DH270.6 bnAb association pathways identify vaccine design targets

The completed simulation set provided detailed structural information for different encounter states and their exchange with nearby states, leading to the known bnAb bound state conformation. An MSM of the association process prepared using the simulated set described this process as time-dependent, memoryless probabilistic jumps between states. States were defined kinetically based on time-dependent changes in distances between known binding site residue contacts using a time-lagged independent component analysis (TICA) transformation method. This transformation identified slowly relaxing degrees of freedom in the system, which could then be clustered for encounter state assignments. These encounter state assignments and their connectivity to other states could then be used to construct a Markov model. Thus, the antibody association pathways could be defined by the relative exchange between states and each state’s probability of moving toward the bound state, defined by the so-called committor distribution, which determines forward transition probabilities based on the MSM³² (Fig. 2A, Supplementary Figs. ⁴–⁶). The association pathway provided a map of transitions that can be used to identify target sites for modification of the association process and, therefore, the association rate, k_a.

Open in a separate window

Fig. 2

The association pathway for the DH270.6 Fv identifies design targets.

A Association pathway for DH270.6 determined from an MSM prepared from the simulated set. Important regions of the gp120_T (blue) include the V1 loop (purple), the V4 loop (light orange), the GDIK motif (gray), and the antibody HCDR3 (bright green). A single gp120 state is used throughout, with 50 representative samples of the antibody shown for each state (ensemble of line structures). Arrows show transition paths with sizes adjusted to indicate each transition’s relative flux. Structures are positioned based on the probability that each state will move forward to the bound state. Transitions targeted for design are highlighted (pink). B Graphical representation of lineage-based design depicting a simplified phylogenetic tree leading from the inferred UCA and intermediates to the mature DH270.6 bnAb. Design targets D1-D4, each targeting low probability mutations (red), with the I3.6 targeting D2 identified as the primary focus for design. The I3.6 intermediate contains two key improbable mutations, the V_H R98T and V_L L48Y. C (left) Representative state from State 1 highlighting the position of the N408-glycan and the contacts between the design target D2 mutation sites V_H T98 and V_L Y48. (middle) Representative state from State 3 highlighting the position of the N442-glycan relative to the N301-glycan and its bound state interactive Y27 residue sidechain. (right) The bound state conformation. D (left) Association rates determined by SPR for the parent CH848 Env SOSIP ectodomain and each design (right) Association kinetics for a CH848 parent and design in which two glycans in the V1 loop have been eliminated via mutation (referred to as DT). Arrows and values indicate the fold change in k_a between the parent and combined mutations. Three replicate measures are shown with the mean and error bars indicating the standard deviation. E (lower left) Simplified graphical representation of encounter transition in the DH270.6 interaction with gp120_T. (upper right) Shapes placed in their relative positions on the gp120_T (orange) map encounter states to the structure. Encounter state 1 maps to the cyan V4 block while Encounter state 2 and 3 map to the orange and yellow. Source data are provided as a ^{Source Data} file.

Our immunogen design targets were defined by functional improbable mutations in each inferred intermediate antibody along the DH270.6 maturation pathway that are critical for achieving neutralization breadth (Fig. 2B)³. Our previous efforts identified a design capable of selecting key mutations in the first intermediate, I5⁸. Specifically, we found that eliminating N-linked glycan sites at positions N133D and N138T in the CH848-d949 V1 loop (DT) enhanced DH270.UCA affinity and enhanced selection of a key heavy chain mutation, G57R, in DH270.UCA knockin mice⁸. We, therefore, focused here on two key mutations in the DH270 B cell lineage intermediate antibody, I3: heavy chain R98T and light chain L48Y (Fig. 2B).

The association path for the V3-glycan bnAb, DH270.6, proceeded through three encounter states, Encounter States 1 through 3, that reached the bound state predominantly through a pre-bound state, state 4 (Fig. 2A). Encounter State 1 involved coupling of the light chain to the gp120_T outer domain, adjacent to the V4 loop (Fig. 2C left, Supplementary Fig. ^7A, Supplementary Movie ²). With the HCDR3 loop oriented toward the N332-glycan, this state acted to present the V_H D115/V_L Y48 N332-glycan binding groove to the glycan (Fig. 2C, Supplementary Fig. ^7A). Next, encounter states 2 and 3 resulted from coupling the N332-glycan to the antibody with various degrees of rotation about the glycan base (Supplementary Fig. ⁷, Supplementary Movie ²). These states accessed the bound conformation through a flexible pre-bound state, Encounter State 4 (Fig. 2A, Supplementary Fig. ⁷). In the MSM, access to the bound state depended on the coupling of the N301-glycan with the light chain and movement of the V1 loop that we previously defined using cryo-EM structure determination (Fig. 2A, Supplementary Fig. ^{7B, C})⁸. These results showed that a series of encounter states that range from being distant to the target epitope to those that are closer to it enables access to the previously observed DH270.6 bound conformation^8,33.

We next designed Envs with faster DH270 lineage antibody association rates based on this pathway. A dominant pathway feature was the importance of early coupling of the N332-glycan D1 arm to the V_H D115/V_L Y48 groove (Fig. 2C left, Encounter State 1). The heavy chain R98T mutation eliminated a salt bridge with D115, allowing D115 to form closer interactions with the N332-glycan, and L48Y added additional hydrogen bonding contacts²⁵. We hypothesized that enhancing access to this state would increase the antibody association rate specifically through interactions with these two residues and would thus yield a strong positive association rate gradient. This would be accomplished by improving access of the V_H D115/V_L Y48 groove to the N332-glycan and by increasing the probability that the encounter state will reach the pre-bound state. Encounter State 1 stood out as a useful target due to its presentation of this groove to the N332-glycan. The N332-glycan acted as a tether for the antibody in the MSM, allowing the antibody to rotate into the bound state. Increasing the likelihood of N332-glycan capture early in the encounter process should, therefore, enhance the association rate.

Access to Encounter State 1 was hindered by an Env glycan at position N408 (Fig. 2C). We thus eliminated the N-linked glycosylation site via an N408K mutation in addition to improving Encounter State 1 heavy chain contacts via Env mutations T290K and E293T, here referred to as the DH270 Encounter design 1 (DE1). We measured binding of DE1 to the V_H R98T/V_L L48Y containing I3.6 intermediate and other clonal lineage antibodies using surface plasmon resonance (SPR; Supplementary Figs. ⁸–¹⁰, Supplementary Data ¹). We found that the N408K mutation resulted in a doubling of the antibody-Env association rate, consistent with the predicted association pathway (Fig. 2D, Supplementary Fig. ⁸, Supplementary Data ¹). Our MD simulation results showed that Encounter State 1 transitioned with the highest probability to the pre-bound state, Encounter State 4, through rotation about the N332-glycan, which occurred by visiting Encounter State 2. This transition was hindered by the close coupling of an Env glycan at position N442 to the antibody light chain (Fig. 2C middle). The N442-glycan interaction with the light chain mirrored that of the N301-glycan in the bound conformation but limited interactions of the antibody with the conserved GDIK motif in the V3 loop of the Env (Supplementary Fig. ^7D). Thus, eliminating the N442-glycan was predicted to increase the probability that the N332-glycan rotation states, Encounter States 2 and 3, would convert to the pre-bound state. Mutation of N442 to alanine, referred to here as DE2, indeed resulted in a 5-fold improvement in the association rate of I3.6 (Fig. 2D). The effect of N442-glycan removal was consistent with data showing that the presence of the N442-glycan correlated with early intermediate neutralization resistance and that the acquisition of I3.6 intermediate mutations reduced this resistance (Supplementary Fig. ¹¹). Additionally, mammalian display-based selection for higher affinity Envs also identified an N442-glycosylation site eliminating mutation, N442D (Supplementary Fig. ¹²). These results show that physical observations in the MSM describe phenomena that significantly impact autologous and heterologous virus neutralization and can direct design as effectively as mammalian display based affinity enhancement.

We next asked whether the MSM faithfully represented the transitions between encounter states. Two mutations are coupled in their effect on binding when the sum of the individual, single-mutant effects differs from the double-mutant effects relative to the unmutated construct. These double-mutant cycles³⁴ provide a powerful tool for probing transition intermediate states such as encounter states. The association pathway between the V4 proximal Encounter State 1 must transition through the N442-glycan hindered Encounter State 2 pathway to reach the bound state. Thus, the MSM indicated the N408K and N442A mutation effects on the association rate should be coupled. The sum of association rate improvements for the DE1 and DE2 designs was ~7-fold. However, when introduced together, the association rate improved by ~13-fold, nearly double that of the individual mutations. A double mutant cycle calculation based on transition state theory³⁵ for these two mutations yielded a G of −2.7kcal/mol, indicating the mutations were indeed coupled, thus confirming the MSM association transition pathway (Fig. 2D).

