (Fig. 4, A and B). From these human naïve B cell clones, we identified two categories of potential BG18-like precursors. The first category shared the same HCDR3 length, D gene, D gene reading frame, D gene position within HCDR3, and JH gene with BG18 (Fig. 4A), exactly matching our initial search criteria when scanning NGS data for BG18-like HCDR3 sequences. Such naïve B cells were termed type I BG18-like precursors. The second category of epitope-specific BG18-like B cells had VL3-25, VL3-1, or VL3-10 and long HCDR3s (≥20 aa) with diverse HC sequences (Fig. 4B). We termed this more diverse class of isolated naïve B cells type II BG18-like precursors. HMP1 was a type I BG18-like precursor (Fig. 4A) with high affinity for the N332-GT2 Env trimer (Kd = 220 nM, Fig. 3F). The type II BG18-like precursors with confirmed binding exhibited a geomean Kd of 10 μM for the N332-GT2 Env trimer (Fig. 3F). Overall, type I and type II precursors accounted for 74% (17 of 23) of the HMP Fabs isolated by N332-GT1 or N332-GT2 and verified as epitope-specific by SPR (Fig. 3, E and F), indicating that such BG18-like precursors may represent a substantial fraction of the human naïve epitope-specific repertoire to these Env trimers. We then isolated additional type I and type II naïve B cell clones using N332-GT5 Env trimer probes with additional blood donors (Fig. 4, A and B). Overall, three type I BG18-like precursors were isolated at a frequency of ~1 in 53 million naïve B cells (HMP1, HMP68, and HMP69; table S4), in good agreement with our initial NGS bioinformatics-based estimate that precursors with BG18-like HCDR3s specific for N332-GT trimers may be present in the human B cell repertoire at a frequency of 1 in 54 million naïve B cells (fig. S5). Type II BG18-like precursors were isolated at a higher frequency of ~1 in 7 million naïve B cells, consistent with their larger sequence space.

Fig. 4 Sequence and structural characterization of type I and type II BG18-like naïve antibodies isolated by N332-GT Env trimers.

(A) HCDR3 sequence and gene segment assignments for three type I BG18-like naïve human precursor antibodies. V, D, and J gene segments are colored blue, red, and green, respectively. Single-letter abbreviations for amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr. (B) HCDR3 sequence and gene segment assignments for 20 type II BG18-like precursor antibodies. (C) Cryo-EM structural analysis of type I (HMP1) and type II (HMP42) precursor antibody LC interactions with N332-GT5 compared with BG18 iGL LC interactions with N332-GT2. Gp120 is colored gray, and the LCs are colored cyan, yellow, and blue for BG18 iGL, HMP1, and HMP42, respectively. (D) Cryo-EM structural analysis of type I and type II precursor HCDR3 interactions with N332-GT5 compared with HCDR3 interactions for BG18+MD39 and BG18iGL+N332-GT2 complexes. Gp120 is colored gray with conserved residues (or the corresponding germline-targeting amino acids) near the base of V3 colored red as in Fig. 1A. Glycans are shown as green sticks.

Structural analysis of BG18-like human precursors bound to immunogens

To gain a structural understanding for the potential of human type I and type II BG18-like precursors (Fig. 4, A and B) to mature into bnAbs, we solved high-resolution cryo-EM structures of the complexes of N332-GT5 bound to HMP1 (type I) and HMP42 (type II), with resolutions of 3.7 and 3.4 Å, respectively (Fig. 4, C and D, and table S2). Both HMPs showed a similar LC binding mode as BG18 iGL, with the LC straddling the V1 loop (Fig. 4C and fig. S10). The HCDR3s of HMP1 and BG18 iGL have nearly identical conformations, supporting HMP1 and type I class naïve antibodies as ideal BG18 precursors (Fig. 4D). The projecting HCDR3 tip of HMP42 interacts with the same Env patch as BG18 but has a slightly different overall conformation that makes additional contacts with the Env V1 loop (Fig. 4D). This structural information supports the hypothesis that some or possibly all type II BG18-like naïve antibodies have a similar binding mode as BG18 iGL. Overall, these findings support the potential for both type I and type II BG18-like precursors to mature into BG18-like bnAbs (bnAbs with a BG18-like binding mode) under an appropriate sequential vaccination regimen. Given that type I and type II BG18-like precursors are enriched among epitope-specific human naïve B cells and have affinities that may confer competitive fitness in GCs, the data indicate that N332-GT Env trimers are strong candidates for priming BG18-like precursors for potential maturation into HIV bnAbs in humans.

Application to vaccine design for pathogens other than HIV

We explored whether our approach to target and prime a diverse pool of antibody precursors may have applicability to other pathogens. To evaluate whether our method of germline-targeting vaccine design could be applied beyond HIV, we carried out sequence and structural analyses for selected bnAb-antigen complexes for several major pathogens. In this nonexhaustive survey, we identified 11 potential antibody targets from five major pathogens, including hepatitis C virus (HCV), influenza virus, malaria, and dengue and Zika viruses (fig. S18). According to our sequence and buried surface area analyses, these antibodies all share the ability to make a series of important contacts with antigens through templated portions of their HCDR3s (portions encoded by D or J genes), which can be targeted by vaccine design. Most of the antibodies we identified are strongly HCDR3 dependent, on the basis of a criterion of HCDR3 contributing >30% of all surface area buried on the antibody. The strong HCDR3-dependence of the antibodies may allow for the development of related antibodies utilizing alternate VH or VL genes (as occurred with BG18) and hence may be advantageous for precursor frequency. With the exception of the dengue and Zika antibody EDE2 A11, all target antibodies have relatively common HCDR3 lengths of ≤22 aa represented by ≥2% of human antibodies (30), suggesting that HCDR3 length will not pose a limitation on precursor frequency. All target antibodies also have mutation levels in VH and VL that are present in ≥1 to 2% of human memory B cells (30), and all but two (9 of 11) lack indels, thus mutation level and indels should not pose a limitation on production of similar antibodies if appropriate precursors can be primed. In some cases, the native antigen has been shown to bind to an inferred-germline or unmutated common ancestor of the target antibody (3133), raising the question of whether a germline-targeting approach would be necessary. We propose that even in such cases, our strategy may improve the design or validation of a vaccine priming candidate. Identification of a diverse set of antibody potential precursors with diverse HCDR3 junctions should allow for testing the breadth of precursor reactivity of the native antigen, and our design and validation strategies may optimize and/or verify breadth.

