Structure and flexibility of the extracellular region of the PirB receptor

Murine paired immunoglobulin receptor B (PirB) and its human ortholog leukocyte immunoglobulin-like receptor B2 (LILRB2) are widely expressed inhibitory receptors that interact with a diverse set of extracellular ligands and exert functions ranging from down-regulation of immune responses to inhibition of neuronal growth. However, structural information that could shed light on how PirB interacts with its ligands is lacking. Here, we report crystal structures of the PirB ectodomain; the first full ectodomain structure for a LILR family member, at 3.3–4.5 Å resolution. The structures reveal that PirB's six Ig-like domains are arranged at acute angles, similar to the structures of leukocyte immunoglobulin-like receptor (LILR) and killer-cell immunoglobulin-like receptor (KIR). We observe that this regular arrangement is followed throughout the ectodomain, resulting in an extended zigzag conformation. In two out of the five structures reported here, the repeating zigzag is broken by the first domain that can adopt two alternative orientations. Quantitative binding experiments revealed a 9 μm dissociation constant for PirB–myelin-associated glycoprotein (MAG) ectodomain interactions. Taken together, these structural findings and the observed PirB—MAG interactions are compatible with a model for intercellular signaling in which the PirB extracellular domains, which point away from the cell surface, enable interaction with ligands in trans.

Ig-like domains. Instead of the six Ig-like domains for mouse PirB, the human LILRB2 ectodomain consists of only four Iglike domains. The first and third domain of PirB are most similar to the first domain of LILRB2 (35 and 43% sequence identity, respectively), whereas the second and fourth domains of PirB are most similar to the second domain of LILRB2 (56 and 43% sequence identity, respectively). The fifth and sixth domain of PirB are most similar to the third and fourth domain of LILRB2 (64 and 57% sequence identity, respectively). Although mouse PirB and human LILRB2 differ in domain number, they interact with a similar set of ligands and have similar functions (1,6,10) Despite the functional importance of PirB, there is no structural information available. To date, the only known structures for LILR family members consist of two consecutive Ig-like domains. For example, two partial ectodomain structures have been published for LILRB2, one of the first two domains, D1-D2 (23), and one of the last two domains, D3-D4 (24). Insight in the structural features of PirB will aid understanding of how the protein is able to interact with such diverse ligands and trigger signaling to carry out its diverse range of functions.
Here we present the crystal structure of the PirB ectodomain and investigate its interactions with the extracellular segment of the MAI MAG. The six Ig-like domains of PirB are arranged in an extended repeating zigzag arrangement. The PirB N-ter-minal domain, D1, has positional flexibility and adopts three alternative orientations in which the angle between domains D1 and D2 varies from 75°to 315°. Using surface plasmon resonance (SPR) experiments we show that the dissociation constant for the PirB-MAG ectodomain interaction is 9 M. The extended conformation of PirB may enable trans-cellular interaction with ligands, such as MAG and MHC-I, to induce intercellular signaling.

PirB adopts an extended zigzag structure
The crystal structure of the PirB ectodomain containing all 6 Ig-like domains (PirB 1-6 ) reveals an extended zigzag conformation with multiple orientations for the N-terminal Ig-like domain ( Fig. 1 and Fig. S1). The PirB 1-6 structure was determined in three crystal forms, denoted PirB cryst1 , PirB cryst2 , Pir-B cryst3 , with maximum resolutions of 3.3, 3.4, and 4.5 Å, respectively (Table 1). In two of the crystal forms, PirB cryst2 and PirB cryst3 , the asymmetric unit contains two monomers, resulting in a total of five unique structures. The six tandem Ig-like domains are arranged in a zigzag shape, with only minor deviations between the five structures (Fig. 2, A-C, and Fig. S2). From tip to tip (Tyr-104 -C␣ to Ser-536 -C␣), the structure measures 146 Å in the extended conformation found in Figure 1. Crystal structure of PirB 1-6 reveals a semi-rigid zigzag of tandem Ig-like domains. A, the PirB ectodomain forms an elongated structure with a repeating zigzag of Ig-like domains D1 to D6. Glycans are shown in stick representation. The model depicted is from PirB cryst1 ; the light blue part of D6 is missing in the density, and has been extended here for viewing purposes using D6 of PirB cryst2-B . B, topology diagram for D2; ␤-sheets (arrows) and polyproline helices (gray cylinders) are indicated. Topology diagrams for the other domains can be found in Fig. S1. C, ribbon drawing of D2, using the same color scheme as in B. D-F, overlay of all six of PirB's Ig-like domains (D1 to D5 from PirB cryst1 and D6 from PirB cryst2 ) illustrates the close structural similarity between domains and domain interfaces. Residues involved in interactions with the previous domain (forming the C-terminal face of the interface) are colored purple, and residues involved in interactions with the next domain (forming the N-terminal face of the interface) are shown in pink. This color scheme is illustrated in the cartoon (G).

