Tailored design of protein nanoparticle scaffolds for multivalent presentation of viral glycoprotein antigens

Multivalent presentation of viral glycoproteins can substantially increase the elicitation of antigen-specific antibodies. To enable a new generation of anti-viral vaccines, we designed self-assembling protein nanoparticles with geometries tailored to present the ectodomains of influenza, HIV, and RSV viral glycoprotein trimers. We first de novo designed trimers tailored for antigen fusion, featuring N-terminal helices positioned to match the C termini of the viral glycoproteins. Trimers that experimentally adopted their designed configurations were incorporated as components of tetrahedral, octahedral, and icosahedral nanoparticles, which were characterized by cryo-electron microscopy and assessed for their ability to present viral glycoproteins. Electron microscopy and antibody binding experiments demonstrated that the designed nanoparticles presented antigenically intact prefusion HIV-1 Env, influenza hemagglutinin, and RSV F trimers in the predicted geometries. This work demonstrates that antigen-displaying protein nanoparticles can be designed from scratch, and provides a systematic way to investigate the influence of antigen presentation geometry on the immune response to vaccination.


Introduction
Multivalent antigen display, in which antigen is presented to the immune system in a repetitive array, has been demonstrated to increase the potency of humoral immune responses 1,2 . This has been attributed to increased cross-linking of antigen-specific B cell receptors at the cell surface and modulation of the immunogen trafficking to and within lymph nodes 3,4 . An ongoing challenge has been to develop multimerization platforms capable of presenting complex oligomeric or engineered antigens [5][6][7] , as these can be difficult to stably incorporate into non-protein-based nanomaterials (e.g. liposomes, polymers, transition metals and their oxides). Attributes such as epitope accessibility, proper folding of the antigen, and overall stability are additional considerations that must be taken into account in any design strategy for antigen presentation. Several reports have utilized non-viral, naturally occurring protein scaffolds, such as self-assembling ferritin [8][9][10] , lumazine synthase 5,11 , or encapsulin 12 nanoparticles, to present a variety of complex oligomeric or engineered antigens. These studies collectively highlight several of the key advantages of using self-assembling proteins as scaffolds for multivalent antigen presentation 13,14 , including the formation of stable, highly monodisperse immunogens, the existence of scalable manufacturing methods, and seamless integration of antigen and scaffold via genetic fusion. More recently, computationally designed one- 15 and two- 16,17 component protein nanoparticles have also been used to present complex oligomeric antigens 18,19 . The high stability of designed proteins 20,21 , versatility and ease of production and purification, increased potency observed upon immunization 18,19 , and the ability to predictively explore new structural space make these materials attractive as scaffolds for multivalent antigen presentation.
Despite this progress, control over the structure of nanoparticle immunogens has remained incomplete, limiting the potential of structure-based vaccine design. Even in the reported cases where computationally designed nanoparticles were used as scaffolds, the nanoparticle subunits were derived from naturally occurring oligomeric proteins.
Nanoparticle scaffolds specifically designed from the bottom up for multivalent display of antigens of interest could be still more effective. For example, for scaffolding homooligomeric class I viral fusion proteins, a large group that includes many important vaccine antigens 22 , designed nanoparticles with a close geometric match between the C termini of the antigen and the N termini of the scaffold would provide considerable advantages.
First, the antigen could be displayed without structural distortion near its base, potentially allowing for better retention of epitopes relevant to protection. Second, they could enable the multivalent display of antigens for which no compatible nanoparticle scaffolds are currently available. More broadly, by allowing atomically precise control over the geometry of antigen presentation, bottom up design of novel nanoparticle scaffolds would enable systematic investigation of the structural determinants of immunogenicity.

