Properdin oligomers adopt rigid extended conformations supporting function

Properdin stabilizes convertases formed upon activation of the complement cascade within the immune system. The biological activity of properdin depends on the oligomerization state, but whether properdin oligomers are rigid and how their structure links to function remains unknown. We show by combining electron microscopy and solution scattering, that properdin oligomers adopt extended rigid and well-defined conformations which are well approximated by single models of apparent n-fold rotational symmetry with dimensions of 230–360 Å. Properdin monomers are pretzel-shaped molecules with limited flexibility. In solution, properdin dimers are curved molecules, whereas trimers and tetramers are close to being planar molecules. Structural analysis indicates that simultaneous binding through all binding sites to surface-linked convertases is unlikely for properdin trimer and tetramers. We show that multivalency alone is insufficient for full activity in a cell lysis assay. Hence, the observed rigid extended oligomer structure is an integral component of properdin function.

Introduction 3 The complement system is an essential aspect of innate immunity providing a first line of defense against 4 invading pathogens as well as maintenance of host homeostasis. The complement cascade is activated 5 when circulating pattern recognition molecules recognizes molecular patterns on a pathogen, dying host 6 cells or immune complexes. Activation can initiate through the classical pathway (CP), the lectin pathway 7 (LP) or the alternative pathway (AP) where the AP also provides an amplification loop for the two other 8 pathways (1). In all three pathways, labile protein complexes known as C3 and C5 convertases are 9 assembled. These convertases conduct proteolytic cleavage of complement component C3 and C5, 10 respectively, resulting in the generation of opsonins (C3b and iC3b), anaphylatoxins (C3a and C5a) and 11 assembly of the membrane attack complex (reviewed in (1)). 12 In the alternative pathway, the C3 convertase C3bBb is formed when a complex between C3b and the 13 serine protease factor B (FB) is activated by factor D. At a high surface density of C3b, this C3 convertase 14 becomes a C5 convertase. Properdin (FP) is a positive regulator of these convertases. FP is a 53 kDa protein 15 composed of an N-terminal TGF-β binding (TB) domain followed by six thrombospondin type I repeats 16 (TSR1-6). The protein is heavily post-translationally modified carrying one N-linked glycan, four O-linked 17 glycans and 14-17 C-mannosylated tryptophan residues in the WxxW motifs present in TSR1-6 (2,3). FP is 18 produced predominantly by monocytes, T-cells and neutrophils and circulates as oligomers and is primarily 19 found as dimers, trimers and tetramer with a 1:2:1 molar distribution in plasma at a concentration of 5-25 20 μg/ml (4,5). The functions of FP in relation to AP convertases are well established. 1) FP enhances the 21 recruitment of FB to C3b and thereby stimulates proconvertase assembly; 2) FP slows the dissociation of 22 C3bBb 5-10 fold; 3) FP directly competes with factor I resulting in decreased irreversible degradation of C3b 23 to iC3b. Other suggested functions of FP are as a C3b independent pattern recognition molecule capable of 24 triggering the AP (reviewed in (4)) and as a ligand for the NKp46 receptor on innate lymphoid cells (6). The 25 importance of FP in innate immunity and homeostasis is demonstrated by individuals with FP deficiency 1 (PD). PD is a rare X-linked disorder which can be divided into three subtypes: type-I (complete lack of FP), 2 type-II (1-10% of normal plasma FP level) and type-III (normal plasma level but dysfunctional FP). All PD 3 types are characterized by reduced AP activity resulting in impaired bactericidal activity and increased 4 susceptibility to Neisseria infections and sepsis (7). 5 Classic negative stain (ns) EM studies of FP dimer, trimer and tetramers revealed that FP oligomers contain 6 compact eye shaped vertexes connected by thin connecting structures (8,9). Alcorlo and coworkers 7 presented the first 3D reconstruction of the FP eye shaped vertices in oligomers and 2D classes of the FP-8 C3bBb convertase complex (10). Recently, crystallographic structures demonstrated that FP oligomers are 9 formed upon interaction of the TB domain and TSR1 from one FP monomer with TSR4, TSR5 and TSR6 from 10 a second FP monomer ( Fig 1A). In addition, it was established that the binding site for C3b is formed by FP 11 TSR5 in conjunction with a large loop from TSR6 (2,3,11). Despite prior attempts to analyze FP dimers and 12 trimers with small angle X-ray scattering (SAXS) and analytical ultracentrifugation (12), detailed information 13 regarding the structure and the dynamic properties of FP oligomers is missing. Here we present for the first 14 time a structural description of intact oligomeric FP obtained through a combination of nsEM and SAXS. 15 Based on the recent crystal structures of monomeric FP, we are able to annotate all FP domains in 16 monomeric, dimeric, trimeric and tetrameric FP in EM 2D classes. Pair distance distributions and atomic 17 models based on solution scattering suggest that the FP oligomers are rather rigid in solution despite their 18 very open structure and that their average conformations have cyclic symmetry. In addition, we 19 demonstrate that the defined structure of FP oligomers is crucial for their biological function. 20

21
