Germline-targeting HIV-1 Env vaccination induces VRC01-class antibodies with rare insertions

Summary Targeting germline (gl-) precursors of broadly neutralizing antibodies (bNAbs) is acknowledged as an important strategy for HIV-1 vaccines. The VRC01-class of bNAbs is attractive because of its distinct genetic signature. However, VRC01-class bNAbs often require extensive somatic hypermutation, including rare insertions and deletions. We describe a BG505 SOSIP trimer, termed GT1.2, to optimize binding to gl-CH31, the unmutated common precursor of the CH30-34 bNAb lineage that acquired a large CDRH1 insertion. The GT1.2 trimer activates gl-CH31 naive B cells in knock-in mice, and B cell responses could be matured by selected boosting immunogens to generate cross-reactive Ab responses. Next-generation B cell sequencing reveals selection for VRC01-class mutations, including insertions in CDRH1 and FWR3 at positions identical to VRC01-class bNAbs, as well as CDRL1 deletions and/or glycine substitutions to accommodate the N276 glycan. These results provide proof of concept for vaccine-induced affinity maturation of B cell lineages that require rare insertions and deletions.


In brief
The induction of bNAbs is essential for a protective HIV-1 vaccine, but current vaccines are unable to induce sufficient B cell maturation. Caniels et al. describe an immunization regimen that elicits neutralizing antibodies toward the CD4bs and isolates monoclonal antibodies with rare sequence features that resemble bNAbs.

INTRODUCTION
Almost 40 years after the identification of HIV-1, the need for a vaccine remains as urgent as ever. A vaccine will need to confer protection against a plethora of HIV-1 strains and it is likely that an essential component of such a vaccine is to induce broadly neutralizing antibodies (bNAbs). bNAbs are generated by a subset of HIV-1-infected individuals after multiple years of HIV-1 replication and many have been cloned and characterized. 1,2 bNAbs can treat and prevent infection in non-human primate studies and are currently being evaluated in clinical trials for HIV-1 treatment and prevention. [3][4][5][6] However, inducing bNAb responses through vaccination in humans remains a major challenge, at least in part because bNAbs require a lengthy and complex process of co-evolution with the virus. [7][8][9][10] The first critical step in bNAb induction is the activation of naive B cells that have the intrinsic capacity to develop bNAbs (germline [gl]-bNAbs). Such B cells are usually present at low frequencies in the human naive B cell repertoire and have no or low affinity for current HIV-1 vaccine candidates, immediately placing these B cells at a selective disadvantage relative to more abundant and higher affinity B cells recognizing other epitopes. [11][12][13][14][15][16] However, gl-bNAbs can serve as templates for the design of immunogens that selectively activate these rare naive B cells (reviewed in 17,18 ). Such immunogens have indeed been generated and in some cases, such as for N332 supersite-targeting gl-PGT121, 19 have led to the induction of NAbs in knock-in (KI) mouse models with high frequencies of HIV-1 bNAb precursors, providing proof-of-concept for ''germline targeting'' strategies. 18 Particularly attractive gl-bNAb precursors are those of the VRC01-class. VRC01-class bNAbs target the conserved CD4 binding site (CD4bs) epitope on the Env trimer and use the IGHV1-2*02 gene segment in combination with a light chain (LC) bearing a short five amino acid LC complementarity determining region 3 loop. 20,21 Some examples of potently neutralizing VRC01-class bNAbs include VRC01, 3BNC60, CH31, 12A12, PGV20, N49P7, and the broadest HIV-1 bNAb described to date, N6. 20-24 All of these VRC01-class bNAbs were isolated from distinct HIV-1 patients, indicating that humans can reproducibly generate such bNAbs. Indeed, VRC01-class precursors can be found in the vast majority of humans at frequencies that are sufficient and practical for germline targeting, an important prerequisite for a viable vaccine strategy. 25 Finally, because of their superior breadth and potency, VRC01-class bNAbs have been the focus of many atomic-level structural studies revealing the precise paratopes and epitopes of such bNAbs and their gl-bNAb precursors, thereby facilitating structure-based vaccine design. 26,27 However, VRC01-class bNAbs often require high levels of somatic hypermutation (SHM) to be broad and potent, sometimes approaching 50%. 20,21 Moreover, many VRC01-class bNAbs require rare insertions and/or deletions (indels) for full activity. For example CH31, 3BNC60, and VRC03 contain insertions in the heavy chain (HC) CDRH1 or FWR3, while others, including VRC01, PGV04, and PGV20, require deletions in the LC CDRL1. 28 It has been proposed that highly improbable insertions, such as in the CDRH1 of CH31 or the FWRH3 of 3BNC60, make additional contacts with a neighboring protomer of the Env trimer. 29 In the case of CH31, it only acquired its neutralization breadth after formation of the large (nine-residue) CDRH1 insert, underscoring the impact insertions can have on recognition and neutralization of HIV-1. 28,30 Similarly, CD4bstargeting bNAbs 1-18 that uses IGHV1-46, the closest related IGHV gene to IGHV1-2, has a six-amino-acid CDRH1 insertion that proved necessary for breadth and potency. 30 Furthermore, VRC01-class bNAbs require small CDRL1 deletions or glycine substitutions, usually in an GXG motif, to accommodate the N276 glycan, a major obstruction to access of the CD4bs. 31,32 Thus, while relatively high levels of SHM can be achieved through vaccination in some circumstances, 4,33 the induction of these rare indels pose a major roadblock to eliciting potent VRC01-class bNAbs by vaccination.
