Assembly and Regulation of the Membrane Attack Complex Based on Structures of C5b6 and sC5b9

Activation of the complement system results in formation of membrane attack complexes (MACs), pores that disrupt lipid bilayers and lyse bacteria and other pathogens. Here, we present the crystal structure of the ﬁrst assembly intermediate, C5b6, together with a cryo-electron microscopy reconstruction of a soluble, regulated form of the pore, sC5b9. Cleavage of C5 to C5b results in marked conformational changes, distinct from those observed in the homologous C3-to-C3b transition. C6 captures this conformation, which is preserved in the larger sC5b9 assembly. Together with antibody labeling, these structures reveal that complement components associate through sideways alignment of the central MAC-perforin (MACPF) domains, resulting in a C5b6-C7-C8 b -C8 a -C9 arc. Soluble regulatory proteins below the arc indicate a potential dual mechanism in protection from pore formation. These results provide a structural framework for understanding MAC pore formation and regulation, processes important for ﬁghting infections and preventing complement-mediated tissue damage.


INTRODUCTION
Proteins of the terminal pathway of complement provide immune protection by forming lytic pores, membrane attack complexes (MACs), in membranes (Esser, 1994). Genetic deficiencies of MAC components lead to recurrent infections (Botto et al., 2009); however, unregulated MAC formation causes tissue damage (Morgan, 1999). Formation of the MAC is a sequential process. Upon complement activation, C5 is cleaved into C5a and C5b by the C5 convertase. C6 then captures a labile binding site in C5b (half-life: 2 min) (Cooper and Mü ller-Eberhard, 1970), followed by C7 association that renders the complex lipophilic (Preissner et al., 1985;Stewart et al., 1987). Binding of heterotrimeric C8abg defines the initial membrane insertion event, with C8b mediating attachment to the assembly precursor (Brannen and Sodetz, 2007;Stewart et al., 1987) and C8a penetrating the bilayer (Steckel et al., 1983). Inserted C5b8 functions as a receptor for C9 and catalyzes its oligomerization, leading to membrane perforation and target cell lysis (Podack et al., 1982;Tschopp, 1984;Tschopp et al., 1985). Off-target assembly of the MAC in solution leads to binding of clusterin and vitronectin, yielding a soluble complex called sC5b9 or sMAC. Dissociation of these chaperones by detergents reconstitutes membrane binding (Bhakdi et al., 1979;Podack and Mü ller-Eberhard, 1980). In addition, sC5b9 and the lytic MAC share a neo-epitope present in C9 that is associated with pore formation (Mollnes et al., 1985), suggesting similarities in how the soluble and membrane-associated complexes are assembled.
MAC proteins and the homologous perforin are proposed to form b-barrel pores on the basis of the structural resemblance between MAC-perforin (MACPF) domains and bacterial cholesterol-dependent cytolysins (CDCs) (Hadders et al., 2007;Law et al., 2010;Lovelace et al., 2011;Rosado et al., 2007;Rossjohn et al., 1997;Shatursky et al., 1999;Slade et al., 2008). Modeling and labeling of the perforin pore based on a cryo-EM reconstruction indicated, however, an inside-out arrangement of the perforin core relative to the proposed pore model for CDCs (Law et al., 2010). Though a recent crystal structure of the heterotrimeric C8abg suggested a CDC-like arrangement of the MAC ring (Lovelace et al., 2011), it lacked the context of the larger MAC assembly. To understand how complement proteins come together to form a lytic pore important for immune defense, we combined crystallographic analysis of C5b6 with electron microscopy (EM) to determine the structure of the sC5b9 complex.

