Vaccination induces HIV broadly neutralizing antibody precursors in humans

Study design

IAVI G001, with ClinicalTrials.gov registry number NCT03547245, was a phase 1, randomized, double-blind, placebo-controlled dose escalation study to evaluate the safety and immunogenicity of eOD-GT8 60mer vaccine adjuvanted with AS01_B in HIV-uninfected, healthy adult volunteers. Two doses of 20 μg or 100 μg eOD-GT8 60mer with AS01_B, or two doses of placebo were given by deltoid intramuscular injection eight weeks apart, with both immunizations given to the same arm. Placebo was the buffer used in the vaccine: DPBS containing 10% sucrose at pH 7.5. The consort diagram is shown in fig. S1.

The primary objectives of the study were to evaluate the vaccine for safety and tolerability and the capacity to induce Immunoglobulin G (IgG) B cell responses from rare precursors for VRC01-class bnAbs. The primary endpoints were the occurrence of adverse events, and the secondary endpoint was induction of eOD-GT8 60mer-specific, eOD-GT8 monomer-specific and CD4-binding-site (CD4bs)-specific serum binding antibody responses. The detection of VRC01-class responses in IgG memory B cells, IgG GC B cells, and plasmablasts were exploratory endpoints but nevertheless were the critical immunological readouts to judge vaccine efficacy. Additional exploratory immunological analyses included: (i) assessing the relative frequencies of the VRC01-class BCRs and competitors; (ii) measuring the changes in somatic hypermutation and eOD-GT8 binding affinities over time for both types of BCRs; (iii) evaluating a wide array of properties of the VRC01-class BCRs, to assess the potential for the BCRs to mature into bnAbs; and (iv) assessing the binding of VRC01-class and competitor BCRs to antigens with CD4bs epitopes closer to native HIV Env compared to the vaccine.

Power analysis and rationale for trial size

Group sizes for the study were selected to measure the primary hypothesis that the vaccine will induce VRC01-class IgG B cells with a response rate of at least 50% in at least one of the adjuvanted protein vaccine arms, as well as to satisfy the need to have enough endpoints for further characterization of that response. We powered the study to have high probability of observing at least 5 participants with a vaccine-induced VRC01-class IgG B cell response among participants receiving eOD-GT8 60mer in study Group 1 or 2 given that the true response rate for this class of B cells was at least 50% or greater among eOD-GT8 60mer recipients in either arm. We assumed a dropout rate of approximately 10%, which translated into an assumption that immunogenicity samples would be obtained from N=16 recipients of eOD-GT8 60mer in each of Groups 1 and 2. Under those assumptions, we determined that at least 5 positive responders for VRC01-class B cells would be required for the 95% confidence interval (CI) about the observed rate to be consistent with a true rate of 50%, because the 95% CI for an observed rate of 5/16=31.25% is 14.2% to 55.6%. Furthermore, power was 96.2% to detect 5 or more positive responders out of 16 when the true rate of response was 50% (fig. S40).

Vaccine and adjuvant

eOD-GT8 60mer was manufactured in accordance with current Good Manufacturing Practice (cGMP) regulations at Paragon BioServices, Inc (Baltimore, MD), as described in detail elsewhere (67). In summary, cGMP manufacture was accomplished by a combination of purification techniques following transient expression in suspension-adapted, cGMP-qualified VRC293 human embryonic kidney 293 (HEK293) cells generated at SAFC (Carlsbad, CA) from a Master Cell Bank generously provided by the National Institutes of Health (NIH) Vaccine Research Center (VRC) within the National Institute of Allergy and Infectious Diseases (NIAID). VRC293 cells were grown in serum-free Expi293 medium (Thermo Fischer), transfected with plasmid DNA encoding eOD-GT8 60mer (cGMP manufactured by Aldevron; Fargo, ND) and Polyethylenimide PEIpro^®-HQ transfection reagent (cGMP manufactured by PolyPlus-Transfection SA; Illkirch, France), and expression was carried out in the presence of 14 μM Kifunensine (cGMP manufactured by GlycoSyn; Graceville, New Zealand). Benzonase endonuclease enzyme (high-purity grade, 250 U/μL, from Millipore; Burlington, MA, USA) was employed to remove residual host cell and plasmid DNA. The eOD-GT8 60mer cGMP clinical material was formulated at 1 mg/mL in 10% sucrose in phosphate-buffered saline (PBS) at pH 7.2, aliquoted at 0.4 mL volume in 2 mL Type 1 glass vials with stoppers (13 mm stopper, Rubber with Flurotec, Afton Scientific), sealed with sterile seals from Afton Scientific Corporation (Charlottesville, VA), and stored at −80°C. The material is currently on a stability testing program as per regulatory guidelines (stable for >36 months).

Quality control procedures were performed on the manufactured eOD-GT8 60mer nanoparticle to confirm its identity, determine protein concentration and purity, establish in-vitro potency, measure the nanoparticle size, and characterize N-linked glycans. Additionally, the clinical trial lot was tested to quantify host cell residual impurities (host cell proteins and host cell DNA), measure bioburden and bacterial endotoxin, determine subvisible particulate matter and confirm sterility. Quality control procedures included sandwich ELISA with GL-VRC01 for potency; high pressure liquid chromatography (HPLC) with tandem mass spectrometry (MS/MS) to confirm amino acid sequence; N-terminal Edman sequencing for further sequence confirmation; size exclusion chromatography (SEC) to assess purity; sedimentation velocity analytical ultracentrifugation (AUC-SV) for particle size distribution; dynamic light scattering (DLS) for average particle size; hydrophobic interaction liquid chromatography coupled to mass spectrometry (HILIC-FLD-MS/MS) for N-linked glycan profiling. eOD-GT8 60mer cGMP material had full glycan occupancy at only five of ten glycosylation sites, and had partial occupancy at three sites (67). Pre-clinical material that performed well in mouse models had full glycan occupancy at three sites and partial occupancy at four sites (43).

AS01_B adjuvant is an adjuvant system composed of two immunoenhancers combined in a liposomal formulation consisting of dioleoyl phosphatidylcholine (DOPC) and cholesterol in phosphate-buffered saline solution. The immunoenhancers are: (1) 3-O-desacyl-4’-monophosphoryl lipid A (MPL), a derivative of lipopolysaccharide from the Gram-negative bacterium Salmonella minnesota, and (2) a saponin molecule (QS-21) purified from the bark of the tree Quillaja saponaria Molina. AS01_B was manufactured and provided by GlaxoSmithKline Biologicals (Rixensart, Belgium). QS-21 was licensed by GSK from Antigenics LLC, a wholly owned subsidiary of Agenus Inc. (Delaware, USA). The administered dose of AS01_B corresponded to 50 μg each of MPL and QS-21.

DPBS with 10% sucrose was manufactured by SAFC Biosciences.

Study procedures

Fine needle aspiration from draining axillary lymph nodes

Aspirates were collected from the draining axillary lymph nodes by fine needle aspiration (FNA) at weeks 3 and 9. Ultrasound was used to visualize the axillary lymph nodes on the same side as the site of the most recent vaccination for lymph node sampling. Following site-specific standard FNA procedures, the overlying skin was swabbed generously with chlorhexidine, betadine, or similar skin disinfectant solution, and 1% lidocaine was injected subcutaneously as local anesthetic. A 22-gauge lumbar puncture needle attached to a 5 mL sterile syringe was then inserted into the largest and most accessible lymph node, and negative pressure was applied by withdrawing the syringe lunger approximately 2–3 mL while collecting sample over a 30 second period of ‘to and fro’ needle movement, or until first appearance of blood in syringe. Immediately after withdrawal of each needle from the study participant, syringe and needle contents were expelled into a 50mL conical tube containing R10 media, flushing out any remaining cells out of the syringe by detaching the lumbar needle from the syringe, withdrawing approximately 2 to 3 mL of fresh R10 media into the syringe barrel, reattaching it to the needle, and expelling the media into the 50mL conical tube; this flushing procedure was repeated a total of 4 times for each aspiration. The collection was repeated up to 4 times for each study participant, using a new sterile needle and syringe for each pass. After collection, the sample was transported to the laboratory on wet ice or cold packs as soon as possible. Lymph node FNA samples ranged in recovery from very few cells up to 4×10⁷ cells (median recovery varied between sites from 2.4×10⁶ to 4.7×10⁶).

