Directed evolution of and structural insights into antibody-mediated disruption of a stable receptor-ligand complex

Omalizumab disruptive potency is modulated by affinity and conformational flexibility

To facilitate omalizumab engineering, we initially assessed the disruptive activities of soluble omalizumab Fab and single chain variable fragment (scFv) proteins (Supplementary Fig. 1b–c). We used a biochemical disruption assay using biotinylated-IgE-Fc_2–4 (bIgE-Fc_2–4) and FcεRIα-conjugated polystyrene beads to evaluate disruptive potency (ID₅₀) (Fig. 1f). In this assay no spontaneous dissociation is observed at low anti-IgE concentrations (Fig. 1f) or with off-target antibody controls (Supplementary Fig. 1d). Incubation with a non-competitive anti-IgE, DARPin E3_53, weakly promotes dissociation of IgE at high concentrations as previously described (Supplementary Fig. 1d)¹⁴. The omalizumab scFv exhibited higher affinity, secondary to a slower dissociation rate (k_d), and a twofold improvement in ID₅₀ for complex disruption as compared to the Fab (Fig. 1g and Supplementary Fig. 1e, Supplementary Tables 1 and 2). However, in these studies disruption efficiency (ID₅₀/K_D) remained unchanged as compared to the more efficient disruptive anti-IgE DARPin E2_79 (Fig. 1k, Supplementary Table 2). These results are consistent with prior studies on the omalizumab scFv and omalizumab Fab variants that showed small improvements in affinity lead to moderate improvements in disruption, although the ID₅₀ of these variants was not calculated to compare efficiencies^15,16. To graphically represent the relationships between affinity (K_D), potency (ID₅₀), and efficiency, we plotted ID₅₀ vs. K_D with lines whose slopes (ID₅₀/K_D) correspond to the efficiencies of the original omalizumab Fab and DARPin E2_79 as benchmarks (Fig. 1g)³. Functional differences between Fabs and scFvs are common, as each format can impose restrictions on the relative orientation of variable heavy (VH) and variable light (VL) domains. To test this possibility, we expressed two additional constructs, a constrained scFv with a disulfide bond at the VH:VL interface (scFv_CC) and a Fab construct with two glycine insertions at the elbow of the Fab heavy and light chains (Fab_H2L2)¹⁷ (Fig. 1h). Consistent with our hypothesis the scFv_CC mutant exhibited a K_D and ID₅₀ similar to the Fab, whereas the Fab_H2L2 construct showed improvements in ID₅₀ and K_D consistent with the scFv (Fig. 1i and Supplementary Fig. 1e, Supplementary Tables 1 and 2). However, the efficiency of disruption (ID₅₀/K_D) remained similar (Fig. 1j and k, Supplementary Table 2). Together these data support the hypothesis that disruptive potency can be modulated by changes in binding kinetics and intrinsic structural features (e.g., conformational flexibility), while larger changes to disruptive efficiency require more substantial changes in binding behavior or molecular structure (e.g., E2_79 vs omalizumab).

Directed evolution of omalizumab variants using a disruptive efficiency screen

The omalizumab scFv and H2L2 variants suggest that antibody affinity and disruptive potency could increase proportionally and track linearly along the same efficiency line (Fig. 2a). However, we hypothesized that directed evolution of disruptive inhibitors could improve or impair affinity and disruptive potency independently. For example improving affinity might not improve disruptive potency (e.g., A → B; Fig. 2a), leading to variants that lie to the upper left of the parental efficiency line. In contrast, variants with improved disruptive efficiency would lie to the lower right of the original efficiency line, with better disruptive potency for a given affinity (A → D; Fig. 2a). Isolating more and less efficient variants could provide insights into sequence, kinetic, and structural features that influence the disruptive mechanism in a given system.

