Methods

Recombinant C9 purification
Human C9 and mouse C9 protein were purified using similar methods (with minor variations). The human C9 gene (P02748) was cloned into pSectag2a (Thermo Fisher Scientific) for expression in mammalian Expi293 cells where the native secretion sequence was replaced with the Igκ leader sequence. Human C9 mutants (F262C, V405C and F262C/V405C) were cloned using QuikChange. The mouse C9 (P06683) sequence was synthesised and cloned into pcDNA3.1 vector (GeneScript) also containing an Igκ leader sequence. Recombinant protein was produced by transient expression in Expi293F cells (Thermo Fisher Scientific) for four days according to the manufacturer’s instructions. The oligonucleotide primers used for cloning can be found in Supplementary Table 3.
The purification methods were essentially the same as previous one13,23 with some exceptions. Following centrifugation, the Expi293 media containing C9 was diluted with an equal volume of 10 mM sodium phosphate pH 7.4, 20 mM NaCl containing cOmplete protease inhibitor tablet (Roche). Then, it was loaded onto an equilibrated, HiTrap DEAE sepharose column (1 mL resin per 100 mL media). Chromatography steps were performed on an ÄKTA FPLC. The protein was eluted from the DEAE column using a linear gradient over 20 column volumes (from 10 mM sodium phosphate, 45 mM NaCl, pH 7.4 to 10 mM sodium phosphate, 500 mM NaCl, pH 7.4). Pooled fractions containing C9 were further purified using hydroxyapatite specifically using a pre-packed Bio-Rad type I CHT column equilibrated in 10 mM sodium phosphate pH 7.0, 100 mM NaCl. The CHT elution was performed over a six column volumes phosphate gradient at pH 8.1 (from 45 mM to 350 mM). Pooled fractions were concentrated using a 30 kDa MWCO concentrator (Amicon) and further purified using a size exclusion column; prepacked Superdex S200 16 mm × 60 mm or 26 mm × 60 mm (GE Healthcare life sciences). Size exclusion chromatography for human C9 was performed in 10 mM HEPES pH 7.2, 200 mM NaCl whereas mouse C9 was purified in 10 mM HEPES pH 7.2, 100 mM NaCl.

Murine C9 crystallisation
The murine C9 gene was synthesised (GenScript) with three mutations (N28E; N243D and N397D) in order to produce a non-N-linked glycosylated protein for crystallisation trials. The purified C9 was consistently observed to have similar activity to human C9 in a haemolytic assay using sheep erythrocyte/antibody/complement 1–8 (EAC 1–8, human C9-depleted serum; ComplementTech). Recombinant murine C9 was purified as described above, and the C9 protein sample was concentrated to ~9 mg mL−1 (~150 µM) for crystallisation trials. Optimised crystals were obtained using the hanging drop vapour diffusion method with a reservoir liquor containing 18% (w/v) PEG 3350, 0.2 M disodium malonate pH 7.5 and 10 mM ZnCl2 using the micro-seeding method. Crystals were flash cooled in liquid N2 with 25% (v/v) glycerol as a cryoprotectant. Data collection was performed at the Australian Synchrotron MX2 Beamline. Experimental phasing of C9 crystals was performed by soaking crystals in tantalum bromide (Jena Biosciences) and uranyl formate (Polysciences, Inc) for two days prior to harvesting.

