Structural Basis for Eculizumab-Mediated Inhibition of the Complement Terminal Pathway Free

This article is featured in In This Issue, p.1

Complement is an ancient part of innate immunity that functions to detect and clear pathogen invasion and altered self-cells. The complement system is activated upon detection of pathogen-associated molecular patterns or danger-associated molecular patterns, which in turn will lead to the formation of the C5 convertases C4b2a3b (1) and C3bBb3b (2, 3), able to cleave C5 and initiate the terminal pathway (Fig. 1A, 1B). Upon C5 cleavage by the C5 convertases, two fragments are formed. The smaller fragment, C5a, functions as a potent proinflammatory anaphylatoxin by signaling through C5aR1 triggering increased vascular permeability on endothelial cells, chemotaxis and oxidative burst on phagocytes, release of proinflammatory molecules, and activation of the adaptive immune system [reviewed by Klos et al. (4)]. The larger fragment C5b can associate with C6, C7, C8, and 12–18 copies of C9 to form the membrane attack complex (MAC), also known as C5b-9. Upon assembly, the MAC inserts into membranes of pathogens without a cell wall, resulting in a lytic pore (5, 6). Host cells are normally protected from MAC lysis by the membrane-associated GPI-anchored regulators CD55 and CD59 and by the soluble regulators vitronectin and clusterin (7–11).

The humanized Ab eculizumab is the first and only therapeutic inhibitor of terminal complement. Eculizumab was approved by the U.S. Food and Drug Administration and European Medicines Agency for the treatment of paroxysmal nocturnal hemoglobinuria (PNH) and atypical hemolytic uremic syndrome (aHUS) in 2007 and 2011, respectively. PNH is a rare form of hemolytic anemia caused by somatic mutations of the PIG-A gene in hematopoietic stem cell lineages (12, 13). This causes a loss of GPI anchors for CD55 and CD59 (12) in the mature erythrocytes bearing the mutation, making them susceptible to lysis by as little as a single MAC. aHUS is a rare disease of complement dysregulation leading to thrombotic microangiopathy and severe kidney damage. Risk factors for aHUS include mutations in factor H, factor I, factor B, C3, and membrane cofactor protein [reviewed in Noris and Remuzzi (14)], as well as infectious environmental stimuli and autoantibodies to complement proteins. The pathological conditions of aHUS are caused by the insertion of MAC into the renal capillary endothelial cells and recruitment of proinflammatory cells by C5a. Eculizumab binds to human C5 with high affinity and prevents its cleavage by complement convertases into C5a and C5b (15), thereby providing a mechanism for the complete inhibition of terminal complement activity in both PNH and aHUS.

Eculizumab was developed from the murine Ab 5G1.1, which was mapped to bind to the N-terminal region of the α-chain of human C5. Moreover, it has been shown to bind C5-derived peptide fragments containing the KSSKC motif (residues 879–883) within the C5 macroglobulin (MG) 7 domain (16), distal to the scissile bond (R751-L752) that is cleaved by the convertase. The recent identification of Arg⁸⁸⁵His/Cys polymorphisms in a small population of eculizumab-resistant patients with PNH (17) provided further evidence that the Ab binds in this region. Eculizumab is also highly specific for human C5, with little to no antagonist activity toward C5 from any other primate species tested (D. Sheridan, unpublished observations).

In order to comprehend in structural detail how eculizumab prevents convertase cleavage of its target, we prepared a Fab molecule with the same sequence as eculizumab in the V region and determined the structure of the complex between this Fab and complement C5. Despite the limited resolution of 4.2 Å, our structure rationalizes existing functional data and genetic data regarding the C5–eculizumab interaction. In addition, we present a complete mutational scan of the eculizumab CDRs and a structure-based interpretation of the observed functional defects induced by individual point mutations.

