Main

Adult humans produce more IgA per day than all other antibody isotypes combined1. IgA is the predominant antibody isotype in mucosal regions, which represent the primary avenue for invasion by many pathogens. The mucosal immune system is responsible for multiple levels of protection against pathogens: induction of immunity in mucosa-associated lymphoid tissues, prevention of pathogen adherence by secretory IgA-mediated immune exclusion, and systemic immune effector functions through the action of serum antibodies2.

Two distinct forms of IgA are found in serum: monomeric and dimeric IgA. Monomeric IgA is analogous to IgG and IgE, but has additional carboxy-terminal tailpieces that interact with J- (joining) chain, forming covalent dimeric IgA3. Co-expression of IgA and J-chain by B cells in the mucosal lamina propria leads to high local concentrations of dimeric IgA in these regions, whereas serum IgA is predominantly monomeric1. Dimeric IgA is specifically bound by pIgR on the basolateral surfaces of the mucosal epithelium, transcytosed through the epithelial layer, and released into mucosal secretions as a covalent complex of dimeric IgA and the cleaved pIgR ectodomain (called bound secretory component, SC)3. The complex of dimeric IgA with bound SC is known as secretory IgA (SIgA) and represents a third distinct form of IgA.

IgA-mediated immune effector responses such as phagocytosis, antibody-dependent cell-mediated cytotoxicity, respiratory burst and cytokine release are mediated through FcαRI (CD89), an IgA-specific receptor that is expressed on monocytes, eosinophils, neutrophils and macrophages4. Both monomeric and dimeric IgA can bind to FcαRI, and monomeric or dimeric IgA immune complexes can activate phagocytosis and other immune responses through the clustering of FcαRI5. SIgA cannot trigger phagocytosis5,6, although it can initiate certain inflammatory responses, such as respiratory burst in polymorphonuclear leukocytes7 and degranulation of eosinophils8, through FcαRI and an integrin co-receptor.

Although FcαRI is distantly related to FcγR and FcɛRI proteins, it shares higher sequence similarity with the leukocyte immunoglobulin-like receptors (LIR/LILR/ILTs), killer inhibitory receptors (KIRs) and other members of the leukocyte receptor cluster9. Unlike FcγRIII and FcɛRI, which form 1:1 receptor:Fc complexes through interactions with hinge-proximal Fc regions10,11, FcαRI forms a 2:1 complex with an Fcα dimer12 by binding to each CH2–CH3 domain interface (called Cα2 and Cα3 in IgA)13,14, reminiscent of the 2:1 complex formed between the neonatal Fc receptor (FcRn) and IgG, in which FcRn binds to the CH2-CH3 interface of Fcγ15. To understand the molecular basis of the FcαRI:IgA interaction, we determined the crystal structures of the FcαRI ectodomain in two crystal forms and a 2:1 FcαRI:Fcα complex at resolutions of 3.0, 3.9 and 3.1 Å, respectively (Supplementary Table 1).

Structure of FcαRI

The FcαRI ectodomain is composed of two immunoglobulin-like domains oriented at approximately 90° to one another (Fig. 1a). The domains of FcαRI have a similar folding topology to the domains of LIR-1 (ref. 16) and other immunoglobulin superfamily Fc receptors11,17,18,19 (Fig. 1b; see Supplementary Table 2 for r.m.s. deviations between the domains of FcαRI and other receptors). In both domains 1 and 2 (D1 and D2), the first β-strand is shared between the two β-sheets. D1 includes short C′ and D strands connecting strands C and E, whereas there is no D strand in D2. There are short stretches of 310 helix in both domains, and a polyproline type-II helix in D2 (Fig. 1a). Ordered carbohydrate is visible at two of the four potential N-linked glycosylation sites in each molecule of free FcαRI in the 3.0 Å structure (N120 and N156 in the first molecule, and N44 and N58 in the second molecule) and at all four sites in the Fcα-bound FcαRI structure. The overall domain orientation resembles that seen in the LIR-1 (ref. 16; Fig. 1c) and KIR20 receptors. Although other immunoglobulin superfamily Fc receptors such as FcɛRI, FcγRIIa, FcγRIIb and FcγRIII also have overall bent structures11,17,18,19, the relative D1–D2 orientation in these receptors is opposite to that observed for FcαRI and LIR-1 (ref. 16; Fig. 1c).