We next asked what effect the V1-glycan deleted DT design mutations had on the association pathway. The CH848-d949 DT design showed a ~3-fold enhanced association rate compared to the unmutated CH848-d949 SOSIP Env with a ~2.5-fold improved dissociation rate for the I3.6 intermediate (Supplementary Fig. ⁹, Supplementary Data ¹). The enhanced association rate suggested that the elimination of the V1 glycans opened new encounter state transitions. We hypothesized that, if the elimination of the V1-glycans in the DT design opened a new, alternative association path, the effect of the DE1 and DE2 mutations would be reduced and these mutations would not show coupling. We, therefore, introduced the DE1 and DE2 mutations, independently and together in the DT design, referred to as DT-DE1-3, respectively, and measured their association rates. The association rate improvements for the I3.6 intermediate were reduced compared to the non-DT design. The DT-DE1 and DT-DE2 constructs showed ~1.5 and 2.5-fold greater association rate, respectively, to I3.6 than the association rate for the DT construct. As predicted, these mutations showed little coupling when combined in the DT-DE3 design, with a transition state G of −0.4kcal/mol (Fig. 2D, Supplementary Data ¹). This loss in coupling confirmed that the elimination of the V1 glycans in CH848-d949 opened a new association pathway and showed that the DH270 lineage acquired the ability to recognize the HIV-1 Env through multiple encounter sites beginning at the I3.6 intermediate (Fig. 2E).

Key DH270 clonal lineage mutations selected in vivo by association pathway Env designs

BCR triggering strength is linearly correlated with association rates³⁶. Enhancing the association rate in immunogen design, therefore, might improve the selection of target bnAb residues. We next asked whether the CH848-d949 N408K/N442A design would yield such an association gradient by showing stronger binding to the I3.6 intermediate where the V_H R98T/V_L L48Y mutations occur, relative to the earlier I5.6 intermediate. We measured the affinity and kinetics of binding of the designed immunogen to the DH270 UCA, as well as to DH270 intermediates I5.6 through I2 and the mature DH270.6 antibody. We compared the binding kinetics of the CH848-d949 N408K/N442A containing DE3 design to those of the unmutated CH848-d949 (Supplementary Figs. ⁸–¹⁰, Supplementary Data ¹). The DE3 design showed a four-fold greater association rate for the I3.6 intermediate compared to the earlier I5.6 intermediate (Fig. 3A). The unmutated CH848-d949 trimer showed a twelve-fold association rate gradient between I5.6 and I3.6 intermediate antibodies, but substantially lower association rates than the DE3 design, at forty-five and thirteen-fold greater rates for the DE3 design, respectively (Fig. 3A, B, Supplementary Fig. ⁸). Thus, the DE3 design showed a more favorable affinity gradient for the selection of I3.6 mutations than the unmutated CH848-d949 trimer.

Open in a separate window

Fig. 3

DH270.6 encounter state design improves the selection of target mutations.

A Association rates for the parent CH8483 Env SOSIP (black) and CH848 DE3 design (red) for the DH270 lineage UCA, intermediates, and DH270.6. B Affinities for the parent CH8483 Env SOSIP (black) and CH848 DE3 design (red) for the DH270 lineage UCA, intermediates, and DH270.6. Log scaling is used for visual clarity. C DH270.UCA Ramos B cell BCR triggering was measured as a percentage of calcium flux relative to the maximum flux observed for an anti-IgM Fab₂. D cryo-EM map of the DE3 design bound to DH270.6 (gray) and a difference map between the parent CH848 bound to DH270.6 and the DE3 design bound to DH270.6 highlighting the N442-glycan missing density (green). Contour levels 1.64 and 0.57, respectively. E Panel of pseudo-virus IC50 neutralization titers for sera from IA4 knock-in mice immunized with either the glycan hole-filled CH848-d949 SOSIP trimer (N=12 mice in group) or the DE3 design (N=13 mice in group). Significance was calculated using a two-tailed, non-parametric Mann-Whitney test and is marked by an asterisk (*p<0.05; CH848.10.17, p=0.203; CH848.10.17 Holes Filled, p=0.320; CH848.10.17.N332T, p=0.220; CH848.10.17.DT, p=0.019; CH848.10.17.DT.N332T, p=0.964; Q23, p=0.011; JFRL, p=0.220; TRO.11, p=0.055; DU156.12, p>0.999; ZM106F, p=0.096; CAAN5342, p>0.999; 92RW20, p=0.264; 6101.1, p=0.024; MLV-SVA, p>0.999). F Mutation frequencies for the IA4 key improbable mutations heavy chain R98T and light chain L48Y. Bars indicate the geometric mean frequencies. Significance was calculated using a two-tailed, non-parametric Mann-Whitney test and is marked by an asterisk (*p<0.05, N=11 mice in the CH848-d949 immunized group and N=13 mice in the DE3 immunized group). Source data are provided as a ^{Source Data} file.

We then measured antigen-induced B cell calcium flux for each Env construct in a Ramos B cell line transfected with an IgM DH270 bnAb lineage UCA antibody compared to the calcium flux induced by binding of an anti-IgM Fab₂ antibody. Though the affinity for the DE1+DT design was slightly lower for DH270.UCA compared to the DT design, at 607 and 474 nanomolar, respectively, DE1+DT showed a ~5-fold enhanced maximum calcium flux over the DT (Fig. 3C, Supplementary Fig. ^8A, Supplementary Data ¹). These data were consistent with the doubling of the association rate for DE1+DT compared to DT despite the faster DE1+DT dissociation rate. Similarly, DE3+DT had an association rate ~3-fold faster than DT but with a ~3-fold faster dissociation rate, giving rise to a similar affinity to DT (Supplementary Fig. ^8A, Supplementary Data ¹). Thus, the DE3+DT design nevertheless showed over an order of magnitude higher calcium flux, indicating the enhanced DE1 and DE3 Env design association rates improved BCR triggering (Fig. 3C).

We next tested whether differences in association and dissociation rates could be related to unintended Env structural changes caused by the designed mutations. We obtained a cryo-EM structure of the CH848-d949 and DE3 SOSIP Env bound to DH270.6 at 3.6 Angstroms and compared it to a previously determined CH848-d949 Env bound to DH270.6³³, and found that the only discernable difference between the structures was the absence of density for the N442-glycan in the DE3 structure (Fig. 3D, Supplementary Fig. ¹³). We then compared the cryoEM structures of these Envs bound to the CD4 binding site VRC01 bnAb to examine the unliganded V3-glycan bnAb epitope structure. Differences in the structures were only visible at the N442-glycan position, where the DE3 design lacked density, and the Fab constant region, which showed a shifted elbow angle (Supplementary Fig. ¹⁴). Thus, the design-enhanced kinetics are attributable to the encounter states observed in the MSM.

We next asked whether the improved association and affinity gradients would enhance selection of key V_H R98T/V_L L48Y I3.6 mutations in vivo in DH270 bnAb intermediate antibody knock-in mice. We immunized fourteen IA4 intermediate knock-in mice four times with either the CH848-d949 10.17 construct containing mutations D230N, H289N, and P291S to fill holes in the glycan shield⁵ or the DE3 design (Supplementary Data ³). IA4 is a mutated version of I5.6 containing V_H S60Y and V_L V100I, neither of which is essential for neutralization breadth or potency^15,26. Mice immunized with the DE3 design displayed an elevated mean titer of CH848-d949 pseudovirus neutralization compared to mice immunized with unmutated CH848-d949 Env and showed greater neutralization breadth and potency for several heterologous HIV-1 viruses (Fig. 3E). Trends in elevated mean values for the DE3 design for total V_H and V_L mutations after vaccination were seen (Supplementary Fig. ^15A). Our primary immunogen design goal was to specifically select for the improbable V_H R98T and V_L L48Y mutations that occur in the I3.6 intermediate. Next-generation sequencing (NGS) of DH270 B cells in the mice immunized with DE3 induced significantly enhanced selection of the target V_H R98T and V_L L48Y mutations (Fig. 3F). To assess whether the DE3 design selected for other known improbable mutations, we examined the frequency of the I2 intermediate mutation G110Y. This mutation was not involved in the relevant DE3 encounter states and did not show significant differences between the groups. (Supplementary Fig. ^15B). These results indicated that the D3 immunogen precisely selected the targeted neutralization-critical R98T and L48Y mutations, as targeted by our association path-based design strategy. It was important to determine if both mutations could be selected together in the same antibody. From a mouse immunized with DE3 that showed considerable neutralization breadth (Fig. 3E), we identified a total of 46 DH270 lineage BCRs from a total of 59 isolated B cells (Supplementary Fig. ^15C). Of these, 10 had both V_H R98T and V_L L48Y mutations, and 9 contained R98T only (Supplementary Fig. ^15D). These results showed that the association pathway-based immunogen design selected for specific mutations targeted in the DH270 bnAb intermediate B cells by vaccination.

