Engineering the Fab fragment of the anti-IgE omalizumab to prevent Fab crystallization and permit IgE-Fc complex crystallization

The omalizumab Fab was engineered to disrupt recurring crystal packing interactions in Fab crystal structures; this led to the eventual structure determination of an omalizumab-derived Fab in complex with its target, IgE-Fc.


Introduction
Immunoglobulin E (IgE) plays a central role in allergic disease through the interaction between its Fc region (IgE-Fc) and the Fc"RI receptor, in which cross-linking of Fc"RI-bound IgE by allergen triggers mast cell and basophil degranulation, with the release of inflammatory mediators (Gould & Sutton, 2008).
The high-affinity interaction between IgE and Fc"RI is a long-standing target in the development of treatments for allergic disease (Holgate, 2014). Omalizumab is an anti-IgE therapeutic monoclonal IgG1 antibody that inhibits the interaction of IgE with Fc"RI and is approved for the treatment of moderate-to-severe persistent allergic asthma and chronic idiopathic urticaria (Holgate et al., 2005;Sussman et al., 2014). Although the binding site for omalizumab had previously been mapped to the C"3 domain (Zheng et al., 2008), and omalizumab was known to bind to a partially bent IgE-Fc conformation (Hunt et al., 2012), the structural basis for its mechanism of action was poorly understood until only recently.
We, and others (Jensen et al., 2015), had attempted to crystallize the complex between the omalizumab Fab and IgE-Fc. However, despite extensive efforts, our crystallization trials of pre-formed omalizumab Fab/IgE-Fc and Fc"3-4 complexes only resulted in selective crystallization of the Fab. The structure of the omalizumab Fab in complex with the Fc"3-4 region of IgE-Fc has been reported, which revealed details of the omalizumab epitope on the C"3 domain (Pennington et al., 2016). However, this Fc"3-4 molecule lacked the (C"2) 2 domain pair and was conformationally constrained by an engineered disulfide bond that locked the C"3 domains into a closed conformation (Pennington et al., 2016). Given the flexible nature of the Fc"3-4 region, and the potential for extreme flexibility in IgE-Fc, which additionally contains the (C"2) 2 domain pair, this structure could thus provide only limited mechanistic insights.
We designed a mutagenesis strategy to disrupt the packing interactions observed in omalizumab Fab crystal structures, without affecting the antigen-binding CDRs, with the aim of crystallizing the complex between the omalizumab Fab and IgE-Fc. The strategy first involved creating a point mutation in a short segment of -strand structure found in the C domain CD loop, followed by two point mutations in the V L domain EF loop.
One omalizumab-derived Fab, termed FabXol3, which contains three point mutations in the light chain, subsequently enabled us to solve the 3.7 Å resolution crystal structure of the complex with IgE-Fc, revealing that omalizumab inhibits binding to Fc"RI allosterically . In this complex, IgE-Fc adopts a partially bent conformation, and the C"3 domains adopt a markedly open conformation, more open than that seen in any other crystal structure thus far.
Here, we report the structural basis and rationale for this mutagenesis strategy. Such an approach could inform the design and structure determination of other Fabs in complex with their target proteins in cases where the pre-formed complex is disrupted by the selective crystallization of one partner, in particular the Fab.
2.3. X-ray data collection, processing, structure determination and refinement Data were collected on beamlines I02, I03, I04, I04-1 and I24 at the Diamond Light Source, Harwell, UK. Data were integrated with XDS (Kabsch, 2010) using the xia2 package (Winter, 2010) or with MOSFLM (Leslie & Powell, 2007), and were scaled with AIMLESS (Evans & Murshudov, 2013) or SCALA (Evans, 2006) from the CCP4 suite . Structures were solved by molecular replacement using MOLREP (Vagin & Teplyakov, 2010) or Phaser (McCoy et al., 2007). Protein atoms from PDB entry 2fjf (Fuh et al., 2006) were used as a search model for the FabXol 1 structure. Subsequent structures were solved using protein atoms (V H , V L , C and C1 domains) from the FabXol 1 structure as a search model, although the CDR residues were removed. The structures were initially refined with REFMAC  and subsequently with Phenix (Liebschner et al., 2019), and refinement was alternated with rounds of manual model building with Coot (Emsley et al., 2010). Model quality was assessed with MolProbity (Chen et al., 2010). Dataprocessing and refinement statistics are summarized in Tables 1 and 2. Interfaces were analyzed with PISA (Krissinel & Henrick, 2007). Figures were produced with PyMOL.