Method overview
We sought to develop a general computational method for generating de novo protein nanoparticles with geometries tailored to display specific antigens of interest, focusing in particular on the challenge of displaying the pre-fusion conformations of the trimeric viral glycoproteins HIV-1 Env (BG505 SOSIP) 23,24 , influenza hemagglutinin (H1 HA) 25 , and respiratory syncytial virus (RSV) F (DS-Cav1) 7 . To make the tailored nanoparticle design problem computationally tractable, we used a hierarchical approach.
First, monomeric protein building blocks were docked into trimeric homo-oligomers with cyclic symmetry (Figure 1a) and computationally screened for configurations featuring N termini with spacings similar to the C termini of the corresponding antigen ( Figure 1b). Rosetta combinatorial sequence design was then used to generate energetically favorable protein-protein interfaces between the monomeric subunits. Upon experimental characterization, trimers found to self-assemble into the desired structure were then paired with second oligomeric components as building blocks for two-component nanoparticle docking in tetrahedral, octahedral, or icosahedral symmetry as previously described (Figure 1c) 16,17 . Rosetta sequence design was employed to optimize interactions between the oligomers, generating secondary designed interfaces to drive nanoparticle assembly. This step-wise computational design protocol yielded twocomponent nanoparticles tailored to present 4, 8, or 20 trimers of viral glycoprotein antigens ( Figure 1d). These steps are described in detail in the following sections.

In silico symmetric docking and trimeric antigen targeting
A topologically diverse set of stable designed repeat proteins was used as scaffolds for symmetric trimer design, restricting trimeric configurations to those compatible with genetic fusion to at least one of the BG505 SOSIP, H1 HA, or DS-Cav1 trimers (see Methods). These viral fusion proteins were selected because of their importance as vaccine antigens, the availability of high-resolution structural data, and the variety they present in C-terminal geometry (31 nm, 15 nm, and 8 nm C-terminal separation distance for each glycoprotein, respectively). C3-symmetric docks of individual repeat proteins were rapidly assessed by the previously described RPX scoring method, which identifies arrangements likely to have good side chain packing 26 . Top-scoring docked configurations with an RPX score above 5.0 were computationally screened for termini geometries compatible with fusion to any of the three selected antigens using the previously described sic_axle protocol 18 . Geometrically compatible docks (those with non-clashing termini separation distances of 15 Å or less) were subjected to full Rosetta C3-symmetric interface design and filtering (see Methods) 26 , and 23 designs were selected for experimental characterization (design strategy presented in Supplementary   Figure 1; sequences for all trimer designs are in Supplementary Table 7).

Structural characterization of designed trimers
Synthetic genes encoding each of the designed trimers were expressed in E. coli and the resulting proteins purified from lysates by Ni 2+ immobilized metal affinity chromatography (Ni 2+ IMAC) followed by size-exclusion chromatography (SEC). 21 designs were found to express in the soluble fraction, and 11 formed the intended  Table  4.

Generation of two-component nanoparticles from designed trimers
Each SAXS-validated trimer (Figure 2a-d) was docked pairwise with a set of previously designed symmetric homo-oligomers 26 to generate tetrahedral, octahedral, and icosahedral arrangements using the TCdock program 16,17 . Icosahedral designs are particularly interesting as vaccine scaffolds as they have the highest valency among the cubic point group symmetries (20 trimers per particle compared to 8 for octahedra and 4 or 8 for tetrahedra), and higher valency has been shown to elicit higher antibody (Ab) titers in immunization studies 18,19,27 . To increase the probability of generating icosahedra, three naturally occurring homopentamers were also included in the docking calculations (PDB IDs 2JFB, 2OBX, and 2B98; sequences in Supplementary Table 7). To identify configurations likely to have a high number of energetically favorable side chain contacts following Rosetta interface design, nanoparticle docks were scored and ranked using the RPX method 26 . High-scoring and non-redundant nanoparticle configurations in which the N-terminal antigen fusion sites faced outward were selected for Rosetta interface design.
The models with the best overall metrics for each unique docked configuration were then

Structural Characterization of Designed Nanoparticles
The five SAXS-validated nanoparticles were structurally characterized using negative stain electron microscopy (NS-EM) 29 Table 1).

