Priming HIV-1 broadly neutralizing antibody precursors in human Ig loci transgenic mice

Main Text

It is widely thought that vaccine elicitation of sustained levels of potent broadly neutralizing antibodies (bnAbs) may protect humans against HIV (1, 2). HIV bnAbs from natural infection show extensive somatic mutation (1–9), and this maturation is required for cross-reactivity to diverse isolates: bnAb inferred-germline variants typically show detectable affinity for few if any HIV envelope protein (Env) antigens tested (3–5, 7, 10–21). Most wild-type Env proteins are therefore poor immunogens to prime bnAb responses (3–5, 17, 22, 23). Reliable vaccine initiation of bnAb responses will likely require design or discovery of “germline-targeting” priming immunogens with appreciable affinity for bnAb germline precursors (3–5, 10, 11, 17, 22–29). Furthermore, consistent bnAb priming may only be possible for bnAb precursors present at reasonable frequency in all or most vaccine recipients (27).

Proof of principle that a germline-targeting immunogen can prime relatively rare bnAb-precursor B cells was shown with the eOD-GT8 60 subunit self-assembling nanoparticle (60mer) that targets precursors of CD4-binding-site-directed VRC01-class bnAbs (26). VRC01-class precursors are defined by their use of a heavy chain VH1–2*02 (or *03 or *04) gene and light chains with unusually short complementarity determining region 3 (CDR3) loops of five amino acids (5, 30–32). In a transgenic mouse model expressing the germline-reverted VRC01 heavy chain paired with wild-type mouse light chains (VRC01 gH mouse), in which the frequency of VRC01-class precursor B cells was estimated to be higher than in humans by a factor of as little as ~5, a single immunization with eOD-GT8 60mer activated VRC01-class precursors and generated VRC01-class memory responses in nearly all immunized mice, while control immunogens presenting a native CD4 binding site failed to activate such precursors (26). The eOD-GT8 60mer activated target precursors less robustly in a different VRC01-class inferred-germline heavy chain transgenic mouse (25). Several properties of both mouse models lowered the bar for germline-targeting compared to the challenges confronted in a human: elevated bnAb precursor frequency, reduced competition from other B cell specificities, limited diversity of bnAb precursors owing to their uniform recombined heavy chains, and an affinity advantage conferred on bnAb precursors due to their mature H-CDR3s.

To better model the conditions for initiation of a VRC01-class bnAb response in humans, we investigated immunization with eOD-GT8 60mer in Kymab mice transgenic for the complete un-rearranged human antibody germline gene repertoire (33). Kymab mice contain human heavy chains paired with either human kappa light chains (HK mice) or human lambda light chains (HL mice), or both (HKL mice). To determine the level of difficulty for VRC01-class bnAb priming in Kymab mice, we first measured the frequency of VRC01-class precursors in the above mouse strains. Using B cell sorting methods, we previously detected eOD-GT8-specific VRC01-class precursors at a frequency of 1 in 2.4 million among human naïve B cells expressing kappa light chains (27). Using similar methods to probe a total of approximately 300 million naïve B cells from the spleens and lymph nodes of unimmunized HK and HL mice (N=3 each), we were unable to isolate any VRC01-class B cells, suggesting that the frequency of such B cells was considerably lower than in humans. The frequency of the V_H1–2*02, *03, or *04 alleles among HK B cells was previously measured as 0.9% (33, 34), lower than the frequency in humans (2.9±1.3%) (fig. S1) (35, 36) by a factor of only ~3. Therefore, to explain the reduced frequency of VRC01-class B cells in Kymab mice, we analyzed the light chain variable gene (V_L) usage and CDR3 length distributions from IgM+ or IgG+ B cells for five HK mice and four healthy humans by next-generation sequencing (NGS) (37). The kappa light chain-CDR3 (L-CDR3) length distributions were broadly similar in HK mice and humans (Fig. 1A), but the frequency of 5-amino acid L-CDR3s was lower in HK mice (0.018 ± 0.013%) than in humans (0.95 ± 0.57%) by a factor of approximately 50 (Fig. 1B). The cumulative frequency of Vκ genes used by known VRC01-class bnAbs (IGVK3–20, IGVK1–33 and IGVK3–15 (32, 38)) was 24.0 ± 3.5% in HK mice, similar to the 23.2 ± 4.9% that we measured in humans (Fig. 1C). However, the frequency of 5-amino acid L-CDR3s associated with known VRC01-class kappa chains was reduced by a factor of approximately 300 in HK mice (0.00089 ± 0.00079%) compared to humans (0.27 ± 0.13%) (Fig. 1D). Restricting the analysis to light chains with five or fewer V_L nucleotide mutations produced similar conclusions (fig. S2). Whatever the cause, our data indicate that VRC01-class precursors are less frequent in HK mice than in humans by a factor of 150 to 900.