Fluorescence-based thermal stability (T m ) measurement
A thermal stability assay was performed using a Quant-Studio 7 Real-Time PCR System (Thermo Fisher). 5 ml of 30Â SYPRO Orange Protein Gel Stain (Thermo Fisher), diluted from 5000Â concentrate with PBS pH 7.4, was added to 45 ml protein sample (0.2 mg ml À1 in PBS pH 7.4) and mixed. 10 ml of this solution was dispensed into an optical 384-well PCR plate. The PCR heating device was set at 20 C and increased to 99 C at a rate of 1.1 C min À1 . A charge-coupled device was used to monitor fluorescence changes in the wells. Fluorescence intensity increases were plotted and the inflection point of the slope was used to generate apparent midpoint temperatures (T m ).

Surface plasmon resonance
Surface plasmon resonance binding experiments were performed using a Biacore T200 instrument (GE Healthcare). Intact omalizumab, the Fabs and scFv were immobilized at similar densities on CM5 sensor chips using an amine-coupling protocol according to the manufacturer's instructions (GE Healthcare). The following immobilization densities were used for these studies: omalizumab, 970 resonance units;

Crystal structures of FabXol (wild-type omalizumab Fab): FabXol 1 and FabXol 2
The structure of FabXol (wild-type omalizumab Fab) was solved in two different crystal forms, which have also been reported by others (Jensen et al., 2015;Wright et al., 2015), and the space groups and unit-cell parameters of these structures, FabXol 1 and FabXol 2 , the latter now reported at a substantially higher resolution, are provided in Table 1. The structures reported here were the result of unsuccessful crystallization trials of the complex between FabXol and an unconstrained Fc"3-4 molecule, but similar crystals were also grown from crystallization trials of FabXol in complex with IgE-Fc.
The FabXol 1 structure (1.85 Å resolution) contains one Fab in the asymmetric unit, which forms two distinct interfaces with symmetry-related molecules (Fig. 1a). In the first interface, with an area of $395 Å 2 , residues from all three heavychain CDRs contact V L and C domain framework residues from a symmetry-related molecule; namely, the V L domain and Tyr102 (CDRH3)-Ser175 (C) (Fig. 1b). The second interface, with an area of $324 Å 2 , includes an extensive network of hydrogen bonds between an edge -strand from the C1 domain (-strand G) and a short segment of -strand structure in the C domain CD loop from a symmetry-related molecule. Here, the -strands are arranged in a parallel manner, with hydrogen bonds between the main-chain atoms of Lys214-Lys218 (C1) and Leu158-Ser160 (C), and between the side chains of Lys217 (C1) and Ser160 (C) (Fig. 1c). This interface is repeated throughout the crystal lattice, as an identical interface forms between Leu158-Ser160 (C) and Lys214-Lys218 (C1) from a symmetry-related molecule.
The FabXol 2 structure (2.3 Å resolution) contains two Fab molecules in the asymmetric unit, which are referred to here as FabXol 2A and FabXol 2B . The CDRs of both molecules adopt similar conformations to those observed in the FabXol 1 structure. CDRH1-3 residues also interact with the V L and C domain framework residues, akin to the first interface observed in the FabXol 1 structure, which for FabXol 2B also Structure of the omalizumab Fab (FabXol). (a) The FabXol 1 structure contains one Fab molecule (pink and blue) in the asymmetric unit. The heavychain CDRs of this Fab contact the V L and C domains (the latter hidden in this view) of one symmetry-related molecule (green and yellow) and the C domain of another (orange and gray). (b) Interface between heavy-chain CDR residues (blue) and V L and C domain framework residues from a symmetry-related molecule (green) in the FabXol 1 structure. Hydrogen bonds are depicted by black lines. (c) Interface between an edge -strand from the C1 domain (blue) and the C domain from a symmetry-related molecule (orange) in the FabXol 1 structure. Hydrogen bonds are depicted by black lines. (d) Interface between heavy-chain CDR residues (gray) and V L and C domain framework residues from a symmetry-related molecule (yellow) for FabXol 2B , which includes a hydrogen bond between His101 (CDRH3) and Gln81 (V L domain). Hydrogen bonds are depicted by black lines. includes a hydrogen bond between His101 (CDRH3) and Gln83 (V L ) (Fig. 1d). The arrangement of Fabs in the FabXol 2 asymmetric unit precludes the propagation of the second, -strand-mediated interface throughout the crystal lattice by a single Fab molecule, as in the FabXol 1 structure. However, interactions between FabXol 2A and FabXol 2B , and different symmetry-related molecules, each display this same -strand interaction, in which Lys214-Lys218 (C1) from FabXol 2A interact with Leu158-Ser160 (C) from one symmetry-related molecule, while Leu158-Ser160 (C) from FabXol 2B interact with Lys214-Lys218 (C1) from a different symmetry-related molecule.