Figure 1: FcαRI structure.
figure 1

a, Ribbon diagram of FcαRI. Disulphide bonds are yellow and carbohydrate residues are shown in ball-and-stick representation. b, FcαRI topology diagram. β-Strands are green, 310 helices are blue, polyproline type-II helices are orange and disulphide bonds are yellow. Green dots show residues that contact Fcα. c, Superimposed Cα traces of FcαRI, LIR-1 (ref. 16), FcɛRI17 and FcγRIII11. d, Superposition of the D1 domains in independently determined FcαRI structures. Regions that differ are highlighted in blue, green or red. e, Simulated annealing electron density omit maps showing the D or C′ strands in D1 of free and Fcα-bound FcαRI, respectively, with ball-and-stick models of the deleted regions superimposed.

A comparison of three independent structures of free FcαRI (two molecules from the 3.0 Å structure, and one from the 3.9 Å structure) reveals moderate flexibility in the D1–D2 interface; the interdomain angles are 97° and 92° for the two molecules in the 3.0 Å structure, and 88° for the low-resolution model. The D1–D2 interface buries 1,170–1,238 Å2, slightly more than interdomain interface areas found in the KIR (1,154 Å2; ref. 20) and LIR-1 (1,012 Å2; ref. 16) receptors. The DE and FG loops in D1 show variable conformations in the independent views of free FcαRI and in FcαRI complexed with Fcα (Fig. 1d).

Structure of Fcα

The structure of Fcα described here is derived from the FcαRI:Fcα complex structure, because attempts to crystallize Fcα alone were unsuccessful. Fcα is a two-fold symmetric dimer of IgA heavy chains, each with two immunoglobulin constant domains, Cα2 and Cα3 (Fig. 2). The Cα3 domains form an extensive dimer interface that buries a total of 2,061 Å2 and associate to form similar dimerization contacts to those seen in the analogous Cγ3–Cγ3 (Fig. 2a) and Cɛ4–Cɛ4 interactions in Fcγ21 and Fcɛ22,23, respectively.

Figure 2: Fcα structure.
figure 2

a, Ribbon diagrams showing front (left) and side (right) views of Fcα (top) and Fcγ (bottom)33. Disulphide bonds are shown in yellow and carbohydrate residues are shown in ball-and-stick representation. b, Topology diagram of Fcα. β-Strands are blue or magenta, 310 and α-helices are light blue and disulphides are yellow. The proposed C241–C241 disulphide bond (not present in our construct) is shown as a dashed yellow line. Blue and magenta dots show residues that contact FcαRI.

Apart from minor variations in loop regions, the primary differences between Fcα and other Fcs of known structure (Fcγ and Fcɛ) involve the positions of interdomain disulphide bonds and N-linked carbohydrates. Like Fcγ, the heavy chains of Fcα are covalently coupled through disulphides linking hinge residues (C241 to counterpart C241; not present in our construct)24,25. However, Fcα also has unusual disulphides between C299 in the DE loop of each Cα2 domain and C242 at the base of the hinge in the opposite heavy chain (Fig. 2b). These interdomain disulphide bonds are reminiscent of those observed linking the IgE heavy chains at the base of the Cɛ2 domains22, although the residues involved are not analogous. Interestingly, Fcα C299 is the counterpart of Fcγ N297 and Fcɛ N394, the residues to which the interdomain N-linked glycans that stabilize the Cγ2 or Cɛ3 domains are attached. Thus, this substitution in Fcα of cysteine for asparagine allows disulphide bond formation and prevents attachment of N-linked glycans between the Cα2 domains, which is a conserved feature of Cγ2 domains and the counterpart Cɛ3 domains. Instead, Fcα contains N-linked carbohydrates attached to N263 on an outer surface of the Cα2 domains (Fig. 2a). Each N-glycan contacts both the Cα2 and Cα3 domains, burying 465 Å2 and 457 Å2, respectively, unlike the Fcγ and Fcɛ N-glycans, which contact only the Cγ2 (or Cɛ3) domains.