We previously estimated the true frequency of eOD-GT8-specific VRC01-class precursors as 1 in 400,000 naïve human kappa-chain B cells, accounting for cell sorter and PCR losses (27). Based on this number and the calculations above, we conclude that the frequency of VRC01-class precursors in HK mice is unlikely to be higher than 1 in 60 million (=150×400,000) B cells and might be as low as 1 in 360 million (=900×400,000) B cells. The spleens of HK and HL mice (7–18 weeks of age) contain 50 million B220+ B cells, of which ~60% are mature B cells (33) that are thought prepared to respond to antigen. Allowing for half that number again in the lymph nodes and periphery, we estimate that each mouse contains approximately 75 million B cells, of which 45 million are mature. Thus we expect a very low average frequency of 0.2 (≈75/360) to 1.3 (≈75/60) eOD-GT8-specific VRC01-class precursors per HK mouse at any given time. Modeling the number of precursors per mouse with Poisson distributions predicts that 27–82% of HK mice will have zero precursors and the remainder 1–4 precursors (Fig. 1E). Thus priming VRC01-class responses in HK mice appears substantially more difficult than in humans, as the precursor frequency is lower (by a factor of 150–900) and the number of precursors per individual is lower (by 3,000–30,000 precursors (27)). Frequency analysis in one HL mouse and two HKL mice reached similar conclusions that were only slightly more favorable for VRC01-class priming (table S1).

We conducted immunization experiments to determine if eOD-GT8 60mer could prime rare VRC01-class precursors in Kymab mice. Our first experiment (Fig. 2A), in both HK and HL mice (28 mice each), evaluated differences in antigen dose (20 versus 4 μg), adjuvant formulation and route (Iscomatrix/sub-cutaneous versus Sigma Adjuvant system also known as “Ribi”/intraperitoneal), and time point of analysis (14 days versus 42 days post-immunization). Unimmunized mice (3 HK and 3 HL) and mice immunized with 20 μg non-germline-targeting eOD-17 60mer in Iscomatrix were controls. Serum ELISA showed robust responses to eOD-GT8 (endpoint titers of [1–5]×10⁻⁵ and IC50s of [5–13]×10⁻⁴ at day 42) and lower responses to eOD-GT8 KO (fig. S3), a mutant with reduced binding to VRC01-class antibodies (26), indicating epitope-specific responses (fig. S4) (37). Spleens and lymph nodes from culled mice were processed, stained, and single-cell sorted for IgM-/IgD- memory B cells that bound to eOD-GT8 tetramers but not to eOD-GT8 KO tetramers (27, 37). Only eOD-GT8 60mer-immunized mice showed evidence of epitope-specific (eOD-GT8⁺/eOD-GT8 KO⁻) memory B cells (Fig. 2B). A control sort of eOD-17 60mer-immunized samples using eOD-17/eOD-17-KO probes identified memory B cells reactive with eOD-17 but not eOD-GT8.

We performed a second experiment (Fig. 2A), in HK and HKL mice, with eOD-GT8 60mer, eOD-17 60mer, the native-like trimer BG505 SOSIP (39–44), and eOD-GT8 d41m3 60mer, a variant of eOD-GT8 60mer with disulfide-stabilization and modification of the underlying nanoparticle (fig. S5) the results for which are combined with those of eOD-GT8 60mer as the two were indistinguishable (fig. S6). Although epitope-specific memory B cell frequencies were slightly (factor of ~3) lower in the second experiment, potentially due to sorting with a different eOD-GT8 KO bait (eOD-GT8 KO2 [fig. S3]), comparison of frequencies between groups (Fig. 2C) supported the findings from the first experiment (Fig. 2B).

To determine whether or not the epitope-specific memory B cells generated by eOD-GT8 60mer immunization contained VRC01-class memory B cells, RNA from single eOD-GT8⁺/eOD-GT8 KO⁻ sorted memory B cells was reverse transcribed and IgG variable genes were amplified and sequenced by NGS (37). The two experiments yielded a total of 10,816 wells from which unique heavy and/or light chain nucleotide sequences could be determined, and of these, 3,122 wells contained a heavy and light chain pair (2,526 pairs for eOD-GT8 60mer-immunized mice, 537 pairs for control-immunized mice and 60 pairs for unimmunized mice) (Fig. 3A and tables S2–S4). From these sequences, we identified 28 nucleotide-unique (26 amino acid-unique) VRC01-class heavy-light paired memory responses among 29% (17 of 58) of eOD-GT8 60mer-immunized mice (aggregating across different doses, adjuvants, timepoints, types of mice and the two experiments) (Figs. 3B and S7, table S5). In contrast, we detected no VRC01-class responses from 32 control mice. eOD-GT8 60mer-induced VRC01-class responses were found at approximately similar frequencies in HK, HL, and HKL mice (Fig. 3C).

