Structure and activation of C1, the complex initiating the classical pathway of the complement cascade

The C1r₂s₂ Tetramer Is Highly Extended.

To gain structural insight into the solution state of C1r₂s₂, we collected synchrotron SAXS data from extensively purified recombinant C1r₂s₂ (Fig. 2A) with the S/A mutation in both proteases making the tetramer an inactive enzyme. Guinier analysis showed a linear plot for C1r₂s₂ in the concentration range of 1.0–6.8 mg/mL, but with slightly steady increasing values of R_g. We extrapolated the data to 0 mg/mL and created a merged dataset from the extrapolated data and the data recorded at 6.8 mg/mL (Fig. 2 E and F and Fig. S1A). For the subsequent steps, we used merged data with q < 2.5 nm⁻¹ [q = (4 × π × sinθ) ÷ λ, where 2θ is the scattering angle]. The Kratky plot of the data showed that C1r₂s₂ is well folded and has limited flexibility (Fig. S1B). The pair-distance distribution function (P(r)) revealed that D_max for the particle is just below 45 nm (Fig. 2F), which was immediate evidence against a compact conformation of the tetramer. To obtain a SAXS-based solution structure of the C1r₂s₂ tetramer, we used rigid-body modeling starting from tetramer models based on known substructures of C1r and C1s (Fig. S1C). The core of the tetramer models was formed by the CUB1-EGF domain from two C1s molecules related by a twofold rotation axis each surrounded by the CUB1-EGF domains from C1r (Fig. 3 A and B). The best results in terms of fit to the experimental data and internal consistency between the resulting models were obtained by dividing both C1r and C1s into four rigid bodies consisting of CUB1-EGF, CUB2, CCP1-CCP2, and the SP domain. Models from 50 rigid-body refinements were sorted into two groups according to χ² values of their fits to the data. The major group of 43 models fitted the SAXS data with χ² =1.78 ± 0.05, whereas the remaining seven models had χ² of 2.60 ± 0.15. The output models in the major group were rather similar with the CCP1-CCP2-SP domains curving to either side of the plane compared with the quite planar starting model (Fig. 3 B and C). Removal of the glycans from the input model resulted in 10 models with χ² = 3.56 ± 0.45, showing that the glycans contributed significantly with respect to matching the models to the data although the 12 Asn-linked glycans constitute less than 10% of the total molecular weight of the tetramer. Importantly, the models obtained in the absence of glycan were similar to those obtained with glycans included. The model with the lowest χ² = 1.68, which is representative, is displayed in Fig. 3D and it is fit to the data in Fig. 2E. The SP domains from both C1r and C1s are protruding from the central CUB1-EGF tetrameric interface. The catalytic sites of the two C1r SP domains are separated by 39 nm, whereas those of C1s are 28 nm apart from each other. The activation site in one C1r is separated by 37 nm from the active site in the second C1r molecule.

We do not know how the C1r CUB1-EGF domain tandem interacts with their C1s equivalent in the core of the tetramer, and C1r₂s₂ may have a different structure in solution than as a component of the C1 complex. We therefore also investigated two alternative tetramer models in which C1r had been translated 27 Å and 54 Å (Fig. S1 D and E) in the plane defined by the C1s CUB1-EGF dimer compared with the first starting model where the C1r-C1s CUB1-EGF arrangement was based on a C1s CUB1-EGF tetramer formed by crystal packing (3). Rigid-body refinements gave average χ² values of 1.68 ± 0.04 and 1.34 ± 0.16 for 27 and 54 Å C1r translation, respectively, for 10 CORAL runs in each case. Hence, in terms of fitting the experimental data, both C1r-translated models were better than models derived from the nontranslated starting model. However, the resulting ensembles of models were structurally more diverse, perhaps reflecting that the C1r hinge points were located further from the center of the molecule and often resulted in twisted tetramers (Fig. S1 D and E). The location of the collagen binding sites in the CUB domains also appeared to be poorly compatible with C1q binding. Nevertheless, the wide separation of the SP domains was similar to the output models obtained from the nontranslated starting model. Overall, the solution studies of C1r₂s₂ suggest that the SP domains of two C1r molecules are widely separated in the tetramer.