Structure of the FcαRI:Fcα complex

The FcαRI:Fcα complex consists of two FcαRI molecules bound to a single Fcα dimer, with one receptor binding at each Cα2–Cα3 interface (Fig. 3a, b). The FcαRI-binding site on Fcα is composed of residues in the AB helix/loop of the Cα2 domain (L256, L257, L258) and several regions of the Cα3 domain: the A strand (E348), the C strand (R382, L384), the CC′ loop (S387, E389), the F strand (M433, H436), the FG loop (E437, A438, L439, P440, L441) and the G strand (A442, F443, T444, Q445) (Figs 2b and 4a, and Supplementary Table 3). These residues are conserved in all human IgA subclasses and allotypes (IgA1, IgA2m(1), IgA2m(2) and IgA2n). Sequence alignment of IgA1 with other antibody isotypes shows that the specificity of FcαRI for IgA appears to arise predominantly from residues in the C strand and CC′ loop, with contributions from residues in the F strand, FG loop and G strand (Fig. 5). The site on FcαRI that binds Fcα involves residues found in the BC loop (Y35), the D strand (R52, R53, L54, K55), the DE loop (F56, W57, carbohydrate linked to N58), and the FG loop (R82, G84, H85) of D1 (Figs 1b and 4a, and Supplementary Table 3). A similar region of LIR-1 has been mapped as the binding site for UL18, a viral major histocompatibility complex (MHC) class I homologue16.

Figure 3: FcαRI:Fcα structure.
figure 3

a, b Ribbon diagrams of the FcαRI:Fcα complex structure seen from the side (two-fold symmetry axis is vertical) (a) or the top (looking down the two-fold symmetry axis) (b). The approximate location of the outer surface of the cell membrane is shown as a grey line in the top panel. ce Ribbon diagrams of the FcRn:Fcγ15, FcγRIII:Fcγ11 and FcɛRI:Fcɛ10 complexes showing proposed orientations with reference to the cell membrane. The receptors are green, and the Fcs are blue and magenta.

Figure 4: FcαRI:Fcα interface.
figure 4

Stereoviews of residues at the FcαRI:Fcα interface (defined as residues with any non-hydrogen atom within 4 Å of the partner domain). Potential hydrogen bonds are shown as black dotted lines. Residues are colour-coded according to protein (a), the chemical character of their side chains (b), or their effects on binding affinity when substituted13,14,26–28 (c).

Figure 5: Alignment of antibody Fc sequences.
figure 5

Residues that make up the FcαRI-binding site in IgA1, or exact matches in other isotypes, are in red. Closely matched residues (that is, conserved hydrophobicity or charge) are in yellow. Similar residues (that is, E → Q substitution) are in blue. The secondary structure of Fcα is illustrated above the sequences. Carets () and asterisks (*) denote residues that contribute 0–4% or 5–12%, respectively, of the binding surface.

The FcαRI:Fcα interface is composed of a central hydrophobic core (FcαRI residues Y35, L54, F56, G84, H85 and the aliphatic portion of K55 packing against Fcα residues L257, L258, M433, L441, A442, F443 and the aliphatic portion of the R382 side chain) flanked by charged residues (FcαRI R52, R53, K55 and R82, and Fcα R382, E389 and E437) (Fig. 4b), burying a total of 1,656 Å2. Several potential intermolecular hydrogen bonds are found at the periphery of the site, but no salt bridges are observed, consistent with binding studies showing that <10% of the free energy of binding is contributed by electrostatic interactions12. The N-glycan attached to FcαRI N58 forms two potential hydrogen bonds and a van der Waals contact with Fcα (Fig. 4a). Although the observed N-glycans on Fcα approach within 8 Å of FcαRI, they do not directly contact the receptor.

The interaction surface identified by the FcαRI:Fcα structure can be used to interpret mutagenesis studies13,14,26,27,28 (Fig. 4c). Substitution of FcαRI Y35, R82 or Y81 to alanine, or H85 to arginine resulted in 100-fold loss of affinity or ablation of binding27,28. Y35 is located at the centre of the hydrophobic patch in the binding site and forms a potential hydrogen bond to the carbonyl oxygen of Fcα L257. R82 forms a hydrogen bond with the carbonyl oxygen of Fcα L256. Y81 is located in the hydrophobic core of FcαRI D1 and is likely to be essential for proper folding of the domain. H85 interacts with the Fcα FG loop. Mutations of FcαRI R52, I83 or G84 to alanine resulted in 8- to 19-fold losses of affinity28. R52 forms two potential hydrogen bonds with the carbonyl oxygen of L258, G84 contacts the FG loop, and I83 appears to stabilize the conformation of the FcαRI FG-binding loop, although it does not contact Fcα. Mutation of Fcα residues in the AB helix/loop (L257R or L258R), the FG loop (P440A, L441R, A442R, L441M/A442N) or the G strand (F443R) resulted in ablation of binding in rosetting assays13,14; these residues are all involved in the observed FcαRI:Fcα interface, and all but one contribute to the hydrophobic core of the binding site.