The FP monomer is pretzel shaped 22 Human FP with a C-terminal His-tag was expressed by HEK293F cells and purified by affinity 23 chromatography. The different FP oligomers were subsequently separated by cation exchange and size 24 exclusion chromatography (SEC). As expected, both recombinant and plasma derived FP eluted in multiple 25 obtained in the absence of hFPNb1 or intermediate between these two extremes. Finally, we analyzed the 1 tetramer FP4 by nsEM and again >90% of the picked particles contributed to 2D classes with flat molecules 2 of pseudo C 4 symmetry with a maximum extent of 38 nm (Fig 2E). For both FP3 and FP4 2D classes, a kink 3 was occasionally observed, presumably at the TSR2-TSR3 interface, in at least one of the TSR2-TSR3-TSR4  4 connections. In summary, our nsEM analysis revealed unambiguous 2D classes for all the naturally 5 occurring FP oligomers, and in all three cases the vast majority of classified particles were present in rather 6 similar classes representing flat molecules with apparent C n symmetry. Furthermore, the oligomer 7 conformations observed in our 2D classes were not rare since the sum of particles used for the 2D classes 8 for all three oligomers represented at least 50% of the particles picked (Fig. 2). Intriguingly, we also 9 observed that the TSR4 specific hFPNb1 could induce alternative folded conformations of the TSR2-TSR3-10 TSR4 arms in FP2 and FP3 that appeared to be in equilibrium through intermediates with their elliptical and 11 flat triangular conformations. 12 The solution conformation of FP oligomers 13 Taking an orthogonal approach to our nsEM analysis of the FP oligomers, we collected SEC SAXS and static 14 SAXS data for FP2, FP3 and FP4 that was SEC fractionated prior to SAXS analysis to obtain samples 15 optimized with respect to the relevant FP oligomer (Figs 3A-C and S2A-G). It is noteworthy that the SAXS 16 data of all three oligomers exhibit characteristic bumps which are also reflected in their p(r)-functions (see 17 inserts to SAXS data and fig S2H). The presence of these pronounced features clearly indicate that the 18 oligomers are rather rigid and well-defined. If conformational freedom had been present, this would smear 19 out the SAXS data. The p(r)-functions of FP1, FP2, FP3 and FP4 all exhibit an initial bump at around 4 nm 20 corresponding well to the repeated distance across the eye that is also visible in the nsEM pictures of all 21 three oligomers (see also fig S2H). The FP2 has a D max value of 23 nm. Along with a second well-defined 22 peak at around 8 nm, which is most likely related to the distance between the two antiparallel arms 23 connecting the two eyes, the D max value indicates that the FP2 solution structure is in better agreement 24 with the extended conformation of FP2 (Fig 2A), than it is with the more compact twisted conformation 25 induced by hFPNb1 ( Fig 2B). Finally, the p(r) of FP2 has a broad peak at 16.5 nm corresponding well to the 1 distance between the two eyes. The D max of the FP3 and FP4 p(r) functions are, respectively, 25 and 36 nm 2 Fig S2H) and in good agreement with their larger sizes also seen in the nsEM 2D classes (Figs 3 2C and 2E). For the FP3 and FP4, the p(r) middle peak at around 8-10 nm appears broader, less pronounced 4 and moves to higher values as the oligomer size increases. FP3 and FP4 have a high p(r) peak at 18 and 19.5 5 nm, respectively, which is most likely the result of neighbor eye-eye distances. The increase of the eye-eye 6 peak position when comparing FP2, FP3 and FP4 suggests that the eyes reorient towards a more planar 7 structure as the oligomeric state increases. Furthermore, the FP4 p(r) function exhibits an additional small 8 shoulder at 28 nm which corresponds well to the less frequent diagonal eye-eye distances. For a quadratic 9 structure, as suggested by the nsEM, these would appear at √2 times the position of the neighbor eye-eye 10 distances at 18-19.5 nm as they indeed do. 11 Using the same strategy and restraints as for FP1, we obtained rigid body models of FP2 in the presence of 12 C 1 (no) symmetry or a C 2 symmetry axis with χ 2 in the range 1.4-3.0 for C 1 symmetry whereas models with 13 C 2 symmetry had χ 2 of 2.4-5.5 ( Fig 3A and Fig S2C). The resulting SAXS-based FP2 models all featured an 14 extended FP2 conformation rather than the double-pretzel FP2 observed by nsEM, even if refinement was 15 started from a double-pretzel conformation mimicking that presented in figure 2B. Interestingly, the FP2 16 SAXS models appeared more curved as compared to the extended conformations present in EM 2D classes. 17 In contrast to the nsEM 2D classes, the rigid body models can not be projected such that the two eye-18 shapes at the opposite ends of FP2 become visible simultaneously. Hence, either rigid body modelling could 19 not reach the conformation observed in nsEM, or the FP2 conformation is influenced by the stain and 20 contacts with the grid and therefore become excessively flat in our 2D nsEM classes. An effect of the grid is 21 in accordance with a lower solution D max compared to the maximum extent of FP2 in nsEM 2D classes. 