2.1. FXII protein expression and purification

DNA constructs spanning the FnII–EGF1–FnI–EGF2–kringle domains (termed FXII^HC5) and FnII domain (termed FXII^FNII) were generated by PCR amplification from the FXII gene and ligated into the pMT-PURO vector using BglII and MluI restriction (Supplementary Fig. S1a). The FXII^HC5 and FnII constructs were expressed using Drosophila Schneider 2 (S2) cells. For each construct, transfections were performed at a cell density of 1 × 10⁶ cells ml⁻¹ of S2 cells in 5 ml serum-free Schneider medium (Invitrogen) supplemented with 10% heat-inactivated fetal bovine serum (FBS) at 28°C using calcium phosphate and grown for 24 h prior to selection with 10 µg ml⁻¹ puromycin (Sigma) to establish stable cell lines. After the selection of stable transformants, for the expression of recombinant proteins the S2 culture medium was gradually replaced with Express Five medium without FBS. After the secretory expression of recombinant proteins, the media were collected and centrifuged to remove cell debris and filtered using a 0.22 µm filter. For the purification of FnII, the resultant sample was diluted with an equal volume of 0.025 M MES pH 6.0, applied onto a Capto-S column (Cytiva) and eluted using a gradient of 0–1.0 M NaCl. Subsequently, the sample was applied onto a HisTrap column (Cytiva) and eluted using a gradient of 0–0.5 M imidazole, followed by a final purification step utilizing gel filtration on a HiLoad Superdex 75 16/60 preparatory-grade column equilibrated with 0.05 M Tris–HCl pH 8.0, 0.1 M NaCl. FXII^HC5 was captured from the medium using an ion-exchange Capto-S column (Cytiva) equilibrated with 0.02 M Na HEPES pH 7.0; a gradient of 0–1.0 M NaCl was used for elution. The fractions were pooled, applied onto a HisTrap column and eluted using a 0–0.5 M imidazole gradient. Finally, FXII^HC5 was purified on a HiLoad Superdex 200 16/60 column in 0.05 M Tris–HCl pH 8.0, 0.2 M NaCl. FXII^HC5 yields were in the range of 5–10 mg pure protein per litre of insect-cell medium. Lastly, a FXII plasmid construct for recombinant FXII^FnII–EGF1 (residues 1–112) was also prepared, expressed and purified in insect-cell medium using the method described above (unpublished data).

2.2. Crystallization, data collection and structure solution of FXII^HC5

Crystallization of FXII^HC5 in 20 mM Na HEPES pH 7.0 was carried out in sitting-drop MRC plates using a Mosquito (TTP LabTech) and commercial protein crystallization screens from both Molecular Dimensions and Hampton Research. FXII^HC5 was concentrated to 8 mg ml⁻¹. Small crystals were observed using condition A8 of the MIDASplus screen (Molecular Dimensions) containing 5% polyacrylic acid sodium salt (PAA) 2100. The crystallization condition was optimized by using PAA polymers with varying molecular weights of 1200, 2100, 5100, 8000 and 15 000 using a grid of concentrations from 3.8% to 6%. In addition, additive screens from both Molecular Dimensions and Hampton Research were utilized to improve the crystal size. The final data set was collected from a single large crystal (0.3 × 0.3 × 0.1 µm) grown in a sitting drop with 2 µl protein solution added to 2 µl of a reservoir solution consisting of 5% PAA1200 with the addition of 0.2 µl Hampton Research Additive Screen Reagent D8 containing 0.1 M urea.

Data collection was performed on beamline I03 at Diamond Light Source to a resolution of 3.4 Å and data were processed using the beamline implementation of autoPROC and STARANISO for reduction in space group I2₁2₁2₁. Molecular replacement with Phaser (McCoy et al., 2007) used the crystal structures of FXII FnII and FnI–EGF2 (PDB entry 4bdw; Beringer & Kroon-Batenburg, 2013) and a homology model of the kringle domain prepared using AlphaFold (Jumper et al., 2021) as templates. Additional density was observed for the FnII domain in only two copies of the FXII^HC5 structure. Manual model building was performed in Coot (Emsley et al., 2010) and structure refinement was performed using Phenix (Liebschner et al., 2019). Evaluation of the quality of the final model was carried out using MolProbity (Table 1; Chen et al., 2010). The final FXII^HC5 electron density is continuous for the main chain, spanning residues 18–278, in two molecules, while the FnII domain is absent in a third molecule (residues 77–278; Supplementary Fig. S4).

