The crystal structure of the globular head of complement protein C1q provides a basis for its versatile recognition properties.

C1q is a versatile recognition protein that binds to an amazing variety of immune and non-immune ligands and triggers activation of the classical pathway of complement. The crystal structure of the C1q globular domain responsible for its recognition properties has now been solved and refined to 1.9 A of resolution. The structure reveals a compact, almost spherical heterotrimeric assembly held together mainly by non-polar interactions, with a Ca2+ ion bound at the top. The heterotrimeric assembly of the C1q globular domain appears to be a key factor of the versatile recognition properties of this protein. Plausible three-dimensional models of the C1q globular domain in complex with two of its physiological ligands, C-reactive protein and IgG, are proposed, highlighting two of the possible recognition modes of C1q. The C1q/human IgG1 model suggests a critical role for the hinge region of IgG and for the relative orientation of its Fab domain in C1q binding.

Innate immunity involves a combination of cell-surface receptors and soluble proteins with the ability to recognize microbial pathogens and thereby to generate signals that both orientate subsequent adaptive immune responses and trigger effector mechanisms (1,2). Most of these molecules are oligomeric and recognize molecular patterns on microorganisms (3). An archetypal molecule of this type is C1q, the recognition subunit of C1, the complex that triggers activation of the classical pathway of complement, a major element of innate immunity. C1q is a 460-kDa protein with the overall shape of a bouquet of flowers, comprising six heterotrimeric collagen-like triple helices that associate in their N-terminal half to form a "stalk," then diverge to form individual "stems", each terminating in a C-terminal heterotrimeric globular domain (4). It is well documented that most of the C1 complex ligands are recognized by these peripheral globular domains, or heads, of C1q, thus triggering activation of C1r and C1s, the proteases associated with C1q (5). It is also established that C1q binds to immune complexes containing IgG or IgM, but not to those having IgA, IgD, or IgE (6). The major C1q binding site on IgG has been mapped to the CH2 domain of the Fc portion of the molecule (7)(8)(9). Although C1q shows marked differences in its reactivity toward IgG subclasses, the reason for this selectivity is not known.
C1q is traditionally known for its ability to bind antibodies. However, it recognizes an amazing variety of other ligands. These include certain bacteria, viruses, parasites, and mycoplasma (6, 10 -12), underscoring its role as an antibody-independent defense protein. C1q also binds to C-reactive protein (CRP) 1 when complexed with exposed phosphocholine residues on bacteria, providing a further means of host defense (13). C1q is also capable of recognizing aberrant structures from self. Thus, in addition to cellular debris and sub-cellular membranes (14), it is established that C1q binds to, and induces clearance of, apoptotic cells (15), thereby playing a major role in immune tolerance. Recent studies also indicate that abnormal proteins such as ␤-amyloid fibrils (16,17) and the prion protein (18,19) are recognized by C1q. There are no obvious structural features shared by these ligands, but the fact that many polyanions are C1q ligands (6) suggests that C1q may function as a charge pattern recognition molecule.
The globular domain of C1q is a heterotrimeric association of protein modules known as gC1q domains found at the C-terminal end of various proteins, including types VIII and X collagens, the adipocyte complement-related protein (ACRP)-30, precerebellin and, multimerin (4). The structures of the globular domains of ACRP-30 (20) and collagen X (21) have been solved by x-ray crystallography, revealing the gC1q fold and indicating that both are homotrimers held together by both hydrophobic and polar interfaces. We report here the x-ray structure of the heterotrimeric globular head of C1q, the domain responsible for the versatile recognition properties of this protein. The structure reveals how three different gC1q modules achieve an assembly homologous to, but structurally more diverse than, the one observed in homotrimers and provides insights into the molecular mechanisms of the recognition function of C1q.

EXPERIMENTAL PROCEDURES
Preparation and Analysis of the C1q Globular Domain-C1q was purified from human serum as described previously (22) and digested with Achromobacter iophagus collagenase (Roche Applied Science) (enzyme/protein ratio ϭ 0.2, w/w) for 24 h at 37°C in 250 mM NaCl, 5 mM * This work was supported by the Commissariat à l'Energie Atomique, the CNRS, and the Université Joseph Fourier, Grenoble. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The N-terminal sequence analysis of the purified C1q globular domain was performed after SDS-PAGE and electrotransfer using an Applied Biosystems model 477 A protein sequencer as described previously (23). Mass spectrometry analysis was performed using the matrix-assisted laser desorption ionization technique on a Voyager Elite XL instrument (PerSeptive Biosystems, Cambridge, MA) under conditions described previously (24).