Concluding remarks

Most antibodies, and most HIV bnAbs, recognize their target in a strongly HCDR3-dependent manner. A central challenge of germline-targeting vaccine design is the large paratope sequence space and structural complexity possible for any set of antibodies targeting a conserved epitope by means of a shared HCDR3-dependent binding modality. Here, we demonstrate the successful design of a germline-targeting immunogen for this general class of antibody recognition. We used the human repertoire and structural features of bnAb-Env binding as guides to identify a pool of potential bnAb precursors and then design an immunogen with affinity for a representative set of those precursors. This procedure was validated by the isolation of three type I BG18-like precursors from naïve human B cells with N332-GT trimers and the demonstration that N332-GT NPs drove a robust BG18-class B cell response in an animal model with rare BG18 precursors. Furthermore, N332-GT trimer–sorted human naïve B cells were also enriched for type II BG18-like BCRs, and such precursors exhibited a BG18-like binding mode, indicating that the pool of potential BG18-like human naïve precursors is larger and more diverse than originally expected. This study does not demonstrate the induction of neutralizing antibodies to WT HIV isolates; the goal for germline-targeting priming immunogens is not to induce bnAbs directly but rather to induce bnAb-precursor B cell responses that have the potential to mature into bnAbs. Induction of bnAbs is the aim for a complete germline-targeting vaccine regimen, which would include a germline-targeting prime and a series of shepherding and polishing immunogens. Overall, we describe a new approach to define bnAb precursors for an epitope of interest and the design of vaccine priming immunogens that take advantage of that information. This approach lays out a generalizable pathway for the development and preclinical validation of germline-targeting immunogens for HCDR3-dependent antibody responses.

Materials and Methods

Experimental design

The overall objective of the study was to test a new, general method for design of germline-targeting immunogens to prime human naïve precursors to known bnAbs. Here we address the study design for the mouse model experiments.

Study objectives and experimental design

These experiments were designed primarily to test whether N332-GT NPs could generate GC responses with detectable levels of BG18gH B cells in GCs, under preimmunization conditions of low BG18gH naïve precursor B cell frequency and high polyclonal competition. BG18gH naïve and GC B cells were identified by cytometry using cell surface markers, including the CD45.2 marker that distinguished these B cells from the WT host mouse B cells that were marked with CD45.1 Additionally, BG18gH naïve and GC B cells were single-cell sorted using N332-GT and N332-GT-KO probes, and the epitope-specific BCRs were sequenced, in order to prove that the HCs were derived from the BG18 iGL2 HC knock-in gene. This was not a foregone conclusion because the adoptively transferred B cells in these experiments were from a heterozygous knock-in with ~30% of B cells expressing the BG18 iGL2 HC variable region. Finally, to assess the degree to which somatic hypermutation led to increased affinities in the BG18gH B cells, soluble Fabs were expressed based on the sorted epitope-specific BCR sequences, and SPR studies were conducted to evaluate binding affinities to N332-GT immunogens. Additional corroborating information was gleaned from these studies by serum ELISA analysis.

Sample size

The number of mice in each group was limited by mouse availability and the costs and time associated with the experiments; however, the number of mice used was judged to be sufficient to detect clear differences between groups.

Randomization and blinding

Animal recipients of adoptive transfer were assigned to groups with no pattern. Neither randomization nor blinding were used, as they were not deemed necessary.

Data exclusion

No data were excluded.

Replicates

Data presented are from two independent experiments. The results have been reproduced in at least two additional experiments in this same mouse model and in a different but related mouse model.

BG18/BG505 SOSIP structural model

The HCDR3 loop of unliganded BG18 (PDB: 5UD9) was aligned to PGT122 (PDB: 4TVP) and several features suggested this as a plausible binding mode. First, ArgL54 in LCDR2 would be positioned in a similar space as LCDR3 ArgL94 in PGT122, a known critical contact residue for PGT122. Second, ArgH29 was positioned close to the N332 glycan, and we confirmed that ArgH29 was important for neutralization, by mutagenesis (fig. S2). Finally, analysis of computationally predicted V1 conformational ensembles including protein and glycan conformational diversity suggested that the LC could plausibly avoid clashing with the N137 glycan.

Design of minimally mutated versions of BG18

Design of minimally mutated versions of BG18 shown in fig. S4 was guided by analysis of the structural model of BG18 bound to BG505 SOSIP (fig. S2B), as this work was carried out before crystal structures were published of BG18 bound to BG505 and B41 SOSIP trimers (18) and before we obtained a cryo-EM structure of BG18 bound to the MD39 trimer (Fig. 1A). Framework and CDR mutations were reverted to germline if structural inspection indicated they were not contributing to the binding interaction. Several HC and LC variants were tested. BG18.11, which we refer to as “minBG18,” was the least mutated variant that retained at least 50% of the breadth of BG18 while retaining similar potency as BG18. BG18.6 was the least mutated variant that showed any neutralization.