Table 1 Crystallographic data-collection and structure-refinement statistics Values in parentheses are for the highest resolution shell.   FXIIHC5 FXIIFnII–Zn2+ Data collection      Space group I212121 P212121  Temperature (K) 93 100  a, b, c (Å) 144.5, 144.6, 155.9 26.3, 40.9, 45.3  α, β, γ (°) 90, 90, 90 90, 90, 90  Resolution (Å) 77.97–3.4 22.78–1.2  Rmerge† 0.179 (1.03) 0.121 (0.395)  〈I/σ(I)〉 7.1 (2.1) 7.5 (2.2)  Completeness (%) 94.2 (66.2) 98.1 (97.0)  Multiplicity 6.4 (6.4) 4.5 (4.6)  CC1/2 0.994 (0.585) 0.993 (0.822)  Total/unique reflections 13995 (701) 15507 (778) Structure refinement  Rwork‡ 0.260 (0.36) 0.174 (0.224)  Rfree 0.320 (0.49) 0.192 (0.240)  R.m.s. deviations   Bond lengths (Å) 0.013 0.005   Bond angles (°) 2.05 0.768  B factor (Å2) 118.9 9.6  Ramachandran plot   Favoured (%) 91.6 98.11   Allowed (%) 8.0 1.89   Outliers (%) 0.4 0  PDB code 8os5 7prj †Rmerge = , where I(hkl) is the observed intensity and 〈Ihkl〉 is the average intensity of multiple observations calculated from symmetry-related reflections. ‡Rwork = , where Fobs and Fcalc are the observed and calculated structure factors, respectively. Rfree is computed as for Rwork, but for only a randomly selected 5% of the reflections, which were omitted during refinement; it was calculated using Phenix.

2.3. Crystallization, data collection and structure solution of FXII^FnII

Sparse-matrix screening with recombinant FXII^FnII–EGF1 and FXII^FnII proteins only resulted in crystals of the FnII domain. Equal volumes of 6 mg ml⁻¹ FXII^FnII and reservoir solution consisting of 0.01 M zinc chloride, 0.1 M sodium acetate pH 5.0, 20%(w/v) PEG 6000 were mixed in a sitting-drop plate and incubated at 19°C. Single crystals grew overnight and were soaked in cryoprotectant comprised of the reservoir supplemented with 30%(v/v) glycerol and subsequently flash-cooled in liquid nitrogen. Data collection for FXII^FnII was carried out on beamline ID23-2 at the ESRF synchrotron and a complete data set was collected to 1.2 Å resolution from a single crystal. Data were indexed and integrated using MOSFLM (Battye et al., 2011). The initial phases were determined with Phaser using a previous crystal structure of FXII^FnII bound to gC1qR (PDB entry 6szw; Kaira et al., 2020). Model building was carried out using Coot, and REFMAC5 and Phenix (Murshudov et al., 2011; Liebschner et al., 2019) were used for refinement. The N-terminal 20 amino acids of FXII are not observed in the electron density and are presumed to be flexible. The refined models were validated by MolProbity (Chen et al., 2010), and PyMOL (https://www.pymol.org) was used to produce figures. The FXII^HC5 and FXII^FnII–Zn²⁺ complex crystal structures have been deposited in the PDB (https://www.rcsb.org) with accession codes 8os5 and 7prj, respectively.

2.4. In-solution deglycosylation of FXII^HC5 under native conditions

PNGase F was used for the digestion of recombinant human FXII^HC5 according to the protocol for nondenaturing reaction conditions provided by the supplier (New England Biolabs, Catalogue No. P0708S). The reaction was set up in the absence of detergent or denaturants, using a tenfold higher enzyme:protein ratio than recommended for denaturing conditions. The reaction was incubated for 6 h at 37°C and subsequently analysed by SDS–PAGE.

