Allosteric mechanism of action of the therapeutic anti-IgE antibody omalizumab

Immunoglobulin E and its interactions with receptors FcϵRI and CD23 play a central role in allergic disease. Omalizumab, a clinically approved therapeutic antibody, inhibits the interaction between IgE and FcϵRI, preventing mast cell and basophil activation, and blocks IgE binding to CD23 on B cells and antigen-presenting cells. We solved the crystal structure of the complex between an omalizumab-derived Fab and IgE-Fc, with one Fab bound to each Cϵ3 domain. Free IgE-Fc adopts an acutely bent structure, but in the complex it is only partially bent, with large-scale conformational changes in the Cϵ3 domains that inhibit the interaction with FcϵRI. CD23 binding is inhibited sterically due to overlapping binding sites on each Cϵ3 domain. Studies of omalizumab Fab binding in solution demonstrate the allosteric basis for FcϵRI inhibition and, together with the structure, reveal how omalizumab may accelerate dissociation of receptor-bound IgE from FcϵRI, exploiting the intrinsic flexibility and allosteric potential of IgE.

Immunoglobulin E (IgE) antibodies play a crucial role in allergic disease, binding to allergens through their Fab arms and expressing their effector functions by binding to receptors for the Fc region (1). The two principal IgE receptors are Fc⑀RI and CD23/Fc⑀RII, commonly referred to as the high-and low-affinity receptors, respectively. On mast cells and basophils, IgE binds to Fc⑀RI so tightly (K D Ϸ10 Ϫ10 M) that such cells are sensitized with pre-bound IgE, requiring only the presence of an allergen to cross-link IgE/Fc⑀RI complexes and elicit an immediate reaction. CD23 is a homotrimer, and thus the intrinsically lower affinity of each IgE-binding C-type lectin-like "head" domain (K D Ϸ10 Ϫ7 M) can be enhanced by an avidity effect when binding to aggregated IgE in immune complexes, nearly matching that of Fc⑀RI for IgE (2). CD23 expressed on B cells is involved in IgE regulation, and expression on airway and gut epithelial cells mediates transcytosis of IgE/allergen complexes (1,2). Fc⑀RI and CD23 are also both expressed on a range of antigen-presenting cells. Thus IgE-receptor interactions are involved in multiple aspects of the allergic response, and IgE is a long-standing target for therapeutic intervention (3).
The Fc region of IgE comprises a disulfide-linked dimer of three domains: C⑀2, C⑀3, and C⑀4. Early FRET studies of a chimeric IgE (4,5), and X-ray solution scattering studies of IgE-Fc (6), indicated a compact, bent structure, and the crystal structure of IgE-Fc later revealed an acutely and asymmetrically bent conformation, with the (C⑀2) 2 domain pair folded back onto the C⑀3 and C⑀4 domains (7). The bend, defined as the angle between the local 2-fold axis of the (C⑀2) 2 domain pair and that of Fc⑀3-4 (the region comprising only the C⑀3 and C⑀4 domains), was found to become even more acute in the crystal structure of IgE-Fc bound to sFc⑀RI␣, the soluble extracellular domains of the IgE-binding ␣-chain of the receptor (8). FRET studies with N-and C-terminally labeled IgE-Fc confirmed this enhanced bend upon sFc⑀RI␣ binding (9).
The Fc⑀RI-binding site spans both C⑀3 domains in the C⑀2proximal region (8,10), although the C⑀2 domain is not directly involved; the engagement of both chains accounts for the 1:1 binding stoichiometry. In contrast, two CD23 molecules bind to IgE-Fc, one in each chain, and at the other C⑀4-proximal end of the C⑀3 domain (11)(12)(13)(14). CD23 binding also causes a conformational change in IgE-Fc (14), but not one that significantly affects the bend (9). However, the relatively "closed" disposition of the C⑀3 domains in the complex with the soluble head domain of CD23 (sCD23), compared with free IgE-Fc, is incompatible with the more "open" arrangement of these domains that is required for Fc⑀RI binding. This partly explains the mutual exclusion of Fc⑀RI and CD23 binding (11,12), although other factors such as local conformational changes and modifications of conformational dynamics (15) also likely contribute to the allosteric communication between the two receptorbinding sites (2).
A more extreme degree of flexibility in IgE-Fc was recently discovered through studies of a complex with an anti-IgE-Fc Fab, termed a⑀Fab (16). Two a⑀Fab molecules bind to IgE-Fc in a symmetrical manner, one on each C⑀3 domain, trapping a fully extended conformation in which the local 2-fold axes of the (C⑀2) 2 domains and Fc⑀3-4 region are virtually coincident. Analysis of the complex formation in solution, together with molecular dynamics simulations of free IgE-Fc, suggests that the (C⑀2) 2 domain pair could "flip" over from one side of the Fc⑀3-4 region to the other (16). The IgE-Fc conformation stabilized by this anti-IgE antibody is incompatible with Fc⑀RI binding, explaining its inhibitory activity (16).
Omalizumab is an anti-IgE monoclonal IgG1 antibody that is approved for therapeutic use (Xolair, Novartis) (17). It binds to free IgE and inhibits both Fc⑀RI and CD23 binding. The site of binding had been mapped to the C⑀3 domain by peptide inhibition and molecular modeling and was recently confirmed by a crystal structure (18 -20). Recently, an inhibitor was discovered that actively disrupted preformed IgE/Fc⑀RI complexes: a Designed Ankyrin Repeat Protein (DARPin) was found to bind to the C⑀3 domain of receptor-bound IgE and accelerate its dissociation from Fc⑀RI (21). The crystal structure of the 2:1 complex of this DARPin (DARPin E2_79) with an Fc⑀3-4 molecule constrained by an engineered disulfide bond (G335C), which artificially locks the C⑀3 domains into a closed conformation, revealed the nature and location of the binding site but left its mechanism of action unclear. It was subsequently reported that omalizumab could also facilitate dissociation of Fc⑀RI-bound IgE, although only at very high concentrations that were substantially greater than those utilized therapeutically (22,23). Omalizumab binding to FRET-labeled IgE-Fc indicated a slight degree of unbending (9) and the potential for allosteric rather than direct inhibition of Fc⑀RI binding. The recent crystal structure of an omalizumab Fab complex is with the same Fc⑀3-4 molecule present in the DARPin complex (20); this constrained Fc⑀3-4 construct lacks the C⑀2 domains and thus cannot report on unbending or other conformational changes.
We report here the crystal structure of the complex between IgE-Fc and a Fab derived from omalizumab. The structure of the complex reveals substantial conformational changes in IgE-Fc, revealing the mechanism of action of omalizumab, both for receptor inhibition and accelerated dissociation of IgE from Fc⑀RI. Solution studies demonstrate that these mechanisms exploit the intrinsic flexibility of IgE.