X-ray data collection and model building
The data were merged and processed using XDS24,25, POINTLESS26,27 and AIMLESS28. Five percent of the datasets were flagged as a validation set for calculation of the Rfree with neither a sigma, nor a low-resolution cut-off applied to the data. Experimental phases (Supplementary Table 4) were obtained by the MIRAS (multiple isomorphous replacement plus anomalous differences) method. Molecular replacement was attempted using the MACPF domain of C6 (PDB ID 3T5O) and with both MACPF domains of C8 (PDB ID 3OJY). None of the MR experiments were successful. The Ta and U heavy atoms were not ordered and the structure was phased using the anomalous signal of the Zn and Ca ions bound to the protein. Two datasets collected at 10,300 eV (which is above both the K-edge of Zn and L-III edge of Ta), were used as derivative 1 and derivative 2 datasets (Supplementary Table 4). Experimental phasing strategies and dataset combinations were evaluated using HKL2MAP29 and final phasing was carried out using the CRANK2 pipeline30; heavy atom positions were located using SHELXC/SHELXD31, substructure refinement was done using BP332. The initial FOM (figure of merit) from phasing was 0.26 and after density modification with PARROT33 this increased to 0.57. Automated model building was carried out using BUCCANEER34 with the initial model consisting of 944 residues with R/Rfree of 34.1/40.0%. Two molecules were found per asymmetric unit. Model building was performed using COOT35 while refinement was performed using PHENIX36, REFMAC37, and autoBUSTER38. Water molecules were added to the model when the Rfree reached 30%. Crystallographic and structural analysis was performed using CCP4 suite39 unless otherwise specified. All Zn atoms were modelled into the omit-map generated using ANODE40 from a dataset collected at 9674.0 eV (1.28162 Å), above the K-edge of Zn (9659.0 eV), and confirmed by the absence of anomalous signal at the Zn sites in a dataset collected below the Zn K-edge at 9643.9 eV (1.28562 Å) (Supplementary Table 5). Figures 1a, b, and 2a; Supplementary Fig. 3; Supplementary Fig. 4 were generated in part using PYMOL and Chimera41. The final model contains two chains: chain A is less flexible with residues 18–226, 248–365, 395–526 modelled into the electron density; chain B residues 18–73, 78–113, 116–205, 214–225, 249–364, and 395–526 modelled. In the final model, the number of residues in the Ramachandran favoured region is 873 residues (out of a total of 874 residues). Structural validation was performed using MolProbity42. The MolProbity score is 0.87 which is in the 100th percentile of structures reported at this resolution.

PolyC9 preparation and data collection
Mammalian cell expressed human C9 (with the two native N-glycans) was buffer exchanged by dialysis into 10 mM HEPES pH 7.5, 50 mM NaCl overnight at 4 °C at a concentration between 100 and 250 µg mL−1. Following dialysis, the human C9 was concentrated between 1.1–1.5 mg mL−1 with a 30 kDa MWCO (Amicon) protein spin filter and 1:9 (v/v) of amphipol A8-35 (Anatrace) was added to a final concentration of 0.015–0.02 mg mL−1. Polymerisation reactions were initiated by incubating at 37 °C overnight and stored at 4 °C.
The polyC9 reaction producing the best grid was from an initial C9 concentration of 1.3 mg mL−1 containing 0.02 mg mL−1 A8-35. Plunge-freezing was performed using a Vitrobot Mark IV (FEI/Thermo Fisher Scientific). PolyC9 (2.5 μL) was added to a freshly glow discharged Quantifoil copper grid (R1.2/1.3, 200 mesh). Data was collected on a Titan Krios (FEI/Thermo Fisher Scientific)) operated at 300 kV at a magnification of 130 K in microprobe EFTEM mode, resulting in a magnified pixel size of 1.06 Å pixel−1. The movies were collected using a Gatan K2 Summit with a quantum energy filter in super resolution mode (for an effective pixel size of 0.53 Å pixel−1). Each movie consists of 20 sub frames and the exposure time was 8 s which amounted to a total dose of 46.4 e− Å−2 at a dose rate of 6 e− Å−2 s−1.