2.5. Mass spectrometry (MS) of FXII^HC5

MS studies were performed at the University of York MS facility. For denaturing MS, FXII^HC5 protein was diluted 1:20 into aqueous 50% acetonitrile containing 1% formic acid. For native-mode mass spectrometry, the protein was buffer-exchanged using Amicon MW spin filters into 1 M ammonium acetate pH 7.0. The protein solution was infused at 3 ml min⁻¹ into a Bruker maXis qTOF mass spectrometer via an electrospray ionization source. Source conditions and ion-optic parameters were adjusted to favour the detection of native protein states, specifically: dry gas, 250°C at 6 l s⁻¹; funnel RF, 400 Vpp; multipole RF, 200 Vpp; quadrupole low cutoff, 1200 m/z; quadrupole ion energy offset, 3 eV; prepulse storage, 50 ms; transfer time, 160 ms. Spectra were summed over 1 min acquisitions at 0.1 Hz. The data were smoothed (0.2 Da, one cycle, Gauss) by maximum-entropy deconvolution to average masses at 200 Da resolution. Separate deconvolutions were performed for the charge-state regions resulting from the protein monomer and dimer and in total cover the mass range 20–80 kDa. Data acquisition was performed using Bruker Hystar and oTof control (version 4.1). Peak picking and spectral processing were undertaken using the Bruker Data Analysis software with MaxEnt deconvolution (version 4.4).

2.6. Small-angle X-ray scattering (SAXS) measurements and analysis

SAXS experiments were carried out on beamline B21 at Diamond Light Source, Didcot, UK using a monochromatic X-ray beam (λ = 0.9408 Å) at an electron energy of 13.2 keV. The beamline was equipped with an EIGER X 4M detector (Dectris, Switzerland) and the sample-to-detector distance was 3.712 m, covering a momentum-transfer range of 0.0045 < q < 0.34 Å⁻¹, where q = (4πsinθ)/λ and 2θ is the scattering angle. In-line gel-filtration-coupled SAXS for human FXII^HC5 at ∼1.5 mg ml⁻¹ was carried out by loading 45 µl protein sample onto a 2.4 ml Superose 6 column (GE Healthcare) equilibrated with a buffer solution consisting of 20 mM Tris–HCl pH 8.0, 150 mM NaCl and coupled to an Agilent 1200 HPLC system with a flow rate of 0.15 ml min⁻¹ at a controlled temperature of 15°C. 600 SAXS data frames were collected with exposure intervals of 3 s and the buffer frames used for the background-subtracted SAXS were collected after 1.5 column volumes. Raw SAXS 2D images were processed with the DAWN package, the processing pipeline available at the beamline, to produce normalized, integrated 1D unsubtracted SAXS curves. Initially, the scattering image frames were spherically averaged, scaled and merged using the in-house software ScÅtter version 4.0. The radius of gyration (R_g) was estimated through the Guinier approximation, I(q) = , qR_g < 1.3, and by using ScÅtter; the same was performed for the pair distribution of the particle, p(r), and the maximum dimension D_max, which also were computed using ScÅtter. The BayesApp (Hansen, 2012) program was used to further fine-tune the pair distribution p(r), together with Guinier analysis, the Porod plot and the Kratky plot generated by Bayesian indirect Fourier transformation (BIFT; Vestergaard & Hansen, 2006). The SAXSMoW server was utilized to calculate the molecular weight (de Oliveira Neto et al., 2022; Piiadov et al., 2019). The rescaled data with potential outliers removed after error assessment (Larsen & Pedersen, 2021) were input to DENSS (version 1.6.3) to reconstruct 20 ab initio 3D electron-density maps from the 1D solution scattering profile and to perform alignment and averaging of the reconstructions (Grant, 2023). The 3D MRC-formatted density maps were visualized using the UCSF Chimera graphical software (Meng et al., 2023) and to fit the FXII^HC5 crystal structure. A summary of the SAXS data collection and processing is given in Table 2. Based on the FXII^HC5 structure, the X-ray scattering profiles were computed and the discrepancies between the experimental and theoretical SAXS curves were quantified by the minimized χ parameter using the FoXS/MultiFoXS web server (Schneidman-Duhovny et al., 2016).