Structure of the PirB ectodomain
PirB cryst1 . Glycans are observed in four of the six predicted N-linked glycosylation sites (Asn-338, -479, -500, and-531), in at least one of the five structures. In PirB cryst1 domain D6 is partially disordered; residues 528 -549 and 567-593 are not visible in the electron density and were omitted from the model. The best-defined structure for domain 6 is observed in the second monomer (chain B) of PirB cryst2 (PirB cryst2-B ), therefore this structure is used for the analysis of D6 outlined below, additionally this structure is used to supplement the PirB 1-6 structure shown in Fig. 1A.
The six Ig-like domains are composed of two antiparallel ␤-sheets, linked together by a disulfide bridge. The six domains are topologically similar to each other ( Fig. 1, B-F, and Fig. S1), and to the Ig-like domains of KIR and other LILR proteins (23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35)(36). ␤-Sheet 1 comprises strands A, B, D, and E and sheet 2 comprises strands AЈ, C, CЈ, F, and G. As is often seen in LILR and KIR family members, the A-AЈ strand is shared between the two ␤-sheets. Additionally, there is a polyproline helix before ␤-strands A and G and a 3 10 -helix in the E-F loop. These secondary structure elements are not present in all domains, for instance, D1, D3, and D6 lack ␤-strand D, in D5 and D6 the G and AЈ strands are unraveled, and in D2 and D6 the 3 10 -helix is missing. Moreover, ␤-strand G is very short in D2 and it does not interact with ␤-strand F, leaving a separate parallel ␤-sheet composed of strands AЈ and G. Despite the differences in secondary structure elements, the three-dimensional structures of all six Ig-like domains are very similar; in PirB cryst1 they have an average r.m.s. deviation on C ␣ 's of just 1.6 Ϯ 0.3 Å (Fig. 1, D-F).

Domain interfaces
Consecutive Ig-like domains interact with each other through topologically equivalent interfaces (Fig. 1, D-F) that are also observed in the two or three domain containing structures for LILR and KIR proteins (23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35). In PirB this "canonical" domain interface is repeated five times, resulting in the observed extended zigzag shape (Fig. 1A). In each interface, the most N-terminal domain contributes residues of the AЈ strand, G strand, and E-F loop (Fig. 1F), whereas the C-terminal side of the interface (located on the following domain) is centered on ␤-sheet 2 and comprises the F, C, and CЈ ␤-strands, the C-CЈ loop, and the polyproline helix preceding strand G (purple in Fig. 1E). These interactions leave sheet 1 (strands A, B, D, and E) and the N-terminal tip of the domain (strand D and connecting loops and the B-C loop) available for interactions with ligands.
The domain interfaces are stabilized by hydrophobic interactions, hydrogen bonds, and salt bridges. On average, the total surface area buried in an interface is 1088 Ϯ 158 Å 2 . All interfaces, except the D4 -D5 interface, have a hydrogen bonding network between residues in strand G or loop E-F on the N-terminal domain to residues in loop F-G on the C-terminal domain. Additionally, in each interface a salt bridge or hydrogen bond is formed between a residue in strand G (or in loop A-AЈ in D1) and a residue in strand CЈ, located at the edge of the interface. Interaction through this canonical interface results in an acute angle between consecutive domains. The orientation is slightly different for each domain pair, with interdomain angles ranging from 62°to 79° (Fig. 2, A-C, and Fig. S2).

