Animals

The animal work was approved by the Emory University Institutional Animal Care and Use Committee (IACUC) under protocol 202100136.

Twelve adult Indian-origin rhesus macaques (Macaca mulatta) (RM) were housed at the Emory National Primate Research Center (ENPRC) and maintained in accordance with NIH guidelines. Animal care facilities are accredited by the US Department of Agriculture (USDA) and the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) International. Animals were treated with anaesthesia (ketamine, 5–10 mg per kg; or telazol, 3–6 mg per kg) and analgesics for procedures including intramuscular (i.m.) and subcutaneous (s.c.) immunization, and blood draws as per veterinarian recommendations and IACUC approved protocols. Rhesus macaques were male and female, aged 3–4 years at the start of the study, with an average weight of 4.8 kg. Animals were grouped to divide age and weight as evenly as possible between the groups receiving either soluble or liposome-conjugated trimers. Animals were housed in pairs for the duration of the study. For the immunization group size, we established the sample size primarily by extensive historical experience in immunization studies, which indicates that the assignment of six animals per immunization group will result in the detection of significant differences in antibody production among different experimental groups. Moreover, we have taken into consideration the three Rs (replacement, reduction and refinement) principle that guides the humane use of animals in scientific research; in particular, we considered reduction, which recommends using the fewest animals necessary to achieve valid experimental results.

Cell lines

HEK293F cells were cultured in FreeStyle 293 Expression Medium (Thermo Fisher Scientific) supplemented with 1× antibiotic–antimycotic solution (Sigma-Aldrich) in a humidified incubator (125 rpm, 8% CO₂, 37 °C). TZM-bl and HEK293T cells were cultured in Dulbecco’s modified Eagle medium (cDMEM) supplemented with 10% fetal BSA and 1× penicillin–streptomycin–glutamine (Gibco) at 5% CO₂ and 37 °C. Expi293F cells were cultured in medium from 293F Expression System Kit in a humidified incubator (125 rpm, 8% CO₂, 37 °C).

Design of NFL trimer immunogens and probes

The Env sequences used for the generation of the NFL trimeric proteins were derived from the following HIV-1 viral sequences obtained from the Los Alamos National Laboratory HIV sequence database: Q23.17 (AF0048885.1), ZM233M.6 (DQ388517), WITO4160.33 (AY835451), 001428_2 (EF117266), 92RW020 (AY669706), BG505.W6M (DQ208456) and 16055_2 (EF117268). These DNA sequences were codon optimized to enhance protein expression and modified as follows to make stabilized NFL trimeric Env proteins: the natural HIV-1 leader sequence was replaced by a CD5 leader sequence. The four-residue furin cleavage site 508-REKR-511 was substituted with a ten-amino-acid flexible linker comprising the sequence G₄SG₄S²⁸. The C terminus of the Env sequence was truncated at residue 664, resulting in the elimination of the membrane proximal external proximal region, transmembrane domain and cytoplasmic tail⁴¹. An additional linker comprising the residues GGGGSHHHHHHHHGSGC was added to the C terminus to facilitate coupling to liposomes through the terminal cysteine residue. Finally, a series of stabilizing mutations was introduced to create highly stable and homogeneous trimeric proteins with near-native conformation. These stabilizing mutations consist of the TD (BG505-trimer-derived residues), helix-breaking glycine and proline substitutions, unnatural disulfides and V3-loop and fusion-peptide-stabilizing mutations^29,30,32.

Trimer expression, purification, characterization and trimer-liposome preparation