Cryo-EM data processing
Unless stated otherwise all processing was performed with RELION (v2.1b.1)43. Movies were down sampled in Fourier space by a factor of 2 and summed after correction of beam-induced motion by MotionCor244. CTF estimation was performed by CTFFIND4.145 and micrographs with ice contamination were discarded by visual inspection of the power spectra. Initially ~1000 particles were manually picked and subjected to reference-free 2D classification to serve as templates for auto picking. A total of ~220,000 particles were extracted from summed micrographs and subjected to multiple rounds of 2D classification. A representative subset of class averages was selected for initial model generation in EMAN2.246 using the common line method. The initial model was low pass filtered to 20 Å and particles were subjected to 3D classification, giving rise to two classes of C22 and C21 symmetry. Initial refinement with C22 symmetry led to a 4.2 Å map. These initial refinements were used to create a solvent mask, which was low pass filtered to 15 Å for subsequent refinements. This final subset of 58,000 particles was selected for masked movie refinement and particle polishing with C22 symmetry, where the MTF of the detector was used to determine a b-factor of −180 Å2. High resolution features were enhanced by sharpening with this b-factor for the purposes of map visualisation. The global resolution was estimated by the Gold Standard 0.143 criterion when comparing the Fourier shell correlation between two independent half maps47. The local variation of resolution was further analysed using blocres using a search box size of 20 voxels and FSC criterion of 0.548.

PolyC9 model building and model validation
Model building of the polyC9 was performed in COOT (0.8.8)49. The crystal structure of murine C9 was manually positioned into the best density of the cryo-EM map and rigid body fitting of individual domains was performed. The TMH1 and TMH2 regions, which significantly alter conformation, were removed for manual building. Non-conserved amino acids were mutated from murine to human residues and their side chains manually positioned to maximise fit in the map. Following initial model building, C22 symmetry was applied to the single subunit of polyC9 using Chimera (UCSF, USA)50 and further real space refinement performed in COOT to minimise clashes between subunits and improve the overall geometry.
The final three-dimensional model of polyC9 was refined into the cryo-EM map using the phenix.real_space_refine programme within PHENIX suite to optimise and correct for poor geometry (Supplementary Table 6)36. During the refinement, standard restrains for covalent geometry, Ramachandran plot and internal molecular (NCS) symmetry were imposed. In addition, secondary-structure restrains were defined for the β-barrel region of the pore (β-strands 186–216, 251–281, 333–363, 379–409) because the map quality towards the end of the pore is of lower resolution. Protein Interactions Calculator51 was used to calculate intermolecular contacts between adjacent molecules of polyC9 (Supplementary Table 1).

Haemolytic assay
Turbidity measurements were performed using sheep EAC1-8 prepared in DHB++ pH 7.4 (Dextrose HEPES Buffer; containing 2.5% (w/v) d-glucose, 5 mM HEPES, 71 mM NaCl, 0.15 mM CaCl2, 0.5 mM MgCl2). EAC1–8 was produced by sensitising 6.5 × 108 cells mL−1 sRBC with equal volume of anti-sheep antibody (0.75 mg mL−1) (Rockland immunochemicals cat no. C220–0002) at 30 °C. Sensitised cells were washed 2 min at 3220×g by centrifugation, and then C9-depleted serum (Complement Tech) added in batch with 1 µL per 3.75 × 106 cells and incubated at 37 °C for 30 min. The absorbance at 620 nm was continuously measured while incubating at 37 °C with intermittent orbital mixing. For unlocking experiments three independently prepared dilutions of C9 were combined with EAC1–8 (3.75 × 106 cells) and a final concentration of 1 mM DTT in a 96-well plate. Competition assays were prepared by combining different ratios of the TMH1 locked C9 (F262C/V405C, [C9mutant]) with a constant amount of wild-type C9 (final concentration 270 ng mL−1) prior to addition of EAC1-8 or BSA which was used as a non-specific binding control. Data are reported as the raw turbidity curves with error reported as the standard error of the mean.

Data availability
Data supporting the findings of this manuscript are available from the corresponding authors upon reasonable request. The datasets generated during the current study are available in the RCSB repository (PDB ID 6CXO) and (PDB ID 6DLW) and the EMDB repository (EMDB ID 7773).