Design of BG18 bnAb variants using alternate VH or VL genes

The VL variant engineering was carried out early in the study and was therefore guided only by our model for the BG18-Env interaction (fig. S2B). To engineer the VL variants shown in fig. S6, the indicated VL gene was substituted for the BG18 VL gene and BG18 mutations were incorporated. The engineering of VH variants was more complicated and was informed by our structural and mutagenesis studies (Fig. 1A and figs. S2, E and F, and S4). These studies indicated that the most important feature of the BG18 VH gene (VH4-4) was HCDR1 with a length of nine amino acids that is rare among human VH genes. We therefore tested whether BG18-like bnAbs could utilize alternate VH genes (4-59 and 4-61) that are closely related to VH4-4 but use the more common HCDR1 lengths of 8 and 10 amino acids found among ~74 and ~19% of human VH genes, respectively. We used mammalian display to screen scFv libraries containing ~104 to 105 HCDR1 sequences for binding to gp120 and native-like trimers based on several isolates (B41, 191084, ZM197, 6811) using the directed evolution design process described previously (22). One VH4-59 library included an NNK codon in the HCDR2 at position 53 in addition to containing HCDR1 sequence diversity. Briefly, libraries were integrated into 293T cells using a dox-inducible lentivirus based system; the scFv was anchored to the cell surface by linking the C-terminus to a PDGFR transmembrane domain; and the cells were incubated with HIS-tagged Env proteins and then stained with anti-HIS PE (miltenyi biotech). With this process, we identified two VH4-59 clones and one VH4-61 clone that when expressed as soluble IgG showed neutralization breadth on a BG18-sensitive virus panel (fig. S6). Because the theoretical number of HCDR1 sequences for lengths 8 and 10 are ~1010 and ~1013, respectively, there are likely to be many HCDR1 sequences that can support neutralization beyond what we identified here. We conclude that BG18-like antibodies with diverse VH and VL genes can achieve broad and potent neutralization. It follows that BG18-like precursors containing alternate VH and VL genes should be targeted by vaccine design.

Immunogen design by mammalian cell surface display directed evolution

BG18 iGL2 had detectable affinity to the 11mutB (PGT121 germline targeting) trimer (22) but no detectable affinity to BG505 MD39 containing a native N332 epitope; therefore, we used 11mutB as a base construct to begin the BG18 germline-targeting design process. The following libraries were screened using a previously described mammalian cell surface display method (22). Briefly, the following libraries were cloned into the pLenti CMVTRE3G Puro Dest plasmid and then stably integrated into rtTA3G-expressing HEK 293T cells using lentiviral transduction. Library 1 was a screen of all 20 aa at a subset of positions in the BG18/PGT121 epitope. It was an NNK codon scan of positions 294, 297, 298, 299, 300, 302, 304, 305, 326, 329, 330, 333, 386, 413, 414, 415, 416, 417, 419, and 420. NNK codons were introduced into BG505-11mutB-gp120 using the QuikChange Site-Directed Mutagensis Kit (Agilent). Library 1 was screened for binding to BG18 iGL2 and PGT121-GLCDR3rev4. Library 2 was intended to sample hydrophobic amino acids underneath the BG18 epitope. It was a combinatorial library with amino acids F/I/L/V introduced at positions 154, 322, 323, 326, 333, 414, 415, and 416. The library insert was assembled with overlapping ultramers (IDT DNA) followed by Gibson cloning (NEB) into BG505-11mutB-gp120. Library 2 was screened for binding to BG18 iGL2 and PGT121-GLCDR3rev4. Library 3 was intended to test all 20 aa at key epitope contact positions predicted by the structural model. It was a combinatorial library with NNK codons introduced at positions 137, 325, and (F/I/L/V) at position 326. The library insert was assembled with overlapping ultramers (IDT DNA) followed by Gibson cloning (NEB) into BG505-MD39-17mutE. Library 3 was screened for binding to the following 24 Abs: BG18 iGL0, BG18 iGL1, pre1 - pre6, pre8, pre10 - pre15, VL2-8, VL2-14, VL3-21, VH1-69, VH3-33, VH4-59, VH5-51, PGT121-GLCDR3mat, and PGT121-GLCDR3rev1. Library 4 was designed to test all 20 aa at key epitope contact positions predicted by the structural model and mutations at position 325 that were isolated in the library 3 screen. It was a combinatorial library with NNK codons introduced at positions 138 and 141 and (P/H/A/D) at position 325. The library insert was assembled with overlapping ultramers (IDT DNA) followed by Gibson cloning (NEB) into BG505-MD39-17mutE-N137K. Library 4 was screened for binding to BG18 iGL0, pre3, pre14, VL3-21, VL2-8, VL2-14, VH3-33. Library 5 was designed to test all amino acids at key epitope contact positions predicted by the structural model and mutations at position 325 that were isolated in the library 3 screen. It was a combinatorial library with NNK codons introduced at positions 138 and 139 and (P/H/A/D) at position 325. The library insert was assembled with overlapping ultramers (IDT DNA) followed by Gibson cloning (NEB) into BG505-MD39-17mutE-N137K. Library 5 was screened for binding to BG18 iGL0, VL3-21, VL2-8, VL2-14, VH1-69, and VH5-51. Library 6 tested all 20 aa at positions not directly in the BG18 epitope to identify mutations that may indirectly effect binding to BG18 precursors. It was an NNK scan of positions 167 to 308. The insert was synthesized at SGI-DNA and Gibson cloned (NEB) into BG505-MD39-N332-GT3. Library 6 was screened for binding to BG18 iGL0, VL2-8. Library 7 screened all 20 aa at positions in and around the BG18 epitope, excluding the V1 loop. It was an NNK scan of positions 309 to 443. The insert was synthesized at SGI-DNA and Gibson cloned (NEB) into BG505-MD39-N332-GT3. Library 7 was screened for binding to BG18 iGL0, pre1, pre2, pre4, pre10, pre15, and VL2-8. All constructs contained a C-terminal myc tag and were anchored to the cell membrane via a C-terminal PDGFR transmembrane domain. Staining of the cell populations was typically done with IgG until saturated binding was obtained at low nanomolar IgG concentrations, and then Fabs were used for staining to maintain selection pressure. IgGs were labeled with Anti-Human IgG-R-PE (Sigma) and Fabs were labeled with Human IgG Fab PE (LSBio). Cell surface protein expression was detected using FITC conjugated chicken anti-CMYC (ICL inc). Typically, libraries were sorted three to five times, and the enriched cell populations were frozen until sequencing could be carried out as described previously (22). Mutations found in the most enriched clones were incorporated into the most recent designs, synthesized at Genscript either as C-terminal His tagged gp120s or MD39/MD64 trimers in the pHLsec vector and expressed and purified as described previously (22).