Results
Despite extensive efforts, crystallization trials for IgE-Fc in complex with the omalizumab Fab resulted in selective crystallization of the Fab fragment only. Others have reported a similar failure to crystallize the complex with IgE-Fc (24). The recently reported structure for the omalizumab Fab complex (20) is with an Fc⑀3-4 molecule that contains a G335C mutation; this mutation artificially locks the C⑀3 domains into a closed conformation. We designed a Fab, derived from omalizumab, with three point mutations, two in the V L domain framework region (S81R and Q83R) and one in the C domain (L158P) (supplemental Fig. S1), with the purpose of disrupting favorable crystal contacts observed in the omalizumab Fab crystal structure. We term this omalizumab-derived Fab FabXol3.

Overall structure of the FabXol3/IgE-Fc complex
We determined the crystal structure of the complex between IgE-Fc and FabXol3 to 3.7 Å resolution ( Fig. 1 and supplemental Movie S1). Two FabXol3 molecules (Fab 1 and Fab 2 ) bind to an asymmetric, partially bent IgE-Fc molecule, and the Fab engages one edge of the exposed face of each C⑀3 domain (Fig.  1). Fab 1 engages the C⑀3 domain of IgE-Fc chain B, whereas    Fig. 2A), which is absent in the Fc⑀3-4 molecule (Fig. 2B), but also shows that the C⑀3 domains adopt a markedly open conformation (Fig. 2, C and E), one that cannot be adopted by the disulfide-bonded C⑀3 domains in the Fc⑀3-4 complex (Fig. 2, D and F). The bending of the (C⑀2) 2 domain pair and the opening of the C⑀3 domains are described in detail below.