22 Using the same strategy we generated models of FP3 and FP4 by rigid body modelling. In contrast to FP2, 23 the χ 2 values for the fit of the best models to the experimental data were comparable and in the range 1.1-24 1.4 indicating that the experimental data could be fitted well with single models. Importantly, the χ 2 values 1 of the output models generated with C 3 /C 4 symmetry were similar to those generated with C 1 symmetry 2 demonstrating that the fit to the data was independent of symmetry ( Fig 3B-C and S2D-E). All models 3 generated with C 1 symmetry were qualitatively similar extended circularized structures with their 3 or 4 eye 4 shaped structures joined by peripheral TSR2-TSR3-TSR4 connecting arms in agreement with the FP3 and 5 FP4 nsEM 2D classes. As for FP2, in the models generated with C 1 symmetry, it was never possible to obtain 6 projections of these SAXS models in which all eyes in FP3 and FP4 were visible. Otherwise the SAXS models 7 had strong resemblance to the 2D classes including an occasional kink in a TSR2-TSR3-TSR4 connection (Fig  8   3B-C). When C 3 /C 4 symmetry was assumed during refinement, some models were trapped in a local 9 minimum and gave rise to highly elevated χ 2 values and models that were distinctly different from those 10 with the lowest χ 2 values. In contrast, the best fitting models were all open and extended with TSR2-TSR3 at 11 the periphery similar to those obtained with C 1 symmetry. For some of these models, it was possible to 12 visualize all eyes simultaneously in projections (Fig 3B-C) which is in agreement with FP3 and FP4 nsEM 2D 13 classes. Overall, our SAXS analysis of FP oligomers were in agreement with the corresponding nsEM 2D 14 classes and suggests that the average solution conformation of FP3 and FP4 -as expressed in the SAXS data 15 -are well approximated by single models with C 3 and C 4 symmetry (S2D-E). With respect to FP2, the 16 situation is less clear as larger differences were observed for rigid body modelling with C 1 and C 2 symmetry 17 and the resulting models appear more curved than those appearing in the nsEM classes. 18 The oligomerization interfaces can undergo very slow exchange

19
The oligomer distribution of FP in plasma is believed to be stable after secretion, but whether monomer-20 monomer interactions occasionally loosen up and whether monomers may exchange between oligomers 21 has never been addressed. Our prior purification of the two-chain monomeric FP molecules FPc and FPhtΔ3 22 (2,11) never suggested that the two chains dissociated under native conditions. However, in size exclusion 23 chromatography performed at low pH it was possible to separate the two FP chains (Fig. S4A) as previously 24 demonstrated for FP oligomers (5). To investigate the stability of the FP oligomerization interfaces, we 25 mixed our two-chain monomer FPc with the two-chain deletion mutant FPhtΔ3 monomer lacking TSR3 (Fig  1   S4B). After incubation at 37 °C, we purified FP molecules containing the His-tagged TSR4-6 tail fragment of 2 FPhtΔ3. Using SDS-PAGE analysis, we observed that over time an increasing amount of the longer head 3 fragment from FPc (TB-TSR1-3) co-purified with the His-tagged tail fragment from FPhtΔ3 while the amount 4 of co-purified short FPhtΔ3 head fragment (TB-TSR1-2) decreased (Fig. S4C). Exchange between the two FP 5 monomers was evident after 30 min and reached equilibrium after 12 hours. To examine monomer 6 exchange into an FP oligomer, we conducted the same experiment with FP2 and FPhtΔ3. We observed that 7 over time, an increasing amount of full length FP from FP2 co-purified with the FPhtΔ3 his-tagged TSR4-6 8 fragment, and in this case exchange was evident after 2 hours and complete in 6-12 hours (Fig. S4D). In 9 conclusion, these experiments for the first time demonstrated that the oligomerization interfaces in FP can 10 open temporarily and even exchange monomers with a different FP molecule under physiologically 11 relevant experimental conditions. However, the exchange occurs rather slowly in our pure system and is 12 probably not a significant reaction in the extracellular environment after secretion. These results add 13 further support to the concept of stable FP oligomer conformations. 14 Convertase binding to FP oligomers 15 A major outstanding question concerning FP biology is whether the strong correlation between biological 16 activity and oligomer size is partly due to simultaneous binding to multiple C3b molecules and convertases 17 deposited on an activator. In FPn, there are n binding sites for C3b each located at the concave face of 18 TSR5, and our prior structures of FP-bound C3bBb revealed that C3b binds with its major axis roughly 19 parallel to the plane of the FP eye formed by the TB domain, TSR1, TSR5 and TSR6 (14). Hence, as illustrated 20 in Fig 3D for FP2, an oligomer with its n convertase binding sites pointing in the same direction will have the 21 optimal architecture for binding simultaneously to n C3b molecules deposited on an activator. Due to the 22 overall flat shape of FP oligomers, the minor principal axis is the C n rotation axis perpendicular to the plane 23 of the models generated with C n symmetry and a pseudo n-fold rotation axis for models generated with C 1 24 symmetry ( Fig 3E). We could therefore quantitate the relative orientation of the convertase binding sites in 25 our FP SAXS models by measuring the angle α between an appropriate vector lying in the plane of each FP 1 eye and the smallest principal axis of the oligomer (Fig 3E). At the maximum value, α=180°, the n 2 convertase binding sites in an FP oligomer will point in the same direction and simultaneously binding at all 3 convertase binding sites by C3b on a planar activator appears possible. At its minimum value, α=90°, the 4 convertase binding sites are parallel to the plane of the FP oligomer defined by the major and intermediate 5 principal axes and simultaneous binding to more than two C3b molecules on a planar activator appears 6 unlikely. We observe a clear decrease in α as a function of oligomer size with α~150°, 128° and 109° for our 7 SAXS-based models of FP2, FP3 and FP4, respectively ( Fig 3F). A decrease in α with increasing multiplicity is 8 in agreement with the shift of a major peak in the pair distance functions ( Fig S2H) described above. This 9 peak largely reflects the separation of neighboring FP eyes, and in curved oligomers this separation will be 10 smaller than in planar oligomers. 