2.7. Gel-filtration analysis of FXII in the presence of polyanions

The complex of FXII in the presence of heparin and polyP was analysed by analytical size-exclusion chromatography (gel filtration) using an ÄKTAmicro FPLC system (Cytiva, Little Chalfont, UK). Narrowly size-fractionated polyP preparations were produced from chemically synthesized polyP using preparative PAGE performed on a Model 491 Prep Cell (Bio-Rad) using a 100 ml gel and a 30 mg polyP load. PolyP concentrations were quantified by measuring inorganic phosphate after complete hydrolysis, as described by Smith et al. (2018). Preparations collected from fractions were sized as described previously (Smith et al., 2018) and are referred to by their polymer lengths, which were 34, 55 and 69 phosphate units. PolyP was added to FXII at different molar ratios from 1:1 to 1:100 with plasma-purified FXII (Enzyme Research Laboratories). Gel filtration was performed on the resulting mixture using a 3 ml Superose S200 column (Cytiva) equilibrated with 0.05 M HEPES–HCl pH 7.0, 0.2 M NaCl at a flow rate of 0.3 ml min⁻¹. Heparin fragment Hep20 was purchased from Iduron (termed dp20) and was added to FXII at different molar ratios from 1:1 to 1:10 and gel filtration was performed as described for polyP. For FXII^HC5 gel-filtration studies with heparin, an ÄKTApure FPLC system (Cytiva, Little Chalfont, UK) was used. Heparin was added to FXII at different molar ratios from 1:5 to 1:10 and gel filtration was performed as described for polyP. The column was calibrated with standard proteins (Cytiva, Little Chalfont, UK) of known Stokes radii: thyroglobulin (669 kDa), ferritin (440 kDa), aldolase (158 kDa), conalbumin (75 kDa), carbonic anhydrase (29 kDa) and aprotinin (6.5 kDa). Blue dextran MW 2000 kDa was used to determine the column void volume (V_o). Elution volume positions were monitored at 280 nm. The partition coefficient (K_av) of FXII and protein standards was calculated using the equation K_av = (V_e − V_o)/(V_t − V_o), where V_e is the elution volume and V_t is the total column volume.

3.1. Crystal structure of FXII^HC5

Constructs spanning different combinations of the FXII N-terminal domains were expressed using insect cells and purified from media as described in Section 2 (Supplementary Fig. S1). FXII^HC5 yields were in the range of 5–10 mg pure protein per litre of insect-cell medium (Supplementary Fig. S1) and the amino-acid sequence was confirmed using mass spectrometry (MS; Supplementary Fig. S2). Data were collected to 3.4 Å resolution from a single crystal (Table 1) and the FXII^HC5 structure was solved by molecular replacement using the available crystal structures of the FXII FnII (Kaira et al., 2020) and FnI–EGF2 domains (Beringer & Kroon-Batenburg, 2013). The FXII^HC5 structure as viewed in Fig. 1(c) has a circular ring or torc shape, with the kringle domain forming a head-to-tail contact via a latch-like loop from the FnII domain (Fig. 1d). The FnII domain is followed by a nine-residue linker loop spanning residues 70–78, which terminates in a sharp 90° angle leading into the tandem arrangement of the EGF1–FnI domains. The structure was analysed for stable interfaces with PDBePISA (Krissinel & Henrick, 2007), which identified that the FXII^HC5 monomers interlock in an antiparallel arrangement, which buries a surface area of 2067 Ų. This FXII^HC5 dimer structure positions the FnII domains radially 75 Å apart, whereas the two kringle domains are located axially 20 Å apart (Figs. 1e and 1f).

In the asymmetric unit there are three copies of FXII^HC5. Two FXII^HC5 monomers are observed with the FnII domain and linker region present. In one FXII^HC5 monomer the FnII domain is absent and the EGF1–FnI–EGF2–kringle domains exhibit a slightly different conformation and a reduced buried surface area of 1408 Ų. The linker-loop residues can be thought of as a flexible strap enabling the FnII domain to wrap around the FnI domain and interlock in a tight-knit arrangement contacting the kringle domain (Supplementary Movie S1). The antiparallel EGF1–FnI domain interface is formed from residues 103–107 and 123–129, with electrostatic interactions in the centre flanked by hydrophobic contacts on either side (Fig. 2a). Hydrogen bonds are formed between the Arg123 side chain and the main-chain carbonyl of Cys102, and the Glu129 carboxyl forms a hydrogen bond to the hydroxyl side chain of Thr107. Additional electrostatic interactions occur between Glu129 and His126 and Lys113.

Figure 2 FXIIHC5 intramolecular interactions and charge distribution. (a) Dimer interfacial interactions between the EGF1 and FnI domains are shown with key residues as sticks and electrostatic bonding as purple dotted lines. (b) A charged molecular-surface representation (blue, positive; red, negative) with a continuous flat surface of positive charge generated centrally by the EGF1–FnI domain dimer. Two radial surfaces of further positive charge are generated by the linker and FnII domains. (c) Cartoon diagram showing the dimer with the two L-shaped FnII–EGF1–FnI polypeptides. Clusters of surface-exposed arginine or lysine residues are shown as sticks coloured blue and labelled as anion-binding exosites (ABE). (d) A schematic diagram illustrates the EGF1 and FnI domains with the location of charged residues for ABE2 and ABE3 indicated in the context of the dimer interface (grey).

