Factor XII and kininogen asymmetric assembly with gC1qR/C1QBP/P32 is governed by allostery

The contact activation system lies at the crossroads of plasma coagulation and innate immunity and consists of proteases factor XII (FXII), prekallikrein (PK), and cofactor high-molecular-weight kininogen (HK).^1-3  Binding of contact factors to the cell surface has been shown to be mediated by the complement C1q receptor (gC1qR; also known as C1QBP, HABP1, and P32) and the urokinase receptor (uPAR).^4-7  gC1qR is a multicompartmental and multifunctional protein essential for mitochondrial function,⁸  but it is also present at the surface of stimulated cells.^9-11  gC1qR has no plasma membrane anchor but forms interactions with other cell surface proteins and receptors (cytokeratin-1, β₁-integrin, dendritic cell–specific ICAM-3–grabbing nonintegrin receptor, and fibrinogen).^4,12-15  The gC1qR crystal structure is a doughnut-shaped symmetrical trimer with both a highly acidic ligand-binding surface and a cell-binding face.¹⁶  gC1qR has been characterized as binding a diverse array of structurally distinct ligands^7,9,15,17,18  and has been proposed to function as a chaperone directing the assembly of multiprotein complexes.¹⁹ 

The domain structures of FXII, HK, and gC1qR are shown in Figure 1A. Crystal structures have been determined for the FXII protease domain^20-22  and the fibronectin type I (FnI) and epidermal growth factor-like domains.²³  As the N-terminal FXIIFnII domain is central to understanding processes of FXII conformational regulation²⁴  we determined the crystal structure in complex with gC1qR. This revealed an asymmetric FXII-binding mode and a novel Zn²⁺-binding site in the gC1qR structure.

The human gC1qR (residues 74-282 with Leu74 substituted to Met74) Escherichia coli expression vector was a gift from Adrian R. Krainer (Cold Spring Harbor Laboratory). Expression and purification of gC1qR was performed as described previously.¹⁶  gC1qR variants were generated by site-directed mutagenesis using the Agilent Technologies QuikChange Kit. HKD5 was expressed in E. coli using the pET28a vector, and D5-1, D5-2, and HK 401-438 were expressed as glutathione S-transferase fusion proteins using the pGEX 4T-1 vector (supplemental Figure 1, available on the Blood Web site). Untagged FXIIFnII domain (residues 1-71) was cloned into vector pMT-PURO for expression in the insect cell–based DES system (Invitrogen) using previously described protocols.²²  FXIIFnII was initially purified from media using a Capto-S column (GE Healthcare) equilibrated with 0.05 M 2-(N-morpholino)ethanesulfonic acid (pH 6.0), and a gradient of 0 to 1.0 M NaCl was used for elution. Subsequently, this was applied to a HiTrap Ni²⁺ column (GE Healthcare) and eluted using an imidazole gradient concentration of 0 to 1.0 M and a final purification step of gel filtration with a HiLoad Superdex 75 16/60 column (GE Healthcare) in 0.05 M Tris-HCl (pH 8.0) and 0.1 M NaCl (supplemental Methods).

FXIIFnII was mixed with gC1qR and 50 μM ZnCl₂, and the complex was isolated using a Superdex increase 200 10/300 column equilibrated with 0.02 M N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid (pH 7.4) and 0.14 M NaCl (Figure 1C). FXIIFnII-gC1qR complex fractions were collected and concentrated to 5.3 mg/mL. Crystals were obtained in 0.1 M NaCacodylate (pH 5.5), 0.1 M CaAc₂, and 12% (w/v) PEG 8000 at 10°C. A data set for a single FXIIFnII-gC1qR complex crystal was collected and processed using the CCP4 suite and the structure determined by molecular replacement with Phaser²⁵  and the gC1qR structure (Protein Data Bank [PDB] accession number 1P32) as a template. Model building (COOT) and refinement (REFMAC)²⁶  provided a final refined model containing the gC1qR trimer, FXIIFnII, and 3 Zn²⁺ (Table 1).

Plasma-purified FXII was purchased from Enzyme Research Laboratories and immobilized onto a CM5 chip using an Amine Coupling Kit (GE Healthcare), and experiments were performed on a Biacore 3000 instrument. The running buffer was 0.02 M N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid (pH 7.4), 0.14 M NaCl, and 50 μΜ ZnCl₂, and the chip surface was regenerated with 1M NaCl and 0.02 M EDTA. To assess any nonspecific binding, the analyte (gC1qR) was also injected over an empty flow cell. Binding curves were analyzed on the basis of the SPR response units recorded at equilibrium for each analyte protein concentration, and a Hill plot was generated using Prism 6 (GraphPad Software).

