PBs skewed to dominant, mutated immunoglobulin G lineages post-RTS,S vaccination
We sequenced the messenger RNA of immunoglobulin (Ig)G-expressing PBs isolated from peripheral blood mononuclear cells (PBMCs) of individuals (n = 45) vaccinated with RTS,S in a phase 2a clinical trial22 using our Immune Repertoire Capture® sequencing platform. In this trial, participants either received three full doses of RTS,S/AS01E 1 month apart (012M; n = 15) or two full doses 1 month apart, followed by a smaller (one fifth, “fractional”) dose 6 months later (Fx017M, n = 30). Vaccinees were challenged with malaria in a controlled human malaria infection (CHMI) model after the third dose. A subset received a fourth dose and were challenged a second time with malaria. PBs were isolated from PBMCs collected 7 days post-third (P3D; n = 22,319 PB) and post-fourth doses (P4D; n = 10,629 PB; Supplementary Table 1) prior to CHMI and were used to generate natively paired heavy and light chain IgG sequences. Almost all (99.2%) of the antibody sequences were divergent from inferred germline precursor sequences (Extended Data Fig. 1, see Methods). Consistent with previous malaria studies15,23–26, specific germline heavy and light chain genes and pairings, including IGHV3-30/33, KV1-5, KV3-20, and LV1-40 were observed frequently in the dataset (Extended Data Fig. 2a–c). No significant associations were observed between protection status and multiple IgG sequence and repertoire features examined (Extended Data Fig. 2d–i).
P3D and P4D PBs were grouped into Ig lineages (n = 18,980), defined here as PB sequences that were likely derived from a common progenitor B-cell clone (see Methods). Lineage size ranged from 1–84 (P3D) or 1–93 (P4D) PBs. As PBs have a short half-life in blood27 (reviewed in28,29) and were isolated from a small volume of blood (~10 ml), detection of lineages with ≥2 PBs indicates recent expansion in lymphoid organs. One-fifth of lineages were cellularly expanded and contained at least two PBs with either the same or divergent B-cell nucleotide sequences (19.4%, n = 3,684 lineages, Fig. 1a). Consistent with antigen-driven selection pressure following vaccination, most of the cellularly expanded lineages also showed evidence of clonal expansion (Fig. 1b), a hallmark of affinity maturation. Furthermore, several lineages had clonal representatives that were observed after both the third and fourth immunisations, referred to here as “recalled” lineages (4.1%–26.6% of vaccinee P3D expanded lineages). In addition, when sequences were compared between the vaccinees, we observed that many of these expanded lineages also show evidence of sequence convergence between ≥2 vaccinees (7.3%–46.7% of vaccinee P3D expanded lineages, see Methods). Not surprisingly, lineages with only a single observed PB in P3D repertoires (n = 10,841) had significantly lower rates of convergence (2.0%–13.8%) and recall (1.2%–18.6%) than the expanded lineages (P < 0.0001 and P < 0.001, respectively, Wilcoxon matched-pairs, two-tailed test) and had higher levels of somatic hypermutation (SHM, Extended Data Fig. 3a). Thus, to increase the chances of identifying antibodies derived against RTS,S antigen, we mainly focused subsequent analyses on expanded lineages (Fig. 1b).
We hypothesised that lineages with the largest number of PBs per vaccinee, referred to here as “dominant lineages”, were more likely to target the vaccine, as they had outcompeted other PB lineages for antigen binding and/or T-cell help in lymphoid organs. Thus, for each vaccinee, expanded P3D lineages (9–99 expanded lineages observed per vaccinee) were rank-ordered by size (“rank-size”). The sum of PBs in the lineages of the top four rank-sizes for each vaccinee constituted 17%–100% of the total number of PBs in each vaccinee’s P3D repertoire of expanded lineages and 33% of the PBs among the P3D-expanded lineages from all vaccinees (Fig. 1c). Because this pattern of PB distribution was consistent across protection status and dose regimens (Extended Data Fig. 3b–c), we generated a mAb screening library for in vitro and in vivo characterisation that was biased toward the dominant P3D lineages of both protected and not protected vaccinees.