Most Env proteins do not engage VRC01-class bNAb precursors, 11,13 unless specifically modified to do so. The lead vaccine candidates for targeting VRC01-class gl-bNAbs include eOD-GT8 multimerized/60-mer, 34 426c TM4DV1-3, 35 and BG505 SOSIP.v4.1-GT1.1, a derivative of BG505 SOSIP.v4.1-GT1. 36 All three are now in human clinical trials (NCT05414786, NCT05471076, and NCT04224701, respectively). eOD-GT8 primes VRC01-class precursors in various KI mouse models, 34,37-39 as does 426c TM4DV1-3. 35 Furthermore, eOD-GT8 can select VRC01-class precursors from the naive B cell repertoire of healthy human donors. 31 eOD-GT8 and 426c TM4DV1-3 are based on Env subdomains, either the gp120 outer domain or the gp120 core, respectively. eOD-GT8 and 426c TM4DV1-3 have not been reported to be able to select for insertions efficiently, but eOD-GT8 priming did select for CDRL1 deletions. 40,41 The design of BG505 SOSIP.v4.1-germline-targeting trimer 1 (GT1) using the native-like trimer BG505 SOSIP platform, is rooted in the hypothesis that a native-like trimer might offer ad-vantages over smaller Env fragments by placing the epitope of choice in the natural native-like trimer context. As such, it constrains the approach angles so that they resemble those of the eventual target of bNAbs: the native Env trimer. Therefore, we previously re-engineered the BG505 SOSIP trimer to specifically engage gl-VRC01 as well as V2-apex gl-precursors. However, while the GT1 trimer was able to engage VRC01-class gl-VRC01, gl-PGV19, and gl-NIH45-46 with nanomolar affinity, it was unable to bind to other gl-bNAbs isolated from different human individuals, including gl-3BNC60, gl-12A12, and gl-CH31. 36 As germline-targeting immunogens are considered to be priming immunogens, it is likely that additional and different immunogens are required to sequentially guide antibody maturation toward neutralization breadth. 16,33,37 Indeed, sequential immunization regimens starting with a germline-targeting immunogen, followed by boosting with shaping and polishing immunogens, including native-like SOSIP trimers, have improved and broadened Env recognition, in one case leading to development of bNAbs in a gl-PGT121 KI mouse model. 19,42 Nevertheless, bNAbs have not yet been consistently induced by vaccination of VRC01-class KI mice with the lead germline-targeting immunogens. [34][35][36][37][38][39] The major impediment to the development of VRC01-class bNAbs in these models is thought to be the N276 glycan that hinders access to the CD4bs. 37,39 VRC01-class precursors could be initiated and matured into Abs that could neutralize viruses from which the N276 glycan was absent, but not wild-type viruses. 33,35,37,39 However, more recently, select VRC01-class recombinant NAbs have been isolated from KI mice that neutralize N276 glycan-bearing viruses with up to $50% breadth, albeit with low potency. 41 Here, we sought to evaluate whether a modified version of the germline targeting GT1 trimer, GT1.2, specifically engineered to expand VRC01-class precursor recognition, could prime a VRC01-class antibody response in a novel gl-CH31 KI mouse model and whether this initial response could be broadened and matured by subsequent immunization with shaping and polishing immunogens. We report that GT1.2 priming followed by these shaping and polishing immunogens resulted in CD4bsspecific serum activity and neutralization of VRC01-signature viruses. This immunization regimen also reproducibly selected for high levels of improbable VRC01-class mutations, including extremely rare multi-residue insertions similar or identical to those observed in VRC01-class bNAbs. Moreover, the vaccination regimen also selected for multi-residue deletions or glycine substitutions in the CDRL1.

RESULTS
Establishment of a gl-CH31 KI mouse model Although VRC01-class precursor KI mouse models exist, including for gl-VRC01 and gl-3BNC60, they exhibit vastly different B cell phenotypes, despite expressing identical IGHV1-2 gene segments. While gl-3BNC60 KI mice would allow for the study of FWR3 insertions as present in 3BNC60, these models display multiple negative B cell selection controls including peripheral deletion and apoptosis, anergy, and extensive LC editing and swapping, cumulatively indicating significant in vivo autoreactivity and profoundly abrogated affinity  Figure S1A), whose naive B cells have enforced the IgH/L locus-specific expression of the inferred UCA rearrangements of the CH31-CH34 bNAb lineage with methods that we previously used to engineer other bNAb-lineage UCA-rearranged KI models. [43][44][45] As opposed to the gl-3BNC60 model and similar to the gl-VRC01 models, gl-CH31 KI mice had largely unperturbed overall B cell development relative to wild-type (WT) C57BL6 controls, including similar developmental subsets ( Figure S1B) and comparable surface IgM and IgD BCR densities ( Figure S1C) with only a modest decrease in total B cell cellularity ( Figure S1D). In contrast with the in vivo tolerizing B cell controls observed in gl-3BN60 KI mice, these findings of largely normal B cell development in gl-CH31 KI mice reinforces the notion that CDR3 specificity is crucial in controlling the developmental the fates of VRC01-class B cell precursors. 46 Importantly, this model allows for the study of the ability of vaccine regimens to recapitulate the acquisition of large HC insertions in VRC01-class bNAb lineages. Finally, to incorporate additional B cell repertoire diversity to this model, we crossed gl-CH31 KI mice with WT C57BL/6 mice, resulting in heterozygous (V H DJ H Because of both this feature (which provides an alternate, unrearranged murine LCk allele) and the manner in which we have knocked in the CH31 V k 1-33/Jk2 rearrangement (Figure S1A), numerous opportunities exist for other LC rearrangements. Indeed, $45% of endogenous LCs are paired with the gl-CH31 HC (Figure S1F), consistent with the number of non-CD4bs cells we detect by flow cytometry ( Figure S1E). In summary, the heterozygous gl-CH31 mouse model is thus suitable to assess the priming potential of a VRC01-class germline-targeting immunogen for its ability to successfully activate gl-CH31 + B cells in vivo.