The expression and purification protocols for HIV-1 Env NFL trimers were described in detail previously^28,30,42,43. NFL trimers used as immunogens were expressed transiently in 293F cells and purified over a lectin column followed by negative selection using the non-neutralizing mAb F105 followed by size-exclusion chromatography (SEC). The trimers were characterized for their structural conformation and antigenicity by SEC, BLI and DSC. In brief, the BLI analysis was carried out on an Octet Red instrument (Sartorius) with IgGs immobilized on anti-human IgG Fc capture sensors (Sartorius). The Env trimers were assessed as free analytes in solution (PBS pH 7.4) at a final concentration of 250 nM. Association and dissociation were measured for 60 s respectively. The data were analysed using ForteBio software v.11.1 and kinetic parameters were obtained using a global 1:1 fit model. The thermal transition temperature (T_m) of the NFL trimers was determined by DSC using a MicroCal VP-Capillary DSC instrument (Malvern Panalytics). The trimer samples were dialysed in PBS, pH 7.4, and 400 μl of the trimer sample at concentration of 0.25 mg ml⁻¹ was loaded into the instrument. The dialysis buffer was used as the reference solution. The DSC experiments were performed at a scanning rate of 1 K min⁻¹ under 3.0 atmospheres of pressure. The data were analysed after buffer correction, normalization and baseline subtraction using MicroCal VP-Capillary DSC analysis software provided by the manufacturer (Origin 7 SR4 v.7.0522). Liposome-conjugated trimers were used as immunogens in NHPs Q7–12. The conjugation protocol for generating trimer-liposomes was described in detail previously³¹. Liposomes comprised 1,2-distearoyl-sn-glycero-3-phosphocholine, cholesterol and 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine-N-[4-(p-maleimidomethyl) cyclohexane-carboxamide]³¹. After chemical conjugation with the NFL trimers, the trimer-liposomes were further purified over the Superdex 200 Increase 10/300 GL column in PBS pH 7.4 buffer to remove unconjugated trimers from the trimer-liposomes. The conjugation of trimer to liposomes was confirmed by imaging the trimer-liposomes using negative-stain electron microscopy (Scripps Research, EM core).

Calcium Flux

Spleen cells from the CH01 UCA dKI mice⁴⁴ and the K46 B cell line expressing PG16⁴⁵ were used for the calcium flux assay. CH01 UCA dKI cells were suspended at 4 million cells per ml in advanced DMEM, labelled with 1.5 μM Calbryte 520 AM (ATT Bioquest, 20651) and Pluronic F-127 (Invitrogen, P3000MP) for 30 min at 37 °C. CH01 UCA cells were washed with Advanced DMEM and stained with TruStain FcX PLUS (BioLegend, 156604) and Alexa Fluor 647 anti-mouse B220 (BioLegend, 103226) for 10 min at room temperature. After the staining protocol, the cells were washed with 2 mM CaCl₂ HBSS. The cells were then incubated at room temperature for 30 min. Two million cells were aliquoted for flow cytometry analysis. Cells were stimulated with NFL trimers or trimer-liposomes and calcium signals were detected for 240 s measuring fluorescence at excitation/emission = 516 nm/533 nm (B2 peak channel) on the Cytek Aurora spectral flow cytometer (Cytek Biosciences). Analysis was performed using FlowJo (Becton Dickinson). For the PG16 K46 B cell line assay, surface expression of the BCR was induced by adding 1 μg ml⁻¹ of doxycycline (Thermo Fisher Scientific, J6380506) 1 day before. The surface PG16 expression was confirmed by binding of FITC anti-human light chain lambda antibody (BioLegend, 316606) using the Cytek Aurora spectral flow cytometer (Cytek Biosciences). Further confirmation of PG16 surface expression was done using biotinylated Q23 NFL trimer conjugated to streptavidin-Alexa Fluor 647 (Molecular Probes, S32357). Trimer- or trimer-liposome-induced calcium flux was measured in a similar way as done for the CH01 UCA spleen cells.