Interface between IgE-Fc and FabXol3
Each FabXol3 molecule engages one edge of the exposed face of the C⑀3 domain (C, CЈ, F, and G strands and base of the Fc⑀RI receptor-binding FG loop). The interface with IgE-Fc is similar to that reported for the constrained Fc⑀3-4 molecule (20). Both the heavy and light chain of FabXol3 are involved, the former contributing ϳ60% to an interface area of ϳ715 Å 2 (Figs. 1 and 3A and supplemental Movie S1).
Arg-419 also plays a key role in the interaction with the FabXol3 light chain (Fig. 3A). Arg-419 is within hydrogenbonding distance of the Tyr-31 (CDRL1) and Asp-32 (CDRL1) main chain carbonyl oxygen atoms, in addition to contacting the Asp-32, Asp-34, and Tyr-36 side chains (forming a hydrogen bond with the Tyr-36 hydroxyl group). Asp-32 also forms van der Waals interactions with Thr-373 and Thr-421 (C⑀3). By contrast, only two CDRL2 residues contribute to the interface, Tyr-53 (CDRL2) contacts Gln-417 (C⑀3), and both Tyr-53 and Tyr-57 form van der Waals interactions with Met-430 (C⑀3); Tyr-57 also forms a hydrogen bond with the Met-430 backbone. As for the heavy chain interaction, there are slight differences in the light chain contacts for Fab 1 and Fab 2 .

Comparison of the FabXol3 interface with other anti-IgE complexes
The binding sites on the C⑀3 domain for FabXol3 and the recently described DARPin E2_79 (21) overlap ( Fig. 3B) and are of similar size at ϳ715 and ϳ753 Å 2 , respectively. The C⑀3 domain residues shared between the two interfaces include Ser-375-Gly-379, Gln-417, Arg-419, Arg-427, and Met-430, but although FabXol3 forms more intimate contacts with the receptor-binding C⑀3 FG loop, the DARPin E2_79 interface extends in the opposite direction to include the C⑀3-4 domain linker.
The overlapping binding sites of FabXol3 and DARPin E2_79 differ markedly from the interface recently described for the anti-IgE-Fc a⑀Fab, which captured IgE-Fc in a fully extended conformation ( Fig. 4) (16). Not only is the a⑀Fab interface area approximately double that of FabXol3 and DARPin E2_79, at ϳ1400 Å 2 , but a⑀Fab engages IgE-Fc at a site centered on Arg-393 in C⑀3, and it also contacts residues in the C⑀2 domain and the C⑀2-C⑀3 linker (16). The crystal structure of another anti-IgE antibody Fab, MEDI4212, in a 2:1 complex with Fc⑀3-4 reveals yet another site for antibody engagement within the C⑀3 domain, this one involving the N-linked oligosaccharide moiety at Asn-394 (25).

IgE-Fc adopts a partially bent conformation when bound to FabXol3
IgE-Fc is predominantly bent in solution (4 -6, 9, 26 -28), and the crystal structure for free IgE-Fc revealed an acutely bent (62°) asymmetric conformation, in which the (C⑀2) 2 domain pair folded back onto the C⑀3 and C⑀4 domains (Fig. 4, A and B), the C⑀2 domain of one chain (chain B) contacting the C⑀4 domain of the other (chain A) (7,8). IgE-Fc becomes even more acutely bent (54°) upon Fc⑀RI␣ engagement (8,9), and the associated conformational changes involve rotation of the C⑀3 domain of chain A together with the (C⑀2) 2 domain pair, as a rigid unit, away from the C⑀3 domain of chain B (8).
In contrast to the a⑀Fab complex, in which IgE-Fc adopts a fully extended, linear conformation (16), IgE-Fc adopts a partially bent conformation in the FabXol3 complex (Figs. 1C and 4, C-F, and Movies S1 and S2). The site to which Fab 1 binds is exposed in free, acutely bent IgE-Fc, but further unbending of IgE-Fc, to just over 90°, is required to render the site occupied by Fab 2 accessible. This unbending of IgE-Fc in the FabXol3 complex is associated with opening of both C⑀3 domains to create an almost symmetrical Fc⑀3-4 region (Figs. 1B and 4, C and D).
In a recent molecular dynamics simulation exploring unbending of IgE-Fc to an extended structure, it was found that although the acutely bent conformation observed in the crystal structure of free IgE-Fc occupied the lowest energy basin, another distinct and well defined energy basin, corresponding to partially bent IgE-Fc conformations, was observed (16). The partially bent conformation adopted by IgE-Fc in the FabXol3/ IgE-Fc complex occupies this particular energy basin (supplemental Fig. S2).
To test whether we could observe these conformational changes in solution, we performed intramolecular FRET measurements with IgE-Fc labeled with donor and acceptor fluoro-  (21) bind to an overlapping interface on the C⑀3 domain. IgE-Fc residues, which only form part of the FabXol3 interface, are colored orange, and those that only form part of the DARPin E2_79 interface, which includes part of the C⑀3-C⑀4 linker, are colored in blue. IgE-Fc residues colored in pink, which include Arg-419 and Met-430, are common to both FabXol3 and DARPin E2_79 interfaces. phores in the C⑀2 domains and at the C termini, respectively. We have previously shown that binding of a⑀Fab results in a nearly fully extended IgE-Fc structure (16); data here indicate a Ͼ90% reduction of the FRET signal (E max ϭ 91%, IC 50 ϭ ϳ1 M) when two molecules of a⑀Fab are bound to IgE-Fc (Fig. 5). In contrast, FabXol3, despite its higher affinity, shows only about 50% reduction of the intramolecular FRET signal under saturating conditions (E max ϭ 52%, IC 50 ϭ ϳ7 nM). These experiments support the hypothesis that binding of FabXol3 induces large-scale conformational changes in IgE-Fc, consistent with a partially bent conformation.