GT1.2 engages gl-CH31 while retaining binding to other gl-bNAbs To assess the activation of gl-CH31 B cells in vivo and to study the selection of indels typical of VRC01-class bNAbs, we designed a trimer that could engage gl-CH31, as the parental BG505 SOSIP GT1 does not engage gl-CH31. 36 Accordingly, we introduced an N279D substitution that establishes an additional contact between GT1 and VRC01-class gl-bNAbs and named the resulting trimer GT1.2 ( Figure 1A). The glycan profile of GT1.2 is similar to that of GT1 and is characterized by complex glycans at the apex and trimer base, which may be caused by deletion of glycans leading to localized enhancement of glycan processing ( Figures 1A, S2B, and S2C). 36,47 The resulting trimer was well formed and remained in a closed, native-like prefusion conformation comparable with GT1 and other native-like HIV-1 trimers, as assessed by negative stain electron microscopy and gel electrophoresis ( Figures 1B and S2A). 36,48 Moreover, its thermal denaturation was also highly comparable with that of GT1 and BG505 SOSIP.664 as evaluated by differential scanning calorimetry, with a melting temperature of 68.0 C ( Figure S2D). 36,49 To assess whether the N279D substitution conferred binding to gl-CH31, we tested GT1.2 binding to different germline precursors and mature bNAbs in a surface plasmon resonance assay ( Figure 1C). Whereas no dissociate constant (K D ) could be derived from the weak GT1 binding to gl-CH31, GT1.2 was able to bind gl-CH31 with slightly higher affinity than eOD-GT8 (K D GT1.2, 4 mM; K D eOD-GT8, 11.9 mM) ( Figures 1C and S2E, Table S1, 34 ). Moreover, the introduction of N279D preserved binding of GT1.2 to other gl-bNAbs such as gl-VRC01 (K D of 1200 nM) and gl-PGV19 (K D of 94 nM), thus effectively expanding the range of VRC01-class precursors that can be engaged with a single substitution ( Figures 1C and S2E, Table S1). This bivalent modeling has been validated 50,51 and used for analyzing the interaction of germline-reverted and mature antibodies with germline-adapted and -unadapted Env trimers. 13,36 To confirm that the N279D substitution indeed allowed engagement of the conserved W100b, we determined the structure of GT1.2 in complex with VRC01-class precursor gl-PGV20 and PGT124 at 3.8 Å resolution ( Figures 1D-1F, Table S2). The structure revealed an additional potential hydrogen bond between N279D GT1.2 and W100b gl-PGV20 as hypothesized, reinforcing the GT1.2/gl-PGV20 contacts, and, by inference, gl-CH31 contacts ( Figure 1E). Moreover, the overall structure of GT1.2 is highly similar to BG505 SOSIP.664 and GT1, with C alpha root-mean-square deviation values of 0.6 Å (GT1.2 vs. BG505 SOSIP.664) and 0.4 Å (GT1.2 vs. GT1) ( Figure 1F). Thus, with the N279D substitution in GT1.2, the range of VRC01-class bNAb precursors that can engage GT1.2 is expanded through an additional conserved W100b contact.
GT1.2 priming induces broadly reactive CD4bs-specific antibody responses Next, we evaluated GT1.2 as a priming immunogen in the gl-CH31 KI mice described above. In a first experiment, five gl-CH31 KI mice received 25 mg GT1.2 trimer formulated in 60 mg poly I:C adjuvant at weeks 0 and 4 followed by boosting with a fully glycosylated BG505 SOSIP trimer at weeks 7, 13. and 18 ( Figure 2A). Serum antibody responses against GT1.2, BG505 SOSIP, a CD4bs knockout (KO) BG505 SOSIP (BG505 D368R), and a candidate shaping immunogen AMC008 GT1 were measured by ELISA 36,52,53 ( Figure S3A-S3C). One GT1.2 immunization led to detectable GT1.2-specific responses in all five mice and the additional immunizations strengthened these responses. However, the sera were only weakly reactive to BG505 SOSIP and equally reactive with the BG505 SOSIP D368R mutant that knocks down VRC01-class Ab binding, indicating that antibodies induced by this regimen did not strongly recognize fully glycosylated trimers, and that those that did were not specific for the CD4bs ( Figure 2B, left and right). Furthermore, only one mouse displayed neutralization of 426c.TM4, a virus used to gauge VRC01-class neutralization signatures in serum ( Figure 2C). The 426c.TM1 virus, which lacks only the N276 glycan, was not neutralized by any of the sera. We concluded that this simple prime-boost regimen using GT1.2 and BG505 SOSIP trimers was insufficient to strongly activate and mature VRC01-class responses in gl-CH31 mice. One implication is that efficient maturation may require shaping immunogens between the priming and polishing stages.
