Mutational Analysis of the Active Site and Antibody Epitopes of the Complement-inhibitory Glycoprotein, CD59

Human CD59 cDNA was obtained from Professor Herman Waldmann (Sir William Dunn School of Pathology, Oxford, UK) and Dr. David L. Simmons (Institute of Molecular Medicine, Oxford, UK). The YTH53.1 hybridoma was also obtained from Professor Waldmann. The anti-CD59 monoclonal antibody BRIC 229 was purchased from IBGRL (Elstree, UK); antibody 2/24 was a gift from Dr. A. Fletcher (Australian Red Cross Society, Sydney, NSW, Australia); antibodies MEM43 and MEM43/5 were gifts from Dr V. Horejsi (Czech Academy of Sciences, Prague); antibody A35 was prepared in house (B.P.M., unpublished); antibodies HC1 and HC2 were gifts from Dr. S. Meri (University of Helsinki, Finland). The polyclonal antisera specific for human CD59 and for Chinese hamster ovary (CHO) cell membranes were prepared in house using standard methods.

A Bam H1 fragment containing the translated region of a CD59 complementary DNA (22) was subcloned into the expression vector pEF-BOS (34). Site-specific mutations were introduced into the gene by in vitro mutagenesis using a commercial kit (Muta-Gene Version 2; Biorad) following the manufacturer's directions. For all mutants, the presence of the desired mutation and fidelity of the part of the gene encoding the rest of the fully processed protein was confirmed by dideoxy sequencing.

CHO cells, in Optimem medium (Life Technologies, Paisley, UK), were transfected with liposomes (Lipofectamine; Life Technologies) containing the expression construct according to the manufacturer's instructions. For analysis of expression levels by FACS^®, the cells were detached 48–72 h after transfection, washed three times in FACS^® buffer (1% BSA and 0.02% sodium azide in PBS) and resuspended at a density of 5 × 10⁶ cells/ml. The cells were then incubated at 4°C for 30 min with an equal volume (100 μl) of polyclonal antihuman CD59 antibody at 10 μg/ml in FACS^® buffer, and then washed three times in FACS^® buffer prior to incubation at 4°C for 30 min with FITC-conjugated rabbit anti–mouse IgG at a final concentration of 10 μg/ml in FACS^® buffer. Fluorescence of the cells was then analyzed on a FACScan^® instrument (Becton Dickinson, Abingdon, Oxford, UK).

CHO cells transfected as described above on day 1 were harvested and replated on day 2, and were then left until the plates were confluent (generally day 4). Cells were then washed twice with serum-free medium, incubated in 0.25 ml complete medium (7.5% FCS) containing a 1/750 dilution of calcein-AM (Molecular Probes; 1 mg/ml stock in DMSO) for 1 h at 37°C, then washed once. Triplicate wells were incubated with either medium alone, a 1/5 dilution of normal human serum (NHS) in medium (no FCS), or with a 1/5 dilution of NHS plus blocking antibodies (BRIC 229 and MEM43 both at 10 μg/ml). All wells contained anti-CHO IgG (0.08 μg/ml) in a total volume of 0.25 ml. Blocking antibodies were not complement-fixing and in the absence of anti-CHO did not trigger NHS lysis. After a 1-h incubation at 37°C, all fluid was removed from the cells and transferred to 96-well plates for calcein measurement. The remaining cells were lysed with 0.25 ml PBS containing 0.1% Triton X-100 during a 15-min incubation at room temperature (RT), then the lysate removed to 96-well plates for calcein release assay.

Calcein fluorescence of supernatants was read in a Denley Wellfluor fluorimeter with the excitation filter set at 485 nm and emission filter at 530 nm. Percent lysis for each well was calculated as specific calcein release/total calcein loading, or:

Mean and standard deviations were determined from triplicate samples.

