DNA origami vaccines program antigen-focused germinal centers

Abstract

Priming rare subdominant precursor B cells in germinal centers (GCs) is a central goal of vaccination to generate broadly neutralizing antibodies (bnAbs) against HIV. Multivalent immunogen display on protein nanoparticle scaffolds can promote such responses, but also generate scaffold-specific B cells that could theoretically limit bnAb precursor expansion in GCs. We rationally designed DNA origami-based virus-like particles (DNA-VLPs) displaying a germline-targeting HIV Env immunogen, which elicited no scaffold-specific antibody responses. Compared with a state-of-the-art clinical protein nanoparticle, these DNA-VLPs increased the expansion of epitope-specific GC B cells relative to off-target B cells and enhanced expansion of bnAb-lineage B cells in a humanized mouse model of CD4 binding site priming. Thus, minimizing off-target responses enhances bnAb priming and indicate DNA-VLPs are a promising vaccine platform.

Introduction:

Generation of broadly neutralizing antibodies (bnAbs) that recognize diverse viral variants is a central goal for successful vaccination against HIV (1) and universal vaccines for other pathogens such as influenza (2, 3), coronaviruses (4), and dengue virus (5). Precursor B cells capable of evolving to produce bnAbs are characteristically rare in the naive repertoire and often have low affinity for target immunogens (6–8). Engaging rare B cell clones that belong to immunologically subdominant bnAb lineages to promote their somatic hypermutation and affinity maturation in germinal centers (GCs) remains an outstanding challenge and a primary aim of rational vaccine design (1, 9–12).

Multivalent display of many copies of an antigen on the surface of protein nanoparticles augments humoral responses and priming of bnAb precursors (13–22). Nanoparticles efficiently transport antigen through lymphatics to draining lymph nodes (22, 23), enhance B cell receptor (BCR) crosslinking (24, 25) inducing signal amplification downstream of the BCR (26), and promote potent valency-dependent B cell activation across a broad range of BCR affinities (27). However, when protein nanoparticles are used as scaffolds to display target antigens, they act as thymus-dependent (TD) repetitively arrayed antigens themselves, eliciting priming of scaffold-specific B cells. These B cells recognize irrelevant protein substrates and potentially compete in GCs against the desired, epitope-specific bnAb precursor B cells (18, 28–32). Scaffold-specific antibody responses against bacterially-derived scaffolds such as ferritin (18, 29, 30) and lumazine synthase (31, 32) and computationally designed two-component nanoparticles (28) have been quantified via serology in animal models and humans. However, it remains poorly understood how these scaffold-specific B cells shape the clonal competition dynamics within GCs, which rely on finite populations of B cells and helper T cells (33). While strategies to limit scaffold-specific responses have been used, such as glycosylation (32, 34) or physical masking of exposed scaffold epitopes (28, 35), they have generally failed to fully eliminate off-target responses.

The effect of scaffold-specific B cells on antigen-specific B cell responses remains unclear. The scaffold-specific B cells might outcompete antigen-specific GC B cells for limited T cell help and antigen availability (33), particularly if the scaffold is more immunogenic than the antigen itself (7). Additionally, they might generate scaffold-specific memory B cells that influence the humoral response upon boosting with the same nanoparticle platform (36, 37). Moreover, they could lead to high titers of anti-scaffold antibodies that facilitate epitope masking, antibody feedback, or carrier-induced epitopic suppression that alters the subsequent immune response (38, 39). Each of these competitive effects are of particular concern when designing a vaccine against immunologically subdominant epitopes (28), such as the CD4 binding site (CD4bs) in HIV Env antigens, which is the target of VRC01-class bnAbs (40).

One strategy to prime these rare subdominant precursors is to immunize with engineered germline targeting immunogens. For example, the engineered outer domain of gp120, eOD-GT8, was designed to bind germline-reverted VRC01-class BCRs with high affinity (12). A protein nanoparticle form of eOD-GT8 (eOD-GT8 60mer) was developed by fusing eOD-GT8 with the bacterial protein lumazine synthase (LumSyn); 60 copies of LumSyn self-assemble to form an icosahedral nanoparticle displaying eOD-GT8 at high density (8, 12, 41). This ~30 nm protein nanoparticle effectively activates VRC01-class B cell precursors in humanized mouse models and humans (21, 31, 42–44), but it also elicits a strong LumSyn-specific antibody response (31, 32, 42). It remains unknown whether this scaffold-specific response may limit the priming of the “on-target” bnAb precursors of interest.

As an alternative to protein-based scaffolds, virus-like particles (VLPs) formed by the programmed assembly of DNA origami (DNA-VLPs) enable user-defined geometries (45–47) with precise antigen display (46, 48, 49) and enhance antibody responses relative to soluble antigens (50, 51) while avoiding anti-scaffold B cell responses (30). Furthermore, DNA-VLPs are minimally immune-stimulatory (52, 53), unless specific nucleic acid motifs are included to activate innate Toll-like receptor pathways (52, 54, 55). Thus, DNA origami may serve as an attractive vaccine scaffold due to its highly programmable geometry and size (45, 56), quantitative spatial precision for display of biomolecular cargoes (57, 58), and immunologically inert scaffold (30). We hypothesized that DNA origami scaffolds could enable a rigorous test of the effects of scaffold-specific competitor B cells on the recruitment and affinity maturation of rare “on-target” precursors in primary GCs. Accordingly, we rationally designed icosahedral DNA-VLPs conjugated with eOD-GT8 as a clinically relevant test immunogen.

Results

Scaffolding eOD-GT8 on DNA origami virus-like particles

We employed a DNA-VLP that had previously shown efficacy in presenting SARS-CoV-2 antigens (30). Thirty equally-spaced copies of the eOD-GT8 antigen were conjugated to the surface of the DNA-VLPs (Fig. 1A). This VLP was computationally designed using DAEDALUS software (47, 56), synthesized using a custom ssDNA scaffold folded with 30 DBCO-modified staple oligonucleotides, and reacted with eOD-GT8 containing an N-terminal azido linker (30, 57) (fig. S1, table S1 and S2). SPAAC reactions were performed with at least 3-fold molar excess of eOD-Azide (fig. S1D), and unreacted antigen was removed by dialysis (fig. S1E). DNA-VLP self-assembly was confirmed through agarose gel electrophoresis (AGE) and dynamic light scattering (Fig. 1B, fig. S2). Antigen-conjugated DNA-VLPs had a hydrodynamic radius of ~40 nm, which we annotate as d40-30mer. The average functionalization efficiency of d40-30mer across independent sample preparations was ~94% (Fig. 1C, fig. S2D). Electron microscopy imaging validated preservation of the DNA-VLPs’ icosahedral geometry post-antigen conjugation and dialysis (Fig. 1D, fig. S2E, fig. S3). Proper presentation of the eOD-GT8 antigen on the particle surfaces was confirmed through gel electrophoresis binding assays with murine VRC01, a broadly neutralizing antibody targeting the CD4 binding site of gp120 that binds with high affinity to eOD-GT8 (12) (fig. S4A). Consistent with previous findings (46), culture of germline Ramos B cells expressing the germline VRC01 B cell receptor with 5 nM eOD-GT8-functionalized DNA-VLPs led to rapid B cell activation as assessed by calcium signaling, while the same dose of soluble eOD-GT8 monomer failed to trigger calcium release (Fig. 1E).

Fig 1. Nanoparticulate eOD-GT8 assembly on icosahedral DNA-VLP scaffolds.

(A) Atomic model of d40-30mer. B) Agarose gel electrophoresis showing bare origami VLPs next to an antigen-functionalized d40-30mer. (C) Antigen functionalization efficiency from five separate DNA-VLP reactions quantified by bicinchoninic acid colorimetric assay. (D) Cryo-electron micrographs of d40-30mer and 2D class averages (scale bar represents 50 nm). (E) Representative calcium flux trace of glVRC01 B cells incubated with 5 nM eOD-GT8 monomer or d40-30mer from n=3 independent experiments. (F) Schematic of immunization study. C57BL/6 mice (n=5/group) were injected s.c. with eOD monomer or d40-30mer (equivalent to 5 μg of eOD-GT8 monomer) together with 5 μg SMNP. This experiment was performed twice and shown are data points from one experiment. (G) eOD-GT8 IgG antibody titer over time. Statistical significance was calculated using multiple Mann Whitney tests with Benjamini, Krieger, Yekutieli correction. Error bars show standard error. (H) Anti-origami IgG responses represented as area under the curve after prime (day 21) and boost (day 42). An anti-dsDNA antibody was used as positive control. (I) C57BL/6 mice (n=5/group) were primed with 5 μg of monomer or d40-30mer and sacrificed on day 14 for analysis of GC responses by flow cytometry. This experiment was performed twice and shown are data points from one experiment. Total counts of GC B cells (B220⁺/CD38^loGL7^hi). (J) Total counts of Tfh cells (CD4⁺/CXCR5^hi PD1^hi). (K) Representative flow plots showing eOD-tetramer staining, frequency of eOD⁺⁺ GC B cells out of all GC B cells and counts of eOD⁺⁺ GC B cells. (L) DNase I WT or KO mice (n=5/group) were immunized with 5 μg d40-30mer. Shown are (L) total counts of GC B cells, (M) counts of Tfh cells, and (N) frequencies and counts of antigen-specific GC B cells. Absolute counts plots are shown with log axes; geometric mean and geometric S.D. are shown. Statistical testing in (I) to (N) was performed with Mann Whitney U tests: n.s is not significant.

d40-30mer DNA-VLPs enhance serum antibody titers in prime-boost immunization but are weak priming immunogens

We expected that the multivalency of d40-30mer would elicit higher antibody responses in mice compared to eOD-GT8 monomer. For immunization studies, we combined DNA-VLPs with saponin-MPLA nanoparticles (SMNP), a potent ISCOMs-like adjuvant (59). We verified that mixing DNA-VLPs with SMNP did not affect the stability of the DNA-VLPs (fig. S4b). C57BL/6 mice were primed and boosted with equimolar doses of eOD-GT8 monomer or d40-30mer co-administered with SMNP, and antigen-specific serum IgG titers were monitored over time. Both eOD-GT8 monomer control and DNA-VLPs resulted in weak responses post-prime, but following boosting the DNA-VLP elicited antibody titers ~1 log higher than the eOD-GT8 monomer (Fig. 1F, G). We assessed antibody responses against the bare DNA-VLP and observed no elevation in anti-origami IgG or anti-dsDNA IgG or IgM after prime or boost (Fig. 1H, fig. S5A-C).