Therefore, we improved the study regimen by selecting affinity-intermediate shaping immunogens in real time, based on the serological reactivity of gl-CH31 KI mice primed with GT1.  Figure 2D). We also assessed improving the functional affinity of GT1.2 by enhancing avidity through the use of two-component I53-50 protein nanoparticles (NPs) that can display twenty SOSIP trimers. 48, 55 We generated GT1.2 I53-50 NPs, which efficiently activated gl-CH31 KI B cells ex vivo ( Figure S3D), and thus were included in a follow-up study ( Figure 2D). We primed a new cohort of gl-CH31 KI mice at weeks 0 and 3 with either 25 mg GT1.2 trimer, equimolar amounts of GT1.2-I53-50 NP, or 25 mg BG505 SOSIP trimer as a control (n = 8 per group). BG505 SOSIP is not optimized for germline-targeting and thus should mainly induce off-target, non-CD4bs Ab responses. To compare these priming immunogens, serological reactivity was measured at week 5 by ELISA. The GT1.2 and GT1.2 NP immunized animals developed strong binding antibody responses against GT1.2, although there was considerable variation between animals ( Figures 2E and S3E). In contrast, BG505 SOSIP did not induce a strong GT1.2 or BG505 SOSIP response in these mice. We did not observe significant serological differences between the GT1.2 trimer and GT1.2 NP groups, possibly because the avidity advantage of NP presentation does not offer a benefit in the context of high precursor frequency 56,57 ( Figure S3E). Therefore, these groups were combined in the analyses below.
GT1.2 priming also induced high binding levels to candidateshaping immunogen BG505 SOSIP-INT3. INT3 contains only three modifications compared to BG505 SOSIP: N276D, T278R, and a seven-amino-acid deletion in the V2 region, all of which are also present in GT1.2. We also detected weak binding to AMC008 GT1 ( Figure 2E). Based on the results, we evenly divided each group into two subgroups and boosted the animals with either BG505 INT3 or AMC008 GT1 with the aim of broadening the CD4bs-directed response and bridging the affinity gap between glycan-deficient GT1.2 and glycan-rich native Envs. This boost indeed strengthened and broadened the serological reactivity; the sera became reactive with AMC008 GT1 and also with AMC008 SOSIP trimers lacking only the N276 glycan ( Figure 2F). However, no significant differences were observed between the two different shaping strategies ( Figure S3F).
To further boost the breadth of the serum response, all animals across groups were immunized with the AMC008 N276D trimer. This additional boost led to the development of reactivity with a clade C trimer that lacked the N276 glycan, ZM197M N276D, indicative of a further broadening of the response (Figure 2G). 58,59 The sera were now also strongly reactive with unmodified BG505 SOSIP trimers, whereas BG505 SOSIP-primed animals did not develop such binding, suggesting that the shaping immunogens had specifically boosted GT1.2-primed CD4bs-directed antibodies ( Figure 2G). We then proceeded to boost all animals with a bivalent cocktail of AMC008 N276D and ZM197M N276D trimers ( Figure S3G) before polishing by immunizing thrice with a cocktail of unmodified, native-like trimers from isolates BG505 and Q23 (both clade A) ( Figure S3), AMC008 (clade B), and ZM197M and DU422 (both clade C) (Figure 2D and Table S3). 59 To verify that the serum response was at least partly CD4bs directed, we compared binding to BG505 SOSIP and BG505 SOSIP D368R at each timepoint. Throughout most of this elaborate immunization schedule, the ratio of BG505/BG505 D368R binding was greater than one, indicating the presence of a CD4bs-directed response ( Figure 2H). In contrast, BG505 SOSIP-primed animals did not exhibit CD4bs specificity at any stage tested, and the ratio of SOSIP/SOSIP D368R binding was significantly lower after seven immunizations (p = 0.002) ( Figure 2H). We observed the same trend with D368R trimers of AMC008 and ZM197M, indicating that a broadly reactive CD4bs-directed serum response was generated in the GT1.2primed animals, but not in the BG505 SOSIP-primed ones (Figure 2I). Finally, a VRC01-class serum neutralization signature was detected in the majority (10/16) of GT1.2-primed animals as measured using the 426c.TM4 virus, whereas only one BG505 SOSIP-primed animal exhibited sustained neutralization of this indicator virus ( Figure 2J, left). Moreover, 7 of the 10 GT1.2-primed animals showed 426c.TM4 neutralization, although none of the BG505 SOSIP-primed animals neutralized the 426c.SM virus only lacking the N276 glycan ( Figure 2J, right). Taken together, priming with GT1.2, but not unmodified BG505 SOSIP, followed by shaping and polishing, induced a CD4bsspecific serum response in gl-CH31 KI mice.
GT1.2 priming followed by shaping and polishing selects for rare VRC01-class sequence features The VRC01-class of bNAbs is characterized by high SHM and many of its members also display indels rarely observed in other viral infections. 28 We examined whether the CD4bs-directed serum response in our experiment had any of these features by performing Illumina next-generation sequencing on CD4bsspecific B cells from splenocytes sorted by using eOD-GT8 with negative selection with an eOD-GT8 CD4bs KO mutant from spleens recovered after study completion. eOD-GT8 was chosen as it has a high affinity for gl-VRC01-class bNAbs. 34 We analyzed three groups of immunized mice described previously: mice receiving the short prime-boosting regimen in our first experiment (from here on designated ''GT1.  Figure 3A). For each of the groups, at least three mice were included. We observed that GT1.2 priming induced significantly more nonsynonymous mutations (%30 amino acids) compared with BG505 SOSIP across Ig subtypes and in IGKV regions ( Figure 3B, left, and Figure S4A). However, the median mutation frequency per mouse was significantly higher for GT1.2 (long) than for GT1.2 (short) (p = 0.0013), and also higher than for BG505 SOSIP priming ( Figure 3B, right). These findings are also reflected in higher sequence diversity for the GT1.2 (long) regimen in 1,000 randomly selected sequences for each mouse ( Figure S4B). The mean number of improbable mutations (defined as those having <2% probability in the absence of selection) defined by ARMADiLLO 60 for each mouse was also significantly higher in GT1.2-primed mice with an extended immunization regimen compared to a simple prime-boost regimen ( Figure 3C).