CSP-reactivity of expanded P3D PBs is associated with lower SHM and lack of protection
A clone from each of 369 unique P3D lineages was chosen, gene synthesised, and recombinantly expressed for testing (see Methods, Extended Data Fig. 4a). This library included almost all (96%) of the largest lineages (rank-size 1) across all vaccinees; approximately half (56%) of the second, third, and fourth rank-size lineages across all vaccinees; a small subset (6.9%) of expanded, sub-dominant lineages (rank-size ≥5), and a few single-PB lineages (0.18% of the 10,841 single-cell lineages). All mAbs were screened in a CSP enzyme-linked immunosorbent assay (ELISA) (Fig 1d, Extended Data Fig 4b), and approximately one-third were screened against the other RTS,S component, hepatitis B surface antigen (HBsAg, Extended Data Fig. 4c). Of the mAbs screened in both assays (n = 130), 52% were reactive to CSP (29/130) or HBsAg (39/130). In total, 38% (139/369) of all mAbs bound to CSP, and binding for an additional 29 mAbs was indeterminate. Of the CSP-reactive mAbs, 73% (102/139) bound peptides from the NANP CR region and 14% (20/139) bound peptides from the C-terminal region (Supplementary Table 2).
Given that expanded lineages were more likely to show evidence of convergence and recall as compared to single PBs, we tested whether those same features were associated with CSP-reactivity. Indeed, mAbs from lineages that show sequence convergence across ≥2 vaccinees were more likely to bind to CSP (54%, 55/102) than clones from lineages that lacked evidence of convergence (31%, 84/267, P = 0.0001, Fisher’s exact, two-sided). Recalled lineages were also more likely to be CSP-reactive (49%, 43/87) as compared to lineages only observed P3D (19%, 16/83 P < 0.0001, Fisher’s exact).
Similar to other reports describing immunisation with whole sporozoites25–28, we found SHM levels of CSP-reactive mAbs were significantly lower than SHM levels of CSP non-reactive mAbs (P < 0.0001, Fig. 1d), and SHM levels of NANP-specific mAbs were lower than SHM levels of C-terminal binding mAbs (P < 0.006, Fig.1d). Consistent with our previous observations about these sequence repertoires16, SHM levels of NANP-binding mAbs were not correlated with vaccinee protection status P3D (P > 0.6, Extended Data Fig. 5b). Further, the percentage of mAbs that were CSP-specific and NANP-specific was surprisingly lower among P3D-protected vaccinees than P3D-non-protected vaccinees (P < 0.0007 for CSP, P < 0.006 for NANP, Fisher’s exact, Fig. 1e–f). This inverse correlation between mAb binding and P3D protection status is also observed when the analysis is restricted to just the mAbs from the most dominant lineages (rank-size 1–4, P < 0.0004, Fig. 1g), as well as when all mAbs, including the 20 from lineages that have only 1 PB, are combined in the analysis (P < 0.0005 Fisher’s exact and P = 0.001 by bootstrap analysis, Fig. 1h). These data suggest that the quality of the CSP-specific antibody repertoire may be more important in driving protection than the overall quantity of circulating, CSP- or repeat-specific PBs.
Sporozoite inhibitory antibodies in P3D PBs are not sufficient for P3D protection
Given this surprising inverse association and the well-reported protective activity of CSP-binding mAbs in both humans8,11 and mice15,18,23,24,30,31, we selected mAbs to advance as potential anti-malaria prophylactics without assuming any correlate of protection. Seventy-seven mAbs (77 unique lineages) were selected that included NANP- and C-terminal-reactive mAbs from protected (n = 26) and not protected (n = 8) vaccinees, and from dominant and sub-dominant lineages of either high or low SHM levels (respectively, ≥20 or <20 nucleotide mutations from germline per antibody, see Methods). As in vitro functional assays have demonstrated limited predictive power for in vivo, anti-malaria activity14,31, we screened for activity using a mouse sporozoite-infection model19,32. Over half of these mAbs (44/77) provided ≥95% inhibition of sporozoite liver burden, and some offered near-complete protection (≥99.9% inhibition). All 44 mAbs bound the NANP-repeat region of CSP and most were derived from the IGHV3-33 germline, while some came from other IGHV3 genes, and one from IGHV1. Thirteen other NANP binding, IGHV3-30/33 mAbs demonstrated limited inhibition of parasite liver burden (80%–95%), and 12 mAbs, including three C-terminal peptide binders, showed minimal, but detectable inhibition (20%–80%, Fig. 2a, Supplementary Table 2).