Individual 24well plates were coated with each of the 8 available antibodies at 10 μg/ml in 0.25-ml aliquots in 25 mM sodium carbonate buffer, pH 9, for 1 h at 37°C. Wells were blocked by incubation in 1% BSA in PBS/0.1% Tween for 1 h at 37°C and washed twice in PBS/Tween, once in PBS alone. Human RBC (0.25 ml of a 1% suspension in PBS), either untreated or preincubated with saturating amounts of each of the antibodies (10 μg/ml, 30 min on ice, followed by two washes in cold PBS), were added to triplicate wells for each coating antibody (and to uncoated, blocked wells as control for nonspecific binding) and incubated at 4°C for 1 h. The wells were washed twice in PBS to remove unbound RBC and then the bound RBC were lysed by addition of water (0.25 ml, 15 min RT). The lysate was removed to a 96-well plate, spun to remove ghosts and the absorbance in the supernatant measured at 412 nm in a plate reader. For each antibody combination, results are expressed as a percentage inhibition derived from the ratio of binding of uncoated RBC/antibodycoated RBC to a particular antibody on the well and are calculated as:

where: Abs1 = absorbance in supernatant from antibody-coated RBC bound to antibody-coated well; Abs2 = absorbance in supernatant from antibody-coated RBC bound to uncoated well; Abs3 = absorbance in supernatant from uncoated RBC bound to antibody-coated well; Abs4 = absorbance in supernatant from uncoated RBC bound to uncoated well. Inhibition was calculated separately for each of the three triplicates and means and standard deviations were determined.

CHO cells were transfected with each of the mutant constructs as described above. 2 d after transfection the cells were replated and, after two additional days, detached, and washed twice with serum-free medium then incubated in 0.25 ml complete medium (7.5% FCS) with each of the antibodies (each at approximately 10 μg/ml). Antibody binding to the cells was confirmed by FACS^® as described above.

Mutant and wild-type CD59-expressing CHO cells were lysed at 4°C in 1% Triton X-100, 1 μg/ml leupeptin, 1 μg/ml pepstatin A and 1 mM phenlymethylsulphonyl fluoride in PBS. Aliquots of the precleared lysate were electrophoresed in non-reducing buffer in 15% SDS-PAGE gels. The proteins were then electrophoretically transferred to nitrocellulose which was then probed with anti-human CD59 antibody MEM43. Bound antibody was detected using a chemiluminescent detection system (Western-Light, Tropix Inc., Bedford, MA).

Student's t test was used to analyze the data.

Since the region(s) of CD59 responsible for its complement-inhibitory activity are controversial, a systematic mutational analysis of amino acids distributed across the entire surface of the extracellular domain of the protein was undertaken. In using this approach it was assumed that CD59 inhibits MAC formation by binding to another protein, such as C8 or C9. Since the sites of most protein–protein interactions that have been characterized in detail are relatively large (700–900 Ų) (33) and the surface area of CD59 is small (∼3200 Ų) (19, 20), it seemed likely that the entire surface could be scanned with a small number of mutations. According to this strategy, a first set of mutations was designed to scan both faces and the edges of the disk-like domain at low resolution. Based on the results of the first round of mutagenesis, a second set of mutations was used to analyze specific regions in greater detail.

The selection of particular residues for mutation was based on several criteria. First, only amino acids with side chains pointing away from the structural core of the protein in the NMR structures were chosen. Assuming that the overall conformation of CD59 in the NMR structure represents that of CD59 when bound to a putative ligand (even though the conformations of the side chains are likely to differ), only side chains of surface residues will directly participate in binding interactions. Second, given the observed lack of strict homologous restriction between certain species (9), it is likely that the active site will include some residues conserved between species so residues representing the most highly conserved of the surface amino acids were selected for mutation (Fig. 1). Third, where possible, mutations were made in regions of the protein implicated in CD59 function by others in order to test their various proposals (19, 20, 23, 31, 32).

Having chosen the sites for mutation, highly nonconservative substitutions were introduced in order to maximize the chances of identifying residues forming the ligand binding surface rather than the smaller subset of residues, identifiable by alanine-scanning mutagenesis, that also contribute significant binding energy to the interaction (35). The likelihood that the mutations would disrupt folding was minimized by confining the substitutions to surface residues. As reasoned previously (36), if a mutant protein is expressed it can generally be expected to have a native-like structure since the drastic changes introduced according to this strategy are unlikely to have subtle effects on the protein fold. The substitutions made in the two rounds of mutagenesis are shown in Fig. 1.

To facilitate mutant screening, an assay for CD59 function in cells transiently expressing wild-type and mutant CD59 genes was developed. In this assay, transfected cells are loaded with calceinAM, a fluorescein-based dye that is converted to the fluorescent compound calcein by esterases in live cells. Calcein is effectively retained by intact cells but is released into the medium by lysed cells, where its fluorescence can be measured. The loaded sensitized cells are incubated with normal human serum as a source of complement and the extent of lysis determined by measuring the fluorescence of the released calcein (see Materials and Methods).