Immunization of NHPs

Twelve NHPs, divided into two groups of six, were inoculated at eight sites (four i.m. and s.c. bilateral immunizations on the deltoids and inner thighs) with 80 μg of Q23 NFL soluble trimer (NHPs Q1–Q6) or 80 μg of Q23 NFL trimer conjugated to trimer-liposomes (NHPs Q7–Q12) in 150 μg SMNP adjuvant. SMNP adjuvant was prepared as previously described³⁴. NHPs were immunized with the same formulation (Q23 NFL) at weeks 2, 4, 6 and 8 as part of the divided-dose regimen. The NHPs were boosted with 150 μg of ZM233 NFL trimer in 375 μg SMNP adjuvant at week 20. Subsequent immunizations with heterologous trimers were performed at 31, 42, 64 and 102 weeks with 100 μg of trimer in 375 μg SMNP adjuvant. Pre-bleeds were collected before immunizations and test bleeds were collected on the day of immunizations and 14 days after each immunization. Blood was collected in Na-citrate CPT tubes (BD Biosciences) for PBMC and plasma isolation. Cells were cryopreserved in FBS with 10% DMSO (Gemini Bio, Fisher Bioreagents). Serum was collected with serum clot tubes (BD Biosciences).

Neutralization

Serum IgG was affinity-purified with protein A Sepharose to approximately physiological levels (10 mg ml⁻¹) as described previously³². Replication-incompetent HIV-1 Env pseudoviruses were produced by co-transfecting HEK293T cells with 15 µg of Env-deficient backbone plasmid (pSG3Δenv) and 5 µg HIV-1 Env plasmid in a ratio of 1:3 (total DNA:Fugene6 transfection reagent). Pseudoviruses were collected 72 h after transfection and incubation at 37 °C by centrifugation of cell culture supernatants at 3,000g for 10 min and stored at −80 °C. Inhibition of entry of HIV-1 Env pseudotyped viruses into standard TZM-bl cells was used to determine the neutralization capacity of sera, purified total serum IgG or mAbs as previously described in a half-well plate format⁴⁶. TZM-bl cell suspension (50 µl; 110,000 cells per ml) in cDMEM was added to each well of white, flat-bottomed tissue culture treated plates 24 h before setting up the sample dilutions and incubated in a 37 °C CO₂ incubator. In a 96-well U-bottomed plate, total serum IgG or antibodies were serially diluted in cDMEM five or six times (with 1:3 or 1:5 dilution factors) resulting in a total of six to seven dilution points. In a separate 96-well U-bottomed plate, the sample dilutions were mixed with prewarmed pseudovirus at a ratio of 1 to 5 (6 µl of sample dilution and 24 µl of pseudovirus), resulting in a maximal final total serum IgG concentration of 2 mg ml⁻¹ or antibody concentration of 50 µg ml⁻¹. The mixtures were then incubated at 37 °C for 1 h. cDMEM was then aspirated from the tissue culture plates containing TZM-bl cells, and 25 µl of the sample-pseudovirus mixture was pipetted directly onto the TZM-bl cells and placed into the 37 °C incubator for 24 h. After 24 h, 75 µl cDMEM was added to all wells of the plates and left at 37 °C for another 24 h. Finally, the cDMEM/sample mixes were aspirated from the plates, and the TZM-bl cells were lysed for 20 min on an orbital shaker at 400–500 rpm. Inhibition of entry was determined using the Promega Luciferase system, with luminescence detected using the Biotek NEO2M plate reader. The resulting Luciferase signals were measured in relative light units (RLU). The serum IgG or antibody concentrations that resulted in a 50% RLU reduction (IC₅₀ values for purified IgG or mAbs) was determined by fitting the neutralization dose–response curves by nonlinear regression using a five-parameter Hill slope equation. Data were analysed using GraphPad Prism (v.10.2.1).

Individualized immunoglobulin genotyping

To determine the germline V, D and J allele content of the studied rhesus macaques, full-length HC VDJ and LC VJ amplicons were generated as described previously³⁶ and analysed using IgDiscover (https://gkhlab.gitlab.io/igdiscover22/). In brief, total RNA was extracted from approximately 5–10 million PBMCs using the Qiagen RNeasy mini kit. Then, 300 ng RNA was subjected to reverse transcription with gene-specific primers for the IgM, IgK and IgL constant regions using the Sensiscript RT kit (Qiagen), the cDNA was purified using the Qiagen MinElute PCR Purification Kit and was eluted in 20 µl of elution buffer. Then, 3 µl cDNA was subjected to the library PCR with 25× cycles using multiplex forward VH, VK and VL primer sets, and gene-specific primers for the IgM, IgK and IgL constant regions as previously described³⁶. The libraries were gel-purified using the Qiagen MinElute Gel Extraction Kit and indexed with 10× PCR cycles according to the instructions of the Illumina MiSeq 2 × 300 bp kit. Indexed library PCR products were purified using the MinElute PCR Purification Kit, followed by AMPure XP magnetic bead (Beckman Coulter) purification before sequencing using the Illumina MiSeq 2 × 300 bp kit. The libraries were analysed using IgDiscover³⁷ and Corecount³⁵ to generate HC V, D and J, and LC V and J genotypes. Input databases for the IgDiscover analysis were KIMBD (http://kimdb.gkhlab.se/) for the HC alleles and IMGT for the LC alleles.