To establish a yeast-based high throughput screening method for selecting more efficient disruptive inhibitors, we considered three potential selection schemes. The first scheme, pure affinity maturation, does not provide any direct selection pressure for disruptive potency or efficiency (Fig. 2b). A second scheme selects variants from a library exposed to ligand:receptor complexes for a fraction of the complex half-life to identify variants with improved disruptive potency and speed (Fig. 2c), as they compete by displacing the receptor. However, this scheme does not control for changes in the disruptive inhibitor binding affinity for the ligand and therefore does not directly select for disruptive efficiency. We therefore devised a third scheme that includes co-staining with a mutant ligand that binds the inhibitor but not the receptor (Fig. 2d). In this scheme, the inactive ligand is titrated to a sub-saturating concentration and stains yeast according to the inhibitor ligand binding affinity, while ligand:receptor complexes report on the disruptive activity of the inhibitor. This dual staining reflects disruption and affinity and could therefore enable direct selection of more efficient (ID₅₀/K_D) variants. Here we used a biotin-labeled IgE-Fc_2–4 as the ligand, a soluble FcεRIα-Ova fusion for the receptor, and a non-FcεRIα binding IgE-Fc_3–4 construct (G335C-IgE-Fc_3–4) as the inactive ligand mutant (Fig. 2e)¹⁸. We have previously shown that omalizumab and E2_79 bind with similar affinity to WT and G335C mutant IgE-Fc_3–4 fragments and both have an epitope confined to the stable Cε3 domain^3,8. Nevertheless it is important to consider that the conformational state of the inactive ligand mutant could favor the selection of antibodies with unanticipated conformational bias.

To validate these staining schemes, we displayed three anti-IgE molecules on S. cerevisiae a weak disruptor (omalizumab scFv), an efficient disruptor (E2_79), and a non-FcεRIα-competing DARPin (E3_53) (Fig. 2f). Each control reagent binds titrations of free IgE or preformed complexes of IgE:FcεRI as expected (Fig. 2g). Staining with the IgE:FcεRI complex alone provided some discrimination of each displayed anti-IgE, however use of the disruptive efficiency stain with IgE:FcεRI complex and the non-FcεRIα binding G335C-IgE-Fc_3–4 significantly improved discrimination of omalizumab and E2_79 (Fig. 2h).

Disruptive inhibitors must bind to intact ligand-receptor complexes briefly prior to disruption. Therefore it is possible that during selections with intact complexes, variants that stabilize complexes or bind complexes non-competitively could also be enriched. Therefore a final “co-binding” assay with the ovalbumin-tagged FcεRIα (FcεRIα-Ova) and a FITC labeled anti-Ovalbumin antibody was used to further discriminate non-competitive binders (E3_53) from competitive and disruptive inhibitors (E2_79 and omalizumab) (Fig. 2i–j).

Using the efficiency stain, we selected omalizumab variants with improved disruptive efficiency and potency. We constructed a series of Error-Prone PCR (EP) omalizumab libraries, and then shuffled the mutations from EP library hits using the staggered extension process (StEP) to generate StEP libraries prior to further rounds of selection¹⁹. All selections were conducted on four libraries named for the mutation technique (EP of StEP) and generation (1 or 2) (Fig. 3a Supplementary Fig. 2).

The first error-prone library (EP1) progressed during four rounds of selection (Fig. 3b), and clones showed improvement by the disruptive efficiency screen (Supplementary Fig. 2b). Subsequent iterations of selections using EP libraries and shuffling were then employed to further improve variants (Supplementary Fig. 2c), and a combination of sequencing data and the efficiency screens isolated improved clones (Supplementary Fig. 2c, d). Clone C02 was selected for further characterization alongside clone 813 as an intermediate comparator.

Omalizumab variants exhibit enhanced affinity and disruptive efficiency

Cloned hits and controls were retransformed into fresh yeast, induced, and stained with the disruptive efficiency stain followed by fixation to reduce sample to sample variation caused by dissociation of bound IgE during staining and data acquisition. In this assay Clone 813 and C02 outperform omalizumab scFv, and C02 appears similar to E2_79 (Fig. 3c). During selections the library (and clone C02) appeared to outperform E2_79 (Supplementary Fig. 2c), however during these experiments fixation was not employed.