The calcein-release assay was used to measure the ability of each of the scanning mutants to inhibit complementmediated lysis of transiently transfected CHO cells (Fig. 2,A). FACS^® analysis indicated that the mutant proteins were all expressed at the cell surface at similar levels (Table 1). When the mutant-transfected cells were tested in the calcein-release assay, most were clearly protected from lysis by the presence of the mutant gene when compared with cells transfected with the expression vector alone (Fig. 2,A, black bars). Consistent with this, co-incubation with CD59blocking antibodies produced statistically significant reversals of the protective effects of these mutants (Fig. 2,A, white bars). These results suggest that most of the scanning mutations did not disrupt the folding or the protective activity of CD59. Although mutant F47E provided statistically insignificant protection compared to cells not expressing CD59 (P = 0.059), this mutant appeared to have some activity since a significant increase in lysis occurred in the presence of blocking antibodies (P = 0.004). Only one of the scanning mutants, W40E, completely failed to provide any protection from complement-mediated lysis. Loss of the inhibitory property of this mutant is not likely to be due to misfolding of the protein since it is recognized by the conformation-sensitive monoclonal antibodies HC1 and MEM43/5 (Table 1). These results indicate that W40 forms at least part of the active site essential for inhibiting the formation of the MAC by CD59. The weak protective effect of the mutant F47E suggests that this residue may be at the periphery of the active site or that it has subtle effects on the conformation of CD59 which indirectly affect its activity.

The sidechain of W40 partially fills a cleft formed by the packing of the α-helix against the three-stranded β-sheet on the glycosylated face of CD59 (19, 20). In a second round of mutagenesis, residues in and around this cleft, specifically K38, K41, R53, L54, and E56, were mutated. Residue D24, which is adjacent to W40 but projects in the opposite direction, was also mutated because it forms part of a conserved loop consisting of residues 20-24 (20, 23). Of these mutants, D24R, R53E, and E56R failed to provide any protection from complement that was reversible by co-incubation with CD59-blocking antibodies (Fig. 2,B). All of the mutations that were tested are represented on the NMR structure (19) in Fig. 3, A–C.

The binding sites of antibodies that block the complement-inhibitory activity of CD59 are likely to at least partially overlap with the active site of CD59, while epitopes of non-blocking antibodies are expected to lie in a region distinct from the active site. A map of the epitopes of such antibodies can therefore be used to independently confirm the location of the putative active site. A panel of anti-CD59 monoclonal antibodies with differing capacity to block the protective effects of CD59 was selected for this purpose. The panel consists of the antibodies YTH53.1, BRIC 229, MEM43, HC2, and 2/24, which all have strong CD59-blocking activity, A35, which blocks weakly, and HC1 and MEM43/5, which have no CD59-blocking activity (Morgan, B.P., unpublished data).

The antibodies were first characterized in a competition assay (Table 2). Competition between the antibodies was determined by measuring the ability of each antibody, prebound to human RBC, to inhibit the binding of the RBC to plates treated with a second anti-CD59 antibody. Whereas none of the blocking anti-CD59 antibodies competed strongly for binding with the non-blocking antibodies, most pairs of the blocking antibodies were competitive, as observed previously with a subset of these antibodies (37), and consistent with the existence of a single active site on CD59. The only pair of blocking antibodies for which competition was not observed was HC2 and BRIC 229. The possibility that HC2 has relatively weak affinity is suggested by the observation that strong competition for binding sites on HC2-coated RBC only occurred on plates coated with the weakly-blocking antibody, A35.