We then assessed whether the SHM was on-track and corresponded with mutations found in VRC01-class bNAbs VRC01, CH31, PGV04, PGV20, 3BNC60, and 12A12 (termed VRC01-class mutations). In SOSIP-primed mice, the majority of the sequences recovered had low numbers of mutations and did not exhibit positive selection of VRC01-class mutations. Similar results were obtained with the GT1.2 (short) regimen-administered group ( Figure 3D, left and middle). In contrast, sequences recovered from the GT1.2 (long) regimen-administered mice were not only highly mutated but Article ll OPEN ACCESS revealed a strong selection for VRC01-class mutations, with some sequences harboring up to 19 substitutions found in VRC01-class bNAbs ( Figure 3D, right). The accumulation of these high numbers of nonsynonymous VRC01-class mutations is strongly suggestive of sequential cycles of recall of previously expanded B cell clones rather than de novo recruitment of naive B cells. Specific selection of VRC01-class mutations was observed in only 29% of BCR sequences from BG505-primed mice, 54% in the group receiving the short GT1.2 regimen, and >90% in that receiving the GT1.2 (long) regimen group ( Figure 3E). GT1.2 priming was able to select for the minimal number of VRC01-class mutations necessary for broad and potent neutralization, as that number is 11 and 14 in minimally mutated (min)VRC01 and min12A21, respectively. 61 However, the exact combination of these mutations in minVRC01 and min12A21 are not found in the monoclonal antibodies (mAbs) described here. Next, we investigated the presence of improbable indel events. We did not observe any CDRH1 insertions in naive mice nor in SOSIP-primed mice ( Figure 3F). However, GT1.2primed mice did show rare CDRH1 insertions, with splenic B cell V(D)J rearrangement sequences from all five mice comprising the GT1.2 (long) group having such insertions ranging in size from four to six amino acids ( Figure 3F). Moreover, some GT1.2-primed mice also showed insertions in the FWR3, a region of low mutability because of high frequency of activationinduced cytidine deaminase cold spots. 60 Furthermore, CDRL1 glycine substitutions that follow the VRC01-class GXG motif and that are also thought to drive accommodation of the N276 glycan, were observed in mice in all groups ( Figure 3F). We assume these substitutions were induced and/or selected as a result of repeated immunization with N276-containing SOSIPs, not by the GT1.2 priming immunogen, which lacks the N276 glycan. Glycine substitutions were common in the groups that received an eight-immunization regimen but rare in the GT1.2 (short) group. Furthermore, in one mouse in the eight-immunization regimen primed with GT1.2, we observed the second known mechanism of N276 glycan accommodation: a two-amino-acid deletion in the CDRL1 ( Figure 3F). We note that mature CH31 itself does not have such a deletion, but has a GXG motif, which may pertain to a preference of the IGHV1-2/IGKV1-33 pairing to resort to GXG motifs as is also observed in IGKV1-33-bearing VRC01-class bNAbs N6 and 12A12. 21,22 In addition to overall on-track VRC01-class mutations and rare indel events, GT1.2 (long) mice more frequently had specific mutations shared with CH31 that are known to contact the trimer than the other groups ( Figure 3G). A number of these mutations, including G56A and T57V, have been implicated previously as key residues in VRC01-class maturation. 20,41 While not all observed frequencies are statistically significantly different, a clear trend is visible where GT1.2 (long) selects for contact mutations more consistently and at up to 1,000-fold higher median frequencies with the exception of Q61W that might rely on N276-glycan exposure ( Figure 3G). Thus, an immunization regimen consisting of GT1.2 priming, boosting with shaping immunogens lacking the N276 glycan and polishing with a cocktail of natively glycosylated Envs selects for on-track VRC01-class mutational patterns, including highly infrequent multi-residue indel events and contact mutations shared with CH31 and other VRC01-class bNAbs.
GT1.2 primes antibodies with VRC01-class neutralizing potential To assess whether these mutational signatures lead to broad binding and possibly neutralization, we sorted single CD4bsspecific eOD-GT8 + /eOD-GT8 KO À B cells from five immunized mice in the GT1.2-primed groups that were used in the next-generation sequencing (NGS) analysis (Figure 3). In total, we acquired unique HC/LC paired sequences for 405 B cells, of which 57 were selected and expressed as mAbs based on interesting sequence features and a representative range in SHM, resulting in mAbs A1-A57 (Table S4). Five of 57 mAbs did not show IgG expression and were not analyzed further. The 52 mAbs had 5-17 VRC01-class mutations, consistent with the B cells analyzed by NGS ( Figure 3D), and in the range of minVRC01 (12), min12A21 (17), and BG24, a mature VRC01-class bNAb with relatively low SHM (22) 62 ( Figure 4A). Apart from accumulating VRC01-class mutations, specific key residues in the HC important for VRC01-class breadth and potency are present in this subset of mAbs, 41 with an average of four of nine key mutations per mAb ( Figure 4B). In this representative panel of selected mAbs, some VRC01-class contact residues, such as T57V and Y33 F/V/L/I, were selected in the vast majority of mAbs, which is consistent with the B cells analyzed by NGS ( Figure 3G). In contrast, amino acid substitutions at other sites such as V37 were rarely observed ( Figure 4B). Seven clonally unrelated mAbs had multi-residue insertions, either in the CDRH1 region or in the FWR3 region ( Figure 4C), and 17 of 52 selected mAbs (33%) had a glycine substitution in the CDRL1.