Roughly a third of the tested mAbs (30%, 23/77) were from not protected vaccinees, including half (7/14) that showed near-complete protection in mice (≥99.9% inhibition, Fig. 2a). These data suggest expression of these inhibitory antibodies by circulating, expanded P3D PB lineages is insufficient to drive protection. For example, the highly effective antibody AB-00031718 is observed in both a protected vaccinee and a not protected vaccinee (Fig 1h, red circles). However, PB expansion levels for this antibody lineage differed between the two vaccinees. In the protected vaccinee, the antibody was a member of the largest PB lineage, while in the not protected vaccinee, the antibody was expressed in the seventh rank-size lineage (respectively, 15.8% versus 1.7% of PB in expanded P3D lineages, and 10% versus 0.9% of all circulating P3D PBs). These data are consistent with the hypothesis that, in addition to the functional activity of an antibody, the number of PBs expressing the antibody may affect protection status by ultimately influencing titre in blood and/or representation in immune memory27,28,33–35.
Inhibitory antibodies from vaccinees bind CSP peptides not present in RTS,S
To explore the developability of these inhibitory mAbs as potential drugs, 35 NANP-repeat-binding lineages were selected for further pharmacology studies from the 52 that demonstrated ≥90% inhibition in the sporozoite-challenge screen. To avoid sequence features that can potentially act as liabilities during the development of mAb drugs, and to survey clones from lineages that have extensive clonal diversity, more than one unique antibody clone was chosen from some (n = 23) lineages. Overall, up to 141 mAbs, from 21 protected vaccinees of both RTS,S dose regimens, representing a range of high and low SHM levels, were tested in binding assays.
Antibodies displayed a broad range of affinities against CSP (KD by surface plasmon resonance [SPR] of 11 pM–9.8 nM, Fig. 2b, Supplementary Table 3). Despite the inverse correlation observed between vaccinee protection status and the percentage of CSP-reactive mAbs observed in the original screening library (Fig. 1e–h), these down-selected, inhibitory mAbs had a significant association between CSP-binding affinity (KD) and SHM levels (P < 0.005, r = –0.26 and P < 0.0001, r = –0.39 for heavy and light chain, respectively, Spearman test), indicating that affinity maturation to CSP occurred following vaccination. These correlations are likely driven by two relationships: binding association rates (kon) to CSP and heavy and light chain SHM levels (Extended Data Fig. 6a–b; Extended Data Table 1) and binding dissociation rates (koff) from CSP and SHM levels in the light chain (P < 0.0005, r = –0.29, Spearman; Extended Data Table 1).
Antibodies were also evaluated for binding to short (12–15 residues) and long (20–24 residues) peptides derived from the varied tetrapeptide-based epitopes (NPNA, NPNV, DPNA)14,15,18,24,26,31,36,37 of the CSP CR and JR (Fig. 2b). Short peptides that were tested included an NPNA-containing major repeat-peptide homologous to epitopes in RTS,S, and two peptides heterologous to RTS,S, a DPNA/NPNV-containing minor repeat-peptide and a DPNA-containing JR peptide. Long peptides tested included an NPNA-containing peptide homologous to RTS,S, and a DPNA/NPNV-containing peptide heterologous to RTS,S. These peptide binding data show correlations between SHM and both kon and koff, with relatively greater coefficients, in some cases, than those seen with CSP (Extended Data Table 1). Specifically, higher levels of SHM are correlated with slower koff to short peptides from both repeat and JR (Fig. 2c–e), while correlations with longer peptides have generally smaller coefficients (Fig. 2f–g, Extended Data Table 1). Indeed, the strongest correlation was observed between SHM and binding rates to the short, homologous peptide even though the long version of the homologous peptide contains more repeats of the same epitope (Extended Data Table 1).
Furthermore, correlations between SHM levels and binding rates to the JR peptide, which is heterologous to epitopes in RTS,S, were stronger than for the long homologous peptide (Extended Data Table 1). These data indicate that B cell receptor maturation of these highly functional mAbs may have been preferentially driven by interactions with short versus long NANP epitopes that benefited maturation to heterologous peptide sequences. In addition, these observations are consistent with reports that some anti-CSP protective mAbs display promiscuous binding across distinct CSP epitopes14,18,24,26,31,36,38, although other reports indicate that such promiscuity may not be required31.