11 Oligomerization alone cannot rescue FP activity 12 The activity of FP oligomers in assays exploring complement-dependent erythrocyte lysis follows the order 13 FP4>FP3> FP2 (5), and the two chain monomer FPc and E244K FP1 are also much less active in convertase 14 stabilization on erythrocytes and bactericidal activity compared to oligomeric FP (11). To investigate the 15 importance of FP oligomerization and structure for activity we linked 2, 3 or 4 copies of hFPNb1 with 16 glycine-serine linkers and showed that noncovalent FPc oligomers of increasing molecular weight linked by 17 the multivalent hFPNb1 molecules could be formed (Fig 4A-B). Next, we compared the activity in 18 erythrocyte lysis in FP deficient serum of free FPc and nanobody linked FPc oligomers to an FP pool 19 containing roughly equal amounts of FP2 and FP3 ( Fig S1F). As expected, the two-chain FPc monomer 20 required a 100 fold higher concentration compared with the FP2/FP3 oligomer pool to elicit a similar 21 degree of lysis ( Fig 4C). The activity of the nanobody-linked FPc oligomers were in between the activities of 22 the FP2/FP3 pool and FPc and increased with the hFPNb1 valency in the oligomers. The erythrocyte lysis 23 activity of these hFPNb1 linked FPc oligomers also correlated well with their increasingly slower 24 dissociation from a biotin-C3b sensor ( Fig 4D). In conclusion, FP activity can only be partially restored by 25 linking FPc monomers through flexible linkers despite that these nanobody-linked monomers exhibited 1 increasingly stronger binding to a C3b coated sensor compared to the FPc monomer. Hence, the well-2 defined extended structure of FP oligomers demonstrated by our SAXS and EM data contributes 3 significantly to the biological activity of FP oligomers. 4

5
Our demonstration of well-ordered EM 2D-classes and our ability to explain solution scattering data with 6 single models make us suggest that FP oligomers adopt a limited number of fairly stable and overall similar 7 conformations in solution. This is a surprising finding considering that EM 2D classes of an FP oligomer have 8 not previously been presented and the large flexibility at especially the TSR2-TSR3 and TSR3-TSR4 9 connections required to reach the very tight FP1 conformation ( were flexible, these oscillations would be less well defined. 14 The solution structures of FP2 and FP3 were earlier investigated by Sun and coworkers with SAXS and 15 analytical ultracentrifugation (12) and gave R g values comparable to those presented in Fig. S2A. Their best 16 fitting FP2 models bear weak overall resemblance to our FP2 models, but their best fitting FP3 model is 17 rather different from the models we obtain. Furthermore, an extended model resembling our models of 18 FP3 did not fit the scattering data in the study by Sun et al. One reason for these discrepancies is beyond 19 doubt that Sun et al did not have a detailed model of FP to base their rigid body modelling on, and in 20 particular they missed the crucial information regarding the stable eye structure formed by the TB domain, 21 TSR1, TSR5 and TSR6 which we used as a single rigid body. They also used a TSR based homology model as a 22 proxy for the N-terminal region that is a TB domain. In addition, the resolution of SAXS data in (12) was 23 much lower with significant noise at q values above 1 nm -1 whereas we had only limited noise in the data in 24 the range q<2.7 nm -1 used for rigid body refinement. 25 The binding sites for C3b is formed by FP TSR5 and TSR6 located in the FP eye and oligomerization is not 1 strictly required for C3b binding, stimulation of C3bB assembly, inhibition of C3bBb dissociation and 2 competition with FI. These activities are supported to some degree by the recombinant monomeric two-3 chain FPc molecule in which TSR2 and TSR4 are not connected by TSR3 (2,11). Furthermore, EM analysis of 4 erythrocytes activated through the classic pathway show tight clusters of 10-40 C3b molecules extending 5 over 40-80 nm (15), which appears well compatible with the separation of C3b/convertase binding sites 6 observed in our structural models (Fig 3A-C). 7 Our demonstration that FP oligomers are structurally well ordered rather than dynamic modular structures 8 (Figures 2-3) and our finding that biological activity cannot be rescued solely by linking of multiple 9 monomeric FP molecules ( Fig 4D) lead us to present a model in figure 4E for the biological function of FP 10 oligomers. Initial monovalent binding occurs when a single FP eye engages with C3b, proconvertase or 11 convertase. Subsequent multivalent binding is likely to be stimulated by FB binding (16) and appears 12 feasible for neighboring convertase binding sites in all three types of oligomers and especially FP2, where 13 the two sites are arranged favorable with respect to each other for binding parallel C3b molecules 14 separated with an average separation of 20 nm (Fig 3D). The higher activity of FP oligomers could derive 15 from; i) An FP oligomer simultaneously binding multiple activator-bound C3b molecules; ii) a high local 16 concentration of C3b binding sites favoring FP rebinding after dissociation; iii) A defined 3D structure of the 17 oligomers e.g. matching the average distribution of deposited C3b. Our prior SPR data showed a 450× lower 18 apparent KD value for immobilized C3b and fluid phase oligomeric FP as compared to the reverse geometry 19 in accordance with contributions from multivalent binding and a high local concentration (11). In contrast, 20 classic binding experiments measuring association to erythrocyte bound C3bB and zymosan-C3b complexes 21 suggested