We next asked whether the DE3 would act as a more effective boost to our previously described CH848-d949 10.17 DT ferritin nanoparticle priming immunogen^5,8. This immunogen is known to effectively select for the key heterologous neutralization breadth enhancing I5.6 mutation V_H G57R. The mice were split into two groups with five mice each and immunized four times with the nanoparticle construct, followed by four immunizations with either the glycan holes filled CH848-d949 10.17 timer or the DE3 trimer (Supplementary Data ³). The groups’ median mutation frequencies for DH270 heavy and light chains were similar (Supplementary Fig. ¹⁶). Examination of key mutation frequencies did not reveal significant differences between the groups, including for the targeted V_H R98T and V_L L48Y mutations (Supplementary Fig. ¹⁶). Significant differences in autologous pseudo-virus serum neutralization titers were not observed between the mice in each study. Of five heterologous pseudo-viruses tested for serum neutralization, only one, the 6101.1 pseudo-virus, showed a significantly greater mean neutralization titer in the glycan holes filled CH848 d949 10.17 construct group (Supplementary Fig. ¹⁶). While not rising to the level of significance, the mean selection frequency of the I5.6 intermediate V_L S27Y was greater in the holes-filled CH848 d949 10.17 group, while the I2.6 intermediate V_H G110Y mutation occurred with a greater mean frequency in the DE3 group. The S27Y mutation, along with the I5.6 G57R mutation, is present in the IA4 knock-in mice, where DE3 immunization showed significantly enhanced selection of the R98T and L48Y mutations compared to the holes-filled CH848 d949 10.17 immunization. The G110Y mutation is not present in the IA4 mice and the mean selection of G110Y was similar in both IA4 groups. The absence of significant differences between the primed and boosted groups may be related to how affinities, and therefore selection, can differ in the presence of differing sets of mutations such as S27Y and G110Y. This would suggest that sequential immunization must consider such coupling to ensure that critical residues are selected in an order that does not hinder the selection of subsequent key mutations.

Association pathway-designed immunogens for bnAb immunogens for the CD4 binding site

To determine the ability of association pathway immunogen design to be effective for targeting Env bnAb sites other than the V3-glycan epitope, we designed boosting immunogens for a CD4 binding site (CD4bs) bnAb B cell lineage. The most potent CD4bs bnAbs attain their neutralization breadth by mimicking CD4 interactions with Env. In contrast to V3-glycan bnAbs, CD4-mimicking bnAbs, are genetically restricted to two VH genes, VH1-2*02 or VH1-46^20,37,38. Both show structural mimicry of the CD4 receptor interaction with the Env trimer^20,39. Antibodies of the VH1-2*02-class use a rare five amino acid Light Chain Complementarity Determining Region 3 (LCDR3) and contain either deletions or multiple glycine mutations in the LCDR1 to accommodate the proximal gp120 loop D and N276 glycan³⁹. The VH1-46-class of bnAbs is an attractive target for HIV-1 vaccine design because neither deletions in the LCDR1 nor atypically short LCDR3s are required for their neutralization breadth, and a high-affinity germline targeting Env has been designed for the UCA of the VH1-46 prototype B cell lineage, CH235⁴⁰. This immunogen selected for a key improbable mutation, K19T, in a CH235 UCA knock-in mouse model⁸, and elicited CD4 mimicking serum-neutralizing CD4bs antibody precursors derived from rhesus VH genes orthologous to human VH1-46⁴¹. Here, we focused on the first two intermediates, the early I60 and the next intermediate antibody, I59, and two later intermediate antibodies that displayed substantial gains in neutralization breadth, I39 and I35 (Fig. 4A) ^20,41.

Open in a separate window

Fig. 4

The association pathway for the CH235.12 Fv identifies design targets.

A Graphical representation of lineage-based design depicting a simplified phylogenetic tree leading from the inferred UCA and intermediates to the mature CH235.12 bnAb. Design targets D1-D4, each targeting low probability mutations (red), that are key for neutralization breadth. B Association pathway for CH235.12 determined from an MSM prepared from the simulated set. Important regions of the gp120_T (blue) include the CD4bs loop (cyan), the V5 loop (red), the loop D (magenta), and the antibody HCDR3 (bright green). A single gp120 state is used throughout, with 50 representative samples of the antibody shown for each state (ensemble of line structures). Arrows show transition paths with sizes adjusted to indicate the relative flux through each transition. Structures are positioned based on the probability that each state will move forward to the bound state along the x-axis. Transitions targeted for design are highlighted (pink). C Representative state from encounter State 2 highlighting the position of the N276-glycan and contact of the HCDR3 with loop D. D Representative state from encounter State 3 highlighting the position of the N386-glycan and contact of the HCDR2 proximal sheet residues with the CD4bs loop. E Mature CH235.12 association rates for the CH505.M5.G458Y parent and the CH505.M5.G458Y CE0 and CE1 designs. Error bars are standard deviations from the mean of three replicate measures. F Aligned structures of the CH505.M5 trimer bound to the early intermediate CH235.I60 Fab or mature bnAb CH235.12 Fab bound to either CH505.M5 or CH505.M5.G458Y highlighting the difference in the V5 loop. The Y458 sidechain is shown for reference. G CH235 UCA, intermediate I60, and mature CH235.12 Fab affinities for the CH505.M5.G458Y parent and the CH505.M5.G458Y CE2 design. Error bars are standard deviations from the mean of three replicate measures. Log scaling is used for visual clarity. Source data are provided as a ^{Source Data} file.

The MD system was prepared similarly to the CH848-DH270.6 system with a truncated CH505TF gp120 and a CH235.12 Fv positioned unbound near the CD4bs and with the antibody HCDR3 oriented toward a gp120_T (Supplementary Fig. ^{17A, B}). A total of 27 adaptive iterations were conducted, yielding 6597 simulations for a total accumulated simulation time of ~1.6ms, including 34 transitions to a near-bound state. A common transition path involved coupling of the Env CD4-binding site loop with a β-sheet region adjacent to the antibody Heavy Chain Complementarity Determining Region 2 (HCDR2), followed by a rotation of the antibody that brought the light chain close to its final point of contact with the Env loop D (Supplementary Movies ³ and ⁴), (Supplementary Fig. ^17C–E). Unlike DH270.6, the fully matured CH235.12 antibody did not transition beyond a near-bound state, here defined as a bound state RMSD below 4Å but without maintaining specific V_L and HCDR3 contacts (Supplementary Fig. ^17D). This appeared to be due to steric hindrance caused by flexible epitope loops that prevented the formation of specific bound state contacts. This near-bound state nevertheless likely represents the final stage of the association process where these contacts are formed. We, therefore, used this state as the model endpoint to interrogate more distant encounter states. The association pathway determined from an MSM of these simulations revealed three encounter intermediate states, States 1-3, that all led to the near-bound state (Fig. 4B, Supplementary Fig. ¹⁸–²⁰). Like the DH270.6 pre-bound state, the overall near-bound conformation was similar to the bound conformation but displayed considerable flexibility (Fig. 4B). The dominant pathway led from diverse structural states surrounding the CD4bs, with the HCDR3 orientated toward the g120_T (Fig. 4B). This state was connected to two distinct encounter states: one in which the HCDR3 interacted with V5 and loop D (Fig. 4C) and the other state was with an HCDR2-adjacent V_H contact with the CD4bs loop (Fig. 4D). Each encounter state had direct access to the near-bound state (Fig. 4B).

In the context of the Env trimer, several conformations in Encounter States 1 and 2 would be partially occluded by the adjacent gp120 protomers, whereas Encounter State 3 is predicted to not be occluded by adjacent protomers (Supplementary Fig. ²¹). We, therefore, selected Encounter State 3 for further design consideration since our vaccine Env immunogen platform is a trimer. Encounter State 3 is stabilized by polar contacts between CH235.12 V_H R56 and N54 sidechains and mainchain gp120 residues 364 to 366 of the CD4bs as well as residues 469 to 471. These contacts are similar to those found in a crystal structure of CH235.12 bound to the clade A/E 93TH057 Env gp120²⁰ and act to anchor the antibody to the Env. These residues are at the periphery of the paratope, creating a hinge-like anchor about which the antibody can rotate into the bound configuration (Supplementary Movie ³). These residues are G54 and S56 in the CH235.UCA and would provide limited stabilization of this encounter. Consistent with an important role for the changes observed in the mature CH235.12, we have shown mutations acquired at these residues enhance Env binding breadth when added to the CH235.UCA²⁰. Their appearance in an early encounter state here suggests they play important roles in both reaching and maintaining the bound state and are, therefore, important targets for design.