Nanoparticle design and purification

To obtain multivalent immunogens, trimers were genetically fused to ferritin from Helicobacter pylori using a short flexible linker. Genes were codon optimized for HEK293 cells and cloned into the pHLsec plasmid (GenScript). MD39-NP DNA was cotransfected with a plasmid encoding human Furin protease into FreeStyle 293F cells (Invitrogen, Cat no. R79007) using 293Fectin (ThermoFisher) and proteins were expressed at 37°C for 4 days. NPs were purified either using snow drop lectin-conjugated agarose beads (Vector laboratories) or HiTrap NHS-Activated HP affinity columns (GE Healthcare) conjugated with PGT145, each followed by gel-filtration using a Superose 6 size-exclusion chromatography column (GE Healthcare). N332-GT2 NP had the Furin cleavage site replaced by a flexible linker (SHSGSGGSGSGGHA) as well as an L545P mutation, both discovered by library screening; hence, N332-GT2 NP was not cotransfected with Furin. NP-assembly was assessed by negative-stain EM and SEC + multiangle light scattering (SEC-MALS) using a Superose 6 10/300 column (GE Healthcare) at a flow rate of 0.5 ml/min followed by DAWN HELEOS II and Optilab T-rEX detectors (Wyatt Technology), correcting for the glycan molecular mass by applying the built-in protein-conjugate analysis (ASTRA).

Neutralization activity

Neutralizing activity of monoclonal antibodies (mAbs) was assessed using a single round of replication in TZM-bl target cells, in the absence of DEAE-dextran except for the assays in Fig. 2K, as described previously (34). Briefly, pseudoviruses were generated by cotransfection of HEK293T cells with an Env-expressing plasmid and an Env-deficient genomic backbone plasmid (pSG3ΔEnv).

Cryo-EM structure determination

High resolution cryo-EM structures were determined for four complexes: (i) MD39 + BG18; (ii) N332-GT2 + BG18iGL; (iii) N332-GT5 + HMP1; and (iv) N332-GT5 + HMP42. Of note, initial attempts to determine even low-resolution EM structures of HMP Fabs bound to N332-GT2 were not successful. However, in the late stages of this study, we found that the N332-GT5 trimer proved capable of forming stable complexes with the two clones, HMP1 and HMP42, representing BG18-type I and type II precursors, respectively. In general, trimers were incubated with a 6-10× molar excess of Fab overnight at room temperature. Complexes containing HMP1 or HMP42 also included RM20A3 Fab, a non-neutralizing trimer base-binding antibody that helps increase orientation sampling of the particles. The following morning, each complex was purified using a HiLoad 16/600 Superdex 200pg size-exclusion column (GE Healthcare) with Tris-buffered saline (50 mM Tris pH 7.4, 150 mM NaCl) as the running buffer, and the peak corresponding to trimer-Fab complex was pooled and concentrated to ~6 to 8 mg/ml. 3.5 μl of each complex was mixed with 0.5 μl of 0.42 mM n-dodecyl β-d-maltoside (DDM; Anatrace), such that final DDM concentration (0.06 mM) is below the critical micellar concentration (CMC). A 4-μl aliquot of the complex was applied to a C-Flat grid (CF-2/2-4C, Electron Microscopy Sciences, Protochips, Inc.) or Quantifoil grid (Q 1.2/1.3-4C, Quantifoil Micro Tools GmBH), which had been plasma cleaned for 10 s using a mixture of Ar/O2 (Gatan Solarus 950 Plasma system), and following a 10-s incubation, the grid was blotted between 4 to 6 s and plunged into liquid ethane using an FEI Vitrobot Mark IV (100% relative humidity, 10°C).

The samples were imaged using either a Thermo Fisher Titan Krios operating at 300 kV or a Thermo Fisher Talos Arctica operating at 200 kV, both with a Gatan K2 Summit direct electron director operating in counting mode. Automated data collection was performed using the Leginon software suite (35). Each micrograph movie (250-ms exposure per frame) was collected at a magnification of 29,000× and a pixel size of 1.03 Å (Krios) or 36,000× and a pixel size of 1.15 Å (Arctica). Data collection statistics for each sample are summarized in table S1. Micrograph movie frames were aligned and dose-weighted using MotionCor2 (36), and CTF models were calculated using GCTF (37).