PirB domain D1 can adopt three different orientations
In two of our PirB structures D1 and D2 do not interact via the canonical interface, resulting in two PirB conformations that deviate from the regular zigzag described above (Fig. 2, A-C, and Fig. S3). In both monomers in PirB cryst2 D1 is bend over to the other side of D2, predominantly rotating around the Trp-122 C-C␣ bond, resulting in an interdomain angle of 315 o in chain A and an angle of 269 o in chain B (Fig. 2, B and C).
The three distinct D1 orientations give rise to three different D1-D2 interfaces (Fig. 2, D-F). The canonical D1-D2 interface, seen in PirB cryst1 , is smaller than the other canonical interfaces, it has an average total buried surface area of 903 Ϯ 51 Å 2 versus 1145 Ϯ 107 Å 2 for the other domain interfaces. The canonical D1-D2 interface consists of a hydrophobic core centered around Trp-208 in strand G of D2, together with a hydrogen bonding network between the backbones of Val-117 and Gly-119 in strand G of D1, and Tyr-206 and Trp-208 in strand G of D2 (Fig. 2D). The domain orientation is further stabilized by a salt bridge between Asp 36 in the A-AЈ loop of D1, and Lys-162 in strand CЈ on D2.
The two noncanonical interfaces are even smaller, with only 542 Å 2 total buried surface area in monomer A and 436 Å 2 in monomer B. The noncanonical interfaces on D1 partially overlap with the canonical interface. In monomer A, this interface comprises strand AЈ, loop E-F, and the domain linker; all but one (Val-39) of the D1 residues are also involved in the canonical interface (Fig. 2E). In monomer B, the interface on D1 comprises loops AЈ-B and E-F, and the domain linker (Fig. 2F). On D2 the noncanonical interfaces are very different from the canonical one. In chain A, Trp-147 in the B-C loop replaces Trp-208 as the central residue in the hydrophobic interface (Fig. 2E). Moreover, residues in the linker region between the domains contribute to the interface; His-120 and Tyr-121 line the interface and Trp-122 interacts through ring stacking with His-148 in loop B-C on D2. In chain B, the hydrophobic part of the interface is formed by Trp-42 in loop AЈ-B of D1, and Trp-147 in loop B-C of D2 (Fig. 2F). Both noncanonical interfaces are further stabilized by salt bridges and hydrogen bonds, although the hydrogen bonding network is not as extensive as the one seen in the canonical D1-D2 interfaces.

Structure of the PirB ectodomain
In summary, our crystal structures show three distinct orientations for PirB D1. Chain B in PirB cryst2 seems to represent an intermediate state between the canonical conformation and the conformation of PirB cryst2 chain A. On the other hand, the main body of the protein, D2-D6, shows very little conformational variation. Taken together, the PirB structures indicate that whereas there is limited flexibility between domains D2-D6, a range of orientations is possible for D1.

PirB self-association
PirB cryst1 and PirB cryst2 reveal dimers with a common intermolecular interface that may represent a PirB cis dimer (Fig. 3). The interface has a total buried surface area of 3531 Å 2 in Pir-B cryst1 and 3708 Å 2 in PirB cryst2 , whereas the area of the second largest interfaces in these crystals are 1991 and 1609 Å 2 , respectively. In the third crystal form this dimer is not observed; here the largest interface is 2087 Å 2 . The putative dimerization interface consists of two patches. The first patch starts at the C-terminal tip of D2 and continues across D3 to the N-terminal tip of D4. The second patch comprises residues at the C-terminal tip of D4 and the N-terminal tip of D5. D1 is not involved in this putative dimerization interface; therefore, the interface is not affected by the observed mobility of this domain.
PirB shows self-association in SPR (Fig. S4). Although, homointeractions can be detected by SPR they cannot be quantified using this technique as self-association of ligands coupled on the chip surface compete with analyte binding. Despite this, both the crystal structures and the SPR data indicate that PirB has propensity to self-associate.