As in the DH270.6 pathway design, we focused on opening this state to collision and enhancing passage from this state to the near-bound state. Two glycans distant from the epitope limited access to the State 3 conformation—one was positioned on the V1/V2 domain at position N197, and the other on the outer domain at N386. Deletion of the N197-glycan was previously shown to enhance virus neutralization by CD4bs targeting bnAbs⁴². We, therefore first examined an N-linked glycosylation site deletion at N197 (N197D)⁴¹. The results showed a nearly 6-fold association rate enhancement for the CH235 UCA and a nearly ~9-fold reduced association rate for the first intermediate, I60 (Supplementary Figs. ²²–²⁴ and ^25A, Supplementary Table ²). However, the following intermediate, I59, and later intermediates, I39 and I35, did not show association rate differences (Supplementary Figs. ²²–²⁴ and ^27A, Supplementary Table ²). The I39 and I35 intermediates showed reduced dissociation rates at 2- and 3-fold, respectively (Supplementary Figs. ²²–²⁴ and ^25A, Supplementary Table ²). The mature CH235.12 bnAb showed a ~4-fold enhanced association rate and no change in the dissociation rate (Supplementary Figs. ²²–²⁴ and ^25A, Supplementary Table ²). The enhanced association rate for CH235.12 was consistent with the CH235.12 association pathway, which was hindered by the presence of the N197-glycan. The substantial differential in the N197D association and dissociation rates between the UCA, intermediates, and mature antibodies suggest a complex interplay between the effects of differing antibody mutations as the CH235 VH1-46 CD4bs B cell lineage develops.

Examination of a representative set of Encounter State 3 structures suggested that replacing the N386-glycan with an arginine could aid in attracting this state through interaction with CH235.12 F_v aspartate and glutamate residues (Supplementary Fig. ²⁶). Consistent with Encounter State 3 structural observations, an N386A mutation, referred to here as CH235 Encounter design 0 (CE0), enhanced the CH235.12 Fab association rate by ~2 fold and a stabilized form of an N386R mutation, referred to here as CE1, which contains the F14 and 2P stabilization mutations^43,44, enhanced the CH235.12 Fab by ~6 fold (Fig. 4E, Supplementary Figs. ²²–²⁴ and ^25B, Supplementary Table ²). The non-stabilized CE1 design showed only ~2-fold enhanced association rate, suggesting stabilization affected the rate (Supplementary Table ²). We, therefore, measured affinities for the F14+2P stabilized, unmutated CH505.M5.G458Y trimer. No changes in CH235.12 association rate were observed for this construct, indicating the stabilization alone did not yield the CE1 association rate. These results show that the N386R and stabilization mutations were coupled in their effect on the CH235.12 association rate (Supplementary Table ²). Substantial impacts on the association rate for the N386A mutation were not observed for the CH235 UCA or intermediates I60 and I59. Modest enhancements began at I39 and I35, suggesting mutations acquired late in CH235 maturation permitted direct use of the N386-glycan occluded encounter site (Supplementary Figs. ²²–²⁴ and ^25B, Supplementary Table ²). Dissociation rates were largely unaffected by the N386 PNGS removal in the UCA, I60, I59, and I39 but showed a two- to four-fold dissociation rate increase, eliminating affinity gains (Supplementary Figs. ²²–²⁴ and ^25B, Supplementary Table ²). The N386-glycan site is adjacent to the CH235 antibody binding CD4bs loop interaction interface. As glycans are known to stabilize the protein fold^45,46, eliminating this glycan may increase CD4bs loop dynamics and thereby impact bound state stability. The increase in the association rates for CH235.12 with N197 and N386 PNGS substitutions is consistent with the simulation-based association pathway that shows these glycans block encounter formation. However, the variability in PNGS removal impacts on the antibody lineage member kinetic rates suggests that these pathways may change substantially as the lineage develops.

We next examined gp120 dynamics in Encounter State 3 and the near-bound state (Supplementary Movie ⁴). The CH235.12 antibody contacts the gp120 CD4bs loop, the V5 loop, and loop D in the bound state. Of these, the most conformationally variable loop was V5 (Supplementary Fig. ^26C). To better understand the impacts of this flexibility on antibody binding, we determined cryo-EM structures of our previous CH505 design that contained an N279K mutation (M5) in loop D bound to the early CH235 intermediate I60 and the mature CH235.12 bnAb in addition to a CH505.M5 design containing a G458Y (CH505.M5.G458Y) mutation in V5 bound to CH235.12 (Supplementary Fig. ²⁷). Examination of the V5 loop backbone trace in these structures revealed marked differences in its conformation (Fig. 4F). In the I60 bound CH505.M5, the V5 loop between residues 458 and 461 is positioned distant from the I60 light chain, showing limited interaction with the heavy chain. In the CH235.12 bound CH505.M5 structure, residues G458 and G459 rest in a groove between the V_H and V_L, packing against the V_L W94 sidechain. This is permitted in the CH235.12 bound structure due to the I39 V_H W47L mutation, which allows the W94 rotamer to shift by ~90°, creating the groove for V5 interaction. The introduction of I39 W47L mutation along with I59 S56R in the CH235 UCA and post I39 CH235 G54W was previously shown to increase recognition from three to five of fifteen heterologous HIV-1 Envs from multiple clades²⁰, suggesting this shift in V5 interaction plays an important role in neutralization breadth. The CH235.12 bound CH505.M5.G458Y V5 loop is positioned intermediate to these two V5 positions, with the aromatic tyrosine sidechain position proximal to the V_H/V_L groove. These results suggest later intermediates can effectively manage V5 conformational variability.

We then asked whether limiting V5 loop flexibility in a CH505.M5.G458Y construct via a disulfide staple at the base of the V5 loop between residues 457 and 465, referred to here as CE2, could enhance early CH235 clonal lineage antibody affinities. A cryo-EM-derived structure of the CE2 design confirmed disulfide bond formation (Supplementary Fig. ²⁸). The V5-loop between residues 458 and 464 was poorly resolved in the cryo-EM map, suggesting these residues are conformationally dynamic.

We next compared CH235 clonal lineage kinetics between the parent CH505.M5.G458Y construct and the CE2 design. The UCA showed a ~14-fold improvement in the dissociation rate with the CE2 design, with no change in the association rate. (Fig. 4G, Supplementary Figs. ²²–²⁴ and ^25C, Supplementary Table ²). The early I60 intermediate showed a ~2-fold enhanced association rate with a ~20-fold improvement in the dissociation rate (Fig. 4G, Supplementary Figs. ²²–²⁴ and ^25C, Supplementary Table ²). While these early-stage CH235 antibodies showed substantial dissociation rate improvements in the CE2 construct, the I59, I39, I35, and mature CH235.12 antibodies did not show dissociation rate improvements, with ~2-fold association rate improvements for each (Fig. 4G, Supplementary Figs. ²²–²⁴ and ^25C, Supplementary Table ²). The elimination of these dissociation rate gains beginning at the I59 intermediate suggests mutations occurring at I59 allow the antibody to manage V5 loop flexibility effectively. These include V_H S56R and D95N and V_L S30R. The V5-loop varies significantly between differing variants and clades, acquiring mutations, insertions, and deletions. While the I59 mutations outweighed the effect of the V5-staple before the acquisition of W47L this later mutation may nevertheless be essential for Env recognition breadth. Alternatively, the D457C mutation may affect the extensive, water-mediated interactions that couple the antibody R56 residue with the Env D457²⁰, thus limiting affinity gains introduced by the V5-staple. These results show that antibody mutation effects can couple with Env residues positively and negatively and that the acquisition of certain antibody mutations can influence affinity gains or losses for additional mutations.

Recombinant HIV-1 envelope SOSIPs, antibodies, and Fab production

The DS stabilized⁶⁴ CH848.10.17.d949 DT and DE1-3 envelope SOSIPs and the 4.1⁶⁵, 2P⁴⁴, and F14 stabilized⁴³ CH505.M5.G458Y, CE0-3 SOSIPs were expressed in Freestyle^TM 293-F cells (ThermoFisher Cat No. R79007). The modified CH505 transmitted founder Env SOSIPs were chimeras containing the CH505 gp120 and BG505 gp41 domains. To express a sufficient amount of protein for immunization, the K279 construct required additional stabilization mutations, F14 and 2P. Before transfection, cells were diluted in Freestyle^TM 293 Expression Medium (Cat No. 12338018) to 1.25×10⁶cells/mL at a volume of 950mL. Plasmid DNA expressing the envelope SOSIP and furin were co-transfected at a 4:1 ratio (650μg and 150μg per transfection liter, respectively) and incubated with 293fectin^TM transfection reagent (ThermoFisher Cat No. 12347019) in Opti-MEM I Reduced Serum Medium (ThermoFisher Cat No. 31985062) to allow for complex formation. The diluted mixture was added to the cell culture which was incubated at 37°C, 9% CO₂ on a shaker at 120rpm for 6 days. On day 6 the cell supernatant was harvested by centrifuging the cell culture at 4000×g for 45min. The supernatant was filtered with a 0.45μm PES filter and concentrated to ~100mL using a Vivaflow® 200 cross-flow cassette (Sartorius Cat No. VF20P2).