Crystal structure of scFvXol (omalizumab-derived scFv)
We also attempted to crystallize the complex between a single-chain form of omalizumab (scFvXol) and IgE-Fc, but were unsuccessful. However, we solved the crystal structure of scFvXol alone, in which the light-and heavy-chain variable domains are connected by a (Gly 4 Ser) 4 linker, to 2.3 Å resolution (Table 1). The scFvXol structure contains one molecule in the asymmetric unit.
In this structure, the -strand-mediated crystal packing interaction observed in the FabXol 1 and FabXol 2 structures is absent, as the construct lacks the C1 and C domains. However, CDRH1-3 residues from a symmetry-related molecule contact the V L domain of scFvXol in a similar manner to the first interface described for the FabXol 1 and FabXol 2 structures, although the interface area is reduced from $395 to $290 Å 2 due to the absence of the C domain in scFvXol.

Mutagenesis strategy I: disrupting the interaction between the Cc1 and Cj domains
Crystallization trials of the complexes between FabXol (omalizumab Fab) and IgE-Fc, between scFvXol (omalizumab-derived scFv) and IgE-Fc, and between FabXol and an unconstrained Fc"3-4 molecule all led to selective crystallization of the Fab or were unsuccessful. Two recurring interfaces in the Fab and scFvXol structures, described in Section 3.1, suggested a route to disrupt crystal packing interactions without mutating the CDR residues responsible for IgE-Fc binding.
We first attempted to disrupt the interface between the edge -strand (-strand G) from the C1 domain (Lys214-Lys218) and the short -strand segment in the C domain CD loop (Leu158-Ser160), observed in the FabXol 1 and FabXol 2 structures. Leu158 from the C domain CD loop was mutated to proline, with the aim of altering its secondary structure, to disrupt the extensive, hydrogen-bond-mediated interactions. This omalizumab-derived Leu158Pro mutant Fab was termed FabXol1.
3.4. Crystal structures of FabXol1 (omalizumab-derived Leu158Pro mutant Fab): FabXol1 1 and FabXol1 2 The Leu158Pro mutation alone was not sufficient to prevent selective crystallization of the Fab, and the structures reported here were the result of unsuccessful crystallization trials of the complex between FabXol1 and IgE-Fc. Two structures were solved for FabXol1, in new crystal forms, and the space groups and unit-cell parameters of these structures, FabXol1 1 and FabXol1 2 , are provided in Table 2.
The FabXol1 1 structure (1.8 Å resolution) contains two Fab molecules (FabXol1 1A and FabXol1 1B ) in the asymmetric unit (Fig. 2a). In this structure, the network of hydrogen bonds observed in the FabXol structures between -strands of the C1 and C domains is indeed disrupted, but the engineered residue, Pro158, now forms other crystal packing interactions.
CDRH1-3 residues in both molecules of the FabXol1 1 structure adopt essentially identical conformations to those found in the FabXol 1 and FabXol 2 (wild-type omalizumab Fab) and scFvXol (omalizumab-derived scFv) structures. They form similar crystal packing interactions to the first interface described for the FabXol 1 structure, in which the heavy-chain CDRs contact the V L domain AB, C 00 D and EF loops, and the C domain DE loop from a symmetry-related molecule. In both molecules, hydrogen bonds form between Ser31 (CDRH1) and Asp17 (V L ), between Tyr54 (CDRH2) and Arg65 (V L ) and between Tyr102 (CDRH3) and Ser175 (C) (Fig. 2d).