C5 was purified as described (18) with modifications. Following elution from the DEAE column, Na₂SO₄ was added to C3/C5 fractions to a concentration of 400 mM and loaded on a 75-ml Phenyl Sepharose FF column (GE Healthcare) equilibrated in Buffer A (20 mM HEPES [pH 7.5] and 400 mM Na₂SO₄). The column was washed with 250 ml buffer A followed by 150 ml 50% buffer B (20 mM HEPES [pH 7.5] and 150 mM NaCl) and eluted with a 150-ml linear gradient from 50–100% buffer B. C5 fractions were adjusted with solid Na₂SO₄ to 400 mM and loaded on a 9-ml SOURCE 15PHE column (GE Healthcare) equilibrated in buffer A. The column was washed in buffer A and eluted with a 120-ml gradient from 0–100% buffer C (20 mM HEPES [pH 7.5] and 50 mM NaCl). Relevant fractions were pooled, concentrated, and subjected to size-exclusion chromatography (SEC) on a 16/600 Superdex 200 column (GE Healthcare) in buffer D (20 mM HEPES [pH 7.5] and 75 mM NaCl). SEC-purified C5 was finally loaded on a 1-ml MonoQ (GE Healthcare) in buffer D and eluted with a 20-ml linear gradient from 75–500 mM NaCl.

A dual CMV expression plasmid encoding an L-chain and a Fab fragment of the H-chain with the same primary sequence as eculizumab (BNJ416) was expressed transiently in Expi293F cells according to the manufacturer’s recommendations (Invitrogen, Grand Island, NY). The Fab fragment was purified from the cell culture supernatant using a κ select resin (GE Healthcare) equilibrated with PBS (pH 7.4) and eluted with 100 mM glycine (pH 2.7). Free L-chain was removed from the κ select pool using cation exchange with SP Sepharose (GE Healthcare). The cation exchange column was equilibrated in Buffer E (50 mM sodium acetate pH 4.8), and eluted with a linear salt gradient of 0–30% Buffer F (50 mM sodium acetate [pH 4.8] and 1 M NaCl). Following purification, the Fab was determined to be >99% pure using both size-exclusion chromatography (SEC-HPLC) and reverse-phase chromatography (reverse-phase HPLC). SEC-HPLC was performed using 10 μg purified sample on a TSKgel GW3000 column (Tosoh) using an isocratic mobile phase consisting of 150 mM NaCl and 10 mM NaPO₄ (pH 7). Absorbance was monitored at 214 nm. Reverse-phase HPLC was performed using 10 μg purified sample on a Vydac C4 column (Grace) with a mobile phase of 64% Buffer G (5% acetonitrile, 0.1% trifluoroacetic acid in water) and 36% Buffer H (80% acetonitrile, 0.1% trifluoroacetic acid in water) and monitoring absorbance at 280 nm.

To isolate the complex, C5 was mixed with the BNJ416 Fab in a 1:1.2 molar ratio. The complex was purified on a 10/300 GL Superdex 200 column (GE Healthcare) equilibrated in buffer D. Complex containing fractions were pooled and concentrated to 8.5 mg/ml.

Initial crystallization conditions obtained with commercial screens from Hampton Research and Molecular Dimensions were optimized with MIMER (19). The best crystals were obtained by mixing the complex at 5–7 mg/ml 1:1 with a reservoir solution consisting of 0.1 M imidazole (pH 6.2), 4% v/v Tacsimate, and 7.8–8% w/v PEG 3350 in sitting drops, which were equilibrated for 2–7 d at 4°C prior to streak seeding. Crystals were cryoprotected in 0.1 M imidazole (pH 6.2), 4% v/v Tacsimate, 21.5% w/v PEG 3350, and 8.5% v/v glycerol and frozen in liquid nitrogen. Data were collected at the European Synchrotron Radiation Facility (ESRF) ID29 at 100 K and processed with XDS (20). The structure was solved by molecular replacement with phenix.phaser (21). The model was improved by rebuilding using the programs ‘O’ (22) and Coot (23) and refined with phenix.refine (24) using noncrystallographic restraints in combination with rigid body, simulated annealing, individual atomic displacement parameters, and translation/libration/screw refinement. Molprobity (25) was used for validation. Figures were prepared with PyMol (http://www.pymol.org). Coordinates and structure factors are available at the Research Collaboratory for Structural Bioinformatics Protein Data Bank (http://www.rcsb.org) with accession code 5I5K.

