A New Map of Glycosaminoglycan and C3b Binding Sites on Factor H¹

Central to immunity is the capacity for discriminating self from non-self. In the case of the alternative pathway of complement activation (1), C3b molecules are continuously deposited onto surfaces. Deposition occurs through a covalent linkage with a hydroxyl or other nucleophilic group, mediated by a thioester that is exposed when C3b is cleaved from C3. Each C3b molecule has the potential to nucleate a positive-feedback loop involving formation of bimolecular C3 convertase complexes (containing C3b and Bb, i.e., C3b,Bb) that enzymatically generate many more molecules of C3b. Complement factor H (fH)³ (2, 3), as well as other members of the regulators of complement activation (RCA) family (4), prevent this amplification of C3b on self-cell membranes and extracellular matrix by perturbing convertase assembly and stability, and by promoting factor I-catalyzed C3b cleavage. On the other hand, C activation proceeds unchecked on many bacterial and other non-self surfaces that lack regulators, precipitating lytic and inflammatory events (5).

Thus, the presence of protective regulators on self surfaces is crucial to self vs non-self discrimination by the C system. Of the RCA proteins that regulate the alternative pathway, membrane cofactor protein (CD46) (6) and complement receptor type 1 (CD35) (7, 8) are integral to the membranes, while decay accelerating factor (CD55) is GPI-anchored (9). Factor H, on the other hand, is a soluble protein (10). It not only modulates fluid-phase C but also has the ability to inhibit C3b amplification selectively on self surfaces (11). This role of fH is particularly significant for self surfaces not protected by membrane-associated RCAs.

Factor H is a 155-kDa, 1213-aa residue glycoprotein (12) composed entirely of 20 complement control protein (CCP) modules (13) (also known as short consensus repeats or SCRs) of ∼60 residues joined in a “string-of-beads” arrangement by 19 short, potentially flexible, (14, 15) linking sequences of three to eight residues. Key to the ability of fH to act specifically at self surfaces is a polyanion binding site mapped to one face of its C-terminal (i.e., 20th) CCP (16, 17). Another important polyanion binding site was localized to CCP 7 (18, 19). Variations among residues that contribute to these two binding sites are linked to three complement-mediated diseases (20): dense deposit disease (membranoproliferative glomerulonephritis type II) (21), atypical hemolytic uremic syndrome (22, 23, 24), and age-related macular degeneration (25, 26, 27, 28, 29). Two further polyanion binding sites were reported, in CCP 9 (30) and between CCPs 12 and 15 (31), although these modules have not been found to contain disease-linked sequence variations. All these putative polyanion binding sites are thought to adhere to electronegative carbohydrates such as glycosaminoglycans (GAGs) and sialic acid in proteoglycans, glycoproteins, and the extracellular matrix. Factor H was reported to additionally carry three C3b binding sites (32): the four N-terminal CCPs bind C3b and (even in the absence of the other 16 modules) also act as cofactor for factor I-catalyzed C3b proteolysis (33, 34, 35) and accelerate decay of alternative pathway C3 (and C5) convertases (36); CCP 20 is thought to, uniquely, have both C3b- and polyanion binding sites (37); and a third C3b binding location was inferred in CCPs 12–14 (32). Current theories for selective activity of fH at self surfaces must therefore attempt to explain the presence of this multiplicity of binding sites.

Research effort has focused on binding sites lying toward either terminus of fH (i.e., in CCPs 1–7 and CCP 20), but C3b- and polyanion-recognition sites within the central modules of fH are underexplored. Evidence for involvement of CCP 13 arose from binding of fH to a heparin analog incorporating a photoactivatable cross-linker (31). That CCP 9 binds heparin is based on studies of recombinant constructs representing CCPs 8–9 and CCPs 9–11 (30) that also include an N-teminal cloning artifact containing two Arg residues. Suggestions of CCPs 12–14 (32) as the third C3b binding site were inferential since no such construct was made and tested directly for binding. To complicate the matter further, there is evidence that some binding sites are cryptic within full-length protein and only available when the C-terminal site is occupied (38).