A MicroCal VP ITC system (Malvern) was used for all ITC experiments performed at 25°C. The ITC cell contained 1.4 mL of wt-gC1qR or gC1qR variant while 300 μL of D5 ligand was loaded into the syringe. The reference power was set at 5 μcal/s, and the syringe stirring speed was set at 300 rpm. An initial pre-equilibration step of 1 hour was followed by 30 × 10 μL injections. Ligand dilution effects were tested by running a ligand to buffer control using the same titration parameters. The ligand to buffer control was subtracted from the experimental data, and any anomalous titration points were removed. Curves for D5-2 and HK 493-516 were fit to a 1-binding-site model, whereas curves for D5 and D5-1 were fit to a 3-site sequential binding model (Table 2).

Blood was drawn from healthy volunteers into vacuettes (Greiner Bio-One) containing 3.2% sodium citrate. To isolate platelet poor plasma, tubes were spun at 1860g for 30 minutes at 4°C. Pooled normal plasma (PNP) was derived from ≥20 donors, portioned in aliquots, and stored at −70°C. Ethical approval was obtained from the University of Aberdeen College Ethics Review Board. The activated partial thromboplastin time assay was performed in a STart 4 coagulometer (Diagnostica Stago). PNP ± gC1qR (50-200 µg/mL) was incubated with Zn²⁺ (50 µM) at 37°C for 180 seconds. Clotting was initiated by addition of CaCl₂ (0.0083 M). Thrombin generation was measured in Calibrated Automated Thrombinoscope. PNP or FXII- or FXI-deficient plasma (Hypen Biomed) ± gC1qR (50-200 µg/mL) and Zn²⁺ (50 µM) was added to thrombin calibrator and MP reagent (Stago United Kingdom) in Greiner 96-well plates. The plate was incubated for 10 minutes followed by addition of FluCa solution, as per the manufacturer’s guidelines. Thrombinoscope software package (Synapse Bv) was used to quantify real-time thrombin activity.

Recombinant FXIIFnII domain (residues 1-71) was produced in insect cells and shown using gel filtration to bind gC1qR in the presence of 50 μM ZnCl₂ (Figure 1B-C). Complex formation was not observed at lower ZnCl₂ concentrations, consistent with the previous report by Joseph et al on the Zn²⁺ dependency of FXII binding to gC1qR.⁷  The FXIIFnII-gC1qR structure was determined to 3Å resolution (Table 1 and Figure 1D). Figure 1E shows FXIIFnII has an asymmetric binding mode and unexpected stoichiometry of 1 FXIIFnII bound to the gC1qR trimer. The asymmetry of binding is achieved in part by the Arg36 and Arg65 side chains of FXII reaching out via the guanidinium groups to form contacts with distinct gC1qR negatively charged pockets termed G1 and G2 (supplemental Videos 1 and 2). gC1qR pocket G1 consists of anionic residues 190 to 202 (termed the G1-loop) together with residues Tyr236 from the αB helix and Trp233, Asp229 from the αB-β7 loop burying a surface area of 489 Ų. The FXIIFnII β2 strand forms main-chain to main-chain interactions resulting in an antiparallel β-sheet contact with gC1qR residues Glu198-Ser201 (Figure 2A-B). Electrostatic complementarity occurs via negatively charged side chains from gC1qR (Asp197, Asp229) forming salt bridge interactions with FXII side chains Lys45 and Arg36, respectively. At the tip of the FXIIFnII β1-β2 hairpin, the Arg36 side chain forms further hydrogen bonds to the main-chain carbonyl of Thr228 and a cation-π interaction with the side chain of Tyr236. Central hydrophobic contacts are made by the side chain of FXII Tyr39 with gC1qR Ala199 and flanking this FXII Gln37 hydrogen bonds to the side chain of gC1qR Ser201 (Figure 2B).

The FXII Arg65 side chain utilizes the guanidinium group to form interactions with 3 gC1qR residues: a salt bridge to Asp249, a hydrogen bond to the Gly247 main-chain carbonyl, and hydrogen bonds to the Ser106 main and side chain (labeled G2 pocket in Figure 2C). Additional gC1qR G2 pocket interactions occur, with FXII residues Asp63 and Gln64 forming a salt bridge and hydrogen bond to the gC1qR Arg122 side chain and Ser106 main-chain nitrogen, respectively, burying a total surface area of 238 A². Between the G1 and G2 pockets, a hydrophobic contact is formed between the FXIIFnII Gln62 and the gC1qR Trp233 side chain (Figure 2D).