Crystallization and Data Collection-The protein was concentrated to 3-5 mg/ml in a buffer containing 50 mM Tris-HCl, pH 7.6, 250 mM NaCl, 2% glycerol, and 100 mM non-detergent Sulfobetaine 195 as a solubilizing agent (25). Crystals were obtained at 20°C in hanging drops containing 0.2-0.4% agarose, 2-3 l of the protein solution, and 2 l of the reservoir solution (28 -41% polyethylene glycol (PEG) 4000, 100 mM Tris-HCl, pH 7.0, 50 mM CaCl 2 , 10 mM ␤-mercaptoethanol). Several native and derivative data sets in various space groups were collected using different European synchrotron radiation facility beamlines. The native data set used to refine the structure (Table I) was collected on the D2AM beamline of the European synchrotron radiation facility (26) in 1998 using an in-house built CCD detector (27).
A platinum derivative was obtained by introducing 0.5 mM K 2 PtCl 4 into the drop containing the crystals. Diffraction data were collected on this soaked crystal on the BM30 beamline of the European synchrotron radiation facility up to 2.6-Å resolution around the platinum LIII absorption edge (1.064 -1.072 Å). Integration of this data set indicated that the crystal had a triclinic P1 space group (a ϭ 48.33 Å, b ϭ 48.42 Å, c ϭ 88.57 Å, ␣ ϭ 91.74°, ␤ ϭ 92.70°, ␥ ϭ 113.54°), with satisfactory overall statistics (R sym ϭ 0.04; 92% completeness) but with only 60% completeness of the anomalous signal.
Structure Determination and Refinement-The structure was solved by molecular replacement using AMoRe (28) in two steps using the multiple sequence alignment of various C1q modules as a guide. First, the ACRP-30 structure (20) was used as a model to solve the platinum derivative in the P1 space group, with two trimers per asymmetric unit. This derivative, with two platinum sites per trimer, allowed us to distinguish the three, A, B, C, chains; indeed, it appeared that each of the two platinum sites was close to a methionine residue (Met A104 and Met B122 ), which in each case was present in only one of the three homologous chains (see Fig. 2B). The model was reduced to its core structure (where the electron density was clear) and then refined carefully using CNS (29) starting from rigid body refinement of the individual position of each chain followed by a round of simulated annealing and minimization using strong non-crystallographic restraints on the corresponding A, B, C chains. This model was then used to solve the structure of the highest resolution C2 space group data set by molecular replacement, with only one trimer per asymmetric unit ( Table I). The models of the ACRP-30 and collagen X structures were superimposed on this minimal model to help tracing and building the complete C1q globular head structure. The automated refinement procedure (30) was used to improve the quality of the maps and to reduce the model bias. Model building was easily carried out into very clear electron density maps using program O (31). Refinement was done using CNS (29) except for the very last steps, which were performed using REFMAC (32) after the introduction of water molecules and of alternative conformations for some amino acids in the model. The following residues have a disordered side chain and have been modeled as Ala: Gln A90 , Arg A92 , Gln A160 , Thr B92 , Gln B93 , Arg B108 , Arg B109 , Arg B150 , Arg B163 , Gln B165 , Lys C89 . Atomic coordinates have been deposited in the Protein Data bank with accession code 1PK6.
C1q Modeling-The C1q collagen-like stem model is based on published statistical information derived from collagen-like structures (33). The relative positioning of the collagen triplets of the A, B, and C chains in the triple helix is the only one compatible with the N-terminal ends of the present globular domain structure. The conformational parameters for amino-rich regions (33) were used for the modeling of segments A40-A48, A55-A60, and A78-A87 and of the corresponding segments B42-B50, B57-B62, B80-B89, C39-C47, C54-C59, and C77-C86. The conformational parameters for imino-rich regions were used for the other segments. The globular domains were equally spaced in a circle with a radius of ϳ100 Å (34). The collagen-like arms were oriented in a symmetrical convergent arrangement toward the center and rotated in such a way as to position the 36 -39-insertion segment of the A chain at the exterior of the kink structure (34).
