Vaccination induces broadly neutralizing antibody precursors to HIV gp41

A key barrier to the development of vaccines that induce broadly neutralizing antibodies (bnAbs) against human immunodeficiency virus (HIV) and other viruses of high antigenic diversity is the design of priming immunogens that induce rare bnAb-precursor B cells. The high neutralization breadth of the HIV bnAb 10E8 makes elicitation of 10E8-class bnAbs desirable; however, the recessed epitope within gp41 makes envelope trimers poor priming immunogens and requires that 10E8-class bnAbs possess a long heavy chain complementarity determining region 3 (HCDR3) with a specific binding motif. We developed germline-targeting epitope scaffolds with affinity for 10E8-class precursors and engineered nanoparticles for multivalent display. Scaffolds exhibited epitope structural mimicry and bound bnAb-precursor human naive B cells in ex vivo screens, protein nanoparticles induced bnAb-precursor responses in stringent mouse models and rhesus macaques, and mRNA-encoded nanoparticles triggered similar responses in mice. Thus, germline-targeting epitope scaffold nanoparticles can elicit rare bnAb-precursor B cells with predefined binding specificities and HCDR3 features.


Vaccination induces broadly neutralizing antibody precursors to HIV gp41
A key barrier to the development of vaccines that induce broadly neutralizing antibodies (bnAbs) against human immunodeficiency virus (HIV) and other viruses of high antigenic diversity is the design of priming immunogens that induce rare bnAb-precursor B cells.The high neutralization breadth of the HIV bnAb 10E8 makes elicitation of 10E8-class bnAbs desirable; however, the recessed epitope within gp41 makes envelope trimers poor priming immunogens and requires that 10E8-class bnAbs possess a long heavy chain complementarity determining region 3 (HCDR3) with a specific binding motif.We developed germline-targeting epitope scaffolds with affinity for 10E8-class precursors and engineered nanoparticles for multivalent display.Scaffolds exhibited epitope structural mimicry and bound bnAb-precursor human naive B cells in ex vivo screens, protein nanoparticles induced bnAb-precursor responses in stringent mouse models and rhesus macaques, and mRNA-encoded nanoparticles triggered similar responses in mice.Thus, germline-targeting epitope scaffold nanoparticles can elicit rare bnAb-precursor B cells with predefined binding specificities and HCDR3 features.
Broad vaccine protection against highly antigenically diverse viruses, such as human immunodeficiency virus (HIV), hepatitis C virus, influenza or the family of betacoronaviruses, has not been achieved in humans but will likely require induction of broadly neutralizing antibodies (bnAbs) that bind to conserved epitopes on otherwise variable membrane glycoproteins.Monoclonal bnAbs for each of the above pathogens have been discovered, and specific genetic and structural features of each bnAb allow binding to its cognate epitope [1][2][3][4] .To use known bnAbs as guides for the design of vaccines that elicit similar responses, strategies to induce bnAbs with predefined genetic properties and binding specificities are needed [5][6][7] .One such strategy, germline-targeting vaccine design, is predicated on molecular design of the 'priming' immunogen to first elicit responses from rare bnAb-precursor B cells with genetic properties needed for bnAb development.Following the prime, sequential boosting with immunogens of increasing similarity to the native glycoprotein aims to guide B cell maturation to produce bnAbs targeting the desired epitope [8][9][10][11] .
Germline-targeting priming in humans was demonstrated for the eOD-GT8 60mer immunogen targeting precursors for VRC01-class bnAbs specific for the HIV envelope CD4-binding site 11 , which was an advance toward the goal of developing precision vaccines that elicit prespecified classes of bnAbs.However, in contrast to the V H -dominant binding mode of VRC01-class bnAbs, most bnAbs to HIV and other viruses exhibit heavy chain complementarity determining region 3 (HCDR3)-dominant interactions with antigen, making it critical to demonstrate induction of HCDR3-dominant bnAb precursors by germline-targeting priming immunogens 7 .An effective HIV vaccine will need to induce several different classes of bnAbs for sufficient coverage against global isolates.Induction of HCDR3-dominant bnAbs to the membrane-proximal external region (MPER) of the HIV-1 envelope protein (Env) might be crucial due to the high breadth of neutralization provided by such bnAbs (for example, approximately 92-98% for bnAbs 10E8 (ref.12), LN01 (ref.13) and DH511 (ref.14)), the relatively high epitope conservation that should reduce the potential of viral escape, and the strong protection by 10E8 in a passive nonhuman primate (NHP) immunization study despite relatively low potency against the challenge virus 15 .However, induction of MPER bnAbs faces challenges, including the recessed location of the MPER at the base of the Env trimer 12,16 , the need to induce antibodies with long HCDR3s bearing specific sequence motifs, and the lack of affinity of most MPER bnAb precursors for their peptide epitopes [17][18][19] .Furthermore, immune tolerance mechanisms block the induction of MPER bnAbs 2F5 and Article https://doi.org/10.1038/s41590-024-01833-w We prioritized one of these epitope scaffolds, T117v2 (ref.26), for further optimization due to its favorable thermal stability, solubility and presentation of surfaces adjacent to the MPER graft that could be engineered to increase contacts with the YxFW motif in the 10E8 HCDR3.T117v2 bound strongly to mature 10E8 (K d = 390 pM) but showed no binding (K d ≥ 100 µM) to 52 10E8-class precursors identified in the NGS database search above paired with the inferred germline (iGL) 10E8 light chain (hereafter NGS precursors; Supplementary Table 1).
We then performed a multistate design and selection process aimed at developing T117v2-based immunogens with the following features: 10 µM affinity or better for the 10E8 unmutated common ancestor (UCA) and as many NGS precursors as possible to enable priming of diverse 10E8-class precursors 7,11,33 ; an affinity gradient for 10E8-class antibodies with the highest affinity for mature 10E8 to favor affinity maturation toward mature 10E8 in vivo 5,7,11,34 ; multivalent display of epitope scaffolds on single-component self-assembling nanoparticles to facilitate mRNA lipid nanoparticle (mRNA-LNP) delivery, improve trafficking to lymph nodes 35 and increase B cell responses 5 ; and N-linked glycosylation sites added to the scaffold and base nanoparticle to reduce off-target responses 36 .Using a combination of structure-based design, computational modeling and directed evolution via yeast surface display 34 , we iteratively optimized binding of T117v2 to 10E8 iGL, UCA and NGS precursors, resulting in a series of immunogens that we refer to as 10E8-GT (Extended Data Fig. 1b-h and Methods).After nine rounds of optimization, 10E8-GT9.2bound with low affinity to a small subset (15%) of NGS precursors (geomean K d = 22 µM; Fig. 2a, Extended Data Fig. 1h, and Supplementary Tables 1 and 2).Further optimization of a pocket designed to contact germline D H 3-3-encoded residues at the tip of the 10E8 HCDR3 that are critical for 10E8 neutralization 12 produced 10E8-GT10.1 and 10E8-GT10.2.These bound to more NGS precursors (22% and 60%, respectively), with geomean K d values of 1.4 µM and 5.4 µM, respectively (Fig. 2a), but compared with T117v2 they bound weakly to mature 10E8 (K d = 27 nM and K d = 247 nM, respectively; Fig. 2a).Optimization of affinity for mature 10E8 generated 10E8-GT11 that had high affinity (K d = 1.4 nM) for mature 10E8 but low affinity (geomean K d = 12 µM) for a minority (6%) of NGS precursors (Fig. 2a,b and Extended Data Fig. 1).Finally, we simultaneously optimized the binding of mature 10E8 and NGS precursors to produce 10E8-GT12 (Extended Data Fig. 1).This final design engaged 46% of precursors with a geomean K d of 4.3 µM and bound strongly (K d = 1.0 nM) to mature 10E8 (Fig. 2a,b).
We multimerized 10E8-GT scaffolds by fusion to self-assembling nanoparticles from hyperthermophilic bacteria (Extended Data Fig. 1c-e,i).10E8-GT10.212mer and 10E8-GT12 12mer, based on fusion to a glycan-shielded variant of the dodecameric 3-dehydroquinase from Thermus thermophilus, and 10E8-GT12 24mer, created by fusing 10E8-GT12 to each terminus of the 3-dehydroquinase nanoparticle protomer, were expressed with high yield (Extended Data Fig. 1c-e,i).We also added N-linked glycosylation sites to scaffold surfaces outside the MPER graft to reduce off-target responses 36 .Site-specific glycosylation analysis by mass spectrometry indicated that approximately 50% of glycosylation sites were occupied (Extended Data Fig. 2).We thus developed self-assembling nanoparticles presenting 10E8-GT scaffolds with broad affinity for 10E8-class precursors.