Whereas only 21% (28/136) of paired sequences with V_H1–2*04 heavy chains had L-CDR3s of 5-amino acids, 88% (28/32) of paired sequences using a 5-amino acid L-CDR3 included a V_H1–2*04 heavy chain (Fig. 3A), suggesting that 5-amino acid L-CDR3s in memory B cells may serve as a proxy for VRC01-like responses. Therefore, we examined the frequency of 5-amino acid L-CDR3s among all paired and unpaired light chain sequences. We identified 62 light chains with 5-amino acid L-CDR3s from 48% (28 of 58) of eOD-GT8 60mer-immunized mice, whereas we found no 5-amino acid L-CDR3s among control mice (Figs. 3A and 3D). eOD-GT8 60mer induced memory B cells with 5-amino acid L-CDR3s at substantial frequencies in all three types of mice (Fig. 3E). All combinations of dose, adjuvant and timepoint produced VRC01-class responses and 5-amino acid L-CDR3 responses (fig. S8). Given the very low frequency of VRC01-class precursors, these results indicate that VRC01-class priming by eOD-GT8 60mer was highly efficient and may have succeeded in all or most mice in which at least one precursor was present.

The VRC01-class antibodies induced by eOD-GT8 60mer shared other characteristic features of VRC01-class bnAbs in addition to the V_H1–2 alleles and 5-amino acid L-CDR3. The L-CDR3 is a key site of affinity maturation in VRC01-class bnAbs (32), and the 28 VRC01-class antibodies showed clear signs of selection toward bnAb sequences in both kappa and lambda L-CDR3 (Figs. 3F). In addition, 21 of 28 VRC01-class pairs had L-CDR1 lengths matching those of the germline V_L genes of known VRC01-class bnAbs (Fig. 3G)(32). Further, 18 of these 28 antibodies used known VRC01-class V_L genes (Fig. 3H), and the H-CDR3 lengths among the VRC01-class pairs (9 to 16 amino acids) were similar to those of known VRC01-class bnAbs (12 to 18 amino acids) (Fig. 3I).

Consistent with a VRC01-class binding mode, all 20 VRC01-class antibodies that we expressed bound to eOD-GT8 but had no detectable affinity for either eOD-GT8 mutant, eOD-GT8 KO or eOD-GT8 KO2 (fig. S9). These VRC01-class antibodies had geometric mean (GM) affinity for eOD-GT8 of 134 nM (geometric standard deviation [GSD], 9.4), a factor of 25 higher than the geometric mean eOD-GT8 affinities of VRC01-class antibodies isolated from naïve human B cells (3.4 μM; GSD, 5.6)(27), possibly due to maturation of the immunogen-induced antibodies (mean±SD mutation levels were 0.8±1.1% in V_H and 1.5±1.1% in V_L). eOD-GT8 60mer-induced VRC01-class antibodies in the VRC01 gH mouse had similar mutation levels (V_H: 1.3±3.2%; V_L:2.1±1.9%) but higher GM affinity for eOD-GT8 by a factor of 22 (GM, 6.0 nM; GSD, 38.7)(26), probably due at least in part to the engineering of eOD-GT8 for ultra-high affinity (9 pM) to germline-reverted VRC01 (27). Thus the Kymab antibody affinities are likely better predictors of the affinities of human responses to a single immunization of eOD-GT8 60mer.

Ultimate elicitation of bnAbs will probably require sequential boosting with more native-like epitope variants to select sufficient bnAb-like mutations (1, 4, 5, 10, 16, 17, 22, 23, 25, 26). Consistent with this expectation and with results of eOD-GT8 60mer priming experiments in knock-in mice (25, 26), the primed VRC01-class antibodies here showed no affinity for the native-like trimer BG505 SOSIP (fig. S10) and no neutralizing activity against the VRC01-sensitive HXB2 HIV strain from which eOD-GT8 was derived (fig. S11). The fact that only ~1% (28/2526) of epitope-specific antibodies were VRC01-class (Fig. 3A) suggests that immuno-focusing strategies may be needed to suppress competing responses during boosting (1, 45), though in humans the higher VRC01-class precursor frequency may mitigate this challenge.

Germline targeting is a promising vaccine strategy, but developing suitable model systems to evaluate targeting of human germline B cells is difficult. Kymab human immunoglobulin loci transgenic mice offer a more stringent and human-like model compared to most knock-in mice. Although Kymab mice under-represent the frequency of VRC01-class precursor B cells compared to humans and possess an average of at most 1.3 such precursors per mouse, the eOD-GT8 60mer still proved capable of priming. The seemingly high targeting efficiency of eOD-GT8 60mer in this mouse model encourages human testing wherein the VRC01-class precursor frequency among B cells and the number of precursors per individual are more favorable for bnAb priming. The results here should also encourage germline targeting for other bnAbs with lower human precursor frequencies.

Supplementary Material