Modeling the C1q-IgG1 and C1q-CRP Interactions-Models of the C1q globular domain interacting with two of its physiological targets with known x-ray structures, human C-reactive protein (35) and human IgG b12 (36,37), were constructed as follows. In each case the mutagenesis data delimiting the C1q binding site (see "Discussion") were taken into account to select the most plausible model. In addition, the position of the C1q globular domain structure was restrained to orientations where the collagen arms had no steric clashes with either the targets or the underlying surface. To take into account these various constraints, the two structures were first manually positioned using program O (31). Subsequently, the automatic protein-protein docking program Hex (38,39) was used to search for solutions in a more exhaustive and objective way. Hex calculates interaction energies that include a hydrophobic excluded volume model derived from the notion of overlapping surface skins with or without soft electrostatic potential complementarity. Bad contacts at the main-chain level are checked by the program. Because most of the solutions obtained from a free rotation of the C1q globular domain were not compatible with the above selection criteria, the search was restricted to the ligand orbit protocol.
The initial C1q-CRP model was built manually by positioning residues Lys A200 , Tyr B175 , and Lys C170 from the top of the C1q head directed toward CRP residues Asp C112 , Tyr E175 , and Asp B112 , respectively. In this configuration, several additional interactions occur between CRP and the periphery of the C1q head. This solution was slightly improved by the docking program. Alternative solutions found by this program were rejected either because of steric conflict between proline side chains or because the resulting position of the collagen arms was incorrect. Moreover, in these alternative models the C1q globular head was too far from the CRP residues experimentally found to contact C1q (40). Because CRP is supposed to have a proper 5-fold axis of symmetry, the C1q head could in principle interact in one of five equivalent orientations, i.e. C1q Tyr B175 could interact with Tyr 175 of the A, B, C, D, or E CRP subunit. These supposedly equivalent configurations were generated using program O and assessed with the docking program Hex. Indeed, as shown in Table II, the corresponding solutions are not equivalent because the CRP structure actually only has pseudo 5-fold symmetry and because these solutions are under severe restricting steric constraints. The solution displayed on Fig. 4 is the top one in Table II, with Tyr 175 of CRP subunits A and D at hydrogen-bond distance from C1q residues Tyr B175 and Lys A200 , respectively, and CRP Tyr E175 at 4 Å from C1q Trp A147 .
Interaction with IgG b12 was constrained by the location of the C1q binding site in human IgG1 as defined by mutagenesis data (9,41,42). The interaction between C1q and IgG is known to involve a major ionic component (7), and Lys 322 in hIgG1 has been identified by several groups as a key residue engaged in an ionic interaction with C1q (9,42). With respect to C1q, the binding site involves Arg residues but no Lys residue (7,43), and Arg A162 , Arg B114 , Arg B129 , Arg B163 , and Arg C156 have been identified as possible interaction sites (44). Analysis of the C1q globular head structure reveals that both Arg A162 and Arg C156 are already engaged in internal salt bridges with Asp A191 and Glu C187 , respectively, and are therefore unlikely to be available for proteinprotein interaction. Other studies based on expression of the individual A, B, and C modules of C1q indicate that, although both modules A and B show significant binding to IgG, only the latter has marked binding selectivity for IgG relative to IgM (45)(46)(47). Taken together the above information led us to the working hypothesis that most of the C1q residues involved in IgG recognition are contributed by module B. The initial model was built manually by positioning the IgG residues Asp 270 and Lys 322 facing C1q residues Arg B129 and Glu B162 , respectively. A cluster of similar models (within 3-Å r.m.s.d. from each other) was obtained with the docking program, and no other alternative solution was found that meets the selection criteria mentioned above. In the proposed model(s), additional ionic interactions possibly form between C1q residues Arg B114 and Arg B161 and IgG residues Glu M195 and Glu 287 , and several hydrophobic residues of C1q (Ile B103 , Val B105 ) and IgG (Leu M154 , Pro M204 ) show a decreased access to the solvent in the assembly. The IgG residues Glu 333 and Lys 326 restrict the access of C1q Glu B162 and Arg B129 , respectively, and improved values are obtained when these two residues are converted to Ala in the computation (Table  II). Although these computed differences are exaggerated because the two models are arbitrarily kept rigid to simplify the modeling process, they are coherent with the observed effects of the corresponding mutations (see "Discussion").