that FP oligomers bind to a C3b opsonized activator in a monovalent fashion (16,17) arguing that 22 the local high concentration of unbound convertase binding sites for monovalent FP-convertase 23 interactions underlies the correlation between biological activity and oligomer stoichiometry. Our cell lysis 24 experiments demonstrate that multivalency alone is not sufficient obtain full biological activity suggesting 25 that the defined 3D structure of the oligomers also contributes significantly to the correlation between 1 activity and oligomer size. Although we do not know in details the oligomer distribution of FP in non-2 mammals, the domain structure with the TB domain followed by six thrombospondin repeats is with 3 certainty also present in properdin sequences from amphibians, reptiles, birds, teleosts and the agnatha 4 Petromyzon marinus. This evolutionary conservation supports that a defined spatial separation of 5 convertase binding sites in FP oligomers is important for its function. For antibodies, it is known that the 6 flexibility of the Fab-Fc hinge enables "walking" over the antigen in search for bivalent attachment on 7 spatially defined multivalent epitopes (18,19). But since our results indicate that FP oligomers are quite 8 rigid, these multivalent molecules seem not to be designed for "walking" over the C3b opsonized activator. 9 Instead, the inherent flexibility of C3b and its attachment through a single covalent bond to the activator 10 may favor multivalent FP-convertase complexes or fast rebinding through a neighboring convertase binding 11 site ( Fig 4E). 12 One reservation with respect to the model of FP structure-function relationships presented in Fig 4E is that  13 our EM 2D classes obtained with the TSR4 binding hFPNb1 nanobody revealed intricate folded 14 conformations and intermediates between the extended flat cyclic conformation observed in the absence 15 of the nanobody (Fig 2B and 2D). This emphasizes, that the FP oligomer structures we present in figure 3 16 reflect their solution conformation, but the structure may be quite different for oligomers bound to surface 17 bound C3b and convertases on activators and NKp46 on innate lymphoid cells. The variable internal 18 structure of TSR4 and rather different orientations of TSR4 relative to TSR5 observed in available crystal 19 structures (2,3) and the MD simulations of FP1 presented here agrees with the ability of a TSR4 binder to 20 induce profound conformational changes in FP oligomers. TSR4 is not directly part of the convertase 21 binding site, but it is linked to TSR5 that harbors the majority of the convertase binding site. In an FP 22 oligomer, a conformational signal may be relayed from TSR4 to TSR5 upon convertase binding and 23 propagate into the rest of the oligomer. The strong preference of FP oligomers for orienting with the plane 24 of the molecule parallel to the grid suggests that achieving high-resolution cryo-EM 3D reconstructions of 25 these unique extended structures will be an extremely challenging task. Alternatively, FP oligomers bound 1 to a C3b or convertase coated activator model may be studied with cryo electron tomography as pioneered 2 by Sharp and Gros for large assemblies of complement proteins (20,21). 3 Interestingly, our structures also predict that empty C3b/convertase binding sites in FP3 and FP4 point in a 4 direction opposite to the occupied site(s) due to their α angle being far from 180° (Fig 3F and 4E). We 5 estimate that such unoccupied convertase binding sites in FP3 and FP4 may protrude more than 30 nm 6 from the thioester-activator linkage of the C3b to which an existing monovalent interaction occurs (Fig 4E). 7 Such empty binding sites appear also to be suitable for bridging a C3b opsonized cell with a second cell. The 8 role of FP mediated agglutination -where C3b tagged bacteria lump together -in bactericidal activity is not 9 well investigated. Historically, FP driven agglutination of erythrocytes was associated with large non-10 physiologically FP oligomers (22). Perhaps more relevant for in vivo activity, FP can also bridge C3b 11 opsonized bacteria with host innate lymphoid cells presenting the NKp46 receptor, an interaction shown to 12 be required for survival in an animal model of Neisseria meningitides infection (6) Acknowledgements, funding sources and conflict of interest 6 We thank the staff at the P12 beamline at PETRAIII for help during data collection and Karen Margrethe 7 Nielsen and Anette Hansen for technical support. We acknowledge access to computational resources from 8 the Danish National Supercomputer for Life Sciences (Computerome) and the ROBUST Resource for 9 Biomolecular Simulations (supported by the Novo Nordisk Foundation). This work was supported by the 10 Lundbeck Foundation (BRAINSTRUC, grant no. R155-2015-2666) and the Novo Nordisk Foundation 11 (NNF16OC0022058). The authors declare no conflicts of interest in relation to this manuscript. 12

13