The FabXol1 2 structure (2.5 Å resolution) contains four Fab molecules (FabXol1 2A -FabXol1 2D ) in the asymmetric unit. In this structure, the packing environment of Pro158 differs from that in the FabXol1 1 structure. Again, the -strand interactions between C1 and C domains are disrupted, but new packing interactions involving Pro158 are formed. In all four molecules of the FabXol1 2 structure, Pro158 forms van der Waals interactions with Pro158-Ser160 (C) from a noncrystallographic symmetry-related Fab (Fig. 3a). In this manner, Pro158 mediates light-chain/light-chain interactions between FabXol1 2A and FabXol1 2C , and between FabXol1 2B and FabXol1 2D . Due to the arrangement of the four Fab molecules in the asymmetric unit, Pro158 from FabXol1 2C is positioned at an interface comprising three Fabs (FabXol1 2A -FabXol1 2C ), and in addition to the interface with Pro158-Ser160 from FabXol1 2A , also contacts Arg87 (V H ) from FabXol1 2B (Fig. 3a).
In molecules FabXol1 2A and FabXol1 2B , the heavy-chain CDRs adopt similar conformations to those in the FabXol, scFvXol and FabXol1 1 structures. CDR residues from FabXol1 2B form a similar interface with V L and C domain framework residues from a symmetry-related molecule; hydrogen bonds form between Ser31 (CDRH1) and Asp17 (V L ), between Tyr54 (CDRH2) and Arg65 (V L ), between His101 (CDRH3) and Gln83 (V L ) and between Tyr102 (CDRH3) and Ser175 (C), burying a surface area of 384 Å 2 . Although FabXol1 2A contacts the V L and C domains of a symmetry-related molecule, the position of this molecule is shifted and the interface area, which is reduced to 274 Å 2 , contains a single hydrogen bond between Tyr102 (CDRH3) and Asp174 (C) (Fig. 3b).
By contrast, the CDRH1 and CDRH3 conformations differ in molecules FabXol1 2C and FabXol1 2D compared with the other structures described thus far. In these molecules, binding of a glycerol molecule causes the Tyr33 (CDRH1) and His101 (CDRH3) side chains to adopt substantially different positions (Fig. 3c), the implications of which are discussed later. Crystal contacts for FabXol1 2C and FabXol1 2D also differ markedly compared with the other Fabs. In FabXol1 2C , Thr30 and Ser31 (CDRH1) form hydrogen bonds with Thr73 and Ser28 (V L ), respectively, from one symmetry-related molecule, while Tyr102 (CDRH3) packs against Gly15 and Gly16 (V H ) from another molecule (Fig. 3d). On the other hand, in FabXol1 2D , only the interaction between Tyr102 and Gly15 and Gly16 from the second symmetry-related molecule is found; the first molecule is positioned further away, precluding hydrogen bonds between Thr30 (CDRH1) and Thr73, and between Ser31 (CDRH1) and Ser28. By contrast, CDRH2 residues do not participate in any crystal contacts, and adopt similar conformations to those in FabXol1 2A and FabXol1 2B .
Despite the different contacts formed by CDRH1 and CDRH3 in molecules FabXol1 2C and FabXol1 2D , the packing environment would not preclude the CDR conformations observed in the FabXol, scFvXol and FabXol1 1 structures, and in molecules FabXol1 2A and FabXol1 2B .

Mutagenesis strategy II: disrupting packing interactions involving the heavy-chain CDRs
Although the Leu158Pro mutation in the short -strand segment of the C domain CD loop disrupted the interaction with the C1 domain edge -strand (strand G), it did not prevent selective crystallization of the Fab. We next attempted to disrupt the interface between the heavy-chain CDRs and the V L and C domain framework residues. As most of this interface involves interactions between the CDRs and the V L domain, and mutating the CDRs could adversely affect the interaction with IgE-Fc, we mutated Ser81 and Gln83 from the V L domain EF loop, which contribute to this interface, to Arg81 and Arg83, respectively, thus incorporating bulkier, charged side chains. We created two omalizumab-derived Fabs, namely FabXol2, with Ser81Arg and Gln83Arg muta-tions, and FabXol3, which additionally contains a Leu158Pro mutation. Thermal stability measurements revealed that the incorporation of these three point mutations, either alone or in combination with one another, did not substantially affect the overall stability of the Fabs (Table 4).