In the present study we expressed recombinantly a set of constructs from fH and assessed their binding to heparin-affinity resin, to defined-length heparin-derived oligosaccharides, and to immobilized C3b. Two segments, CCPs 7–8 and CCPs 19–20, were shown to have binding sites both for polyanions and for C3b (although the C3b binding site in CCPs 7–8 is very weak), while CCPs 1–4 bind only C3b. No further C3b or heparin binding sites of significant affinity were found in seven segments of various lengths representing CCPs 9–15, spanning the full region of previous contention over the existence of such sites. This improved map of fH binding sites allows construction of a new model for fH action at cell surfaces.

The DNA sequences encoding the appropriate segments of fH (see Fig. 1 and Table I; residues are numbered according to encoded protein sequence, i.e., before removal of secretion signal) were cloned into the Pichia pastoris expression vector pPICZα (Invitrogen). In the case of fH CCPs 8–9 (i.e., fH 8–9) the native Lys at position 446 in fH was engineered into a previously prepared fH 8–9 construct lacking this residue using a QuickChange site-directed mutagenesis kit (Stratagene) to substitute the Gly from the cloning artifact with the native Lys. In the case of fH m1–4h, a hexahistidine tag was added to the C-terminal end and a myc-epitope tag was added to the N-terminal end of a previously prepared fH 1–4 clone using a QuickChange site directed mutagenesis kit. Expressed proteins were directed to the secretory pathway by placing the coding sequence behind the Saccharomyces cerevisiae α-mating factor secretion sequence.

Following transformation into P. pastoris strain KM71H (Invitrogen), proteins were expressed in shaker flasks or a fermentor. Where appropriate, proteins were isotopically labeled (using ¹⁵N-ammonium sulfate, and D₂O in the case of fH 6–8, in the growth media) in batches of 0.8 liter (initial volume) of cell culture. Cation- or anion-exchange chromatography was used as a first purification step (except in the case of fH m1–4h, where a HisTrap kit (Amersham Biosciences) was used in the initial step) and was followed by another ion-exchange chromatography step or gel-filtration chromatography. Proteins that exhibited N-linked glycosylation were deglycosylated before purification, or between the first and second purification steps, by incubating 100 ml supernatant with 6000 U EndoHf (New England Biolabs) at 37°C for 3 h. Yields were typically in the region of 0.1–0.5 mg of pure protein per gram of wet cells.

Purified proteins were suspended in 50 mM ammonium bicarbonate, then reduced and alkylated using 5 mM DTT and 15 mM iodoacetamide. Following digestion with trypsin (1/100 (w/w) trypsin-protein, 2 h, 37°C), a 0.5-μl aliquot of the digest was mixed with 0.5 μl α-cyano-4-hydroxycinnamic acid matrix (10 mg/ml in 50% (v/v) acetonitrile containing 0.1% (v/v) trifluoroacetic acid) on a MALDI sample plate. Samples were then analyzed on a Voyager-DE STR biospectrometry workstation MALDI-TOF mass spectrometer (Applied Biosystems), and processed spectra were searched against the National Center for Biotechnology Information nonredundant database or in-house database using ProteinProspector (University of California, San Francisco, http://prospector.ucsf.edu) or Mascot (Matrix Science, http://www.matrixscience.com).

All the NMR spectra were acquired on either Avance (Bruker BioSpin) 600-MHz or 800-MHz spectrometers fitted with 5-mm cryogenic probes. Data were acquired on samples at concentrations ranging from 50 to 125 μM (for one-dimensional ¹H-NMR) or from 100 to 600 μM (for two-dimensional (¹H,¹⁵N)-NMR) at 25°C or 37°C. Samples were buffered in either 20 mM potassium phosphate (pH 6.6) (all except fH 6–8 and fH 19–20); 20 mM potassium phosphate, 1 mM EDTA (pH 6.2) (fH 6–8); or 20 mM sodium acetate, 200 mM NaCl (pH 4.5) (fH 19–20).