Crystal Structure of C5b6
C5b6 was assembled in vitro, purified, and crystallized. Crystals diffracted anisotropically to a resolution between 3.5 and 4.2 Å , and the structure was solved by molecular replacement (Table  S1). The final model, consisting of all 12 domains of C5b and all 10 domains of C6, was refined to a final Rwork/Rfree of 25.6/27.0 and displayed good geometry ( Figures 1A-1C, Figure S1A, and Table S1). Because of the limited resolution, we restrict our discussion of the structure to the level of individual domains.
C5 undergoes marked domain rearrangements upon cleavage into C5b and formation of the C5b6 complex. C5 consists of two peptide chains, denoted b (residues [res.] 1-674) and a (res. 678-1676), that form 13 domains (Fredslund et al., 2008) ( Figure 1B). Similar to the conversion of homologous C3 to C3b (Janssen et al., 2006;Janssen et al., 2005), the domains of the a chain undergo major relocations, while most of the b-chain forms a stable ring-like structure ( Figure 1E, Table S2, Movie S1). Removal of the anaphylatoxin (ANA/C5a) domain results in extensive movements of the macroglobulin (MG) 3, MG7, MG8, and ''complement C1r/C1s, Uegf, Bmp1'' (CUB) domains and the thioester-like domain (TED/C5d); C5 lacks the prototypical thioester present in this domain in other members of the C3/ a2-macroglobulin protein family. Although the concerted movement of MG7 and CUB resembles that observed for the C3-C3b transition, the position of the connected TED differs dramatically ( Figures 1D and 1E, Table S2). In C5b6, TED is positioned halfway up the b-ring, in contrast to C3b, where TED lies at the ''bottom'' (Figures 1D and 1E). A requirement for this conformation to be caught and stabilized by C6 may explain the short halflife of the C6 binding site in C5b (Cooper and Mü ller-Eberhard, 1970). Without C6 binding, C5b will irreversibly decay to a state incapable of binding C6 (Cooper and Mü ller-Eberhard, 1970).
The structure of C6 in the C5b6 complex reveals that its ten domains can be divided into two functional parts. The first consists of the six N-terminal domains and includes the ''core'' region common to all MACPF proteins in the MAC; these six domains are thrombospondin (TSP) domain 1, TSP2, lowdensity lipoprotein receptor class A domain (LDLRa), MACPF, epidermal-growth factor (EGF) domain, and TSP3. The second, C-terminal part consists of two complement-control-protein (CCP) and two factor I/MAC (FIMAC) domains ( Figures 1A, 1C, and 1F). These regions are separated by a long flexible linker. A comparison with free C6 (Aleshin et al., 2012) reveals that whereas the N-terminal region is highly similar, the C-terminal region has a strikingly different arrangement (Figures 1A and 1F,Table S3) and forms the major interface with C5b.
The C5b-C6 interface buries ± 3100 Å 2 of solvent accessible surface area. The core of C6 binds to the ''bottom'' of C5b, in between MG1, MG4, and TED, to a highly conserved patch (Figure 2A). A second, major interface is formed by the linker and the CCP1-2 domains of C6 that wrap around the TED domain of C5b ( Figure 2A). The CCP1 domain is wedged in between TED, CUB, and MG2 of C5b, where it seems to stabilize the observed position of TED (Figures 1 and 2). The importance of this interaction is supported by data showing that the CCP domains are essential for C6 activity (DiScipio et al., 1999). The linker preceding CCP1 also makes extensive contacts with TED (Figures 2A and 2C), where it interacts in part with a distinct b-hairpin that forms a unique insertion in TED ( Figures 2C, 2D and S1B-S1D). We tested the importance of the linker region by mutating several conserved residues, all showing a reduced activity in a hemolytic assay ( Figure 2B).
The observed association of C6 to C5b positions the MACPF domain with its putative b-barrel forming transmembrane segments (TMS) at the periphery of the complex ( Figure 1A). The two TMS regions in C5b6 (C6 res. 255-312 and res. 381-438) adopt helical conformations loosely folded onto the central curved four-stranded b sheet of the C6 MACPF; a similar arrangement is observed in structures of free C6 (Aleshin et al., 2012), C8 (Hadders et al., 2007;Lovelace et al., 2011;Slade et al., 2008), perforin (Law et al., 2010), and CDCs (Rossjohn et al., 1997). Importantly, we do not observe a structural rearrangement of these regions compared to free C6, suggesting that binding to C5b is not the trigger for unfolding of these segments. Thus, we interpret the MACPF fold of C6 in C5b6 as a premembrane attack state, which is in agreement with the soluble nature of C5b6.