Despite selections with intact IgE:FcεRIα complexes, clone 813 bound intact complexes to a similar extent as omalizumab scFv in co-binding assays (Fig. 3d). In contrast C02 showed a modest enhancement of co-binding (Fig. 3d). In multicolor experiments staining for both bound IgE and FcεRIα receptor, each omalizumab variant showed a degree of co-binding signal proportional to the IgE binding signal in 2D plots, suggesting that co-binding can occur at low levels prior to omalizumab mediated displacement (Fig. 3e), consistent with prior reports that omalizumab inefficiently binds intact complexes¹⁵.

As a control for affinity maturation without explicit selection for disruptive efficiency we also produced a high-affinity omalizumab variant (HAE) previously engineered by phage display and taken into clinical trials²⁰. When displayed on yeast the HAE scFv exhibits a weaker disruptive efficiency signal than C02 and looks comparable to 813 in terms of efficiency (Fig. 3f). The HAE control also showed enhanced co-binding to intact bIgE-Fc_2–4:FcεRIα-Ova complexes (Fig. 3g).

To evaluate the success of the yeast-based selections and screening assays purified scFvs were produced (Supplementary Fig. 3a–b) and used in binding and disruption studies. Compared to omalizumab, clone 813 showed a modest improvement in affinity driven primarily by slower dissociation, an almost two-fold improvement in ID₅₀, and a small increase in disruptive efficiency (Fig. 3h and Supplementary Fig. 3c–d, Supplementary Tables 3 and 4). Clone C02 showed a fourfold improvement of affinity with slower dissociation, a tenfold improvement in ID_50, and a five-fold improvement in efficiency, making it almost as efficient as the DARPin E2_79 (Fig. 3h and Supplementary Fig. 3c–d, Supplementary Tables 3 and 4). In contrast HAE scFv dissociated extremely slowly, bound IgE more than 5-fold tighter than C02, yet was less than twice as disruptive and thus exhibited a disruptive efficiency more similar to the parental omalizumab and clone 813 (Fig. 3h and Supplementary Fig. 3c–d, Supplementary Tables 3 and 4). These observations match the ranking predicted in yeast-based assays and C02 and HAE demonstrate that binding kinetics and disruptive potency can be engineered independently. Consistent with their divergent disruptive profiles, C02 and HAE contain two mutated VL CDR1 residues in common (Q27K, Y31S/G), but no overlapping VH mutations, highlighting that each antibody improved its affinity and disruptive potency via distinct paths (Fig. 3i–j).

Structural analysis of high-affinity disruptive omalizumab variants

To identify structural correlates of disruptive potency and efficiency we crystallized native omalizumab scFv, clone 813, clone C02, and clone HAE in complex with a smaller IgE Fc fragment, IgE-Fc_3–4, which contains the full omalizumab epitope (Supplementary Table 5). Prior structural studies suggested that the degree of potential steric overlap between inhibitor and FcεRIα correlated with their disruptive-potency and efficiency²¹. We therefore aligned all antibody:IgE pairs relative to the Cε3 domain to analyze conformational arrangements that could impact steric overlap with FcεRIα. The binding pose of omalizumab is well conserved in CDR proximal regions in all structures, but conformational flexibility is present in regions distal to the binding interface (Fig. 4a). Despite the flexibility across crystal structures we observed no correlation between disruptive potency and the binding pose of individual VH or VL domains, or the combined VH + VL structures relative to the bound IgE Cε3 domain. (Supplementary Fig. 4a). Consistent with this observation the predicted steric overlap with receptor for each scFv in structural alignments was extremely similar (Fig. 4b).

Conformational rearrangements within the IgE-Fc have been extensively studied and are well known to effect receptor binding^22,23,24,25. Therefore we also assessed if scFv variants could modulate “open,” and “closed,” conformations of the IgE-Fc²², as these conformation rearrangements are known to modulate IgE:FcεRIα interactions. This analysis revealed no correlation between the Fc conformational states and disruptive potency (Supplementary Fig. 4b, c). Rearrangements of the Cε2 domains are also known to regulate receptor binding, however these domains are not present in these structures, and we cannot exclude the possibility that these scFvs differentially interact with or alter the orientation of these domains.