3.6. Crystal structures of FabXol2 (omalizumab-derived Ser81Arg, Gln83Arg mutant Fab) and FabXol3 (omalizumabderived Ser81Arg,Gln83Arg,Leu158Pro mutant Fab) Complexes between IgE-Fc and both of the omalizumabderived Fabs that contained the Ser81Arg and Gln83Arg mutations were eventually crystallized. Crystals with a similar morphology were grown for each complex, although the FabXol3/IgE-Fc complex crystals diffracted to higher resolution, and we recently reported the crystal structure of the complex to 3.7 Å resolution .
To understand the effects of the Ser81Arg and Gln83Arg (V L ) mutations on Fab crystal packing interactions, we solved the structures of FabXol2 and FabXol3 alone. Both FabXol2 and FabXol3 crystallized in the same crystal form (Table 2), with one Fab molecule in the asymmetric unit. With the exception of the light-chain residue 158, which is leucine in FabXol2 and proline in FabXol3, the structures are otherwise essentially identical.
The packing interactions that involve V L domain residues 81 and 83 in the FabXol and FabXol1 structures are substantially different in the FabXol2 and FabXol3 structures. In contrast to Ser81, which contacts Ser31 (CDRH1) and Tyr54 (CDRH2), Arg81 instead forms hydrogen bonds with Asn156 (C, symmetry-related molecule) (Fig. 4a). In FabXol3, Arg81 contacts Pro158 (C), while Leu158 is partially disordered in FabXol2. Furthermore, and in contrast to Gln83, which contacts Tyr33 (CDRH1), Tyr54 (CDRH2) and His101 (CDRH3) in the FabXol and FabXol1 structures, Arg83 does not participate in any crystal packing interactions in the FabXol2 and FabXol3 structures (Fig. 4a). As the overall structures of FabXol2 and FabXol3 are similar, further discussion will be limited to the FabXol3 structure, which was solved at higher resolution (1.45 Å for FabXol3 compared with 2.05 Å for FabXol2).
On the other hand, Asp55 (CDRH2) forms a salt bridge with Lys211 from the C domain of a different symmetry-related Fab, and together with Gly56 (CDRH2) packs against Pro117 and Ser118 (Fig. 4c).
The FabXol3 CDRH1 and CDRH3 conformations are markedly different to those in the FabXol, scFvXol and FabXol1 structures; the nature and implications of these conformational differences are discussed later.

Conformational diversity in the CDRs: comparison of unbound and bound Fab structures
In the FabXol, scFvXol and FabXol1 1 structures, and in the molecules FabXol1 2A and FabXol1 2B , the heavy-chain CDRs adopt similar conformations (Figs. 1b, 1d, 2d and 3b). However, substantial conformational diversity is observed for CDRH1 and CDRH3 in molecules FabXol1 2C and FabXol1 2D , and in FabXol3.
In molecules FabXol1 2C and FabXol1 2D , a glycerol molecule occupies a structurally equivalent position to Ser378 and  Structure of the omalizumab-derived Ser81Arg, Gln83Arg, Leu158Pro mutant (FabXol3). (a) In the FabXol3 (gray) and FabXol2 (pink) structures, Arg81 forms hydrogen bonds with Asn156. In the FabXol3 structure, Arg81 contacts Pro158, while Leu158 is partially disordered in the FabXol2 structure. Arg83 does not form any crystal packing interactions. (b) In the FabXol3 structure, CDRH1 and CDRH3 residues (gray) contact the V L domain of a symmetry-related molecule (green). Hydrogen bonds are depicted by black lines. (c) Asp55 (CDRH2) forms a salt bridge with Lys211 from the C domain of a symmetry-related molecule (blue). Table 4 Thermal stabilities of the omalizumab-derived Fabs. Gly379 from the C"3 domain in the complex between the omalizumab-derived Fab and IgE-Fc , altering the position of Tyr33 (CDRH1), which adopts a similar position to that in the IgE-Fc-bound Fab (Fig. 5a). The conformations of Ser31 (CDRH1) and Gly32 (CDRH1) are also similar to those in the complex, presumably due to the conformational change involving Tyr33. In the complex with IgE-Fc, Gly32 and Tyr33 from CDRH1 contribute to the interface with the C"3 domain, packing against Ala377 and Ser378. The glycerol molecule, close to Tyr33, also causes the His101 (CDRH3) side chain to adopt a different position (Fig. 5a); however, the overall conformation of CDRH3 is otherwise similar to that in the unbound FabXol, scFvXol