Ligand binding
Ligand interaction is critical to PirB's role as an extracellular receptor. To verify the interaction of the PirB ectodomain for ligand binding, SPR experiments were performed for PirB's known binding partner MAG (Figs. S5 and S6). C terminally biotinylated PirB 1-6 was immobilized at three different densities on a streptavidin-coated SPR chip, to probe interactions with increasing concentrations of the extracellular segment of MAG in the mobile phase. The entire extracellular MAG segment, containing five extracellular Ig-like domains (MAG 1-5 ), was used. MAG 1-5 -PirB 1-6 binding was found to have a K D of 9 Ϯ 2.9 M (Fig. S6).

Discussion
PirB is a versatile receptor that binds to numerous, structurally varied ligands. We have solved the structure of PirB's ectodomain in three conformations and performed binding experiments to gain insight into PirB-ligand binding. What sets the PirB ectodomain structure apart from other structures of LILR and KIR family members is that it is the largest of its kind to be elucidated and it is the only reported KIR/LILR structure with multiple orientations for one of its domains.
The six Ig-like domains that make up PirB's ectodomain are arranged in a regular zigzag repeat. Interestingly, the first domain can break this regular pattern and adopt multiple orientations. In our crystal structures three distinct orientations are observed for PirB D1. In all three crystal forms the D1 domain is involved in crystal packing and this may stabilize conformations of PirB that are only sparsely populated in solution. The conformations in PirB cryst1 and PirB cryst2-A seem to represent the extremes in a range of conformations, whereas PirB cryst2-B represents an intermediate state between the two. To our knowledge flexibility of one of the Ig-like domains has not been reported for any of the LILR or KIR family members. Furthermore, there are no indications of such flexibility in the structures for LILR and KIR family members. This is surprising, as, at first glance, the PirB D1-D2 interface is very similar to the D1-D2 interfaces of other LILR family members, by comparing the sequences or the available structures.
The conformational variability in PirB may be explained by the size of the interface and a number of residues unique to PirB. The canonical D1-D2 interface is less extensive than the other canonical interfaces in PirB, with an average total buried surface area of 903 Ϯ 51 Å 2 instead of 1145 Ϯ 107 Å 2 for the other interfaces. However, it is in the same range as the D1-D2 interface in other LILR structures, where the total buried surface area is 839 Ϯ 37 Å 2 (23, 34 -36). Moreover, the threedimensional organization of the interface is very well conserved between the LILRs and PirB. On closer inspection, four residues unique to PirB, but with important roles in the three different interfaces (Fig. 2, D-F) stand out; His-120 and Trp-42, -122, and -147. His-120 is part of the linker region between D1 and D2 and points outward into the solvent in the canonical structure. The corresponding residue in other LILR family members (human LILRs A1, A2, A3, A4, B1, B2, B3, and B4 and mouse LILRB4), is an alanine that does not participate in the D1-D2 interface, or a phenylalanine (human LILRs A5, A6, and B5 and chimpanzee LILR A6 and B5), that engages in extensive hydrophobic interactions with domain 1. As His-120 is bulky, unlike