Scaffold-antibody interactions mimic epitope and the HCDR3 motif
To assess the structural mimicry of the MPER helix within the epitope scaffold and the interaction of the epitope scaffold with the 10E8-class D gene YxFW motif, we determined three crystal structures and one cryoelectron microscopy (cryo-EM) structure of 10E8-GT epitope scaffolds complexed with 10E8-class human antibodies (Fig. 3 and Extended Data Fig. 3).These included a 2.62-Å-resolution crystal structure of the early-stage design 10E8-GT4 bound to a variant of 10E8 iGL bearing the mature 10E8 HCDR3 (10E8-iGL1; complex 1), a 4.0-Å 4E10, potentially due to lipid reactivity 20 , raising concerns that other more potent MPER bnAbs, such as 10E8, might also face tolerance barriers [21][22][23][24] .Here, we developed and validated germline-targeting epitope scaffold nanoparticle priming immunogens to induce 10E8-class HCDR3-dominant bnAb-precursor responses.These immunogens represent candidates for human vaccination and demonstrate design and evaluation processes that could be applied to other bnAb targets.

10E8-class naive precursors are present in most humans
Structural 12,16,25,26 and mutational 12,18 data indicate that 10E8 binds to its MPER helical peptide epitope primarily through a germline D H 3-3-encoded binding motif YxFW positioned near the tip of a long (22-amino acid (aa)) HCDR3, required to access the sterically occluded epitope at the base of full-length membrane-bound Env (Fig. 1a).The activity of 10E8 bnAb further requires a PP motif in the junction between D and J genes within the HCDR3, which could have arisen either during V(D)J recombination or somatic hypermutation (SHM), and germline-encoded HCDR1 and framework region 2 residues and somatically mutated HCDR2 residues within the gene encoding V H 3-15.We therefore defined 10E8-class heavy chain precursors as heavy chains with a V H gene closely related to V H 3-15 and an HCDR3 length of 21-24 aa with a YxFW motif at the equivalent position within the HCDR3 as 10E8 (Extended Data Fig. 1a).This definition allowed for diverse V-D and D-J junctions and did not require the PP motif that can arise during SHM.To determine if heavy chains with these properties were present in humans, we searched an ultradeep next-generation sequencing (NGS) dataset of primarily naive IgM heavy chains from 14 HIV-seronegative donors 7,27 .Heavy chains matching the 10E8-class properties were found in all donors, with a geomean frequency of 1:68,000 (Fig. 1b).
The 10E8 light chain contributes to binding of membrane-associated Env by contacting the virion lipid membrane and conformationally stabilizing HCDR3 (ref.26).The range of germline light chains that have the potential to acquire mutations to mediate such contacts is unclear but could be large.In two paired heavy chain-light chain datasets 28,29 , human light chains within the V L 3 family used by 10E8 were paired with V H 3-15 heavy chains at a frequency of approximately 1:7.5, suggesting that the frequency of 10E8-class heavy chain-light chain precursors was approximately 1:510,000.Thus, 10E8-class precursors are present in healthy humans at a substantial frequency.

Germline-targeting immunogens bind 10E8-class precursors
Because the MPER region is sterically occluded at the base of full-length membrane-bound Env and absent from most soluble native-like trimers 30 , epitope scaffold immunogens were previously designed to conformationally stabilize and expose the C-terminal MPER helix 26,31,32 (Fig. 1a).

Article
https://doi.org/10.1038/s41590-024-01833-wcryo-EM structure of 10E8-GT10.2 in complex with mature 10E8 and a scaffold-specific 'off-target' monoclonal antibody (mAb) to help image processing (complex 2), a 2.7-Å-resolution structure of nonglycosylated 10E8-GT10.1 in complex with the NGS precursor 10E8-NGS-03 (complex 3), and a 1.9-Å-resolution structure of 10E8-GT11 bound to 10E8-iGL1 (complex 4; Fig. 3 and Extended Data Fig. 3).In all four complexes, the overall structures of the epitope scaffold and the MPER helix were similar to the original T117v2 scaffold complexed with 10E8, with backbone root mean square deviation (bb-r.m.s.d.) values of 0.73, 0.89, 0.75 and 0.62 Å for the 10E8-GT antigens, respectively (Fig. 3).In all except complex 3, the antibody engaged the epitope scaffold at an angle closely resembling the interaction between mature 10E8 and the MPER peptide (Fig. 3), and the D gene YxFW motif interacted with the engineered D gene binding pocket and adopted a conformation similar to that of mature 10E8 bound to MPER peptide, with all-atom r.m.s.d.values computed over YxFW of 0.5, 0.47 and 0.32 Å for complexes 1, 2 and 4, respectively (Fig. 3).In complex 3, the FW portion of the YxFW motif also accurately mimicked the interaction between mature 10E8 and MPER peptide (all-atom r.m.s.d.= 1.04 Å over FW), but the Yx portion of the motif was divergent, owing to the antibody approaching the MPER from an angle differing from mature 10E8 by ~46°, potentially due to a different conformation of the immature HCDR3 (Fig. 3).In conclusion, despite variability in the angle of binding of diverse precursors, the 10E8 scaffolds stabilized the MPER in the 10E8-bound conformation and consistently engaged the hydrophobic tip of 10E8-class HCDR3s in a manner closely resembling the interaction of mature 10E8 with gp41.
We also searched 10E8-GT-binding sequences for signatures of other MPER bnAb lineages, including precursors of LN01-class MPER bnAbs, which are genetically and structurally distinct from 10E8-class bnAbs but share key features such as an D H 3-3-encoded 'FW' motif at the tip of a long (20-aa) HCDR3 (ref.13).Although immunogens were optimized for engagement of 10E8-class precursors, we detected LN01-class HCDR3s in all six samples of 10E8-GT12-sorted naive IgM + B cells, with a median frequency of 3.3% (Extended Data Fig. 4g,h).Accounting for HCDR3 properties, the V H 4-39 gene of LN01 and the frequency of epitope-specific B cells, the total frequency of 10E8-GT12-binding LN01-class IgH precursors among naive B cells was 1:41,000, which was only slightly lower than the frequency of 10E8-like IgH precursors (1:15,000; Fig. 4g).

10E8-class B cells function in vivo
To evaluate 10E8-class B cell development and activation in vivo, we created IgH knock-in (KI) mice using one of the highest-affinity human 10E8-class naive precursors, MPER HuGL18, identified through sorting human naive B cells that bound to 10E8-GT9.2(Supplementary Table 1).B cell development in the bone marrow of IgH MPER-HuGL-18/WT IgK WT/WT mice (hereafter MPER-HuGL18 H mice) was similar to that in wild-type mice (Extended Data Fig. 6a), and frequencies of Live/Dead − TCRβ − B220 + B cells, other B cell subpopulations and T cells among splenic lymphocytes in heterozygous MPER-HuGL18 H mice were comparable to those in wild-type mice (Fig. 5a).These data indicated normal B cell development in MPER-HuGL18 H mice, in contrast to previously developed MPER bnAb IgH or IgHK KI mice [22][23][24] .

10E8-GT 12mer induces 10E8-class BCRs in rhesus macaques
Two of five known homologs of human D H 3-3 in Indian rhesus macaques (D H 3-41*01_S8240 and D H 3-41*01_S4389) encode the YxFW motif, whereas the remaining alleles encode YxIW (Fig. 7a) 41 , permitting testing of 10E8-GT immunogens in some rhesus macaques.BCR sequencing of sorted 10E8-GT10.2epitope-specific naive CD20 + IgG − IgD + B cells from unimmunized rhesus macaques (Extended Data Fig. 9a) indicated that sorted BCRs were enriched for long HCDR3s (Fig. 7b).BCRs with 10E8-class HCDR3s (length of 21-24 aa with a YxFW motif at the equivalent position within the HCDR3 as 10E8) were detected in eight of nine macaques with a median frequency of 0.0078% among naive B cells (Fig. 7c), 18-fold lower than their frequency of 0.14% in the human naive B cell repertoire, which made rhesus macaques a viable, although challenging, model to assess 10E8-GT immunogens.