and FabXol1 1 structures and in the molecules FabXol1 2A and FabXol1 2B . In FabXol3, residues Ser25-Gly32 (CDRH1) adopt a markedly different conformation compared with the other unbound and bound Fab structures, which alters the positions of Tyr27 and Ile29; the Phe79 side chain, adjacent to CDRH1, also adopts a different position (Fig. 5b). On the other hand, Tyr33 adopts a similar position to that in the FabXol1 2C and FabXol1 2D molecules and the bound Fab structures. Comparison of the FabXol3 structure with the structure of the complex with IgE-Fc  reveals that the positions adopted by Ser25-Ser31, and Tyr33 in FabXol3 would not preclude an interaction with the C"3 domain; however, Gly32 would clash with Ser378. This particular CDRH1 conformation thus appears to be incompatible with IgE binding. By contrast, in FabXol3, CDRH3 adopts a strikingly different conformation compared with the other Fab structures reported here (Fig. 5c). In these Fab structures, the CDRH3 conformation is incompatible with IgE binding due to steric clashes with the C"3 domain. However, the CDRH3 conformation in the unbound FabXol3 structure is similar to Conformational diversity in the omalizumab CDRs. (a) In molecule FabXol1 2C (blue), the binding of a glycerol molecule (GOL) alters the position of Tyr33 (CDRH1), which adopts a similar position to that in the IgE-Fc-bound Fab (yellow; Davies et al., 2017). The FabXol 1 structure (gray) is shown for comparison. (b) Compared with the FabXol1 structure (gray), molecule FabXol1 2C from the FabXol1 2 structure (blue) and FabXol3 from the complex with IgE-Fc (yellow), CDRH1 adopts a conformation in the unbound FabXol3 structure (pink) that alters the positions of Tyr27 and Ile29 (CDRH1). Phe79 also adopts a different position. By contrast, Tyr33 (CDRH1) adopts a similar position in FabXol1 2C (blue), unbound FabXol3 (pink) and FabXol3 bound to IgE-Fc (yellow). Tyr33 adopts a substantially different position in FabXol 1 (gray). (c) In the FabXol3 structure (pink), CDRH3 adopts a different conformation compared with that in the FabXol 1 structure (gray). (d) The conformation adopted by CDRH3 in the unbound FabXol3 structure (pink) is similar to that in the FabXol3-IgE-Fc complex (yellow). The surface of the C"3 domain from the complex is colored orange . the conformation adopted by CDRH3 in the FabXol3/IgE-Fc complex Fig. 5d); a conformational change in the CDRH3 main chain causes a dramatic rearrangement in the positions of side-chain residues, particularly His101, Tyr102 and Phe103, which contact the C"3 domain in the complex.
In contrast to the structural diversity displayed by CDRH1 and CDRH3, the conformation of CDRH2 is conserved in the unbound Fab and scFv structures, and in the complexes of the omalizumab Fab with the constrained Fc"3-4 molecule (Pennington et al., 2016) and of FabXol3 with IgE-Fc . Like CDRH2, the light-chain CDR conformations are also conserved; similar conformations are adopted in the 12 independent views reported here and in other unbound Fab structures (Jensen et al., 2015;Wright et al., 2015), which are similar to those in the complexes between the omalizumab Fab and the constrained Fc"3-4 molecule (Pennington et al., 2016) and between FabXol3 and IgE-Fc . Nevertheless, the FabXol2 and FabXol3 crystal structures show substantial conformational diversity in the heavy-chain CDRs, and together with the FabXol1 2 structure reveal how conformations compatible with IgE binding are adopted in the unbound Fab.

Interaction of the omalizumab-derived Fabs and scFv with IgE-Fc in solution
The aim of our mutagenesis strategy was to disrupt the crystal packing interactions observed in the wild-type omalizumab (FabXol) crystal structures, without mutating the CDR residues responsible for IgE-Fc binding and significantly affecting the affinity for IgE-Fc. We have previously demonstrated that the kinetics of the interaction between omalizumab and IgE-Fc are biphasic, with one high-affinity ($1 nM) and one lower-affinity ($30 nM) interaction , and that FabXol3 has a slightly higher affinity for IgE-Fc than FabXol (wild-type omalizumab Fab) and intact omalizumab .