To screen for contact residues in eculizumab important for binding to C5, each amino acid in all six CDRs was subjected independently to substitution with histidine and screened for altered binding kinetics to C5. Briefly, full-length IgGs with the same primary sequence as eculizumab containing single amino acid substitutions were expressed transiently in Expi293F cells by cotransfecting 1:1 ratios of CMV expression plasmids encoding the Ab L- and H-chains. Ab expression levels were quantified by biolayer interferometry (BLI) on protein A sensors (catalog number 18-5010; ForteBio) mounted on an Octet QK (ForteBio) and normalized to 2.4 μg/ml in 1× kinetic buffer (0.01% BSA and 0.002% [v/v] Tween 20 in PBS buffer [pH 7.4]). The Abs were then immobilized on protein A sensors and exposed to C5 at 26 nM to determine binding kinetics on an Octet Red with association in 1× kinetic buffer (pH 7.4) and dissociation in 1× kinetic buffer (pH 7.4 or 6).

Binding kinetics on a small subset of representative Abs with single amino acid substitutions were assayed using surface plasmon resonance (SPR) on a BIAcore 3000 instrument (GE Healthcare). The experiments were performed using an anti-Fc human capture method at pH 7.4. Anti–Fc-Human (KPL #01-10-20) diluted to 0.1 mg/ml in 10 mM sodium acetate (pH 5) was immobilized on two flow cells of a CM5 chip for 8 min by amine coupling. The parental Ab with the same primary sequence as eculizumab (EHL000) and single histidine substitution variants were diluted to 0.5 μg/ml in running buffer (HBS-EP [pH 7.4]) and injected onto one flow cell to achieve a capture level of 50 RUs on the immobilized anti-Fc human surface. The second flow cell was used as a reference surface. C5 in a range of 8–10 concentrations per Ab sample were injected over both flow cells to determine kinetics. The surface was regenerated each cycle with 20 mM HCl and 0.01% P20 (200 μl injection at 100 μl/min). The data were processed with a 1:1 Langmuir model using BIAevaluation 4.1 software (Biacore) with double referencing subtraction of both the reference flow cell and a 0-nM C5 (blank) cycle.

The same subset of mutated Abs analyzed by SPR were assayed for their potential to inhibit terminal complement activity in a classical pathway chicken RBC (cRBC) hemolysis assay in normal human serum (NHS). The EHL000 Ab was used as a positive control. Ab-sensitized cRBCs were prepared for each assay from 400 μl chicken whole blood in Alsever’s buffer (Lampire Biologicals), washed four times with 1 ml gelatin veronal-buffered saline plus calcium and magnesium (GVB++; Complement Technology) at 4°C, and resuspended in GVB++ at 5 × 10⁷ cells/ml. To sensitize chicken erythrocytes, a polyclonal anti-chicken RBC Ab (Rockland Immunochemicals) was added to the cells (150 μg/ml) and incubated for 15 min on ice. After washing with GVB++ once, the cells were resuspended in GVB++ to a final volume of 3.6 ml.

Abs were serially diluted in neat NHS at a ratio of 1:2 to final concentrations ranging from 300 μg/ml to 2.343 μg/ml, and then 20 μl of each Ab/serum sample was diluted in 80 μl GVBS and incubated at room temperature for 20 min. Sensitized cRBCs were added to the Ab/serum mixture at 30 μl/well (2.5 × 10⁶ cells) and incubated at 37°C for 30 min. The plates were centrifuged at 1800 × g for 3 min, and 75 μl supernatant was transferred to a new flat-bottom 96-well plate. The absorbance was measured at 415 nm. Samples containing serum without anti-C5 Abs with or without 10 mM EDTA were used as no lysis or complete lysis controls, respectively. Sample conditions were run in triplicate, and each sample was tested twice in the same assay.