The FcαRI:Fcα structure reveals other residues in the binding site that had not been identified previously, including FcαRI residues in the D strand (L54, K55, F56, W57) and N58-linked sugar, and Fcα residues in the Cα2 AB helix/loop (L256), the Cα3 A (E348), C (R382, L384), F (M433, H436) and G (Q445) strands, and the CC′ (S387, E389) and FG loops (E437, A438, L439). In particular, FcαRI residues L54, K55 and F56 seem to be important for binding, as each makes three to five contacts with Fcα and contributes 16%, 9% or 10%, respectively, of the buried surface area in the interface (Supplementary Table 3).

Conformational changes on complex formation

Several conformational changes occur within FcαRI D1 on binding Fcα. Superposition of D1 in the free and bound receptors shows that the FcαRI D strand and the DE and FG loops adopt different conformations (Fig. 1d, e). The D strand in free FcαRI shifts from the front β-sheet (strands ABED) to the back β-sheet (strands A′GFCC′) on binding Fcα, forming instead a C′ strand in the complex. The main secondary structural elements of FcαRI that participate in binding to Fcα (the D/C′ strand and the DE/C′E and FG loops) show inherent flexibility in free FcαRI, and they adopt new conformations on binding (Fig. 1d). A comparison of the structure of FcɛRI in several crystal forms showed similar structural variability in the analogous regions of D2 (ref. 29). No significant change in the angle between FcαRI domains occurs on binding to Fcα; the D1–D2 angle when bound to Fcα (96°) is nearly identical to that observed for one molecule of free FcαRI (97°) and similar to the other two molecules of free FcαRI (88°, 92°).

Free Fcγ24 and Fcɛ22,23 show significant conformational changes on binding to FcγRIII11 or FcɛRI10, respectively. These conformational changes induce asymmetry in the Fc molecules when the FcR binds. By contrast, the Fcα dimer in the FcαRI:Fcα structure is two-fold symmetric. Since the structure of free Fcα is not known, it is not clear whether any conformational changes occur when FcαRI binds. However, the disulphide bonds at the top of the Cα2 domains and the large interface between the Cα3 domains burying a total of 2,061 Å2 would be expected to constrain the conformational freedom of Fcα.

Comparison to other FcR:Fc interactions

The Cα2–Cα3 interface of Fcα where FcαRI binds is analogous to the Cγ2–Cγ3 interface on Fcγ that serves as the binding site for a number of structurally distinct proteins, including FcRn15 (Fig. 3c), protein A30, protein G31, rheumatoid factor32 and Fc-III, a peptide selected for high-affinity IgG binding33. The binding of these proteins to Fcγ involves residues in the AB loop of the Cγ2 domain and the G strand of the Cγ3 domain that create a hydrophobic and highly accessible surface33. Although the FcαRI-binding site on Fcα also incorporates these regions, none of the residues is conserved exactly between FcαRI:Fcα and the other complexes. However, the Fcα and Fcγ binding surfaces share the common property of being primarily hydrophobic with outlying charged residues.

FcαRI is structurally similar to FcγRIII and FcɛRI, yet the FcαRI:Fcα binding interaction is completely different from the FcγRIII:Fcγ and FcɛRI:Fcɛ complexes, in which a single Fc receptor binds across both Cγ2 (or Cɛ3) domains in a region adjacent to the Fcγ hinge or the Fcɛ Cɛ2 domains10,11 (Fig. 3d, e). Furthermore, Fcα is bound to FcαRI in an ‘upright’ orientation with its C termini presumably near the membrane, whereas both Fcγ and Fcɛ are bound by their receptors in the opposite orientation10,11 (Fig. 3). The ‘upside-down’ orientation of receptor-bound IgG and IgE antibodies probably does not cause steric hindrance between the Fabs and the cell surface, because the hinge region of IgG antibodies is highly flexible34, and the Fc region of IgE has a sharp bend between the Cɛ2 and Cɛ3–Cɛ4 domains22. By contrast, IgA1 has a heavily O-glycosylated hinge region, which probably adopts a rigid conformation35. Thus, the IgA1 hinge is unlikely to be flexible enough to allow binding in the mode observed in the FcγRIII:Fcγ and FcɛRI:Fcɛ complexes. It is potentially relevant that IgA2 has a 13-residue deletion in the hinge region, eliminating the O-glycosylated region1; binding of FcαRI to the Cα2–Cα3 junction therefore allows identical interactions with both subclasses of IgA, which would be unlikely if FcαRI bound IgA in a hinge-proximal region.