To corroborate our SAXS analysis, we performed single-particle electron microscopy on C1r₂s₂. We recorded negative stain images of the C1r₂s₂ complex at low voltage conditions (60 kV) to achieve optimum contrast. Furthermore, we selected specimen regions with a typical single carbon layer appearance to avoid flattening effects from additional carbon films covering the particles that would influence the apparent size. We computed class averages of the particle dataset with on average 34 images per class by using unbiased clustering techniques independent of any SAXS-based model (Fig. 3E, Upper). Clustering enhances the signal-to-noise ratio and serves as a statistical tool that brings out recurrent particle views. All class averages show particles with an elongated shape and a mean maximum dimension of approximately 43 nm in agreement with a D_max of 45 nm observed by SAXS. The particles show a central broadening and small peripheral domains protruding from the complex.

To compare the EM data to the SAXS model, we generated a density map from a representative atomic SAXS-derived model by using the same sampling as for the EM images, and computed projection images from this map to illustrate how the SAXS structure would appear in EM under different angles of view. For comparison between EM and SAXS, we manually assigned a computed projection to each class average (Fig. 3E, Lower). The comparison shows that the SAXS-derived model exhibits the same shape and overall dimensions as the particles imaged by EM. Overall, the SAXS and EM models appear in line with each other up to the imposed twofold symmetry in the SAXS model, i.e., some EM class averages suggest a deviation from a perfect twofold symmetry in particular in the outer protuberances.

C1 Is a Hollow Particle with the SP Domains Pointing Away from the Core.

We next examined the solution structure of nonactivated C1 with SAXS. C1 was assembled by mixing C1q purified from plasma with purified recombinant C1r₂(S/A)s₂(S/A) tetramer (Fig. 2 A and B) in which both C1r and C1s were inactive because the serine in the catalytic triade was mutated to alanine. We used sucrose gradient centrifugation to separate the reconstituted C1 from unbound C1r₂s₂ (Fig. 2C). The SAXS data were recorded inline immediately following elution of the C1 from a size exclusion chromatography (SEC) column. This approach allowed us to separate a small amount of oligomeric C1 from monomeric C1. We selected scattering curves from the center of the elution peak for merging to optimize the signal-noise ratio and to minimize the possible contribution from unbound C1q and C1r₂(S/A)s₂(S/A) eluting after the C1 complex. Furthermore, a stable value for the radius of gyration R_g (Fig. S2A) argued against significant C1 dissociation in the late eluting part of the peak, and interestingly, the R_g value of C1 was smaller than those observed for C1rs and C1q (Figs. S2A and S3 and Table S1). The Guinier plot of the C1 data were linear (Fig. S2B), and for subsequent data analysis, we used data with q < 2.5 nm⁻¹. The Kratky plot of the data suggested that C1 is a multidomain protein with several well-folded domains with some flexibility (Fig. S2C). Ab initio modeling produced a model featuring a hollow shape similar to a saddle, which is incompatible with a densely packed core in which the C1r and C1s protease domains are tightly packed in the center of C1 (Fig. S2D).

To elucidate the structural arrangement of the domains within C1, we performed rigid-body modeling assuming a central twofold axis running in parallel with the N-terminal segment of the C1q collagen stem. We started from the same tetramer model as for C1r₂s₂ except that the SP domain formed a single rigid body together with the CCP domains for both C1r and C1s, and glycans were omitted to reduce computational complexity. The collagen stems of C1q were restrained to the CUB domains of C1r₂s₂ (Fig. 4A). Common to all of the resulting models derived by rigid-body refinement was a hollow core as also observed in the ab-initio reconstruction, with the protease domains of C1r and C1s protruding away from the core. Fifty models were divided into two groups according to their fit to the experimental data. The first group comprised 44 models with χ² = 2.38 ± 0.050, whereas the remaining 6 models had χ² = 2.6 ± 0.082. In the major group, there was a clear twist of the tetramer compared with the planar starting model of the C1r₂s₂ tetramer (Fig. 4 B and C). Generally, the protease domains stretch out to form a longer and slimmer particle perpendicular to the twofold symmetry axis compared with the starting model. In a few models, the protease domains of C1s and C1r are pointing upwards toward the N-terminal collagen stem of C1q (Fig. S2E). In 4 of the 44 models, the protease domains occupy the same general area as the globular heads of C1q. In these models, there is a shift upwards of two or more of the globular heads to compensate for the SP domains of C1r₂s₂ such that the mass distribution between gC1q and SP domains is roughly conserved.