Factor H binding to C3b was monitored by surface plasmon resonance (SPR) using a Biacore T100 instrument (GE Healthcare). The sensor surfaces were prepared by immobilizing human C3b (Complement Technology) in two or three of the four flow cells of Biacore series S carboxymethylated dextran (CM5) or carboxymethylated matrix-free (C1) sensor chips (GE Healthcare), using standard amine coupling and as summarized in Table II; the reference surface in each case was prepared in the remaining flow cell(s) by performing a dummy coupling reaction in the absence of any proteins. Experiments were performed at 25°C using a flow rate of 30 μl/min (after performing a flow-rate study to check for mass transport limitation). Duplicate injections of fH samples in 10 mM HEPES-buffered 150 mM saline with 3 mM EDTA and 0.05% (v/v) surfactant p20 (HBS-EP+) were performed at the concentrations indicated. A contact time of 90 s was used, as this was found to be sufficient to achieve steady-state conditions, followed by a dissociation time of 600 s with HBS-EP+ as running buffer. The chips were regenerated between sample injections by two injections of 1 M NaCl for contact times of 45 s. Data were processed using Biacore T100 evaluation software version 1.1. Reporter points for affinity measurements were set to 2 s before injection began and 2 s before the injection period finished. Dissociation constants were calculated by fitting steady-state binding levels derived from the background-subtracted traces to a one-to-one binding steady-state model.

Protein samples (50–78 μg, 1 ml) in 20 mM potassium phosphate buffer (pH 7.4) were loaded individually onto either a HiTrap heparin-affinity chromatography column (7 × 25 mm, GE Healthcare) or a Poros 20HE heparin-affinity chromatography column (4.6 × 100 mm, Applied Biosystems) equilibrated with 20 mM potassium phosphate buffer (at pH 7.4) and subsequently eluted with a linear gradient of 0–1 M sodium chloride in 20 mM potassium phosphate buffer (pH 7.4).

Oligosaccharides were prepared from low-molecular-weight heparin by partial digestion with heparinase I followed by size fractionation on a BioGel P10 gel filtration column (Bio-Rad) (39). Fluorophore-labeled species were produced by attachment of 2-aminoacridone to the oligosaccharide reducing end (39), and GMSA were performed, as described previously (40). Briefly, 2-aminoacridone tagged oligosaccharides were combined with the recombinant segments of fH at a range of concentrations in a volume of 10 μl of PBS containing 25% (v/v) glycerol for 15 min (at room temperature). Samples were then loaded on a 1% agarose gel in 10 mM Tris-HCl (pH 7.4) and 1 mM EDTA. Electrophoresis was performed (200 V, 8–15 min) in a horizontal agarose electrophoresis system using an electrophoresis buffer comprising 40 mM Tris/acetate, 1 mM EDTA (pH 8.0). Immediately thereafter, the fluorescent oligosaccharides were visualized.

We set out to reexamine locations of putative GAG and C3b binding sites within the fH molecule. We therefore decided to overexpress proteins corresponding to a mostly overlapping array of short segments within fH (see Fig. 1), to check their structural integrity, and to test for binding to GAGs and C3b in vitro. Clones for the module pair consisting of CCPs 7 and 8 (i.e., fH 7–8) and the module pair fH 19–20 were already available (17, 19). These two recombinant proteins have been shown unambiguously to fold properly and bind heparin under approximately physiological condition. Hence, they served as positive controls in the present study. The triple module fH 6–8 was also used as a positive control on GMSA since its ability to bind heparin had been previously well characterized (19, 29). Further positive controls were provided by full-length fH purified from plasma and the construct fH m1–4h (see Table I) containing modules 1–4 (known to bind C3b) with an N-terminal Myc tag and a C-terminal hexahistidine tag. All proteins were purified to the extent that they ran as single bands when overloaded on nonreducing SDS-PAGE. Upon ageing, four constructs (fH 8–15, fH 10–15, fH 11- 14, and fH 13–15) exhibited evidence of dimer and trimer formation under nonreducing conditions (data not shown), necessitating fresh sample preparations for all assays. Under reducing conditions, SDS-PAGE of these four constructs (each of which includes CCP 14) revealed a small proportion of degraded protein that was absent from the non-CCP 14-containing constructs. Specifically, samples of fH 8–15, fH 10–15, fH 11–14, and fH 13–15 exhibited <5% degradation. The primary structures in all cases were confirmed by tryptic digestion and MALDI-TOF mass spectrometry.