Biochemical studies have previously identified contributions to polyanion binding from the FnII, EGF1 and FnI domains (Shamanaev et al., 2022). A distinctive feature of our FXII^HC5 dimer structure is the symmetric presentation of three areas of positively charged residues [termed the anion-binding site (ABE); Figs. 2b and 2c]. The FnII domain arranges His35, Arg36, His40, Lys41, His44 and Lys45 into ABE1, which is a triangular-shaped flat surface (Supplementary Movie S2). The linker-region residues Lys73, Lys74 and Lys76 cluster with His78, Lys81 and His82 in the EGF1 domain to generate an area of positive charge, and these residues have been identified as the polyP-binding site (ABE2; Shamanaev et al., 2023). A third cluster of basic residues, Lys113, Lys127, Lys145 and Arg141, centred around the FnI domain results in an extended pocket of symmetric positive charge close to the dimer axis (ABE3; Fig. 2d). The FXII^HC5 dimer that we observe may not occur in full-length FXII due to steric and structural constraints imposed by the presence of the C-terminal PRR and protease domain.

3.2. The FXII^HC5 kringle domain contains a lysine-binding site

The FXII kringle domain contains a lysine-binding site, with a groove formed by Trp257 and Trp268 flanked by acidic residues Asp251 and Asp253 that interacts with Lys81 from EGF1 on an adjacent FXII^HC5 structure (Figs. 3a and 3b). In the FXII^HC5 structure two kringle domains are arranged such that the lysine-binding sites point outwards separated by a distance of 50 Å (Fig. 3a). This is a characteristic orientation, with the extended lysine alkyl chain interacting with the Trp257 and Trp268 side chains and the amine forming electrostatic interactions with the side chains of Asp251 and Asp253. The FXII kringle domain lysine-binding site bears a close resemblance to the tissue-type plasminogen activator (tPA) kringle crystal structure (de Vos et al., 1992) and the tandem EGF2–kringle inter-domain angle of 90° has been observed previously in the crystal structure of the amino-terminal domains of the urokinase plasminogen activator (uPA; Barinka et al., 2006). Compared with the plasminogen kringle domain 1 structure, the FXII kringle domain does not have the required basic residues (plasminogen kringle 1 Lys35 and Arg71) for terminal lysine binding (Mathews et al., 1996).

Figure 3 FXIIHC5 kringle lysine-binding site. (a) Cartoon diagram of FXIIHC5 in the region of the EGF2 (black) and kringle (red) domains showing the outward-facing lysine-binding sites as sticks engaging the Lys81 side chain (green spheres). (b) Enlarged view of EGF1 Lys81 (green) and the kringle domain lysine-binding site residues shown as sticks. (c) The interface between the EGF1 and the kringle domains involves positive charges on Lys76, Lys81 and His82 from the EGF1 domain coordinating to the kringle domain residues. Electrostatic interactions are shown as purple dotted lines. (d) Kringle-mediated interactions result in three FXIIHC5 dimer forms being arranged into a cyclic hexamer. This hexameric form of the FXIIHC5 structure is represented as a ribbon with the subunits coloured purple, grey, green and yellow and the FnII-less FXIIHC5 dimer coloured cyan and salmon. (e) Two views of the hexameric form of FXIIHC5 shown as a molecular surface.

The FXII^HC5 kringle–EGF1 interaction resembles a crystal contact and buries a small surface area whereby EGF1 α-helix residues Lys76 and Lys81 interact with the kringle Asp251 and Asp253 side chains (Fig. 3c). Additional interactions are formed by the packing of the Tyr228 and Lys81 side chains, and the Tyr228 main-chain carbonyl hydrogen-bonds to the side chain of His82. These intermolecular FXII^HC5 inter­actions involving the EGF1–kringle domains assembles three FXII^HC5 dimers into a doughnut-shaped hexamer (Figs. 3d and 3e).