Env-specific memory B cell sorting by flow cytometry

To produce Env trimer probes for B cell sorting, 10 µg of biotinylated Q23, BG505 and 16055 NFL trimers were conjugated to streptavidin–APC (Invitrogen) or streptavidin–BV421 (BioLegend) in five sequential steps, each incubation proceeded for 20 min at 4 °C. Frozen single-cell suspensions from blood mononuclear cells (PBMCs) from animals Q7, Q9, Q10 and Q12 were thawed at 37 °C, washed twice in prewarmed RPMI 1640 medium (HyClone) supplemented with 10% FBS (HyClone) and penicillin–streptomycin (100 IU ml⁻¹, 100 µg ml⁻¹) (Gibco). The cells were washed with PBS (Sigma-Aldrich) and counted using Trypan Blue exclusion of dead cells by a Countess II cell counter (Thermo Fisher Scientific). Cells were suspended in PBS and incubated for 30 min at 4 °C with Live/Dead Fixable Aqua Dead Cell Stain Kit (Life Technologies) according to the manufacturer’s instructions. Cells were washed with FACS buffer (PBS + 1% FBS) and surface stained with the following antibodies: CD3 FITC (SP34-2), CD14 FITC (M5E2), CD20 BV421 or PerCP-Cy5.5 (2H7), CD27 PE-Cy7 (M-T271), IgG PE-CF594 (G18-145) (all from BD Biosciences). Staining was performed for 30 min at 4 °C. After washing with FACS buffer, cells were subsequently stained with the fluorescently conjugated NFL trimers. Live CD3-CD14⁻CD20⁺CD27⁺IgG⁺ENV⁺ single cells were sorted on a four-laser FACSAria Fusion cell sorter (Becton Dickinson) into 96-well PCR plates (Eppendorf) containing 4 µl per well of ice-cold cell lysis buffer (0.5× PBS, 10 mM DTT and 2 U µl⁻¹ RNAsin (all from Thermo Fisher Scientific)). After sorting, the 96-well plates were centrifuged, sealed and immediately frozen on dry ice and stored at −80 °C until use.

Single B cell RT–PCR and mAb cloning

For cDNA synthesis the 96-well plates, containing single B cells, were thawed on ice. The reverse transcription (RT) was performed using SuperScript IV reverse transcriptase (Thermo Fisher Scientific) using random hexamers, oligodT, dNTPs (Invitrogen), Igepal CA-630 (Sigma-Aldrich) and RNAsin (Thermo Fisher Scientific). IgG HC and LC V(D)J sequences were amplified separately in 20 μl nested PCR reactions using 4 μl of cDNA for HC and 3 μl for LC in the first-round PCR and 1 μl PCR product in the second-round PCR using KAPA HiFi HotStart ReadyMix 2× (Roche). PCR products from positive wells were purified, Sanger sequenced (Genewiz) and analysed. HC and LC V(D)J sequences were cloned into expression vectors containing the human IgG1, Igκ1 or Igλ2 constant regions⁴⁷. The sequences, engineered with overhangs complementary to the linearized vector ends, were assembled using the Gibson Assembly Master Mix (New England Biolabs). The reaction mixture, consisting of 50 ng of vector and 30 ng of insert in 20 μl reaction mix, was incubated at 50 °C for 1 h. After incubation, 1 µl of the diluted (1:3) reaction mix was transformed into XL10-Gold ultracompetent cells (Agilent Technologies) by heat shock at 42 °C for 30 s. Screening of transformed colonies was assessed by PCR, positive clones were expanded and plasmids were isolated using the Plasmid Plus Midi Kit (Qiagen). The correct sequences were confirmed using Sanger sequencing (Genewiz).