We used surface plasmon resonance analysis to characterize further the interaction between IgE-Fc and the omalizumabderived Fab and scFv constructs. As we have shown previously, at the highest concentration tested (100 nM IgE-Fc), the omalizumab-derived Fabs and scFv all display the same mode of interaction with IgE-Fc, i.e. a biphasic model with one higher-affinity and one lower-affinity binding interaction . When these data were normalized to have the same maximum binding values, it was found that the association rates were similar to those for intact omalizumab  Table 5). However, a statistically significant trend of increasingly slower dissociation rates was observed: the dissociation rate for the omalizumab-derived Fab (FabXol) is slower than that for intact omalizumab, FabXol2 has a slower dissociation rate than FabXol and FabXol3 is even slower, while the scFvXol dissociation rate is the slowest of all (Table 5 and Supplementary Fig. S1).

Discussion
After unsuccessful attempts to crystallize the complex between the Fab fragment of the therapeutic anti-IgE omalizumab and IgE-Fc, and the Fc"3-4 region, we designed a mutagenesis strategy to disrupt the substantial, and recurring, crystal packing interactions observed in different omalizumab Fab structures. We targeted crystal packing interactions at two different interfaces. The first interface comprised hydrogen bonds between an edge -strand from the C1 domain (-strand G; Lys214-Lys218) and a short segment of -strand structure in the C domain CD loop (Leu158-Ser160). The second interface involved the omalizumab heavy-chain CDRs and V L domain AB, C 00 D and EF loops and C domain DE loop. Our mutations were designed to disrupt these packing interactions without significantly affecting the affinity of omalizumab for IgE, and as such were distal to the antigenbinding CDRs.
Packing interactions similar to that between the C1 domain edge -strand (strand G) and the C domain CD loop are found in a number of other crystal structures containing Fab fragments (see, for example, Hall et al., 2016;Lee et al., 2017;Li et al., 2009;Sickmier et al., 2016). Indeed, a variety of packing interactions involving hydrogen-bond networks between -strands have been detected in crystal structures of intact antibodies and their fragments (Edmundson et al., 1999;Wingren et al., 2003), including antiparallel arrangements between edge strands in C and C1 domains (see, for example, Faber et al., 1998), V H domains (see, for example, Harris et al., 1998) and V L domains (see, for example, Bourne et al., 2002).
We mutated Leu158 from the omalizumab C domain CD loop to proline (omalizumab-derived mutant FabXol1) to disrupt the interface with strand G from the C1 domain, and although this was achieved, the FabXol1 molecule still crystallized preferentially, in different packing arrangements stabilized in part by the presence of Pro158.
We next targeted the crystal packing interactions between the omalizumab CDRs and V L and C domain framework residues (V L domain AB, C 00 D and EF loops and C domain DE loop) from symmetry-related molecules. We mutated Ser81 and Gln83 from the omalizumab V L domain EF loop to arginine and created two omalizumab-derived mutants: FabXol2 contained the Ser81Arg and Gln83Arg mutations, while FabXol3 additionally contained the Leu158Pro mutation. The IgE-Fc protein was successfully crystallized in complex with both FabXol2 and FabXol3, and the 3.7 Å resolution crystal structure of the FabXol3/IgE-Fc complex research communications Table 5 Kinetics of omalizumab, the omalizumab-derived Fabs and scFv binding to IgE-Fc.