Induced BCRs acquire affinity for a boosting candidate
Germline-targeting priming immunogens should consistently induce bnAb-precursor memory and/or GC B cells susceptible to boosting by immunogens more similar to the native viral protein (native-like) than the priming immunogen 11 .10E8-class bnAb precursors induced by 10E8-GT nanoparticles in either hD3-3/J H 6 mice or NHPs had no neutralizing activity against HIV pseudoviruses (Supplementary Table 6), which was expected because the 10E8 epitope on the priming immunogens was substantially modified from wild-type and lacked steric constraints imposed by the membrane and ectodomain of HIV Env.To determine whether 10E8-GT nanoparticle immunization selected for 10E8-class BCRs that bound to more native-like immunogens, we tested binding of post-prime antibodies to an epitope scaffold (10E8-B1) with a 10E8 peptide epitope that was fully native, except for one mutation required for solubility (W680N, Hxb2 numbering; Extended Data Fig. 1h).10E8-B1 had no detectable affinity for early 10E8 lineage members but had increasing affinity for artificial intermediate 10E8 lineage members with three or more mutations (Supplementary Table 1) and ultrahigh affinity (K d = 80 pM) for mature 10E8 (Fig. 8a).
A 2.63-Å-resolution crystal structure of 10E8-B1 in complex with mature 10E8 bnAb showed that the structure of 10E8-B1 agreed well with the original T117 scaffold (0.61 Å bb-r.m.s.d.) and that the interaction between 10E8-B1 and 10E8 closely resembled the interaction between 10E8 and MPER peptide (Fig. 8b and Extended Data Fig. 3), suggesting that 10E8-B1 has appropriate antigenicity and structure to probe for 10E8-like maturation in 10E8-GT-induced antibodies.10E8-B1 bound to 10% of 10E8-class antibodies induced by 10E8-GT nanoparticles, including 10E8-GT10 12mer or 10E8-GT12 24mer in hD3-3/J H 6 mice (Fig. 8c and Supplementary Table 2), and to 25% of 10E8-class antibodies primed by 10E8-GT10 12mer in macaques (Fig. 8c).Binding of 10E8-B1 to germline-reverted 10E8-class antibodies from macaques was significantly weaker than to the matching antibodies primed by 10E8-GT10 12mer in macaques (Fig. 8c), indicating that binding by 10E8-B1 was due to SHM acquired by these antibodies.Thus, 10E8-GT nanoparticle immunization selects for affinity maturation that confers affinity for an antigen with a more native-like 10E8 epitope.

Discussion
By combining germline targeting with epitope scaffolding and nanoparticle design, we developed immunogens that consistently induced 10E8-class HIV bnAb precursors with bnAb-associated genetic and structural features, including long HCDR3s with specific binding motifs that confer the potential to develop into bnAbs, in rhesus macaques and two mouse models.We also showed that the priming immunogens selected for productive directional affinity maturation, such that at least a subset of the induced bnAb precursors had affinity for a more native-like antigen.These findings provide proof of principle that epitope scaffolds can be designed to induce responses from rare, HCDR3-dominant bnAb precursors and select for a degree of favorable maturation in those precursors, extending the functionality of the epitope scaffold approach 31,[44][45][46][47] .
Development of B cells expressing precursors for MPER bnAbs 2F5 and 4E10 was reported to be blocked by tolerance barriers [21][22][23][24] .We found normal B cell development for 10E8-class precursors.The 10E8-GT epitope scaffolds also induced precursors for a related yet genetically distinct class of bnAb, LN01, demonstrating the capacity for multi-bnAb precursor priming without obvious interference from tolerance mechanisms, consistent with the low poly-or autoreactivity exhibited by 10E8-and LN01-class lineages 12,13 .
Our observations validated 10E8-GT nanoparticle germlinetargeting priming immunogens for consistent induction of diverse bnAb precursors across vaccinated mice and NHPs.Germline-targeting vaccine design posits that bnAbs can be elicited by first priming bnAb precursors with the necessary bnAb-associated genetic and structural features and then using a series of boosters of increasing similarity to the native glycoprotein to select for the necessary SHM to produce bnAbs.Hence, additional work is needed to develop sequential heterologous boosting regimens to induce 10E8-class bnAbs.We envision boosting immunogens to include an epitope scaffold nanoparticle with a more native-like MPER epitope such as 10E8-B1, followed by one or more membrane-bound envelope protein(s) to select for maturation Based on these data, we confirm the MPER as an attractive HIV vaccine target and propose 10E8-GT nanoparticles as MPER vaccine priming immunogens.Our epitope scaffolds bound to and isolated human naive bnAb precursors from human PBMCs, suggesting that the positive immunization data from mice and macaques have the potential for translation to humans.Our finding that 10E8-GT12 24mer delivered by mRNA-LNP induces similar 10E8-class B cell responses as SMNP-adjuvanted protein immunization in a stringent mouse model supports the potential for rapid clinical testing.The data further encourage the development of germline-targeting epitope scaffold nanoparticles to induce bnAb precursors and initiate bnAb induction for other epitopes that are sterically occluded or poorly immunogenic in the context of native viral glycoproteins, such as the MPER of Filoviridae 48 , the influenza A hemagglutinin anchor 3 or the relatively conserved S2 subunit in betacoronaviruses 4 .

Methods
This work complies with all relevant ethical regulations.Animal experiments were approved by the Institutional Animal Care and Use Committees (IACUCs) of Harvard University, Massachusetts General Hospital (MGH), Alpha Genesis and Emory University.Experiments involving human samples were approved by the La Jolla Institute for Immunology (LJI) Institutional Review Board.

Human immunoglobulin repertoire bioinformatic analysis
MPER bnAb-precursor frequencies were estimated from publicly available NGS data of ~1.1 × 10 9 heavy chain sequences from 14 HIV-seronegative donors, as previously described 7,51