Protein production and SEC assays 14 DNA encoding FP with C-terminal TEV-His sequence was generated by site directed mutagenesis and was 15 expressed by transient expression in HEK293F cells as described in (11). FP was purified from cell 16 supernatants using HisExcel column (GE Healthcare) and 1 ml Mono S column (GE Healthcare) as described 17 (14). Fractions containing FP1, FP2, FP3 or FP4 from the Mono S column were pooled and further purified 18 by SEC performed on a 24 mL Superdex 200 increase column (GE Healthcare) at 4 °C with a flow rate of 0.25 19 mL/min in a buffer containing 20 mM HEPES, 150 mM NaCl pH 7.5. The monomeric FP variants FPc and 20 FPthΔ3 used for exchange experiments were expressed and purified as described in (2,11), respectively. 21 hFPNb1 were expressed and purified as described in (30). SEC analysis of FPc-hFPNb1 complexes were 22 performed with 200 µL samples containing 14 µg FPc in complex with hFPNb1 or its multivalent derivatives 23 in a 4-fold molar excess with respect to the number of hFPNb1 subunits. Samples were incubated in SEC 24 buffer at room temperature for 15 min before injection. 25 SAXS data acquisition, analysis and rigid body analysis 1 SAXS data was collected at the EMBL beamline P12 at PETRA III in Hamburg, Germany (31). The 2 temperature for the sample changer and exposure unit was set to 8°C, and the detector and X-ray energy 3 was configured to give a q-range of 0.023 to 7.332 nm -1 , with = 4 sin (θ)⁄ , where is the wavelength 4 of the X-ray beam and θ is the half scattering-angle. The dimer data collected using the sample changer was 5 inspected for radiation damage and averaged and background subtracted using primus in the ATSAS suite 6 (Franke et al, 2017). SEC-SAXS data was collected with an in-line 24 ml Superdex 200 increase column 7 operated at a flow rate of 0.5 ml/min. The trimer and tetramer data from SEC-SAXS were reduced using the 8 chromixs tool from the ATSAS suite (SEC curves are presented in S2F-G). Water was used as reference to 9 convert data to absolute scale units of cm -1 (32). Initial Guinier analysis to verify that no larger aggregates 10 were present (data not shown) and indirect Fourier transformation to determine the p(r) functions were 11 performed using the BayesApp software (33) available through the "GenApp.Rocks" server maintained by 12 Emre Brookes at University of Texas. For plotting of the scattering data, the ~800 data points were 13 logarithmically rebinned into ~180 points with better high-q statistics. Also, the data are only plotted in the 14 fitted range out to about 0.25 Å -1 . Central model independent parameters of the SAXS analysis are 15 presented in Figure S2A. 16 Rigid body modelling of FP monomer and oligomers was performed in CORAL (34). The FP eye formed by 17 the TB domain, TSR1, TSR4, TSR5 and TSR6 were used as a single rigid body. TSR2 and TSR3 from the crystal 18 structure of FPc (entry 6RUS) formed two additional bodies that were linked to each other and the eye with 19 distance restraints. In addition, an intact Asn linked complex glycan (entry 3RY6) formed a fourth rigid body 20 linked to FP Asn428, whereas mannosyl groups linked to tryptophans and fucose-glucose disaccharides 21 linked to serine and threonine were included in the same rigid body as the TSR domain they form a 22 covalent bond with. The starting models were constructed such that two residues to be connected across 23 the rigid bodies were in proximity. Starting models of FP oligomers were generated by C n symmetry and 24 oriented with their rotation axis along the z-axis. For each FP system, six different eyes derived from entries 25 6S08, 6S0A, 6S0B, 6RUS, 6SEJ, 6RUR (2,3) were first evaluated with a consistent set of distance restraints to 1 identify the optimal eye for the final rigid body refinements. The α angle in Figure 3E-F was calculated as 2 the angle between a vector connecting the C α atoms of FP residues Ala402 and Ser345 and the shortest 3 principal axis of the SAXS rigid body models. If the average α was <90, the model was flipped to yield an 4 average α>90. Ser345 and Ala402 were chosen for definition of the FP eye vector, as their difference vector 5 is in the plane of the FP eye and roughly parallel to the long axis of the C3b in C3bBbFP complex (2). 6 Scattering data together with an example of output model and fit to the experimental data are deposited in 7 the SASBDB for the FP dimer, trimer and tetramer. Scattering data for the FP E244K monomer is available 8 as SASBDB entry SASDB69. 9 Modeling and MD simulations of the FP E244K monomer 10 The FP1 monomer obtained by CORAL rigid-body modeling was used as the template to construct an initial 11 atomistic model of FP E244K with Modeller9.18 (35) in which missing residues in loops connecting domains 12 were added and the disulfide bond Cys132-Cys170 was established . Glycosylations were added including 13 the Asn linked-glycan at Asn428, O-linked Glucoseβ1-3Fucose at Thr92, Thr151, Ser208 and Thr272, and C-14 mannosylations at Trp83,Trp86,Trp139,Trp142,Trp145,Trp196,Trp199,Trp202,Trp260,Trp263,Trp321,15 Trp324, Trp382, Trp385, and Trp388. The glycan at Asn428 was modeled as a complex glycan. FP1 E244K 16 was placed into a periodic cubic box with sides of 14.6 nm solvated with TIP3P water molecules containing 17 Na + and Clions at 0.15 M, resulting in ~300,000 atoms in total. The CHARMM36m force field (36) was used 18 for the protein. Force field parameters for N-and O-linked glycans were generated using the Glycan 19 Modeler module in the CHARMM-GUI web interface (37). Force field parameters for C-Mannosyl Trp were 20 obtained from Shcherbakova et al. (38). Neighbor searching was performed every 20 steps. The PME 21 algorithm was used for electrostatic interactions with a cut-off of 1.2 nm. A reciprocal grid of 128×128×128 22 cells was used with 4th order B-spline interpolation. A single cut-off of 1.2 nm was used for Van der Waals 23 interactions. MD simulations were performed using Gromacs 2019.4 or 2019.5 (39). The temperature and 24 pressure were kept constant at 300 K using the Nose-Hoover thermostat and at 1.0 bar using the Parrinello-25 Rahman barostat with a time constant of 5 ps and a frequency of 20 for coupling the pressure, respectively. 