Characterization of Viral Glycoprotein-Nanoparticle Fusions
To explore the capability of the designed nanoparticles to display viral  Table   9). The secreted fusion proteins were then purified using a combination of immuno-affinity chromatography and SEC. The second component for each nanoparticle was produced recombinantly in E. coli as described above for the designed trimers, and in vitro assembly reactions were prepared with equimolar mixtures of the two components and incubated overnight (see Supplementary Methods).
Assembled nanoparticles were purified by SEC and analyzed by NS-EM as described in the Methods section to assess particle assembly and homogeneity. ~1,000 particles were manually picked and used to perform 2D classification and 3D classification/refinement in Relion 31 . Reconstructed 3D maps with docked nanoparticle (cyan for fusion component and gray for assembly component) and BG505 SOSIP trimer (green) models are displayed in Figure 5 (left). BG505 SOSIP trimers are clearly discernible in 2D class-averages and reconstructed 3D maps. However, the trimers appear less well-resolved than the corresponding nanoparticle core in the three reconstructions, likely due to the short flexible linkers between the BG505 SOSIP trimer and the trimeric nanoparticle components (sequences in Supplementary Table 9

Geometries
Because previous work involving icosahedral nanoparticle scaffolds displaying HIV-1 Env trimers has shown that antigen crowding can modulate the accessibility of specific epitopes and thereby influence the humoral immune response 19 Figure 6a). Next, a panel of anti-Env mAbs targeting epitopes ranging from the apex to the base of the BG505 SOSIP trimer were immobilized on SPR sensor chips (Figure 6b).
BG505 SOSIP-T33_dn2A trimer or BG505 SOSIP-T33-dn2 nanoparticle was then flowed over the mAbs and the ratio of macromolecule bound was calculated from the binding signal as previously described 19 . For mAbs that target apical, V3-base, and CD4binding site epitopes (PGT145, PGT122, 2G12, and VRC01), the number of molecules of trimer or nanoparticle bound was relatively similar (ratio ~1). However, for mAbs that target more base-proximate epitopes in the gp120-gp41 interface (ACS202, VRC34, and PGT151), an inter-protomeric gp41 epitope (3BC315), and the main autologous neutralizing Ab epitope in the glycan hole centered on residues 241 and 289 (11B), the accessibility was reduced in the nanoparticle format. Binding to a mAb directed to the trimer base (12N) was obliterated on nanoparticle BG505 SOSIP-T33_dn2 (Figure 6b).
We also compared this binding to historic data for BG505 SOSIP-I53-50 using six different representative mAbs 19 . As for BG505 SOSIP-T33_dn2, mAbs to the apex, V3base, and CD4-binding site (PGT145, PGT122, and VRC01) gave molar ratios ~1 for BG505 SOSIP-I53-50. However, for mAbs that target the more base-proximate epitopes in the gp120-gp41 interface (VRC34 and PGT151), there was a nearly 3-fold higher epitope accessibility on T33_dn2 than I53-50 nanoparticles (Figure 6c). Further down the trimer, no accessibility difference was again observed for a mAb that targets the gp41 inter-protomeric epitope (3BC315), which was relatively inaccessible on both nanoparticles, likely due to steric hindrance by neighboring trimers. These findings demonstrate the ability to modulate epitope accessibility by varying the designed scaffold geometry; this capability conferred by the panel of designed nanoparticles could be extremely useful in structure-based approaches to eliciting epitope-targeted humoral immune responses.

Discussion
Strong BCR signaling is required for eliciting successful humoral immune increments, respectively. For each sample, the distance between the C terminal residue of the target immunogen and N terminal residue of the docked trimer was measured until a minimum, non-clashing distance was determined (Supplemental Figure 1). Solutions for docks that were less than or equal to 15 Å for one or more of the three immunogens were selected for full Rosetta symmetric interface design as described in previously published methods 26 . Individual design trajectories were filtered by the following criteria: difference between Rosetta energy of bound (oligomeric) and unbound (monomeric) states less than -30.0 Rosetta energy units, interface surface area greater than 700 Å 2 , Rosetta shape complementarity (sc) greater than 0.65, and less than 50 mutations made from the respective native scaffold. Designs that passed these criteria were manually inspected and refined by single point reversions, and one design per unique docked configuration was added to the set of trimers selected for experimental validation.