3.3. MS analyses of recombinant FXII^HC5

Analysis of the interfaces in the FXII^HC5 crystal structure using PDBePISA (Krissinel & Henrick, 2007) indicates that the large surface area buried by the structure shown in Fig. 1(f) should be stable as a dimer. We thus utilized MS (Rostom & Robinson, 1999; Karch et al., 2022) to probe whether a FXII^HC5 dimer could be detected in the gas phase. To gain a precise measurement of the mass, we first used denaturing MS, revealing a major peak for FXII^HC5 at ∼34.8543 kDa (Supplementary Fig. S3). In the presence of PNGase F a shift in mass indicated the removal of a single N-glycan (Supplementary Fig. S3). Next, native MS was used to analyse oligomer formation of FXII^HC5 (Supplementary Fig. S3) and showed both charged-state and deconvoluted spectra and detected a monomer peak at 34.9 kDa and a small dimer peak at a molecular weight of 67.1 kDa. We next purified mouse FXII^HC5 for comparison, which has ∼70% sequence identity to human FXII^HC5. Mouse FXII^HC5 also revealed a monomer peak of 35.9 kDa and a small dimer peak that is precisely double this at 71.9 kDa (Fig. 4 and Supplementary Fig. S3). In data acquisition under native-favouring conditions there is a shift in the charge-state distribution towards a lower charge state and range (higher m/z), which is typical of native structures that are more folded and so have less accessibility for protonation. Overall, these analyses revealed that the signal is dominantly native low-charge monomer and relatively less poorly resolved low-charge dimer, as annotated in the spectrum obtained. Although protein multimerization can be measured by MS in the gas phase, this is very much protein-dependent.

Figure 4 Native mass spectrum of human FXIIHC5. (a) Coomassie-stained SDS–PAGE gel for purified recombinant human and mouse FXIIHC5. (b, c) Native mass spectrum of (b) human FXIIHC5 and (c) mouse FXIIHC5 showing multiple charge-state distributions; different species are labelled as monomer and dimer peaks.

3.4. SAXS characterization of FXII^HC5

We next used SAXS to investigate whether FXII^HC5 dimer formation could be detected in solution. A single broad peak was observed during gel filtration (Fig. 5a) and the resulting SAXS data were of high quality, with no evidence of aggregation from the Guinier plot (Fig. 5b). The radius of gyration (R_g) values from both Guinier and Porod analysis are in good agreement at 28.03 and 29.10 Å, respectively, with a maximum particle dimension (D_max) of 105 Å (Figs. 5c and 5d, Table 2). This compares favourably with the calculated FXII^HC5 dimer R_g of 26.95 and longest dimension of 96 Å. The smaller value of the calculated R_g compared with the experimental value may be explained by hydration and the flexible N-terminus (residues 1–17) and C-terminus (residues 278–295), which are not defined in the crystal structure. The molecular-weight estimate calculated via SAXSMoW is 51 kDa (Piiadov et al., 2019; de Oliveira Neto et al., 2022).

Figure 5 SAXS analysis of human FXIIHC5. (a) SEC-SAXS signal plot. Each point represents the integrated area of the ratio of the sample SAXS curve to the estimated background. The frames used as the buffer background are in grey with the average represented as a grey horizontal line. (b) Guinier plot. (c) Distance distribution function fit to the data with rescaled errors. (d) Distance distribution. (e) Kratky plot. The plots in (b)–(e) are taken from BayesApp. Fit of the interlocking dimeric crystal structure (f) into the 32 Å density envelope obtained from DENSS. (g) Comparison of the experimental SAXS curve (circles) with the curve calculated from the crystal monomeric (red) and dimeric (blue) structure models for data covering a momentum-transfer range of 0.0045 < q < 0.15 Å−1. The goodness-of-fit parameter from FoXS (χ2) is 1.78 for the monomer, with Rg = 25.54 Å, c1 = 0.99 and c2 = 3.23, and 4.75 for the dimer, with Rg = 26.95 Å, c1 = 1.05 and c2 = −2.00.