RESULTS
Overall Structure-The C-terminal globular domain of C1q was obtained after digestion of the collagenous part of the protein with collagenase, treated with sialidase, and purified by ion-exchange chromatography as described under "Experimental Procedures." N-terminal sequence analysis of the purified material after separation of the three chains by SDS-PAGE yielded the following sequences: Gly-Asn-Ile-Lys-Asp-Gln (A chain), Gly-Pro-(OH)Lys-Gly-Glu-Ser (B chain), and Gly-Glu-Pro-Gly-Glu-Glu (C chain). Analysis by mass spectrometry yielded three major peaks with mass values of 17,339 Ϯ 20 Da (A chain), 16,812 Ϯ 20 Da (B chain), and 15,600 Ϯ 20 Da (C chain). Both analyses were consistent with each other and indicated that the purified material comprised residues Gly 85 -Ala 223 of the A chain, Gly 81 -Ala 226 of the B chain, and Gly 78 -Asp 217 of the C chain.
The crystal structure of the C1q globular domain was solved by molecular replacement and refined to 1.9-Å resolution. The final R work and R free factors are 0.199 and 0.238, respectively, and the refined model has excellent stereochemistry (Table I).
Residues Gln A90 to Ser A222 , Thr B92 to Asp B223 , and Lys C89 to Asp C217 show clear and continuous electron densities, with only a few disordered side chains (see "Experimental Procedures"). The N-terminal collagen-like triplets not digested by collagenase are absent from the electron density map and, consequently, were not included in the model. The structure reveals a tight heterotrimeric assembly with non-crystallographic pseudo-3-fold symmetry, the subunits arranged clockwise in the order A, B, C when viewed from the top (Fig. 1A). The assembly exhibits a globular, almost spherical structure with a diameter of about 50 Å (Fig. 1B). As observed in the case of the ACRP-30 and collagen X homotrimers (20,21), the N and C termini of the three subunits emerge at the base of the trimer. A further feature reminiscent of the collagen X structure (21) is the presence of a Ca 2ϩ ion bound to the apical side of the trimer (Fig. 1B).
The subunit structure shows a 10-stranded ␤ sandwich with a jellyroll topology homologous to the one described initially for tumor necrosis factor (48,49), consisting of two five-stranded ␤-sheets (AЈ, A, H, C, F) and (BЈ, B, G, D, E), each made of anti-parallel strands ( Fig. 2A). Compared with each other the C1q subunits show r.m.s.d. values of 0.73-0.94 Å, based on their overall structures, and of only 0.56 -0.71 Å, based on the ␤-strands. These comparisons indicate strong conservation of the latter and significant variability in the loops, particularly A-AЈ and G-H on the apical side ( Fig. 2A). Compared with ACRP-30 and collagen X, the ␤-sheets of the C1q modules show r.m.s.d. values of 0.59 -0.70 Å, indicating strong structural homology within the gC1q family. The free cysteines homologous to those found in ACRP-30 and collagen X (Fig. 2B) are essentially buried in the structure, consistent with the fact that they are not alkylated, despite treatment of the protein with iodoacetamide (see "Experimental Procedures"). A specific feature of the C1q modules is that they contain two extra cysteines that form a disulfide bond (Cys 150 -Cys 168 in module A) (Fig.  2B). These cysteines are conserved in the sequences of the C1q chains from other species, except for Cys 168 , which is missing in the mouse C1q A chain (50). In agreement with previous studies (51,52), a carbohydrate chain is attached to Asn A124 in the BЈ-B loop, as shown by the observed partial electron density corresponding to the proximal two N-acetylglucosamine residues.