We next evaluated germinal center responses in draining lymph nodes two weeks after a single immunization (fig. S6). Immunization with d40-30mer did not increase the number of GC B cells, follicular helper T (Tfh) cells), or antigen-specific GC B cells over eOD-GT8 monomer vaccination (Fig. 1I–K). Alternative adjuvants including alum, AddaVax, AS01b, and CpG ODN 1826 did not enhance GC B cell responses elicited by the DNA-VLPs (fig. S7).

A primary concern for DNA origami-based nanoparticles is their in vivo stability in the face of high levels of extracellular nucleases. Endonucleases have the highest activity towards wireframe DNA-VLPs (60, 61). Model DNA substrates incubated in sera isolated from mice deficient in the endonucleases DNase I, DNase IL3, or DNase I/IL3 (62, 63) revealed that DNase I contributed the most to the degradation of both ssDNA and dsDNA (fig. S8A). Even though nuclease expression may differ between serum and lymphoid tissues, we reasoned that DNase I-deficient mice would model reduced systemic nuclease activity and therefore assessed vaccine responses to DNA-VLPs in these animals (fig. S8B). The d40-30mer remained intact in DNase I KO serum for longer than seven days (fig. S8C). DNase I KO mice established robust GCs in response to protein vaccination (fig. S8D). To determine if DNase-mediated particle degradation was a limiting factor in vivo, we primed WT or DNase I KO mice with d40-30mer and analyzed germinal centers 14 days later. Total GC size, Tfh responses, and antigen-specific GC B cell responses elicited by d40-30mer in DNase I KO animals were low and identical to responses in WT littermates (Fig. 1L–N). These data suggested that the d40-30mer nanoparticle design may provide sufficient avidity to expand antibody-producing cells upon boosting, but the nanoparticle design, irrespective of serum stability, was not optimal for supporting primary GC responses.

d40-30mer DNA-VLPs have restricted access to follicular dendritic cells

To understand why the d40-30mer DNA-VLP poorly primed GC responses, we considered the biological barriers limiting antigen delivery to B cells in vivo. Antigen arriving at draining lymph nodes during a primary immunization is often quickly cleared by lymph flow (64), protease activity in the tissue (65, 66), or by cellular uptake and intracellular degradation (67). However, antibody- or complement-opsonized antigen can be captured by follicular dendritic cells (FDCs) and presented to B cells for prolonged periods to support GC responses (20, 68–70). We analyzed the biodistribution of fluorescently-labeled antigen in draining lymph nodes (dLNs) following immunization, comparing eOD-GT8 monomer and d40-30mer DNA-VLPs with the protein nanoparticle eOD-GT8 60mer (hereafter, p60mer). Two hours post immunization, eOD-GT8 monomer was primarily detected in central non-follicular areas of dLNs, largely colocalizing with F4/80⁺ medullary sinus macrophages (MSMs, Fig. 2A and B). By contrast, eOD-GT8 delivered as d40-30mer was present at substantially higher levels, and colocalized with CD169⁺ subcapsular sinus macrophages (SSMs) lining the edges of the lymph node and MSMs (Fig. 2A and B). The eOD-GT8 60mer is known to become decorated with complement in vivo via the lectin pathway, leading to rapid, robust accumulation on FDCs that express high levels of complement receptors (20). In agreement with this prior work, we found that p60mer particles colocalized with SSMs but were already at this early timepoint accumulating on the dendrites of CD35⁺ FDCs (Fig. 2A, lower panel, white arrows). Quantitative pixel analysis revealed that d40-30mer and p60mer had similar levels of antigen colocalization with SSMs (Fig. 2C), but only p60mer showed enrichment on FDCs (Fig. 2C). By day 7, p60mer had become highly concentrated on FDCs, whereas follicles in lymph nodes immunized with d40-30mer had no retention of fluorescent antigen (Fig. 2D). Thus, antigen trafficking and retention of DNA-VLPs was distinct from similarly-sized protein nanoparticles.

Fig. 2. d40-30mer DNA-VLPs are poorly retained on follicular dendritic cells in draining lymph nodes.

(A) C57BL/6 mice (n=3 mice, 6 lymph nodes/group) were primed s.c. with AF647-labeled eOD-GT8, d40-30mer, or p60mer. At 2 hr post-injection, inguinal lymph nodes were flash frozen, cryosectioned, and stained with anti-CD35 antibody (blue) and anti-CD169 antibody (green). Shown are representative lymph node images collected on a laser scanning confocal microscope, with close-up images of subcapsular sinus regions and follicles (lower panel). White arrows point to antigen (magenta) colocalized with FDC dendrites. Scale bars represent 200 μm. (B) Lymph node sections from mice immunized with AF647 labeled eOD-GT8 monomer or d40-30mer (magenta) were stained with anti-CD169 (green) and anti-F4/80 (blue) antibodies for visualization of subcapsular sinus or medullary sinus macrophages, respectively. Arrows point to antigen colocalization with MSM or SSM populations. (C) Quantification of antigen colocalized with SSMs (left) or FDCs (right) at 2 hours in inguinal lymph node sections stained with anti-CD169 and anti-CD35 antibodies. Occupancy is defined as the fraction of area positive for markers of SSMs (CD169⁺) or FDCs (CD35⁺) that contained fluorescent antigen signal. Each data point represents measurements from one entire lymph node section from a single experiment. (D) Optically cleared inguinal LNs 7 days after immunization with AF647-labeled d40-30mer or p60mer. Follicles were labeled in situ by injection of anti-CD35 antibody prior to fixing and clearing. Scale bar = 200 μm. (E) MBL deposition measured by ELISA. Antigen formulations were immobilized on assay plates and treated with dilutions of fresh mouse serum. Anti-MBL antibody was used to detect MBL deposited on the antigens. Data points show technical replicates. (F) C3 deposition measured by ELISA. Antigen formulations were immobilized on assay plates and treated with dilutions of fresh mouse serum. Anti-C3 antibody was used to detect C3 bound to the antigens. Data points show technical replicates. Error bars indicate standard error. Statistics were calculated with one-way ANOVA with post-hoc Tukey test.

We hypothesized that trapping of d40-30mer in the subcapsular sinus could either be due to high expression of DNA-binding scavenger receptors on macrophages and/or lymphatic endothelial cells or because of inadequate triggering of complement pathways to mediate transport of the particles to FDCs (69, 71). To investigate these possibilities, we passivated the d40-30mer by coating it with polylysine-PEG polymer (72), which has been used in vivo with other DNA origami formulations for stabilization, charge neutralization (54), and avoidance of macrophage uptake (73, 74). PEGylated d40-30mer exhibited enhanced stability in mouse serum and reduced association with murine macrophage cells in vitro (fig. S9A-C). However, d40-30mer nanoparticles with or without PEGylation showed similar patterns of early accumulation in medullary and subcapsular sinus regions (fig. S9D), and PEG-coated DNA-VLPs elicited weaker GC responses compared to the uncoated d40-30mer (fig. S9E-H), likely due to interference of the polylysine-PEG coating with antigen recognition on BCRs (61).

Heavily glycosylated nanoparticle-antigens become decorated with complement in vivo via the lectin pathway, when mannose binding lectin (MBL), an innate immune protein present in blood and lymph, binds to the particles and triggers complement deposition (20, 71). We measured binding of MBL and the downstream complement protein C3 from naive mouse serum to eOD-GT8 monomer, d40-30mer, or p60mer particles immobilized on ELISA plates. MBL and C3 both showed reduced binding to eOD-GT8 monomer and d40-30mer compared to p60mer (Fig. 2E and F). Similarly, very little binding of recombinant MBL to eOD-GT8 monomer or d40-30mer was detected by ELISA compared to p60mer (fig. S10A). Thus, DNA-VLPs exhibited considerably less effective engagement of the lectin pathway for complement activation than p60mer nanoparticles.