Implications for signalling through FcαRI

Both monomeric and dimeric forms of serum IgA are able to bind and initiate immune effector functions through FcαRI5. The primary function of SIgA seems to be immune exclusion, an FcαRI-independent process in which SIgA binds to pathogens and sterically prevents their adherence to the mucosal epithelia1,2. SIgA is unable to trigger phagocytosis5,6, although it can initiate respiratory burst in polymorphonuclear leukocytes7, and degranulation of basophils and eosinophils8,36. Binding of polymorphonuclear leukocytes to immobilized SIgA7 and degranulation of eosinophils8 both show an absolute requirement for the integrin co-receptor Mac-1 (or its β-subunit, CD18), suggesting that SIgA is unable to bind or activate FcαRI alone without the integrin co-receptor. The inability of SIgA to bind or activate FcαRI can be explained based on the structural information presented here. Bound SC in SIgA interacts with IgA residues in the DE and FG loops37,38 and forms a disulphide bond with C311 of IgA3. As described above, the FG loop comprises a major part of the FcαRI-binding site on Fcα (contributing 23% of the total binding surface; Supplementary Table 3), and C311 is located only 10 Å away; therefore, bound SC occludes some or all of the FcαRI-binding site(s) in SIgA, apparently preventing binding and activation of FcαRI in the absence of the Mac-1 co-receptor.

FcαRI binds IgA with a 2:1 stoichiometry, as shown here by the co-crystal structure and by independent techniques in solution12. The 2:1 stoichiometry raises an interesting question: if IgA alone (in the absence of antigen) is able to dimerize FcαRI, what prevents constitutive activation of FcαRI by free IgA? There are several potential explanations. First, the C termini of the two FcαRI molecules in the FcαRI:Fcα complex are separated by 124 Å, which may be too far apart to register as a receptor dimerization event triggering downstream signalling. Second, the concentration of IgA in serum is typically 20 µM, which is 50- to 100-fold above the Kd for the FcαRI:IgA binding interaction12. Under such conditions, 1:1 complexes of FcαRI with free IgA would dominate. Due to the moderately fast on- and off-rates of the FcαRI:IgA binding reaction (2 × 105 (M s)-1 and 0.04 s-1, respectively)12, rapid sampling of free IgA would occur. By contrast, the multivalent nature of interactions between FcαRI and an IgA-immune complex would lead to a very slow off-rate, allowing stable binding and phagocytosis of an IgA-bound antigen even in the presence of high free IgA concentrations in serum. Finally, the cytoplasmic domain of FcαRI interacts with the cytoskeleton, and this interaction can be modulated by addition of cytokines such as granulocyte–macrophage colony-stimulating factor (GM-CSF)39. GM-CSF stimulates a rapid increase in phagocytosis and affinity of FcαRI for IgA40, indicating that cytokine priming may initiate a change in local FcαRI surface density and, therefore, avidity for IgA. These findings suggest that the tethering of FcαRI to the cytoskeleton could spatially restrict FcαRI, preventing clustering until priming of the immune cell with an activating cytokine occurs12.

Methods

Crystallization and data collection

The ectodomain of FcαRI and the Fc fragment of IgA1 were expressed and purified as described12. Crystals of FcαRI were grown in hanging drops containing 20% PEG-8000 and 0.1 M MES, pH 6.0 (FcαRI crystal form I; space group I41, two molecules in the asymmetric unit) or 20% PEG-8000 and 0.1 M sodium phosphate, pH 6.0 (FcαRI crystal form II; space group I212121, two molecules per asymmetric unit). Crystals of the FcαRI:Fcα complex were grown from 2:1 molar ratio mixtures of FcαRI and Fcα in 1.7–1.85 M magnesium sulphate heptahydrate and 0.1 M MES, pH 6.5–6.7, in space group P3 with one 2:1 FcαRI:Fcα complex per asymmetric unit. Data processing and scaling was carried out using the HKL package41 (Supplementary Table 1). Although the complex dataset was only 72% complete in the highest-resolution shell (3.2–3.1 Å), the data were 90% and 99% complete in the next-highest-resolution shells (3.3–3.2 Å and 3.5–3.3 Å, respectively).