Structural validation was based on NMR spectroscopy. ¹⁵N-labeled samples were prepared for heteronuclear NMR studies of the key constructs fH 6–8, fH 8–9, fH 12–13, and fH 13. In a (¹H,¹⁵N) heteronuclear single quantum coherence (HSQC) spectrum, a cross-peak is expected for each backbone amide group; for small folded proteins, cross-peaks are discrete and well dispersed. As may be seen from Fig. 2, the (¹H,¹⁵N) spectra of fH 6–8, fH 8–9, and fH 12–13 and fH 13 are of a similar high quality to the spectrum of fH 19–20. Therefore, all modules within each construct are properly folded and the expected pattern of disulfide formation may be confidently inferred. Given that P. pastoris-expressed CCPs 6, 7, 8, 9, 12, and 13 all fold properly, the use of one-dimensional ¹H spectra (Fig. 2) was deemed adequate for structural assessment of the other fH segments expressed in this study. Except in the case of fH 13–15, line widths and signal dispersion are consistent with monodispersed, well-folded protein molecules. The fH 13–15 construct shows evidence of aggregation or unfolded material in addition to a high proportion of properly folded monomeric protein.

We first attempted to confirm a previous report that CCP 9 harbors a GAG binding site (30). Fully characterized and authenticated fH 8–9 eluted from both HiTrap (Fig. 1,B) and Poros heparin-affinity columns (data not shown) in 20 mM phosphate buffer (pH 7.4) with no additional salt; on the other hand, >>150 mM NaCl was required to elute positive controls, fH 7–8 and fH 19–20 (Fig. 1, B and D). In a consistent result, fH 8–9, unlike positive controls, failed to retard the mobility of a range of purified, defined-length heparin-derived oligosaccharides in a GMSA (Fig. 3, A and C). This lack of detectable affinity for heparin by fH 8–9 is in apparent contradiction to the previously published report of significant binding of heparin by several constructs that encompass CCP 9 (30). Explanations based on misfolding of fH 8–9 are eliminated on the basis of its NMR-authenticated structure. To investigate further, we expressed the same sequence as was used in the key experiment of the previous study. Thus, we produced fH rr8–9 (Table I), which incorporates an N-terminal sequence artifact containing two Arg residues. This construct does indeed bind to a HiTrap heparin-affinity column (although only slightly to a Poros column, data not shown) significantly better than does fH 8–9 (Fig. 1,B) and almost as tightly as the positive controls. It also binds to sulfated heparin fragments according to GMSA (Fig. 3 C). The clear implication is that the extraneous dibasic sequence contributes nonspecifically to heparin binding affinity of fH rr8–9. To investigate further, we tested a synthetic peptide of sequence EFTWPSRPSRIGTKT for binding to heparin-affinity resin. This sequence matches the non-native sequence at the N terminus of fH rr8–9 plus two native residues (Lys and Thr). The peptide does not have a strong affinity for heparin (data not shown). It is therefore concluded that the heparin binding site in fH rr8–9 is a composite of non-native N terminus and native sequence.