Envelope SOSIPs were purified using a PGT145 or PGT151 affinity chromatography equilibrated in 15mM HEPES, and 150mM NaCl (pH=7.1). A PGT145 and PGT151 Ab affinity column was made by coupling mAbs to CNBr-activated Sepharose 4B (Cat No. 170430-01, GE Bio-Sciences) and packed into a Tricorn column (GE Healthcare). The supernatant was applied over the column at 2mL/min using an AKTA go chromatography system (Cytiva) followed by three-column volume wash steps with 15mM HEPES, and 150mM NaCl. Protein was eluted off the column using 3M MgCl₂ and diluted in 15mM HEPES, and 150mM NaCl buffer. The protein sample was buffer exchanged into 15mM HEPES, and 150mM NaCl by ultrafiltration using a 100kDa MWCO Amicon® Ultra-15 Centrifugal Filter Unit (Millipore Aldrich Cat No. UFC9010) and concentrated to <0.5mL for size exclusion chromatography. Size exclusion chromatography was performed using a Superose 6 10/300GL Column (Cytiva) on an AKTA go system in 15mM HEPES, and 150mM NaCl. Fractions containing trimeric SOSIP were collected. Lots produced were subjected to quality control including analytical SEC, SDS-PAGE, thermal shift analysis, biolayer interferometry, and negative stain electron microscopy to assure the presence of well-folded Env trimers.

Recombinant antibody and antibody Fab production

Antibodies were expressed in Expi293FTM cells in the Expi293TM Expression System (ThermoFisher Cat No. A1435101). Cells were diluted to 1.25×10⁶cells/mL at a volume of 100mL. Heavy and light chain plasmid DNA, at a ratio of 1:1 (50µg each per 100mL transfection), were mixed in Opti-MEM I Reduced Serum Medium (ThermoFisher Cat No. 31985062) and incubated with Expifectamine 293 transfection reagent (ThermoFisher Cat. No. A14525) to create the DNA-reagent complex. The mixture was added to the prepared cell culture which was transferred to be incubated at 37°C, 8% CO₂ on a shaker at 120rpm for 6 days. On day 6, the cell supernatant was harvested by centrifuging the cell culture at 4500rpm for 45min. The supernatant was filtered with a 0.45μm PES filter.

Each antibody was purified with a Cytiva HiTrap Protein A HP antibody purification column (Cytiva Cat. No. 29048576) equilibrated in 1X PBS. The harvested supernatant was applied over the equilibrated column at a flow rate of 0.5mL/min using an AKTA start chromatography system (Cytiva), followed by 3 column volume wash steps of 1X PBS. The antibodies were then eluted off with Pierce IgG Elution buffer, pH 2.8 (ThermoFisher Cat. No. 21004). The sample buffer was neutralized with Tris-pH 9 at 100µL/mL (ThermoFisher Cat. No. J60707). Lots produced were subjected to quality control including SDS-PAGE and thermal shift analysis.

Antibody Fab was prepared using Lys-C digestion. A digestion mixture of Lys-C enzyme (ThermoFisher Cat. No. 90307) and IgG (1:2000 ratio) was combined in 600µL 1X PBS for each sample. Each sample was incubated at 37°C for 2h. Following the incubation, the digested Mab was removed from the incubator and a proteinase inhibitor is added to the solution to stop the digestion process. The sample was then purified with a Cytiva HiTrap Protein A HP antibody purification column (Cytiva Cat. No. 29048576), collecting the Fab fragments that flow through. The Fab fragments are then buffer exchanged into 15mM HEPES, and 150mM NaCl. Each Fab lot produced was subjected to quality control by SDS-PAGE and thermal shift analysis.

Thermal shift assay

Thermal shift assay was performed using Tycho NT.6 (NanoTemper Technologies). Envelope ectodomain SOSIPs were diluted (0.15mgml⁻¹) in 15mM HEPES buffer with 150mM NaCl at pH 7.1. Intrinsic fluorescence was recorded at 330nm and 350nm while heating the sample from 35 to 95°C at a rate of 3°Cmin⁻¹. The ratio of fluorescence (350/330nm) and the inflection temperatures (Ti) were calculated by Tycho NT.6.

Molecular dynamics

Encounter state lifetimes can range from several to hundreds of microseconds. MD simulations at these timescales are accessible only by specialized computing hardware that is not widely available. This was accomplished by building MSMs from the aggregated simulations to identify movements toward transition states. The states are used in the next iteration. Since simulations rarely access transition states and typically return to metastable states when they approach transition states, launching simulations from near-transition states increases the likelihood of moving to and beyond a true transition state.

The system size determines the calculation rate for atomistic MD simulations. As the number of atoms in the simulation increases, the computing time needed to complete a simulation increases geometrically. Therefore, it is essential to prepare the smallest, physically realistic system possible to achieve sufficiently robust information that is of biological relevance. We prepared systems of N- and C-terminal truncated gp120 domains extracted from Env SOSIP closed-state structures (PDB IDs 6UM6 and 6UDA for CH848.10.17.d949 and CH505TF, respectively). Antibody Fv structures were prepared from crystal structures (PDB IDs 5TQA and 5F96 for DH270.6 and CH235.12, respectively). Missing loops were added using MODELER⁶⁶, Unbound antibody Fvs were positioned near the relevant epitope in PyMol⁶⁷. The Fv-gp120_T system was glycosylated using the CHARMM-GUI Glycan Modeler^68–73 based Glycosite-predicted⁷⁴ glycosylation sites with either mannose-9 or mannose-5 glycans, for the CH848 and CH505 systems, respectively. The glycosylated system was immersed in a 150Å octahedral water box, neutralized, and brought to a NaCl concentration of 150mM. The Amber22 software package with pmemd CUDA^75–77 using the CHARMM36 force field^78–81 and TIP3P water model⁸² was used for simulations. Electrostatic interactions were calculated with the Particle Mesh Ewald method⁸³ with a cutoff of 12Å and a switching distance starting at 10Å. The SHAKE algorithm⁸⁴ with hydrogen mass repartitioning⁸⁵ was used to constrain hydrogen atoms and allow for a 4fs timestep. To equilibrate the systems, 10,000 steps of energy minimization were performed with 1000 steps of steepest descent minimization with 500kcal/mol•Å restraints on protein followed by an unrestrained energy minimization consisting of 10,000 steps and 1000 steepest-descent steps. Minimization was followed by 20ps of NVT heating with 10kcal/mol•Å restraints applied to protein atoms and a subsequent 5ns NPT equilibration with no restraints, with a constant temperature of 298.15K maintained using a Langevin thermostat⁸⁶, and pressure of 1atm was maintained with isotropic position scaling. A total of 200–400 250ns unrestrained simulations in the NVT ensemble were performed to equilibrate the system and generate a wide ensemble of antibody orientations relative to the Env gp120. The cpptraj tool in AmberTools21⁸⁷ was used for trajectory unwrapping.

Adaptive sampling

The last frame from each of the 200–400 250ns was extracted for use in the first adaptive sampling routine epoch. Adaptive sampling was performed using the High-Throughput Molecular Dynamics (HTMD v. 1.24.2) package⁷⁷. Each subsequent epoch consisted of ~200–400 independent simulations of 250ns. At the completion of each epoch, simulations from each iteration were first projected using a distance-based metric between gp120 and Fv amino acid residues gp120 glycan atoms. This was followed by a TICA⁷⁸ projection using a lag time of 5ns, retaining five dimensions. MSMs were then built using a lag time of 50ns for the selection of new states for the next iteration. A total of 17 adaptive epochs were performed for the DH270.6-CH848-d949 and CH235.12-CH505 systems, yielding total simulation times of ~2ms and ~1.1ms for the CH848-d949 system and CH235.12-CH505 systems, respectively. Simulations were visualized in VMD and PyMol and selected using tools in the HTMD software.