Anti-sporozoite activity correlates with CSP-peptide binding and SHM levels
Seventy mAbs, representing 33 of the 35 protective lineages evaluated in binding studies, were directly compared in an intravenous sporozoite-challenge mouse model to the highly efficacious mAb AB-00031718,19,26,31,37,39–41 in order to prioritise inhibitory, mAbs for development as anti-malaria drugs. Antibodies inhibited 44.1%–97.5% of sporozoite liver burden (47.4%–103.8% of AB-000317 inhibition, Supplementary Table 3). Overall, about half of the mAbs demonstrated inhibition comparable to AB-000317 (n = 32), while the other half demonstrated significantly weaker inhibition (n = 36), and one, AB-000224, showed activity that was superior to AB-000317 (Fig. 3a–b, Supplementary Table 3). Serum concentrations for most mAbs were at least 1000-fold higher than the CSP KD of the respective mAbs (Supplementary Table 3, Fig. 3c), indicating that antibodies demonstrating weak inhibition were not likely due to low levels of circulating antibody. Lineages with at least one mAb that demonstrated activity consistent with AB-000317 were considered for further advancement.
To determine if RTS,S-driven affinity maturation contributed to mAb inhibition, we assessed whether percent inhibition compared to AB-000317 correlated with peptide binding kinetics or SHM levels. Indeed, relative activity was associated with slower koff from CSP (Fig. 3d), with slower koff from the short homologous peptide, NPNA3 (Fig. 3e), and with slower koff from the JR and the other short, heterologous peptide (Fig. 3f–g, Extended Data Table 1). Strikingly, no significant correlations were observed between inhibitory activity and binding kinetics with the long homologous peptide, NANP6 (P > 0.3 [koff]; P > 0.7 [kon], Spearman and Pearson, Extended Data Table 1), despite this peptide being the most representative of both RTS,S and CSP. Taken together, the data evaluating these inhibitory antibodies suggest that while binding to NPNA epitopes may be required31, mutations which favour binding to heterologous peptides may be preferred over mutations that simply improve binding to the longer, homologous NPNA epitopes36,42.
Indeed, affinity maturation via SHM likely underlies the correlations between in vivo function and binding kinetics, as inhibitory activity significantly correlates with heavy and light chain nucleotide and amino acid changes from germline (Fig. 3h–i, Extended Data Table 1). Consistent with this observation, low SHM mAbs were more likely to demonstrate significantly weaker inhibition compared to AB-000317 than mAbs with higher mutational burden (86% [12/14], versus 44% [24/55], P = 0.007; Fisher’s exact, two-sided). Taken together, these correlations between higher SHM levels and binding kinetics to homologous (Fig. 2c, f) and heterologous epitopes (Fig. 2d–e, g), and between higher SHM levels and inhibitory activity (Fig. 3h–i), suggest that affinity maturation to epitopes of RTS,S includes bystander maturation to heterologous epitopes that may be functionally important.
Despite the correlations between SHM levels, inhibitory activity, and koff from CSP and short peptides, some mAbs with high SHM levels are exceptions. In some cases, high SHM mAbs have relatively fast koff, and slow kon, and are comparatively poor inhibitors like many of the low SHM mAbs (Fig. 3j). These antibodies may have resulted from inefficient affinity maturation and/or aberrant selection mechanisms limiting survival in and recall from memory36,43–46 (Fig. 1e–h). In other cases, some high SHM mAbs have relatively fast koff and slow kon to short peptides but are still relatively good inhibitors despite their unfavourable binding kinetics (Fig. 3j). In these latter cases, affinity maturation toward antibody homotypic Fab–Fab interactions, not CSP epitopes, may contribute to the relatively strong activity. Inter-antibody binding events can contribute to anti-CSP-binding potency and increased functional activity25,47, and have been reported for some mAbs described here41,47 (Martin et al., 2022, https://www.biorxiv.org/content/10.1101/2022.09.20.508747v1, in review with Nature Communications). Such homotypic interactions may not be reflected in binding kinetics to short NPNA3 peptides, which due to their short length, cannot sterically accommodate multiple simultaneous binding events 25,47 (Martin et al., 2022, https://www.biorxiv.org/content/10.1101/2022.09.20.508747v1, in review with Nature Communications). Indeed, four mAbs that have relatively fast koff to short peptides, but are comparable to AB-000317 in activity, are from a lineage containing a mAb that binds via Fab–Fab homotypic interactions (AB-00039941, Fig. 3j, red circles) (Martin et al., 2022, https://www.biorxiv.org/content/10.1101/2022.09.20.508747v1, in review with Nature Communications). Overall, the data are consistent with mAb affinity maturation via multiple different modes of binding18,39,41,47, and reveal several mAbs (>30) with activity comparable to that of AB-000317 and the potential to be developed into clinical leads.