Omalizumab 3.3 Â 10 5 2.9 Â 10 5 7.0 Â 10 À4 1.2 Â 10 À2 FabXol 5.7 Â 10 5 4.4 Â 10 5 5.6 Â 10 À4 1.2 Â 10 À2 FabXol2 5.1 Â 10 5 3.3 Â 10 5 4.5 Â 10 À4 1.1 Â 10 À2 FabXol3 9.7 Â 10 5 2.7 Â 10 5 3.3 Â 10 À4 9.0 Â 10 À3 scFvXol 6.9 Â 10 5 3.1 Â 10 5 2.9 Â 10 À4 8.7 Â 10 À3 was recently reported . Engineering the Ser81Arg and Gln83Arg mutations in the V L domain of the omalizumab Fab clearly disrupted the interactions seen in the FabXol structure, but these residues also formed new packing interactions in the FabXol2 and FabXol3 structures that were seen when these molecules were crystallized alone. Presumably, however, these packing contacts were collectively weaker than those in either FabXol or FabXol1, since they were unable to compete with the pre-formed Fab/IgE-Fc complexes and their crystallization. Unbound IgE-Fc adopts an acutely bent conformation, in which the C"2 domains fold back against the Fc"3-4 region (Doré et al., 2017;Holdom et al., 2011;Wan et al., 2002). IgE-Fc is more acutely bent in the crystal structure of the sFc"RI/ IgE-Fc complex (Holdom et al., 2011), less acutely bent when in complex with sCD23 (Dhaliwal et al., 2017), partially bent when in complex with FabXol3  and fully extended in the complexes with the anti-IgE Fabs a"Fab and 8D6 (Chen et al., 2018;Drinkwater et al., 2014); these structures demonstrate that IgE-Fc is conformationally dynamic. However, despite this flexibility, IgE adopts a predominantly bent conformation in solution (Beavil et al., 1995;Davis et al., 1990;Holowka & Baird, 1983;Holowka et al., 1985;Hunt et al., 2012;Zheng et al., 1991Zheng et al., , 1992. The propensity for IgE-Fc to adopt such a bent conformation might account for the selective crystallization of the omalizumab Fab and the omalizumab-derived mutant FabXol1. Bending of IgE-Fc, from the partially bent conformation observed in the FabXol3/IgE-Fc complex to the acutely bent structure, would disrupt one of the omalizumab binding sites on the C"3 domain. In the FabXol3/ IgE-Fc complex, Arg81 and Arg83 from one FabXol3 molecule contact one of the C"2 domains, in addition to the omalizumab binding site on the C"3 domain. This additional interaction might stabilize the partially bent conformation in the complex. In IgE-Fc and Fc"3-4, the C"3 domains adopt a range of conformations relative to one another, from closed to open (Chen et al., 2018;Cohen et al., 2014;Davies et al., 2017;Dhaliwal et al., 2012Dhaliwal et al., , 2014Dhaliwal et al., , 2017Doré et al., 2017;Drinkwater et al., 2014;Garman et al., 2000;Holdom et al., 2011;Jabs et al., 2018;Wan et al., 2002;Wurzburg & Jardetzky, 2009;Wurzburg et al., 2000;Yuan et al., 2013); this conformational diversity is crucial for the allosteric regulation of IgE binding to its receptors, Fc"RI and CD23 (Borthakur et al., 2012;Dhaliwal et al., 2012). The flexibility of the C"3 domains could account for our failure to crystallize the complex between the omalizumab Fab and the unconstrained Fc"3-4 molecule, which lacks the C"2 domains. Notably, the reported omalizumab Fab complex (Pennington et al., 2016) is with an Fc"3-4 molecule that contains an engineered disulfide bond, which locks the C"3 domains into a closed conformation, thus reducing the overall flexibility of the complex.
Fab fragments are invaluable tools as chaperone proteins for crystallization, and are used for their ability to trap different conformations or reduce flexibility in the target protein ( Uysal et al., 2009). However, in our case, crystallization trials of our conformationally flexible target protein, IgE-Fc, in complex with the Fab fragment of the therapeutic anti-IgE antibody omalizumab resulted in the disruption of pre-formed complexes and selective crystallization of the Fab alone.
Here, we have described a successful mutagenesis strategy in which framework regions of the omalizumab Fab were engineered to disrupt recurring crystal packing interactions in the Fab crystal structures, without significantly altering the stability of the Fab, nor its affinity for IgE-Fc. Although disrupting the hydrogen-bond-mediated interactions between -strands did not prevent selective crystallization of the Fab, the recurring interface between the light chain and CDRs was disrupted by introducing bulkier residues through point mutations in the light-chain framework regions.
This approach, of introducing point mutations distal to the antigen-binding CDRs to disrupt undesired crystal packing interactions, could assist in the structure determination of Fabs in complex either with similarly conformationally flexible, or indeed inflexible, target proteins.