All but two mAbs that expressed efficiently during transient transfection of HEK293F cells showed binding to GT1.2 in a CD4bs-dependent fashion as illustrated by the absence of binding to GT1.2 with two knock out mutations for VRC01-class bNAbs, D279A and D368R ( Figure 4D). Most mAbs bound strongly (median effective concentration [EC 50 ] between 0.01 and 0.1 mg/mL) to shaping immunogen AMC008 GT1 and trimers lacking the N276 glycan. However, when the N276 glycan was present, many (31/52) of these mAbs lost the ability to bind (EC 50 > 50 mg/mL), showing that these mAbs had not advanced toward the accommodation of the N276 glycan. However, a proportion (21/52) was able to recognize one or more fully glycosylated native-like trimers.
We tested the ability of these mAbs that bound at least one fully glycosylated trimer to neutralize the autologous viruses. Of these 21 mAbs, 17 neutralized GT1.2 at a half-maximal inhibitory concentration (IC 50 ) below 0.004 mg/mL ( Figure 4E, left) and 12 mAbs neutralized BG505 N276D with an IC 50 of <1 mg/mL. Moreover, a number of mAbs neutralized fully glycosylated native viruses and some neutralized up to five autologous viruses from clades A, B and C, although we note that this neutralization was weak in all cases (10 mg/mL < IC 50 < 200 mg/mL) ( Figure 4E, right). We confirmed that these mAbs target the CD4bs as they potently neutralize the VRC01-class signature viruses 426c.TM1 (N276D/N460D/N463D) and 426c.SM (N276D) but not their CD4bs KO counterparts (N279K) ( Figure 4F). Some mAbs showed sporadic neutralization of heterologous viruses from the nine-virus global panel when the N276 glycan was removed (N276Q), although no neutralization was observed for the parental viruses ( Figure 4G). Thus, although some mAbs have evolved single or double glycines in their CDRL1, the N276 glycan remained a major hurdle in the neutralization of heterologous viruses. The genetic and functional properties of these mAbs collectively suggest that they have progressed substantially on the path from VRC01-class germline precursors to VRC01-class bNAbs and might require few additional mutations to overcome the N276-glycan barrier and acquire neutralization breadth and potency.

Rare indels induced by GT1.2 priming play a major role in antibody binding and neutralization
Although the functional consequences of the two amino acid CDRL1 deletion signature found in many VRC01-class bNAbs are well understood, 32,62 the large CDRH1 and FWR3 insertions have not been the subject of extensive in-depth study. It has been proposed that the CDRH1 insertions of CH31 and 1-18 and the FWR3 insertion of VRC03 and 3BNC60, are involved in establishing contacts with the adjacent Env protomer. 29,30 As far as we know, the mAbs described here are the first reported mAbs with CDRH1 or FWR3 insertions after vaccination. Article ll OPEN ACCESS We tested binding and neutralization of mAbs A7, containing a four amino acid CDRH1 insertion; A23 with a two amino acid deletion in CDRL1; and A27 with a six amino acid insertion in FWR3. Binding to GT1.2 was not negatively impacted by reversion of the respective indel event, with the binding of A23 with full-length CDRL1 in fact being slightly higher ( Figure 5A, left), suggesting that the indels were not selected during the initial GT1.2 priming. However, removal of the indel event led to a 2.5-fold and 22-fold decrease in binding to fully glycosylated BG505 SOSIP for A23 and A27, respectively ( Figure 5A, middle). However, while A27 shows a similar 15-fold reduction to background levels in binding to AMC008, A7 was not impacted by the removal of its four amino acid insertion ( Figure 5A, right). Bio-layer interferometry experiments with BG505 SOSIP confirmed our ELISA findings: A23 showed a modest reduction in binding when the two amino acids in CDRL1 were restored, and A27 showed a large reduction in binding when the six amino acid FWR3 insertion was removed ( Figure 5B). These data were further corroborated by neutralization experiments. The neutralization capacity of A27 was severely impacted when the FWR3 insertion was removed, resulting in a failure to neutralize BG505 N276D at IC 50 of <5 mg/mL and showing a reduction in neutralization potency against native viruses of up to 10-fold ( Figures 5C and S5). Performing a nearest-neighbor search with the A27 HC sequence across all $250,000 recovered NGS reads from the repertoire of mouse V11417 revealed that the FWR3 insertion in A27 did not arise until significant SHM was achieved, with the closest-related sequence found exhibiting 14 amino acid substitutions, among which are 6 VRC01-class substitutions in the CDRH2 ( Figure 5D). Many of these mutations, such as T57V and Q61Y, are contact residues in VRC01-class bNAbs and are highly improbable ( Figure 5D). These analyses support the supposition that the insertion was not induced by or selected for by the priming immunogen, but rather during the shaping or polishing phases ( Figure 5A).
Finally, to determine whether the A27 FWR3 insertion contributes to antibody-antigen interaction, we used AlphaFold-Multimer to predict the structure of A27. 63 AlphaFold is an artificial intelligence software that attempts to model the structure of a protein based on its amino acid sequence. 64 The AlphaFold-predicted structure of A27 aligns well with experimentally determined structures of VRC01-class bNAbs VRC01, CH31, and 3BNC60, with the variable domains overlapping to a greater extent than the C H 1 and C L domains of the modeled IgG1 Fab ( Figure 5E). Similar to the FWR3 and CDRH1 insertions of 3BNC60 and CH31, respectively, the six inserted residues in the FWR3 of A27 extend to the adjacent gp120 protomer (Figure 5F), possibly allowing hydrogen bonding or salt bridge formation between the D72 A27 residue and K207 gp120 on the neighboring protomer as judged from their close proximity. Removal of the inserted residues in A27DFWR3 abolishes this interaction, potentially weakening the overall A27-gp120 interactions leading to reduced binding and neutralization. Thus, we show that the incorporation of highly improbable insertions and deletions during vaccination is not simply a byproduct of high SHM, but a selection-driven mechanism to improve engagement of natively glycosylated HIV-1 Env trimers.