The minor group of output models is much more heterogeneous, but generally, the same features are observed. We observe the same orientations of the SP domains, where they can either be in the central plane, pointing upwards, but never packed inside the central void delimited by the collagen stems, the globular heads of C1q, and the CUB1-EGF domains of C1r₂s₂. To confirm that the input model did not bias our results, we conducted another 50 rigid-body refinements where the C1r CCP domains and the SP domain were folded into the core of C1 similar to the prevalent C1r zymogen model (10), with only minor adjustments due to geometrical constraints (Fig. S4A). In all of the resulting rigid-body output models, the C1r CCP1-CCP2-SP domains had swung out from the core to point away from the core as seen with the first input model (Fig. S4B).

We also recorded negative stain images of C1 molecules under identical conditions as for C1r₂s₂. Again, we selected specimen regions with a typical single carbon layer appearance to avoid flattening effects. We clustered the resulting dataset into an average 28 images per class by using unbiased clustering techniques independent of any SAXS model. The resulting class averages show maximum dimensions of 32–36 nm (Fig. 4D), which is slightly lower than the D_max derived from SAXS. In the predominant views, 8–10 globular domains are arranged around a central network of densities. The central portions of the negatively stained class averages appear rather dark, which may be explained by an accumulation of heavy metal ion stain in this region. This particular staining behavior is thus in agreement with a rather hollow structure of the particle as suggested by both SAXS ab-initio and rigid-body modeling.

When comparing the EM class averages to a representative SAXS model of the C1 complex, the SAXS model predicts up to 10 globular domains that protrude from the stalk of C1 in top/bottom views of C1. Furthermore, the SAXS model predicts side views, where the C1q hub is seen as one of the protruding domains (compare Fig. 4C, Left, with e.g., upper protuberance labeled “h” in Fig. 4D). An EM class average with 10 individual globular protrusions is shown in row 3, view 3 of Fig. 4D. However, the majority of class averages show eight to nine discernible peripheral domains. For interpretation of views with less than 10 protruding domains, it is relevant to note that in certain projection views, some domains may be superimposed and, thus, masked in the class averages.

We interpreted class averages, where all protruding domains appeared as small roundish densities as top/bottom views (Fig. 4D), which implies that the central densities may represent overlays of the C1q hub with central portions of C1r₂s₂. The frequent occurrence of top/bottom views may be explained by a preferred binding of the C1q globular domains to the carbon film. The distances between the globular domains appear to be somewhat variable, indicating a certain degree of conformational variability. Such variability is not reflected in the SAXS models where twofold symmetry was imposed to minimize the number of rigid bodies to reduce overfitting and computational complexity.

Although the level of detail does not allow distinguishing C1q head domains and C1r₂s₂ SP domains, the class averages we obtained for C1 in negative stain strongly argued against a compact conformation of C1r₂s₂ within the complex. Such a compact C1 conformation would give rise to projection views showing up to six protruding domains dependent on the angle of view and degree of overlay (2). Moreover, opposing globular domains were typically separated by a projected distance of at least 29 nm, which likewise argues in favor of a rather long distance between C1r₂s₂ SP domains in the C1 complex.