Immunogen design
Visual inspection of Protein Data Bank (PDB) IDs 4G6F and 5T85 suggested a clash between K52 of 10E8-iGL1 (Kabat numbering) and F673 of the MPER (HxB2 numbering, F121 of T117).Computational modeling using the Rosetta Software suite 52,53 resulted in 10E8-GT2 that resolved this clash and bound weakly (1.5 µM) to 10E8-iGL1.Further Rosetta fixbb design yielded GT3 and GT4, with improving affinities for 10E8-iGL1.Next, directed evolution by yeast surface display 34,54 was used to select for variants capable of binding to increasingly more germline-reverted variants of 10E8.Selection with 10E8-iGL1 led to 10E8-GT5 that bound to 10E8-iGL1 with 81 nM affinity.Next, we divided the epitope on 10E8-GT5 into surface patches of three or four residues and performed 'combinatorial NNK patch scanning' by yeast display.In contrast to traditional deep scanning mutagenesis, which only analyzes the effect of single point mutations, these combinatorial NNK patches contained all combinations of all 21 aa (including the stop codon) at the respective positions, thereby allowing potential compensatory mutations to occur.By analyzing four different patches in parallel (residues 71-74, 93-96, 111-114 and 114-115 + 117-118 of T117, respectively), most contact residues were optimized.Each combinatorial NNK patch was enriched against antibodies 10E8-iGL1, 10E8-iGL2 and 10E8-iGL3.The best overall resulting variant, 10E8-GT7, resulted from patch 3 and bound 10E8-iGL1 with 8.2 nM affinity.
One design goal was the multimerization of the immunogen into self-assembling nanoparticles displaying the germline-targeting epitope scaffold.However, the T117v2 scaffold was not well suited for this goal because both N and C termini of the scaffold are near the epitope; hence, genetic fusion of the epitope scaffold to a self-assembling protein would result in poor exposure of the epitope on the nanoparticle.With generation 10E8-GT8, we switched from T117v2 to T298, a previously described circularly permuted variant of T117 with N and C termini opposite the epitope 32 .However, the original T298 suffered from low expression levels and relatively low thermal stability (48 °C), and it dimerized in solution 32 .Resurfacing of T298 using the dTERMen algorithm 55 was used to improve stability and solubility of the scaffold.We genetically fused a circularly permuted immunogen, 10E8-GT8, to several self-assembling nanoparticle platforms and managed to purify high yields (~24 mg l -1 ) of fully assembled particles after fusion to a glycosylated variant of the dodecameric 3-dehydroquinase from T. thermophilus.
Another key design goal was the introduction of N-linked glycosylation sites onto the scaffold to focus B cell responses on the MPER epitope.We initially explored the introduction of single artificial N-linked glycosylation sites into irrelevant surfaces of the epitope scaffold of 10E8-GT8.1.Sites that decreased affinity for 10E8-iGL2 by no more than 1.3-fold and decreased expression yields by no more than 40% were selected for further investigation.We next tested combinations of multiple glycosylation sites on 10E8-GT8.1 and obtained 10E8-GT8.2with four N-linked glycosylation sites.We further increased the number of N-linked glycosylation sites with each subsequent generation of immunogens.
Although 10E8-GT8.2showed strong binding to 10E8-iGL1 (K d = 13 nM), it did not bind the 10E8 UCA nor any of the 55 NGS-derived precursors tested, all of which differ from 10E8-iGL1 in their respective HCDR3s but are otherwise identical (Supplementary Tables 1 and 2).The key residues of 10E8 are the D H -encoded residues at the tip of the HCDR3, which were present in all 10E8-class precursors used in this study.We hypothesized that additional interactions of the scaffold with these key residues might increase the affinity and breadth for precursors that share these features.At this stage, we had not yet taken full advantage of the combinatorial NNK patch scanning yeast display optimization of 10E8-GT5 performed earlier, because 10E8-GT7 and 10E8-GT8 only contained mutations from patch 3, ignoring results from the other patches.Patch 2 had analyzed residues 93-96 (equivalent to 75-78 in circularly permutated T298), which form a loop of the scaffold in close proximity to the D gene-encoded 'FW' motif of bound 10E8.Enrichment of this library with 10E8-iGL3 had resulted in complete redesign of the 75-78 loop with the final sequence 'GWYQ', which we hypothesized to form a pocket that would result in additional beneficial contacts with the D H -encoded key residues.
We combined this new D H binding loop into a combinatorial yeast library with other promising residues from the remaining combinatorial NNK patch screenings and enriched for binding to 10E8-class precursors.Enrichment with 10E8 UCA resulted in 10E8-GT9.1 that bound to the 10E8 UCA and several NGS precursors.Reversion of each MPER mutation revealed that mutation N132I (N677I in Hxb2 numbering) had no impact on binding affinity; hence, this mutation was subsequently removed.To improve expression and solubility, we manually inspected models of T298-GT9.1 and reverted exposed hydrophobic residues to the respective amino acids in the original T298 before resurfacing.By combining these changes with further refinement of the glycan shield, we obtained 10E8-GT9.2,which bound weakly to the 10E8 UCA and several NGS precursors (21 of 33 tested precursors; Fig. 2a).Further optimization of the D H gene binding pocket by yeast display of an error-prone PCR library (Genemorph II Random Mutagenesis kit, Agilent) added mutation P41Q, resulting in 10E8-GT10.1,which bound weakly to 28 of 33 10E8-like precursors and had modest affinity for the 10E8 UCA (K d of ~600 nM).However, nanoparticle constructs presenting 10E8-GT10.1,created by fusion to several self-assembling proteins, failed to express in 293F cells.We therefore further refined the scaffold for increased expression by removing exposed hydrophobic patches, adding additional glycosylation sites and further improving the D binding pocket by incorporating mutation V42A, discovered from yeast display of an error-prone PCR library of GT10.1 that was simultaneously enriched with 10E8 UCA Fab (directly conjugated to Alexa Fluor 647 (AF647)) and NGS-57 (stained with PE-conjugated anti-human Fcγ secondary antibody; Jackson ImmunoResearch).Fusion of the resulting protein, termed 10E8-GT10.2,to the glycosylated 3-dehydroquinase from T. thermophilus via a linker that incorporated the PADRE 56 epitope yielded homogenous particles of the expected molecular weight, termed 10E8-GT10.212mer (of note, unexpectedly, the addition of PADRE to the linker in the GT10.2 12mer increased expression levels substantially).
Although 10E8-GT9 and 10E8-GT10 immunogens bound to several NGS precursors, they bound much more weakly to mature and artificial intermediate 10E8 lineage members than to previous immunogen generations.We therefore transferred the MPER of 10E8-GT8, the most https://doi.org/10.1038/s41590-024-01833-wadvanced version that retained strong binding to mature and intermediate 10E8 variants, onto 10E8-GT10.2,moved an N-linked glycosylation site into the scaffold surface patch engaged by 10E8-NGS-03 and added mutation W680N to the MPER graft.The resulting construct, termed 10E8-GT11, bound with high affinity to mature 10E8 (K d = 1.4 nM) but interacted only weakly with 10E8-class precursors.Additional optimization by yeast display was performed by identifying beneficial mutations using an error-prone PCR library, followed by screening of a combinatorial library that combined identified mutations using NGS precursors.The resulting epitope scaffold, termed 10E8-GT12, engaged 46% of 10E8-class precursors tested with affinities comparable to GT10 while binding strongly (K d = 1.0 nM) to mature 10E8 (Fig. 2).We fused 10E8-GT12 to the same PADRE-containing 12mer nanoparticle platform described above and obtained well-formed particles, termed 10E8-GT12 12mer (Extended Data Fig. 1d).We also created 10E8-GT12 24mer by fusing 10E8-GT12 to each terminus of the 3-dehydroquinase nanoparticle protomers.In the 10E8-GT12 24mer, rather than using linkers with PADRE, we included exogeneous T-help peptides derived from Aquifex aeolicus lumazine synthase that were found to be broadly immunogenic in humans 57 .
We also developed antigens with more native 10E8 epitope grafts to serve as candidate boost immunogens to follow the prime and to serve as tools to probe the maturation of 10E8-class antibodies induced by the prime.To reduce binding of irrelevant antibodies, we resurfaced T298 using the dTERMen algorithm 55 , and we eliminated remaining hydrophobic surface patches manually.Similar to the original T117 (ref.31) and T298 (ref.32) scaffolds, many designs formed dimers or aggregates in solution.Inspection of the previously published crystal structure of T298 (PDB ID 3T43) led to inclusion of bulky residues at positions 55 (methionine) and 77 (N-linked glycosylation site) to disrupt dimer formation without altering 10E8 bnAb binding.We also added the D binding pocket from GT12 to maintain the critical interaction with the FW motif within the HCDR3s of 10E8-class antibodies.We removed N-linked glycosylation sequons that we had found to be unoccupied in 10E8-GT12.We fused a series of such candidate epitope scaffolds to the same glycosylated 3-dehydroquinase nanoparticle described above, either at the nanoparticle C terminus (12mer) or at both N and C termini (24mer).Incorporation of a consensus 10E8 epitope graft led to aggregation, but we obtained homogenous nanoparticles by including a single germline-targeting mutation (W104N; W680N in Hxb2 numbering).This nanoparticle was termed 10E8-B1 24mer, and the corresponding monomeric epitope scaffold was termed 10E8-B1.

Protein expression, purification, biotinylation and biochemical characterization
Genes of proteins and antibodies were synthesized and cloned into pHLSec or its variant pCWSec by Genscript using codons optimized for expression in human cells.Proteins were expressed and purified as described in detail previously 51 .Briefly, plasmids were transfected into FreeStyle 293F cells (Thermo Scientific), and expression was performed in protein-free chemically defined FreeStyle medium (Thermo Scientific).His-tagged proteins were purified from clarified supernatants using immobilized metal affinity chromatography followed by size-exclusion chromatography (SEC), nanoparticles were purified using Galanthus nivalis lectin affinity chromatography (Vector Laboratories) followed by SEC, and antibodies were purified by protein A affinity chromatography followed by buffer exchange into Tris-buffered saline.High-throughput expression of antibodies was performed in 96-well plates using the ExpiCHO system (Thermo Scientific) and purified by protein A affinity purification as previously described 38 .Sorting probes were expressed with a C-terminal AviHis tag (GSGGSGLNDIFEAQKIEWHEGSGGHHHHHH**, where '*' denotes a stop codon) and purified by metal affinity chromatography and SEC, as described above.Matching KO probes for each immunogen were generated that incorporated five KO mutations (672A, 673R, 675R, 680E and 683D; Hxb2 numbering) in the MPER.Purified proteins were biotinylated by BirA enzymatic reaction (Avidity) according to the manufacturer's protocol and purified by SEC.Immunogen candidates were characterized by SEC coupled with multiangle light scattering on a Dawn 18 instrument (Wyatt Labs) and Optilab dRI detector using ASTRA 7.1.1.3software, as previously described 51 .Protein stability was determined by dynamic scanning calorimetry (DSC) on a MicroCal VP-Capillary DSC (Malvern Instruments) as described previously 51 .

SPR
All K d values for antibody-antigen interactions presented in main text figures were measured on a Carterra LSA instrument using HC30M or CMDP sensor chips (Carterra) and 1× HBS-EP+ (pH 7.4) running buffer (20× stock from Teknova, H8022) supplemented with bovine serum albumin at 1 mg ml -1 .Carterra Navigator software instructions were followed to prepare chip surfaces for ligand capture.In a typical experiment, approximately 2,500 to 2,700 RU of capture antibody (SouthernBiotech, 2047-01) at 25 µg ml -1 in 10 mM sodium acetate (pH 4.5) was amine coupled using a commercial Amine Coupling kit (GE, BR-1000-50) but using tenfold diluted NHS and EDC concentrations.Regeneration solution was 1.7% phosphoric acid injected three times for 60 s per each cycle.The solution concentration of ligands was around 1 µg ml -1 , and contact time was 5 min.Raw sensograms were analyzed using Carterra Kinetics software (Carterra), interspot and blank double referencing, Langmuir model.For fast off-rates (>0.009 s -1 ), we used automated batch referencing that included overlay y aline and higher analyte concentrations.For slow off-rates (≤0.009 s -1 ), we used manual process referencing that included serial y aline and lower analyte concentrations.After automated data analysis by Kinetics software, a custom R script was used to remove datasets with maximum response signals smaller than signals from negative controls.Some of the K d values in supplementary tables were determined on a ProteOn XPR36 (Bio-Rad) using a GLC Sensor Chip (Bio-Rad) and ProteOn Manager software or Biacore 4000 with CM5 Series S Sensor Chips, as described previously 51 .The same analyte-ligand pair would produce similar K d values on all systems tested within a factor of two.