mAb expression and purification

mAbs were expressed by co-transfecting equal amounts of HC and LC plasmids (18 µg each) into 30 ml FreeStyle 293-F cells (Thermo Fisher Scientific) (HEK293-F) cultured in FreeStyle 293 Expression Medium (Thermo Fisher Scientific) supplemented with 1× antibiotic–antimycotic solution (Sigma–Aldrich) at a density of 1.2 million cells per ml at >95% viability. Cultures were maintained in a humidified shaking incubator (125 rpm, 8% CO₂, 37 °C). Transfections were carried out with FreeStyle Max reagent (Invitrogen) in Opti-MEM medium (Gibco). Then, 7 days after transfection, the mAbs were purified from the supernatant using Protein G Sepharose columns (Cytiva). Next, 4 μg of each purified antibody was analysed using SDS–PAGE under reducing conditions using NuPAGE 4–12% Bis-Tris polyacrylamide gels (Invitrogen).

ELISA

Ninety-six-well plates (Half area Corning) were coated with lectin (Galanthus Nivalis Lectin, Vector laboratories) at 2 μg ml⁻¹ for 1 h. The plates were blocked for 1 h with 150 μl PBS buffer comprising 2% non-fat milk and 5% FBS. The plates were washed four times with 150 μl of PBS buffer supplemented with 0.02% Tween-20 between each of the subsequent incubation steps. Soluble trimer or trimer-liposomes at 2 μg ml⁻¹ were captured on the plates for 1 h. The plates were incubated for 1 h with fivefold serial dilutions of selected mAbs with a starting concentration of 50 μg ml⁻¹. Antibody binding was detected using horseradish-peroxidase-conjugated anti-human IgG secondary antibody (Jackson ImmunoResearch) at a dilution of 1:5,000 for 1 h, followed by another wash and developed with TMB (Life Technologies) substrate solution. The plates were visually examined for development of colour, and the reaction was stopped with 0.3 N sulfuric acid, and the absorbance was measured at 450 nm. Data were plotted using GraphPad Prism v.9.4.1.

EMPEM analysis

Total serum IgG was digested to polyclonal Fab using papain according to the manufacturer’s instructions (Pierce Fab Preparation kit, Thermo Scientific). For each immune complex, 1 mg of polyclonal Fab was incubated with 15 µg of Env NFL trimer overnight and purified the next morning using the Cytiva Superdex 200 Increase size-exclusion column. Fractions corresponding to immune complexes were pooled and adsorbed onto 400-mesh carbon-coated copper grids (Electron Microscopy Sciences) at a concentration of around 0.02 mg ml⁻¹ for 10 s before blotting off the excess liquid. Grids were stained for 45 s using 2% (w/v) uranyl formate (Electron Microscopy Sciences) before blotting. Imaging was performed using a Thermo Fisher Scientific Glacios transmission electron microscope operating at 200 kV, equipped with the Thermo Fisher Scientific Falcon 4 camera (×73,000 magnification, 1.89 Å pixel size). Automated data collection was performed using EPU (Thermo Fisher Scientific) and data processing was performed using Relion (v.4.0)⁴⁸, according to standard procedures for reference-free particle picking and 2D classification. After three rounds of 2D classification, particles from classes corresponding to immune complexes were subjected to 3D refinement with C₃ symmetry and a 40 Å low-pass-filtered map of HIV Env ectodomain as the initial model. The initial model is based on PDB coordinates 6V0R, converted to a map using the molmap feature in UCSF ChimeraX⁴⁹. After 3D refinement, C₃ symmetry expansion was applied to the particles and 7 separate focused 3D classification skip align jobs were run (K = 10), each with a 40 Å diameter spherical mask over key HIV Env epitopes. The names of the epitopes and reference structures used for orienting the masks are: (1) gp41-base (PDB: 7L8Z), (2) gp41-GH (PDB: 7L8U), (3) gp41-FP (PDB: 7L8T), (4) gp120-GH (PDB: 7L8B), (5) C3V5 (PDB: 7L8Y), (6) CD4bs and gp120 interface (PDB: 7L8X), (7) V1V3 (PDB: 7L8E) and (8) V2-apex (PDB: 9D1W). For each epitope 3D classification, classes with visible Fab density were selected and subjected to 3D refinement, 2D classification and a second round of 3D classification. If the 3D refinement resulted in partial Fab density relative to the Env trimer, classes were selected from the subsequent round of 3D classification, and this was repeated until the reconstruction improved, or no change was noted. Final reconstructions were visually inspected and assigned to the correct epitope label. The number of final particles belonging to each epitope was divided by the total number of particles in the initial (C₃) 3D refinement. This value, which can range from 0 to 3 due to C₃ symmetry expanded particles used in the 3D classification steps, is described as the EMPEM magnitude. It is also assumed that most epitopes are represented three times on an Env trimer. An exception is made for the V2-apex epitope, and the EMPEM magnitude is normalized using a 3× multiplier, as V2-apex bNAbs are known to bind only at a ratio of 1 Fab per trimer. The total magnitude is the sum of each individual epitope EMPEM magnitude for a given animal and timepoint. Representative maps have been deposited to the Electron Microscopy Data Bank.