To further investigate the indicated C1 model, we also performed native cryo-EM (Fig. 4E, columns 1 and 3). A particular challenge was the low contrast of the C1 complex under cryo conditions, which can be explained by the extreme sparsity of protein density in this complex, as a mass of only 774 kDa is distributed in a sphere of 38 nm diameter. For comparison, the C1 complex is more than 50% larger than a prokaryotic ribosome by size while containing only ∼1/3 of its mass. We used highly diluted samples of C1 to minimize aggregation, resulting in approximately 1–10 particles per exposure. The cryo class averages (Fig. 4E) comprise 65 images per class. As cryo images are taken at a higher defocus compared with negatively stained particles, there is a somewhat higher amount of blurring, which, in turn, means that individual globular domains may appear superimposed. Nevertheless, we could distinguish six to nine protruding domains in most class averages (Fig. 4E, columns 1 and 3). Moreover, the angular distribution of views appeared to be more uniform than in the negative stain data, leading to a mean diameter of ∼38 nm in the cryo class averages. As for the C1r₂s₂ complex, we converted a representative C1 SAXS model into an identically scaled density map and computed projections under different angles of view (Fig. 4E, columns 2 and 4). Again, a computed projection was manually assigned to each class average. In this comparison, it can be seen that there is a good overall agreement in the outlines of the class averages and assigned SAXS projections. However, the exact position of the individual globular domains may vary between the EM class averages and the single SAXS model used here for projection. These discrepancies may originate from structural heterogeneity of C1, and the individual flexibility of both the C1q and C1r₂s₂ components may lead to a considerable deviation from the twofold symmetry assumed for the SAXS rigid-body modeling. Such a conformational flexibility is in good agreement with the variations found between the different SAXS models that we computed as described above.

Expression and Purification of Proteins.

The genes according to the sequence for C1r (NM_001733.4) and C1s (NM_201442.2) were synthesized by GenScript Inc. USA. The synthesized genes were cloned into the vector pcDNA3.1/myc-His(-) A (Invitrogen) using the restriction sites EcoRI/XbaI. Inactive C1r and C1s were generated by mutating the active serine of the catalytic triad in C1r(S637A) and C1s(S632A). The human embryonic kidney cell line HEK293-F (Invitrogen) was used for transient expression of recombinant inactive versions of C1r and C1s. HEK293-F cells were grown in protein-free medium (FreeStyle 293 Expression Medium; Gibco) with agitation at 37 °C and 8% CO₂. Transfection was done by using PEI (25 kDa; Polysciences) as transfection reagent in a PEI:DNA ratio 3:1. After transfection, the cells were grown for 4–5 d, and the supernatants were harvested after centrifugation.

C1 and C1r₂s₂ Purification for SAXS Studies.

The tetramer C1r₂(S637A)C1s₂(S632A) was purified according to Bally et al. (7), whereas C1q was purified from serum according to Tenner et al. (27). To reconstitute C1, C1q was mixed with a molar excess C1r₂s₂ and loaded on a 10–30% (wt/vol) sucrose gradient in 50 mM EPPS (4-(2-hydroxyethyl)-1-piperazinepropanesulfonic acid) (pH 8.5), 145 mM NaCl, 3 mM CaCl₂, and centrifuged in a TH-660 rotor (Sorvall, ThermoFisher Scientific) at 185,795 × g for 17 h at 4 °C. After ultracentrifugation, fractions were collected from the bottom of the centrifuge tube by using a peristaltic pump and analyzed by SDS-PAGE. Before in-line SAXS analysis of C1, it was dialyzed against 50 mM EPPS (pH 8.5), 145 mM NaCl, 3 mM CaCl₂, and concentrated immediately before SAXS measurements. C1r₂s₂ and C1q were purified like C1 on a 10–30% (wt/vol) sucrose gradient in 50 mM Tris⋅HCl (pH 7.4), 145 mM NaCl, 3 mM CaCl₂. Before SAXS data collection, the samples were dialyzed against 50 mM Tris⋅HCl (pH 7.4), 145 mM NaCl, 3 mM CaCl₂, and concentrated immediately before SAXS measurements.

SAXS Data Collection and Processing.