Preparation of lymphocytes from fine needle aspirates from axillary lymph nodes

The 50mL conical FNA sample was spun at 325 × g for 10 mins at 4°C (no brake). Supernatant was gently aspirated with a serological pipet down to about 500uL. If the cell pellet had any redness, cells were resuspended in 5mL of cold 1x red blood cell (RBC) lysis buffer (eBioscience) and incubated for 5 minutes at ambient temperature while agitating sample by gently pipetting up and down periodically with a P1000 pipette or using a tube rocker. Cells were washed with 40 mL cold 10% FBS/ 1xPBS. Samples were centrifuged at 325 × g for 10 minutes at 4°C. Supernatant was gently aspirated with serological pipette and the cell pellets were resuspended with 2 mL cold 10% FBS/ 1xPBS. Cell numbers and viability of the sample were determined using a hemocytometer, Muse cell analyzer, or Nexcelom Auto 2000.

Production of protein reagents for immunological assays

Serum binding analysis by BAMA

Serum IgG responses were measured via a multiplex antigen panel (i.e. eOD-GT8, eOD-GT8-KO11, eOD-GT8 60mer, Lumazine synthase) to determine specificity and magnitude. A binding antibody multiplex assay (BAMA) (68–70) was modified by lengthening the primary antibody incubation period from 30 to 120 minutes for enhanced detection sensitivity of early bnAb precursors. The assay was validated for accuracy, specificity, precision, robustness and limit of detection/quantitation (LLOD/LLOQ). The LLOQ for detection of early bnAb precursors was 0.0041–0.0866 μg/ml. Antibodies were measured at day 0 (day of first vaccination), day 14 (two weeks post-first vaccination), day 28 (four weeks post-first vaccination, day 56 (day of second vaccination), day 70 (two weeks post-second vaccination), and day 112 (eight weeks post-second vaccination). Samples were serially diluted (1:50, 1:250, 1:1250, 1:6250, 1:31250, and 1:156250) and incubated for 120 minutes with a mixture of carboxlyated fluorescent MagPlex microsphere sets (Luminex) that were each covalently coupled to one of the antigens. Antigen-specific IgG was detected using a biotinylated detection antibody to human IgG Fc (Southern Biotech), followed by washing and incubation with Streptavidin PE (BD Pharmingen). Samples were acquired on a Bio-Plex instrument (Bio-Rad), and antibody levels were measured as median fluorescent intensity (MFI) from two wells and then averaged using the mean. The readout was background-subtracted median fluorescent intensity (MFI), where background refers to the antigen-specific plate-level control (i.e., a blank well containing antigen-conjugated beads run on each plate plus detection antibody). Additionally, a blank or reference bead was included to estimate non-specific antibody binding. Area under the titration curve (AUTC) was used as the magnitude measure of interest. AUTC was calculated using the trapezoid method with truncation in the case of negative background-adjusted MFI minus background-adjusted blank MFI (MFI*) values or MFI* values > 22000. Samples with blank MFI > 5000 were excluded from the analysis. The positive controls were GL-VRC01 (germline VRC01 mAb), VRC01 mAb, anti-6X HIS epitope tag (for His-tagged proteins), CH31 mAb, 2G12 mAb.

The positivity of a response was defined based on background-adjusted MFI values at the screening dilution level (1:50) except Lumazine synthase. Lumazine synthase was observed to have high baseline values across participants at the screening dilution, so filtering for high baseline MFI* was not applied and the response call for this antigen was made at the first dilution for which baseline MFI* was < 6500. Antigen-specific positivity thresholds were computed as the maximum of 100 and the 95th percentile of the baseline MFI* by antigen except Lumazine synthase which was set to 100. Samples from post-enrollment visits were declared to have a positive binding antibody response by BAMA if they met three criteria:

• 1

  MFI* values were greater than the antigen specific positivity threshold.

• 2

  MFI* values were greater than 3 times the baseline MFI* values.

• 3

  Background-adjusted MFI values were greater than 3 times the baseline background-adjusted MFI values.

For differential binding to the CD4 binding site (CD4bs), defined as the difference in AUTC for binding to eOD-GT8 (AUTC_Ref) and eOD-GT8-KO11 (AUTC_KO), the positivity of response was defined as a positive response to eOD-GT8 plus an additional criterion:

• 4

  ΔAUTC > 0, where ΔAUTC was defined as AUTC_Ref − AUTC_KO.

Pseudovirus production and neutralization assays

Neutralizing antibodies against HIV-1 pseudoviruses were measured as a function of reduction in Tat-regulated luciferase (Luc) reporter gene expression in TZM-bl cells (71, 72). Neutralization ID50 titers were measured from specimens obtained at days 0 (visit 3, baseline), 14 (visit 4, 2 weeks post-1st vaccination), and 70 (visit 8, 2 weeks post-2nd (final) vaccination). Neutralization titer was defined as the serum dilution at which relative luminescence units (RLUs) were reduced by 50% (ID50) relative to the RLUs in virus control wells (cell + virus only) after subtraction of background RLUs (wells with cells only).

A specialized panel of 426c Env-pseudotyped viruses was used to detect and characterize early intermediates of VRC01-class bnAbs. Some VRC01 early intermediates neutralize 426c in a manner that requires deletion of one or more N-glycans in the vicinity of the CD4bs. Neutralization may be enhanced by Man5-enrichment of remaining N-glycans that otherwise are processed into larger complex-type glycans. Man5-enrichment was achieved by producing Env-pseudotyped viruses in 293S GnTI⁻ cells. 293T cells were used to produce Env-pseudotyped viruses with fully processed glycans. Mutation D279K was used to confirm CD4bs specificity. The viruses 426c.N276D/GnTI⁻ and 426c.N276D.N460D.N463D/GnTI⁻ can detect germline-reverted forms of VRC01, VRC07 and VRC20 but not germline-reverted forms of other CD4bs bnAbs (73).

A specialized panel of CH505 Env-pseudotyped viruses was used to detect and characterize early intermediates of CH235 and CH103 bnAb lineages. Neutralization by CH235 early intermediates is dependent on an N279K.G458Y double mutation and also requires Man5-enrichment of N-glycans that are otherwise processed into larger complex-type glycans. Man5-enrichment was achieved by producing Env-pseudotyped viruses in 293S GnTI⁻ cells. 293T cells were used to produce Env-pseudotyped viruses with fully processed glycans. Mutation N280D was used to confirm CD4bs specificity. CH505.N279K.G458Y/GnTI⁻ detects germline reverted and early intermediates of CH235 but not germline-reverted forms of other CD4bs bnAbs (74).

Neutralization by early intermediates of CH103 requires deletion of four N-glycans in the vicinity of the CD4bs, combined with Man5-enrichment of remaining N-glycans that otherwise are processed into larger complex-type glycans. Mutation S365P was used to confirm CD4bs specificity. CH505TF.gly4/GnTI⁻ detects germline-reverted and early intermediates of CH103 (74).