DISCUSSION
While germline targeting is a promising vaccination strategy for bNAb induction, it will be challenging to recapitulate how bNAbs naturally develop in some HIV-1-infected people. Specifically, developing VRC01-class NAbs with appropriate breadth and potency requires the initial selection of BCRs with a specific genetic signature, adaptation to glycans surrounding the CD4bs, and high levels of SHM that frequently involve rare indels. 28, 30 We show here that gl-CH31 KI mice are an excellent animal model for testing whether vaccination can recapitulate, and perhaps even provide a shortcut for, the complex sequence of events that can occur during chronic infection.
Eliciting VRC01-class bNAbs will likely require carefully designed immunization regimens. A priming immunogen should expand genetically favorable B cell populations, followed by shaping and/or polishing immunogens that gradually expose these B cells to glycan barriers and increasingly diverse CD4bs epitopes. While some designs focus on activating large numbers of B cells through a high-affinity interaction between naive B cell and immunogens, others aim to impose structural constraints early on, expanding lower numbers of B cells, yet imposing a more stringent initial selection. While priming immunogens aim to engage precursor B cells by removing CD4bs-adjacent glycans, boosting immunogens need to gradually impose structural constraints that guide favorable SHM while maintaining the ability to activate primed B cells. We observed no specific CD4bsdirected serum neutralization signature after priming with GT1.2 and boosting with fully glycosylated BG505 SOSIP. Hence, the affinity gap between GT1.2 and a fully glycosylated native Env may be too large, requiring boosting with immunogens lacking one or more glycans to enable GT1.2-primed B cells to continue VRC01-class development.
Although we observed N276 glycan-coping mechanisms present in bNAbs (i.e., CDRL1 deletions and/or glycine substitutions) ( Figure 3F), the N276 glycan still poses a major hurdle to the development of NAb breadth, possibly indicating a need for earlier introduction of Env immunogens carrying this glycan, either in native complex form or in shorter Man 5 isoforms. 54 However, the frequent observation of N276 glycan-coping mechanisms in our NGS dataset is encouraging and provides proof of concept that adaptation to the N276 glycan by sequential vaccination is possible.
Recent studies have highlighted the possibility of naturally arising VRC01-class bNAb lineages with relatively low SHM (10%-14% in the HC). 62,65 Other than their relatively low SHM levels, they exhibit canonical VRC01-class signatures, such as IGHV1-2*02 use, a CDRH3 W100b motif and classical N276 glycan-coping strategies. One example is BG24, a VRC01-class bNAb with 13.4% SHM at the nucleotide level that evolved a six amino acid CDRL1 deletion despite relatively low SHM. 62 A second is the PCIN63 lineage with 10%-15% SHM, where a GXG motif emerged to accommodate the N276 glycan. 65 Here, we demonstrate that the SHM levels observed in these VRC01-class bNAbs, as well as the extreme selection pressure coinciding with indel development can be reproducibly achieved through sequential vaccination, but requires priming with germline-targeting immunogen GT1.2. A similar study that used eOD-GT8 Article ll OPEN ACCESS as a priming immunogen showed isolation of NAbs that neutralize heterologous N276 glycan-bearing viruses with up to 54% breadth, albeit at low potency. 41 Although the numbers of immunizations and the animal models used are similar, no B cell clones with multi-residue insertions were isolated after eOD-GT8 priming. Hence, priming, shaping and polishing with native-like SOSIP trimers that impose appropriate steric constraints might be particularly favorable for the selection of such insertions. We need to understand how and when CDRH1 and FWR3 insertions arise in our gl-CH31 KI immunization model. For instance, given that our existing regimen can reproducibly induce CDRH1 insertions that potentially offer a shortcut compared with those accumulating multiple single residue changes, 29 it will be key to understand if and how they can be acquired through vaccination earlier. We also need to investigate whether the full nine amino acid insertion found in mature CH31 offers advantages over the four amino acid insertion found in this study. Additionally, gaining insight into the molecular mechanisms involved (i.e., V(D)J recombination-related, SHMassociated, both and/or other) may allow for developing approaches to modulate their promotion during vaccination. Considering the disfavoring of indel formation over base substitutions during SHM, these results indicate that strong selection pressure induced through trimer-based sequential immunizations can reproducibly elicit these rare events, which in turn also suggests it drives the same memory B cells to re-enter germinal centers for further rounds of SHM and affinity maturation.
The eight-dose regimen used here would be extremely hard to implement in humans. However, several factors could simplify the design of a practical vaccine regimen. For instance, are two primes and/or three finishing boosts truly required, or might fewer be needed? Additionally, to further increase the chances of translating these approaches in humans, several innovative methods have shown promise in animal models. For example, adoptive transfer experiments with our gl-CH31 model can be used to better mimic low gl-bNAb frequencies, similar to those typically found in human repertoires. Second, osmotic pumps that release small amounts of antigen over time have been shown to result in more robust T follicular helper cell development, germinal center B cells with increased Env affinity and up to 20-fold higher NAb titers in animal models. 66 The simultaneous presentation of different immunogens on a mosaic NP has been shown to increase breadth in the context of SARS-CoV-2 and influenza vaccination. [67][68][69] This technique may allow multiple germline-targeting phases (i.e., priming-shaping or shaping-polishing) to be triggered by a single immunogen, effectively decreasing the number of immunizations needed to achieve serum breadth. Our current regimen thus holds substantial near-term promise and a strong basis for further iterative pursuit of a truly practical trimer-based strategy, especially given the (in this study) unexplored effects of optimal timing, adjuvanting, and delivery platforms.