Based on a previous report of heparin binding to CCP 13 or CCP 14 of factor H (31), we decided to further investigate this potential GAG interaction site. Of note is that CCP 13 has more positively charged residues than do other CCPs in fH. Surprisingly, fH 12–13 failed to adsorb to either of the two heparin-affinity columns used in this study at physiological salt concentration and pH (Fig. 1,C). Nor did fH 12–13 bind to defined-length (from 4 to 12 sugar units) heparin-derived oligosaccharides in the GMSA (Fig. 3,B), even at a 4:1 ratio of protein to sugar. It is conceivable that their long (eight-residue) intermodular linker allows CCPs 12 and 13 to arrange themselves side-by-side in the context of the isolated pair such that the putative GAG binding site on CCP 13 is inaccessible. This possibility was eliminated by our observation that when expressed as a single module CCP 13 (fH 13) was unable to bind GAGs (Fig. 3, B and D). It was also a possibility that residues from module 14, or from the 13–14 linking sequence, are required to complete a GAG binding subsite in CCP 13. This was excluded by a study in which it was observed (Figs. 1,C and 3,C) that fH 11–14 is neither retained to a significant extent on a heparin column nor does it bind to heparin in a GMSA. Next, the possibility that CCP 14 rather than CCP 13 might be central to a longer putative binding site was considered. To this end, fH 13–15 was produced, but it was also found not to bind the heparin-affinity column or to produce more than a hint of binding according to GMSA (Figs. 1,D and 3 C). The presence (as judged by NMR) of some unfolded or aggregated material in the fH 13–15 sample would be most unlikely to explain this lack of binding by the majority of folded fH 13–15 material that is also present.

Thus, these results show clearly that modules 13 and 14 of fH and their immediate neighbors do not constitute a discrete GAG binding site comparable to the ones present in CCP 7 and CCP 20.

Having failed to detect heparin binding within CCPs 8 or 9 or CCPs 11–15, we chose to examine intervening modules and expressed fH 10–12. This construct was not retained on a heparin column nor was it positive by GMSA (Figs. 1,C and 3,C). We then reasoned that CCPs 12–13 (small modules joined by a long linker) might be the hinge where the fH molecule folds back upon its self, allowing non-neighboring modules to form a composite GAG binding site. To investigate this, we included two CCPs on either side of fH 12–13 by expressing fH 10–15. Size-exclusion chromatography of freshly prepared fH 10–15 (data not shown) implied a globular, rather than an extended, shape of fH 10–15, in support of the case for higher order structure. Nevertheless, this construct did not bind heparin with significant affinity (Figs. 1,C and 3,C). Finally, we expressed fH 8–15 to test whether CCPs 9 and 13 might individually be relatively weak GAG binders but nonetheless contribute to a common, higher affinity GAG binding site. Despite encompassing two previously reported GAG binding modules, this longer construct does not have significant heparin affinity (Figs. 1,B and 3 C).

When acting as a C regulator on self surfaces, fH presumably binds simultaneously to both GAGs and C3b. We first investigated, using SPR, the affinities for C3b of various fragments of fH, employing constructs fH m1–4h and fH 19–20 as positive controls. For these exploratory experiments we did not attempt to measure K_d values; instead, we compared affinities based on the number of response units (RU) measured (that correlates with the amount of analyte that binds to the sensor chip) when a 10 μM solution of the fragment was passed over the chip surface. Fig. 4,A illustrates the quality of the sensorgrams; Fig. 4 B summarizes the size of the response (normalized to the binding of fH 19–20) obtained both before and after adjustment to take account of the M.W. of the analyte.