Cryo-EM of the Soluble Regulated Pore, sC5b9
To understand how the MAC proteins arrange to form a pore, we examined by EM sC5b9 purified from activated serum (Figure S2). Broadly, sC5b9 is a thin, square-shaped complex with a single protrusion at one corner, as suggested by two-dimensional (2D) images ( Figure 3, Figures S2A-S2E). Three-dimensional structures of both negatively stained and cryo-EM sC5b9 ( Figure S2F) further define two prominent features of the square-like central region. The protrusion ( Figure 3A, indicated by a brown arrow) connects to the core at one corner of an arc-shaped crescent, while large connected densities form a butterfly arrangement below the arc ( Figure 3A, indicated by a gray surface).
Docking of the C5b6 crystal structure into the 24 Å cryo-EM molecular envelope clearly defined the identity of the protrusion as C5b6. Strikingly, the labile conformation of C5b trapped by C6 in the C5b6 crystal structure is preserved in the larger complex ( Figures 1A and 3E). The orientations of the C-terminal FIMAC domains of C6 and the C345C domain of C5b, likely affected by crystal packing, did not fit the density of the sC5b9 reconstruction (Figures S1E-S1G). In addition, the C6 TSP1 domain, thought to regulate assembly of MAC precursors (Aleshin et al., 2012), was also out of density. Our map could accommodate these domains ( Figure 3E, indicated by asterisks); however, low resolution precluded modeling their orientations in the sC5b9 reconstruction.
Next, we generated a model of multiple MAC proteins that contain the conserved TSP-LDLRa-MACPF-EGF domain architecture ( Figure S3) by duplicating the MACPF-MACPF orientation of C8ab (Lovelace et al., 2011). Five MAC proteins fit the arc below the protrusion ( Figure 3E). Labeling with a monoclonal anti-C9 antibody, recognizing a neo-epitope present in both sC5b9 and the lytic pore, identified C9 as the MAC protein in the arc furthest from the C5b protrusion (Figures 3C-3E and S4). Previous biochemical data showed that C9 binds to C8a  (Slade et al., 2006) and that C8b binds C5b7 (Brannen and Sodetz, 2007;Stewart et al., 1987), indicating C8ba as the two MAC components preceding C9. Independent docking of the five-MAC model and C5b6 into the EM map superimposes the core of C6 onto the first MAC position ( Figure S3A), suggesting that C6 is the first MAC protein and C7 is in the remaining unoccupied position, the second position between C6 and C8b. Models involving a six-MAC protein arc correlated less well with the EM density (correlation coefficients of 0.88 and 0.84 for five-and six-membered arcs, respectively). It is noteworthy that models involving the C6-C7-C8b-C8a-C9 arrangement, in which the position of C6 was defined by docking C5b6 as a rigid body, correlated 7% better with the map than those ordered C7-C6-C8b-C8a-C9 as previously proposed (Aleshin et al., 2012).
Density in the center of the arc accommodates the lipocalin fold of C8g ( Figure 3E, indicated as a solid gray surface), a MAC component that is flexibly, but covalently, attached to C8a (Lovelace et al., 2011;Slade et al., 2008) (Figure S3C). C8g enhances lysis, but it is not essential (Parker and Sodetz, 2002) for MAC activity, and its orientation in sC5b9 suggests a role in stabilizing the MACPF-MACPF interactions before closure of the ring. Finally, density present at a ridge along C5b ( Figure 3E, indicated with a dashed orange line) could account in part for the unmodeled CCP1-2 and FIMAC1-2 domains of C7, known to interact with the C345C domain of C5b (Thai and Ogata, 2004).
Density of the butterfly-shaped region of the sC5b9 map can be attributed to regulatory proteins, vitronectin and clusterin, known to bind exposed lipophilic regions of MAC precursors. This is supported by previous EM analysis of gold-labeled vitronectin localizing oligomers to this region (Preissner et al., 1989). Moreover, the location of regulatory proteins is consistent with the positioning of predicted transmembrane segments below the MACPFs in the arc and the interpretation that these segments are either disordered or flapped out into their b-hairpin conformation (Figure 4). The structural data, therefore, indicate a potential dual mechanism in protection from pore formation: the two ''wings'' of the butterfly-shaped regulatory region cap the ends of the MAC arc and thereby block C9 oligomerization, and they enwrap the lipophilic segments to prevent membrane interaction.