The omalizumab IgE footprint was also well conserved across structures, yet the buried surface area of C02 contained a relatively larger VH and smaller VL footprint (Fig. 4c, d). Minor changes were observed near epitope proximal mutations, across scFv copies related by non-crystallographic symmetry (NCS), or in regions adjacent to solvent exposed side chains that could not be modeled (Fig. 4d). Although the antibody footprint was similar, there were substantial differences in the distribution of interface and interface adjacent mutations. C02 accumulated mutations within the heavy chain HCDR3 and forms polar contacts via the H101N mutation and intramolecular hydrogen bonds within the HCDR3 loop via the S100N and R98K mutations (Fig. 4e). C02 also acquired mutations in the HCDR1 loop which introduced a N-linked glycan proximal to the interface, yet the sugar does not appear to form stable contacts with IgE or produce a dramatic reorganization of adjacent residues (Fig. 4e). In contrast, HAE contains a cluster of different mutations within HCDR2 and only one extends to the omalizumab interface (T53K) (Fig. 4e). The remaining HAE HCDR2 mutations are distal to the interface and form intrachain contacts, but surround Y54, a key interface residue found in all omalizumab:IgE complexes.

Within the VL domain, 813, C02, and HAE converged on a Y31S or G substitution that reduces steric bulk proximal to residue 32 (Fig. 4f). This substitution facilitates the formation of intrachain hydrogen bonds in HAE and C02 at VL position D/E 32 and Y36 which form the posterior edge of the binding pocket for R419 of IgE, a residue shown to be critical for omalizumab binding (Supplementary Fig. 4d)⁷. Adjacent to these core mutations, C02 and HAE both converged on an upstream Q27K mutation at the base of the LCDR1, and HAE contains additional mutations in and proximal to the LCDR1 (S28P, M37L) which are not directly in contact with IgE (Fig. 4f). These data demonstrate that alterations that stabilize the LCDR1 loop are critical to affinity and disruptive potency enhancement in two omalizumab variants isolated through distinct selection methods.

Together these studies highlight shared LCDR1 mutations that enhance affinity and demonstrate that small changes to the VH and VL binding footprints and polar bonds help distinguish the efficient disruptive omalizumab variant C02 from HAE, 813, and omalizumab scFv. However, these IgE bound structures do not reveal the relevance of these mutations during disruption. We therefore reasoned that the effect of each mutant might be revealed in structures of the predicted trimolecular omalizumab:IgE:FcεRI complex along the disruption pathway.

A disulfide “locked” IgE-Fc2–4:FcεRIα complex enables the selection of omalizumab variants that trap a disruption intermediate

To study transient intermediates along the disruption pathway, we designed a disulfide engineered IgE-Fc_2–4:FcεRIα “locked” complex. The covalently stabilized locked complex should be resistant to full dissociation and allow trapping of partially disrupted intermediates for structural studies. A favorable disulfide bond was predicted between IgE residue G335 and FcεRIα residue W156²⁶. In the smaller IgE-Fc_3–4 fragment the G335C mutation forms an interchain disulfide bond and traps the IgE-Fc in a closed conformation, but in the IgE-Fc_2–4 fragment the distance between G335 residues is larger and constrained by adjacent Cε2 domains. We therefore hypothesized that when co-expressed with the FcεRIα mutant W156C a fraction of IgE-Fc_2–4 mutant G335C might form a intermolecular IgE:FcεRIα disulfide bond, covalently linking the two proteins proximal to site-two of the asymmetric IgE:FcεRIα complex (Fig. 5a).

Using a two-step affinity and ion-exchange purification scheme a homogenous complex containing a His-tagged FcεRIα and unlabeled IgE-Fc_2–4 was isolated (Supplementary Fig. 5a, b). This species was stable in SDS, yet reducible to monomeric components in the presence of DTT (Supplementary Fig. 5c, d). The locked complex exhibited binding characteristics similar to wild type IgE-Fc_2–4:FcεRIα complexes with robust binding to the non-competitive inhibitor E3_53 and weak binding to omalizumab scFv, Fab, and E2_79 at concentrations ~50–100 fold higher than their respective K_D for free IgE-Fc_2–4 (Fig. 5b). We also assessed the binding affinity of yeast-displayed anti-IgE agents for biotinylated locked-complex as compared to free bIgE-Fc_2–4. In agreement with biolayer interferometry (BLI) studies E3_53 displays similar binding affinity for both the locked-complex and free-IgE, while omalizumab scFv binds free-IgE well, but not the locked-complex (Fig. 5c). The disruptive variants C02 and HAE display an intermediate profile with robust free-IgE binding and enhanced locked-complex binding as compared to omalizumab scFv, further indicating that the locked complex can segregate agents by their disruptive potency (Fig. 5d).