Single particles were selected using DoGPicker (38) from the whole-frame aligned and summed micrographs, and particles extracted using Relion 3.0 (39) using a box size of 288 or 320 pixels. After numerous rounds of 2D and 3D classification, final reconstructions were performed in Relion 3.0, and after postprocessing, the final resolution estimates (FSC 0.143) are ~3.9 Å for N332-GT2 + BG18iGL (C3 symmetry), ~4.4 Å for MD39 + BG18 (C3 symmetry), ~3.7 Å for N332-GT5 + HMP1 + RM20A3 (C3 symmetry), and ~3.4 Å for N332-GT5 + HMP42 + RM20A3 (asymmetric). Additional data processing statistics are summarized in fig. S3.

Atomic models were built and refined into the high-resolution reconstructions by creating homology models based off deposited coordinates of BG505 SOSIP.664 (PDB 5cez) and 354BG18 Fab (PDB 5ud9), as well as docking of an HMP42 Fab crystal structure from this study (table S6), followed by an iterative cycle of manual building in COOT (40) and real space refinement in Phenix 1.13 (41) and Rosetta Relax 3.10 (42). Glycans were validated by CARP (43) and Privateer (44), and overall structures were evaluated using EMRinger (45) and MolProbity (46). Buried surface area calculations were performed using UCSF Chimera (47).

ELISA

For analysis of serum responses from immunized mice

N332-GT2–specific antibody titers were detected by ELISA, using anti-His Ab (2 μg/ml) to capture N332-GT2 or N332-GT2-KO antigen (2 μg/ml) on the plate. Mouse sera were incubated for 2 hours and alkaline phosphatase conjugated anti-mouse IgG (Jackson ImmunoResearch, #115-055-071) was incubated another hour. Titers were determined from the dilution curve in the linear range of absorbance. All noncommercial ELISA plates were developed with p-Nitrophenyl Phosphate (Sigma, # N2770). Absorbance at 405 nm was determined with a plate reader (BioTek).

For analysis of mAb binding

mAb binding ELISAs were performed by capturing antigen (1 µg/ml) onto plates precoated with anti-His antibody (1 µg/ml; Genscript) and blocked with blocking buffer (5% skim milk, 1% fetal bovine serum, 0.2% tween 20 in PBS). Dilution series of mAbs were added as indicated and labeled with peroxidase-conjugated goat anti-mouse IgG (1:5000; Jackson ImmunoResearch). Wells were developed with 1-Step Ultra TMB-ELISA substrate (Thermo Scientific) diluted 1:4 in H2O and stopped by addition of 0.5M H2SO4. Absorbance was read at 450 nm and reference absorbance measured at 570 nm was subtracted from each well.

Surface plasmon resonance (SPR)

Kinetics and affinities of antibody-antigen interactions were measured on a ProteOn XPR36 (Bio-Rad) using GLC Sensor Chip (Bio-Rad) or Biacore4000 (GE) with Series S Sensor Chip CM5 (GE). We used 1× HBS-EP+ pH 7.4 running buffer (20× stock from Teknova, Cat. No. H8022) supplemented with BSA at 1 mg/ml. Following manufacturer’s instructions for Human Antibody Capture Kit (Cat. No. BR-1008-39 from GE), we immobilized about 6000 RUs of capture mAb onto each flow cell of GLC Sensor Chip or about 10,000 RUs in the case of the CM5 Sensor Chip. In a typical experiment on the ProteOn XPR36 system, about 300 to 400 RUs of mAbs were captured onto each flow cell, and analytes were passed over the flow cell at 50 μl/min for 3 min followed by a 5-min dissociation time. Regeneration was accomplished using 3 M magnesium chloride with a 180-s contact time and injected four times per cycle. Raw sensograms were analyzed using ProteOn Manager software (Bio-Rad), including interspot and column double referencing, and either Equilibrium fits or Kinetic fits with Langmuir model, or both, were used when applicable. For the Biacore4000 instrument, we used similar conditions but lower ligand capture levels. In the case of Fab-antigen kinetic and affinity measurements on ProteOn XPR36 or Biacore4000, we used a similar ligand-capture technique with several modifications. The capture reagent was His-tag Rabbit pAb (GenScript Cat. No. A00174). It was amine coupled to the Sensor Chip surface using the same protocol from the GE Human Antibody Capture Kit referenced above. Our regeneration solution was phosphoric acid 0.85% with a 30-s contact time, four injections per cycle. In the case of the ferritin nanoparticle experiment, we used the ProteOn XPR36 system and Human Antibody Capture protocol described above with one additional step. We captured PGT128 IgG at 1300 RU level in all channels, including reference, followed by NP (as ligand) capture at 1600 RU. All other steps were the same as in the Human Antibody Capture protocol. Analyte concentrations were measured on a NanoDrop 2000c Spectrophotometer using an absorption signal at 280 nm (8).

NGS dataset of human BCR HCs

This work utilized a large NGS dataset of 1.1 × 109 amino acid sequences of BCR HCs from 14 healthy, HIV-uninfected donors. In this dataset, 255 million sequences from 10 donors were obtained from (21), which used the HiSeq sequencing platform and an amplification strategy including unique identifiers (UIDs) to enable discrimination of unique mRNA transcripts from PCR artifacts. These 10 donors were evenly divided between males and females and nearly evenly divided between Caucasians and African Americans, and had ages ranging from 18 to 30 (21). These sequences were collapsed by UID, assigned to VDJ gene segments with Abstar (21), and then rendered unique by clustering at the 99% amino acid identity level within each of six biological replicates per donor. Thus the 255 million sequences were unique at the amino acid level within biological replicates. JSON output files from Abstar were converted to parquet format and uploaded to the Amazon S3 storage cloud. To query databases, Amazon Elastic Map Reduce (EMR) 5.15.5 was used to configure a Spark cluster with added PySpark and Zeppelin configurations. Zeppelin was used to assemble PySpark scripts to query the database with custom scripts. An additional 858 million sequences from four additional donors were obtained here by both HiSeq and NextSeq sequencing platforms without the use of UIDs, as described below.