C⑀3 domains adopt a markedly open conformation in the FabXol3/IgE-Fc complex
In crystal structures of IgE-Fc and the Fc⑀3-4 sub-fragment, the C⑀3 domains adopt a range of different orientations (7,8,10,11,13,14,16,25,29), a property associated with allosteric regulation of IgE binding to its two principal receptors, Fc⑀RI and CD23 (8,(11)(12)(13)(14). Both the distance between the C⑀3 domains and their positions with respect to the C⑀4 domains have been used to describe the variety of conformations observed for the Fc⑀3-4 region (29) (a full description for these measurements is provided in supplemental data). In the FabXol3/IgE-Fc complex, the C⑀3 domains are positioned further away from one another than in any other crystal structure containing IgE-Fc or Fc⑀3-4 and thus adopt the most open conformation observed thus far (Figs. 1B and 4, C and D, and supplemental Movie S2); this conformation is significantly more open than the conformation for Fc⑀RI-bound IgE-Fc (supplemental Fig. S3).

Effect of FabXol3 on Fc⑀RI and CD23 receptor binding
Omalizumab inhibits not only the interaction between IgE-Fc and Fc⑀RI, but also the interaction between IgE-Fc and CD23 (30). Consistent with the latter, superposition on the C⑀3 domains from the FabXol3/IgE-Fc structure, and the previously reported structure of CD23 in complex with an Fc⑀3-4 molecule (11), reveals steric clashes between FabXol3 and CD23 at both sites of CD23 engagement on the C⑀3 domain. Furthermore, C⑀3 domain residues Arg-376, Ser-378, and Lys-380 are involved in both FabXol3 and CD23 binding (Fig. 6, A and B) (11,13,14,31). Thus, omalizumab inhibits CD23 binding by orthosteric blocking.
In contrast to CD23 binding to IgE, Fc⑀RI␣ binds across both C⑀3 domains. However, in the FabXol3/IgE-Fc complex, the C⑀3 domains adopt a conformation that is more open than in Fc⑀RI-bound IgE-Fc, which precludes simultaneous engagement of both chains (Fig. 6, C-E). Moreover, superposition of the FabXol3/IgE-Fc and sFc⑀RI␣/IgE-Fc (8) complexes reveals potential steric clashes; for example, FabXol3 would clash with the (C⑀2) 2 domain pair from the acutely bent conformation found in Fc⑀RI␣-bound IgE-Fc.
Also in contrast to CD23, the binding sites for omalizumab and Fc⑀RI␣ do not actually overlap, although FabXol3 CDRL1 residues are positioned immediately adjacent to the Fc⑀RI␣binding C⑀3 domain FG loop. This loop, in chain B, contributes to a hydrophobic "proline sandwich" interaction, in which Pro-426 in C⑀3 packs between two tryptophan residues of Fc⑀RI␣ (Fig. 6E). Asp-32 (CDRL1) contacts Thr-421; Gly-33 (CDRL1) contacts Pro-426, Arg-427, and Ala-428; and Asp-34 (CDRL1) contacts Arg-427 and Ala-428. These interactions alter the position of the C⑀3 domain FG loop and would further compromise the binding of IgE to Fc⑀RI.
Thus, omalizumab binding stabilizes a conformation of IgE-Fc, which is incompatible with Fc⑀RI binding.