Engineering DNA-VLPs with high antigen density induces follicle targeting and augments antigen-specific GC B cell responses

We hypothesized that poor complement activation by DNA-VLPs could be a result of insufficient antigen density on the DNA origami scaffold (71), as stable MBL binding is highly sensitive to the density of glycans on a viral or bacterial surface (75). We synthesized a set of 4 different particles, with diameters of approximately ~23 or ~34 nm, functionalized with either 30 or 60 copies of eOD-GT8 (Fig. 3A), providing a range of predicted glycan densities (0.04 to 0.14 glycans/nm²) and a range of inter-antigen distances from 11 nm (the original d40-30mer) down to 4.5–6 nm between neighboring antigens (d30-60mer, Fig. 3B, Supplementary Text 1 and fig. S2D). Upon antigen conjugation, the hydrodynamic diameters of these particles were approximately ~30 nm and ~40 nm, respectively, denoted as d30 and d40 (fig. S2E). Each of the four DNA-VLP designs elicited indistinguishable calcium signaling in germline VRC01-Ramos B cells (Fig. 3C). Notably, antigen attachment sites in d30-60mer are spaced similarly to the distances between CBD domains on MBL (76), and this design also closely mirrors the spacing of eOD-GT8 antigen in the p60mer.

Fig. 3. High density antigen display enhances lectin pathway activation, FDC targeting, and antigen-specific GC B cell responses.

(A) Atomic models of icosahedral DNA origami nanoparticles with diameters of 30 or 40 nm and eOD-GT8 valency of 30 or 60. (B) Predicted glycan densities (assuming four fully occupied N-linked glycans per eOD, distributed over the icosahedral surface) and inter-antigen distances calculated as the average distance to the five nearest neighbors based on atomic models. (C) Representative calcium signaling of glVRC01 B cells incubated with eOD formulations from 3 independent experiments. (D) MBL deposition and (E) C3 deposition on VLPs following incubation in fresh mouse serum detected by ELISA. Each data point shows the area under the curve of a technical replicate. (F) AF647-labeled eOD-GT8 monomer, DNA-VLPs, and p60mer were injected s.c. into C57BL/6 mice (10 μg, n=5 mice/group) with SMNP. Anti-CD35 antibody was injected prior to necropsy for in situ labeling of FDCs. Inguinal lymph nodes were collected at 96 hrs and optically cleared for whole tissue imaging. Shown are representative follicles (maximum projection over 80 μm) from 18-20 follicles areas imaged per group. Scale bar represents 200 μm. Each data point represents an by generating a binary mask and determining the overlapping pixel areas to calculate FDC occupancy (fraction of CD35⁺ cell area containing antigen, bottom left) and the percent of antigen colocalized with FDCs (bottom right). This experiment was performed twice, and data from one experiment is shown. (G) C57BL/6 mice (n=5/group) were primed with equimolar antigen doses (equivalent to 5 μg eOD-GT8) with SMNP. Inguinal LNs were analyzed on day 14. Shown are representative flow plots of eOD-tetramer staining of GC B cells. (H) Total counts of GC B cells (B220⁺CD38^loGL7^hi, left) and Tfh cells (CD4⁺CXCR5⁺ PD1⁺), right. (I) Frequency (left) and total counts of eOD⁺⁺ GC B cells (right). Absolute counts plots are shown with log axes; geometric mean and geometric S.D. are shown. Statistical significance was determined by 1-way ANOVA with post-hoc Tukey test.

Using our ELISA assays for MBL and complement binding, we observed a monotonic increase in both MBL and C3 deposition on DNA-VLPs as antigen density increased, with the highest amount of complement depositing on d30-60mer (Fig. 3D–E and fig. S10B). MBL binding with d30-60mer was nearly identical to p60mer; however, the amount of C3 that bound to these particles was lower. We next immunized mice with the panel of fluorescently-labeled DNA-VLPs, alongside eOD-GT8 monomer and p60mer controls, and harvested dLNs on day 4 after injection for imaging. eOD-GT8 signal was detected in lymph nodes from all DNA-VLPs, but colocalization with FDCs was only observed for d30-60mer and p60mer particles (Fig. 3F, fig. S11). However, the follicular area occupied by antigen and the percentage of total antigen per tissue section localized to FDCs was lower than p60mer nanoparticles (Fig. 3F, lower panel), consistent with the in vitro C3 deposition data. Thus, as observed with other synthetically mannosylated protein nanoparticles (71), greater MBL deposition on DNA-VLPs in vitro correlates with increased antigen capture on FDCs in vivo.

We next assessed how antigen density impacts GC formation. Mice immunized with eOD-GT8 monomer or DNA-VLPs showed similar total counts of GC B cells and Tfh cells two weeks post immunization, but overall lower cell counts compared to p60mer (Fig. 3G–H). Note that while it appears that more cells are found in the eOD⁺⁺ gate for p60mer due to the overall larger total GC response primed by the protein particle (Fig. 3), a lower frequency of these cells was antigen specific. However, d30-60mer immunization elicited a ~15-fold increase in the frequency and 10-fold increase in total counts of eOD-specific GC B cells compared to the eOD-GT8 monomer control, and a ~3-fold increase in eOD-specific B cell frequencies compared to the p60mer (Fig. 3I, fig. S12). Thus, a DNA-VLP design that promoted particle capture by FDCs correlated with substantially stronger expansion of antigen-specific GC B cells. Furthermore, this effect was driven by antigen density, rather than valency, since the d40-60mer failed to elicit antigen capture on FDCs or expansion of antigen-specific cells.

Incorporation of synthetic T cell epitopes expands GCs primed by DNA-VLPs

The observation that DNA-VLPs elicited lower total GC B cell and Tfh cell counts compared to the p60mer, despite nearly identical antigen display, prompted us to further iterate our d30-60mer design. B cell expansion and somatic hypermutation in GCs is governed by signals received from follicular helper T cells (33, 77). DNA-VLPs are T-independent scaffolds; therefore, all T cell epitopes must be derived from the conjugated protein antigen, and T cell help may be limiting for a small antigen such as eOD-GT8. In contrast, protein scaffolds, such as LumSyn that forms the basis of the p60mer, provide abundant scaffold-derived T cell help (42, 78). Therefore, we tested whether the incorporation of the universal helper peptide pan-HLA DR-binding epitope (PADRE) on the C-terminus of the antigen could amplify GC responses elicited by DNA-VLPs without introducing off-target B cell epitopes (table S3). PADRE binds with high affinity to a broad spectrum of human and mouse MHC haplotypes and has been shown to be safe in human clinical trials (79, 80). We compared GC responses induced by three different 60mer formulations: d30-60mer, d30-60mer carrying eOD-PADRE (d30-60mer-PADRE), and p60mer. Fusion of the T cell epitope to eOD-GT8 had no effect on the DNA-VLP formulation, glVRC01 Ramos cell activation, or MBL/C3 binding (Fig. 4A, fig. S2, and S10C).

Fig. 4. Incorporating synthetic helper T cell epitopes boosts GC size after DNA-VLP vaccination.

(A) Representative calcium signaling traces from glVRC01 Ramos B cells incubated with 5 nM eOD-GT8 equivalent concentrations of indicated particles from n=3 independent experiments. (B) C57BL/6 mice (n=8 mice/group) were primed s.c. with equimolar doses (equivalent to 5 ug of eOD) of the 60mer formulations with SMNP and sacrificed on day 14 for serum titer analysis and germinal center analysis by flow cytometry. This experiment was performed twice, and data from one experiment is shown. Anti-eOD-GT8 IgG titers were determined by ELISA (against eOD-GT8 without PADRE). Shown are AUC values (log10). (C) Anti-PADRE IgG titers determined by ELISA against biotinylated PADRE peptide. Shown are AUC values (log10). (D) Representative flow plots, frequencies (%GC out of total B cells), and counts of germinal center B cells (B220+/CD38lo GL7^hi). (E) Representative flow plots, frequencies (%Tfh out of CD4 T cells), and counts of Tfh cells (CD4⁺ /CXCR5⁺ PD1⁺). (F) Representative flow plots, frequencies (%eOD⁺⁺ out of GC B cells), and counts of antigen-specific eOD⁺⁺ GC B cells. Absolute counts plots are shown with log axes; geometric mean and geometric S.D. are shown. Data was analyzed with one-way ANOVA with post-hoc Tukey test.

We primed mice with the three 60mer formulations and the addition of PADRE enhanced early anti-eOD-GT8 IgG responses compared to d30-60mer (Fig. 4B), while no IgG responses were detected against the PADRE epitope itself (Fig. 4C). Mice primed with d30-60mer-PADRE had increased frequencies of GC B cells and Tfh cells compared to d30-60mer, though the total GC B cell count remained 4-fold lower than that elicited by LumSyn-scaffolded p60mer (Fig. 4D, E). However, GCs initiated by d30-60mer-PADRE expanded a much higher frequency of eOD-specific GC B cells compared to the p60mer, leading to higher total numbers of eOD-specific GC B cells compared to the p60mer (Fig. 4F). Thus, engineering of T cell help in DNA-VLPs enabled priming of a robust antigen-focused GC response.

DNA-VLPs prime epitope-focused germinal centers in humanized mice

We hypothesized that the increased frequency of antigen-specific B cells in GC activated by DNA-scaffolded compared to protein-scaffolded antigen may reflect an “immune focusing” effect due to the lack of scaffold-specific B cells. To test the implications of a more antigen-focused response on the maturation of bnAb precursor B cells, we immunized transgenic mice expressing the germline human IGHV1-2*02 gene segment knocked into the mouse Ig locus (V_H1-2 mice), which is paired with endogenous mouse light chains (81) (Fig. 5A). This mouse models the diversity of potential bnAb precursor B cells present in the human B cell repertoire, with a low frequency of bona fide bnAb precursors.

Fig. 5. DNA-VLPs produce focused germinal centers and prime VRC01-class precursors in humanized mice.