The FXIIFnII-gC1qR structure is consistent with previous enzyme-linked immunosorbent assay data from Gebrehiwet et al showing the single point mutation of gC1qR Trp233Gly and G1-loop deletion variants disrupted FXII binding.²⁷  The involvement of the gC1qR G2 pocket residues in the interaction with FXII is novel, and we developed an SPR assay to quantitate the gC1qR-FXII interaction with a series of gC1qR variants. Plasma-purified full-length FXII was amine coupled to a CM5 sensor chip (GE Healthcare), and a reference cell was prepared by blank amine coupling. A recombinant gC1qR wild-type (WT-gC1qR) dilution series was conducted, and analysis of sensorgrams resulted in an equilibrium dissociation constant (K_D) of 120 ± 12 nM in the presence of 50 μM ZnCl₂ (Figure 3a). Consistent with the gel filtration data, no complex was observed in the absence of Zn²⁺ or with EDTA in excess (5 mM). To test the contribution of the gC1qR G1 and G2 pockets to binding FXII, we prepared 2 gC1qR variants replacing key residues with alanine: (1) T228A, D229A, W233A, and Y236A (variant gC1qR-G1-4Ala) and (2) S106A and D249A (gC1qR-G2-2Ala). The structural integrity of the recombinant gC1qR variants was confirmed by gel filtration, revealing native-like characteristics of the trimer. Both the gC1qR G1 and G2 pocket variants failed to elicit a significant signal response or a clear association binding curve under the same conditions of the SPR binding assay to FXII, illustrating the contribution of residues from the G1 and G2 pocket to FXII binding (Figure 3B-C).

Kumar et al showed gC1qR displays specific binding to Zn²⁺ ions,²⁸  and divalent cations such as Ca²⁺ and Mg²⁺ do not bind or support formation of the complex with FXII. Utilizing the structure factors collected for the FXIIFnII-gC1qR complex, we observed the 3 highest peaks in an anomalous difference Fourier map were consistent with a Zn²⁺ coordinated by residues Asp185 and His187. The measured bond lengths are consistent with Zn²⁺ tetrahedral coordination geometry with 2 water molecules also bound. To test the contribution of the gC1qR Zn²⁺-binding site to FXII binding, we prepared a gC1qR H187A variant. SPR experiments were used to assess binding to immobilized full-length FXII and revealed attenuated binding of gC1qR H187A compared with wt-gC1qR, which could not be fitted to derive a K_D (Figure 3D). The Zn²⁺ is located close to the base of the acidic G1-loop adjacent to the FXIIFnII-binding site (Figure 3E). In the previously published gC1qR crystal structure (PDB accession number 1P32¹⁶ ), no metal ion is observed bound at the Asp185-His187 site. We also determined a 1.7-Å-resolution gC1qR structure in the presence of Ca²⁺ ions, and as expected, no metal ion was bound between residues His187 and Asp185 (unpublished results).

A comparison of the ligand-free gC1qR structures with the FXII-gC1qR complex reveals that the Zn²⁺ is replaced by the Arg207 side chain, which forms a salt bridge with Asp185 (Figure 3F). Arg207 does not directly contact FXII, but in the FXII-gC1qR complex, it forms a salt bridge to Glu190, which could indirectly influence FXII binding via the G1-loop. The Trp233 side chain is observed to have multiple conformations in the unbound crystal structures and resembles a flexible tip of the thumb-like helix αB, which can open and close to allow ligand access to the G1 pocket (supplemental Video 3). The other major difference is that the anionic G1-loop becomes ordered upon FXIIFnII binding, and this loop is not resolved in the unbound gC1qR crystal structures and is assumed to be flexible. These 2 sets of conformational changes in the region of the G1 pocket suggest an indirect/allosteric basis for the Zn²⁺ modulation of FXII binding.