The data for C1r₂s₂ and C1q were collected in batch mode (ranging from 1.0 to 6.8 mg/mL) at the ESRF BM29 beamline by using a PILATUS 1M pixel detector and λ = 0.992 Å in a temperature-controlled capillary at 4 °C. The sample-to-detector distance was 2.867 m, covering a range of momentum transfer 0.004 < s < 0.5 Å⁻¹ [q = (4 × π × sinθ) ÷ λ, where 2θ is the scattering angle]. For the C1 complex, in-line SAXS size exclusion chromatography was done at the PETRA III/EMBL P12 beamline by using a Superose 6 10/300 GL column, equilibrated in 50 mM EPPS (pH 8.5), 145 mM NaCl, 3 mM CaCl₂. The flow rate was 0.3 mL/min. The data were recorded at 20 °C by using a PILATUS 2M pixel detector (DECTRIS) and λ = 1.240 Å. The sample-to-detector distance was 3.0 m covering 0.002 < s < 0.48 Å⁻¹. Each frame covers a 0.995-s exposure performed every second during the SEC run. Normalization and radial averaging was performed at the beamline by using the automated pipeline (28, 29). Data reduction to produce the reduced and buffer-subtracted scattering profile were performed by following standard methods (30). The background and C1 data were obtained from frames 3800–3900 and frames 1851–1950, respectively.

SAXS Ab Initio Modeling and Rigid-Body Refinement.

In the C1r₂s₂ tetramer, two C1s molecules are located centrally with one C1r molecule on the outside of each C1s molecule (2, 3). The input model for CORAL refinement of the C1r₂s₂ tetramer was constructed in three steps (Fig. S1C). First, the CCP1-CCP2-SP fragments were constructed by the combination of structures of zymogen C1r (PDB ID codes 1GPZ and 1MD7), whereas for zymogen C1s, ID code 4J1Y was used. Second, a CUB1-EGF tetramer was constructed based on a crystal packing tetramer in the structure of the C1s CUB1-EGF-CUB2 fragment (ID code 4LMF) suggested to represent a possible model for the C1r₂s₂ CUB1-EGF tetramer (3). The C1r CUB1 and CUB2 domains were modeled by using phenix.sculptor (31) starting from the structure of the Ca²⁺ bound C1s CUB1-EGF-CUB2 (ID code 4LOR). These models were then combined with the structure of the C1r EGF domain (ID code 1APQ). In the ID code 4LMF, the CUB2 domain is bended out of the plane of the CUB1-EGF tetramer. We modeled an approximate in-plane orientation of both the C1r and C1s CUB2 domains by comparison with the structure of MAP-1 (ID code 4AQB), which is better compatible with simultaneous binding of collagen stems to both CUB1 and CUB2 domains. In the third step, the C1r₂s₂ tetramer model was completed by combination of the CUB1-EGF-CUB2 tetramer with the CCP1-CCP2-SP fragments of C1r and C1s using the structure of the C1s CUB2-CCP1-CCP2 fragment (ID code 4LOT) as a guide. Finally, smaller missing regions in this tetramer model were manually modeled within the program “O” (32), and two Asn-linked glycans on C1s and four on C1r were modeled as described (33). However, because the spatial relation between the C1r CUB1-EGF domains and the corresponding domains from C1s is unknown, we also constructed two alternative start models with 27 and 54 Å in-plane translations of C1r relative to the fixed C1s (Fig. S1 D and E).

The backbone and β-carbons of the C1q collagen stems were modeled by using the interactive Triple-Helical collagen Builing Script (THe BuScr) (34). The C1q collagen stem was modeled in two separate pieces with one fragment consisting of the amino acids before kink, and the other consisting of the amino acids after the kink. The chains were modeled with the glycine of chain A hydrogen bonding to the next amino acid in the tripeptide unit in chain C, the glycine of chain C hydrogen bonding to chain B, and the glycine of chain B hydrogen bonding to chain A as described in ref. 4. An all-atom model was obtained from the backbone and β-carbon model by using a side chain rotamer library (35) followed by the addition of the hydroxyprolines with THe BuScr. In the final model of the collagen stem of C1q, the collagen stem after the kink was added as two separate pieces to allow for better solvation in CORALXL. The kink was not modeled in details, instead the two parts of C1q was held together by two distance restrains to allow flexibility in the angle between the two stems. Finally, the complete C1 molecule was assembled by placing lysine residues in the C1q collagen stems known to be important for interaction close to the calcium sites in CUB1 and CUB2 domains in C1r and the C1s CUB1 domain (7) to establish a starting geometry for interaction between the collagen stem and CUB domains as observed in ID codes 4LOR (C1s CUB1-collagen) and 3POB (MASP-1 CUB2-collagen). Only the C1r₂s₂ tetramer model with a C1r translation of 0 Å was used as subsequent placement of the collagen stems onto the CUB domains in the two models with C1r translations was found to require a tight clustering of the collagen stems at opposite ends of the CUB1-EGF-CUB2 tetramer.