Interaction of FabXol3 with IgE-Fc in solution
Recently, binding of omalizumab to Fc⑀RI␣-bound Fc⑀3-4 has been reported (22), although it is difficult to see how omalizumab might be able to engage Fc⑀RI-bound IgE based on the static crystal structures of IgE-Fc in complex with sFc⑀RI␣ (8) and FabXol3. We therefore studied the solution state binding of FabXol3 to IgE-Fc, and we characterized the interaction between FabXol3 and the IgE-Fc/Fc⑀RI complex. Our results provide insights into the mechanism of action of omalizumab.
We characterized the IgE-Fc/FabXol3 interaction in two different ways, either by directly immobilizing FabXol3, FabXol (omalizumab Fab), or intact omalizumab on an SPR 5 sensor surface and binding IgE-Fc, or by binding FabXol3 to Histagged captured IgE-Fc. The binding characteristics of IgE-Fc to FabXol3, FabXol, and omalizumab were compared (Fig. 7, A-C). Not surprisingly, in competition binding experiments, all three molecules competed for the same binding sites and showed broadly similar binding affinities; the FabXol3 construct demonstrates slightly higher affinity compared with FabXol and intact omalizumab (Fig. 7, A-C). Consistent with the crystal structure, two FabXol3 molecules bind to IgE-Fc; the binding is clearly biphasic with a high-affinity (ϳ1 nM) interaction observed at low ligand concentrations and a second (weaker) binding site (ϳ30 nM) observed at higher concentrations (Fig. 7D). It might be speculated that the higher affinity interaction corresponds to the binding of Fab 1 , which would have unimpeded access to a bent IgE-Fc molecule, whereas the lower affinity and slower on-rate corresponds to Fab 2 , but we cannot be definitive about this. 5 The abbreviations used are: SPR, surface plasmon resonance; TR-FRET, time resolved-FRET. A sandwich-style SPR experiment allowed the two FabXol3binding sites to be observed and characterized separately. Using this approach, FabXol3 was covalently immobilized on a sensor surface, and IgE-Fc was flowed over this surface. At low concentrations, under these conditions, the high-affinity site dominates the interaction, and the binding curves can be described by monophasic interaction kinetics (K D ϳ1 nM, k on ϳ1.2 ϫ 10 6 M Ϫ1 s Ϫ1 , and k off ϳ 8 ϫ 10 Ϫ4 s Ϫ1 ). This 1:1 FabXol3/IgE-Fc complex, captured on the SPR biosensor surface, could then be used to measure the binding of the second FabXol3 molecule, the binding of which is significantly weaker (K D ϳ30 nM, k on ϳ 2 ϫ 10 5 M Ϫ1 s Ϫ1 , and k off ϳ 6 ϫ 10 Ϫ3 s Ϫ1 ) than the first (Fig. 7E). Again, the slower association rate constant measured for the second (weaker) interaction would be consistent with the Fab binding to the less accessible of the two binding sites, i.e. the Fab 2 site in the crystal structure.
In characterizing the binding of the two different FabXol3 molecules, we observed that binding of the second FabXol3 molecule destabilized the 1:1 FabXol3/IgE-Fc complex; this destabilization of a pre-formed complex is the same phenomenon of accelerated dissociation that has been seen in IgE in relation to Fc⑀RI binding (21,22). We also saw that the FabXol3-mediated accelerated dissociation of the FabXol3/ IgE-Fc complex was highly temperature-dependent, with essentially no FabXol3-mediated accelerated dissociation occurring at 5°C but marked accelerated dissociation occurring at 35°C (Fig. 8). Because of their physically distal binding sites and the strong temperature dependence of the phenomenon, the ability of the second FabXol3 molecule to mediate accelerated dissociation of the 1:1 FabXol3/IgE-Fc complex must be an allosterically mediated process.