1 Two independent MD simulations (one microsecond for each) were performed to collect the 2 conformational ensemble. These sampled conformations were used for further ensemble refinement using 3 the Bayesian Maximum Entropy (BME) method guided by experimental SAXS data as described (40)(41)(42). By 4 tuning the regularization parameter in the BME reweighting algorithm, we adjusted the conformational 5 weights in variant degrees to improve the fitting with the experimental SAXS data for FP E244K. VMD and 6 PyMol were used for visualization of the conformational ensemble and movie preparation. The theoretical 7 SAXS curve for each frame was back calculated using Crysol3 (43). 8 Single particle negative stain EM data acquisition and analysis 9 All samples were purified on a 24 ml Superdex 200 increase size exclusion column equilibrated in 20 mM 10 HEPES pH 7.5, 150 mM NaCl and subsequently adsorbed to glow discharged carbon coated copper grids, 11 washed with deionized water and stained with 2 % (w/v) uranyl formate. Images were acquired with a FEI 12 Tecnai G2 Spirit transmission microscope at 120 kV, a nominal magnification of 67.000x and a defocus 13 ranging from 0.7 to 1.7 µm. Automated image acquisition was performed using leginon (44). For the FP1 14 monomer and its hFPNb1 complex, CTF estimation and subsequent particle picking and extraction was 15 carried out with cisTEM (45). For the remaining samples, CTF estimation, manual particle picking and 16 extraction were performed with RELION (46). Initial 2D classes were generated and used to set up 17 template-based particle picking. For all samples, 2D classification was performed in RELION. 18 FPc and FP exchange assay 19 The stability of FPhtΔ3 under acidic conditions was evaluated by SEC on a 24 mL Superdex200 column (GE 20 Healthcare) equilibrated in 20 mM HEPES, 150 mM NaCl pH 7.5 or in 100 mM glycine pH 2.3. Samples of 21 100 µL FPhtΔ3 at 1 mg/mL were injected and eluted at 0.5 mL/min at room temperature. Fractions were 22 analyzed by SDS-PAGE followed by silver staining using the SilverQuest silver staining kit (Thermo Fisher). 23 Fractions containing 0.1 M glycine pH 2.3 were neutralized with 100 µL 2 M Tris pH 8.5 before SDS-PAGE 24 analysis. Five µg of FPc or dimeric FP were mixed with 5 µg of the his-tagged FPhtΔ3 in 100 µL of 20 mM 25 1 h, 24 h or 7 days. As a positive control, 5 µg of FPc or FP were mixed with 5 µg of FPhtΔ3 in 100 µL 20 mM 2 HEPES, 150 mM NaCl pH 7.5 and subsequently acidified by adding 100 µL 0.1 M Glycine pH 2.3 to cause 3 oligomer dissociation (5). The sample was incubated for 5 min at room temperature before 20 µL of 2M Tris 4 pH 8.5 was added for neutralization. Pull downs were performed on all samples using 50 µL of Ni-NTA 5 beads. The beads were transferred to a 1 mL spin columns (Bio-Rad) and equilibrated in 100 mM HEPES, 0.5 6 M NaCl, 30 mM imidazole pH 7.5. The samples were then transferred to the columns and incubated for 2 7 min, followed by a 30 s centrifugation step at 70 g. The beads were washed 5 times with 500 µL of 100 mM 8 HEPES, 0.5 M NaCl, 30 mM imidazole pH 7.5 before bound protein was eluted with 80 µL of 100 mM HEPES, 9 0.5 M NaCl, 400 mM imidazole pH 7.5. The eluates were retrieved from the columns by a 30 s 10 centrifugation step at 70 g. The eluates were re-applied to the column and the centrifugation step was 11 repeated. The samples were analyzed under non-reducing conditions on a 12 % SDS-PAGE gel (GenScript) 12 using the SilverQuest silver staining kit (Invitrogen). The oligomeric FP used in this assay was approximately 13 90 % dimer and 10 % trimer as judged by SEC analysis performed on 24 mL Superdex200 increase column 14 (Fig. S1G). A SEC standard (Bio-Rad) was used for comparison. 15 Bio-layer interferometry assays 16 Bio-layer interferometry experiments were performed on an Octet Red96 (ForteBio) at 30 ⁰C and shaking at 17 1,000 RPM. Binding of hFPNb1:FPc complexes were tested on streptavidin sensor tips (SA, ForteBio) 18 equilibrated in assay buffer (PBS supplemented with 1 mg/mL BSA and 0.05 % Tween 20) and coated with 19 biotinylated C3b at 16 µg/mL for 10 min. FPc or hFPNb1:FPc complexes were prepared by size exclusion 20 chromatography and diluted to 50 µg FPc/mL in assay buffer. Association was monitored for 120 s followed 21 by 120 s dissociation in assay buffer. 22 Erythrocyte lysis assay for AP activity 23 Rabbit erythrocytes (Er) in 11.38 mM D-Glucose, 2.72 mM mono basic sodium citrate, 2.19 mM citric acid, 24 7.19 mM NaCl (Alsever's Solution, Statens Seruminstitut) were washed and re-suspended in AP assay buffer 25 (5 mM barbital, 145 mM NaCl, 10 mM EGTA, 5 mM MgCl 2 , pH 7.4, with 0.1 % (w/v) gelatin) to obtain a 6 % 1 (v/v) suspension. Samples of 20 μL human FP-deficient serum diluted in assay buffer and supplemented 2 with WT FP, FPc or hFPNb1-FPc complexes were prepared separately and then transferred to a V-shaped 3 bottom 96-well microtiter plate (Nunc) in duplicates. The WT FP used for this assay was approximately 50 % 4 dimer and 50 % trimer as judged by SEC analysis performed on a 24 mL Superdex200 increase (Fig. S1F). 5 Ten micro liter of the E r -suspension was then transferred to the assay plate. The plate was mixed well and 6 incubated for 2 h at 37 °C, and shaken every 30 min. Hemolysis was stopped by adding 40 µL ice-cold 0.9 % 7 NaCl, 5 mM EDTA to each well. The plate was centrifuged at 90 g for 10 min, and 50 µL of each supernatant 8 were subsequently transferred to a flat-bottom microtiter well plate (Nunc). Hemolysis was then 9 determined from the A 405 measured on a Victor3 plate reader (PerkinElmer). Results are expressed relative 10 to total hemolysis (obtained with water alone) and to background hemolysis (EDTA sample). properdin, the third component of complement (C3), and its physiological activation products. 47 Biochem J 252, 47-54 48 18.