BCR HC sequencing for four donors

Full leukopaks (three blood volumes) were obtained from four human subjects (AllCells LLC or Hemacare, Inc.) under a protocol approved by the Institutional Review Board of the respective commercial provider. All subjects were healthy, HIV-negative adults with no reported acute illness in the 14 days prior to leukapheresis, and samples were deidentified prior to shipment. The Institutional Review Board of The Scripps Research Institute determined that research with these samples did not constitute human subjects research. Immediately upon receipt of the leukopak, peripheral blood mononuclear cells (PBMCs) were purified by gradient centrifugation and cryopreserved in aliquots of approximately 5 × 108 PBMCs. The junctional regions of antibody heavy chain libraries were amplified as in Willis et al. (48). SPRI-purified sequencing libraries were initially quantified using fluorometry (Qubit, Thermo Fisher Scientific) before size determination using a bioanalyzer (Agilent 2100). Libraries were requantified using qPCR (KAPA Biosystems) before sequencing on either an Illumina HiSeq (2 × 150–bp chemistry) or NextSeq (2 × 150–bp chemistry). Sequences were merged with PANDAseq using the default (symple_bayesian) merging algorithm before annotation with Abstar (21). Identical amino acid sequences from the same donor and biological replicate were collapsed into a single unique amino acid sequence.

BG18 precursor frequency estimate

The NGS dataset of human BCR HCs was queried by bioinformatic searches to gain information on the frequency of BG18-like HCDR3s in the human B cell repertoire (fig. S5). HCDR3s meeting the definition of BG18-like feature set i in fig. S5A, constituting a broad set of potential BG18-like HCDR3 precursors, were identified in 14 of 14 donors (fig. S5C). The geomean frequency was 1 in 58,000 among the 10 donors sequenced by Briney et al. (21) using UIDs. To refine this frequency estimate, we considered that only 11 of 14 BG18 iGL variants with NGS-derived HCDR3s differing in the HCDR3 junctions (fig. S5B) exhibited binding to N332-GT2 (Fig. 1D). BG18-like feature set ii (HCDR3 junction features) in fig. S5A characterized amino acids present in the nontemplated junction regions of BG18 and its somatic variants (17) and in the 11 precursors that bound to N332-GT2. The frequency of HCDR3s meeting the definitions for both feature sets i and ii was found to be lower than those within set i by a factor of 104. Because the VL gene plays an important role In the BG18 V1-loop straddling binding mode, we incorporated VL gene usage into the frequency estimate, as feature set iii (“VL gene family”). We made the conservative assumption that only VL3 LCs can support the BG18-class binding mode, as all VL3 LCs tested bound with high affinity to N332-GT2. The frequency of all VL3-derived Abs in the HC-LC paired sequences in DeKosky et al. (49) was 1 in 9 (13845 VL3s in 127701 sequences). We also assumed that any VH gene can support this binding mode, because when five of the most common human VH genes were substituted into BG18 iGL1, all five variants showed low nanomolar binding to N332-GT2 (Fig. 1D). Therefore, no frequency factor was imposed for VH gene usage. Multiplying the frequencies of all three feature sets together gave our best estimate for the frequency in the human B cell repertoire of BG18-like precursors that could be targeted by N332-GT2: 1 in 54 million.

N332-GT–specific naïve human B cell sorting and BCR sequencing

LRS (leukoreduction) tubes were obtained from the San Diego Blood Bank from healthy, HIV-seronegative human donors. These studies do not constitute human subjects research, as determined by the Institutional Review Boards of both La Jolla Institute and The Scripps Research Institute. More than 1 billion peripheral blood mononuclear cells were regularly recovered from each donor. CD19+ B cells were isolated using a positive-selection magnetic-bead separation kit (Miltenyi Biotec) and resuspended in complete RPMI media with 10% FBS.

Avi-tagged protein immunogens were biotinylated using the Bulk BirA kit (Avidity, LLC). N332-GT5 and N332-GT5-KO probes were used in N332-GT5 sorting experiments. N332-GT2 and N332-GT2-KO probes were used in N332-GT2 sorting experiments. N332-GT1 and MD39 probes were used in N332-GT1 sorting experiments. 11mutB and MD39 probes were used in 11mutB sorting experiments. Biotinylated protein immunogens were individually premixed with fluorescently labeled streptavidin to form tetramer probes. Multiple tactics were used to avoid false positives: (i) used two “positive” probes, (ii) each “positive” probe used a different protein tag (His-tag or Strep-tag) to avoid tag specific B cells, (iii) used a “negative” probe to identify N332-epitope specific B cells, (iv) chose independent (no tandems) fluorochromes for all probes to avoid fluorochrome specific B cells (29). For example, N332-GT2 sorting experiments used the follow probes: N332-GT2–StrepTag-biotin + streptavidin Alexa Fluor 647 (“N332-GT2-S-A647”), N332-GT2–HisTag-biotin + streptavidin Brilliant Violet 421 (“N332-GT2-H-BV421”), and N332-GT2-KO–StrepTag-biotin + streptavidin phycoerythrin (“N332-GT2KO-S-PE”).