Competition between the FabXol3-and Fc⑀RI␣-binding sites and the formation of a FabXol3/IgE-Fc/Fc⑀RI␣ complex
We next investigated the capacity of FabXol3 to affect the interaction between IgE-Fc and Fc⑀RI␣. In solution competi- tion binding experiments, increasing concentrations of FabXol3 inhibited binding of IgE-Fc to Fc⑀RI␣ (Fig. 9A). Mechanistically, FabXol3 affects both the number of available binding sites (B max ) and the apparent K D value of the IgE-Fc/Fc⑀RI␣ interaction. Reduction in B max values is indicative of an allosteric inhibitory process, and a decrease in the apparent affinity of the interaction is most commonly associated with direct competition for a shared binding site (i.e. orthosteric inhibition) but can also be seen for some allosteric inhibitors (32). Although we cannot rule out an orthosteric contribution, considering the lack of overlap between the binding sites observed in the crystal structures, it is likely that FabXol3 inhibits IgE-Fc binding to Fc⑀RI using primarily allosteric mechanisms.
Competition between the omalizumab-and Fc⑀RI␣-binding sites has been described in many publications but was generally interpreted as direct competition between binding sites that were presumed to be identical, or at least overlapping. This interpretation was often used to explain why omalizumab cannot bind to IgE-Fc⑀RI complexes on cells. We observed, however, that FabXol3 can indeed bind, and with high affinity, to IgE-Fc that is pre-bound to Fc⑀RI␣ to form a trimolecular complex (Fig. 9B, inset). The data indicate that although the binding of IgE-Fc to Fc⑀RI␣ did not significantly change the affinity of FabXol3 for IgE-Fc, it did markedly change the number of available binding sites for FabXol3 in the population of Fc⑀RI␣bound IgE-Fc molecules. We compared the K D and B max binding values for an IgE-Fc molecule captured by an anti-His tag antibody with one captured by sFc⑀RI␣, and we found that Fc⑀RI␣-bound IgE-Fc had less than 10% of the FabXol3-binding sites compared with the His tag captured IgE-Fc, which, as expected, showed binding levels consistent with 2:1 stoichiometry (Fig. 9B). Therefore, it is not that omalizumab does not bind to mast cell-bound IgE because the Fc⑀RI␣ and omalizumab-binding sites overlap, or because of steric clashes between two ligands bound to adjacent sites (20). Instead Fc⑀RI␣ acts on IgE-Fc allosterically, changing a dynamic equilibrium of different IgE-Fc conformations, resulting in a substantially reduced number of omalizumab-binding sites in a population of Fc⑀RI␣-bound IgE-Fc molecules. Kim et al. (21) reported that DARPin E2_79 could accelerate the disassembly of pre-formed complexes of IgE/Fc⑀RI. Following up on this observation, Eggel et al. (22) later showed that omalizumab at high concentrations could also promote disso- ciation of IgE from Fc⑀RI. We have found that when FabXol3 binds to the IgE-Fc/sFc⑀RI␣ complex, it accelerates the dissociation of IgE-Fc from sFc⑀RI␣ (Fig. 9C) and, furthermore, that FabXol3 does this more efficiently than FabXol, and much more efficiently than intact omalizumab (Fig. 9D). One Fab engages the IgE-Fc/sFc⑀RI␣ complex but does not accelerate the dissociation of IgE-Fc from sFc⑀RI␣. Strikingly, it appears that accelerated dissociation occurs only after occupancy of the second binding site (i.e. the low-affinity site). The (FabXol3) 2 / IgE-Fc/sFc⑀RI␣ tetramolecular complex must alter the energy landscape of IgE-Fc in such a way as to markedly reduce the energy barrier for IgE-Fc/sFc⑀RI␣ dissociation, resulting in a rapid dissociation of this otherwise very stable complex.