DISCUSSION
Pore formation for MACPF-containing proteins involves a dramatic conformational change in which helical bundles transform into a membrane-spanning b-barrel. In contrast to the current model for immune pore formation, which is based on perforin (Law et al., 2010), our structural analysis of C5b6 and the sC5b9 complex supports a model for the MAC that resembles bacterial CDC pores (Figure 4). Docking of C5b6 into the complex, the curvature of the arc, and density for C8g inside the pore are all in concordance with the CDC orientation of MACPF proteins in the membrane. Despite its similarity to simpler bacterial pores, the complement-mediated immune response has evolved complex assembly and regulatory mechanisms that are likely required to prevent host tissue damage yet effectively clear pathogen infections. Our model suggests that the ability of C6 to capture a labile binding site in C5b to form an assembly competent state provides the first checkpoint in MAC formation. Next, C7 binds the C5b6 complex, making extensive contacts to C5b through its C-terminal CCP and FIMAC domains, thereby aligning MACPF domains of C6 and C7. Binding of C7 then drives rearrangement of the TMS regions, making C5b7 lipophilic and creating the novel hybrid-binding site for C8. Subsequent association of C8abg through alignment of the C8b and C8a MACPF domains relocates C8g, which in the soluble C8abg complex may serve to inhibit C9 association before incorporation into the larger assembly. Finally, host regulatory proteins clusterin and vitronectin can prevent pore formation by blocking both hairpin insertion into the membrane and oligomerisation of C9. Together, our data show how the MAC is assembled and regulated in blood, providing a framework for understanding the role of complement in microbial infection and inflammatory disorders.

Purification of C5b6
C5b6 was isolated from a mixture of purified human C5 and C6 in which the C5 was activated by addition of a mixture of cobra venom factor (CVF), factor B, and factor D in the presence of 0.5 mM MgCl 2 . C5 and C6 were purified from pooled normal human serum as previously described (Kolb et al., 1982;Tack et al., 1980). The formed C5b6 complex was separated from the other components by ion exchange chromatography over a Mono Q column (GE Healthcare) and subsequently by gel filtration over BioGel A0.5 m (BioRad). Fractions were pooled on the basis of C5b6 functional activity (Rawal and Pangburn, 2000), concentrated to 0.7 mg/ml and dialyzed against 10 mM HEPES (pH 7.2), 120 mM NaCl, and 0.02% w/v NaN 3 .

Crystallization and Data Collection
The C5b6 complex was crystallized by vapor diffusion in hanging drops consisting of 2.5 ml protein (0.7 mg/ml) mixed with 0.5 ml 1M HEPES-NaOH (pH 7.8). Drops were equilibrated at 18 C over 300 ml reservoir solution consisting of 0.1M HEPES-NaOH (pH 7.8) and 250 mM NaCl. Crystals grew to maximum dimensions of 800 3 80 3 20 mm in $3 weeks and were cryoprotected by brief incubation in reservoir solution supplemented with 30% (v/v) ethylene glycol, followed by flash freezing in liquid N 2 . A complete data set was collected at ESRF beamline ID29 in seven wedges of 15 that were collected along the length of the crystal through the use of 1 oscillations. The diffraction data were integrated and scaled by XDS (Kabsch, 2010) and Aimless (Collaborative Computational Project, Number 4, 1994). The crystals belong to space group  The diffraction data were strongly anisotropic, extending to 4.2 Å resolution in the direction of a*, 3.8 Å in the direction of b*, and 3.5 Å in the direction of c*. These resolution limits were determined by applying a cutoff based on either a mean intensity correlation coefficient of half-data sets > 0.5 or F/sF = 3. Both methods gave the same value for resolution cutoffs.