The antibody epitopes were mapped by measuring the ability of each antibody to bind to CHO cells transfected with each of the mutants. Although this approach cannot delineate their borders precisely, the data show that the epitopes of the blocking antibodies are overlapping but distinct (Table 1 and Fig. 4). All include W40, plus differing portions of continuous surface defined by R53 and other residues adjacent to or in close proximity to W40, suggesting that this region overlaps with the active site of CD59. This is consistent with the results of the calcein-release assay, in which mutants W40E and R53E were unable to protect cells from complement-mediated lysis. It may be significant, however, that mutation of D24 to R did not prevent the binding of any of the blocking antibodies, although this change disrupts CD59 activity. The binding data also indicate that the epitopes of the non-blocking antibodies both lie in a distinct region of the protein which includes L33. However, mutation of E56 to R, which disrupts the function of CD59, also prevents binding of the non-blocking antibody HC1. This implies that the epitope of HC1 and the active site of CD59 partially overlap but in such a way that antibody binding does not completely block CD59 activity. Although the epitope for HC1 appears discontinuous in the figure, this cannot be concluded since the role of intervening residues in antibody binding was not tested.

To test the role of the glycans at the highly conserved N-glycosylation site in the complement-inhibitory function of CD59, N18 was mutated to Q. The N18Q mutant form of CD59 runs as a single sharp band of ∼14 kD on SDS-PAGE, at least 5 kD smaller than wild-type CD59 expressed in CHO cells, which runs as a ladder of bands in the molecular mass range 20–25 kD, and erythrocyte CD59, which runs as a broad smear in the molecular mass range 19–25 kD (Fig. 5). The mobility and banding patterns of all of the other mutants were indistinguishable from those of wild-type CD59 (data not shown). The mobility of N18Q CD59 is similar to that of deglycosylated CD59 (28), indicating that the mutant is not glycosylated. The expression of mutant N18Q at the cell surface, where it binds all of the polyclonal and monoclonal antibodies (Table 1), demonstrates that N-glycosylation is not required for native-like folding and expression of human CD59. Mutant N18Q retains wild-type levels of complement-inhibitory activity (Fig. 2,B) and binds all of the blocking antibodies at levels similar to that of wild-type CD59 (Table 1), suggesting that the N-glycan does not contribute to the active site. This is consistent with the ability of YTH53.1 and BRIC 229 to immunoblot CD59 from endoglycosidase F-treated erythrocytes (37). These data suggest that N-glycosylation is not required for the maintenance of the structure or complement-inhibitory function of human CD59.

CD59 blocks the complement-mediated lysis of host cells by preventing assembly of the MAC, presumably through binding to C8 and/or C9 molecules in the nascent complex and preventing recruitment of the additional C9 molecules required for pore formation. Whilst it remains formally possible that CD59 has two distinct binding sites for C8 and C9, the present data strongly suggest that there is a single active site and that this site lies in the vicinity of W40. Mutation of W40 to E eliminates the complement-inhibitory function of CD59 and prevents the binding of all the function blocking anti-CD59 antibodies tested. Although W40 is partially buried in the low resolution NMR structures (19, 20), its mutation to E results in a protein that is probably folded in a predominantly native-like structure since it is expressed at the cell surface and is recognized by the polyclonal antiserum and both of the non-blocking antibodies to similar extents (60–75% of the wild-type level, Table 1). Mutation of the adjacent residue, R53, which clearly protrudes away from the core of the structure (19, 20), to E also inhibits the complement-inhibitory activity and reduces the binding of four of the six CD59blocking antibodies, implicating this residue as part of the active site of CD59. A third residue likely to be involved is E56, which lies only 7.5 Å (Cα-Cα) from R53 on the same face of the protein. However, E56 must lie at the periphery of the active site because it constitutes part of the epitope of a non-blocking antibody (Table 1; Fig. 4). Moreover, on the basis of recent carbohydrate analyses of erythrocyte CD59, it has been proposed that, in some tissues, residues T51 and T52, which occupy the region between E56 and R53 but were not mutated in the present study, may be O-glycosylated (37a). If this is the case the O-linked glycans are likely to disrupt any contiguous protein surface that includes W40, R53 and E56.

The role of the fourth residue whose mutation disrupted the function of CD59, D24, is less clear as this residue does not form part of a contiguous surface with W40, R53 and E56 (Fig. 3). D24 lies on the edge of the extracellular domain of CD59 only 6 Å from W40 (Cα-Cα; Fig. 3, A–C) but is separated from this residue by F23, K41, and F42, all of which were mutated without disrupting the function of the protein. Several explanations could account for this observation. First, it is possible that D24 is part of a discontinuous active site, or that CD59 has a second active site which includes D24. However, both possibilities seem unlikely since none of the epitopes of the panel of six blocking antibodies includes D24. Second, it is conceivable that the mutation of D24 indirectly alters the conformation of key active site residues. However, conformational effects, if these were to occur at all, can only be very restricted given that the D24R mutant binds a series of conformation-sensitive antibodies (Table 1). Finally, the possibility needs to be considered that residue D24, which forms part of a flexible loop (19, 20), is in a sufficiently mobile region of the molecule to allow its direct involvement in forming the active site. Additional experiments are required to determine the role of D24 in the inhibition of complement-mediated lysis.