In order to obtain structural insight into how eculizumab binds C5 and prevents its activation by complement convertases, we generated a Fab fragment with the same sequence as found in eculizumab and crystallized the Fab–C5 complex. After screening >200 crystals, we obtained x-ray diffraction data extending to a maximum resolution of 4.2 Å (Table I). We found two copies of the complex in the asymmetric unit that turned out to have virtually identical structures. The electron density of the MG1 and MG3 and especially the MG4 and MG5 domains and the C-terminal C345c domain of C5 are better in one Fab–C5 complex compared with the second complex, but comparable at the interface between C5 and the Fab (Supplemental Fig. 1). The two complexes can be superimposed with a root mean square deviation (rmsd) of 1.24 Å for 1669 Cα atoms, and at the Fab–C5 interface, the two complexes are even more similar, as 272 Cα atoms from the C5 and the Fab domains forming the interface overlay with an rmsd of 0.35 Å. In the following, we describe the complex containing chain B (C5) and chains H and L (Fab) (Fig. 1C). Due to the limited resolution of the data, fine details concerning intermolecular interactions cannot be derived, but the structure unambiguously reveals the regions of C5 and the Fab engaged in contacts. We consider the main chain tracing at the intermolecular interface to be reliable, whereas the direction of the side chains may differ in a high-resolution structure (Fig. 1D, Supplemental Fig. 2A). In addition, we cannot account for water-mediated interactions. Thus, intermolecular interactions described in the following should be considered putative.

The complex is highly nonglobular, with the Fab fragment extending in an almost orthogonal manner from the MG7 domain as compared with the major axis of C5 (Fig. 1C). The conformation of C5 is quite similar to that of C5 in complex with cobra venom factor (CVF), with the C-terminal C345c domain stably associated with the C5 MG7 and MG8 domains. Comparison of the Fab-bound C5 with CVF-bound (26) and unbound C5 (27) indicate rotations of the CUB and C5d domains relative to the MG1–MG6 domains of 4° for Fab-bound C5 relative to CVF-bound and 11° relative to unbound C5. Compared to our previous C5-containing structures, we observe a change in the scissile bond region (26, 28), the linker between the MG6 and MG7 domains, and we have also corrected a small register error in the first strand of the CUB domain. The same features are observed in the C5–Rhipicephalus appendiculatus C5 inhibitor (RaCI)–Ornithodoros moubata complement inhibitor (OmCI) complex (29). Hence, these changes are unlikely to be caused by Fab binding. At the Fab–C5 interface (Fig. 1C), the variable domains of both H- and L-chain exclusively interact with the MG7 domain of C5. Although caution must be taken due to the low resolution, analysis with the program PISA (30) suggests that ∼1800 Ų is buried in the interface, with the H- and L-chain accounting for 65 and 35% of the binding surface area, respectively.

The C5 residues recognized by the Fab are all part of the antiparallel four-stranded β-sheet in the MG7 domain. Residues 851–858 and 882–888 are found in two antiparallel β-strands held together by a disulfide bridge, whereas a third region (residue 915–920) connects two β-strands in a hairpin loop (Fig. 2A, 2B). Except for L-chain CDR2, all CDR regions are engaged in interaction with C5, with the most prominent interactions being formed by H-CDR3 (Fig. 2C–F). Residues 101–107 of H-CDR3 together with residues 30–32 of L-CDR1 and Leu⁹² in L-CDR3 form a binding pocket for C5 Trp⁹¹⁷ and Phe⁹¹⁸ (Fig. 2D). This explains the exquisite specificity of eculizumab for human C5, because a Trp⁹¹⁷Ser substitution is found in most other species (Fig. 2B, Supplemental Fig. 3). H-CDR3 is also engaged in forming the binding pocket for C5 Arg⁸⁸⁵ (Fig. 2E). Our structure suggests that the resistance to eculizumab in patients with PNH with Arg⁸⁸⁵His/Cys polymorphisms (17) can be explained by a disruption of the C5–eculizumab interface due to the histidine/cysteine being too small to fill the arginine-binding pocket. This most likely causes the extended conformation of H-CDR3 to change significantly with a concomitant loss of affinity. Moreover, as Arg⁸⁸⁵ is spatially close to the disulfide bridge C856–C883 (Fig. 2A), the Arg⁸⁸⁵Cys polymorphism could result in an altered disulfide bridge pattern, potentially perturbing the integrity of MG7.