(A) Humanized V_H1-2 mice (n=8/group) were primed s.c. with 5 μg d30-60mer-PADRE or p60mer with SMNP, and inguinal lymph nodes were harvested on day 14. The lymphocytes were split into four samples and each stained with GC markers and fluorescently labeled antigen probes (eOD-60mer, bare lumazine synthase, or bare DNA-VLP (d30) probes). This experiment was performed twice, and data from one experiment is shown. (B) Representative GC B cell gating (left) and total counts of B220⁺/CD38^loGL7^hi GC B cells (right). (C) Representative eOD-60mer probe staining of GC B cells (left), eOD-CD4bsKO-60mer probe staining (center), and frequency of CD4bs-specific cells (eOD-60mer⁺⁺KO^- ) out of IgM⁻ IgD⁻ GC B cells (right). (D) Representative histogram of antigen probe staining in CD4bs⁻ specific GC B cells and quantification of probe staining MFI (right). (E) Representative LumSyn staining of GC B cells (left) and frequency of LumSyn⁺⁺cells out of IgM⁻ IgD⁻ GC B cells (right). (F) Representative DNA-VLP (d30) staining of GC B cells (left) and frequency of DNA-VLP⁺⁺ cells out of IgM⁻IgD⁻ GC B cells (right). (G) Fraction of germinal center B cells that bound to each set of antigen or scaffold probes represented as parts-of-whole plot (left) and ratio of CD4bs-specific B cells to competitor CD4bsKO⁺ and LumSyn⁺⁺ B cells (right). Statistical significance was determined using one-way ANOVA with post-hoc Tukey test for B-G.

We stained GC B cells with fluorescently-labeled eOD-60mer nanoparticle probes and an eOD-60mer-CD4bsKO nanoparticle probe, which contains mutations in the CD4bs that ablate binding by true epitope-specific B cells; binding to this KO probe identifies B cells specific for eOD epitopes outside the CD4bs (Fig. 5C). To quantify scaffold-specific competitor B cells, we stained GC B cells with fluorescently-labeled “bare” LumSyn nanoparticles or bare DNA-VLPs (Fig. 5D–E). As observed in WT mice, the p60mer primed overall slightly larger total GC responses (Fig. 5B). However, GCs elicited by d30-60mer-PADRE immunization were composed of nearly ~60% CD4bs-specific GC B cells, compared to ~20% expanded by p60mer (Fig. 5C). Furthermore, the epitope-specific B cells expanded by the DNA-VLP formulation exhibited brighter staining by the antigen probes (Fig. 5D). Following p60mer immunization, LumSyn-specific B cells contributed to ~13% of the GC, comparable in magnitude to the epitope-specific CD4bs-directed response (Fig. 5E). When staining with bare DNA-VLP probes, some background staining was detected in all groups, but this staining was the same in mice primed with either DNA-VLPs or p60mer, and similar low levels of background binding were detected on naïve B cells (Fig. 5F, fig. S13), possibly resulting from low-level scavenger receptor expression in B cells (82). Compared to the p60mer, DNA-VLP immunization promoted GCs that contained 25-fold higher ratios of CD4bs-specific B cells to off-target B cells (Fig. 5G), whereas in p60mer the frequency of epitope-specific B cells was enriched only 2-fold.

eOD display by DNA-VLPs leads to more efficient priming of bnAb precursors

We inquired whether the robust expansion of epitope-specific B cells in GCs produced by DNA-VLPs led to better expansion and/or maturation of VRC01-class precursors.VRC01-class precursors are identified as eOD-GT8 CD4bs-specific BCRs employing IGHV1-2*02 heavy chains combined with a 5-amino acid CDRL3 in the light chain (LC). These characteristics provide the best available measurement of B cells with the potential to evolve toward production of VRC01-like bnAbs. The V_H1-2 mouse model has a diversity of V(D)J-rearranged human V_H1-2 heavy chains and a fully murine LC repertoire, where only about 0.15% of B cells express a 5-amino acid CDRL3 (81). We sorted and sequenced B cells from naive V_H1-2 mice and measured the frequency of VRC01-class precursors in these animals to be ~0.0006%, or 6 in 1 million B cells (fig. S14A); VRC01-class B cells with diverse CDRL3s (fig. S14B) were present at a similar frequency and affinity as those of VRC01-class precursors in humans (fig. S14A, C) (8, 83, 84). We immunized V_H1-2 mice with d30-60mer-PADRE or p60mer, sorted class-switched CD4bs-specific GC B cells on day 14, and performed single-cell sequencing to analyze the BCR repertoire (fig. S54, table S4). A total of 444 and 465 BCR sequences were recovered and analyzed for d30-60mer-PADRE and p60mer, respectively.

We found diverse IGHV1-2*02 BCR clonotypes (defined as shared V_H + shared CDRH3) expanded by each immunogen (Fig. 6A–B, data S1). However, in DNA-VLP vaccination, several public clonotypes contained more expanded representatives (Fig. 6A), indicating proliferation of clonally related B cell lineages in the GCs. Further, within these clonotypes, DNA-VLP immunization enriched for the five amino acid CDRL3 VRC01 class signature (Fig. 6C), whereas the p60mer did not (Fig. 6D). The p60mer was previously reported to enrich for this signature in V_H1-2 mice after a single prime immunization, but expansion of these lineages required longer times (up to 8 weeks) (81). Within the clonotypes expanded by DNA-VLP vaccination, the five amino acid CDRL3 signature was enriched by expansion of public B cell clones (BCRs with shared V_H + CDRH3 + V_L + CDRL3 re-occurring within the independent vaccine recipients; Clone IDs - 70, 78, 116, 62, 19 and 137) (Fig. 6A, right, fig. S16). Expansion of multiple genetically identical B cell clones in different mice points to a reproducible mechanism for triggering of VRC01-class precursors with DNA-VLPs. The 5 AA CDRL3 included enrichment of the QQYXX motif (Fig. 6C) and was associated with usage of mouse IGKV4-61*01 (fig. S17), as described previously (41). We also observed SHM in the vaccine-expanded B cells primed with both formulations, but they were not different at this early timepoint. We also identified enrichment of key VRC01-class HC mutations (80) in both formulations, particularly S55R/N mutations (fig. S18).

Fig 6. BCR sequencing reveals DNA-60mers effectively prime VRC01-class precursors in early GCs.

(A) BCR sequencing analysis of CD4bs-specific GC B cells (B220⁺ /GL7^hi CD38^lo/ IgM⁻ IgG⁺/ eOD⁺⁺ KO) isolated from V_H1−2 mice (n=2) immunized with d30-60mer-PADRE at day 14 (Mouse 1 = 191 BCRs, Mouse 2 = 253 BCRs; 444 BCRs in total). This experiment was performed once. Shown are IGHV1-2*02 clonotypes; IGHV1-2*02-expressing public B cell clonotypes (shared VDJ in both vaccine recipients) are shown in color and the excised and asterisk-marked pie slices reflect public B cell clones (shared VDJ + VJ origin in both vaccine recipients) within each clonotype that also contain the short 5 amino acid CDRL3 signature of VRC01-class precursors and A phylogenetic analysis of these public clonal pathways within each excised slice is presented in fig. S12. (B) V_H1−2⁺ clonotype diversity in CD4bs-specific GC B cells from the VH1−2 mice immunized with p60mer at day 14 (n=2 VH1−2 mice; Mouse 1 = 263 BCRs, Mouse 2 = 202 paired BCRs; 465 BCRs in total). (C) Frequency of 5 amino acid CDRL3s in the IGHV1−2*02 CD4bs-specific GC B cells from d30–60mer-PADRE immunization (444 BCRs, n=2 mice). Sequence conservation in light chains represented as a logo plot showing enrichment of QQYXX from murine LC repertoire (44, 64). (right). (D) Frequency of 5 amino acid CDRL3s in the IGHV1−2*02 CD4bs-specific GC B cells from p60mer immunization. Expansion of 5aa CDRL3 was not detected by the p60mer at this early timepoint.

We repeated immunizations of V_H1-2 mice with DNA-VLPs or p60mer (8 mice per immunization group) and carried out 10x sequencing of CD4bs-specific GC B cells 2 weeks post immunization. This additional analysis revealed increased clonal expansion of recovered B cells (fig. S19A), and a readily detectable population of IGHV1-2*02-expressing B cells with 5 AA CDRL3 elicited by DNA-VLPs that was nearly absent following p60mer immunization (fig. S18B). Within these short CDRL3s, the CQQYXX motif was also enriched for these VRC01-class B cells expanded by the DNA-VLP (fig. S19C), and a similar total number of SHM mutations was observed between both groups at this early time post-prime (fig. S19D, E). Thus, epitope-focused GCs primed by DNA-VLPs can augment the recruitment of bnAb precursors.

Discussion

Anti-scaffold B cell responses have been reported for almost all protein-based nanoparticle scaffolds (18, 28, 29, 31, 85, 86). Thus far, anti-scaffold antibodies have been thought to be minimally concerning for immunodominant antigens (28), or even beneficial by improving antigen capture in lymph nodes or masking scaffold epitopes in boosts (38, 87). However, for subdominant antigens like HIV Env, anti-scaffold responses can dominate the serum antibody response (28), and it remains unclear whether they influence the B cell response evolving in germinal centers. Strategies such as glycosylation (32, 34), PEGylation (28, 35), and PASylation (28) have been introduced to reduce scaffold-specific responses. Even with perfect masking of exposed scaffold epitopes, degradation of these nanoparticles by tissue proteases present in lymph (65, 66) may expose additional epitopes that elicit off-target scaffold-specific responses. This has been observed with p60mer, which elicits robust anti-LumSyn B cells responses in mice and humans, even though intact particle probes mostly bound B cells that are not LumSyn-specific (32), suggesting that LumSyn-specific antibodies arise from partial degradation of nanoparticles in vivo.