Structure of the PirB ectodomain
the much smaller alanine, and is not able to contribute to the interface with hydrophobic interactions like phenylalanine, it might destabilize the PirB cryst1 D1-D2 interface. The three tryptophans unique to PirB (42,122, and 147) are not involved in the canonical interface, instead they provide hydrophobic patches for D1 interaction in the noncanonical orientations. Because of the location of Trp-147 in loop BC, which is of variable length, the corresponding residues in other LILR family members can only be determined by structural comparison. In LILRs with published structures the corresponding residues are glutamine or arginine, with extended hydrophilic side chains that are consistent with exposure to the solvent. For Trp-42 the corresponding residue in other LILR family members is always an arginine and for Trp-122 the corresponding residues are either hydrophilic (Glu, Ser, Asn) or small (Pro, Ile, Ala). These corresponding residues would not be able to interact with D1 in the same manner as Trp-42, -122, or -147. In conclusion, three tryptophans that are unique to PirB provide alternate interfaces for D1-D2 interaction and, together with His-120, may promote conformational mobility of D1.
Our crystal structures provide evidence for PirB dimerization, and SPR data also indicate that PirB 1-6 has a propensity for self-association. Although our experimental data on PirB dimerization are not conclusive, oligomerization (e.g. dimerization) could be part of the signaling mechanism in (some of) PirB's roles as a receptor when expressed on the cell surface. Nonetheless, to our knowledge, dimerization has not been documented to have a functional role for any of the LILR family members, nor for the related KIR proteins. Also, there is no evidence for dimerization in the (partial) structures of LILR and KIR proteins. PirB 1-6 is separated from the cell surface by a 19-amino acid linker that may confer flexibility to the orientation of PirB with respect to the membrane. Therefore, whereas the maximum distance that the six Ig-like domains can span in the extended PirB conformation is 146 Å, it is not clear if PirB projects that far from the membrane. If PirB adopts a dimer conformation as observed in PirB cryst1 and PirB cryst2 the orientation relative to the membrane becomes much more restricted. A PirB dimer in the extended zigzag conformation (PirB cryst1 ) would protrude 130 Å from the plane defined by the most membrane-proximal tips of the protein (Fig. S7). For the noncanonical orientations of D1, this distance is less; 100 Å in the fully flipped conformation (PirB cryst2-A ) and 120 Å in the intermediate state (PirB cryst2-B ).
Neuronal PirB interacts with MAIs, expressed on oligodendrocytes, in trans. Although interactions of MAIs with the Nogo Receptor (NgR) have been well studied (37,38), less is known about binding of these proteins to PirB. Binding affinities of 14 nM have been reported for Fc-tagged MAG to PirB (10), but this interaction is most likely enhanced by artificially dimerizing MAG. Indeed, weaker interactions of 33 M have been reported for binding of untagged MAG to PirB (39) although in their experiments the maximum concentration of MAG used, 3 M, is too low for accurate affinity determination. Using higher MAG concentrations we obtain an affinity of 9 M for the MAG-PirB interaction.
There are no crystal structures available of complexes involving MAG and PirB to provide additional clues to how these proteins might interact (Fig. 4A). Structures are available of LILR family members in complex with MHC-I loaded with a (viral) peptide. The LILR-MHC-I trans binding mode is highly conserved among all studied complexes (40). As a member of the LILR family, PirB is likely to display the same binding mode to form complexes with MHC-I (Fig. 4B). Interestingly, PirB would only be able to bind MHC-I complexes in the extended zigzag conformation; in the noncanonical conformations, seen in PirB cryst2 , D1 would clash with the MHC-I complex. PirB dimerization as observed in PirB cryst1 and PirB cryst2 on the other hand would not interfere with MHC-I binding.
In conclusion, our data reveal that the extracellular segment of PirB is extended and that PirB has a propensity to self-associate. The dimerization mode revealed in the crystal structures of PirB 1-6 is compatible with dimer formation of full-length transmembrane PirB on the cell surface in cis, but the importance of PirB dimerization for signaling has not been investigated further. In full-length PirB the mobile N-terminal domain D1 is likely positioned furthest away from the cell surface, poised for interaction with ligands on other cells. Possibly, the mobility of D1 contributes to its ability to interact with a diverse set of binding partners. Taken together, our structural and interaction data are compatible with a model for intercellular signaling in which PirB has an extended conformation on the cell surface to enable interaction with ligands in trans (Fig. 4).

Cloning of constructs
The soluble ectodomain constructs for PirB and MAG were generated using polymerase chain reaction (PCR). The templates and construct boundaries are listed in Table 2. All constructs were subcloned into pUPE107.30 (U-Protein Express) (C-terminal His 6 tag) using BamHI/NotI restriction sites. For SPR experiments the original PirB 1-6 construct was subcloned into pUPE107.62, a vector containing a C-terminal His 6 tag for purification and a C-terminal biotin acceptor peptide tag for biotinylation.