Statistical analysis of neutralization assay data

Response to a viral isolate was considered to be positive if the neutralization titer was greater than or equal to 20. Response rates were calculated with 95% Wilson score intervals. Response magnitude analyses included vaccine-recipients only and included responders and non-responders (with truncated titer measurements). There were no detectable neutralizing antibody responses to the eOD-GT8 vaccine (low-dose or high-dose) that indicated elicitation of early intermediates of VRC01-class, CH235, or CH103 bnAbs. There was one low-level low-dose response (ID50 = 22.47) to CH0505TF.gly4.S365P.2/293S/GnTI⁻ (CH103 precursor knock-out mutant) after the 2nd vaccination, but this was hard to interpret in the absence of a response to CH0505TF.gly4 (CH103 precursor). Negative results do not necessarily indicate a lack of bnAb precursors, since the virus panel only detects a subset of CD4bs bnAb precursors. There were no Tier 1A (MW965.26/293T/17 and MN.3/293T/17), placebo, or baseline responses. Due to overall lack of neutralization response, response rate and magnitude testing were not performed.

Overview of the VRC01-class B cell assay

VRC01-class B cell precursors were identified first based on their ability to bind to the CD4bs of the eOD-GT8 protein, determined using fluorescently-labeled eOD-GT8 proteins and a modified eOD-GT8 protein containing mutations specifically in the CD4bs epitope (eOD-GT8 KO11). Differential staining of B cells to the eOD-GT8 protein but not to the CD4bs mutant (eOD-GT8 KO11) indicated that B cells were specific to the CD4bs. Therefore, fluorescently-labeled eOD-GT8 and eOD-KO11 were combined with a flow cytometry antibody panel to identify and single-cell sort the specific B cell populations of interest: circulating memory B cells, plasmablasts and germinal center (GC) B cells. CD4bs-specific B cells bound two eOD-GT8 tetramers (GT8⁺⁺) but not tetramers of eOD-GT8 KO11 (KO⁻). Since the B cell sorting and sequencing needed to be performed on fresh samples for plasmablasts and lymph node (LN) FNAs, the sorting and sequencing was established and performed at two laboratories that were in close proximity to the clinical sites. Procedures and reagents were harmonized (except where noted otherwise) and verified at the two laboratories (figs. S5–S10 and tables S10–S19).

PBMC samples from weeks -4, 4, 8, 10, and 16 after the first vaccination were sorted for CD4bs-specific IgG memory B cells; draining axillary lymph node cells, acquired by FNA at weeks 3 and 11, were sorted for CD4bs-specific IgG germinal center (GC) B cells; and PBMC samples at week 9 (5–8 days after the second vaccination) were sorted for CD4bs-specific IgD− plasmablasts (PBs) (Fig. 1A and tables S10–11). For each sample, RT-PCR and DNA sequencing were applied to as many CD4bs-specific IgG cells as possible, up to a maximum of two 96-well plates. DNA for HCs and LCs (kappa and lambda) was sequenced by the Sanger method, and selected samples from the low dose vaccine group were re-sequenced using next-generation sequencing. BCR sequences were subjected to quality filtering and bioinformatic analysis. Additional details of these procedures are described below.

Tetramer probe preparation

Tetramers were prepared at a 4:1 molar ratio of monomeric eOD-GT8 or KO11 proteins to streptavidin. The total volume of protein, 1x PBS, and protease inhibitor (catalog# 539131–10VL; MilliporeSigma; Burlington, MA, USA) was added to a microcentrifuge tube. Twenty percent of the total streptavidin volume was added and incubated with continuous rotation for 20 min at 4°C in the dark. The incremental addition of streptavidin was repeated 4 times until the total amount of streptavidin had been added to the protein. Tetramers were made fresh for each experiment (VRC) or stored at 4°C and kept to up to 1 month (FHCRC).

Monoclonal bead controls

Bead assays were performed on the day of sorting experiments to confirm the functionality of the tetramers by flow cytometry. Anti-mouse Ig kappa beads (BD Bioscience, La Jolla, CA, USA) were washed with 3 mL R10 in polystyrene FACS tubes and then centrifuged at 650 × g for 5 mins. Supernatants were discarded and beads were resuspended in 100μL R10. Each bead control was given 1μg of mouse anti-human IgG and incubated for 15 mins at room temperature. Beads were washed with 3mL R10 and resuspended with either 1μg of glVRC01 mAb (to test eOD-GT8 probes) or KG064 mAb (to test eOD-GT8-KO11 probes) in 100μL. Beads were incubated for 15 mins at room temperature, washed with 3mL R10 and then resuspended in 100μL R10. Both eOD-GT8 probes were added to the glVRC01 positive control beads as well as a negative control beads that did not receive glVRC01. The KO11 probe was added to the KG064 mAb-labeled beads as well as a bead-only control that did not receive KG064. Each tube was incubated for a minimum of 25 mins at 4°C. Beads were washed with 3mL R10, centrifuged, decanted and resuspended in up to 300μL R10 for collection.

Enrichment of B cells from cryopreserved PBMC samples

All five cryopreserved PBMC samples from a single participant were sorted in an experimental batch on a given day. Cryopreserved PBMCs were thawed, either by water bath or with use of Thawsome cryovial adapters, into warm benzonase (MilliporeSigma; Burlington, MA, USA) supplemented R10 [RPMI 1640 with 25mM HEPES buffer and L-glutamine (Thermo Fisher Scientific; Waltham, MA, USA) and 10% fetal bovine serum (FBS; Nucleus Biologics, San Diego, CA, USA)]. Samples were centrifuged at 300 × g for 12 minutes. The supernatant was decanted, and the cells were washed with R10. The cell recovery and viability were optionally determined using the Muse Cell Analyzer (MilliporeSigma; Burlington, MA, USA). B cells were enriched using the EasySep Human Pan-B Cell Enrichment Kit (StemCell Technologies; Vancouver, CA) and the Big Easy magnet (StemCell Technologies; Vancouver, CA) following the manufacturer’s instructions summarized here. Samples were resuspended at a concentration of 5×10⁷ cells/mL with EasySep buffer. Fifty microliters of Enrichment Cocktail (per mL of sample) were added and mixed. The samples were incubated at room temperature for 10 min. Magnetic particles were vortexed for 30 seconds, and 75 μL magnetic particles (per mL of sample) were added and mixed with the samples and incubated at room temperature for 5 mins. EasySep Buffer was added to the sample up to 5 mL for samples under 2 mL (<10⁸ cells) or 10 mL for samples greater than or equal to 2 mL (>10⁸ cells). The sample was loaded into the EasySep magnet without the lid and incubated at room temperature for 5 mins. With the tube on the magnet, the cell suspension was pipetted off and into a new conical. The sample conical was removed from the magnet, and the beads were resuspended with the same amount of EasySep Buffer used previously and then mixed and re-incubated on the EasySep magnet (without lid) at room temperature for 5 mins. With the tube on the magnet, the cell suspension was pipetted off and combined with the previous matching sample. The cell number and viability of the sample was optionally determined.

Flow cytometry staining procedures

Cells, either fresh PBMCs from the plasmablast timepoint (week 9), fresh lymph node mononuclear cells (from FNAs at weeks 3 and 11), or enriched B cells from previously cryopreserved samples (at weeks -4, 4, 8, 10 and 16), were resuspended in 100uL 10%FBS/1x PBS and stained with the fluorescently labeled eOD-GT8 KO11 tetramer for 30 minutes at 4°C. Cells were washed with 10% FBS/1xPBS and then resuspended in a staining cocktail of antibodies and remaining eOD-GT8 tetramers diluted in Brilliant Buffer (BD Bioscience, La Jolla, CA, USA). The antibody staining panels were tissue dependent and included a panel for PBMC samples (memory B cells and plasmablasts) (table S10), and a separate panel for lymph node FNAs (table S11). After surface staining, the samples were resuspended at about 2×10⁶ cells/mL in R10 containing the viability dye (7AAD) and maintained at 4°C until processed by flow cytometry.