HK binding to gC1qR has been quantified by a number of different techniques,^7,29-32  and SPR data in the study published by Pixley et al revealed that binding to full-length HK was in the range of 0.7 to 0.8 nM, which could be abolished in the presence of chelating agent EDTA.³³  The HKD5 domain and constituent peptides have been characterized as the key cell and Zn²⁺-binding^34-36  sites of HK. To build on the previous data, conduct fine mapping of HKD5-binding regions, and determine the stoichiometry of the interaction with the gC1qR trimer, we expressed a series of HKD5 constructs (Figure 4A). Using gel filtration, we were able to detect the coelution of the HKD5-gC1qR complex when applied mixed in a 2:1 ratio, whereas a 3:1 ratio revealed excess unbound HKD5 eluting separately (Figure 4B). To explore this further, we used ITC with gC1qR in the sample cell titrated with HKD5. These data confirmed that HKD5 binds to gC1qR in the absence of Zn²⁺, and binding was unaffected by an excess of EDTA in the ligand buffer (Figure 4C). Examination of the binding isotherm revealed a multistep binding curve that could be modeled as 3 sequential binding steps with very different affinities (K_D values: ∼1.9 nM, ∼64.9 nM, and ∼1.01 μM). The symmetrical gC1qR trimer is therefore unable to accommodate HKD5 at the 3 equivalent sites, demonstrating allosteric effects between sites resulting in asymmetric binding. The first event is of high affinity (K_D of ∼1.9 nM) and is associated with a strongly exothermic interaction and a compensating negative entropy change, typical of the binding and immobilization of a flexible ligand (Table 2). The second and third binding events are 35- and 500-fold weaker, with much diminished enthalpy and entropy changes, consistent with partial steric occlusion at these 2 sites once the first high-affinity site is occupied.

To determine the origin of the allosteric component of the interaction, we expressed recombinant N- and C-terminal fragments of HKD5 in which D5-1 (residues 401-473) contains the His-Gly-rich region and D5-2 (residues 474-531) is His-Gly-Lys rich (Figure 4a). HKD5 has a His-rich nature (21% of the HKD5 sequence) and has previously been shown to bind Zn²⁺ ions.^34,35  Using electrospray ionization mass spectrometry with the whole HKD5 we detected polypeptide species with 1, 2, and 3 bound Zn²⁺ ions and 2 Zn²⁺ bound in each case to D5-1 and D5-2 constructs, respectively (supplemental Figures 3-5).

We examined the binding of the D5-1 and D5-2 fragments to gC1qR by gel filtration and showed that both components could bind independently. The coelution of the N-terminal D5-1 with gC1qR had a Zn²⁺ ion dependency, whereas D5-2 did not. Coelution of D5-1 with gC1qR could be eliminated in the presence of EDTA, but D5-2 was unaffected by EDTA (Figure 4D). This was confirmed by ITC measurements, in which titrations in buffer containing EDTA eliminated the interaction with D5-1 but had no effect on D5-2 (Figure 4E-F).

The binding isotherms of the 2 fragments also showed highly distinct ITC profiles with HKD5-1 retaining the complex 3-phase binding curve observed for full-length HKD5, whereas the C-terminal D5-2 fragment could be modeled as a single-phase binding event, with an estimated stoichiometry of between 2 and 3 (N = 2.3) gC1qR sites each with a K_D of ∼763.4 nM. The 3-phase interaction of D5-1 with gC1qR showed a >40-fold reduction in affinity for the first high-affinity step compared with full-length HKD5 but only a 25 to threefold reduction for the lower affinity steps 2 and 3. These data show that the allosteric effects are largely associated with the His-Gly-rich D5-1 motif and occur in a Zn²⁺-dependent manner (Table 2). The observation that the D5-2 fragment binds in a Zn²⁺-independent fashion suggests that this basic His-Gly-Lys-rich D5-2 motif may be interacting at a different location on the gC1qR structure to D5-1. In the presence of Zn²⁺, we were able to detect a tricomplex among gC1qR, D5-1, and D5-2 by gel filtration, with all components comigrating in the same fractions (Figure 4D).

To further delineate the site of interaction on D5, we considered still shorter peptide motifs derived from D5-1 (HK residues 401-438, HK 439-455, and HK 457-475) and D5-2 (HK 493-516) and studied the interaction by ITC. None of the peptides derived from D5-1 revealed any binding in the presence or absence of Zn²⁺ ions to gC1qR, suggesting that the binding site covers a much larger proportion of the D5-1 fragment (Figure 4G). However, the His-Gly-Lys-rich peptide HK 493-516 derived from D5-2 produced a Zn²⁺ independent single-site binding isotherm comparable to that for D5-2, differing by only twofold in binding affinity, which similarly indicated ∼3 equivalent sites on the gC1qR trimer. The HK 493-516 sequence resembles other peptides rich in Gly-Lys that have been characterized as binding to gC1qR^37,38  without Zn²⁺, and an alignment of these sequences is shown in supplemental Figure 7C.