Here, we used inert DNA origami vaccine scaffolds to assess the impact of distracting scaffold epitopes on the potency of antigen-specific GC responses. We found that high antigen valency and density on DNA-VLPs was critical for promoting complement-mediated follicle trafficking and expansion of antigen-specific B cells in GCs. The optimal d30-60mer nanoparticle design closely mimicked the geometry of the clinical p60mer (8), and both particles were effectively recognized by murine MBL. However, C3 deposition on DNA-VLPs and overall accumulation of antigen on FDCs was lower compared to the protein nanoparticle, which may arise due to faster degradation of DNA-VLPs or due to their porosity, specific surface chemistry and negative charge of the origami structure, and/or interaction of DNA-VLPs with complement binding proteins which may inhibit downstream complement activation (75, 88). Second, we found that augmenting the small eOD-GT8 immunogen with synthetic T cell epitopes substantially improved GC priming without introducing off-target epitopes for B cells.

Given that the optimized DNA-VLP 60mer closely mirrored the clinical p60mer nanoparticles in its antigen valency, spacing, geometry, and presence of T cell help, we assessed the effect of the scaffold material on clonal competition in GC responses in a physiologically relevant V_H1-2 mouse model (81). GCs expanded by DNA-VLPs were more epitope-specific than those induced by p60mer, where a substantial frequency of LymSyn-specific GC B cells was observed. We hypothesize that epitope focusing achieved by DNA-VLPs is promoted by the fact that T-cell help will be provided primarily to eOD-specific B cells, while following p60mer immunization, antigen- and scaffold-specific GC B cells will compete for Tfh help.

BCR sequencing analysis confirmed that DNA-VLP immunization enhanced the priming of bnAb precursors compared to p60mer. These cells were enriched for the CQQYXX CDRL3 motif important for VRC01-class bnAbs. However, we observed no differences in SHM, nor enrichment of the X65E mutation that plays an important role in recognition of loop D in Env (21, 89), likely due to the early time point at which we sampled GC B cells. It will be an important direction for future work to determine if DNA-VLPs can elicit this and other further VRC01-class mutations at later time points post priming or following boosting with shepherding immunogens.

The preferential recruitment of clones bearing 5-amino acid CDRL3s warrants further investigation, as it may reflect reduced competition from scaffold-specific B cells, enabling broader participation of epitope-specific clones in the GC and increasing the likelihood of priming rare VRC01-class precursors. However, it may also be possible that lower complement opsonization of DNA-VLPs compared to p60mer reduces potential partial masking of the CD4bs by complement, facilitating greater access for these structurally constrained BCRs (90). Interestingly, human vaccination with mRNA-LNP encoding eOD-GT8 60mer followed by a heterologous LumSyn boost was efficient in both priming and boosting of VRC01-class responses, despite pre-existing serum reactivity to LumSyn and strong increases in serum scaffold-specific responses with each vaccination (21). Our data suggests that even for a very effective immunogen like eOD-GT8, further gains in bnAb precursor recruitment can be made through further engineering designed to reduce competitor responses.

Several important issues related to potential clinical translation of DNA-VLPs merit future work, including their stability against degradation by extracellular nucleases. Additionally, while in our work we did not detect any class-switched anti-DNA or -VLP antibody responses, autoimmune risks posed by antibody responses to DNA/protein hybrids must be carefully evaluated.

Materials and Methods

Mice

Six-to-ten-week-old female C57BL/6 mice were purchased from The Jackson Laboratory (strain no. 000664). V_H1-2 mice (gift courtesy of Dr. Frederick Alt) were bred in-house and genotyped using Transnetyx. DNase I KO mice were provided by the Reizis lab at NYU Langone (original source CMMR MGI:103157) and rederived for subsequent experiments. For DNase I KO experiments, heterozygous mice were bred, pups were genotyped using PCR (primers provided in Supplementary Information), and wildtype littermates were used as controls. Mice were housed at 5 animals/cage in pathogen-free conditions with standard diet and were euthanized by CO₂ asphyxiation followed by cranial dislocation. All animal studies were performed under an institutional animal care and use committee-approved animal protocol (MIT CAC protocol # 2303000488) following local, state, and National Institutes of Health guidelines for the care and use of animals. This research does not include human participants.

Study Design

Animals were randomly assigned to groups, and blinding was not used due to logistical limitations. No statistical methods were used to pre-determine sample sizes for animal studies, which were often limited by mouse colony size and cost of analyses (group size ranges from n=2 to n=8, depending on experiment). Our sample sizes were similar to those previously reported (30, 66, 91), anticipating similar effect size. All analytical flow cytometry experiments and confocal microscopy experiments were performed at least twice, with the exception of DNase I KO experiments due to limits in animal cohorts. BCR sequencing experiments were each performed one time due to cost of analysis.

ssDNA scaffold synthesis

Custom-length DNA scaffolds for DNA-VLPs were synthesized as previously described (92). Briefly, SS320 E. coli cells were co-transformed with the corresponding plasmid and the M13cp helper plasmid (originally provided by Andrew Bradbury, Los Alamos National Laboratories). Pre-cultures of the transformed cells were grown overnight at 37 °C in 2x YT medium containing 100μg/ml ampicillin and 15 μg/ml chloramphenicol). Cells were pelleted by centrifuging three times at 4000 g for 3 min and subsequently discarded. Phage was precipitated from the supernatant in presence of 6% (w/v) PEG_8kDa and 3% (w/v) of NaCl by stirring at 4 °C for 1 h and harvested by centrifugation at 20,000 g at 4 °C for 1 h. After resuspension in TE buffer, ssDNA was extracted via the EndoFree GigaPrep purification protocol with the following modifications: Proteinase K was added to buffer P1 followed by incubation at 37 °C for 1 h, addition of buffer P2 and incubation at 70 °C for 10 min. After ssDNA purification, Triton X-144 was used to remove residual endotoxins to levels less than 0.2 EU/μmol DNA scaffold, corresponding to less than 0.000015 EU/injection into mice. Endotoxin levels were measured using ToxinSensor Gel Clot Endotoxin Assay Kits. Purity of the scaffolds was analyzed by agarose gel electrophoresis (AGE)(1.6% agarose, TAE buffer with 12 mM MgCl₂, EtBr, 60 V for 150 min at room temperature).

Oligonucleotide synthesis

DBCO-modified oligonucleotides were fabricated in-house. A (#10-1000), C (#10-1010), G (#10-1020), T (#10-1030), and DBCO-TEG (#10-1941) phosphoramidites and ancillary synthesis reagents were acquired from Glen Research and used following the manufacturer’s recommendations. Phosphoramidites were dissolved and diluted in anhydrous acetonitrile to 0.1 M prior to synthesis. Oligonucleotides were synthesized on a Biolytic Dr. Oligo 192c oligonucleotide synthesizer, at a 200 nmol scale with CPG 1000 Å standard base supports (#20-2001, #20-2011, #20-2021, #20-2031 from Glen Research) in normal mode under nitrogen. Synthesis success and yield was observed by monitoring the penultimate trityl-cleavage. Following the synthesis, oligonucleotides were cleaved and deprotected under pressure at 55°C for 2 hr in 30% ammonium hydroxide (Thermo Fisher Scientific, cat# 423305000,). The cleaved strands were desalted with acetonitrile and eluted in nuclease-free water. The DBCO-modified oligonucleotides were purified by Glen-Pak^™ DNA 30 mg in a 96-Well Plate (#60-5400-01) under reduced pressure following the manufacturer’s protocols. After purification, the oligonucleotides were dried and resuspended in 1x TE and stored at 4°C. Oligonucleotide concentrations and yields were determined via UV-Vis measurements (Tecan Spark). Oligonucleotide purity was confirmed by HPLC (Waters Acquity System).

Immunogen synthesis

eOD-GT8 monomer (with N-terminal 6x-His and cysteine), eOD-PADRE monomer (with N-terminal 6x-His and cysteine and C-terminal PADRE), and eOD-60mer were synthesized as previously reported (41). Briefly, plasmids were transiently transfected into Expi293F cells. After 5 days of culturing in conditions described above, cell culture supernatants were collected and protein was purified in an AKTA pure chromatography system using HiTrap HP Ni sepharose affinity column, followed by size exclusion chromatography using Superdex 75 Increase 10/300 GL column (GE Healthcare). Endotoxin levels in purified protein were measured using Endosafe Nexgen-PTS system (Charles River) and were ensured to be less than 5EU/mg protein. The eOD-60mer was affinity purified by incubating with Galanthus nivalis lectin-conjugated agarose beads (Vector Laboratories, AL-1243-5,) overnight under gentle agitation at 4°C and eluted with lectin elution buffer containing 1 M methyl a-D-mannopyranoside (Millipore Sigma, M6882,). The resulting solution was dialyzed in PBS and further purified by size-exclusion chromatography using Superdex 200 Increase 10/300GL. Bare lumazine synthase was produced by BlueSky Bioservices as previously described (31). Briefly, lumazine synthase from Aquifex aeolicus was expressed in E.Coli, cells were lysed, supernatant was heat-treated for 30 min at 75°C and supernatant was again clarified by centrifugation. Fully assembled particles were purified by two successive size-exclusion chromatography steps, using Superdex 75 and Sephacryl 500 columns, respectively.