Cells were incubated with N332-GT probes for 20 min at 4°C. Without washing, anti-CD19 (PE-Cy7,ThermoFisher, clone HIB19) and anti-CD20 (PE-Cy7, ThermoFisher, clone 2H7), in addition to anti-IgG (APC-Cy7, Biolegend, clone HP6017), anti-CD3 (APC eFluor780, ThermoFisher, clone UCHT1), anti-CD14 (APC eFluor780, ThermoFisher, clone 61D3), anti-CD16 (APC eFluor780, ThermoFisher, clone eBioCB16), and Live/Dead (APC eFluor780, ThermoFisher) for exclusion, were added for an additional 20 min. A BD FACSAria was used for all cell sorting. Cells were sorted at a flow rate of 1500 events/s using an 85-μm nozzle. Sorting stringency was set to a strict setting to obtain one cell per well. Single B cells were sorted directly into cold lysis buffer or N332-specific clonal B cell lines were generated and interrogated as in (9). cDNA synthesis, nested BCR PCR, Sanger sequencing and sequence analysis was carried out as in (15). Sorting was done with FACSDiva (BD) software and post-sort analyses were done with FlowJo (FlowJo, LLC).

In order to determine a normal HCDR3 length distribution for naïve human B cells, we combined sequences from DeKosky et al. (49) and Jardine et al. (30). For control VL gene frequencies, only H+L paired sequences from DeKosky et al. (49) were used. Statistical analysis of LC V gene usage (Fig. 3D) was done by applying Fisher’s exact test (two-sided) to each immunogen-probe B cell set (i.e., N332-GT1, N332-GT2, N332-GT5) compared to the reference normal population.

Adoptive transfer experiments and immunization

For adoptive transfer experiments, B cells were isolated from CD45.2 C57BL/6J (“WT”) or BG18gH KI mice of 8 to 10 weeks of age, and cells were resuspended in 150 μl of PBS and counted by The NucleoCounter NC-200 (ChemoMetec USA Inc). The 150-μl cell suspensions were injected i.v. into CD45.1 B6.SJL-Ptprca Pepcb/BoyJ recipient animals (5 × 103 cells per mouse for BG18gH transfers and 50 × 103 cells per mouse for C57BL/6 transfers). One day later, recipient mice were injected i.p. with 10 μg GT2- or MD39-NPs with Sigma adjuvant (Sigma, # S6322 SIGMA). After 8 days, mice were sacrificed to harvest spleen samples. Blood samples were taken from the submandibular vein on days 0 and 14 after immunization. Four immunization conditions were tested in two independent experiments, with the following total number of mice in each condition: (i) BG18gH B cell transfer, N332-GT2 NP immunization (N = 6 for day 8 GC analysis, N = 5 for day 14 ELISA); (ii) BG18gH B cell transfer, MD39 NP immunization (N = 3 for day 8 GC analysis, N = 3 for day 14 ELISA); (iii) WT B cell transfer, N332-GT2 NP immunization (N = 5 for day 8 GC analysis, N = 5 for day 14 ELISA); (iv) WT B cell transfer, MD39 NP immunization (N = 3 for day 8 GC analysis, N = 3 for day 14 ELISA). We complied with all relevant ethical regulations. The animal studies were approved by the Institutional Animal Care and Use Committee of Massachusetts General Hospital.

Antigen specific single-cell sorting for BCR sequencing in mouse experiments

Antigen tetramers were prepared by conjugating for 1 hour (room temp.) biotinylated N332-GT2 and N332-GT2-KO trimers with fluorescently labeled streptavidins (Alexa Fluor 488, Alexa Fluor 647, eBioscience; Alexa Fluor 568, Thermo Fisher Scientific) in a 4:1 molar ratio. The same streptavidins conjugated with biotinylated Fab anti-IgM and biotinylated BSA were used as positive and negative staining controls, respectively (data not shown). Single-cell suspensions generated from spleen samples were depleted of red blood cells by ACK lysis, Fc blocked (BD Biosciences), and stained in FACS buffer (2% FCS/PBS) with antigen tetramers for 30 min at 4°C, 50 nM concentration. Next, a cocktail of mAbs was added for 30 min at 4°C. For staining of splenocytes from naïve mice (Fig. 2A), the cocktail was B220 PerCP-Cy5.5 (Clone RA3-6B2, Biolegend), IgD PE-Cy7 (Clone 11-26c.2a, Biolegend), CD4 APC-eFluor780 (Clone RM4-5, eBioscience), CD8a APC-eFluor780 (Clone 53-6.7, eBioscience), F4/80 APC-eFluor780 (Clone BM8, eBioscience), Ly-6G APC-eFluor780 (Clone RB6-8C5, eBioscience). For staining of splenocytes from immunized mice (fig. S13), the cocktail was CD38 Alexa Fluor 700 (Clone 90, Invitrogen), CD45.2 PE (Clone 104, Biolegend), CD45.1 PerCP-Cy5.5 (Clone A20, Biolegend), B220 PB (Clone RA3-6B2, Biolegend), CD95 PE-Cy7 (Clone Jo2, BD Bioscience), CD4 APC-eFluor780 (Clone RM4-5, eBioscience), CD8a APC-eFluor780 (Clone 53-6.7, eBioscience), F4/80 APC-eFluor780 (Clone BM8, eBioscience), Ly-6G APC-eFluor780 (Clone RB6-8C5, eBioscience). Live-Dead staining kits (Thermo Scientific) were used to identify dead cells for exclusion from the analysis. Data acquisition and single-cell sorting were performed on FACS ARIA II (BD Bioscience) and analyzed with FlowJo v. 10 (Tree Star). Single-cell sorting and single-cell PCR were carried out as described previously (25).