Standardized templates for flow cytometry analysis

Standardized templates were designed using BD Diva software for analysis of bead assays, internal controls, and experimental samples for each sample type (fig. S7). These templates were replicated for each flow cytometry experiment to ensure consistent gating and labeling of samples. Templates were tested and compared between both test sites using aliquots of the internal control sample before trial samples were used. Since antigen-specific events were rare in the internal control sample which made setting gates difficult, we developed a standardized gating system that could be implemented at both sites referred to as the M.A.R.I.O gate.

Mathematically Articulated and Reasoned but Independently Optimized (M.A.R.I.O.) Gating

The analysis of fresh samples for the clinical trial required the use of equipment that was located near the clinical trial sites. The physical differences between flow cytometers, and the subjectivity of gating rare and hard to identify cell populations, can lead to high levels of variability in the sorting of populations like antigen-specific B cells. For this trial, we developed a method of standardizing the gating of antigen-specific B cells by flow cytometry. eOD-GT8 proteins were conjugated to two fluorophores, allowing us to identify fluorescently double-positive events and reduce the fluorophore-specific background that would be selected by using either fluorophore individually. Due to the variability between flow cytometers, developing a method to standardize the gating of antigen-specific events by flow cytometry was important for this trial. Here we utilized FMP (fluorescence minus probe) stained samples (fully stained samples without the addition of fluorescent probes) to determine the background for each channel being used to detect antigen-specific B cells. We then developed a gate based on the ratio of these fluorophores for each flow cytometer which could identify fluorescently well-balanced antigen-specific B cells while also reducing the amount of antigen-non-specific B cells captured by the gate.

FMP samples were used to calculate the coordinates for the 99^th percentile of each probe-specific fluorophore using Flowjo software. This serves as the upper threshold of the probe negative gate (fig. S7A; C1). Similarly, twice the 99^th percentile of each probe specific fluorophore was determined to serve as the lowest acceptable threshold for probe double positive-specific events (fig. S7A; C2). Axis points for each individual fluorophore were placed at coordinates representing ten times the 99^th percentile for each individual fluorophore (fig. S7A; C4, C5). Additional axis points were added which are based on the proportion of the 99^th percentiles for each fluorophore (fig. S7A; C6, C7). These bounds were used to define the area in which acceptable fluorescent probe double positive events were acceptable for sorting by each test facility and account for each facility’s individual cytometer characteristics (fig. S7B).

FACS Sorting

Individual target cells were index-sorted into empty skirted 96-well plates (Thermo Fisher Scientific; Waltham, MA, USA). Several wells were left empty for subsequent PCR controls. PBMC samples from weeks -4, 4, 8, 10, and 16 after the 1^st vaccination were sorted until either four 96-well plates were filled with eOD-GT8 CD4bs-specific-specific IgG memory B cells (CD19⁺/CD20⁺/IgD⁻/IgG⁺/KO⁻/GT8⁺⁺; fig. S8) or the sample was exhausted; two 96-well plates were subjected to RT-PCR and BCR sequencing, while the other two plates were reserved at −80°C. The number of PBMCs thawed and stained depended on the timepoint: 200×10⁶ (FHCRC) or 250×10⁶ (VRC) at week -4, 100×10⁶ at week 4, 100×10⁶ at week 8, 200×10⁶ at week 10, and 100×10⁶ at week 16.

In planning the trial procedures, we expected that GC B cells from LN FNAs and plasmablasts (PBs) would have lower levels of surface IgG, and hence we were uncertain if sorting of these samples would have sufficient detectable antigen-specific staining by flow cytometry. We therefore decided to sort two plates of phenotype-specific cells without regard to tetramer binding, followed by two plates of phenotype-specific cells that were also CD4bs-specific as determined by differential tetramer staining. Therefore, up to two plates of total GC B cells (CD19⁺ IgD⁻ IgG⁺ CD20^hi CD38^hi) were sorted, as well as CD4bs-specific GC B cells (CD19⁺ IgD⁻ IgG⁺ CD20^hi CD38^hi KO⁻ GT8⁺⁺) (fig. S9A). From fresh PBMCs at week 9, two plates of PBs (CD19⁺ CD27⁺ IgD⁻ CD38^hi) were sorted without regard to tetramer binding followed by the sorting of CD4bs-specific PBs (CD19⁺ CD27^hi CD38^hi IgD⁻ KO⁻ GT8⁺⁺) (fig. S9B). Note that the PB gating strategy, which included CD38^hi and did not gate on CD20, may have included CD38^hiCD20^hi B cells that have been described as activated B cells as opposed to PBs (10). A maximum of four 96-well plates of B cells were sorted for cDNA synthesis per time point. Plates were sealed, centrifuged at 800 × g for 1 min and immediately transferred to dry ice. Plates were stored at −80°C for at least overnight before proceeding. A total of 884 (384 at VRC; 500 at FHCRC) 96-well plates of sorted cells were produced and stored.

Standardized and direct analysis of sorting flow cytometry data

To allow for direct analysis of the flow cytometry data as it was analyzed during the sorts, the flow cytometry experiments were exported from Diva software as xml files along with the FCS files for the total recorded events and the index files for the sorted cells. These primary data were processed and concatenated at the Vaccine Immunology Statistical Center (VISC), using the Cytoverse suite of R packages. CytoML (75) is an R/Bioconductor package that enables cross-platform import, export, and sharing of gated cytometry data. Using CytoML we parsed the XML files created during the Diva experiments, which contained data transformations, compensation matrices, gates and their hierarchical relationships, sample meta-data and other information required to reproduce the gating analysis. Once the workspaces were imported into R, the analysis was faithfully reproduced for each sample, the gated cytometry data was visualized, and positive proportions for all populations were calculated. As a quality control measure, the calculated percent positive populations were compared to exported tables of populations from the Diva software using Spearman correlations. This data was processed using R (v4.1.1) and Cytoverse packages CytoML (v2.5.4), flowWorkspace (76) (v4.5.3), ggcyto (77) (v1.20.0), and cytolib (v2.5.3). Frequencies were measured either during the period of sorting (FHCRC) or until the samples were exhausted (VRC).

PCR and sequencing positive and negative controls

Early in the trial, the lysates from an immortalized VRC01-class naive B cell line, referred to as Immo-A1 (78), and sorted pooled donor CD19⁺ B cells were used as PCR controls. These were found to be insufficient as controls, so we developed synthetic positive controls to use for the remainder of the trial. Synthetic positive control DNA sequences compatible with our nested PCR protocol were generated for B cell receptor heavy and both light chains (kappa and lambda). Synthetic sequences were based on natural sequences identified from VRC01-class antibody sequences isolated from human PBMCs. The CDR3 amino acid junctions were altered to distinguish each of the synthetic controls from naturally occurring sequences and from each other by replacing them with amino acid “VRC” “VIP” and “DL” segments separated by chain-specific amino acids. Heavy chain CDR3 sequence tags coded for histidine (H) leading and separating the segments. Similarly, kappa chain CDR3 sequence tags coded for lysine (K) and lambda chain CDR3 sequences coded for leucine (L) leading and separating the segments. The sequence tags are shown below, and the full nucleotide sequences for the synthetic controls are given in table S19.

Sequence Tags:

1. 5’ GVRCGVIPGDL 3’ 5’ ggcgtgcgctgcggcgtgattccgggcgatctg 3’

2. 5’ KVRCKVIPKDL 3’ 5’ aaagtgcgctgcaaagtgattccgaaagatctg 3’

3. 5’ LVRCLVIPLDL 3’ 5’ ctggtgcgctgcctggtgattccgctggatctg 3’

To further differentiate the synthetic controls by sequence as well as by physical size using gel electrophoresis for quality control purposes, novel sequences were inserted at both ends of the sequences. Inner and outer primer binding sites were attached to ensure these sequences would amplify with our nested PCR primer sets. Ultimately control sequences were ~50% mutated from natural sequences distributed throughout the sequence to ensure identification even with high mutation and poor sequence quality.