Discussion
We report the structure, at 3.7 Å resolution, of the complex between IgE-Fc and a Fab fragment derived from the therapeutic anti-IgE antibody omalizumab; we call this Fab fragment, which contains three point mutations in framework regions distal to the antigen-binding site, FabXol3. The structure reveals two FabXol3 molecules in complex with IgE-Fc, one bound to each of the two C⑀3 domains, and provides an explanation for the ability of omalizumab to inhibit the binding of IgE to both Fc⑀RI and CD23. IgE-Fc is also found to adopt a partially bent conformation in the FabXol3 complex, consistent with intramolecular FRET measurements in solution.
Recently, our understanding of the conformational flexibility of IgE-Fc was profoundly enhanced when we solved the structure of a fully extended conformation, captured in a complex with an anti-IgE-Fc Fab (a⑀Fab) (16). A molecular dynamics simulation, exploring IgE-Fc unbending from the acutely bent to the extended conformation, revealed energy basins corresponding to partially bent conformations. The FabXol3/IgE-Fc complex reported here, in which the bend between the (C⑀2) 2 domain pair and the Fc⑀3-4 domains is ϳ90°, corresponds to a distinct energy basin in this simulation (16) (supplemental Fig. S2) and is consistent with our intramolecular FRET measurements (Fig.  5). Intriguingly, the location of the FabXol3-binding site would not preclude further unbending to the fully extended conformation (supplemental Movie S3), and it is therefore possible that IgE-Fc can undergo further substantial changes in conformation even when in complex with omalizumab.
In addition to the bending of the (C⑀2) 2 domain pair relative to the C⑀3 and C⑀4 domains, the various IgE-Fc, Fc⑀3-4, and receptor complex structures have demonstrated that the C⑀3 domains can adopt a range of relative orientations, from closed to open (7,8,10,11,13,14,16,25,29). Opening and closing of the C⑀3 domains contributes to the allosteric regulation of receptor binding in IgE-Fc (11,12); in the CD23 complex they are relatively closed (11,13,14), and in the Fc⑀RI complex they are more open (8,10). Comparison of the structures of the CD23/Fc⑀3-4 and FabXol3/IgE-Fc complexes shows that the CD23 and omalizumab sites overlap, and competition binding experiments indicate that inhibition of IgE binding to CD23 by omalizumab is straightforwardly orthosteric.
However, inhibition of Fc⑀RI binding is mechanistically different. In the FabXol3 complex, the C⑀3 domains adopt a more open conformation than seen in any previous structure, so much so that the two sub-sites of interaction between IgE-Fc and Fc⑀RI, one involving each C⑀3 domain, cannot engage simultaneously. In addition to large scale domain motions, local conformational changes induced by FabXol3 binding, such as those in the Fc⑀RI-binding FG loop, may also contribute to this inhibition. Although there is a possibility of steric clashes if FabXol3 and Fc⑀RI␣ bind simultaneously to IgE-Fc, the crystal structure of the FabXol3/IgE-Fc complex demonstrates that omalizumab's mechanism of inhibition is principally allosteric.
SPR experiments to investigate the mechanism of the inhibition of IgE-Fc binding to Fc⑀RI␣ by FabXol3 revealed a reduction in the number of available sites for FabXol3 on IgE-Fc (reduced B max ) when in complex with Fc⑀RI␣. The inhibition of IgE binding to Fc⑀RI by omalizumab has frequently been interpreted in terms of direct competition for overlapping sites, but there have been reports that indicate that omalizumab can bind to receptor-bound IgE (22). Here, we have demonstrated directly the ability of FabXol3 to bind to IgE-Fc when it is already bound to Fc⑀RI␣ to form a trimolecular complex. The effect of the pre-binding of IgE-Fc to Fc⑀RI␣ is to reduce the number of FabXol3-binding sites on IgE-Fc to less than 10% of those available in free IgE-Fc; this effect can only be due to allosteric modulation.
The nature of the interaction of FabXol3 with the IgE-Fc/ Fc⑀RI complex provides insights into the mechanism of accelerated dissociation of IgE from Fc⑀RI. This phenomenon was first reported for a DARPin and subsequently for omalizumab (21,22), the latter at substantially greater concentrations than those used therapeutically (23), and it is now shown here for omalizumab Fab fragments. We further conclude that the dissociation occurs only after binding of the second (lower affinity) FabXol3 molecule. Stated another way, a tetramolecular complex, (FabXol3) 2 /IgE-Fc/Fc⑀RI␣, must be formed for significant accelerated dissociation to occur.
Based on our observations with FabXol3, IgE-Fc, and sFc⑀RI␣, we envisage the following mechanism occurring for omalizumab, IgE, and Fc⑀RI. IgE binds to Fc⑀RI, and under these conditions a small population of the bound IgE molecules adopt a conformation to which omalizumab molecules can bind. When a second omalizumab molecule binds to form the tetrameric complex, the energy landscape of IgE is changed such that the interaction with Fc⑀RI is destabilized and a rapid dissociation of IgE from Fc⑀RI occurs. The inhibitory activities of omalizumab thus take advantage of the intrinsic flexibility of IgE and, for the process of accelerated dissociation, the dynamics of the IgE/Fc⑀RI complex. IgE has a number of unusual structural characteristics compared with other antibody isotypes, including the presence of the C⑀2 domains and the uniquely conformationally dynamic molten globule-like character of the C⑀3 domains (33). Together, these properties create an allosteric communication pathway that prevents simultaneous engagement of CD23 and Fc⑀RI; this is essential to avoid allergen-independent mast cell activation by cross-linking of Fc⑀RI-bound IgE by the trimeric CD23 molecule (12). Other functional advantages associated with the dynamics of IgE have been proposed for the membrane-bound IgE B cell receptor (16). We have shown here that omalizumab does not utilize the expected orthosteric mechanism for inhibition of the IgE/Fc⑀RI interaction, but rather it exploits these unusual dynamic properties of IgE. Furthermore, omalizumab can actively dissociate IgE from Fc⑀RI, albeit at concentrations higher than used therapeutically, by employing allostery and the intrinsic flexibility of IgE that persists even when in complex with its receptors.