The construct fH 11–14 incorporates the previously inferred C3b binding modules 12–14, but its affinity for C3b was insignificant compared with that of fH m1–4h or fH 19–20 (Fig. 4). The putative 12–14 binding site had been inferred from module-deletion experiments, the interpretation of which may be complicated by neighboring-module effects. We therefore tested fH 10–12 and fH 13–15 but found these to have no affinity for C3b. Nonetheless, the possibility of a composite site for C3b, requiring non-neighboring modules, could not be eliminated on the basis of results obtained with shorter constructs. We therefore tested fH 10–15 for C3b binding but obtained a response (at a target loading of 4200 RU) of <4 RU that we interpreted as reflecting negligible affinity (Fig. 4). Since we had produced fH 8–15 for the heparin binding study, this construct was also assayed; we were able to detect some evidence of binding when a 10 μM sample was passed over the sensor chip (Fig. 4), although the response was significantly smaller than that obtained from injection of even 1 μM positive controls (data not shown). This implies that CCPs 8 or 9 contribute to a C3b binding site. On the other hand, the double module fH 8–9 did not have a measurable affinity for C3b in this assay. Hence, we cannot rule out some degree of cooperativity between low-affinity sites in module 8 or 9 and another low-affinity site within the 10–15 region.

To investigate the role of module 8 more thoroughly, we assayed the C3b binding behavior of fH 7–8. Surprisingly, since the possibility has not received serious attention before, fH 7–8 (Y402) binds C3b. Taking into account its smaller mass (two modules in fH 7–8 compared with eight in 8–15), the sensorgrams indicate that significantly more fH 7–8 molecules bind to the C3b-coated chip compared with fH 8–15 under comparable conditions, although fH 7–8 binding is still much weaker compared with fH 19–20. This implies that module 7 is a far more important contributor to C3b binding than are modules 9–15. We further compared the relative binding of H402 and Y402 allotypic variants in the context of the triple module fH 6–8, and we found that they bind with approximately equal strengths (data not shown).

To further investigate the extent to which the binding to C3b of CCPs 1–4 and 19–20 dominates the fH-C3b interaction, we measured the K_d values for fH m1–4h and fH 19–20 and compared them to that of full-length fH. These experiments were repeated at several C3b densities and on two different types of sensor chip (Fig. 5, Tables II and III). The averaged K_d values of fH m1–4h and fH 19–20 (as measured on the CM5 chip) are approximately 8- and 2-fold, respectively, weaker than that of full-length fH. These results are not inconsistent with a modest avidity effect on the fH-C3b interaction based predominantly on mutual contributions of these two binding sites. We repeated these measurements on a C1 chip, and we minimized the loading on the C1 chip of C3b in an attempt to achieve a situation where adjacent molecules of C3b in appropriate orientations are unlikely to be available for binding to a single fH molecule. The K_d values obtained on the C1 chip are basically comparable to those measured on the CM5 chip with a slightly larger but still modest avidity effect (Table III).

The use of C regulators to selectively protect self from indiscriminate amplification of surface-deposited C3b via the alternative pathway is a rudimentary but effective strategy for immune surveillance (41). The first four CCPs of fH are necessary and probably sufficient for the ability of this protein to act in the fluid phase as a cofactor for factor I, and they have some ability to accelerate decay of convertases (34, 36, 42). The remaining 16 modules ensure that regulatory potential is delivered selectively on self surfaces; the presence of three or four polyanion binding regions (at CCPs 7, 9, and/or 13 and 20), along with C3b binding sites at CCPs 12–14 and CCPs 19–20 (reviewed in Ref. 3), have all been previously proposed to contribute in some way to this selectivity of fH action. The current data demonstrate, however, that modules 9–15 lack discrete strong C3b or heparin binding sites. The results suggest that any contribution they make to binding is relatively small. How can our data be reconciled with previous findings, and what does the revised map of binding sites suggest in terms of mechanisms of recognition?