Purification and crystallization of unliganded Fabs

Fabs were expressed in Expi293F cells (Thermo Fisher Scientific, A14527) transiently transfected with plasmid DNA encoding the Fab heavy and light chains at a 1:1 ratio. Transfections were performed using the ExpiFectamine 293 Transfection Kit in Opti-MEM medium. Then, 6 days after transfection, the culture supernatants were collected and sterile-filtered through a 0.22-µm membrane filter. Fabs were purified using the CaptureSelect CH1-XL affinity matrix (Thermo Fisher Scientific, 1943462050), followed by SEC on the Superdex 200 16/600 column equilibrated with 20 mM Tris, 150 mM NaCl, pH 7.5. The fractions corresponding to Fab proteins were pooled and concentrated to 12 mg ml⁻¹ for crystallization. Crystallization screening was performed using our Rigaku CrystalMation robotic system with JCSG Core Suites 1–4, and Top96 Cryo screens at 20 °C. Crystals were cryoprotected, if necessary, by supplementing the reservoir solution with 15% ethylene glycol and then flash-cooled in liquid nitrogen for storage prior to data collection.

X-ray data collection and structure determination

X-ray diffraction data were collected at beamline 17-ID-1 (AMX) of the National Synchrotron Light Source II at 100 K using a wavelength of 0.9197 Å. Crystals of Fabs Q9M-023, Q10M-055, Q12QBM-007 and Q12BBM-069 diffracted to resolutions of 2.34 Å, 1.63 Å, 1.54 Å and 1.79 Å, respectively. The crystallization conditions were as follows: Q9M-023 Fab crystallized in 40% MPD, 0.1 M cacodylate buffer at pH 6.5 and 5% (w/v) PEG-8000; Q10M-055 Fab in 100 mM HEPES at pH 7.5, 10% (v/v) ethylene glycol and 20% (w/v) PEG-8000; Q12QBM-007 Fab in 50% PEG-200 and 0.1 M citrate at pH 5.5; and Q12BBM-069 Fab in 1.26 M ammonium sulfate and 0.1 M cacodylate at pH 6.5. Data processing, including indexing, integration and scaling, was performed using autoPROC⁵⁰. Structure determination was carried out by molecular replacement using Phaser within the Phenix software suite⁵¹, with an initial model generated by AlphaFold 3⁵². Subsequent model building and refinement were performed through iterative cycles using Coot and Phenix.refine, respectively^53,54. The quality of the final structures was assessed using MolProbity⁵⁵, and additional validation was carried out through the PDB validation server. Ramachandran statistics showed that 97.7%, 97.4%, 98.0% and 97.5% of residues for Q9M-023, Q10M-055, Q12BQM-007 and Q12BBM-069, respectively, were in favoured regions, with the remaining 2.3%, 2.6%, 2.0% and 2.5% in allowed regions. Data collection and refinement statistics are summarized in Supplementary Table 2.