Overall, this proof-of-concept study demonstrates that priming with a prefusion-stabilized germline-targeting SOSIP trimer can drive the maturation of VRC01-class Abs, including the selection of multi-residue insertions and the induction of key VRC01-class mutations. We show that these indels are functional, possibly stabilizing the interaction between antigen and antibody, in particular in the context of natively glycosylated trimers. These observations, together with other recent studies showing the elicitation of heterologous NAbs in lower-bar animal models, 33,41 pave the way to develop more feasible, potent, and broad vaccine strategies to elicit anti-HIV-1 bNAbs in humans.

Limitations of this study
Immunization-driven VRC01-class affinity maturation, including functionally important indel formation, represents an important step forward in HIV-1 vaccine research and a conceptually novel avenue in creating universal vaccines to diverse pathogens. However, the gl-CH31 precursor frequency in the relatively low-bar KI mouse model used here is far from the frequency present in a human population. 22,25 Indeed, precursor frequency modulation through adoptive transfer of KI B cells (and/or use of KI mice with fully humanized Ig loci) to achieve near-physiological levels of gl-bNAb precursors will thus prove very informative. Moreover, the total numbers of mice used are suboptimal to draw conclusions from on a population level, even though statistical significance was achieved for many readouts (Figures 2 and  3). Finally, the fact that this first GT1.2 priming vaccine protocol described here through iterative serologic screening does not induce the exact (full-length) nine-amino acid CDRH1 insertion acquired during original human CH30-34 bNAb maturation is also a potential limitation of the study. Nevertheless, it encouraging that the insertions our vaccine protocol produced are in the exact same location in CDRH1 as the one that occurred in CH31 in the patient. Thus, optimization of timing, adjuvanting, and delivery platform modalities, coupled with a deeper basic understanding of vaccine-driven indel formation, should help to improve our current regimen.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

DECLARATION OF INTERESTS
Amsterdam UMC has filed a patent application related to germline-targeting HIV-1 Env trimers.

RESOURCE AVAILABILITY
Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contacts, Dr. Laurent Verkoczy (laurent.verkoczy@absinstitute.org) and Dr. Rogier W. Sanders (r.w.sanders@amsterdamumc.nl).

Materials availability
The MAbs generated in this study will be available under an MTA with Amsterdam UMC.
Data and code availability This paper does not report original code. NGS data have been deposited at NCBI sequence read archive (SRA) under accession numbers 32312762-32312784 and are publicly available as of the date of publication. The crystal structure reported in this manuscript has been deposited at the Protein Data Bank (PDB) under accession number 8E1P. Any additional information is available from the lead contacts upon request.

Mice
The heterozygous gl-CH31 KI (heterozygous knock-in; V H DJ H +/x V k J k +/-) vaccination model was generated on the C57BL/6 CD45.2 + background, based on previously-described Ig locus-directed gene-targeting techniques. 44,70-72 Briefly, gl-CH31 ''HC only'' (i.e. V H DJ H +/+ ) KI mice were first generated by knocking in the published V H DJ H rearrangement of the inferred gl-CH31, via replacement of the mouse J H cluster with a gl-CH31 HC expression cassette (containing the promoter and split leader sequences of the J558 V H family H10, positioned 5 0 of the rearranged gl-CH31 V H DJ H mini-gene segment sequences), and intrabred to achieve homozygosity. In parallel, recombinant ES cells bearing the murine LC kappa locus-targeted inferred gl-CH31 V k J k rearrangement sequence were generated by replacing Jk1 and Jk2 with the gl-CH31 LC expression cassette (comprised of the VOx1 promoter and split leader located 5 0 of the pre-recombined gl-CH31 VkJk rearrangement), and intra-bred to derive homozygous gl-CH31 ''LC only'' i.e. V k J k +/+ KI mice. Finally, homozygous gl-CH31 ''HC only'' and 'LC only'' KI strains were repeatedly inter-crossed until a fully homozygous gl-CH31 KI (V H DJ H +/x V k J k +/+ ) breeding colony was established, and in order to characterize pre-immune/naïve B-cell development and V(D)J repertoire diversity, relative to age and gender-matched wild type C57BL/6 mice ( Figure S1). To generate heterozygous gl-CH31 vaccine cohorts, fully homozygous gl-CH31 base breeders were crossed to wild type C57BL/6 mice. All gl-CH31 KI animals reported in this manuscript were 8-12 weeks of age (either homozygous ones used for naïve/pre-immune developmental and repertoire characterizations or heterozygous cohorts at the start of all vaccine studies), with equal numbers of males and females distributed across all experimental groups. All mice were housed in Duke University (Division of Laboratory and Animal Resources) facilities or the ABS vivarium, both under pathogen-free environments, 12h light/dark cycles at 20-25 C, in accordance with NIH guidelines. All animal procedures performed were approved by Duke University or ABS Institutional Animal Care and Use Committee (IACUC)-approved protocols.
Cell lines HEK293F cells (ThermoFisher) were used to produce recombinant proteins and antibodies as described below, as per the manufacturer's instructions.

REAGENT or RESOURCE SOURCE IDENTIFIER
Adobe Illustrator 2021 Adobe https://www.adobe.com