Evidence for the GAG binding of CCP 13 originated from crosslinking studies involving fH and a heparin analog (31). It is striking that fully folded CCP 13, with a prominent electropositive patch on one face (C. Schmidt, A. Herbert, C. Fenton, D. Uhrin, P. Barlow, our unpublished data), does not bind well to heparin. It suggests that specifically positioned side-chains, rather than complementary charge alone, are required for GAG binding at physiological salt concentrations. That CCP 13 in the context of fH 12–13, and of four larger fragments, does not bind heparin eliminates the possibility of neighboring modules completing a partial GAG-recognition site in CCP 13. It remains possible, however, that types of GAG other than the heparin/heparan sulfate family, or those with different sulfation patterns for example (43), interact with CCP 13, although heparin is rarely, if ever, a weaker ligand than heparan and dermatan sulfates for GAG binding proteins. It is also possible that a very weak binding site in CCP 13 cooperates in GAG binding with modules 7 and 20 once fH is anchored at the cell surface via the latter modules. However, CCP 13 should not be considered in the same vein as modules 7 and 20 in terms of GAG binding by fH.

Unlike CCP 13, CCP 9 appears on the basis of its pI as an unlikely candidate for a heparin binder. Indeed, a construct consisting of CCPs 1–6 followed by 8 and 9 (i.e., the ΔCCP 7 version of fH 1–9) was previously reported not to bind heparin (44). The more recent results of Ormsby et al. identifying CCP 9 as a GAG binding module (30) were therefore surprising. We were, however, able to convert fH 8–9 from a non-GAG binding construct to a GAG binding one by addition of non-native cationic N-terminal residues present in the Ormsby et al. constructs. This suggests that the non-native Arg residues were critical for the previously reported interaction. It remains possible that CCP 8 contributes to the well-explored GAG binding site centered on neighboring CCP 7 (29), but we are unable to find evidence that CCP 9 has even a partial binding site; indeed, the absence of GAG binding by CCP 8–15 would suggest that it does not.

An earlier study noted depleted binding to C3b immobilized via its thioester to an E surface by an fH deletion mutant lacking CCPs 6–10 (i.e., fH Δ6–10) (45). It was unclear whether this result reflected a loss of direct interaction with C3b, since there is a concurrent loss in this mutant of the GAG binding site of CCP 7 that could contribute to fH association with C3b in the context of a GAG-bearing cell surface. Subsequent experiments performed by Jokiranta et al. (32) and in the current study, in which C3b is immobilized via amine coupling to a carboxymethylated dextran SPR sensor chip, measure direct C3b-fH interactions. Note that we detected no significant differences in the SPR-derived K_d for fH 19–20 when comparing amine-coupled C3b to C3b immobilized via a biotinylated thioester linkage to an avidin chip (data not shown). The previous SPR studies indicated that fH 8–20 and fH 19–20 bind C3b, but fH 8–11 and fH 15–18 do not (32). Note that these results do not conflict with the ones reported in the present study. In the previous work (32), however, a C3b binding site in CCPs 12–14 was inferred on the basis that fH 8–20 binds both C3d and C3c while fH 19–20 binds C3d but not C3c (C3c and C3d are distinct fragments of C3b). That none of our constructs (fH 10–12, fH 10–15, fH 11–14, or fH 13–15) binds C3b with an affinity remotely comparable to that of fH m1–4h or fH 19–20 falsifies the hypothesis that CCPs 12–14 constitute an autonomous C3b binding site. On the other hand, our observation that fH 7–8 and fH 6–8 have significant affinities for C3b immobilized on a chip is consistent with original studies on fH Δ6–10 (45). It remains possible that fH 12–14 binds to C3c (since new binding sites are exposed when the thioester domain is cleaved from C3b (46, 47)), but this site needs no longer figure in proposed mechanisms for recognition of C3b at the cell surface. The K_d measurements for fH m1–4h, fH 19–20, and fH support the predominant role of the N- and C-terminal C3b binding sites but they do not rule out small contributions from weaker binding sites such as fH 6–8, or even from other potential contact sites whose affinities as individual entities are too weak to measure. A summary of re-mapped fH binding sites is shown in Fig. 6.