To improve the binding of scFv variants to the locked complex for structural studies, we produced a small shuffled library of C02 and HAE. After five rounds of selection the majority of clones converged on variants employing C02-VH and HAE-VL sequences with sporadic additional mutations (Supplementary Fig. 5e–f), yet clones with the tightest locked-complex binding in yeast (Fig. 5e) shared a novel mutation distal to the binding interface in the E-strand of the VL chain (clone A4:D74G, clone 7:D74H, and clone 16:D74Y) (Fig. 5f and Supplementary Fig. 5f). Unlike the other clones, clone 7 was identical to C02 beyond the D74H mutation, which was independently identified in early selections (clone F04, D74H Supplementary Fig. 2b).

Confirmatory BLI binding studies with scFvs and the locked complex could not be fit by 1:1 binding models, although we had previously hypothesized that only a single epitope would be accessible during the disruption of IgE:FcεRIα complexes. Both C02 and HAE appear to have one high and one low affinity binding site within the locked-complex (56.9 nM–113 nM and 2.86 nM–25.8 nM respectively) (Fig. 5g, Supplementary Table 6). The D74H mutation increases the binding affinity of clone 7 compared to C02 (Fig. 5g) and binding models for all clones with the VL E-strand mutations (A4, 7, and 16) estimate extremely slow dissociation rates in a subset of binding events (Supplementary Fig. 5g, Supplementary Table 6). These measurements indicate that disruptive variants can bind IgE:FcεRIα complexes at low concentrations prior to disruption, and that non-equivalent omalizumab epitopes exist in the intact IgE:FcεRIα complex.

We then tested each clone in yeast for receptor co-binding using wildtype bIgE-Fc_2–4:FcεRIα-ova complexes. While both clones A4 and 16 showed a modest increase in co-binding of FcεRIα-ova compared to C02 or HAE, clone 7 displayed significantly greater co-binding with a phenotype intermediate to C02 and the non-competitive binder E3_53 (Fig. 5h). Consistent with these observations soluble scFvs of clone A4 and 16 showed a disruptive ID₅₀ similar to C02 or HAE, while the ID₅₀ of clone 7 increased five-fold relative to C02 despite displaying a higher affinity for free-IgE (Fig. 5i, Supplementary Fig. 5g, h, and Supplementary Tables 7 and 8). The resulting disruption efficiency of clone 7 is lower than any other omalizumab variant and this decrease in efficiency is caused by a single point mutation outside the antigen binding site. These results suggests that clone 7 not only binds tightly to the locked complex but also stabilizes an intermediate state along the disruption pathway in native protein complexes.

Cryo-EM structure of a partially disrupted antibody:IgE:FcεRI complex

We selected clone 7 for further structural studies given that it varied from C02 by a single amino acid, was capable of disruption, and appeared to stabilize intact complexes in wildtype IgE:FcεRIα binding studies. Stable clone 7:IgE-Fc_2–4(G335C):FcεRIα(W156C) complexes could be detected by SEC (Supplementary Fig. 6a) and the structure of the clone 7-scFv₂:IgE-Fc_2–4(G335C):FcεRIα(W156C) complex was determined by single particle cryo-EM to a nominal resolution of ~7.3 Å (Supplementary Table 9, Fig. 6a) with regions proximal to antibody and receptor interfaces resolved at higher local resolution (<7 Å) (Supplementary Fig. 6c–e). Although size exclusion chromatography suggested a binding stoichiometry between 1 and 2 scFvs per complex, no particle classes containing only one scFv were identified, consistent with the complex binding behavior observed in BLI studies. Density for all domains within the complex was well resolved and even allowed modeling of the core IgE-Fc-glycans at N394 (Fig. 6a), glycans on FcεRIα, and glycans on each scFv (Supplementary Fig. 6g), but was notably worse for Cε2 which had the poorest fit to the density (Supplementary Fig. 6f)