Expression and purification
The soluble ectodomain constructs (in vector pUPE107.30) were expressed by transient transfection in HEK293-ES cells (U-Protein Express) and grown in FreeStyle 293 Expression Medium (Thermo Fisher). HEK293-ES is a cell line that lacks the N-acetylglycosaminyltransferase I enzyme and therefore produces proteins with short, homogeneous oligomannose glycans (41). After 6 days, cells are spun down (10 min at 1000 ϫ g) and the medium was harvested. The medium was concentrated 10-fold and diafiltrated against IMAC binding buffer (500 mM NaCl, 50 mM Hepes pH 7.8) using a Quixstand benchtop system (GE Healthcare). Cell debris was removed by centrifugation (30 min at 9500 ϫ g). The protein was purified from the cleared concentrate by Ni-affinity chromatography on a 5-ml HiTrap column (GE Healthcare) followed by SEC on a Superdex 200 column (GE Healthcare) equilibrated with SEC buffer (150 mM NaCl, 25 mM Hepes, pH 7.5). The purified protein was concentrated using a spin concentrator with the appropriate MWCO, snap frozen in liquid nitrogen, and stored at Ϫ80°C until use.

Structure of the PirB ectodomain
PirB 1-6 for SPR was expressed in HEK293-ES cells (U-Protein Express), in small 4-ml cultures. To allow in vivo biotinylation of these constructs, the cells were co-transfected with the Escherichia coli BirA biotin ligase and the cell medium was supplemented with 25 g/ml of biotin. After 6 days, the medium was harvested as described above and diluted 10-fold with IMAC binding buffer. The protein was purified from the diluted medium by batch binding to Ni-Sepharose beads (GE Healthcare). After incubation at room temperature for 2 h, the beads were transferred to spin columns (Thermo Fisher) and washed 4 times with binding buffer, followed by elution with 50 l of binding buffer supplemented with 200 mM imidazole.

Crystallization and data collection
Crystals of PirB 1-6 were grown at 20°C using the hanging drop vapor diffusion method and a concentration of 9 to 10 mg/ml of PirB 1-6 . PirB cryst1 , in space group P4 1 22, was obtained against a reservoir solution of 1.1 M LiCl, 14% PEG 6000, 0.1 M citric acid buffer, pH 5.5. A dataset to 3.3 Å resolution was collected at Diamond light source beamline I04 (wavelength 0.9795 Å, temperature 100 K). PirB cryst2 , in space group P2 1 , was obtained against a reservoir solution of 1.2 M LiCl, 12% PEG 6000, 0.1 M citric acid buffer, pH 5.5. A dataset to 3.4 Å resolution was collected at Swiss light source beamline PX-I (wavelength 1.0000 Å, temperature 100 K). PirB cryst-3 , also in space group P2 1 , was obtained against a reservoir solution of 14.6% polyacrylate 5100 sodium salt and 0.07 M Tris, pH 8.0. An incomplete dataset to 4.5 Å was collected at Swiss light source beamline PX-I (wavelength 1.0000 Å, temperature 100 K). The missing degrees were collected at ESRF beamline ID23-1 (wavelength 0.9686 Å, temperature 100 K). The two datasets were merged in AIMLESS and further processed as described below.