Ghebrehiwet et al used deletion mutants to identify gC1qR residues 144 to 148, 196 to 202, and 204 to 218 as being important for whole HK binding.²⁷  To determine the location of the recombinant HKD5- and D5-1– and D5-2 fragment–binding sites, we prepared 4 similar gC1qR variants, removing negatively charged residues from the anionic loop regions: gC1qR-G2-5Ala (residues 146-148 and 156-157 mutated to Ala), gC1qRdelG3 (G3-loop residues 214-224 deleted), gC1qRdelG1 (G1-loop residues 196-200 deleted), and a G1-loop variant (residues 196-200), with 5 acidic residues substituted for Ala termed gC1qR-G1-5Ala (supplemental Figure 6). ITC experiments revealed only the gC1qRdelG3 variant had a significant effect in abrogating binding of HKD5 with other variants reproducing the multiphase binding isotherms evident for WT-gC1qR (Figure 5A-C). We repeated the titration with the fragments D5-1 and D5-2 and observed no detectable binding to D5-1 and significant attenuation of binding to D5-2 (Figure 5B-C). This indicates that the central acidic G3-loop that defines an inner pocket (G3 pocket) plays a significant role in the interaction with HKD5, particularly for the binding of the N-terminal D5-1 region (Figure 5D; supplemental Video 4). The placement of the G3-loop at the center of the gC1qR trimer is consistent with the allosteric nature of HKD5 and D5-1 binding, as steric occlusion would reduce the ability of subsequent HKD5 ligands to cobind (Figure 5E). Overall the anatomy of the gC1qR monomer is such that it resembles a hand with the FXII-binding site formed between the index finger (G1 loop) and the thumb (αB), and the palm of the hand contains the Zn²⁺-binding site, adjacent to which is the little finger (G3-loop), defining the principal HK-binding site (Figure 5D; supplemental Video 4).

As both HKD5 and the FXIIFnII domain exhibit asymmetric binding to the gC1qR trimer but by different mechanisms, we next tested whether the FXIIFnII and HKD5 could bind simultaneously to gC1qR. If an excess of HKD5 is present, then FXIIFnII cannot compete for binding, and this is consistent with multiple interaction sites for HKD5 identified by ITC. However, lower stoichiometric ratios of HKD5 reveal both HKD5 and FXIIFnII coeluting in fractions from the gC1qR complex peak shown in Figure 6A. We next extended these studies to a series of gel filtration experiments using the full-length plasma-purified FXII and HK in isolation and in combination with gC1qR. The molecular weights of FXII, HK, and gC1qR estimated by reducing sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) migration are 80 kDa, 110 kDa, and 33 kDa, respectively, and equivalent gel filtration estimates are 80 kDa (FXII monomer), 200 kDa (HK dimer³⁹ ), and 90 kDa (gC1qR trimer). A mixture of full-length FXII, HK, and gC1qR in the presence of Zn²⁺ resulted in comigration of all 3 proteins in a single peak corresponding to a ∼500-kDa complex (Figure 6B). The stoichiometry of the species in this peak is equivalent to a 1:2:6 ratio of FXII-HK-gC1qR shown schematically in Figure 6C. The concept of a ternary complex formed by FXII-HK-gC1qR is consistent with previous in vitro experiments by Joseph et al showing that efficient gC1qR stimulation of PK enzymatic activity requires the presence of PK, FXII, cofactor HK, and Zn²⁺ ions.⁴⁰ 

We next tested the effects of gC1qR on plasma coagulation in the absence of additional stimuli and observed a dose-dependent shortening of clotting time with PNP in the presence of Zn²⁺ (Figure 7A). Thrombin generation experiments revealed a significant shortening in the lag time on addition of gC1qR and Zn²⁺ to PNP, which influenced several additional parameters, including peak thrombin, endogenous thrombin potential, and velocity of thrombin generation (Figure 7B-F). The lag time of thrombin generation in the presence of gC1qR and Zn²⁺ was delayed by a factor of 2.8 in FXII-deficient plasma and absent in FXI-deficient plasma, indicating a dependence on the intrinsic pathway (Figure 7C). Peerschke et al also showed that gC1qR did stimulate plasma coagulation, but not in a FXII-dependent manner,⁴¹  and the reason for this discrepancy may relate to the addition of Zn²⁺. Analysis of 2 gC1qR variants revealed that the gC1qRdelG3 variant (but not gC1qRdelG1) was unable to stimulate thrombin generation to the same degree as wt-gC1qR, implicating the central G3-loop as being functionally important for stimulation of coagulation. It is unknown whether the concentration of endothelial cell bound gC1qR is high enough to support thrombin generation on the vessel wall.