In this disruption-intermediate, clone 7 scFvs occupy both site-1 and site-2 epitopes of the IgE-Fc homodimer in a pseudosymmetric manner. One of the scFvs is oriented with its VL domain adjacent to the IgE-Cε2 domains (site-1), while the other scFv VL domain is adjacent to FcεRIα (site-2). Both of these novel VL interfaces involve the unique D74H mutation in the E-strand of clone 7 (Fig. 6b), providing a structural explanation for the increased binding affinity of clone 7 to the locked complex and enhanced co-binding of wildtype IgE:FcεRIα complexes.

In their native binding poses on IgE, omalizumab and FcεRIα would physically overlap each other, suggesting that one or both molecules must be displaced in this intermediate state. Within the antibody, we observe a distortion of the scFv VH/VL conformation relative to the native-C02 structure. Although the low resolution of the EM structure precludes detailed models of these conformational rearrangements, the density and alignment to existing structures supports the following conclusions: 1. The VH domain binding pose remains relatively unperturbed 2. The VL-CDR1 loop, and the site of the most convergent scFv mutations is displaced from the free-IgE binding pose at both sites (Supplementary Fig. 6h–j). Therefore, the adjacent FcεRIα and Cε2-domains must partially block omalizumab binding and impose distortions on the VH/VL conformation.

Despite rearrangements of the scFv, alignment of the site-2 scFv to the native IgE:FcεRIα structure by the proximal IgE-Cε3 domain demonstrates that steric clashes between the native FcεRIα position would persist (Fig. 6c), indicating that FcεRIα displacement was required for clone 7 binding. We measured FcεRIα displacement from the native binding pose at site-2 or site-1, by aligning the disruption-intermediate structure to the native IgE:FcεRIα structure through the corresponding Cε3 domains (Fig. 6c–e). After alignment we visualized the relative displacement of FcεRIα across models and quantified the displacement using angle_between_domains from Pymol Script Collection (PSICO). Although the low resolution limits a detailed analysis of this alternate conformation, it is clear that FcεRIα undergoes substantial rearrangements to fit large portions of the main chain into the experimental density (Fig. 6d). The entire FcεRIα chain is displaced relative to both Cε3 binding sites in the disruption-intermediate, with more pronounced displacement at site-2 (θ = 16.5˚, displacement = 5.5 Å), where the steric conflicts between the clone 7 scFv and FcεRIα need to be resolved (Movie 1). Additional predicted site-2 clashes occur between FcεRIα N-linked glycans (N42 and N166), which retain similar orientations across all prior IgE:FcεRIα and FcεRIα structures^3,21. Notably, in the disruption-intermediate, density at these glycan sites suggests that they reorient and adopt distinct conformations to accommodate clone 7 binding (Fig. 6f).

Adjacent to the site 1 proximal omalizumab epitope, distal to FcεRIα, the Cε2 domains adopt a conformation not seen in prior IgE-Fc structures to accommodate antibody binding (Supplementary Table 10). Alignment of the disruption-intermediate complex to site-1 of the native IgE:FcεRIα structure reveals that this Cε2 conformation could accommodate the simultaneous binding of FcεRIα and omalizumab without overlap of Cε2 and FcεRIα (Fig. 6g). This is distinct from the Cε2 conformation observed in omalizumab(Xol3):IgE-Fc_2–4 structure, where the Cε2 domains pack between the tips of the extremely open Cε3 domains, overlapping the native FcεRIα position¹⁵ (Fig. 6g). This observation suggests that non-disruptive omalizumab binding could occur at site 1, although antibody binding would block Cε2 dimer interactions with Cε3 and Cε4 domains that are known to stabilize the complex²³.