Structure determination and refinement

Phases for crystal form I of FcαRI were determined by a combination of multiwavelength anomalous diffraction (MAD) and multiple isomorphous replacement with anomalous scattering (MIRAS) using trimethyl lead acetate and uranyl acetate derivatives. Five lead positions were determined by the program SOLVE42 using all three wavelengths in the MAD data set. Five uranium positions were then determined by anomalous difference Fourier analysis. MIRAS phases were calculated using SHARP43, giving an overall figure of merit of 0.24 from 25–3 Å. Solvent flipping with SOLOMON44 gave a readily interpretable map containing essentially the entire first FcαRI molecule and most of the second FcαRI molecule. Refinement was carried out using all reflections with F > 0 (no sigma cuts were applied) using cycles of torsional, positional and grouped B-refinement with a bulk solvent correction in CNS45 with the maximum likelihood Hendrickson-Lattman (MLHL) target using MIRAS phases without solvent flipping. The refinement initially used noncrystallographic symmetry (NCS) constraints; later stages incorporated NCS restraints (300 kcal mol-1 Å2), with separate NCS operators for the first and second immunoglobulin-like domains. Regions that differed between the two molecules were not restrained. The final model (Rcryst = 23.5%, Rfree = 29.1%) contains residues 6–198 of the first molecule; residues 2–111 and 123–193 of the second molecule; six N-acetyl glucosamine residues; and one Tris molecule. The overall B-factors by domain, including all protein atoms in the first FcαRI molecule (and the second molecule), are: D1, 35.0 Å2 (33.6 Å2) and D2, 31.7 Å2 (61.4 Å2).

The structure of crystal form II of FcαRI was determined by molecular replacement using AMoRe46 with coordinates from the form I structure; the molecular replacement search and subsequent rigid-body refinement gave significantly higher correlation coefficients and lower R-values in I212121 than in I222. Both molecules in crystal form II were subjected to rigid-body refinement followed by a single round of torsional refinement with a bulk solvent correction and NCS constraints (Rcryst = 35.2%, Rfree = 39.0%). The starting model had B-factors arbitrarily fixed at 40 Å2; other than the bulk solvent correction, no B-factor refinement was carried out. The resulting model was not refined further owing to the limited resolution of the data.

The FcαRI:Fcα complex was solved by molecular replacement using AMoRe46. A polyalanine version of Fcγ (1DN2.pdb33) and the refined structure of FcαRI (crystal form I) were used as search models, yielding a clear solution in space group P3. The model for a half-complex (one Fcα heavy chain and one FcαRI molecule) was subjected to rigid-body refinement followed by torsional, positional and grouped B-factor refinement using all reflections with F > 0. We incorporated a bulk solvent correction and used NCS constraints initially and NCS restraints (300 or 200 kcal mol-1 Å2) in later rounds, yielding a final model (Rcryst = 25.1%, Rfree = 28.4%) that includes residues 242–450 of each Fcα heavy chain; residues 6–195 of FcαRI; 20 N-acetyl glucosamine, 20 mannose, two fucose and two sialic acid residues. The overall B-factors by domain for all protein atoms in the half-complex (and the NCS-related half-complex) are: Fcα Cα2, 88.5 Å2 (89.4 Å2); Fcα Cα3, 71.4 Å2 (72.3 Å2); FcαRI D1, 77.0 Å2 (79.3 Å2); FcαRI D2, 84.0 Å2 (82.2 Å2).

For analyses of interdomain angles, contacts and buried surface areas, FcαRI D1 was defined as residues 6–100 and D2 was defined as residues 101–198; Fcα Cα2 was defined as residues 242–342 and Cα3 was defined as residues 343–450. Interdomain contact residues were defined as residues within 4 Å of another domain and identified using CNS45. Potential hydrogen bonds were identified according to established criteria47, using the program HBPLUS47. Buried surface areas were calculated using AREA from the CCP4 program suite44 with a 1.4 Å probe. Angles between domains were calculated using DOM_ANGLE48, which determines the angle between the long axes of adjacent domains that are approximated by ellipsoids calculated from the coordinates. Figures were generated using Molscript49 and Raster3D50.