The affinity of CD59 for the nascent MAC has not been measured directly, although it has been shown that soluble forms of CD59 block the formation of the MAC at relatively low concentrations (∼0.5 μM [27]). This suggests that CD59 participates in a protein–protein interaction of relatively high affinity compared to other cell surface protein–protein interactions (38). Sites of protein–protein interaction are often defined by regions of hydrophobic and irregular protein surface (33). In the NMR structures of human CD59 W40 lies at the base of a hydrophobic groove, formed by residues F23, C39, W40, and L54, which is lined on one side by a ridge of hydrophilic residues (K41, H44, R53 and R55; Fig. 3,D). Sequence alignments suggest that the hydrophobic groove is among the best conserved regions of CD59 (Fig. 3,E) with only F23 exhibiting any variation among the fully sequenced homologues (substitutions to leucine or glutamine; reference 23, Fig. 1). Given the relatively poor overall levels of CD59 sequence conservation and the observed lack of strict homologous restriction between species with known sequences, the conserved hydrophobic groove centred on W40 represents a reasonable candidate for the active site of CD59. Residues 53 and 56, which vary to some extent between species, may contribute some specificity to recognition by CD59. Although mutation of residues F23 and L54, which form part of this hydrophobic groove, did not abrogate the activity of CD59, the putative active site may not have severe topological constraints and, given that these residues are located at the ends of the groove, it is conceivable that the mutants retained sufficient hydrophobic character to be functional.

The NMR analysis of a form of CD59 which includes at least part of the non-lipid component of the GPI-anchor indicated that much of the seven residue COOH-terminal stalk is packed against the non-glycosylated face of the extracellular domain of the molecule (20). This organization of the molecule places the glycosylated face containing the proposed active site in a well exposed, membrane distal position. The NMR study also indicated that in all calculated structures containing model oligosaccharides similar to those present on the protein, the region of the proposed active site remains fully exposed. Consistent with this, the normal expression and full activity of the unglycosylated mutant, N18Q, suggests that the N-linked glycan does not directly contribute to protein folding or complement inhibition by CD59. However, the bulk of the N-linked glycan would be expected to dampen any flexibility at the base of the stalk which could, in turn, keep the extracellular domain in an optimal orientation for interaction with the nascent MAC. Removal of the N-linked glycan could therefore be expected to reduce the affinity of this interaction without eliminating it completely. A partial reduction in the affinity of the N18Q mutant for the MAC could be masked by overexpression of the protein and therefore might have gone undetected in the present study. An indirect role for the N-linked glycan in facilitating the inhibition of complement by CD59 may account for the conservation of the glycosylation site but fails to explain the observation that the enzymatically deglycosylated protein loses most of its activity (25).

The structural analyses of CD59 confirmed that Ly-6SF proteins have topological similarities to snake venom neurotoxins (19, 20). In the crystal structure of a complex between the snake toxin fasciculin and its receptor acetylcholinesterase, the residues that contact the receptor lie on the concave face of the toxin (39). The analogous face is the site of the proposed active site of CD59. Examples of conserved active site locations have been noted previously for other protein–protein recognition molecules including, for example, immunoglobulin superfamily adhesion molecules and cadherins (40, 41). However, the fasciculin-acetylcholinesterase crystal structure (39) and functional studies of modified forms of other neurotoxins (42, 43) have revealed that the primary contacts between the toxins and their receptors involve residues of loop II, defined as strands C and D and their connecting loop. In contrast, in CD59, the equivalent loop is significantly shorter and the proposed active site lies instead on the loop between strands D and E on the opposite end of the face. This loop is significantly longer and more elaborate than the corresponding loop in the neurotoxins, which lack the α-helix in this region. These results suggest, not unexpectedly, that the determinants of ligand recognition have diversified considerably since the emergence of protein–protein recognition molecules with this topology.