Correlates of potency and safety in allergic effector cell desensitization

The multiple binding sites observed in the locked complex structure suggests that binding events during antibody-mediated disruption could drive receptor crosslinking and spontaneous activation (anaphylactogenicity). We therefore produced full IgG1 antibodies of the most efficient disruptive variant (C02), a high-affinity variant (HAE), and a variant that was potently disruptive but bound tightly to the locked complex (clone 16). We then tested each antibodies’ ability to spontaneously activate mouse bone marrow-derived mast cells transgenic for human FcεRIα (BMMCs^tg) and isolated primary human basophils (Supplementary Fig. 7a–f). Given the role of VH/VL flexibility in the ID₅₀ and affinity of omalizumab, we also produced H2L2-IgG1 constructs with flexible Gly-Gly linkers at each Fab elbow for C02 and HAE variants to better reflect the flexibility of scFvs used for selections. The IgG constructs displayed a similar trend to scFv variants in bead-based disruption assays, yet the C02_H2L2_IgG variant was modestly improved and showed similar potency to HAE_IgG (Fig. 7a). Similar to omalizumab, most variants did not spontaneously activate either BMMCs^tg or human basophils. However, despite showing robust disruption in bead-based assays, clone 16 induced dose-dependent activation of BMMCs^tg and human basophils (Fig. 7b–c). Notably the modification of HAE_IgG to the HAE_ H2L2_IgG format also led to activation in some basophil donors (Fig. 7c). Given the similar disruption profiles of HAE_IgG to the HAE_ H2L2_IgG in bead-based disruption experiments, this data suggests that even small changes in the balance of binding and disruption events at the cell surface may favor receptor crosslinking and promote cell activation. These data suggest that HAE rests on the verge of cell activation, consistent with sporadic hypersensitivity events that terminated phase 2 clinical trials²⁷ and warrants further investigation in more subjects.

Given that some clones were anaphylactogenic, we measured IgG binding to intact IgE:FcεRIα complexes by BLI (Fig. 7d). In these experiments, a loss in signal can occur following removal of IgE from IgE:FcεRIα complexes. However, simultaneous non-disruptive antibody binding events could produce net-positive or net-neutral BLI binding signals. Strikingly, both anaphylactogenic and non-anaphylactogenic variants show pronounced complex formation over a short timeframe, while omalizumab shows little detectable association. A control off target anti-CoV2 antibody (D10) showed no significant association or dissociation of IgE:FcεRIα complexes on the tip in the same assay, while the non-competitive DARPin E3_53 formed stable complexes with IgE:FcεRIα complexes (Supplementary Fig. 7g). In these BLI assays the stability of the inhibitor binding correlates well with the spontaneous activation profile in BMMCs^tg and human basophils. In particular, the most anaphylactogenic variant, clone 16, has the most prolonged net-positive signal and HAE variants show an intermediate profile. These results suggest that the balance of disruptive versus non-disruptive binding events, and in particular the dwell time on receptor complexes, are critical safety correlates for non-activating disruptive antibodies.

We then used the most potent non-anaphylactogenic variants (HAE_IgG or C02_ H2L2_IgG), to assess their ability to rapidly desensitize BMMCs^tg. Although omalizumab can accelerate IgE dissociation, this effect is minimal over the course of hours at physiological concentrations. In contrast the observed ID₅₀ of HAE_IgG1 and C02_H2L2_IgG1 for stripping IgE from BMMCs^tg falls in the nanomolar range after a 6-h treatment (216 nM, 95%CI [174.2–263.2 nm] and 316 nM, 95%CI [243.4–394.1 nM] respectively) (Fig. 7e). Furthermore, these agents are able to suppress activation of BMMCs^tg with half-maximal inhibition in the mid nanomolar range (550.5 nM, 95%CI [425–642.7 nM] and 624.9 nM, 95%CI [412.8–805.4 nM] respectively) (Fig. 7f).

We also isolated primary human basophils from three grass-allergic donors to measure the inhibitory profile of each antibody in human cells. These experiments confirmed that both agents were capable of completely removing cell surface IgE and suppressing IgE dependent activation in the mid nanomolar range (Fig. 7g, h). Interestingly, C02_H2L2 was consistently more potent than HAE in basophil stripping and inhibition experiments although both exhibited similar function in BMMCs^tg. Furthermore, both HAE and C02_H2L2 significantly desensitized cells at a concentration of 500 nM in 6 h (Fig. 7i). In comparison, omalizumab had little effect on basophil IgE levels and signaling even at the highest concentrations studied.