The Heterotrimeric Assembly-The ␤-sheet formed by the BЈ, B, G, D, and E strands and loop A-AЈ form the exterior of the trimer, whereas the second ␤-sheet (strands AЈ, A, H, C, F), loops E-F, and G-H are essentially buried and, together with the C-terminal half of strand E, account for most of the interfaces between the three modules (Fig. 1A). The heterotrimer involves a very tight association of the modules, with a total buried surface of 5490 Å 2 equally contributed by each module and corresponding for the most part (70%) to non-polar interactions. Going from the top to the base, the central interface involves a series of interactions distributed along the 3-fold axis of the trimer. (i) On the top, loops E-F from the three modules are connected through a network of hydrogen bonds involving water molecules. (ii) A Ca 2ϩ ion is coordinated by six oxygen ligands contributed by one of the side-chain oxygens of Asp B172 , the side-chain carbonyls of Gln A177 and Gln B179 , the main-chain carbonyl of Tyr B173 , and two water molecules, with an average bond distance of 2.58 Å (Fig. 1C). The nearby carboxyl group of Asp A169 at 5.6 Å may provide electrostatic compensation. The Ca 2ϩ binding site is therefore asymmetrical relative to the trimer, since Ca 2ϩ bridges strand F of module A to strands E and F of module B but is not connected to module C. Whereas the Ca 2ϩ cluster observed in collagen X is buried (21), the single Ca 2ϩ ion of C1q is well exposed to the solvent and defines the upper entrance of a discontinuous central channel. (iii) Main-chain polar interactions take place between residues of loops E-F before the channel closes at the level of Ser A180 , Thr B182 , and Ser C176 . (iv) In the central part of the assembly a pocket is formed containing several water mole-cules bridging residues mainly located on strands C and F through a network of hydrogen bonds. (v) Tyr A138 , Tyr B142 , and Tyr C139 are found at the boundary between the upper, hydrophilic and the lower, hydrophobic parts of the central interface, the latter mainly involving residues from the C-terminal half of strand H. In addition to this central interface, a number of lateral contacts take place between the three modules. Again, these interactions are hydrophobic near the base and become more polar toward the top, with a number of hydrogen bonds and two intermodular salt bridges, Glu B209 -Lys C160 and Asp A211 -Lys B166 .
The heterotrimeric C1q assembly has many characteristics in common with the collagen X homotrimer (21), notably the different nature of the upper and lower halves of the structure. However, comparison of the overall ␤-sheet structures in the C1q and collagen X trimers shows a r.m.s.d of 0.78 Å, indicating subtle differences in the relative positioning of the three modules in the two structures. Tentative modeling of homotrimeric C1q assemblies clearly results in a loss of shape complementarity and in destabilization of the lateral salt bridges. Amino acid sequence comparisons (Fig. 2B) reveal that residues conserved in the gC1q modules of C1q, ACRP-30 and collagen X either belong to the hydrophobic core or participate in hydrophobic interactions in the lower half of the trimers, but they indicate substantial variability in the residues found at the interfaces.
Surface Properties-The three C1q modules exhibit marked differences in their electrostatic surface potentials. Thus, the pseudo 3-fold symmetry seen at the scaffold level disappears when the charge distribution at the C1q surface is considered. Module A (Fig. 3A) shows a combination of arginine and acidic residues scattered on its external face. The top part of the module is also markedly charged, with a predominance of lysine residues. The surface of module B (Fig. 3B) shows a net predominance of positive charges with, in particular, a contin-uous patch of three basic residues, Arg B101 , Arg B114 , Arg B129 , flanked by Asp B116 and Glu B127 . Arg B114 and Arg B129 as well as Arg B163 on the opposite edge of module B have all been proposed to be involved in the interaction with IgG (44). Whereas Arg B163 is disordered, Arg B129 and Arg B114 markedly protrude outside the structure (Fig. 3, B and C) and are involved in crystal contacts with Glu B162 . Several arginine residues are also present on the top. Module C (Fig. 3C) shows a combination of basic and acidic residues scattered on the surface, with Lys C170 protruding on the top.
Several hydrophobic residues are exposed on the external side of each subunit, with most of them located in concave areas of the structure. Notable exceptions to this include Pro A103 , Met A104 , Pro A199 (Fig. 3A), Ala C105 , Pro C106 , and Val C185 (Fig.  3C) as well as a hydrophobic cluster (Ile B103 , Val B105 , Pro B106 ) extending over the Arg B101 , Arg B114 , Arg B129 triad in the upper part of module B (Fig. 3B). Another hydrophobic patch, Val B118 , Ile B119 , Pro B128 , lies in between the charged cluster of module B and the B-C interface. Only module C displays solvent-accessible aromatic residues (Tyr C155 , Trp C190 ) on its equatorial side (Fig. 3C). The C1q structure does not exhibit the strip of solvent-exposed aromatic residues extending across each subunit interface found in collagen X (21).