Synthetic controls were tested for primer binding capacity, in frame amino acid translation, Ig chain homology and gene usage and potential gBlock production issues. All controls were titered and rigorously tested prior to use in the trial.

Single-cell sequencing

In preparation for cDNA synthesis, plates were removed from −80°C storage and again centrifuged at 800 × g for 1 minute before being placed on wet ice. The cDNA synthesis reaction mix was constructed using the SuperScript III Reverse Transcriptase Kit (Invitrogen). The reaction mix consisted of 1x SuperScript III First Strand Buffer, 4.8 mM DTT and 200 U SuperScript III, as well as 20 U RNaseOUT (Invitrogen), 450 ng Random Hexamers (Gene Link) and 0.77 mM dNTP mix (GeneAmp). The mix was added to every well, including wells 12E through 12H to be used as negative controls. Thermal cycling parameters were 42°C for 10 min, 25°C for 10 min, 50°C for 60 min and 94°C for 5 min. Once complete, plates were frozen at −20°C. A total of 531 (267 at VRC, 264 at FHCRC) unique cDNA plates were generated for subsequent PCR amplification.

Amplification proceeded by three immunoglobulin chain-specific PCRs. The reaction mix generally consisted of 2 U HotStarTaq Plus DNA polymerase and 1x HotStar Plus PCR Buffer (Qiagen; Hilden, Germany), as well as 0.25 mM dNTP mix and 1.5 mM MgCl₂. Each chain was amplified using a chain-specific pool of forward primers (IDT) and isotype-specific reverse primers (IDT) (tables S13, S15, S17). One reaction mix each was constructed containing 2.1 μM forward and 0.2 μM IgH reverse primers, 0.3 μM forward and 0.625 μM reverse Igκ primers, and 0.8 μM forward and 0.625 μM reverse Igλ primers. The template input was 4 μL of cDNA, including carrying forward 4 μL of each negative control from wells 12E through 12H. The cDNA plates were briefly vortexed and centrifuged before template addition. Thermal cycling parameters were 95°C for 5 min and 50 cycles of 95°C for 30 sec, either 52°C (IgH) or 58°C (Igκ and Igλ) for 30 sec and 72°C for 55 sec, followed by a final extension of 72°C for 10 min. Once complete, plates were frozen at −20°C.

An additional nested PCR was performed for all chains to increase specificity. The reaction mix consisted of 1 U Phusion High-Fidelity DNA Polymerase and 1x Phusion High-Fidelity Reaction Buffer (New England Biolabs; Ipswich, MA, USA), as well as 0.2 mM dNTP mix and 1x Q Solution (Qiagen; Hilden, Germany). Each chain was again amplified using a chain-specific pool of forward primers and isotype-specific reverse primers, with one reaction mix each containing 5.5 μM forward and 1.0 μM IgH reverse primers, 4.0 μM forward and 5.0 μM reverse Igκ primers, and 3.5 μM forward and 5.0 μM reverse Igλ primers (tables S14, S16, S18). The template input was 4 μL of chain-specific PCR product, again including carrying forward 4 μL of each negative control from wells 12E through 12H. In addition, chain-specific positive PCR controls (synthesized by IDT) were included in wells 12A through 12D. These synthetic constructs were added at 400 copies per well. The PCR plates, including the positive PCR controls, were briefly vortexed and centrifuged before template addition. Thermal cycling parameters were 98°C for 30 sec and 35 cycles of 98°C for 30 sec, either 58°C (IgH and Igλ) or 52°C (Igκ) for 30 sec and 72°C for 55 sec, followed by a final extension of 72°C for 10 min. Once complete, the plates were frozen at −20°C and shipped to Genewiz (South Plainfield, NJ) on dry ice. Genewiz conducted an enzymatic clean-up on the PCR products before conducting uni-directional Sanger sequencing using the reverse chain-specific constant region primers (3’Cg CH1, 3’CK 494, and 3’CL shown in tables S14, S16, S18). Sequencing data were obtained for a total of 3186 PCR plates (6x the total number of cDNA plates), not accounting for repeats. Quality control measures were taken for all sequence data to determine if sequencing of specific plates needed to be repeated. When PCR controls were not available, we required a minimum of 50% of the sequences on any plate to have identifiable and functional B cell receptor sequences. When PCR controls were available, we required >50% of the positive controls to have identifiable control sequences (2 to 4) and the negative controls to have <50% of the wells with identifiable sequences (1 or 0) (VRC) or we required the negative controls to have <50% of the wells with identifiable sequences (1 or 0) (FHCRC). Sequences quality was first checked with the Genewiz quality report to determine overall sequencing success for the plate. Sequences were then analyzed by IMGT/V-QUEST for productive B cell receptor sequences or sequences with identifiable B cell receptor gene usage. When synthetic PCR controls were used, all sequences were screened for contamination by the spiking of the controls using the CDR3 amino acid junction sequence tags mentioned previously. All data was transferred to the Vaccine Immunology Statistical Center (VISC) for data repository and analysis. However, sequencing plates that did not pass our criteria were investigated further and either sequenced again directly by Genewiz or PCR amplified from an earlier step with remaining cDNA or PCR material. The new plates were submitted to Genewiz for sequencing, and new sequence data was identified with an increased “Replicate” number.

Additional PCR and NGS BCR sequencing for low dose group

To control for potential sequencing errors in the uni-directional/single-read Sanger sequencing data, we resequenced selected samples from the low dose group using next generation sequencing (NGS). To perform NGS on selected cells, cDNA generated for Sanger sequencing was used as the starting point. Plates were removed from −80°C storage and centrifuged at 800 ×g for 1 minute before being placed on wet ice. Selected wells from different plates were collected into a new plate, and PCR was performed using the same nested method used for Sanger sequencing. The nested PCR product from the individual heavy and light chains were pooled in equal volumes of 10 μL from each plate into a single plate. The pooled plate was purified using AmpureXP beads (Beckman coulter Cat: A63881) in the ratio of 0.8X beads to PCR product for a total volume of 24 μL of beads to 30 μL of pooled PCR product. The purified product was eluted into 20 μL of RNase-free water. Each purified plate was quantified using the QuantIT dsDNA High sensitivity kit (Thermofisher Cat: Q33120) and normalized to 0.2ng/μL for optimal NGS library generation using the Illumina Nextera XT kit (Illumina Cat: FC1–31-1096). The normalized plates were processed for NGS library preparation using 1.25 μL of the product with 2.5 μL of Tagmentation buffer and 1.25 μL of the Amplicon Tagmentation mix provided in the Nextera kit. The plates were centrifuged at 800 ×g for 1 minute and placed in the thermocycler at 55°C for 10 minutes. Following the incubation, 1.25 μL of Neutralization buffer was added to the wells and incubated at room temperature for 5 minutes. The tagmented product was indexed using custom primers generated to multiplex 10 plates together in a single sequencing run. Each plate was indexed with 1.25 μL of a unique 10 base pair Mi7 index at a concentration of 10 μM and each well of the plate was indexed with 1.25 μL of a unique 10 base pair Mi5 index at a concentration of 10 μM, along with 3.75 μL of the NPM reaction mix provided in the Nextera XT kit. The PCR plates were centrifuged at 800 ×g for 1 minute and placed in the thermocycler for indexing. The thermocycler parameters used were 72°C for 3 minutes, 12 cycles of 95°C for 30 sec, 50°C for 30 sec, and 72°C for 1 min, followed by a final extension of 72°C for 5 min. The indexed libraries were further PCR purified using the AmpureXP beads at 0.8X concentration and eluted into 10 μL of RNase-free water. The purified product from 10 indexed plates was pooled together for loading onto the Illumina Miseq for sequencing. The final concentration of the pooled library was determined using the Kapa QPCR kit (Roche Cat: KK4835) and diluted to 4nM for sample denaturating according to the Illumina loading guidelines. The denatured sample was further diluted to 10 pM concentration for loading onto the Miseq V3 600 cycle kit (Illumina cat: MS-102–3003). Demultiplexed fastq files were retrieved from the Miseq and aligned to immunoglobulin genes using BALDR (79).