In an orthogonal evaluation of the contribution of CDR residues to the C5 binding, we created 66 single-point mutants in which each CDR residue had been mutated to histidine, allowing us to selectively study the effect of a positive (pH 6) or a neutral side chain (pH 7.4) mutation on the dissociation of the C5–Ab complex. A total of 28 of the 66 point mutants showed a significant effect on association, dissociation, or both (Fig. 3A–C). All of the important mutations increased dissociation at pH 6, whereas 60% of them also had an effect on the dissociation at pH 7.4 (dissociation of Asp¹¹⁰ from H-CDR3 could not be assessed as its association with C5 was highly compromised). Precise determination of the binding kinetics was determined for a subset of the most important mutant Abs using SPR, and the results are in agreement with the findings from the BLI assays (Fig. 3D, Table II). EHL000 bound to C5 with an affinity of 17.6 pM. This affinity decreased up to 7700-fold for the tested mutant Abs ranging from 0.443 nM for Thr⁹⁴His of L-CDR3 to 136 nM for Trp¹⁰⁷His of H-CDR3 (Table II). Both the association and dissociation rate constants determined by SPR were in good agreement with the more qualitative BLI data, and we conclude that increases in the dissociation rates are the primary contributor to the loss of affinity rather than a decrease in association rates (Fig. 3A, 3B, 3D, Table II). It is worth noting that the mutants seem to fall into two classes as observed from the SPR curves. The first class behaves somewhat similar to the wild-type Ab and includes the Glu⁵⁹His mutant of H-CDR2 and the Thr⁹⁴His mutant of L-CDR3. These Abs have slower on-rates and markedly slower off-rates as compared with the second group, including the Trp³³His mutant of H-CDR1, the Phe¹⁰¹His and Trp¹⁰⁷His mutants of H-CDR3, and the Ala³²His mutant of L-CDR1. For the same subset of Ab mutants tested using SPR, the impact of loss of binding affinity on the functional inhibition of terminal complement was assessed in a chicken RBC hemolysis assay at pH 7.4 (Fig. 3E). These results clearly show that the effects on inhibitory activity are in agreement with the effect each mutation has on binding to C5, where mutations that significantly affect binding (H-chain Trp¹⁰⁷His, Phe¹⁰¹His, Trp³³His, and L-chain Ala³²His) severely limit the ability to inhibit hemolysis. Mutations conferring smaller effects on binding kinetics (H-chain Glu⁵⁹His and L-chain Thr⁹⁴His) also have a less pronounced impact on function in the hemolysis assay.

In order to rationalize the effect of these mutations, we inspected the surroundings of the affected residues in our structure (Fig. 3C, Supplemental Table I). Trp³³ (H-CDR1), Tyr⁹⁹ (H-CDR3), and Trp¹⁰⁷ (H-CDR3) are prominent examples of affected residues in direct contact with C5. We predict that mutation of these residues to histidine causes a disruption of the binding pocket mentioned above accommodating Arg⁸⁸⁵ of C5 (Fig. 2E). Furthermore, a positive charge of these residues at pH 6 might result in electrostatic repulsion of the arginine. Histidine substitution of the noninteracting Pro⁹⁵ (L-CDR3) might cause a conformational change in the main chain, perturbing the C5-interacting residues Thr⁹⁴ and Leu⁹⁶ (L-CDR3) and the neighboring Glu⁵⁹ (H-CDR2), and a positive charge at the proline position might disrupt the potential electrostatic interaction between Glu⁵⁹ (H-CDR2) and C5 Lys⁸⁸⁷ (Fig. 2F). The same effect might occur upon Thr⁹⁴His (L-CDR3) substitution. A second residue not in direct contact with C5 is Glu⁵⁰ (H-CDR2); however, Glu⁵⁰ appears to electrostatically stabilize Arg⁸⁸⁵ of C5, suggesting that a positively charged histidine substitution might cause repulsion of the C5 arginine (Fig. 2E). Substitution of Asp¹¹⁰ from H-CDR3 severely impacted the Ab’s ability to associate with C5; however, this residue is at least 10 Å from the nearest C5 atom. We suggest that the effect of mutating Asp¹¹⁰ is indirect and caused by destabilization of Arg⁹⁸ at the N-terminal end of H-CDR3, which, according to our structure, is packed between Tyr²⁷ and Tyr³² from H-CDR1 (Supplemental Fig. 2B). This is proposed to affect the conformation of H-CDR3 and possibly also H-CDR1, causing an effect upon association with C5. Overall, the effects of the mutations are in excellent agreement with the Fab–C5 interface observed in our crystal structure. Even for residues not involved in direct contacts with C5, our structure still provides plausible explanations for the observed effects (Supplemental Table I). In summary, our structure and the comparison with the C5–CVF complex (Fig. 4A) suggests that eculizumab functions by exerting steric hindrance for C5 binding to the noncatalytic subunit of the C5 convertase.