To add a terminal azide for SPAAC reactions, eOD-Cys monomers were reduced with 5 molar excess of TCEP, then desalted into PBS + 10 mM EDTA using Zeba 7K MWCO desalting columns (Thermo Fisher, cat# 89883). The number of free cysteines was confirmed using an Ellman’s assay. The azide linker was formed by reacting 60 mM SMCC (Thermo Fisher Scientific, cat #A35394) with 1.1x excess of Azido-PEG3-Amine (Broadpharm, cat#BP-20580) for 1 hour at room temperature. Reduced antigen was reacted with 5 molar excess of azide linker for 4 hours at room temperature, and unreacted linker was then removed by spin filter centrifugation in Amicon 10K MWCO centrifugal filters (Millipore Sigma, cat# UFC801008). Antigen was concentrated to approximately 300 μM using spin filter centrifugation (Amicon 10K MWCO) for SPAAC reactions.

DNA-VLP synthesis and characterization

DNA-VLPs were assembled as previously described. All designs were generated using DAEDALUS (47) (d40 VLP) or ATHENA (56) (d30 VLP) with staples manually adjusted in Tiamat software to have outward nick positions (2 per edge). To fold origami, 30 nM of scaffold was mixed with 5–10x excess of each oligonucleotide staple in TAE buffer with 12 mM MgCl₂ and thermally annealed as follows: 95°C for 5 min, 80–75°C at 1°C per 5 min, 75–30°C at 1°C per 15 min, and 30–25°C at 1°C per 10 min. DNA-VLPs were purified into PBS using Amicon Ultra 100 kDa centrifugal filters (Millipore Sigma, cat #UFC810024) spun at 2000g, concentrated to ~1.5 μM, and stored at 4 °C. Concentration of the DNA-VLPs was determined by triplicate A260 measurements from a Nanodrop Spectrophotometer. Purity and dispersity of DNA-VLPs were validated by AGE (1.6% molecular biology grade agarose, TAE buffer with 12 mM MgCl₂, SYBR Safe, 65V for 150 min) and dynamic light scattering. For antigen conjugation, the concentrated stock of eOD-Azide was diluted to an appropriate concentration and added to the VLPs at 1:1 v/v so that final molar equivalents of Azide: DBCO was at least 3:1, yielding final concentrations (22.5 μM DBCO and 67.5 μM eOD-Azide (for 30mers) and 45 μM DBCO and 135 μM eOD-Azide). Excess antigen was removed by drop dialysis into 1X PBS for 16 hours (Millipore Sigma, mixed cellulose ester, 0.025 μm), and the concentration of purified nanoparticles was determined by A260 measurements, after which they were diluted to 100 nM and stored at 4°C. Antigen functionalization was validated using AGE (1.6% agarose, TAE buffer with 12 mM MgCl₂, SYBR Safe, 65V for 150 min) and quantified using bicinchoninic acid colorimetric assay with a standard curve made from known dilutions of eOD-GT8. Functionalization efficiency (FE) was calculated as the determined eOD concentration/theoretical eOD concentration assuming full conversion based on # of DBCO oligos) in accordance with equation eq. S1 in the Supplementary Text.

Negative stain transmission electron microscopy

Uranyl formate staining was performed as previously described (30). Briefly, DNA-VLPs were diluted to 3–5 nM in PBS and 5 μL was applied onto glow-discharged electron microscopy grids. After 30 seconds, the grids were blotted with filter paper and washed with 5 μL of fresh 2% uranyl formate solution containing 5 mM NaOH for 30 seconds. The uranyl formate solution was removed by blotting on filter paper, and the grids were stored with a desiccant before imaging. TEM was performed on an FEI Tecnai G2 Spirit Twin. Shown are images at 30K, 42K, or 67K magnifications.

CryoEM imaging

Three microliters of the folded and purified DNA-VLP solution (approximately 300 nM) was applied onto the glow-discharged 200-mesh Quantifoil 2/1 grid, blotted for four seconds and rapidly frozen in liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific). Grids were screened and imaged on a Talos Arctica cryo-electron microscope (Thermo Fisher Scientific) operated at 200 kV at a magnification of 79,000× (corresponding to a calibrated sampling of 1.76 Å per pixel). Micrographs were recorded by EPU software (Thermo Fisher Scientific) with a Gatan K2 Summit direct electron detector in counting mode, where each image is composed of 24 individual frames with an exposure time of 6s and a total dose ~63 electrons per Ų. We used a defocus range of −1.0 to −2.5 μm to collect images, which were subsequently motion-corrected using MotionCor2. CTF estimation was performed in RELION (v3.0.8) using the CTFFind 4.1 algorithm. Particles were manually picked from motion-corrected micrographs using the RELION (v3.0.8) manual picking GUI. Extracted particles were boxed at 320 pixels without down-sampling and normalized against background (pixels outside 256 pixel-diameter circle) prior to classification. Reference-free 2D classification to generate 50 classes was performed in RELION with 25 iterations, using 14,000 picked particles and a regularization parameter of T=2. The final class averages were assessed visually, and a subset of well-resolved classes showing characteristic DNA origami features was selected for presentation.

In vitro B cell activation assays

Calcium flux assays were conducted as previously described (46). Ramos B cells expressing germline VRC01 BCRs (provided by Lingwood lab, Ragon Institute) were stained with 10 μM Fluo-4 AM (Thermo Fisher Scientific, cat# F14217) for 30 min at 37 °C. After washing twice in serum-free RPMI medium, calcium flux assays were performed in triplicate on a Tecan plate reader at 37 °C on a 96-well microplate with 160 μl of Fluo-4 labelled Ramos cells at 2 million cells per ml. A baseline fluorescence was then recorded for 1 min, and 40 μl of antigens were added to the cells for a final concentration of 5 nM of antigen. Raw calcium traces were normalized to a common baseline by subtracting the PBS time trace at every time point, then dividing the time trace at every point by the average of the time points before antigen addition.

Dye labeling of nanoparticles for imaging and flow probes

Protein antigens (eOD-GT8 monomers, eOD-60mer, and lumazine synthase) at 1 mg/mL were desalted into PBS and diluted 1:1 with 0.2M sodium bicarbonate buffer. Dye stocks (AF488-NHS-ester, AF647-NHS ester, AF555-NHS ester) (Thermo Fisher Scientific, cat# A20000, A20006, A20009) were prepared in DMSO, and added at 5-molar excess and reacted with antigens at 1 hour at room temperature. Due to lower efficiency of labeling lumazine synthase, reactions were performed overnight with shaking (500 rpm). Excess dye was removed by passing proteins through Zeba Spin Desalting Columns (Thermo Fisher Scientific). DNA-VLPs (d30) used as flow probes were formulated with 6 DBCO-modified oligos and reacted with 10 molar excess of AF488-Azide or AF555-Azide triethylammonium salt (Thermo Fisher Scientific, cat# A10266, A20012). Excess dye was removed by drop dialysis into 1X PBS for 16 hours (Millipore Sigma, mixed cellulose ester, 0.025 μm). For flow staining, eOD-tetramers were produced by biotinylating Avi-tagged eOD-GT8 monomers (without PADRE) with BirA (Avidity, Inc) and mixing an excess of biotinylated eOD-GT8 monomer with fluorescently labeled Streptavidin-PE, Streptavidin-AF647, or Streptavidin-BV421 (BioLegend cat# 405203, 405237, 405226) step-wise on ice.

Immunizations

SMNP adjuvant was prepared as previously described (59). Antigen formulations were prepared by diluting 5 μg of the indicated antigens mixed with 5 μg of SMNP adjuvant per mouse in 1X PBS and gently mixed with a pipette. Mice (6-10 weeks old) were shaved at the tail base and immunized subcutaneously at the left and right sides of the tail base (50 μL per side). For adjuvant screening, the following formulations were used and prepared for 100 μL total volumes per mouse: SMNP (5μg/mouse), alum in the form of Alhydrogel (InvivoGen, 100 μg/mouse), AddaVax (InvivoGen, 1:1 v/v), AS01b (GSK, 1:1 v/v), ODN 1826 (IDT, 50 μg/mouse).