The top of the heterotrimer (Fig. 3D) reveals a striking predominance of positive charges mainly contributed by lysine residues interspersed with a few acidic residues. In addition the top exhibits several exposed hydrophobic patches with, in particular, a remarkable continuous stretch (Ala C105 , Pro C106 ,  Fig. 1. The free cysteines and the disulfide bonds of subunits A, B and C are superimposed, and only the cysteines of subunit C are displayed for clarity. B, structure-based sequence alignment of the gC1q domains of C1q, ACRP-30, and collagen X. The residue numbering (A, B, C, from top to bottom) and the secondary structure elements shown are those of the C1q chains. Residues in italics have a disordered side chain and have been modeled as Ala. C-terminal residues Ala A223 , Met B224 , Glu B225 , and Ala B226 have no matching electron density. Residues involved in the C1q interfaces are highlighted in orange. Ca 2ϩ binding ligands in C1q and collagen X are in red. Asterisks (*) indicate residues conserved in C1q, collagen X, and ACRP-30. The free Cys residues are indicated by SH, and the carbohydrate attachment site in subunit A of C1q is indicated by CHO.

DISCUSSION
The three new gC1q module x-ray structures described here and those determined previously for ACRP-30 and collagen X (20,21) confirm the generality of the topology identified initially from these latter structures. Remarkably, the trimeric globular domains assemble in similar ways in C1q, ACRP-30, and collagen X despite the fact that interactions are achieved by three identical modules in the latter two proteins but by three different modules in C1q. In this respect, tentative assembly in silico of C1q homotrimers reveals in all cases a number of severe steric hindrances, especially at the level of lateral contacts, for example, in the A-A-A pseudo homotrimer, where Met A183 and Leu A165 clash with Ile A115 and Ile A98 , respectively. This provides a structural basis for the specific property of the C1q subunits to associate only as heterotrimers (4). Interestingly, the highest degree of conservation between C1q and its homotrimeric homologues among interface residues is seen at the level of the hydrophobic interfaces near the base of the trimers, suggesting a critical role for this region in the alignment of the three chains in the globular and collagen regions on either sides. In this respect, the C1q structure provides information that is essential to model the collagen-like triple helix on the N-terminal side of the globular domain (see "Experimental Procedures"). The overall similarity between C1q and collagen X along the interface also suggests a common assembly process driven by hydrophobic interactions near the base and locked by hydrophilic contacts toward the top. Interestingly, the total buried surface in C1q (5490 Å 2 ) is similar to that of ACRP-30 (5320 Å 2 ) but considerably less than that of collagen X (7360 Å 2 ), confirming that the exceptional stability of the collagen X NC1 trimer arises from a particularly compact structure compared with other C1q-like proteins (21).
A further interesting feature of the C1q structure is the presence of a Ca 2ϩ binding site at the top of the assembly. This finding is consistent with the fact that crystallization was performed in the presence of Ca 2ϩ ions and accounts for previous data (53), indicating that Ca 2ϩ binding is an intrinsic property of the C1q molecule. Whether this site only contributes to the stability of the heterotrimeric assembly or has also a functional role remains to be determined. In this respect it is noteworthy that Ca 2ϩ is coordinated by six ligands, among which is a single carboxyl group. Although these characteristics are shared by many of the known protein Ca 2ϩ binding sites (54), the observation that the Ca 2ϩ ion is fully accessible to the solvent (Fig. 1C) opens the possibility that some of the charged targets recognized by C1q may interact directly with the Ca 2ϩ ion by displacing one or both of the water molecules (Fig. 1C). It is also noteworthy that the residues involved in Ca 2ϩ binding in C1q are homologous to their counterparts in collagen X. Because the latter define a consensus sequence that is also found in ACRP-30, it appears likely, therefore, that the partially disordered structure observed at the top of ACRP-30 (20) arises from the absence of Ca 2ϩ in the structure.
The heterotrimeric structure of the C1q globular head has direct implications in generating the versatile recognition properties of this protein. Each of the three subunits exhibits particular surface patterns in terms of charged and hydrophobic residues and may, therefore, be expected to display specific individual recognition properties. In addition, the compact trimeric structure of the C1q head clearly allows ligand recognition through residues contributed by two or even three subunits, thereby broadening the recognition spectrum of C1q. This may explain in part the observation that many of the C1q ligands exhibit significant interaction with several of the three subunits of its globular domain (47). The diversity of the recognition modes of C1q is illustrated below by the proposed models of interaction between the top of the C1q head and CRP (Fig. 4) and between the lateral side of subunit B and human IgG1 (Fig. 5).