Markov state modeling

MSMs were prepared in HTMD using inverse exponential distances between amino acid residues and glycan units identified as contacts in relevant antibody-bound Env subunits (PDB IDs 6UM6 and 6UDA for DH270.6-CH848-d949 and CH235.12-CH505, respectively: HXB2 residues 135, 136, 137, 138, 139, 148, 149, 299, 300, 301, 321A, 322, 323, 324, 325, 326, 327, 328, 329, 330, 332, 415, and 417 and the N138, N156, N301, N332, and N442 glycans for CH848, DH270.6 V_H residues 33, 52, 55, 57, 59, 84, 101, 103, 104, 105, 106, 107, 108, 109, 110, 111, 113, and 115, and DH270.6 V_L residues 27, 32, 48, 51, 52, 57, 58, 93, 96, and 97 and HXB2 residues 280, 282, 365, 368, 429, 458, and 470 for CH505 TF, CH235.12 V_H residues 30, 55, 57, 62, 65, 72, 102, 104, 105, and 106 and CH235.12 V_L residues 92 and 94). MSMs were prepared in HTMD using a TICA lag time of 25ns retaining five dimensions followed by K-means clustering using 500 cluster centers. The implied time scales (ITS) plots were used to select a lag time of 100ns for MSM building. Models were coarse-grained via Perron cluster analysis (PCCA++) using 5–6 states and validated using the Chapman-Kolmogorov test. State statistics were collected for mean first passage times, stationary distributions, and RMSDs relative to known bound state structures for each respective system. The RMSD and contact metric means were model-weighted. Weighted state ensembles containing 50–250 structures were collected for visualization in VMD.

Cryo-electron microscopy sample preparation, data collection, and processing

Purified Envelope SOSIP ectodomain preparations were prepared at concentrations of 4–5mg/ml in 15mM HEPES buffer with 150mM NaCl at pH 7.1 and mixed with Fab at a 1:5 molar ratio. A total of 2.5µl of the complex was deposited on a Quantifoil 1.2/1.3 grid that had been glow-discharged for 15s in a PELCO easiGlow glow discharge cleaning system. After 30-s incubation in >95% humidity, the excess protein was blotted away for 2.5s before being plunge-frozen in liquid ethane using a Leica EM GP2 plunge freezer (Leica Microsystems). Frozen grids were imaged in a Titan Krios (Thermo Fisher) equipped with a K3 detector (Gatan). Individual frames were aligned, dose-weighted, and CTF corrected, followed by particle picking, 2D classification, ab initio model generation, heterogeneous refinements, homogeneous 3D refinements, and local resolution calculations in cryoSPARC⁸⁸ (Supplementary Tables ³, ⁴, and ⁵).

Cryo-electron microscopy structure fitting and analysis

The antibody-bound SOSIP structures for DH270.6 and CH235 lineage antibodies (PDB IDs 6UM6 and 6UDA, respectively) structures were used to fit the cryo-EM maps in ChimeraX (114, 115). Mutations were made in PyMol or Coot⁸⁹. Coordinates were then fitted using Isolde (116) followed by iterative refinement using Phenix⁹⁰ real-space refinement. Structure and map analyses were performed using PyMol⁶⁷, Chimera⁹¹, and ChimeraX⁹².

Surface plasmon resonance

The SPR binding curves of the DH270 and CH235 lineage Fabs against CH848 SOSIPs and CH505M5 SOSIPs, respectively, were obtained using either a Biacore S200 or T200 instrument (Cytiva) in HBS-EP+ 1X running buffer (Cytiva). For the DH270 lineage Fabs (UCA3, I5.6, I3.6, I2.6 and DH270.6), PGT151 was immobilized onto a CM5 chip to ~10,000RU. The CH848 SOSIPs were injected at 5µl/min and captured to a level of 100–800RU with an average capture level of 450RU. Using the single cycle injection type, five sequential injections of the Fabs diluted from 25 to 1500nM were injected over the captured CH848 SOSIP Trimers at 50µL/min for 120s per concentration followed by a dissociation period of 600s. The Fabs were regenerated with a 12s pulse of Glycine pH2.0 at 50µL/min. Fab binding to a surface with negative control Flu antibody, Ab82, immobilized to a similar level as PGT151 as well as buffer binding were used for double reference subtraction to account for non-specific antibody binding and signal drift. Curve fitting analyses were performed using the 1:1 Langmuir model with a local Rmax. One interaction (DE1 with DH270UCA3) used the heterogeneous ligand model with the faster kinetics reported. The reported binding curves and kinetics for all DH270 Fabs are representative of at least one data set. For the CH235 lineage Fabs (UCA, I60, I59, I39, I35 and CH235.12), PGT151 was immobilized onto a CM5 chip to ~5000–10,000RU. The CH505M5 SOSIPs were injected at 5µl/min and captured to a level of 100–800RU with an average capture level of 400RU. Using the single cycle injection type, five sequential injections of the Fabs diluted from 31.25–3000nM were injected over the captured CH505M5 SOSIP Trimers at 30µL/min for 120s per concentration followed by a dissociation period of 600s. The Fabs were regenerated with a 12s pulse of Glycine pH2.0 at 50µL/min. Fab binding to a surface with the negative control Flu antibody, Ab82, immobilized to a similar level as PGT151 as well as buffer binding were used for double reference subtraction to account for non-specific antibody binding and signal drift. Curve fitting analyses were performed using the 1:1 Langmuir model with a local Rmax. The reported binding curves and kinetics for all CH235 Fabs are representative of at least three data sets (Supplementary Data ²).

B cell activation and calcium flux measurements with Ramos cells

Calcium flux experiments were performed using the FlexStation 3 Microplate Reader (Molecular Devices) and Softmax Pro v7 software (Molecular Devices) in conjunction with the FLIPR Calcium 6 dye kit (Molecular Devices). On the day of the experiments, a cell count for the Ramos cells was performed with a Guava Muse Cell Analyzer (Luminex) to ensure cell viability was greater than 95% and to calculate the volume of cells needed for a concentration of 1×10⁶cells/mL. The appropriate volume of cells was then pelleted at 1500rpm for 5min after which the supernatant was decanted and the cells were resuspended at a 2:1 ratio of RPMI media (Gibco)+FLIPR Calcium 6 dye (Molecular Devices). The cells were plated in a clear, U-bottom 96-well tissue culture plate (Costar) and incubated at 37°, 5% CO₂ for 2h. Antigens were separately diluted down to a concentration of 500nM in 50uL of the 2:1 ratio of RPMI media (Gibco)+FLIPR Calcium 6 dye (Molecular Devices) and plated in a black, clear bottom 96-well plate. The final concentration of antigen would be 250nM based on the additional 50uL of cells added during the assay. A positive control stimulant, anti-human IgM F(ab’)₂ (Jackson ImmunoResearch) was also included in the antigen plate. Using the FlexStation 3 multi-mode microplate reader (Molecular Devices), 50µL of the cells were added to 50µL of protein or Anti-human IgM F(ab′)₂ diluted in RPMI/dye and continuously read for 5min. Calcium flux results were analyzed using Microsoft Excel and GraphPad Prism v9. The relative fluorescence of a blank well containing only the RPMI/dye mixture was used for background subtraction. Once subtracted, the antigen fluorescence was then normalized with respect to the maximum signal of the IgM control, and calcium flux values were presented as a percentage (% of Anti-hu IgM Fab2 Max). Calcium flux data are representative of at least 2 measurements for the DH270UCA3 cell line.