Zhang, P., Liu, X., Liu, P., Wang, F., Ariyama, H., Ando, T., Lin, J., Wang, L., Hu, J., Li, B., and Fan, C. 1 (2020) Capturing transient antibody conformations with DNA origami epitopes. Nat Commun 11 ,  2  3114  3 19. Preiner, J., Kodera, N., Tang, J., Ebner, A., Brameshuber, M., Blaas, D., Gelbmann, N., Gruber, H. J., 4 Ando, T., and Hinterdorfer, P. (2014) IgGs are made for walking on bacterial and viral surfaces. Nat 5 Commun 5  with FP1 with the number of particles indicated. A magnified view of the 2D class marked by star is shown 5 to the right. C) As for panel B, but for the FP1-hFPNb1 complex. Compared to FP1, an additional mass marks 6 the location of hFPNb1 and hence TSR4 enabling assignment of the TB domain and the six thrombospondin 7 repeats in the magnified view to the right. D) Representative atomic model of FP1 E244K derived by rigid 8 body modelling against the SAXS data. E) Comparison of SAXS experimental data (11) and the fitted 9 scattering curve corresponding to the FP1 E244K model presented in panel E. Insert to E: p(r) function 10 derived from the SAXS data. F) Conformational ensemble of FP1 E244K sampled by a one μs MD simulation 11 represented by 100 frames with 10 ns interval shown as transparent tubes. The starting model is displayed 12 as a cartoon with the glycans and glycosylated residues in grey stick representation. Disulfide bridges are 13 represented by yellow sticks. G) Comparison of SAXS experimental data and the scattering curve obtained 14 from MD ensemble after refinement using the Bayesian maximum entropy approach, the two curves fit 15 with χ 2 = 1.4. The minor difference in the experimental data apparent at the highest q-values in panels E 16 and G is due to subtraction of a constant by CORAL in panel E. 17 magnified view of the 2D class marked by star is shown to the right with the putative domain assignment 20 indicated for one of the two monomers in the dimer. Notice the difference in curvature at TSR2 and TSR4 of 21 the connecting arms, which facilitates the domain assignment. B) As panel A, but with the FP2-hFPNb1 22 complex. The double pretzel conformation is present in two classes, while others feature the elliptical 23 shape or intermediates. C) The 2D classes obtained with FP3 reveal a flat molecule structure of apparent C 3 24 symmetry. D) The FP3-hFPNb1 complex. The 2D class to the left reveals an FP3 conformation with an 25 apparent C 3 symmetry that is radically different from the cyclic appearance of FP3 in panel C, but 2D classes 1 presenting flat cyclic FP3 and intermediates are also present. E) The 2D classes obtained with FP4 suggest a 2 planar extended molecule with apparent C 4 symmetry. 3 Figure 3. Models of FP oligomers obtained by SAXS rigid body refinement. A-C) Left: SAXS data (blue) and 4 model fits (orange) corresponding to FP2, FP3 and FP4 structures obtained with C 1 symmetry. Inserts plot 5 the corresponding p(r) functions. Middle: Models fitted using C 1 symmetry. Right: Models fitted assuming 6 C n symmetry. The C 1 models were chosen as the most representative model in the largest cluster from the 7 10 generated, whereas the C n models were those amongst 10 or 30 models that according to their χ 2 value 8 were in best agreement with the experimental data, see panel F. In panel A, two orientations are displayed 9 to illustrate the curvature of FP2 models. D) Hypothetical model of FP2 from panel A bound to two C3 10 convertases to illustrate that such a complex could be formed on an activator to which the two C3b 11 molecules (green) carrying the Bb protease (orange) are bound through their Gln1013 (red sphere). The 12 two FP monomers are colored light blue and grey, respectively. To the left is presented a side view and to 13 the right a top view. The principal axes of FP2 are displayed as a grey Cartesian coordinate system. E) 14 Enlarged view of the area outlined with a grey box in panel D. The vertical dashed vector is parallel to the 15 smallest principal axis (labelled "z") of the FP2 molecule, the orange stick represents the vector in the plane 16 of the FP eye used for calculation of α, the angle between the two vectors. F) The average α values 17 calculated from the SAXS rigid body models. The number of models used for statistics from a pool of 10 or 18 30 models is indicated by the number n, remaining models were discarded as they had χ2 values that were 19 significantly higher than the n well fitting models. 20   HEPES,150 mM NaCl,pH 7.5 (black) or 100 mM glycine pH 2.3 (red), demonstrating that the two chains of FPhtΔ3 dissociate at low pH (top). Fractions from the SEC experiments were analyzed by SDS-PAGE and silver staining showing that at neutral pH the two fragments of FPhtΔ3 co-elute (bottom, left) whereas at low pH FPtHis (25 kDa) elute first and FPhΔ3 (18 kDa) elute last (bottom, right). B) Overview of FP constructs used in panels C-D. C) A silver-stained SDS-PAGE gel following the exchange of chains between FPc and FPhtΔ3. The exchange was monitored by the appearance of the FPh chain (25 kDa) from FPc and the disappearance of the FPhΔ3 chain, after pull-down on Ni-NTA beads using the His-tag on the C-terminus of FPtHis. D) A silver-stained gel from SDS-PAGE demonstrating the exchange of FP chains between FP2 and FPhtΔ3. The exchange was monitored by the appearance of the full length FP (50 kDa) after pull-down on Ni-NTA beads using the C-terminal His-tag of FPtHis from FPhtΔ3.