Site-specific glycosylation profiling
Two methods were used to analyze glycosylation profiles: single-site glycan profiling and DeGlyPHER.
Single-site glycan profiling was performed as described previously 58 .Briefly, proteins were denatured, reduced and alkylated, followed by enzymatic digestion using trypsin, chymotrypsin or α-lytic protease.Peptides were analyzed by nanoLC-ESI mass spectrometry with an UltiMate 3000 HPLC (Thermo Fisher Scientific) system coupled to an Orbitrap Eclipse mass spectrometer (Thermo Fisher Scientific).Peptides were separated using an EASY-Spray PepMap RSLC C18 column (75 µm × 75 cm) with an in-line trapping column (PepMap 100 C18 3 µM, 75 µM × 2 cm).Data were analyzed using protein metrics Byos software (version 3.5).The relative amounts of each glycan at each site as well as the unoccupied proportion were determined by comparing the extracted ion chromatographic areas for different glycopeptides to an identical peptide sequence.Glycans were categorized according to the composition detected.
DeGlyPHER was performed as described previously 59 .Briefly, proteins were deglycosylated with Endo H, digested with proteinase K and deglycosylated again with Endo H, followed by lyophilization and resupension in PNGase F-containing H 2   18   O.Samples were analyzed on a Q Exactive HF-X mass spectrometer.Protein and peptide identification were performed using the Integrated Proteomics Pipeline (IP2) using the automated GlycoMSQuant (Baboo et al. 59 ) implementation.GlycoMSQuant summed precursor peak areas across replicates, discarded peptides without NGS, discarded misidentified peptides when N-glycan remnant mass modifications were localized to non-NGS asparagines and corrected/fixed N-glycan mislocalization where appropriate. https://doi.org/10.1038/s41590-024-01833-w

Structure determination by cryo-EM
10E8-GT10.2(120 µg) was incubated with the on-target mature 10E8 Fab (300 µg) and an off-target W6-10 Fab (300 µg) in an equal molar ratio (1:1:1) overnight at room temperature.The complex was then purified over a Superdex 200 Increase column (GE Healthcare) and concentrated to 2.5 mg ml -1 .Next, 3 µl of the complex was mixed with 0.5 µl of 35 µM lauryl maltose neopentyl glycol (Anatrace; final concentration of 5 µM) before deposition onto 1.2/1.3UltrAuFoil 200 grids (EMS; glow-discharged for 10 s), directly preceding vitrification using a Vitrobot Mark IV (Thermo Fisher Scientific) with the following settings: 4 °C, 100% humidity, 10-s wait time, 6-s blot time and blot force of 2. Once the sample was deposited, the grids were blotted and plunged into liquid ethane to immobilize the particles in vitreous ice.Movie frames were collected using EPU image acquisition software (Thermo Fisher Scientific) at a nominal magnification of ×190,000 with a Thermo Fisher Scientific Falcon 4 detector mounted on a Thermo Fisher Scientific Glacios operating at 200 kV.Counting mode was used, with a total exposure dose of 53 e -Å -2 .In total, 4,249 micrographs were motion, dose and CTF corrected using cryoSPARC Live imported into cryoSPARC 65 (Extended Data Fig. 3).Template Picker was used to pick 956,668 particles, which were then extracted and two-dimensionally classified.The particles in selected two-dimensional classes were further filtered by ab initio reconstruction using C1 symmetry, resulting in 56,628 particles subjected to nonuniform refinement.The final reconstruction was estimated at ~4.0-Å resolution using Fourier shell correlation and a 0.143 cutoff (Extended Data Fig. 3).
For model building, an initial model of 10E8 Fab and MPER scaffold was generated using PDB 5T85 and docked into the cryo-EM map using UCSF ChimeraX 66 .Coot 0.9.8 (ref.67), Phenix 68 and Rosetta 52,53 were used for model building and refinement (Extended Data Fig. 3).The final model and map have been deposited in the PDB and Electron Microscopy Data Bank under accession codes 8SX3 and EMD-40825, respectively.

Human B cell repertoire screening and sorting
Leukoreduction (LRS) tubes from healthy donor samples were obtained from the San Diego Blood Bank from consenting participants, in accordance with protocols approved by the LJI Institutional Review Board.PBMCs were isolated from blood by the LJI Blood Processing Core and were frozen and stored in liquid nitrogen until analysis.
Cryopreserved PBMCs were thawed and recovered in RPMI medium containing 10% fetal bovine serum (FBS) supplemented with 1× penicillin/streptomycin (pen/strep) and 1× GlutaMAX (R10).Fluorescently labeled antigen probes were prepared by mixing fluorophore-conjugated streptavidin with small volumes of biotinylated antigen probes in 1× PBS at room temperature, with additions every 15 to 20 min for a total of 45 min to 1 h depending on the human naive B cell screening experiment.
For 10E8-GT9.2 and 10E8-GT10.1 human naive B cell screening using direct lysis sorts and single-cell BCR amplification, cells were first stained with either 10E8-GT9.2-KOor 10E8-GT10.1-KOprobe and then the respective wild-type probes for a total of 45 min.Without washing, the cells were then stained with the antibody master mix for an additional 30 min.Cells were then washed twice in R10 and sorted on a FACSAria II (BD Biosciences).
For the 10E8-GT10.1 human naive B cell screening using 10x Genomics single-cell BCR sequencing, anti-AF647 antigen-specific B cell enrichment was performed.Cells were first incubated with 10E8-GT10.1-AF647probe in R10 medium for 60 min at room temperature and then washed with 1% bovine serum albumin in PBS.The cells were then incubated with anti-Cy5/anti-AF647 microbeads for isolation of 10E8-GT10.1-AF647probe-binding cells, following the user guide provided (Miltenyi Biotec).The purified cells were counted and stained with a mix of tetramer probes (the other 10E8-GT10.1 probe and a 10E8-GT10.1-KOprobe).Without washing, the antibody master mix was added to the cells for another 30 min.Anti-human TotalSeq-C hashtag antibodies (BioLegend) were also added at this time at a concentration of 0.1 µg per 1 million cells.Cells were then washed twice in R10 before sorting on a FACSAria II (BD Biosciences).
For 10E8-GT12 human naive B cell screening, total B cells were enriched by negative selection using an EasySep human B cell isolation kit (StemCell).Purified B cells were then counted and incubated with the fluorescently labeled antigen probes.First, 10E8-GT12-KO probe was added for 15 min at 4 °C and then incubated with wild-type 10E8-GT12 probes for an additional 15 min.Without washing, Fc Block (BD Biosciences) was added for 5 min and stained with surface antibodies for an additional 30 min at 4 °C.Cells were washed twice with FACS buffer (PBS + 2% FBS + 1 mM EDTA) and sorted on a FACSymphony S6 (BD Biosciences).

Human BCR sequencing using 10x Genomics
Sorted cells were prepared for 10x single-cell V(D)J sequencing similar to previously published protocols 70 .For the 10E8-GT10.1 human naive B cell screening, single indexed V(D)J and Feature Barcode libraries were generated following the user guide for the Chromium Single Cell V(D) J Reagent kits with Feature Barcoding technology (Legacy version, 10x Genomics).All libraries were pooled, and sequencing was performed on a NovaSeq Sequencer (Illumina).The V(D)J contigs were assembled and annotated using Cell Ranger v3.0.2 using an immunoglobulin library compiled from IMGT references.The constants.pyfile was modified to increase the maximum CDR3 length to 110 nucleotides.A Python script was used to associate hashtag read counts with the productive assembled V(D)J sequences, compiling the data into a tabular format 70 .For 10E8-GT12 human naive B cell screening, V(D)J libraries were prepared using the Dual Indexed 10x Genomics V(D)J 5′ v.2 according to the manufacturer's protocol (10x Genomics).Raw sequencing data were processed using Cell Ranger v6.1.2.V(D)J contigs were generated and aligned to the prebuilt human reference (refdata-cellranger-vdj-GRCh3 8-alts-ensembl-7.0.0).V(D)J output was further processed following the Immcantation pipeline 71 .Briefly, contigs were annotated using IgBlast on the IMGT database, and only productive sequences were kept for downstream analysis.Heavy and light chain contigs were paired, and cells with more than one heavy chain sequence were removed.Final analysis was performed using Sequencing Analysis and Data library for Immunoinformatics Exploration (SADIE), and sequences were filtered for IgM + sequences based on the 'c_call_heavy' field.