Surface plasmon resonance
SPR experiments were carried out on a Biacore T200 instrument (GE Healthcare). Specific surfaces were prepared either by covalently coupling proteins using the amine coupling protocol (GE Healthcare), with coupling densities Ͻ300 resonance units, or capturing His-tagged proteins using an anti-His sensor surface. For capturing His-tagged ligands, an anti-His tag monoclonal antibody was employed and immobilized according to the manufacturer's instructions (Biacore His Capture Kit, GE Healthcare). In binding experiments, association times of 180 -240 s were typically used, and dissociation components were monitored for at least 500 s. Injections were performed at a flow rate of 25 l min Ϫ1 in a running buffer of 20 mM HEPES, pH 7.4, 150 mM NaCl, and 0.005% (v/v) surfactant P-20 (GE Healthcare). Most experimental binding measurements were performed at 25°C; some binding experiments were performed over a range of temperatures (5-35°C) to control the degree of accelerated dissociation in the system; low temperatures minimize this phenomenon, and higher temperatures increase it. In all cases, standard double referencing data subtraction methods were used (35), and kinetic fits were performed using Origin software (OriginLab).
TR-FRET inhibition assays were performed by competing 1 nM terbium-labeled IgE-Fc and 0 -20 nM Alexa Fluor 647labeled sFc⑀RI␣-IgG 4 -Fc with a range of concentrations of FabXol3. Assays were conducted in 384-well hi-base white plates (Greiner BioOne) using Lanthascreen buffer (Invitrogen) as a diluent. The plate was left to incubate overnight at room temperature and read by an Artemis plate reader (Berthold Technologies). TR-FRET ratios were then calculated for each well as the emission of acceptor at 665 nm divided by the emission of donor at 620 nm multiplied by 10,000.

Intramolecular FRET
Measurements of intramolecular FRET were performed essentially as described in Drinkwater et al. (16), using an IgE-Fc mutant (E289C) biotinylated at the C terminus using a BirA tag (Avidity). This protein was then fluorescently labeled using a thiol-reactive terbium chelate (Invitrogen) and bound to a monovalent, amine-reactive Alexa Fluor 488 (Invitrogen)labeled streptavidin (37). The anti-IgE Fabs at various concentrations were added to IgE-Fc in PBS, to give a final concentration of 25 nM IgE-Fc, and incubated for 120 min at 25°C. FRET was measured on an Analyst HT microplate reader (LJL Biosystems) with an excitation wavelength of 330 nm and emission wavelengths of 485 and 520 nm. Each sample was measured in quadruplicate, and in at least two separate experiments.

Crystallization
Crystals up to 400 m in length were grown at 18°C using the sitting drop vapor diffusion method. The reservoir contained 50 l of 4% (w/v) PEG 8000 and 0.03 M sodium fluoride, and the drop contained 100 nl of protein solution and 300 nl of reservoir. Despite extensive efforts at optimization, the diffraction quality of the crystals could not be further improved beyond that used for this study. Crystals typically started to grow after a few days and often dissolved in their drops, but they could be stabilized in 4 M trimethylamine N-oxide, which was successfully used as a cryoprotectant.

X-ray data collection and processing
Data were collected at beamlines I02 and I03 at the Diamond Light Source (Harwell, UK). Integration was performed using XDS (38) as implemented in the xia2 package (39). The crystals diffracted anisotropically, and data from multiple crystals were merged. The data were scaled to 3.7 Å resolution with AIM-LESS from the CCP4 suite (40,41) and then truncated to resolution limits of 3.7 Å (a*), 3.9 Å (b*), and 4.2 Å (c*) using the UCLA Diffraction Anisotropy Server (42). Calculation of the Matthews coefficient indicated a solvent content of ϳ62%, for a single 2:1 FabXol3/IgE-Fc complex (molecular mass of ϳ170 kDa) in the asymmetric unit.

Structure determination, model building, and refinement
The structure was solved by molecular replacement with PHASER (43) and MOLREP (44) from the CCP4 suite (40) using protein atoms from PDB entry 2WQR (8) and a 1.9 Å resolution omalizumab Fab structure, belonging to the same space group as published crystal structures (19,24) as search models. Refinement was initially performed with REFMAC (45) and later with PHENIX (46) and alternated with manual model building in Coot (47). The quality of the model was assessed with MolProbity (48) and POLYGON (49). Data processing and refinement statistics are presented in Table 1. A region of the electron density map is shown in supplemental Fig. S4. Interfaces were analyzed with PISA (50), figures were prepared with PyMOL (51), and movies were prepared with Chimera (52), PyMOL (51), and the eMovie plugin (53) for PyMOL. a Values in parentheses are for the highest resolution shell. b Refinement was performed with data truncated to resolution limits of 3.7 Å (a*), 3.9 Å (b*), and 4.2Å (c*). c R free set comprises 5% of reflections.