Flow cytometry analysis of GC B and Tfh cells

Mice were euthanized at the indicated time point post-injection by CO₂ asphyxiation, and inguinal lymph nodes were collected and mechanically digested into a single-cell suspension using Biomasher tubes and a motorized tissue grinder. Lymphocytes were strained twice through 60 μm Multi-Screen Mesh filter plates (Millipore Sigma, cat# MANMN6010) and washed once in 1X PBS. The cells were then stained with 1:750 dilution of Zombie UV Live Dead Stain (BioLegend, cat# 423107) diluted in PBS for 10 minutes at room temperature. Excess dye was removed by washing once with FACS buffer (1X PBS, 2% FBS, 0.01% sodium azide). The cells were then resuspended in 25 μL FACS buffer containing Fc block (BioLegend, cat# 156603) and incubated on ice for 15 minutes. Antibody mixes (2X stock) were prepared in FACS buffer and contained the following antibodies: anti-CD4 BUV737 (BD Biosciences, RM4.5, cat #612844), anti-B220 APC-Cy7 or PE-Cy7 (BioLegend, RA3-6B2, cat# 103224 and #103222), anti-CD38-AF488 or BV421 (BioLegend, 90, cat# 102714 and #102732), anti-GL7 PerCpCy5.5 (BioLegend, GL7, cat# 144610), anti-CXCR5 BV605 (BioLegend, L138D7, cat# 144513), anti-PD1-AF647 or BV421 (BioLegend, 29F.1A12, cat# 135230 and #135221), and antigen tetramer probes. For the experiment shown in Fig. 5, anti-IgM-PE-Cy7 (BioLegend, MA-69, cat# 408610) and anti-IgD-BV785 (BioLegend, 11-26c.2a, cat# 405757) were also included in the antibody cocktail. For cell staining for cell sorting, an additional anti-IgG-BV605 was included to gate on class-switched GC B cells (BD Biosciences, clone X56, cat# 742477). Tetramers were added to antibody mix so that the final amount per sample was 50 ng. For nanoparticle probes (Fig 5 & 6), nanoparticle probes were added at 5 ng/ sample. The cells in Fc block were mixed with antibody and probe mixture and incubated for 30 minutes on ice, then washed twice with FACS buffer. The cells were fixed with 2% PFA, washed once and stored in FACS buffer until flow cytometry analysis. The cells were transferred to U-bottom 96-well plates and mixed with CountBright Absolute Counting beads (Thermo Fisher Scientific, cat# C36950) before analysis. Flow cytometry was carried out on a BD LSR Fortessa or Symphony A3 cytometer in plate-mode. Full-minus-one (FMO) staining controls were included in every experiment for drawing gates.

Lymph node microscopy

For lymph node histology, mice were immunized with 5 μg of fluorescently labeled eOD-GT8 monomer, DNA-VLPs, or protein nanoparticles and sacrificed at the indicated time points. Mice were euthanized by CO₂ asphyxiation, and inguinal lymph nodes were flash frozen in OCT embedding medium and sectioned on a Leica Cryostat (10 μm thick sections) and stored at −80°C. For immunostaining, the sections were quickly thawed, fixed with 10% neutral buffered formalin, and blocked/permeabilized with PBS containing 2% BSA and 0.01% Triton-X. The sections were then stained with 1:100 dilutions of anti-CD35-BV421 (BD Biosciences clone 8c12, cat#740029), anti-CD169-AF488 (BioLegend clone 3D6.112, cat# 142419), or anti-F4/80 (BioLegend clone BM8, cat# 123131) in block/perm buffer for 1 hour in a humidity chamber. The slides were washed three times in PBS and mounted with ProLong Diamond Antifade (Thermo Fisher Scientific, cat# P36970) and secured with sealant. Slides were stored in the dark and imaged on a Leica Sp8 Laser Scanning Confocal Microscope with 25x water-based objective.

For cleared lymph node imaging, mice were immunized with 5 or 10 μg of fluorescently labeled nanoparticles (specified in captions). For in situ follicle labeling, 4 μg of anti-CD35-BV421 (BD, clone 8c12, cat#740029) was injected subcutaneously 12–16 hours prior to tissue harvesting. Inguinal lymph nodes were isolated and fixed in 4% PFA for 24 hours and cleared using DISCO as previously described (20). Briefly, LNs were washed twice in PBS and excess fat and connective tissue were removed. Nodes were then gradually moved into solutions containing successively higher concentrations of methanol until they were incubated for half an hour in pure methanol. Nodes were then bleached in hydrogen peroxide for one minute before being returned to methanol for half an hour. They were then gradually transferred into solutions containing increasing concentrations of tertiary-butanol before eventually being incubated in pure tertiary-butanol 0.4% α-tocopherol for one hour. Nodes were then removed from solution and allowed to dry completely before being placed in dichloromethane. After the lymph nodes dropped to the bottom of tubes following swirling, they were stored in dibenzyl ether with 0.4% α-tocopherol. Follicles were imaged using an Olympus FV1200 Laser Scanning Confocal Microscope at 10x magnification over a 300 μm distance. Lasers were set to minimize pixel saturation in the brightest samples. Images were analyzed using ImageJ software. Antigen occupancy was quantified by applying a Gaussian filter to each channel to remove auto pixels, creating a binary mask to identify antigen area, FDC area based on CD35 or SSM area based on CD169 signal, and calculating the fraction of area occupied by antigen signal, as previously described (20). Representative binary masks of follicle areas are shown in fig. S11A.

Passivation of DNA-VLPs with polylysine-PEG

DNA-VLPs (d40-30mer) were coated with polylysine-PEG1K or 5K (Alamanda Polymers) at 1:1 N:P ratio, as previously described (72), and complexes were analyzed by AGE (1.6% agarose, TAE buffer with 12 mM MgCl₂, SYBR Safe, 60V for 150 min at room temperature). For serum stability analysis, d40-30mer was incubated with or without polymer coating at 37°C in the presence of 10% mouse serum (freshly isolated from C57BL/6 mice) and analyzed by AGE as described previously. For macrophage association, RAW264.7 murine macrophages (ATCC) were seeded into 24-well plates and incubated with 10 nM AF647-labeled d40-30mer with or without PEG coating. After 2 hrs of incubation, the cells were thoroughly washed, and nanoparticle signal was analyzed by flow cytometry. Lymph node imaging and measurement of GC responses induced by vaccination with uncoated or coated DNA-VLPs were performed as described above.

ELISA analysis of serum antibody responses

Blood samples were collected from immunized mice via retro-orbital or submandibular bleeds and serum was isolated using centrifugation through Serum Gel tubes (Sarstedt) and stored at −20°C. MaxiSorp plates (Thermo Fisher Scientific) were coated with 2 μg/ml protein immunogens (eOD-GT8 monomer, eOD-CD4bs-KO monomer, or Lumazine Synthase) overnight at 4°C. For anti-dsDNA and anti-VLP ELISAs, ELISA plates were coated first with 100 ug/mL poly-D-lysine in 1X PBS and incubated at 37°C for 1 hr. The plates were washed once with PBS and then incubated with 10 μg/mL of bare DNA-VLPs (d30) or calf-thymus DNA (Sigma Aldrich) overnight at 4°C. The plates were blocked with Casein block buffer (G-Biosciences). Plates were washed four times in 1x PBS containing 0.2% Tween-20, and dilutions of serum in blocking buffer were added and incubated for two hours. A commercial dsDNA antibody (Abcam, clone 35I9, cat# ab27156) was used as a positive control in anti-DNA assays, and murine VRC01 was included in all anti-eOD-GT8 ELISAs. Plates were washed four times and an HRP-conjugated goat anti-mouse IgG (BioRad, cat# 31430) was added and incubated for one hour. Plates were then washed and TMB was added. The reaction was stopped with sulfuric acid once the wells containing the lowest dilutions of TMB began to develop visually and the absorbance of each well was determined. All titers reported are inverse dilutions where A450nm – A540nm (reference wavelength) equals 0.2 or as area under the curve measurements (specified in figure captions). Anti-DNA IgM ELISAs were performed using a commercial assay (Alpha Diagnostic International, cat# 5130) following manufacturer instructions.

C3 and MBL ELISAs

ELISA plates were coated with 100 μg/mL poly-D-lysine in 1X PBS and incubated at 37°C for 1 hr. The plates were washed with DPBS, then coated with nanoparticles overnight at 4°C (DNA-VLP concentrations were normalized so that final eOD concentration was 1 μg/mL). The plates were blocked with 1X PBS containing calcium and magnesium with 1% BSA, then washed 4 times with 1X PBS containing 0.2% Tween-20. Fresh mouse serum was collected from naive C57BL/6 mice in Sarstedt serum gel tubes and kept on ice. Serum was diluted in blocking buffer (starting at 30% v/v) and further diluted in 1:2 dilution series. Serum dilutions were transferred to ELISA plates and incubated for 1 hr at 37°C. For recombinant mouse MBL assays, MBL2 (Biotechne R&D, cat #2208) was diluted in blocking buffer starting at 100 ug/mL concentration and diluted 1:2. The plates were washed 4 times as before, then incubated with 2 μg/mL rat anti-mouse MBL antibody (Abcam, clone 14D12, cat# ab106046) or rat anti-mouse C3 antibody (Abcam, clone 11H9, cat# ab11862) diluted in blocking buffer. The plates were washed four times, then incubated with 1:5000 HRP-conjugated mouse-adsorbed goat anti-rat IgG (Bio-Rad, cat# STAR72). Plates were then washed and TMB was added. The reaction was stopped with sulfuric acid once the wells containing the lowest dilutions of TMB began to develop visually and the absorbance of each well was determined. Complement deposition is reported as an area under the curve.