Structure solution and refinement
The data were integrated using iMOSFLM (42) or XDS (43), followed by scaling and merging using the AIMLESS pipeline (44). In brief, the strategy for solving and refinement of the structures was as follows; the three structures were solved in the  (23). HLA-G is composed of an ␣ chain (green) and ␤2-microglobulin (cyan) and is loaded with a peptide (pink). In A and B the PirB structure shown is from PirB cryst1 ; the semi-transparent part of D6 is missing in the density, and has been extended here for viewing purposes using D6 of PirB cryst2-B . The number of amino acids between the structured part and the membrane attachment site is indicated near each linker. Structure of the PirB ectodomain order in which the datasets were collected (first PirB cryst3 , then PirB cryst2 , and finally PirB cryst1 ). Each time, the previously solved and partially refined structure was used as a molecular replacement (MR) model for the next dataset. Resulting in the "best" PirB structure from PirB cryst1 . To improve the other two structures, the domains of the refined PirB cryst1 structure were placed back in density for PirB cryst2 and PirB cryst3 followed by further refinement to yield the structures reported here. The structure for PirB cryst3 was solved with MR in Phaser (45), using structures of LILRA5 (35) and LILRB2 (24) as MR models. The structure was refined using iterative rounds of manual building in Coot (46) and refinement in REFMAC 5 (47). The resulting structure was used as a model for MR to solve the structure of PirB cryst2 with Phaser. To prevent bias in the domain orientations, single domains were used as MR models. The resulting structure was refined in Coot and REFMAC. The same domain-by-domain approach was used to solve the structure from PirB cryst1 , with PirB cryst2 as a MR model. The new structure was refined in REFMAC and phenix.refine (48) and manual model building in Coot.
To improve the structure from PirB cryst2 , D1-D5 of this structure were substituted with D1-D5 from PirB cryst1 . This was done by individually superimposing the domains, followed by extensive refinement in REFMAC, phenix.refine, and manual model building in Coot. A similar approach was used to improve the structure from PirB cryst3 , domains from the higher resolution PirB cryst1 (D1 and D2-D5) and PirB cryst2 (D6) were used to rebuild the structure. To avoid overfitting of the low resolution PirB cryst3 structure, refinement was kept to a minimum. After an initial round of rigid body refinement (one body per domain) in phenix.refine, a limited number of adjustments were made in Coot. Namely, the missing domain connections were added; sugars were added where they were visible in the density; His-120 and Tyr-121, which are Ramachandran outliers in the model because of a crystal contact, were fixed; and loop 170 -178, which is displaced in the PirB cryst3 structure because of a crystal contact, was modeled in the visible density using the corresponding region in PirB cryst2-A as a template. After this model building the structure was subjected to one round of jelly body refinement in REFMAC and one more round of rigid body refinement in phenix.refine.

Structure analysis
The domain interfaces for all five unique monomers as well as the dimer interfaces were analyzed using the PISA server (49). Angles between Ig-like domains were determined from the angles between the largest principle axes of these domains. Glycosylation predictions were preformed using the NetNGlyc 1.0 Server and the NetOGlyc 4.0 Server (50). Structure r.m.s. deviations were calculated in gesamt (51), using the C␣ atoms for alignment. All figures of the structures were generated using PyMol (DeLano Scientific LLC).

Surface plasmon resonance
SPR experiments were performed on an MX96 (IBIS Technologies), using a SensEye Sensor (IBIS Technologies) with a streptavidin-coated dextran matrix. Biotinylated PirB 1-6 was coupled to the chip as the SPR ligands using a Wasatch Micro-fluidics continuous flow microspotter. This method creates multiple spots, or regions of interest on the chip surface, each with a ligand density of choice. Purified analyte (MAG 1-5 or PirB 1-6 , Figs. S4 and S6) was flowed over the chip at a constant temperature of 25°C in running buffer (150 mM NaCl, 20 mM Hepes, pH 7.2, 0.001% Tween 20). Equilibrium binding experiments were performed to measure the binding affinities of MAG to PirB. Using the analyte in a concentration range of 0.8 -108 M for MAG [1][2][3][4][5] .
Data processing was started in SprintX (IBIS Technologies), where the data were blanked once, using reference spots close to the regions of interest. The data were then zeroed before each injection of analyte, and exported to Scrubber (BioLogic). In Scrubber the amount of bound MAG 1-5 was determined when equilibrium is reached at the end of the association phase (Fig. S6). A saturation curve was fitted with a 1:1 Langmuir binding model in GraphPad Prism to determine the maximum analyte binding (B max ) and equilibrium binding constant (K D ) of MAG 1-5 -PirB 1-6 interaction.