Phylogenetic analysis

We used Clustal Omega to create a multiple sequence alignment (input amino acid sequences from heavy chain PCR) and iTol (EMBL) to plot phylogeny trees: https://itol.embl.de/help/gkw290.pdf. The distance between nodes in the trees in Fig. 2I is a representation of dissimilarity (or evolutionary distance); the scale indicates the number of substitutions per site.

Analysis of Ab-antigen complexes for pathogens other than HIV

A broad but nonexhaustive set of antibody-antigen complexes for diverse major human pathogens were analyzed. If the HCDR3 was judged to play an important structural role in the interaction, then the HCDR3 sequence was analyzed for similarity to human germline D genes, checking all reading frames for all D genes listed at IMGT (50). We did not have access to nucleotide sequences for all Abs, so amino acid sequences were used for this analysis. If a recognizable D gene could be identified, then the complex was subjected to buried surface area (BSA) analysis. Antibody BSA analysis was carried out using PDBePISA (51). Glycan interfaces were not included in the BSA analysis. All complexes analyzed for buried surface area were included in fig. S18, except if the HCDR3 length was >20 amino acids and the Ab BSA analysis indicated that VH, VL, and HCDR3 all contributed substantially to the buried area on the Ab, in which case the complex was not considered a promising target and was not considered further. CDRs and FWs were specified according to IMGT convention (52). Somatic hypermutation (SHM) was determined by aligning the VH and VL genes to the IMGT human germline VH and VL genes and calculating the % difference in amino acid sequence from the most similar human germline gene. Insertions or deletions relative to the most similar human germline VH or VL gene were identified similarly.

Statistical analysis

Statistical parameters including the mean (or geometric mean), the SEM (or geomeric standard deviation), and in some cases the P value, are reported in the figures. Statistical analyses were performed using Prism (GraphPad Software) and compared with Student’s t test (Fig. 2, D to F) or chi-square test (Fig. 3D); P values < 0.05 were considered significant. Correspondences between the number of asterisks and the P values are stated in the figure legends.

Supplementary Materials

science.sciencemag.org/content/366/6470/eaax4380/suppl/DC1

Supplementary Text

Figs. S1 to S18

Tables S1 to S7

References (5362)

References and Notes

  1. See materials and methods.
Acknowledgments: We thank H. Gristick and P. Bjorkman for providing atomic coordinates of unliganded BG18 Fab in advance of publication (17) and C. Corbaci for graphical design assistance for the summary. Funding: This work was supported by the National Institute of Allergy and Infectious Diseases (NIAID) UM1 Al100663 (Scripps Center for HIV/AIDS Vaccine Immunology and Immunogen Discovery) and UM1 AI144462 (Scripps Consortium for HIV/AIDS Vaccine Development) (to W.R.S., F.D.B., S.C., A.B.W., and D.R.B) and NIAID R01 AI113867 (to W.R.S.); by the Ragon Institute of MGH, MIT, and Harvard (to F.D.B., W.R.S., and D.R.B.); by the International AIDS Vaccine Initiative (IAVI) Neutralizing Antibody Consortium (NAC) and Center (to W.R.S., A.B.W., I.A.W., and D.R.B.); and through the Collaboration for AIDS Vaccine Discovery funding for the IAVI NAC Center (to W.R.S., A.B.W., I.A.W., and D.R.B.). Author contributions: J.M.S. and W.R.S. conceived the study. J.M.S., Y.-C.L., C.H.-D., S.P., G.O., D.R.B., A.B.W., S.C., F.D.B., and W.R.S. designed the study. J.M.S., D.W.K., S.R., A.R., and W.R.S. designed immunogens. J.M.S. designed Abs. Y.-C.L., S.P., S.K., E.M., and F.D.B performed immunization studies. C.H.-D. and L.T. performed naïve B cell–sorting studies. G.O. and J.L.T. performed cryo-EM studies. B.B. performed NGS. J.R.W. performed bioinformatics analyses. D.S., E.L., and J.U. performed neutralization assays. A.L., O.K., and X.H. characterized immunogens and Abs. E.G., N.P., Y.A., and M.K. purified proteins. S.M.B. and I.A.W. contributed structural information. J.M.S. and W.R.S. wrote the manuscript. All co-authors edited the manuscript. Competing interests: S.P. is now employed by GSK Vaccines S.r.l., a company that might benefit indirectly from this research. D.R.B. is a paid consultant of IAVI. J.M.S. and W.R.S. are inventors on a patent application submitted by IAVI and The Scripps Research Institute that covers the N332-GT immunogens developed in this manuscript. Data and materials availability: Coordinates and maps for the structural data presented in this manuscript have been deposited to the Protein Data Bank under accession codes 6DFG, 6DFH, 6NF5, 6NFC, and 6OC7 and to the Electron Microscopy Data Bank under accession codes EMD-7875, EMD-7876, EMD-7884, and EMD-7885. Antibody sequences discovered during this study have been deposited to GenBank under accession numbers MN495018 to MN495471 (BG18gH mouse antibodies) and MN514889 to MN514945 (human naïve B cell antibodies binding N332-GT immunogens). Custom scripts for the NGS database query will be made available from W.R.S upon request. NGS sequencing data used in this manuscript and example analysis methods are available at https://github.com/SchiefLab/SteichenScience2019. All other data are available in the main text or supplementary materials.

Stay Connected to Science

Navigate This Article