High quality NGS sequences with at least 100 Nextera reads were obtained for a total of 1,998 samples and were regarded as perfectly accurate. Comparison of high quality NGS sequences to corresponding Sanger sequences revealed perfect agreement for 82.9% of Sanger reads at the nucleotide level and 85.9% of Sanger reads at the amino acid level. There were differences of 1, 2, 3, 4, or 5 nucleotides for 7.8%, 2.3%, 1.3%, 0.65%, or 0.75% of Sanger reads, respectively, and differences of 1, 2, 3, 4, or 5 amino acids for 6.7%, 1.8%, 1.3%, 0.70%, or 0.50% of Sanger reads, respectively. Hence, the NGS data confirmed that >94.2% of the Sanger sequencing reads were accurate to within 3 nucleotides, and >95.6% of the Sanger sequencing reads were accurate to within 3 amino acids. We concluded that it was not necessary to carry out NGS resequencing on the remaining low dose samples or on any of the high dose samples. Nevertheless, a small fraction (<4.4%) of Sanger reads differed by more than 3 amino acids from the corresponding high quality NGS sequences. High quality NGS sequences were used in place of the corresponding Sanger sequences for downstream analysis in cases where Sanger and NGS differed by 1 to 19 nucleotides, as described in the NGS module of the BCR bioinformatic analysis pipeline below.

Sample identify confirmation for one case of accidental sample mislabeling

Comparison of serum antibody binding data from BAMA with antigen-specific B cell frequency data from cytometry suggested that samples from visits 8 and 10 (weeks 10 and 16) for placebo-recipient PubID_164 might have been accidentally swapped with samples from the same timepoints for vaccine-recipient PubID_153 during flow cytometry analysis. Short Tandem Repeat (STR) analysis along with confirmatory flow cytometry testing was used to confirm the identify of the samples. Duplicate samples were thawed and divided. Small aliquots were lysed and transported to two independent facilities for STR testing to confirm donor sample identity. Remaining samples were processed by trial flow cytometry protocols for comparison of cellular phenotypic data. Both STR testing facilities were in agreement with each other and the confirmatory flow cytometry data to confirm sample identities. This analysis confirmed that a sample labeling error of transposing the participant IDs had occurred during analysis by flow cytometry, effectively swapping the samples for the two participants at the timepoints listed above. The sample identity confirmation provided justification for the bioinformatic pipeline to correct for the effective “sample swap”, as described in the SWAP module below.

Bioinformatic BCR sequence processing

Computing VRC01-class frequencies by combining frequencies from FACS and BCR sequencing

Key B cell frequencies reported

B cell frequencies and B cell receptor (BCR) signatures were measured by flow cytometry (FACS) and BCR sequencing, respectively. In this section we list the key frequencies reported in the manuscript and/or in the supplementary data files. Additional frequencies not listed here are provided in the supplementary data files. Methods to estimate frequencies based on combined cytometry and BCR sequence analyses are given in the preceding section, Computing VRC01-class frequencies by combining frequencies from FACS and BCR sequencing.

The key frequencies reported for PBMC samples (weeks -4, 4, 8, 10, 16) were as follows.

From cytometry analysis:

• Percent of IgG memory B cells that were GT8⁺⁺ (regardless of KO binding status)

• Percent of GT8⁺⁺ IgG memory B cells that were KO⁻

• Percent of IgG memory B cells that were CD4bs-specific

From combined cytometry and BCR sequence analysis:

• Percent of CD4bs-specific IgG memory B cells that were VRC01-class

• Percent of GT8-specific IgG memory B cells detected as VRC01-class

• Percent of IgG memory B cells detected as VRC01-class

• Percent of B cells detected as VRC01-class

The key frequencies reported for FNA samples (weeks 3 and 11) were as follows.

From cytometry analysis:

• Percent of IgG GC B cells that were GT8⁺⁺ (regardless of KO binding status)

• Percent of GT8⁺⁺ IgG GC B cells that were KO⁻

• Percent of IgG GC B cells that were CD4bs-specific

From combined cytometry and BCR sequence analysis:

• Percent of CD4bs-specific IgG GC B cells that were VRC01-class

• Percent of GT8-specific IgG GC B cells detected as VRC01-class

• Percent of IgG GC B cells detected as VRC01-class

• Percent of B cells detected as VRC01-class

• The key frequencies reported for PB samples (week 9) were as follows.

From cytometry analysis:

• Percent of IgD⁻ PBs that were GT8⁺⁺ (regardless of KO binding status)

• Percent of GT8⁺⁺ IgD⁻ PBs that were KO⁻

• Percent of IgD⁻ PBs that were CD4bs-specific

From combined cytometry and BCR sequence analysis:

• Percent of CD4bs-specific IgD⁻ PBs that were VRC01-class

• Percent of GT8-specific IgD⁻ PBs detected as VRC01-class

• Percent of IgD⁻ PBs detected as VRC01-class

• Percent of B cells detected as VRC01-class

VRC01-class response calls

The criteria for determining if detection of one or more VRC01-class B cells in a sample represented a vaccine-induced VRC01-class response were as follows.

Pre-trial evaluation of baseline VRC01-class IgG B cell frequency

During pre-trial development of the B cell sorting and BCR sequencing assay, we sought to determine an estimate for the baseline frequency of VRC01-class IgG B cells in humans. We applied the assay to leukapheresis samples from 35 healthy, HIV-unexposed donors, including approximately 12 billion PBMCs and approximately 29.2 million IgG⁺ memory B cells. From 142 BCRs with HC and LC sequenced out of 171 CD4bs-specific cells sorted, we detected no VRC01-class B cells, indicating a relatively low baseline frequency of VRC01-class B cells in the human IgG memory B cell repertoire. In contrast, we verified by sorting of human naive IgM+ B cells that the same assay isolated VRC01-class human naive precursors at frequencies similar to those reported previously (not shown).

IGHV1–2 genotype analysis using IgDiscover

Genotyping of the G001 participants was performed under an ethics approval from the National Ethical Review Agency of Sweden (decision #2021–01850). Total RNA was isolated from frozen PBMC samples prepared from each participant (n=48) using the Qiagen RNeasy kit and protocol. 200 ng of total RNA was used for cDNA synthesis with an IgM constant region primer that contained a Unique Molecular Identifier (UMI), according to the procedures described previously (87). Two independent 5’ multiplex libraries (L and U) were produced for each case using either a IGHV leader region specific primer set (L) or a primer set that targeted the 5’ UTR (U). The libraries were individually indexed and sequenced on the Illumina MiSeq platform using the V3 2 × 300 cycle kits (Illumina). The Read 1 and Read 2 files for each library were processed using the IgDiscover analysis pipeline as described previously (87, 88). The IgDiscover program (version 0.12.4.dev266+ge9e3119) enabled the analysis of IGHV1–2 allele content in each case, using the two independent libraries for high confidence allele identification.