How the contact factors assemble on the cell surface and become activated remains one of the fundamental and unanswered questions underpinning several pathways driving inflammation and thrombosis. The FXIIFnII-gC1qR-Zn²⁺ complex provides the first structural insight into a gC1qR ligand interaction and shows a stoichiometry of 1 FXIIFnII bound to the gC1qR trimer. The asymmetry of FXII binding to gC1qR involves 2 negatively charged surface pockets, whereas for HKD5, this is driven by the central location of a critical negatively charged loop and steric occlusion. Overall, the gC1qR ligand-binding mode we observe for FXII and HK is consistent with the molecular mechanism observed for the chaperone heat shock protein 90 (HSP90), whereby multiple client proteins can be colocalized by asymmetric binding onto the HSP90 dimer.⁴²  The parallel with HSP90 is pertinent, as extracellular HSP90 has been described as a chaperokine⁴³  involved in inflammatory processes and was shown to initiate PK-HK activation.⁴⁴  gC1qR has a requirement for FXII, PK, and HK, whereas HSP90 requires PK and HK, but not FXII.⁴⁴ 

FXII and HK binding to gC1qR is Zn²⁺ dependent, and our data show Zn²⁺ is bound to gC1qR residues Asp185 and His187 in between the FXII and HK-binding pockets (supplemental Video 4). These observations build on data from Kumar et al that describe a Zn²⁺-dependent conformational change in gC1qR²⁸  and thus provide an allosteric mechanism for modulation of the interaction with FXII. The biological context of Zn²⁺ modulation of the FXII-gC1qR interaction is proposed to originate from activated endothelial cells or secreted platelet granules.^45,46  How gC1qR is anchored to the cell membrane has been reported to occur via interaction with other cell receptors¹  or cytokeratin-1.⁴  uPAR has also been shown to act as a receptor for FXII, and this interaction has been studied for endothelial,⁴⁶  neutrophil,⁴⁷  and dendritic cell⁴⁸  function. It is unknown whether gC1qR, uPAR, and cytokeratin-mediated binding of contact factors are functionally aligned.

Gel filtration using the full-length proteins identified a FXII-HK-gC1qR complex with an overall molecular weight in the 500 kDa range, which may represent FXII bound to 2 gC1qR trimers and a HK dimer (Figure 6C). The ability of gC1qR to cluster FXII and HK into a planar ternary complex is conceptually familiar to the way vitamin K–dependent hemostatic proteases from the extrinsic pathway have their protease domains aligned with the substrate activation loop by a Ca²⁺-dependent process on a planar phospholipid surface.⁴⁹  Targeting a chaperone to disable the function of client proteins involved in a pathogenic mechanism is established for chaperones protein disulfide isomerase⁵⁰  and HSP90,⁴²  and our data provide a scaffold for a similar approach to target gC1qR.¹⁴ 

The authors acknowledge the Diamond Light Source for provision of synchrotron radiation in using the Beamline I03.

This work was funded by the British Heart Foundation (grant RG/12/9/29775) (J.E.). Part of this work was funded by the Centre for Membrane Proteins (COMPARE) (J.E.) and an Engineering and Physical Sciences Research Council (EPSRC) studentship (A.S.).

Contribution: B.G.K. performed the experiments with FXIIFnII and gC1qR; A.S. performed the experiments with HKD5 and gC1qR; B.G.K., A.S., I.D., and K.R.M. designed experiments, analyzed the data, and critically reviewed the manuscript; U.S. and N.J.M. performed plasma-based assays; and M.S. and J.E. contributed to the design of experiments, analysis of the data, preparation of the figures and to the writing of the manuscript.

Conflict-of-interest disclosure: J.E. receives consultant’s fees from several pharmaceutical companies. The remaining authors declare no competing financial interests.

Correspondence: Mark Searle, School of Chemistry, Biodiscovery Institute, University of Nottingham, University Park, Nottingham NG7 2RD, United Kingdom; e-mail: mark.searle@nottingham.ac.uk; and Jonas Emsley, Room C52, School of Pharmacy, Biodiscovery Institute, University of Nottingham, University Park, Nottingham NG7 2RD, United Kingdom; e-mail: jonas.emsley@nottingham.ac.uk.