HEp-2 cell staining assay
The HEp-2 cell staining assay was performed using kits purchased from Aesku Diagnostics, according to the manufacturer's instructions.These Aesku slides use optimally fixed human epithelial (HEp-2) cells (ATCC) as a substrate and affinity-purified, fluorescein isothiocyanate (FITC)-conjugated goat anti-human IgG for detection.Briefly, 2.5 µg or 25 µl of 100 µg ml -1 mAb and controls were added to wells and incubated on HEp-2 slides in a moist chamber at room temperature for 30 min.After incubation, the slides were removed from the incubator chamber and rinsed with PBS buffer.To prevent cross-contamination, a stream of PBS buffer was run along the midline of the slide, allowing the buffer to run off the lower edge of the slide.After the washing procedure, 25 µl of FITC-conjugated goat anti-human IgG was immediately applied to each well, and the slide was returned to the incubator chamber.The slides were allowed to incubate at room temperature in a moist chamber for another 30 min.Subsequently, the slides were washed in the same manner as described above and then mounted on coverslips using the provided mounting medium.
Slides were viewed at 20× magnification and photographed on an EVOS f1 fluorescence microscope at a 250-ms exposure with 100% intensity.Positive-and negative-control sera were provided by the vendor.Samples that demonstrated fluorescence greater than the negative control were considered positive for HEp-2 staining.

MPER-18 IgH KI mouse construction, characterization and immunization studies
MPER-HuGL18 H mice were generated following published protocols 39,72 .In brief, the targeting vector 4E10 was modified by the incorporation of human rearranged MPER HuGL18 V(D)J (heavy chain construct) sequences downstream of the promoter region and by elongation of the 5′ and 3′ homology regions using the Gibson assembly method (New England Biolabs).The targeting vector DNA was confirmed by Sanger sequencing (Eton Bioscience).Next, fertilized mouse oocytes were microinjected with a donor plasmid containing the prerearranged MPER HuGL18 IgH with the mouse VHJ558 promoter, two pairs of single guide RNAs (25 ng ml -1 ) targeting the H locus and AltR-Cas9 protein (50 ng ml -1 ) and injection buffer 39 .Following culture, resulting zygotes were implanted into the uteri of pseudopregnant surrogate C57BL/6J mothers.
For experiments, male B6.SJL-Ptprc a Pepc b /BoyJ mice (CD45.1 +/+ ) 8-12 weeks of age were purchased from The Jackson Laboratory.F0 mice from the MPER-HuGL18 H mouse (CD45.2+/+ ) colony were bred at the animal facility of the Gene Modification Facility (Harvard University), and breeding for colony expansion and experimental procedures was subsequently performed at the Ragon Institute of Mass General, MIT and Harvard.Ear or tail snips from MPER-HuGL18 H mice were genotyped by TaqMan assay under a fee-for-service agreement (TransnetYX).TaqMan probes for the genotyping assay were developed by TransnetYX.CD45.2 + B cells from MPER-HuGL18 H donor KI mice were enriched using a Pan-B Cell Isolation kit II (Miltenyi Biotec), counted, diluted to the desired cell numbers in PBS and adoptively transferred retro-orbitally into CD45.1 + recipient mice, as reported previously 33 .All experiments were performed under the approval of the IACUC of Harvard University and the MGH (animal study protocols https://doi.org/10.1038/s41590-024-01833-w2016N000022 and 2016N000286) and were conducted in accordance with the regulations of the Association for Assessment and Accreditation of Laboratory Animal Care.
Following single-cell sorting of antigen-specific B cells, the genes encoding the variable region of the heavy and light chains of IgG were amplified through reverse transcription PCR.In brief, first-strand cDNA synthesis was performed using SuperScript III Reverse Transcriptase (Invitrogen) according to manufacturer's instructions.Nested PCR reactions consisting of PCR-1 and PCR-2 were performed as 25-µl reactions using HotStarTaq enzyme (Qiagen), 10 mM dNTPs (Thermo Fisher Scientific) and cocktails of Iggand Igk-specific primers and thermocycling conditions described previously 73 .PCR products were analyzed on 2% 96-sample precast E-Gels with SYBR Safe (Thermo Fisher Scientific), and wells with bands of the correct size were submitted to GENEWIZ for Sanger sequencing.Heavy chain products were sequenced using the heavy chain reverse primer from PCR-2 (5′-GCTCAGGGAARTAGCCCTTGAC-3′), and the light chain was sequenced using the light chain reverse primer (5′-TGGGAAGATGGATACAGTT-3′) from PCR-2.Reads were quality checked, trimmed, aligned and analyzed using Geneious software (Biomatters).IMGT/V-QUEST (http://www.imgt.org) 74,75was used for mouse/human immunoglobulin gene assignments.

hD3-3/J H 6 rearranging mouse construction, characterization and immunization studies
To generate the hD3-3/J H 6 mice, a cassette containing hD3-3 and hJ H 6 (Extended Data Fig. 7) was integrated via homologous recombination into the mouse DQ52/J H locus of an F1 (129/Sv × C57BL/6) embryonic stem (ES) cell line 76 .As a result, hD3-3 substitutes for mouse DQ52, and hJ H 6 substitutes for mouse J H 1-4. Correct integration of the hD3-3/J H 6 cassettes was verified by Southern blotting.An ES clone with hD3-3/ J H 6 was injected into Rag2-deficient blastocysts to yield chimeric mice 77 , which were subsequently crossed with 129SVE mice to give germline transmission.For the B cell analysis shown in Extended Data Fig. 7b-e, splenocytes from wild-type mice were stained with the following antibodies: FITC anti-B220 and APC anti-Thy1.2(Extended Data Fig. 7b); FITC anti-B220, PE anti-IgM and APC anti-IgD (Extended Data Fig. 7c); FITC anti-B220, APC anti-CD93, PE/Cy7 anti-CD23 and PerCP/ Cy5.5 anti-CD21 (Extended Data Fig. 7d); and FITC anti-B220, biotin anti-IgK, APC streptavidin and PE anti-IgL (Extended Data Fig. 7e).Dead cells were gated out by Sytox blue staining.Flow cytometry was performed in a AttuneNxT instrument, and data were analyzed with FlowJo10 software.For IgH repertoire analysis shown in Extended Data Fig. 7f,g, genomic DNA was isolated from splenocytes of hD3-3/ J H 6 mice.Repertoire analysis was performed with the HTGTS-rep-seq technique 78,79 .Library synthesis was initiated with a primer downstream of hJ H 6: 5Biosg/CAA CCT GCA ATG CTC AGG AA.Illumina MiSeq adaptors were added to the ends of library DNA, and sequencing was performed on a MiSeq instrument.Sequencing data were analyzed with the HTGTS-rep-seq pipeline.
Mice were killed with compressed CO 2 (100%) in a clear chamber to allow for visualization of respiration and subsequent death via respiratory cessation.Blood was collected from the chest cavity before removal of the spleen and lymph nodes (mesenteric, inguinal and popliteal; RNA injections only, left leg only).Tissues were placed in 3 ml of FACS buffer (1× PBS (calcium/magnesium free), 1 mM EDTA, 25 mM HEPES (pH 7.0) and 1% heat-inactivated FBS) in a 15-ml polypropylene tube on ice.Tissues were disassociated using the rough ends of two sandblasted microscope slides in a 5-ml Petri dish and returned to the same 15-ml polypropylene tube for centrifugation (460g for 5 min at 4 °C).Red blood cell lysis was performed using 1 ml of ACK buffer (Quality Biological) for 2 min on ice in a 15-ml polypropylene tube.Lysis was halted by adding 14 ml of FACS buffer per sample.After lysis and centrifugation (460g for 5 min), cells were resuspended in 3 ml of Bambanker freezing medium (Bulldog Bio) before filtration through a cotton-plugged, borosilicate Pasteur pipette into a borosilicate glass test tube.One milliliter of filtered cell solution was subsequently divided into three cryovials per mouse, which were precooled in a styrofoam rack on dry ice.Cells were stored at −80 °C for 2-7 days before long-term storage in liquid nitrogen.Sera were collected by spinning the blood at 14,000 rpm for 30 min.Sera were stored at −20 °C.All work followed IACUC guidelines associated with animal protocol number 20-0001.
Streptavidin-conjugated baits were prepared by combining biotinylated monomeric baits or Env trimer baits with fluorescent streptavidin at room temperature for at least 1 h in the dark.Wild-type baits were complexed with streptavidin-AF647 (Invitrogen, S21374) and streptavidin-BV421 (BioLegend, 405225).KO baits were conjugated with TotalSeq-C hashtagged streptavidin-PE (BioLegend, 405261).Baits were conjugated with streptavidin at a 4:1 (bait/streptavidin) ratio and used at a final bait concentration of 200 nM for staining.
During the addition of antibody master mix, a unique TotalSeq-C anti-mouse hashtag antibody (BioLegend) was added to each sample at a concentration of 2.5 µl per up to 20 million cells.Following antibody staining, 1 ml of 1:300 live/dead stain (LIVE/DEAD Fixable Aqua, Invitrogen, L34966) was added to each sample and incubated for an additional 15 min at 4 °C.At the end of staining, cells were washed with 10 ml of FACS buffer and resuspended in 500 µl of FACS buffer.
All samples were sorted on a BD FACSMelody using FACSChorus software.Single-color compensations were performed using C57BL/6 splenocytes with matched antibodies.For channels used for bait detection, cells were stained with biotinylated anti-CD19 (BioLegend, 115503), followed by secondary staining with the appropriate fluorescent streptavidin.Samples were filtered through a 35-µm mesh-cap FACS tube (Falcon, 352235) before being loaded on the sorter.A maximum of 15,000 cells was sorted using purity mode into a PCR plate well containing 20 µl of 0.2-µm-filtered FBS.Event rates were typically maintained at ~1,000 events per s and no more than 1,500 events per s to ensure high sorting efficiencies.
Sorted samples were prepared for BCR sequencing by the 10x Genomics Single Cell Immune Profiling platform.After cell sorting, DPBS was added up to near the top of the sample collection well (~100 µl) and gently mixed to dilute the FBS catch buffer.The plate was sealed, and cells were spun down for 2 min at 2,000 rpm, after which the excess buffer was removed except for the ~38 µl required for the 10x Genomics GEM reaction.Samples were processed according to manufacturer's user guide for Chromium Next GEM Single Cell 5′ Reagent kits v2 (Dual Index) with Feature Barcoding with two previously described main modifications 82 .The number of PCR cycles in the cDNA amplification step was determined by assuming that only 20% of the total number of cells sorted would be recovered.This modification was made based on the observation that, on average, the number of unique paired BCR sequences recovered from the 10x Genomics platform was typically ~20% of the total number of cells sorted.In the second modification, the number of PCR cycles for each of the V(D)J amplification steps was increased to ten cycles if the number of cells sorted (according to the sorter) was fewer than 1,000 cells.Pooled libraries were sequenced on an Illumina NextSeq 2000 using NextSeq 1000/2000 Control Software and a 100-cycle P3 reagent kit (Illumina, 20040559) with a target depth of 5,000 paired-end reads for both the V(D)J and Feature Barcode Libraries and run using read parameters indicated in the 10x Genomics user guide.
Raw sequencing data were demultiplexed, processed into assembled V(D)J contigs and counts matrix files and assigned to specific animal IDs based on TotalSeq-C antibody hashtag counts using Cell Ranger (v6.1) and scab, as previously described 82 .Epitope-KO-positive cells were identified based on unique molecular identifier counts of the hashtagged PE-streptavidin probe using scab 82 .Gene assignment, annotation and formatting into Adaptive Immune Receptor Repertoire format 83 for paired heavy and light chain antibody sequences were performed using SADIE with a custom hD3-3/J H 6 mouse germline reference database 11 .SADIE outputs were summarized using custom Python scripts.Animals with low cell viability (<25,000 total live B cells) or low sequence recovery (<20 total sequences, where indicated <100 total sequences) were excluded.In addition, all non-IgG sequences were discarded.Sequences matching the criteria listed in Extended Data Fig. 1a were counted, and the frequencies were plotted in GraphPad Prism 9.5.1.Multipanel figures were assembled using Adobe Illustrator.