Eculizumab is known to function by preventing cleavage of C5 by the complement C5 convertases; however, in our structure, no residue in the Fab comes closer than 30 Å to the scissile bond region (Fig. 1C). This suggests that eculizumab functions by sterically hindering the convertase from associating with C5 but not by directly preventing access to the scissile bond. The primary noncatalytic subunit of the C5 convertases (C3b in the alternative pathway and C4b in the classical pathway) are believed to interact with their substrates through a two-point interaction similar to that observed in the C5–CVF complex (26), where CVF serves as a model for the noncatalytic subunit of the C5 convertase (Fig. 1B). The MG4 and MG5 domains in C5 may interact with the same domains in C3b/C4b, whereas the C5 MG7 domain is likely to contact the MG6 and MG7 domains of C3b/C4b (Fig. 4A). A comparison of the interaction sites for CVF and the Fab on the C5 MG7 domain reveals significant overlap of the two binding sites (Fig. 4B, 4C). This further strengthens the idea that eculizumab prevents the interaction of C5 with the convertase by directly competing for binding to the MG7 domain and possibly indirectly by averting the MG4 and MG5 recognition by the convertase. It has very recently been suggested that the role of the additional/nearby C3b in the C5 convertase as compared with the C3 convertase is to prime the conformation of C5 such that it can bind a nearby C3 convertase rather than inducing a particular conformation in the C5 convertase with high affinity for the substrate (29). Both models are equally compatible with our suggestion that steric hindrance is the predominant mechanism through which eculizumab inhibits C5 cleavage.

This mechanism of inhibition based on steric hindrance is shared with the bacterial C5 inhibitor SSL7. Although SSL7 binds to the MG1, MG5, and MG6 domains in C5, quite far from the putative convertase-interacting residues in the MG4, MG5, and MG7 domains, SSL7 elegantly hijacks IgA into a ternary IgA–SSL7–C5 complex and in this way introduces steric hindrance for convertase binding (28). The tick protein OmCI also prevents C5 cleavage by the convertases, and based on small-angle x-ray scattering data, we previously suggested OmCI to bind near the C345c domain of C5, far from the scissile bond (27). This is now confirmed by the crystal structures of the C5–OmCI–RaCI complexes (29). Interestingly, these structures also reveal that three different RaCI inhibitors all bind in the same pocket located between the C5d, MG2, and MG1 domains, again very far from the scissile bond region. They might function by locking C5 into a conformation that prevents the conformational change needed to allow the C5 scissile bond region to unfold and insert into the catalytic site of the catalytic subunit of the convertase. OmCI appears to work differently as compared with the RaCI inhibitors as it does not prevent inhibit C5 cleavage by CVFBb, but this may relate to the positional stabilization of the C5 C345c domain upon OmCI binding (27, 29). Perhaps flexibility of this domain is required for cleavage by the endogenous C5 convertases but not by CVFBb.

In conclusion, our structure provides an explanation of both the exquisite selectivity of eculizumab for human C5 and the mechanism of eculizumab resistance observed in a small subset of patients with PNH with polymorphisms encoding mutations at Arg⁸⁸⁵. With further refinement, this model may provide structure-based insights to guide engineering efforts to make even more efficient versions of eculizumab by pinpointing the key residues that interact with C5.

We thank the beamline staffs at ESRF for support during data collection and Tristan Croll for advice on model building.

J.A.S.-J. and G.R.A. declare a collaboration with Alexion. Y.Z., K.J., A.N., and D.S. are employees at Alexion Pharmaceuticals, marketing the eculizumab Ab.