Lead antibodies prioritised for development
To identify the most optimal lineage(s) for clinical candidate development, we compared mAbs for pharmacological and developability characteristics. Using the sporozoite liver burden data, we further down-selected 26 mAbs representing 15 lineages for evaluation in the parasitaemia challenge model as an alternate endpoint for assessing in vivo function19,32. This set included AB-000317, AB-000224, 23 other mAbs with liver burden inhibitory activity similar to AB-000317, and one mAb with weaker activity than AB-000317. All except two mAbs were significantly more likely to prevent parasitaemia than the negative control, yet no mAbs were significantly better than AB-000317 (Supplementary Table 3). Serum concentrations for almost all mAbs (25/26) at the time of infection were at least 1000-fold higher than the respective mAb’s KDCSP-SPR (Supplementary Table 3), indicating that mAbs more efficacious than AB-000317 were likely not missed due to low levels of circulating antibody. Overall, the data suggest that AB-000317 has in vivo activity at or near maximal efficacy among this set of lead antibodies. Seven mAbs displayed a trend towards superior protection versus AB-000317 (non-parametric log-rank hazard ratios <1 versus AB-000317, Fig. 3k, Supplementary Table 3). Three of these mAbs belonged to the AB-000224 lineage, the only mAb that demonstrated significantly better activity than AB-000317 in the liver burden model (Fig. 3a–b). Given these functional assessments, AB-000224 was considered the prime lead molecule. Only one other mAb from a separate lineage, AB-007088, demonstrated a similar trend to superiority over AB-000317 in a repeated parasitaemia challenge experiment (Fig. 3l, Supplementary Table 3).
A subset of antibodies (19 mAbs from 14 lineages) evaluated in the parasitaemia model was also compared for drug properties important in developing medicines48, including biophysical characterisation assays relevant to drug stability (i.e., protein conformational and solution colloidal stability, see Methods). Although none of the data indicate any of the leads should be eliminated as potential drugs due to definitive development risks, the prime lead, AB-000224, and its siblings, generally performed less favourably than many other mAbs in several assays (Fig. 4a–b). Thus, we selected AB-007088 as a backup molecule given its more favourable biophysical characteristics (Fig. 4b) and efficacy (Fig. 3k–l).
Clinical candidate engineered for optimised developability
While functional potency is essential for any effective drug, biophysical properties like manufacturability, stability and formulation are equally critical for successful drug development. Because our lead antibodies demonstrated protective in vivo activity comparable to AB-000317, we prioritized improving biophysical stability and cell line manufacturing properties by mutating specific residues in the antibody framework regions per Just – Evotec Biologics’ Abacus™ design platform without impacting mAb activity (Fig. 4c–d, see Methods). For AB-000224 and AB-007088, a total of 17 and 5 clonal variants, respectively, were engineered and tested in the same biophysical and pharmacologic assays used previously.
Importantly, engineered mutations improved both conformational and colloidal stability of many variants, including enhanced thermal stability, solubility, and aggregation profiles during storage (Fig. 4e–h). Most of the variants retained parental mAb binding profiles against a subset of tested peptides (NPNA3 and NVDP2NANP3; Extended Data Table 2). Like the parental mAbs, activity was not significantly different than AB-000317 for the subset that was tested in the parasitaemia challenge model; however, unlike the previous screening result that showed AB-000224 to be more efficacious than AB-000317 in the liver burden model (Fig 3b), average percent inhibition for almost all variants of AB-000224 was comparable to AB-000317 across repeated experiments (Extended Data Fig. 7 and Extended Data Table 3). As sera concentrations of the variants at the time of challenge were consistently lower than those of AB-000317 (Extended Data Table 3), mutations engineered in the variants were unlikely to contribute to a reduction in functional activity compared to the parental mAbs. Variants generated as human IgG1 with an Fc mutation to extend half-life (Xtend49) were used to make a panel of stable transfectant cells. Expression and metabolic data were collected (Fig. 4i–j and Extended Data Fig. 8) to identify the best pools for the generation of a production cell line (see Methods). Optimisation of stability violations in AB-000224 greatly improved production titres (Fig. 4i), which importantly, can reduce cost per dose. Three cell lines from the panel of 22 variants were selected for clone generation. The top clonal cell line from one of the engineered mAbs, MAM01, advanced into production following good manufacturing practices (GMP) to support studies for a clinical development of an anti-malaria drug suitable for use in paediatric populations living in low-middle income countries (LMICs).