The rigid-body refinements were done by using data range in the range s < 0.25 Å⁻¹ with CORALXL, a custom-made version of CORAL (36) allowing 30 rigid bodies and 1,000 distance restraints compiled by Maxim Petoukhov at the EMBL Hamburg. A twofold symmetry operation was applied by using P2 symmetry to the input models to minimize the number of rigid bodies. Distance restraints were used to maintain the interface between symmetry-related C1s CUB1-EGF domains across the twofold rotation axis. For C1r₂s₂, three distance restraints were applied, whereas 28 restraints were used for C1. Several values of these restraints (connectivity and distance) were tested before the final refinements and evaluated based on internal consistency between the resulting models and biological relevance. A representative output model and the data used for rigid-body refinement are deposited at the SASBDB for both C1r₂s₂ and the C1 complex (Table S1). Plots were prepared with GraphPad Prism 5.03. The Porod volume was determined by in-house software written by J.S.P. A Guinier expression was fitted to the low-q part of the data to determine I(0), and the Porod law I(q) ∝ q⁻⁴ with an added constant was fitted to the high-q part to determine the constant to be subtracted from the data so that they follow the Porod Law. The Porod invariant was calculated by numerically integrating the experimental data multiplied by q² with appropriate extrapolations to q = 0 and q = ∞.

EM.

For negative-stain EM of the C1r₂s₂ complex, a peak fraction from size exclusion chromatography on a 24-mL Superdex 200 column (GE Healthcare) equilibrated in 10 mM Hepes (pH 7.4), 150 mM NaCl, and 3 mM CaCl₂ was diluted appropriately. Carbon film prepared on mica was floated on freshly prepared sample for 2 min, and grids were prepared by using the sandwich method (37). To prepare C1 samples for negative-stain EM and cryo-EM, the C1 complex was subjected to sucrose gradient centrifugation as described above for preparation of the SAXS sample. Fractions containing the C1 complex (Fig. 2C) were subjected to SEC on a Superose 6 10/300 GL column, equilibrated in 50 mM EPPS (pH 8.5), 145 mM NaCl, 3 mM CaCl₂, and the peak fraction was prepared according to the sandwich method for the negative-stain EM. Grids were imaged in a Tecnai T12 microscope (FEI/Thermo Fisher Scientific) at room temperature at a high tension of 60 kV on a Gatan Multiscan 794 CCD camera (Gatan) at a nominal magnification of 42,000× resulting in a pixel size of 4.2 Å per pixel. For C1r₂s₂, a defocus of −1.5 to −2.1 µm was used. A total of 7,551 particles were manually selected from the images, corrected for defocus (38) and clustered as described (12) into on average 34 images per class. For the C1 complex, a defocus of −1.2 to −1.4 µm was used, and 14,331 particles were selected and clustered into on average 28 images per class.

For cryo-EM analysis of C1 material from SEC was also used. A continuous carbon film was mounted on Quantifoil (Jena, Germany) 3.5/1 grids. Appropriately diluted sample was adsorbed on a glow-discharged grid for 1 min, followed by manual blotting and freeze-plunging into liquid ethane by using a Leica EM CPC system (Leica Microsystems). The sample was imaged in a Titan Krios EM at 200 kV (FEI), a defocus of −5 to −8 µm and 59,000× nominal magnification on a Gatan US4000 camera resulting in a final pixel size of 1.2 Å per pixel. Because the C1qrs complex generates a rather low contrast under cryogenic conditions, a low-pass Wiener filter was applied to the images, and 12,890 images were selected manually. Images were clustered into on average 65 images per class.