Computational Design of Nanoparticles Validated Trimers. Two-component
nanoparticle docks were scored and ranked using the RPX method 26 as opposed to prior methods involving only interface residue contact count 16 . High-scoring and nonredundant nanoparticle configurations were selected for Rosetta interface design with an added caveat that they included trimers with outward-facing N termini for antigen display.
The design protocol took a single-chain input .pdb of each component, and a symmetry definition file 42 containing information for a specified cubic point group symmetry. The oligomers were then aligned to the corresponding symmetry axes of the architecture using the Rosetta SymDofMover, taking into account the rigid body translations and rotations retrieved from the .dok file output from TCdock 16,17 . Each configuration was then subjected to a symmetric interface design protocol similar to that previously described 26 .
Individual design trajectories were filtered by the following criteria: difference between Rosetta energy of bound and unbound states less than -30.0 Rosetta energy units, interface surface area greater than 700 Å 2 , sc greater than 0.6, and less than 50 mutations made from each native scaffold. Designs that passed these criteria were manually inspected and refined by single point reversions for mutations that did not appear to contribute to stabilizing the bound state of the interface. The sequence with the best overall metrics for each unique docked configuration was then selected for experimental characterization. NS-EM of HA-I53_dn5. The HA-I53_dn5 complex was adsorbed onto glow-discharged carbon-coated copper mesh grids for 60 s, stained with 2% uranyl formate for 30 s, and allowed to air dry. Grids were imaged using the FEI Tecnai Spirit 120 kV electron microscope equipped with a Gatan Ultrascan 4000 CCD Camera. The pixel size at the specimen level was 1.60 Å. Data collection was performed using Leginon 45 with the majority of the data processing carried out in Appion 46 . The parameters of the contrast transfer function (CTF) were estimated using CTFFIND4 47 . All particles were picked in a reference-free manner using DoG Picker 48 . The HA-I53_dn5 particle stack from the micrographs collected was pre-processed in Relion. Reference-free two-dimensional (2D) classification with cryoSPARC was used to select a subset of particles, which were used to generate an initial model using the Ab-Initio reconstruction function in

NS-EM of
CryoSPARC. The particles from the best class were used for non-uniform refinement in CryoSPARC to obtain the final 3D reconstruction. was set with the resulting pixel size of 1.03 Å at the specimen plane. Total dose was set to ~50 e -/Å 2 with 250 ms frames. Nominal defocus range was set to -0.6 to -1.6 µm for all 3 nanoparticle samples. Automated data collection was performed using Leginon software 45 . Data collection information for acquired datasets is shown in Supplementary

Cryo-EM of Designed
where n is the number of moles of macromolecules, R the response at 300 s (RU), m the mass bound per area and RU (g/(mm 2 RU)), A the interactive area of the chip (mm 2 ), and M the molar mass of the macromolecule (g/mol). This analysis corrects for the greater 23 mass (and thereby greater signal) for each bound nanoparticle such that the number of binding events by differing macromolecules can be directly compared.      . Tuning epitope accessibility through antigen display in monovalent, tetrahedral, and icosahedral formats. a, NS-EM micrographs of BG505 SOSIP-T33_dn2A with and without VRC01 Fab bound, 2D class averages, and models fit into 3D maps. b, Representative sensorgrams of indicated antigen binding to anti-Env mAbs. c, Relative accessibility of epitopes on BG505 SOSIP-T33_dn2 nanoparticles and BG505 SOSIP-I53-50 nanoparticles (reproduced from prior publication for comparison 19

Structure Deposition Information
Crystal structures have been deposited in the RCSB protein data bank with the PDB IDs