That fH modules 9–18 have now been shown not to have a significant involvement in binding to C3b or GAGs begs the question, What is the biological purpose of these modules? A notable feature of modules 12–14 is the presence of relatively small modules connected by longer than average linking sequences. For example, between modules 12 and 14 there are 66 residues, of which only 51 belong to CCP 13. Together with biophysical data that are consistent with a bent-back conformation of the parent molecule (48) and the size-exclusion chromatography profile of fH 10–15, this observation suggests that the CCPs 12–14 region is the site of a bend or hinge in fH. The presence of such a hinge allows spatial proximity of the N- and C-terminal portions that between them contain all the experimentally proven, discrete binding sites for GAG and C3b. The previous observation that an Ab to CCP 20 can block the cofactor activity of fH (38) is consistent with the relevant sites being close in space at least when fH is engaged with the surface-associated convertase. That fH 19–20 can overcome the protective effects of fH against C-mediated lysis of E (16) points to a key role for this region in anchoring fH to the surface, perhaps via a composite binding site consisting of C3b and GAGs (37) with specific patterns or densities of sulfation. A bend in fH would allow CCP 7 to bind nearby, perhaps acting in a “proofreading” role in that it could recognize a second composite C3b-GAG binding site, again containing a particular distribution of sulfates. Binding of this nature would then place the N-terminal four CCPs at a specific position relative to the C3b(C3d) binding site in the C-terminal module and to the surface-bound C3b. Such positioning might be critical for efficient operation of the N-terminal functional unit in its cofactor and decay-accelerating roles.

Involvement of two (or more) sites in binding of fH to a C3b molecule on the CM5 chip (Fig. 6,B) could explain the 2-fold stronger K_d of the full-length fH protein compared with that of the tightest binding fragment, fH 19–20. On the other hand, the high density of the C3b molecule population tethered to the flexible dextran matrix of the CM5 sensor chip, or on the surface of a cell membrane, could facilitate bridging of two (or more) C3b molecules by one molecule of fH (Fig. 6,C). On the matrix-free surface of the C1 chip, C3b molecules were less flexibly, and more sparsely, attached; it is therefore noteworthy that, even at our lowest target density (130 RU, equivalent to 130 pg/mm² (49)), a modest but significant avidity effect is manifested (as judged from the K_d values on the C1 chip for fH and fH 19–20 in Table III). Under these conditions, one would expect individual C3b molecules to be tethered to the chip at an average distance from one another of around 500 Å, equivalent to 12 CCP module lengths. Thus, while it is still conceivable that fH spans two appropriately orientated neighboring C3b molecules, bi- or multivalent one-to-one binding (Fig. 6,B) appears more probable at these low C3b densities. This does not rule out incidences of the Fig. 6 C scenario predominating at high C3b densities on the cell surface.

Thus, the principal role of some of the central modules could be to act as a set of spacers, projecting the functional regions away from a hinged central region also composed from CCPs. This configuration allows the key binding sites to approach each other so as to act cooperatively in selectively engaging and destabilizing the self surface-associated convertase complexes (Fig. 6, B or C). Such an arrangement is reminiscent of the mechanism employed by the RCA C4b binding protein, in which functional sites at the tips of seven arms (each consisting of eight CCPs) cooperate in recognizing a composite surface of GAGs and C4b (50). The complement receptor type-1 also employs cooperation between sites that are remote in its sequence, for example to accelerate decay of the C5 convertase (51). The potential for cooperation between CCPs 7 and 20 suggested by Fig. 6 is interesting in that such an arrangement has an enhanced capacity for combinatorial recognition of specific types of GAG molecules among diverse possibilities. This supports a model in which complement regulation in tissues displaying different GAGs could be differentially susceptible to mutations and polymorphisms within modules 7 and 20; between them, these two modules account for nearly all the disease-related sequence variations in fH.

We acknowledge Jon A. Deakin (University of Manchester, Manchester, U.K.) for technical assistance.

The authors have no financial conflicts of interest.