Fig. 5(f) shows two views of the SAXS-derived electron-density map (∼32 Å resolution). The left-hand view showing the profile is triangular in shape and the interlocking FXII^HC5 dimeric model from the crystal structure can be comfortably fitted into the 3D reconstruction (Fig. 5f). The right-hand view in Fig. 5(f) also shows that the dimensions match the model, but the narrowing of the density in the middle may represent an average of both monomer and dimer. To summarize, three-dimensional electron-density reconstruction using DENSS, which does not strictly rely on assumptions of globularity, yielded a volume consistent with an average particle size indicative of both monomeric and dimeric forms. The fitted electron-density map corroborates the dimensions and spatial arrangement consistent with the interlocking dimer in the crystal structure, reinforcing the presence of dimeric forms in solution. The SAXS data were further analysed by calculating theoretical SAXS data from the crystal structure using FoXS/MultiFoXS. Comparison of the experimental SAXS curve (circles) with the curve calculated from the monomeric (red) and dimeric (blue) structure models for data covering a momentum-transfer range of 0.0045 < q < 0.15 Å⁻¹ is shown in Fig. 5(g). The goodness-of-fit parameter from FoXS (χ²) is 1.78 for the monomer and 4.75 for the dimer, respectively (Fig. 5g).

3.5. Gel filtration of FXII with polyanions of different lengths

The full-length FXII zymogen is monomeric (Colman et al., 1997; Kaira et al., 2020), but oligomerization has been reported to occur in solution in the presence of soluble polyanions (Samuel et al., 1992; Wang et al., 2019). We utilized fractionated polyP in a series of gel-filtration experiments to characterize FXII oligomerization at increasing polyP concentrations and chain lengths of 34, 55 and 69 phosphate units. Gel filtration was performed with a Cytiva ÄKTAmicro system and a 3 ml Superose S200 column. The column was calibrated using commercial gel-filtration standards (Cytiva): thyroglobulin (669 kDa), ferritin (440 kDa), aldolase (158 kDa), conalbumin (75 kDa), carbonic anhydrase (44 kDa) and aprotinin (6.5 kDa). The FXII zymogen has an elution volume (V_e) of 2.2 ml and the calculated Stokes radius is 3.2 nm, which is very close to the conalbumin standard (75 kDa), corresponding to an FXII monomer (the calculated molecular weight based on the amino-acid sequence alone is 65 733 Da). Addition of the shortest polymer polyP34 to FXII at increasing concentrations up to a ratio of 1:100 did not result in any difference in V_e compared with FXII alone (Fig. 6a). In contrast, polyP55 and polyP69 resulted in the appearance of a second peak with a V_e of 1.92 ml similar to the gel-filtration standard ferritin (1.92 ml) with a molecular weight of 440 kDa.

Figure 6 Gel filtration of FXII complexes with polyP and heparin fragments. (a, b, c) Gel-filtration elution profiles for full-length plasma-purified FXII in the presence of increasing molar ratios of polyP with different chain lengths of (a) 34, (b) 55 and (c) 69. Elution volume (Ve) measured in millilitres is shown on the x axis and UV absorption (mAU) is shown on the y axis. (d) Gel-filtration elution profiles for FXII in the presence of increasing molar ratios of heparin fragment Hep20. The second peak arising at a Ve of 1.92 ml is equivalent to the standard gel-filtration standard ferritin, which has a molecular weight of 440 kDa.

To extend these FXII oligomerization observations to a second polyanion, we utilized a commercially available (Iduron) fractionated low-molecular-weight heparin composed of ∼20 saccharide units (ten repeats of the main disaccharide unit; MW of ∼5750 Da). A mixture of FXII in the presence of increasing molar ratios of Hep20 resulted in the appearance of second peak at a V_e of 1.94 ml which became a single peak at a 1:10 ratio excess of Hep20. Hep20 requires a 1:10 ratio to drive FXII oligomer formation and results in a complete conversion to the oligomeric FXII. In contrast, polyP requires a 1:100 molar ratio for the FXII oligomer to appear and the FXII monomer is still present in the elution profile. Interestingly, both the polyanions polyP55 and Hep20, which are of similar molecular weight, give a very similar observed increase in the Stokes radius of the FXII–polyanion complex to 6.2 and 6.1 nm, respectively. The standard ferritin with a molecular weight of 440 kDa has a Stokes radius of 6.2 nm. Additionally, a small subsidiary peak is observed at ∼2.5–2.6 ml; however, this peak corresponds to a significantly smaller estimated size than the FXII domain: ∼7.0–4.0 kDa. This peak did not contain any protein fragment when analysed by SDS–PAGE and may represent potential breakdown products, aggregates or buffer constituents etc., and was treated as irrelevant to the experiment. In these assays, the concentrations of FXII and polyanions were 10 µM and 0.01–1 mM, respectively.