Rhesus macaque repertoire characterization and immunization studies
Indian rhesus macaques (Macaca mulatta) for the 10E8-GT10.2and MD39 immunization groups were housed at Alpha Genesis and maintained in accordance with National Institutes of Health (NIH) guidelines.This study was approved by the Alpha Genesis IACUC.All macaques were between 2 and 3 years old at the time of the immunization.
For the 10E8-GT10.2immunization group, eight macaques were used in the study, with four females and four males.Immunizations were given subcutaneously in the left and right deltoids with a total dose of 50 µg of 10E8-GT10.212mer nanoparticle and 375 µg of SMNP 43 per side.For the MD39 control group, four macaques were used in the study, with two females and two males.Macaques were immunized subcutaneously in the left and right midthighs with a total dose of 50 µg of MD39 and 375 µg of SMNP 43 per side.For priming, a seven-dose 12-day escalating dose strategy was used 42 .Data from the MD39 group have been previously published 50 .
For the 10E8-GT12 immunization group, macaques were housed at the Yerkes National Primate Research Center and were maintained in accordance with NIH guidelines.This study was approved by the Emory University IACUC.Six male macaques were used in the study and were aged between 3 and 4 years at the time of immunization.Immunizations were given subcutaneously in the left and right midthighs with a total dose of 50 µg of 10E8-GT12 12mer and 187.5 µg of SMNP-QS21 per side using a seven-dose 12-day escalating dose strategy 50 .
For the naive B cell screening study, macaques were housed at the Yerkes National Primate Research Center and were maintained in accordance with NIH guidelines.The study was approved by the Emory University IACUC.Blood was drawn from nine unimmunized macaques, processed for PBMCs and frozen and stored in liquid nitrogen until further analysis.

Rhesus macaque lymph node fine needle aspiration
Lymph node fine needle aspirates were used to sample either the left or right draining axillary lymph nodes or the left and right draining inguinal lymph nodes, depending on the route of immunization.Draining lymph nodes were identified by palpation, and fine needle aspirates were performed by a veterinarian.A 22-gauge needle attached to a 3-ml syringe was passed into the lymph node up to five times.Samples were dispensed into RPMI medium containing 10% FBS and 1× pen/strep.ACK lysing buffer was used if the sample was contaminated with red blood cells.Lymph node fine needle aspirate samples were frozen and stored in liquid nitrogen until analysis.

Rhesus macaque IGHD3-3 genotyping
To genotype IGHD3-3 ('IGHD3.41'),we used targeted long-read Pacific Biosciences single-molecule real-time sequencing data generated for each macaque in the study cohort.Sequencing data were generated by adapting the published human immunoglobulin loci-targeted enrichment protocol 84,85 .Briefly, a custom oligonucleotide probe panel was designed ('HyperExplore', Roche) using immunoglobulin heavy chain (IGH), κ (IGK) and λ (IGL) genomic region sequences from the rhesus macaque genome reference build (RheMac10) and alternative haplotype assemblies from Cirelli et al. 42 as sequence targets.
High-molecular-weight genomic DNA was isolated from PBMCs collected from each macaque using a DNeasy kit (Qiagen).DNA (1-2 µg) was then sheared using g-tubes (Covaris) and size selected using a Blue Pippin instrument (Sage Science).Size-selected DNA was end repaired and A-tailed using the standard KAPA library protocol (Roche), followed by ligation of sample-specific sequence barcodes and universal primers.PCR amplification was performed for eight to nine cycles using PrimeSTAR GXL polymerase (Takara), and the resulting products were further size selected and purified using 0.7× AMPure PB beads (Pacific Biosciences).Target enrichment hybridization was performed using IGH/IGK/IGL-specific oligonucleotide probes (Roche).Target fragments were recovered using streptavidin beads (Life Technologies), followed by a second round of PCR amplification for 16-18 cycles using PrimeSTAR GXL (Takara).Long-read sequencing libraries were prepared using a SMRTbell Express Template Preparation kit 2.0 (Pacific Biosciences), including Damage Repair and End Repair mix to repair nicked DNA, followed by the addition of an A-tail and overhang ligation with SMRTbell adapters.Libraries were then treated with a

Fig. 1 |
Fig. 1 | 10E8-class bnAb precursors are present in most humans.a, Schematic of the epitope scaffold design showing antibody 10E8 (gray) and Env (blue), including the MPER (purple) that was grafted onto an unrelated epitope scaffold (cyan).b, Frequency of 10E8-class IgH precursors in 14 NGS datasets 7,27 of heavy chains from HIV-seronegative humans defined as sequences with genes encoding V H closely related to 10E8 and HCDR3 lengths of 21-24 aa with a YxFW motif at the correct position.Lines indicate the median and 25 and 75% quantiles; HCs, heavy chains.