BCR sequence hierarchical clustering

Average linkage agglomerative clustering (Fig. 3 and figs. S19–S20 and S24–S25) was performed using the SADIE cluster module. The module uses scikit-learn (89) to perform clustering specifically on BCR sequences with the AIRR annotation schema. Sequences were split into VRC01-class and non-VRC01-class and then grouped by HCDR3 and LCDR3 length. For each subgroup, the distance between two antibodies was the sum of Levenshtein distances between the corresponding HCDRs and LCDRs (HCDR1 distance + HCDR2 distance + HCDR3 distance + LCDR1 distance + LCDR2 distance + LCDR3 distance). To facilliate clustering of both mutated and un-mutated BCRs with a single distance cutoff, the “somatic pad” option was used to subtract all common somatic mutations from the total Levenshtein distance for any pair of BCRs (90). With this option, the minimal distance between two BCRs was zero. The sparse upper-triangular distance matrix was then saved into memory. Clustering was then performed using scikit-learn with the pre-computed distance matrix using average-linkage clustering and a distance cutoff of 3. The cluster labels were extracted and added to the output dataframe. The centroid of every cluster was computed as the lowest mean distance to every other member in the cluster. The all-vs-all, intra-cluster, and inter-centroid distances were then computed from the saved distance matrix.

Bioinformatic BCR sequence analysis other than VRC01-calls or clustering

BCR V gene assignments and mutation levels were determined using SADIE, which was also used in the ANNOTATION module described above. Mutation analysis for VH1–2 genes was made more accurate by accounting for the personalized IGHV1–2 genotypes as described in the PERSONALIZE module above. Statistical quantile analyses throughout the manuscript were carried out using R and Pandas (https://github.com/pandas-dev/pandas). Confidence intervals for the non-VRC01-class response rate (fig. S19C) were computed using the Wilson score method (91). Statistical comparisons for BCR mutation levels at different timepoints (Fig. 5, fig. S20, tables S46–S49), were carried out using a Wilcoxon signed-rank test for paired data using the VISCfunctions R package (https://github.com/FredHutch/VISCfunctions). Analyses of VRC01-class BCR features (Fig. 6 and figs. S16 and S24–S29) were carried out by custom python functions available as a package in the data repository for G001 (htttps://github.com/SchiefLab/G001). We detected few recurring VRC01-class clones at different timepoints, and therefore we did not carry out analyses of intra-clonal SHM over time as have been reported elsewhere (64). Inferred germline amino acid sequences were computed by reverting templated V, D, and J gene segments to their germline sequences for the alleles predicted on the basis of the antibody nucleotide sequence; in this process, VH1–2 allele predictions were made more accurate by restricting the allowed alleles for each participant to those experimentally determined by VH1–2 genotyping for that participant. When referring to control distributions from OAS (92, 93), we restricted to human, non-vaccinated, no disease state data from OAS. In data we report from OAS, each symbol refers to a separate individual.

Antibody selection and production for SPR

For IgG expression, we selected at least two VRC01-class and one non-VRC01-class BCR from each participant at each post-vaccination timepoint, attempting to represent the genetic diversity and mutation levels of the low dose BCRs. Although some mAbs failed to express, we carried out SPR analyses for 267 VRC01-class and 145 non-VRC01-class BCRs across all participants and post-vaccination timepoints (Fig. 7, fig. S29, and Data S2). To assess affinities of potential naive precursors to the post-vaccination BCRs, we evaluated eOD-GT8 binding to inferred-germline antibodies (iGLs) for 118 VRC01-class and 55 non-VRC01-class post-vaccination BCRs. For comparison we also measured eOD-GT8 binding to pre-vaccination IgG memory BCRs (6 of 7 VRC01-class BCRs from both dose groups and 53 of 55 non-VRC01-class BCRs from the low dose group) and human naive VRC01-class (n=62) and non-VRC01-class (n=72) precursors specific for the CD4bs of eOD-GT8 and isolated previously from 12 other HIV-unexposed individuals (21, 43) (Fig. 7 and Data S2).

Genes encoding the antibody Fv regions were synthesized by GenScript and cloned into antibody expression vectors pCW-CHIg-hG1, pCW-CLIg-hL2, and pCW-CLIg-hk. Monoclonal antibodies were produced by GenScript using transient transfection in the Expi293F system (ThermoFisher) and were purified using HiTrap MabSelect SuRe columns (Cytiva). Additional antibody production was conducted in house using transient transfection of HEK-293F cells (ThermoFisher) and purification using rProtein A Sepharose Fast Flow resin (Cytiva). We note that the antibodies were produced as IgG1, but our B cell sorting and BCR sequencing workflow did not distinguish between IgG subclasses.

Antigen production for SPR

His-tagged monomeric and trimeric antigens were produced by transient transfection of HEK-293F cells (ThermoFisher) and purified by immobilized metal ion affinity chromatography (IMAC) using HisTrap excel columns (Cytiva) followed be size-exclusion chromatography (SEC) using either Superdex 75 10/300 GL or Superdex 200 Increase 10/300 GL columns (Cytiva).

SPR

We measured the kinetics and affinities of antibody-antigen interactions on a Carterra LSA instrument using HC30M or CMDP sensor chips (Carterra) and 1x HBS-EP+ pH 7.4 running buffer (20x stock from Teknova, Cat. No H8022) supplemented with BSA at 1 mg/ml. We followed Carterra software instructions to prepare chip surfaces for ligand capture. In a typical experiment, approximately 2500 to 2700 RU of capture antibody (SouthernBiothech catalog #2047–01) in 10 mM Sodium Acetate pH 4.5 was amine coupled. The critical detail here was the concentration range of the amine coupling reagents and capture antibody. We used N-Hydroxysuccinimide (NHS) and 1-Ethyl-3-(3-dimethylaminopropyl) carbodiimidehydrochloride (EDC) from Amine Coupling Kit (GE order code BR-1000–50). As per kit instruction 22–0510-62 AG, the NHS and EDC should be reconstituted in 10 ml of water each to give 11.5 mg/ml and 75 mg/ml respectively. However, the highest coupling levels of capture antibody were achieved by using 10 times diluted NHS and EDC during surface preparation runs. Thus, in our runs the concentrations of NHS and EDC were 1.15 mg/ml and 7.5 mg/ml. The concentrated stocks of NHS and EDC could be stored frozen in −20C for up to 2 month without noticeable loss of activity. The SouthernBiotech capture antibody was buffer exchanged into 10 mM Sodium Acetate pH 4.5 using Zeba spin desalting columns 7K MWCO 0.5ml (catalog # 89883 from Thermo) and was used at concentration 0.25 mg/ml with 20 minutes contact time. Phosphoric Acid 1.7% was our regeneration solution with 60 seconds contact time and injected three times per each cycle. Solution concentration of ligands was around 5 ug/ml and contact time was 3 min. Raw sensograms were analyzed using Kinetics software (Carterra), interspot and blank double referencing, Langmuir model. Analyte concentrations were quantified on NanoDrop 2000c Spectrophotometer using absorption signal at 280 nm. A typical SPR run tested 6 different analyte concentrations using a dilution factor of 4. Maximum analyte concentration for eOD-GT8 was generally 10 μM, except for weak binders which were generally re-run at higher maximum analyte concentrations of 50 or 118 μM. For eOD-GT6 and eOD-GT6 variant analytes, maximum analyte concentrations were 9 to 16 μM, but weak or expected weak binders were run at 37–63 μM. For MD39 trimer, maximum analyte concentration was 4 or 11 μM, but most interactions were tested at 11 μM. For core-gp120, maximum analyte concentration were 11 or 46 μM. For best results, analyte samples were buffer exchanged into the running buffer using desalting columns or dialysis. We typically covered a broad range of affinities in our runs, and the best referencing practices were different depending on how fast the off-rate was for each particular ligand. For fast off-rates (>9×10⁻³ 1/s 1×10⁻² 1/s) we used automated batch referencing that included overlay y-align and higher analyte concentrations. For slow off-rates (≤ 9×10⁻³ 1/s), we used manual process referencing that included serial y-align and lower analyte concentrations. After automated data analysis by the Carterra Kinetics software, we also performed additional filtering to remove datasets with highest response signals smaller than signals from negative controls. This additional filtering was performed automatically using an R-script. The script also identified ligands for which capture on the sensor chip was insufficient, and measurements involving those ligands were excluded from subsequent analyses. Many interactions were measured more than once. Whether or not multiple measurements were available, we used the following algorithm to select the representative measurement for any particular interation:

1. If no measurement resulted in a kinetic-fit K_D from the Carterra Kinetics software analysis, the K_D value was set to ≥100 μM.