Cryo-electron microscopy

All Fabs used for structural analysis were prepared by cleaving the IgG using papain (Thermo Fisher Scientific, 44985). The following four immune complexes were incubated overnight: (1) 0.2 mg of BG505 NFL TD CC3+ with 0.25 mg of either Q12BBM-069, Q12QBM-007 or Q9M-023; (2) 0.2 mg of Q23 NFL TD CC3+ with 0.25 mg Q10M-055. Complexes were purified the next morning using the HiLoad 16/600 Superdex 200 pg (Cytiva) gel-filtration column. The fractions corresponding to each immune complex were pooled and concentrated using an Amicon 100 kDa MWCO centrifugal device to between 4 and 6 mg ml⁻¹. A complex of Q7M-675 Fab with WITO NFL CC3+ was prepared without the gel-filtration column step to account for the lower affinity expected based on the weaker neutralization titre. The samples were vitrified using the Vitrobot Mark IV (Thermo Fisher Scientific). The temperature was set to 4 °C and the humidity was maintained at 100% during the freezing process. The blotting force was set to 1 and the wait time was set to 10 s. The blotting time was varied from 3 to 6 s. Detergent lauryl maltose neopentyl glycol (Anatrace) or octyl-β-glucoside (Anatrace) was added to the sample to a final concentration of 0.005 mM or 0.1% (w/v), respectively, shortly before freezing. UltrAuFoil 1.2/1.3 (Au, 300-mesh; Quantifoil Micro Tools) grids were used and glow discharged (PELCO easiGlow, Ted Pella) for 40 s before sample application. Then, 0.5 µl of detergent was mixed with 3.5 µl of the samples and 3 µl of the mixture was immediately loaded onto the grid. After blotting, the grids were plunge-frozen into liquid-nitrogen-cooled liquid ethane.

The samples were loaded into a Thermo Fisher Scientific Glacios 2 TEM operating at 200 kV equipped with a Thermo Fisher Scientific Falcon 4i direct electron detector. The exposure magnification was set to ×190,000 with a pixel size at the specimen plane of 0.718 Å. EPU software (Thermo Fisher Scientific) was used for automated data collection. Micrograph video frames were motion corrected, dose weighted and CTF correction was performed using cryoSPARC Live⁵⁶. cryoSPARC was used for the remainder of data processing. Particle picking was performed using blob picker initially followed by template picker. Particles were initially down-sampled by a factor of 4 and multiple rounds of 2D classification were performed followed by 3D ab initio reconstruction. The remaining particles were then re-extracted, with optional Fourier cropping to reduce the memory requirements of large box sizes, resulting in image pixel sizes of 1.034 Å (Q12BBM-069 + BG505; Q10M-055 + Q23), 1.005 Å (Q12QBM-007 + BG505) or 0.718 Å (Q9M-023 + BG505; Q7M-675 + WITO). 3D non-uniform refinement was performed. For all cryo-EM maps, the global resolution was estimated using half maps and a Fourier shell correlation cut-off of 0.143, and the local resolution was estimated using the cryoSPARC Local Resolution tool with 0.143 FSC cut-off. Relatively lower resolution (4.3 Å global FSC estimate), due to preferred orientation and low amounts of Fab-bound trimers, prevented accurate model building into the Q7M-675 + WITO complex. Initial models were generated using AlphaFold3 and docked into the cryo-EM maps using UCSF ChimeraX^49,52. Manual building was performed in Coot v.0.9.8 and real space refinement in Phenix^57,58. Final models were validated using MolProbity and EMRinger in the Phenix suite. Buried surface area calculations and interface analyses were performed using UCSF ChimeraX. Data collection, processing and model statistics are summarized in Extended Data Fig. 5 and Supplementary Table 3.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.