Extended Data Fig. 1 | 9 Extended Data Fig. 3 |Extended Data Fig. 4 |. 6 |
Design and properties of immunogens.a, Overview of 10E8-class and LN01-class antibody categories.HCDR3 motifs are shown as regular expressions that were used to query the database.If multiple amino acids were allowed at the same position, they are shown in square brackets; positions in which all amino acids were allowed are indicated as '.'.b, Schematic of the development of MPER-GT scaffolds.c, Schematic overview of nanoparticle formation by genetic fusion of the immunogen (T2983 − GT) to each terminus of the 3-dehydroquinase nanoparticle from Thermus thermophilus (NP) via flexible linkers containing exogeneous T-help peptides derived from Aquifex aeolicus lumazine synthase (link).The epitope scaffold is shown in light blue, the MPER graft in purple, the linker in green, the nanoparticle in red and glycans in dark blue.d, SEC-MALS traces of 10E8-GT NPs.Normalized UV 280 absorptions are shown as dotted lines and protein molecular weights of main peaks are shown as solid lines.e, DSC measurements of the indicated monomers and nanoparticles with results from a fit indicated in light grey.f, Amino acid sequences of 10E8-GT epitope scaffolds through generation 7, none of which had the circular permutation present in later generations.Germline-targeting mutations are highlighted in red; N-linked glycosylation sites are blue; the D-gene binding pocket is green; resurfacing and/or solubility enhancing mutations are orange.All sequences are succeeded by a 6x His-tag, unless the protein ends in stop codons (denoted by symbol *).10E8-GT8.2through 10E8-GT12 are preceded by a mammalian secretion signal.g, Resurfaced T298v2 sequences compared to previously published T298 32.Colors as in f. h, Amino acid sequences of monomeric immunogens based on resurfaced circularly permutated T298v2, with colors as in f. i, Amino acid sequences of multivalent nanoparticles.Sequences are wrapped over multiple lines.The 3-dehydroquinase is shown in purple; additional T help epitopes are brown; epitope KO mutations are cyan; and all other colors are shown as in f.Extended Data Fig. 2 | Glycosylation sites on 10E8-GT nanoparticles vary in occupancy.Site-specific glycan analysis was measured using the single site glycan profiling (SSGP) and DeGlyPHER methods.Positions of N-linked glycosylation sites are indicated as relative positions within the epitope-scaffold (ES) or the nanoparticle (NP, indicated with a black box).The 24mer contains two independent copies of the scaffold, which cannot be distinguished by either method and therefore averaged values are shown.Data collection, refinement and validation.a, Data collection and refinement statistics of x-ray crystallography.a Numbers in parentheses refer to the highest resolution shell; b CC 1/2 = Pearson correlation coefficient between two random half datasets; c From MolProbity 64 .b, Summary statistics of data collection, refinement and validation of the cryo-EM reconstruction of 10E8-GT10.2 in complex with 10E8 and W6-10 Fabs.c, Fourier Shell Correlation, d, angular sampling and e, map colored according to local resolution (units Ångstrom) of the cryo-EM reconstruction of 10E8-GT10.2 in complex with 10E8 and W6-10 Fabs.https://doi.org/10.1038/s41590-024-01833-wEx vivo evaluation of 10E8-GT scaffolds, related to Fig. 4. a, Representative gating scheme.b, Enrichment of HCDR3 lengths among epitope-specific B cells over unsorted controls as in Fig. 4d.c, Frequency of long (>=20aa) HCDR3s among epitope-specific B cells (sorted) as in b, or among total IgM+ naïve B cells (unsorted).Symbols represent n = 3 (GT9), n = 4 (GT10.1),n = 6 (GT12) or n = 14 (unsorted controls) independent donors.* p = 0.046, **** p < 0.0001, Kruskal-Wallis test with Dunn's multiple comparison correction.d, Percentage of HCDR3s containing the YxFW motif among epitope-specific B cells as in c. * p = 0.03, *** p = 0.0001, Kruskal-Wallis test with Dunn's multiple comparison correction.e, Percentage of 10E8-class V H among epitope-specific BCRs with 10E8-like HCDR3s (with length 21-24 aa and YxFW at correct position within HCDR3) within datasets obtained using the 10x Genomics sequencing method as in c. **p = 0.002, two-sided Mann-Whitney test.f, Percentage of IGVL3 family light chains among 10E8-GT12-sorted BCRs that are either 10E8-class (10E8-class H3) or lack the YxFW motif (non-YxFW).n = 6 independent donors, ns not significant, two-sided Wilcoxon test.g, Percentage of 10E8-GT12-specific naive IgM + BCRs with 10E8-class or LN01-class HCDR3s.Symbols represent n = 6 (GT12) or n = 14 (unsorted controls) independent donors.h, Frequency of 10E8class or LN01-class B cells among IgM+ naive B cells, detected through 10E8-GT12 sorting as in g. i, SPR-measured monovalent K d values for 10E8-GT9, 10E8-GT10.2,and 10E8-GT12 monomer binding to 10E8-class and non-10E8-class (competitor) antibodies isolated by the respective scaffolds.Symbols represent different antibodies; lines represent median values.Extended Data Fig. 5 | Poly-and Auto-reactivity of 10E8-class precursors.a, Polyspecificity reagent binding as measured by ELISA.PGT121, VRC01 and PGT128 served as negative controls, MPER bnAb 4E10 as a positive control.NGS-1 through −22 correspond to human NGS precursors described in the main text.These are 10E8-class heavy chain (HC) precursors identified from searching next-generation sequencing (NGS) datasets of primarily naive IgM HCs from 14 HIV-seronegative human donors 5,27 paired with the inferred-germline 10E8 light chain (LC); 10E8-HuGL: bona fide HC/LC pairs isolated by epitope-specific sorting of naïve human B cells (n = 25).b, Mean fluorescence intensity (MFI) of antibodies as in a in a HEp-2 cell autoreactivity assay.c, raw images of data shown in b.Immunization of MPER-HuGL18 H B cell adoptive transfer recipient mice with 10E8-GT10.212mers.a, Flow cytometry analysis of bone marrow cells from WT (n = 4) and MPER-HuGL18 H (M18, n = 7) mice; gating strategy shown on the left.B-cell progenitors (B220 + ) were divided into immature (CD43 + ) and mature (CD43 − ) cells.Early (CD43 + ) B-cell progenitors were subdivided into Hardy populations A (CD24 − BP-1 − ), B (CD24 + BP-1 − ), and C (CD24 + BP1 + ).Late (CD43 − ) B-cell progenitors were subdivided into Hardy populations D (IgM − IgD − ), E (IgM + IgD int ), and F (IgM + IgD+).Right bars represent quantifications of these populations, error bars indicate SD. b, Frequency of CD45.2 + B cells among splenic B cells, one day after adoptive transfer of 200,000 CD45.2 + MPER-HuGL18 H B cells into CD45.1 + WT mice.Symbols represent individual animals, error bars indicate SD. c, Germinal center (GC) response to immunization in CD45.1 + WT mice adoptively transferred with 2 × 10 5 CD45.2 + MPER-HuGL18 H B cells on Day 21 after immunization with 10E8-GT10.212mer or negative control 10E8-GT9-KO 12mer (KO).Left column shows the frequency of total GC (CD38 lo CD95 + ) among B cells gated from SSL; right column shows the frequency of CD45.2 B cells among total GC.Extended Data Fig. 8 | Immunogenicity of 10E8-GT nanoparticles in hD3-3/J H 6 mice.a, Representative gating strategy of splenic B cells sorted for sequencing of BCRs from hD3-3/J H 6 mice six weeks after immunization with 10E8-GT12 24mer.b, Percentage of 10E8-GT-binding cells (10E8-GT9-KO ++ , 10E8-GT10.1 ++ or 10E8-GT12 ++− ) among IgM − IgD − B cells, 42 days after immunization of hD3-3/ J H 6 mice with 10E8-GT9-KO 12mer (control, n = 3), 10E8-GT12 12mer (n = 12), 10E8-GT12 12mer (n = 5) or 10E8-GT12 24mer (n = 12) delivered as protein in SMNP, respectively, or 10E8-GT12 24mer delivered by mRNA (n = 11).Each symbol indicates an animal, lines indicate median values.c, Percentage of 10E8-class HCDR3s among all IgM − IgD − B cells after immunization as in b. d, Enrichment ratio for HCDR3 amino acid (aa) length distribution for epitope-specific (10E8-GT10.1 ++ 10E8-GT10.1-KO− or 10E8-GT12 ++ 10E8-GT12-KO − ) IgG + B cells from animals immunized with the indicated 10E8-GT immunogens, relative to HCDR3 amino acid length distribution for epitope-specific (10E8-GT9 − KO ++ 10E8-GT9 − ) IgG + B cells from animals immunized with 10E8-GT9-KO 12mer as in b.HCDR3 lengths longer than 22 were only found in the 10E8-GT-immunized groups, precluding calculation of enrichment scores for longer HCDR3s.e, Frequency of long (>=20 aa) HCDR3s among epitope-specific IgG + B cells as in d. f, Percentage of HCDR3s containing the YxFW motif among epitope-specific + B cells as in c. *p < 0.05, Kruskal-Wallis test with Dunn's multiple correction.value (control vs. GT10 12mer): 0.02, p value (control vs. GT12 24mer): 0.01.g, Foldchange in SPR-measured K D for 10E8-GT immunogens binding to 10E8-iGL3 upon addition of the indicated HCDR3 mutations.h, Percentage of epitope-specific (10E8-GT9-KO ++ 10E8-GT9 − ,10E8-GT10.1 ++ 10E8-GT10.1-KO− , or 10E8-GT12 ++ 10E8-GT12-KO − ) with 10E8 − class or LN01-class HCDR3s among IgG + BCRs from day 42 after immunization of hD3 − 3/J H 6 as in b.Symbols represent individual animals; bars indicate median values.