2. If all available measurements had kinetic-fit K_Ds from the Carterra Kinetics software analysis that were ≥5 times the maximum analyte concentration used in the measurement, the K_D value was set to ≥100 μM. The lowest max analyte concentration for which this was invoked was 37 μM, in which case the kinetic-fit K_D of >185 μM was reported as ≥100 μM.

3. If at least one measurement had a kinetic-fit K_D from the Carterra Kinetics software analysis, and if k_on and k_off were within range for the instrument (10 min⁻¹M⁻¹< k_on< 1×10⁹ min⁻¹M⁻¹; 1×10⁻⁵ min⁻¹< k_off < 5×10⁻¹ min⁻¹), the measurement with the lowest chi-squared fit value was chosen as the representative measure.

4. If all available measurements with a kinetic-fit K_D had a k_off that was out of range (this only occured with fast k_off), then if the kinetic-fit K_D was within a factor of 3 to the equilibrium-fit K_D, the ratio of kinetic K_D to equilibrium K_D was calculated, the measurement with ratio closest to 1 was chosen as the representative, and equilibrium K_D was reported for that measurement. If no kinetic-fit K_D was within a factor of 3 to the equilibrium-fit K_D, the K_D was reported as ≥100 uM. The reporting of equilibrium-fit K_Ds was not common: in Fig. 7, n=2 of 267 post-vaccination VRC01-class K_D values, and n=10 of 145 post-vaccination non-VRC01-class K_D values, were reported as the equilibrium-fit K_D.

5. We had no cases of k_on out of range, thus the algorithm did not have to deal with that case.

Regression analyses involving SPR-measured quantities

Linear mixed effects models (LMEs) were used to analyze the linear dependence of one variable on another, for the following dependencies: (i) dependence of K_D for eOD-GT6 or its variants on K_D for eOD-GT8, for VRC01-class BCRs (Fig. 8B); (ii) dependence of iGL K_D for eOD-GT8 on the number of mutations in the BCRs from which the iGL was inferred, for VRC01-class and non-VRC01-class BCRs (fig. S30B); (iii) dependence of eOD-GT8 K_D on k_on and k_off, for VRC01-class and non-VRC01-class BCRs (fig. S32C); (iv) dependence of eOD-GT8 K_D, k_on, and k_off, on the number of BCR V gene mutations (figs. S34–S36). LMEs were used to estimate the slope of each association [e.g. change in log(K_D) per mutation]. Estimated associations only included antibodies for which the K_D value was measureable (K_D<100 μM). Each LME had fixed effects for the intercept and slope. Random effects for the intercept and slope were included for each participant, with the exception of estimation for a given week (figs. S34B, S35B, and S36B) or for non-VRC01-class iGL (fig. S30B), where a random effect for only the intercept was included. For non-VRC01-class BCRs at week 16 in figs. S34B, S35B, and S36B, only one antibody per participant was available, so estimates were based on a linear model (K_D, k_on, and k_off). The estimated regression line and shaded 95% prediction interval were generated using a semi-parametric bootstrap method (using 1000 bootstrap datasets). LMEs were fit and p-values testing the null hypothesis that the fixed effect for slope is zero were evaluated using Satterthwaite’s degrees of freedom method using the lmerTest package in R (94). For a log₁₀-transformed y-value depending linearly on $\mathrm{x}$, as in figs. S34–36 the effect of $\mathrm{x}$ on $\mathrm{y}$ is given by $\mathrm{l}\mathrm{o}\mathrm{g}\left({\mathrm{y}}_2\right)-\mathrm{l}\mathrm{o}\mathrm{g}\left({\mathrm{y}}_1\right)=\left[{\mathrm{x}}_2-{\mathrm{x}}_1\right]\cdot \mathrm{s}$, where $\mathrm{s}$ is the slope. This is equivalent to ${\mathrm{y}}_2/{\mathrm{y}}_1={10}^{[\mathrm{x}2-\mathrm{x}1]\cdot \mathrm{s}}$. Thus, in figs. S34–S36 where $\mathrm{x}$ represents mutations and $\mathrm{y}$ represents log(K_D) or log(k_on) or log(k_off), an increase of $\mathrm{n}$ mutations going from ${\mathrm{x}}_1$ to ${\mathrm{x}}_2$ will lead to a change in $\mathrm{y}$ as: ${\mathrm{y}}_2/{\mathrm{y}}_1={10}^{\text{n}\cdot \text{s.}}$. For a log₁₀-transformed y-value depending linearly on a log₁₀-transformed x-value, as in Fig. 8B, the effect of $\mathrm{x}$ on $\mathrm{y}$ is given by $\mathrm{l}\mathrm{o}\mathrm{g}\left({\mathrm{y}}_2\right)-\mathrm{l}\mathrm{o}\mathrm{g}\left({\mathrm{y}}_1\right)=\left[\mathrm{l}\mathrm{o}\mathrm{g}\left({\mathrm{x}}_2\right)-\mathrm{l}\mathrm{o}\mathrm{g}\left({\mathrm{x}}_1\right)\right]\cdot \mathrm{s}$, where $\mathrm{s}$ is the slope. This is equivalent to ${\mathrm{y}}_2/{\mathrm{y}}_1={\left({\mathrm{x}}_2/{\mathrm{x}}_1\right)}^{\mathrm{s}}$. Thus, in Fig. 8B, where $\mathrm{x}$ represents the K_D for eOD-GT6 or a GT6 variant, and $\mathrm{y}$ represents K_D for eOD-GT8, an increase in the K_D for eOD-GT6 or a GT6 variant by a factor of α will lead to a change in the K_D for eOD-GT8 according to ${\mathrm{y}}_2/{\mathrm{y}}_1={\mathrm{\alpha }}^{\mathrm{s}}$.

Statistical analysis

Placebos in both groups were pooled into a single control group for all statistical analyses. Confidence intervals for frequencies and response rates were based on the Wilson score method (91). Spearman correlations were used to evaluate the association between immune responses. To assess if response rates differed by group, Barnard’s exact test was used. To assess if response rates differed over time within a group, McNemar’s test was used to account for paired data. Response magnitude comparisons between groups were compared using the Wilcoxon rank-sum test. Response magnitude comparisons between time points were performed using the Wilcoxon signed-rank test to account for paired data. Comparisons of percent mutation levels over time (Fig. 5, fig. S20, and tables S46–S49) were performed using the Wilcoxon signed-rank test to account for paired data. Regression analyses were performed as described in Regression analyses involving SPR-measured quantities. Statistical quantile analyses were carried out using R and Pandas (https://github.com/pandas-dev/pandas). The R language and tidyverse R packages were used for graphical and statistical analysis (95, 96).

Figure generation

Most figures were generated with either Matplotlib (97) or a custom port of the Seaborn package that incorporates Wilson confidence intervals into the statistical analysis [ (98); https://github.com/jwillis0720/seaborn-fork], and with Adobe Illustrator for final composition. Figures S4 and S17 were produced using the ggplot package in R. Figure generation can be found in the accompanying data analysis repository (github.com/SchiefLab/G001/figures). Figures S5, S6, S8, and S9 were produced using Intaglio (https://intaglio.en.softonic.com/mac). Fig. S10 was produced using BioRender (BioRender.com), and the publishing license can be found in the data repository.