3.6. FXII Zn²⁺-binding sites

Biochemical analyses have identified Zn²⁺-binding sites in the FnII and EGF1 domains (Røjkjaer & Schousboe, 1997) which greatly enhance the enzymatic reactions of FXII (Wang et al., 2019; Shore et al., 1987). The precise location of the Zn²⁺-coordinating residues is unknown, but the FXII sequence is histidine-rich (Samuel et al., 1993) and the FXII^HC5 structure shows that the FnII–EGF1–FnI domains present ten histidine resides on one contiguous surface (Fig. 7a). A cluster of residues, His78, His82, His99 and His110 in the EGF1 domain and His40, His29, His35 and His44 in the FnII domain, represent potential Zn²⁺ ion coordination sites. To investigate, we expressed untagged FXII FnII domain (residues 1–71, termed FXII^FnII) and confirmed that the FnII protein did bind to a Zn²⁺ affinity column (unpublished observation). The purified FXII^FnII sample was crystallized in the presence of ZnCl₂ and the structure was determined using molecular replacement (Table 1). The 1.2 Å resolution FXII^FnII structure revealed the expected FnII domain fold of four β-strands arranged as two antiparallel β-sheets, β1–β2 and β3–β4, at right angles, together with a short α-helix (Fig. 7b). Utilizing the 1.2 Å resolution structure factors, we were able to identify two bound Zn²⁺ ions in the anomalous difference Fourier map. Fig. 7(b) illustrates the Zn²⁺ ions as grey spheres; site 1 is coordinated by His17 and His40, with the remaining coordination completed by two water molecules (Fig. 7c); the site 2 Zn²⁺ ion is coordinated by residues His44 and Glu71 with additional coordination from His35 of a symmetry-related molecule. The two Zn²⁺ ions are placed on the same face of the FXII^FnII structure as the basic residues Arg36, Lys41 and Lys45, forming a concentrated triangular surface of positive charge (Fig. 7d).

Figure 7 FXII Zn2+-binding sites. (a) Cartoon diagram of the FXIIHC5 structure showing the dimer with the two L-shaped FnII–EGF1–FnI polypeptides, with clusters of surface-exposed histidine residues as potential Zn2+-binding sites shown as sticks (cyan). The black dashed line labelled N represents the N-terminus. (b) Cartoon diagram of the isolated FXIIFnII crystal structure with bound Zn2+ ions shown as grey spheres. Residue Arg47 is shown in light blue and Pro48 is in green. Other residues are shown as sticks in dark blue. (c) Enlarged view of the His17 and His40 residues shown as sticks, with the coordination sphere of the Zn2+ ion (grey sphere) completed by two water molecules (red spheres). Purple dotted lines represent electrostatic interactions formed with Zn2+. (d) A charged molecular-surface representation (blue, positive; red, negative) showing the FXIIFnII ABEI domain as a triangular surface of positive charge with two Zn2+ ions as grey spheres. Comparison of the FnII domain cation-binding site in the FXIIHC5 crystal structure, where it is occupied by Pro48 (e), compared with the Zn2+-bound isolated FXIIFnII (f), where a loop rearrangement occurs and the the Arg47 guanidinium occupies the cation-binding site.

An interesting feature of the FXII^FnII Zn²⁺ sites is that they both coordinate residues from the FnII domain flanking sequences. His17 is not present in the FXII^HC5 structure and residues 1–17 are not visible in the FXII^HC5 electron density (shown as a dotted line in Fig. 7a), and Glu71 is in the linker region. FnII domains can interact with ligands via a surface pocket formed between residues Trp53 and Trp66 from strands β4–β3 and Phe60 from the α-helix (Morgunova et al., 1999; Wah et al., 2002). This FXII^FnII pocket is occupied by the Arg47 side chain, which extends to form a cation–π interaction with Trp66. A comparison of FXII^FnII with the FXII^HC5 structure reveals significant conformational change in the FnII latch-loop region. The FXII Arg47-Pro48-Gly49-Pro50 sequence rearranges such that Arg47 ratchets out of the FnII pocket in FXII^HC5 to form a head-to-tail salt bridge with the kringle domain residue Asp264. In FXII^HC5 the main-chain carbonyl of Pro48 forms a hydrogen bond to the side-chain hydroxyl of Tyr68 (Figs. 7e and 7f). This 4 Å movement of Arg47 and Pro48 is animated as a molecular morph between the two structures in Supplementary Movie S3.