CRP is a major acute phase plasma protein in man that binds to phosphocholine head groups of membrane phospholipids and is in turn recognized by C1q (13). Its crystal structure (35,55) shows a pentraxin fold. The C1q binding site identified by mutagenesis experiments (40) lies close to the central pore of its pentameric structure, on the face opposite to that attached to the phosphocholine-bearing surface. As illustrated on Fig. 4, the top of the C1q head structure, predominantly basic, can be accommodated by the negatively charged central pore of CRP, with a striking shape complementarity between the two proteins. Asp 112 and Tyr 175 have been identified as major C1q contact residues (40). The model shown in Fig. 4 allows these residues from CRP subunits A and E to come into direct contact with appropriate residues from C1q subunits B and A (see "Experimental Procedures" for details). The enhanced complement activation observed after mutation to Ala of CRP Lys 114 may be due to its interference with the C1q lysine residues directed toward the negatively charged residues of the pore (Fig. 4). Although the access of the top of C1q into the pore of the phosphocholine-bound CRP structure is under severe steric restraints in the proposed model (see "Experimental Procedures"), it may be expected to be more easily accommodated after a slight conformational change in the CRP structure in a physiological context (55).
Interaction with IgG, on the other hand, provides an example of a recognition mediated by the equatorial region of a single subunit of the C1q globular head. The macroscopic model of C1q shown in Fig. 5C suggests that the B subunit is found at the outer portion of the C1q molecule, a location that appears particularly well adapted for IgG recognition in the context of immune complexes. The most plausible model(s) is obtained by positioning the two molecules in such a way that Asp 270 and Lys 322 of IgG form salt bridges with Arg B129 and Glu B162 of C1q, respectively (Fig. 5A). In this configuration, there is remarkable shape complementarity between C1q and IgG (Fig.  5B), with Arg B129 acting like a wedge in between the CH2 and CL domains. This model is fully consistent with (i) the physicochemical characteristics of the ionic component of the C1q-IgG interaction (7), (ii) the mutagenesis experiments locating the C1q binding epicenter in human IgG1 around residues Asp 270 , Lys 322 , Pro 329 , Pro 331 (9,41,42), and (iii) the fact that the interaction involves two arginine residues of the C1q head (Arg B114 , Arg B129 ) proposed to mediate IgG recognition (44). The location of Tyr 278 at the interface with C1q (Fig. 5B) is also consistent with the observation that nitration of the tyrosine residues of human IgG abolishes its C1q binding activity (56). The fact that, in this model IgG1 Glu 333 and Lys 326 severely interfere with the access of C1q Glu B162 and Arg B129 , respectively, is consistent with the observations that a double mutation E333S/K326W increases the C1q binding ability of an IgG1 molecule and confers a non reactive IgG2 molecule the ability to bind C1q (41). Opening the way to further analysis of the C1q-IgG interaction by site-directed mutagenesis, the proposed model has several major implications in terms of IgG recognition by C1q. First, it raises the possibility that C1q binds not only to the Fc region but also to the Fab arm of IgG through the interaction with the CL domain, as illustrated in Fig. 5B.
Because the IgG b12 structure shows extreme interdomain flexibility (37), the latter binding interactions may be variable and possibly do not apply to all IgG subtypes. Nevertheless, the Fab/Fc orientation is clearly a critical factor of the recognition because it will condition access of the C1q globular head to the major target site in the CH2 domain. This is illustrated in Fig.  5D by the structure of Mcg, a human IgG1 with a deleted hinge that is unable to interact with C1q (57) because the Fab arm obstructs the C1q binding site. A further illustration of the critical role of the hinge region is that a double mutation L234A/L235A in IgG1 b12 abolishes C1q binding ability (42). These residues form a hydrophobic cluster at the Fab/Fc hinge (Fig. 5B) and are, therefore, expected to condition the relative positioning of the Fab arm. The observation that, unlike its isolated Fc region, human IgG4 does not recognize C1q to a significant extent (58) may also lead to a similar structural interpretation. A further interesting feature of our model is that it places the C1q globular head in the vicinity of the antigenic sites (Fig. 5C), raising the possibility of direct additional contacts between C1q and the antigen itself.
In summary, as exemplified above by the models proposed for CRP and IgG, the x-ray structure of the globular domain of C1q provides a structural basis for the versatility of its recognition properties and opens the way to decipher its interaction with many of its immune and non-immune ligands.