Single cell BCR sequencing of naïve V_H1-2 mice

For quantification of VRC01-class precursor frequency in V_H1-2 mice, ten V_H1-2 mice were euthanized using 100% compressed CO₂ in a clear chamber to allow visual monitoring of respiration and confirmation of death via respiratory cessation. Spleens were harvested and placed in 3 mL of resuspension buffer (1× PBS without Ca²⁺/Mg²⁺, 1 mM EDTA, 25 mM HEPES, pH 7.0, 1% heat-inactivated fetal bovine serum) in 15 mL polypropylene tubes on ice. Spleens were mechanically dissociated using the rough edges of two sandblasted microscope slides in a 5 mL petri dish, then transferred back to the same tubes for centrifugation at 460 × g for 5 minutes at 4 °C. Red blood cells were lysed by incubating samples with 1 mL of ACK buffer (Quality Biological, Cat# 118-156-721) for 2 minutes on ice. Lysis was quenched by adding 14 mL of resuspension buffer to each sample. Post lysis cells were pelleted at 460xg for 5 minutes and resuspended in 5 mL of resuspension buffer prior to filtration through a cotton-plugged, borosilicate Pasteur pipette into a borosilicate glass test tube. Samples were stored on ice until preparation for FACS. Samples were enriched for B cells, stained, and sorted for antigen specificity (CD19⁺/IgD⁺/IgM⁺/eOD-GT8⁺⁺) by FACS as previously described (84). BCRs from sorted antigen specific naïve B cells were sequenced using the 10X Genomics Single Cell Immune Profiling platform as previously described (84). Raw sequencing data were demultiplexed and processed into assembled VDJ contigs and count matrix files using Cell Ranger (v6.1). Individual cells were assigned to specific animal IDs based on TotalSeq-C antibody hashtag counts using and scab, as previously described (93). Gene assignment, sequence annotation, and formatting of paired heavy and light chain antibody sequences into the Adaptive Immune Receptor Repertoire (AIRR) format were performed using the Sequencing Analysis and Data library for Immunoinformatics Exploration (SADIE) (31), with a custom V_H1-2 mouse germline reference database. Naive VRC01-class mAbs were produced and measured for affinity to eOD-GT8 using a Carterra LSA as previously described (84).

Single cell BCR sequencing of V_H1-2 mice immunized with DNA-VLPs or p60mer

For evaluation of VRC01-class responses after immunization with DNA-VLPs or p60mer using Smart-Seq2 (Fig. 6), CD4bs-specific GC B cells were FACS-sorted from inguinal lymph nodes harvested from V_H1-2 mice immunized subcutaneously with 5 μg p60mer or d30-60mer-PADRE, gated on B220⁺/CD38^loGL7^hi/IgM⁻IgG⁺/eOD⁺⁺KO⁻ populations, stained with antibodies as described in the above flow cytometry section. We generated BCR libraries from whole transcriptome amplification (WTA) products produced using the Smart-Seq2 protocol on the FACS isolated B cells (94). The WTA products underwent two 0.8x (v/v) SPRI bead-based cleanups and were verified using High Sensitivity D5000 ScreenTape (Agilent Technologies Inc) and quantified and normalized using the Qubit dsDNA HS Assay kit (Thermo Fisher Scientific, cat # Q32854). To enrich the BCR sequences (FR1 to CDR3), corresponding heavy and light chains were amplified (HotStarTaq Plus Master Kit, Qiagen, cat #203645) with a pool of partially degenerate primers specific against all possible IGHV (human) or IGLV (mouse) and IGKV (mouse) segments in the FR1 region (final concentration: 10 μM each) and reverse primers against the heavy or light constant regions (final concentration: 10 μM each)(94). These primers were also built with attachments to the Illumina P7 (V region) and P5 (constant region) sequences. The amplicons were quantified and normalized following BCR amplification and SPRI cleanup, after which cellular barcodes and Illumina sequencing adapters (Nextera XT Index Adapters, Illumina Inc.) were added to each amplified heavy and light chain using step-out PCR (Kapa HiFi HotStart ReadyMix; Fisher Scientific cat # 50-196-5217). After another SPRI cleanup the HC and LC samples were pooled and the single-cell BCR libraries were sequenced via paired-end 250x250 reads and 8x8 index reads on an Illumina MiSeq System (MiSeq Reagent Kit v2 (500-cycle), cat# MS-102-2003). The BCR heavy and light chains reads were then paired using the barcodes and the overlapping sequencing reads were reconstructed with PandaSeq (95), and aligned against the human or mouse IMGT database (96). Sequencing error correction was performed with MigMAP, a wrapper for IgBlast (https://github.com/mikessh/migmap). Consensus V_H and V_L/V_K chain for each single cell was achieved by collating all reads with the same CDR3 sequence and then calling the top heavy and light chain sequences by frequency. The BCR nucleotide sequences were then aligned using the ClustalW alignment tool in the MEGAx (97) and further subdivided into clonotypes (= shared VDJ: VH1-2*02 usage + shared CDRH3 sequence) and clonal lineages (= shared VDJ/VJ: VH1-2*02 usage + shared CDRH3 sequence + shared VL + shared CDRL3 sequence)(91). Public BCR features were defined as the same clonotype and/or clone expanded in more than one vaccine recipient. Phylogenetic trees for the public clones were constructed using the neighbor-joining method with Poisson Model in the MEGAx (97). The reliability of the tree nodes was tested using the Felsenstein bootstrap method with 500 replicates.

For 10x analysis of B cells from immunized V_H1-2 mice (fig. S18), CD4bs-specific GC B cells were FACS-sorted from inguinal lymph nodes harvested from VH1-2 mice immunized subcutaneously with 5 μg p60mer or d30-60mer-PADRE. LNs were mechanically dissociated, filtered, and stained with NIR live dead stain, followed by a cocktail of primary antibodies and fluorescein-labeled eOD-60mer probes. Cells were also stained with Totalseq C anti-mouse hashing antibodies (BioLegend, 1:100) before FACS. CD4bs-specific GC B cells were gated on live B220+/CD38loGL7hi/IgG+/eOD++KO− populations. Cells were subsequently processed according to the 10x Genomics 5’ Immune Profiling v3 protocol. The gene expression libraries, VDJ libraries, and cell hashing libraries were pooled according to the manufacturer’s recommendation and sequenced by Illumina Nextseq 500. The sequenced gene expression libraries were aligned to the GRCm39 reference genome and quantified using the Cell Ranger Multi pipeline v9.0.1 along with the cell hashtag oligo (HTO) libraries. Data processing, analysis, and visualization were done using Python v3.10.0, R v4.3.0, Seurat v5.0.1, and ggplot2 v3.5.1. The gene expression count matrix was processed using the Seurat (v5.0.1) package in R. The initial quality control filtered out genes that were detected in fewer than 3 cells and removed cells with fewer than 100 genes. Cells were normalized using the NormalizeData() function. The HTO count matrix was added to the Seurat object and normalized. The HTODemux() function was used to assign HTO to each cell. Only singlet cells by HTO assignment were kept for downstream analysis. The variable genes were identified using the FindVariableFeatures() function. The ScaleData() function was used to regress out RNA feature counts and percent of mitochondrial genes before performing principal component analysis (PCA) using the RunPCA() function. For the initial cell lineage analysis, thirty principal components (PCs) and 500 decision trees were used for constructing the nearest-neighbor graph with the FindNeighbors() function. Thirty neighboring points and twenty PCs were used to generate uniform manifold approximation and projection (UMAP) with the RunUMAP() function. Unsupervised clustering was determined using Louvain clustering as implemented in the FindClusters() function. The VDJ library sequence reads were constructed into BCR contigs using the Cell Ranger VDJ de novo mode. The annotation of contigs followed a three-step process: 1) IgBLAST v1.14.0 was used to annotate V(D)J genes and CDR regions based on IMGT Mus musculus v3.1.42 database, 2) a custom R script performed local alignment against the human IGHV1-2*02 gene, using the pairwiseAlignment() function from Biostrings package with the same scoring metric as that of IgBLAST, and 3) the murine V gene assignment for contigs with high alignment scores was replaced by IGHV1-2*02. Cellular BCR contig consensus was established by ranking the contigs by IGHV1-2*02 alignment score (only for heavy chain contigs), UMI count, and read count (in this order) and picking the top contig.

To quantify somatic hypermutation (SHM) counts, we developed a computational pipeline that reconstructs germline V(D)J sequences for each cell and measures nucleotide mismatches relative to observed contig sequences. The germline sequences for each cell were reconstructed by concatenating the full-length V and J gene segments. The D genes were masked with “N” characters with equivalent length and not counted, accounting for uncertainty in precise D gene boundaries and somatic rearrangement. SHM was quantified as the number of nucleotide mismatches between the observed contig sequence and the reconstructed germline, excluding positions in the masked D region. Clonal relationships among B cells were determined by heavy chain only. Private clones represent clonally related B cells within individual animals. Cells were assigned to the same private clone if they originated from the same animal and shared an identical V gene, J gene, and full nucleotide CDR3 sequence. Public clones were defined to capture convergent antibody responses across animals. These are clonal groups that share the identical V gene, J gene, and full amino acid CDR3 sequence, but occur in two or more distinct animals.

Statistics

Statistics were computed in GraphPad Prism v.9 as denoted in the figure captions. For flow cytometry analyses, comparisons between two groups were made using a two-sided Student’s t-test or Mann Whitney U test if the sample size was too small to pass a normality test. For experiments with comparison with 3 or more groups, we used one-way analysis of variance (ANOVA) with host-hoc Tukey test for multiple comparisons. For data shown on log axes, geometric means and geometric S.D. values are shown. No outliers were excluded. One mouse from Fig. 4 was excluded due to accidental pregnancy. Exact P values are denoted in the figures. For all figures, NS is not significant (P > 0.05).

Data and materials availability:

All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. Associated raw data files, including raw cryoelectron microscopy data and class averages, are deposited in Dryad (98). BCR sequences identified in Fig. 6 are deposited in GenBank (Accession numbers PX287649-PX289030). Data from the 10x scRNA-seq experiment (fig. S18) are available in the Gene Expression Omnibus under accession number GSE305729. DNase I KO mice were shared from the Reizis lab (NYU, with material transfer agreement from CMMR), and V_H1-2 mice were obtained from the Alt lab (with a material transfer agreement from Boston Children’s Hospital).