Generation of the DH270 IA4 mice

The DH270 IA4 mice were generated by integrating pre-rearranged V(D)J exons of DH270 IA4 antibody heavy chain (HC) and light chain (LC) (Supplementary Table ¹) into the mouse immunoglobulin heavy chain (IgH) J_H and kappa light chain (Igk) J_k loci respectively. The integration was performed in mouse embryonic stem (ES) cells. In the integration construct, the DH270 IA4 HC or LC V(D)J exons were preceded by mouse V_H promoter (for HC) or Vk promoter (for LC), which drive the expression of V(D)J exons in B cells. The expression cassette, which consists of the V promoter and V(D)J exon, were flanked by DNA fragments that were derived from the mouse J_H or Jk regions. Upon transfection of the construct into ES cells, homologous recombination mediates the integration of the DH270 IA4 V(D)J exons into the J_H or Jk loci. At the integration site, the DH270IA4 HC expression cassette replaces the whole mouse J_H region, from 877bp upstream of J_H1 to 237bp downstream of J_H4; the DH270 IA4 LC expression cassette replaces the entire mouse Jk region, from 114bp upstream of Jk1 to 286bp downstream of Jk4. ES clones with correct integration events were identified by Southern blotting. The neomycin resistance gene, which was used to select for stable integration of the construct, was subsequently removed by cre recombinases-mediated excision through flanking loxP sites; this step is necessary to prevent potential interference of local transcriptional regulation by the drug selection marker. The final ES clone was injected into Rag2 deficient blastocysts to generate chimeric mice. Because Rag2 is required for V(D)J recombination, all B and T cells in the chimeric mice are derived from injected ES cells, and the chimeric mice can therefore be used for analysis of lymphocyte function. The chimeric mice were subsequently bred with C57BL/6 mice for germline transmission. Since the original ES cell was derived from an F1 mouse that resulted from a cross between 129Sv and C57BL/6 mice, the mouse line is of mixed 129Sv and C57BL/6 genetic background. To enumerate DH270.IA4-bearing B cells in DH270.IA4 het/het mice, we labeled splenocytes from knock-in mice with B cell tetramers comprising CH848DT SOSIP trimers and a panel of mAbs to identify various B cell compartments. For comparisons, we also labeled splenocytes from C57BL/6 and DH270.UCA3 het/het⁸. DH270.IA4 knock-in mice exhibited similar signs of impaired B-cell differentiation as DH270.UCA3 mice, including decreased numbers of transitional and follicular B cells as compared to BL/6 mice, although the reductions in cell numbers in DH270.IA4 mice were less dramatic (Supplementary Fig. ^{32A, B}). Notably, marginal zone and B1 B cell numbers were comparable across the three mouse strains. B cells expressing knock-in BCRs were readily identified by co-labeling with CH848DT B cell tetramers (Supplementary Fig. ^32C), with DH270.UCA3-bearing B cells representing ~6% of total B cells and DH270.IA4-bearing B cells representing ~20% of total B cells (Supplementary Fig. ^32D). In both DH270.UCA3 and DH270.IA4 knock-in mice, the majority of CH848DT-labeled B cells exhibited the phenotype of transitional B cells (Supplementary Fig. ^32D), expressing CD93 and IgM, but lacking expression of IgD and CD23 (Supplementary Fig. ^32E), consistent with previous observations of impaired B cell development⁸. Preparation of mouse cells for flow cytometry: Spleens from C57BL/6, DH270.UCA3 het/het, and DH270.IA4 het/het mice were processed into single-cell suspensions and treated with ACK lysis buffer to remove red blood cells. Splenocytes were suspended in PBS/2%FBS. Fluorochrome-mAbs were prepared at optimized dilutions, then added to cells (Supplementary Table ²). Splenocytes were co-labeled with B cell tetramers comprising CH848DT SOSIP trimers conjugated to two different fluors, VB515 and AF647. Cells were incubated at 4°C for 20min, then washed with PBS. Cells were resuspended in 100μL PBS containing Near-IR Live/Dead (ThermoFisher) at 1:1000, and incubated at room temperature for 20min. Cells were washed in PBS/2%FBS, then resuspended in PBS/2% formaldehyde. Cells were analyzed on a BD LSRII (BD Biosciences). Data were analyzed using FlowJo v10.7.1 (FlowJo).

Knock-in mouse immunizations

The CH235 UCA KI mouse model has been described previously⁸. 8–17 week old mice were used. 10 CH235 UCA KI mice were split into two groups (N=5 each group), with each group having at least one mouse. The synthetic Toll-like receptor 7/8 agonist 3M-052 absorbed to ALUM (3M-052-ALUM) was used as the adjuvant for the vaccine immunogens. Mouse vaccination studies were performed intramuscularly with 3M052-ALUM adjuvanted N279 CE3 design or the K279 CE3* design proteins. Vaccine immunogens were administered at 25mcg and formulated with 0.5mcg of adjuvant. Mice were immunized every 2 weeks for five times. Blood samples were collected 7 days prior to the first immunization (pre-bleed) and 7 days after each immunization. The DH270 IA4 KI mice are described above. 8–17 week old mice were used. 25 DH270 IA4 KI mice were split into two groups, each group had at least one female mice. The synthetic Toll-like receptor 7/8 agonist 3M-052 absorbed to ALUM (3M-052-ALUM) was used as the adjuvant for the vaccine immunogens. Mouse vaccination studies were performed intramuscularly with 3M052-ALUM adjuvanted CH848-d949 construct (N=12) or the DE3 design (N=13) proteins. Vaccine immunogens were administered at 17.5–25mcg and formulated with 0.5mcg of adjuvant. Mice were immunized every 2 weeks for six times. Blood samples were collected 7 days prior to the first immunization (pre-bleed) and 7 days after each immunization. Necropsy was performed 1 week after the third immunization, and blood, spleen, and lymph nodes were collected. All mice were cared for in a facility accredited by the AAALAC (Supplementary Data ³). All study protocols and all veterinarian procedures were approved by the Duke University IACUC. All rooms were on a 12/12 light cycle. Heat and humidity were maintained within the parameters outlined in The Guide for the Care and Use of Laboratory Animals, and animals were fed a standard rodent diet. Rodents were housed in individually ventilated micro-isolator caging on corn cob bedding.

Antibody isolation and production

Single antigen-specific B cells were sorted as previously described⁸. Briefly, inguinal and axillary lymph nodes were processed into single-cell suspensions and cryopreserved. Upon thawing, cells were counted then stained with optimal concentrations of the following fluorophore-antibody conjugations: BB700 anti-IgG1 (clone A85-1, BD Biosciences), BB700 anti-IgG2a (clone R2-40, BD Biosciences), BB700 anti-IgG3 (clone R40-82, BD Biosciences), PE anti-GL7 (clone GL7, Biolegend), PE-Cy7 anti-IgM (cloneR6-60.2, BD Biosciences), APC-R700 anti-CD19 (clone 1D3, BD Biosciences), BV510 anti-IgD (clone 11-26C.2a, BD Biosciences), BV605 anti-IgK (clone 187.1, BD Biosciences), and BV650 anti-B220 (clone RA3-6B2, BD Biosciences). Cells were also labeled with fluorophore-labeled CH848-d949 DT (10.17) in two colors (VB515 and Alexa Fluor 647). Dead cells were identified by labeling with LIVE/DEAD Fixable Near I/R stain (Thermo Fisher Scientific). Class-switched B cells were identified as single, viable lymphocytes that were B220+CD19+IgM−IgD−IgG+; antigen-specific B cells were identified as cells that co-labeled with CH848DT baits conjugated to VB515 and AlexaFluor647. Single cells were sorted on a BD FACS AriaIIu running DIVA version 8.0. Cells were sorted into cell lysis buffer and 5X first-strand synthesis buffer in individual wells of a 96-well PCR plate. Plates were frozen in a dry ice/ethanol bath immediately and stored at −80°C until reverse transcription of RNA. Immunoglobulin genes were amplified from singly sorted antigen-specific B cells of DH270 IA4 knock-in mice by RT-PCR as previously described with some modifications^8,93,94. Briefly, immunoglobulin genes were reverse-transcribed with Superscript III (Thermo Fisher Scientific, Waltham, MA) using random hexamer oligonucleotides as primers. The complementary DNA was then used as template to perform nested PCR for amplification of antibody heavy and light chain genes using AmpliTaq Gold 360 (Thermo Fisher Scientific, Waltham, MA). In parallel, PCR reactions were done with mouse immunoglobulin-specific primers and DH270 variable region-specific primers. In one set of PCR reaction, mouse variable region and mouse constant region primers were used to amplify mouse heavy, kappa, and lambda chain genes. In another set of PCR reaction, human variable region and mouse constant region primers were used to amplify heavy and lambda chain genes of the knock-in DH270 IA4 and its derivates. Positive PCR amplification of immunoglobulin genes was identified by gel electrophoresis. Positive PCR products were purified in Biomek FX Laboratory Automation Workstation (Beckman Coulter, Indianapolis, IN) and sequenced by Sanger sequencing. Contigs of the PCR amplicon sequence were made, and immunogenetics annotation was performed with software Cloanalyst⁹⁵ using both the human and mouse Ig gene libraries. Antibody genes were categorized as human or mouse based on the species of the V gene with the highest identity to the antibody sequence. The immunoglobulin gene amplicons were further used for protein expression by generating linear expression cassette of each gene. The linear expression cassettes were constructed by overlapping PCR to place the PCR-amplified VH and VK/L chain genes under the control of a CMV promoter along with heavy chain IgG1 constant region or light chain constant region and a BGH ploy A signal sequence. The linear expression cassettes of each paired heavy chain and light chain were then co-transfected into 293T cells. The supernatant containing recombinant antibodies were cleared of cells by centrifugation and used for binding assays. The genes of selected heavy chains were synthesized as IgG1 (GenScript). Kappa and lambda chains were synthesized similarly. Plasmids with the cloning of synthesized antibody genes were prepared using the Megaprep plasmid plus kit (Qiagen) for transient transfection. The paired heavy and light chain plasmid DNA was co-transfected into 293i cells using ExpiFectamine™ (Life Technologies, Carlsbad, CA) by following the manufacturer’s protocol. Antibody proteins were purified from the cell culture supernatant by protein A beads for binding, neutralization, and other immunological assays.