Structure Determination and Refinement
The structure of C5b6 was solved by molecular replacement through the use of PHASER (McCoy et al., 2007). A solution could be found by using the known structures of C5 (PDB code 3CU7; Fredslund et al., 2008) and C6 (PDB code 3T5O; Aleshin et al., 2012), while prior to the publication of the C6 structure a model of C6 had been generated starting from fragments of homologous structures. The C5 coordinates were separated into TED (res. 986-1305), the b-chain (res. 20-607), and the MG7 (res. 822-931), MG8 (res. 1374-1512), CUB (res. 932-985, 1308-1368), and C345C domains (res. 1530-1676), while the C6 coordinates were separated into the core (res. 22-629), the CCP domains (res. 641-765), and the FIMAC domains (res. 771-934). The model was completed by iterative model building in Coot (Emsley and Cowtan, 2004) and refinement in Phenix (Afonine et al., 2010) and autoBuster (Blanc et al., 2004). Although the data set used for refinement was strongly anisotropic, (4.2-3.5 Å ; see above), we used all data up to 3.5 Å for refinement, as anisotropic truncation did not improve refinement statistics or map quality. Initial refinement runs were heavily restrained and involved the use of both secondary structure and Ramachandran restraints as implemented by Phenix (Afonine et al., 2010). In later stages of refinement, the model from Phenix was used to generate LSSR-based target restraints as implemented by autoBuster (Blanc et al., 2004). The refinement strategy further included individual positional and B factor refinement and TLS refinement using 11 TLS groups. The model was refined to an Rfree of 27.0% and displays good geometry, with 88.9% of the residues in the allowed and 0.5% of the residues in the disallowed regions of the Ramachandran plot.
Expression and Purification of C6 C6 constructs were expressed as His6-tagged N-terminal fusions in transiently transfected suspension cultures of N-acetylglucoaminyltransferase-I-deficient HEK293E cells (Utrecht-ProteinExpress). Medium was harvested six days after transfection, then concentrated $10-fold and buffer-exchanged with the use of a 30 kDa cutoff filter (Quixstand hollow fiber; GE Healthcare). The proteins were purified by Ni-SepharoseTM 6 Fast Flow (GE Healthcare) and size exclusion chromatography with the use of a SuperdexTM 200 10/300 column equilibrated in 20 mM HEPES-NaOH (pH 7.4) and 150 mM NaCl. Fractions containing the C6 construct were pooled, concentrated, and flash frozen in liquid N 2 . All proteins were stored at À80 C until use.
Hemolytic Assay C6 was assayed for hemolytic activity with the use of antibody-sensitized sheep erythrocytes (EA) and C6-depleted human serum (Complement Technology), as described previously (Rawal and Pangburn, 2000). Hemolytic titers (ng C6 required for lyse of 50% of the EA) were determined, and the activities were normalized to recombinant wild-type C6. Each sample was tested at six different concentrations (n = 6). Standard errors were calculated with a nonlinear fitting program (GraFit 5.0, Erithacus software).

Purification of sC5b9
Blood was collected from healthy volunteers and allowed to clot, and serum was separated within 1 hr of collection. To activate complement by both classical and alternative routes, zymosan (10 mg/ml; Sigma) and heat-aggregated human IgG (1mg/ml; made in house) were added to the serum and incubated overnight at 37 C. Particulates were removed by centrifugation and filtration (0.2 mm). Serum was then applied on an affinity column (HiTrap; GE Healthcare) to which 40 mg mouse anti-human C8 monoclonal antibody E2 (generated in house) was coupled. Protein was eluted in 0.1 M glycine (pH 2.5) and neutralized by collection into 0.5 M Tris buffer (pH 8.0). sC5b9 containing fractions were pooled, concentrated, and applied to a preparative scale gel filtration column (Superdex-200 matrix in a XK16/70 column; GE Healthcare) equilibrated in PBS (137 mM NaCl, 10 mM phosphate, 2.7 mM KCl [pH 7.4]). Immediately before analysis, sC5b9 was further purified by two successive purifications on a Superose 6 PC 3.2/30 gel filtration column in a buffer containing 150 mM NaCl, 50 mM Tris (pH 7.5), 1 mM CaCl 2 , and 1 mM MgCl 2 . Antibody-labeled sC5b9 complexes were generated by incubating 5 mg of monoclonal antibody, aE11 (Hycult Biotech), with 7.5 mg sC5b9 for 20 min at