Mammalian cell surface display of HIV-1 envelope

A site-saturation library was designed using CH848.3.D0949.10.17 gp140 with N133D and N138T substitutions (10.17DT) as a template⁸. N133D and N138T substitutions were added to enable HIV-1 bnAb DH270 inferred precursor binding⁸. The site-saturation envelope library was synthesized (GenScript) to contain envelopes with a single substitution. The substitutions introduced could be any of the other 19 possible amino acids at any one position within the gp120 subunit. The stabilized, cleaved HIV-1 envelope gp140 was fused to a C-terminal c-myc tag and HRV-3C cleavage site, followed by a platelet-derived growth factor receptor transmembrane region that ensured cell surface presentation of the envelope. The open reading frame encoding the envelope library was cloned into the mammalian expression vector VRC8400 CMV/R.

Sixteen micrograms of CMV/R vector carrying HIV-1 SOSIP library were co-transfected with 67µg of carrier DNA (CMV/R vector) without the insert and 3.7µg of plasmid coding Furin into Freestyle 293F cells using a 293fectin Transfection Kit (Thermo Scientific) according to manufacturer’s instruction. Cells were cultured for 48h with shaking at 120rpm in an incubator filled with 8% CO2 at 37°C. Twenty million harvested transfected cells were stained with 500µl of 100µg/ml DH270 UCA and 10µg/ml chicken anti-c-myc. DH270 UCA and anti-c-myc antibody binding were detected with 500µl of 10µg/ml goat-anti-human PE (Sigma) and 10µg/ml goat anti-chicken Alexa 647 respectively. In separate tubes, two million cells were stained with 50µl of the above antibodies for analytical flow cytometry. Gates for sorting were set such that cells that bound to DH270 with a higher fluorescence intensity than the unmutated envelope (prototype) were sorted. RNA was then isolated from sorted cells using PurelinkTM RNA micro kit (Invitrogen). RNA was subjected to reverse transcription primed by CMVR-SOSIP-rev (cacagcagatct tcatcagcggggctttttctg) using AccuScript Reverse Transcriptase (Agilent) to generate cDNA. Double-stranded DNA was amplified from cDNA with CMVR-SOSIP-fwd (gtcaccgtcgtcgacgccacc atggaaaccgatacactgct) and CMVR-SOSIP-rev using Herculase II Fusion DNA Polymerases (Agilent) and subcloned into CMVR vector by SalI/BglII digestion and ligation. The resulting constructs were transformed into electrocompetent DH10B using electroporation followed by culture in 500mL of Luria broth. Plasmid was then extracted for another round of transfection and sorting. To get stronger binding envelopes, the selection stringency was increased by decreasing the concentration of DH270 UCA used to sort positive cells in each round. Sorted cells were sequenced by PacBio long-read sequencing.

PacBio Sequencing of library variants. RT-PCR products from the last cell sorting were purified using a gel-extraction kit (Qiaqen) and sequenced using PacBio long-read sequencing. The frequency of mutations at desired sites was calculated. DNA samples were quantified using Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA, USA) and sample purity and integrity were checked using Nanodrop and Agilent TapeStation, respectively. DNA library preparations, sequencing reactions, and initial bioinformatics analysis were conducted at Azenta Life Sciences (Formerly Genewiz) (South Plainfield, NJ, USA). Following slow annealing, PacBio SMRTbell amplicon libraries for PacBio Sequel were constructed using SMRTbell Express Template Prep Kit 2.0 (PacBio, Menlo Park, CA, USA) using the manufacturer-recommended protocol. The pooled library was bound to polymerase using the Sequel Binding Kit 3.0 (PacBio) and loaded onto PacBio Sequel using Sequel Sequencing Kit 3.0. Sequencing was performed on the required PacBio Sequel SMRT cells. The generated subreads were demultiplexed and clustered. Long Amplicon reads were obtained using PacBio laa pblaa v2.4.2 with default parameters. The generated subreads were demultiplexed and aligned to the reference sequence using the PacBio software pbmm2. Variant calling was performed using the PacBio software variantCaller arrow using default parameters. Variants that passed the default filters were parsed into a filtered vcf file for each sample. The substitutions observed in the most abundant variants were introduced into soluble HIV-1 envelope for characterization.

HIV-1 neutralization assays

Neutralizing antibody assays were performed with HIV-1 Env-pseudotyped viruses and TZM-bl cells (NIH AIDS Research and Reference Reagent Program contributed by John Kappes and Xiaoyun Wu) as described previously^96,97. Neutralization titers are the reciprocal sample dilution (for serum) or antibody concentration in µg/mL (for mAbs) at which relative luminescence units (RLU) were reduced by 80% or 50% (ID80/IC80 and ID50/IC50 respectively) compared to RLU in virus control wells after subtraction of background RLU in cell only control wells. Serum samples were heat-inactivated at 56°C for 30min prior to assay.

Next generation sequencing of mouse B cell receptor repertoires

We performed NGS on mouse antibody heavy and light chain variable genes using an Illumina sequencing platform. RNA was purified from splenocytes using the RNeasy Mini Kit (Qiagen, Cat# 74104). Purified RNA was quantified via Nanodrop (Thermo Fisher Scientific) and used to generate Illumina-ready heavy and light chain sequencing libraries using the SMARTer Mouse BCR IgG H/K/L Profiling Kit (Takara, Cat# 634422). Briefly, 1μg of total purified RNA from splenocytes was used for reverse transcription with Poly dT provided in the SMARTer Mouse BCR kit for cDNA synthesis. Heavy and light chain genes were then separately amplified using a 5′ RACE approach with reverse primers that anneal in the mouse IgG constant region for heavy chain genes and IgK for the light chain genes (SMARTer Mouse BCR IgG H/K/L Profiling Kit). The DH270 IA4 KI mouse model has the light chain gene knocked into the kappa locus, therefore kappa primers provided in the SMARTer Mouse BCR kit were used for light chain gene library preparation. 5µl of cDNA was used for heavy and light chain gene amplification via two rounds of PCR (18 and 12 cycles per round). During the second round of PCR, Illumina adapters and indexes were added. Illumina-ready sequencing libraries were then purified and size-selected by AMPure XP (Beckman Coulter, Cat# A63881) using kit recommendations. The heavy and light chain libraries per mouse were indexed separately, thus allowing us to deconvolute the mouse-specific sequences during analysis. Libraries were quantified using QuBit Fluorometer (Thermo Fisher). Mice were pooled by groups for sequencing on the Illumina MiSeq Reagent Kit v3 (600 cycle) (Illumina, Cat# MS-102-3003) using read lengths of 301/301. 20% PhiX was spiked in to increase sequence diversity of the libraries due to the predominance of the antibody knockin-derived reads in the repertoires.

Annotation of antibody sequences with immunogenetic information, clonal clustering, and clonal tree reconstruction were performed using the software Cloanalyst (https://www.bu.edu/computationalimmunology/research/software/). Due to variable recovery rates of heavy and light chain reads using the bulk single chain NGS sequencing of mouse BCR repertoires and the depth required for sufficient statistics for analysis of rare mutational events, mouse repertoires that had <1000 unique and functional bnAb UCA knock-in derived reads recovered were excluded from the data analysis. Functional heavy and light chains were defined by the presence of required immunogenetic characteristics as determined by Clonanalyst including: V gene regions of at least 200 base pairs in length, presence of invariant cysteines, and tryptophan (heavy) or phenylalanine (light), CDR3s in reading frame 1, non-zero CDR3 length, and absence of stop codons. Reported mutation frequencies were calculated at the nucleotide level from the beginning of the antibody sequence through the invariant cysteine codon that precedes CDR3. Analysis of the probability of mutations was performed using the computational program ARMADiLLO as previously described^3,8 (Supplementary Figs. ³³–³⁵).

Cryo-EM data were collected at the Duke Krios at the Duke University Shared Materials Instrumentation Facility (SMIF), a member of the North Carolina Research Triangle Nanotechnology Network (RTNN), which is supported by the National Science Foundation (award number ECCS-2025064) as part of the National Nanotechnology Coordinated Infrastructure (NNCI). This study used the computational resources offered by Duke Research Computing (http://rc.duke.edu; NIH 1S10OD018164-01) at Duke University. This project was supported by NIH, NIAID, Division of AIDS Consortia for HIV/AIDS Vaccine Development (CHAVD) Grant UM1AI144371 (B.F.H.), R01AI145687 (P.A.), U54AI170752 (R.C.H. and P.A.), DP2-AI164323-03 (R.H.), and Translating Duke Health Initiative (R.C.H. and P.A.).