. A Model for MAC Formation
The complement terminal pathway is initiated by the cleavage of C5 to C5b. C6 traps a labile conformation of the C5b TED domain to form C5b6, a platform for the stepwise assembly of components C7, C8, and C9. Regulatory proteins in the plasma block MAC assembly in solution by binding exposed hydrophobic regions and sterically inhibit C9 oligomerization. Binding of C5b8 to membranes recruits multiple C9 molecules whose MACPF domains arrange to form a b-barrel pore similar to that of CDCs. room temperature. Excess antibody was removed by gel filtration as described above.
Negative Stain Electron Microscopy Immediately after gel filtration, a volume of 2.5 ml of sC5b9 (16 mg/ml) or antibody-labeled sC5b-9 (9 mg/ml), were applied to glow-discharged carboncoated copper grids. Grids were negatively stained with 0.75% uranyl formate. Images were taken under low-dose conditions ($10 e À /Å 2 per exposure) at a nominal magnification of 72,500 on a JEOL JEM-1230 operated at 100 kV. Images were recorded on a 4k 3 4k TemCam-F416 camera (TVIPS) and 2.28 Å /pixel. 11330 sC5b9 and 1104 immune-labeled windowed particles were each subjected to reference-free alignment with the use of EMAN2 (Tang et al., 2007) and classified into 149 and 31 classes, respectively. The standard EMAN2 initial-model-generation program (e2initialmodel.py) was used to obtain an initial template for refinement. With the use of this methodology, several models were constructed from a series of randomly generated Gaussian blobs, masked according to the sC5b9 particle diameter, and used to initiate the angular assignment of reference-free-generated 2D class averages. The resulting models were ranked on the basis of the agreement of the projection with the class average. The top choice was used as template for the refinement of negatively stained sC5b9 single particles with the use of EMAN2. Handedness of the map was determined on the basis of an 8% difference in correlation coefficient used for measuring the agreement of the C5b6 crystal structure with the reconstruction.

Cryo-Electron Microscopy
The identical preparation used in the negative stain EM experiments described above was also subjected to analysis by cryo-EM. Aliquots (4 ml) of purified sC5b9 (0.1 mg/ml) were applied to glow-discharged holey carbon grids (QUANTIFOIL R 1.2/1.3) and vitrified in liquid ethane with an FEI Vitrobot. Images were acquired on a JEM-200FS FEG electron microscope (JEOL) operating at 200kV, equipped with an in-column energy filter (OMEGA). Images were recorded with a defocus range of À3.0 to À6.0 microns underfocus and at a magnification of 54,400 on a 4k 3 4k ULTRASCAN 4000 SP CCD camera, corresponding to 2.76 Å /pixel. A total of 18,983 individual sC5b9 particles were windowed with e2boxer (EMAN2), corrected for the contrast transfer function (CTF) with Bsoft (Heymann and Belnap, 2007), and lowpass filtered to 11 Å . The sC5b9 negative stain reconstruction served as a reference for the refinement of cryo-EM data through the use of projectionmatching algorithms in Xmipp (Scheres et al., 2008). All fitting and rigid-body refinement of pseudo-atomic models into the cryo-EM envelope were performed in Chimera (Pettersen et al., 2004). Figures were generated withPymol (PyMOL Molecular Graphics System, version 1.3, Schrö dinger).

ACCESSION NUMBERS
Coordinates and structure factors for the crystal structure of C5b6 have been deposited in the Protein Data Bank under accession code 4A5W. The cryoelectron microscopy density map for sC5b9 has been deposited at the Electron Microscopy Data Bank under accession number EMD-1991.

SUPPLEMENTAL INFORMATION
Supplemental Information includes four figures, three tables, and one movie and can be found with this article online at doi:10.1016/j.celrep.2012.02.003.

LICENSING INFORMATION
This is an open-access article distributed under the terms of the Creative Commons Attribution 3.0 Unported License (CC-BY; http:// creativecommons.org/licenses/by/3.0/legalcode).