Crystal Structure of the CUB1-EGF-CUB2 Domain of Human MASP-1/3 and Identification of Its Interaction Sites with Mannan-binding Lectin and Ficolins*

MASP-1 and MASP-3 are homologous proteases arising from alternative splicing of the MASP1/3 gene. They include an identical CUB1-EGF-CUB2-CCP1-CCP2 module array prolonged by different serine protease domains at the C-terminal end. The x-ray structure of the CUB1-EGF-CUB2 domain of human MASP-1/3, responsible for interaction of MASP-1 and -3 with their partner proteins mannan-binding lectin (MBL) and ficolins, was solved to a resolution of 2.3Å. The structure shows a head-to-tail homodimer mainly stabilized by hydrophobic interactions between the CUB1 module of one monomer and the epidermal growth factor (EGF) module of its counterpart. A Ca2+ ion bound primarily to both EGF modules stabilizes the intra- and inter-monomer CUB1-EGF interfaces. Additional Ca2+ ions are bound to each CUB1 and CUB2 module through six ligands contributed by Glu49, Asp57, Asp102, and Ser104 (CUB1) and their counterparts Glu216, Asp226, Asp263, and Ser265 (CUB2), plus one and two water molecules, respectively. To identify the residues involved in interaction of MASP-1 and -3 with MBL and L- and H-ficolins, 27 point mutants of human MASP-3 were generated, and their binding properties were analyzed using surface plasmon resonance spectroscopy. These mutations map two homologous binding sites contributed by modules CUB1 and CUB2, located in close vicinity of their Ca2+-binding sites and stabilized by the Ca2+ ion. This information allows us to propose a model of the MBL-MASP-1/3 interaction, involving a major electrostatic interaction between two acidic Ca2+ ligands of MASP-1/3 and a conserved lysine of MBL. Based on these and other data, a schematic model of a MBL·MASP complex is proposed.

The lectin pathway of complement is increasingly recognized as an important component of innate immunity against pathogens. This pathway is triggered by oligomeric lectins that recognize patterns of neutral and acetylated carbohydrates on the surface of pathogens and share the ability to associate with and trigger activation of modular proteases termed mannanbinding lectin-associated serine proteases (MASPs) 3 (1,2). Four such oligomeric lectins have been described: mannanbinding lectin (MBL) and ficolins L, H, and M (3)(4)(5)(6)(7). These proteins all assemble as oligomers of homotrimeric subunits, each comprising N-terminal collagen-like triple helices prolonged by recognition domains endowed with lectin-like binding activities. There are three different MASPs (MASP-1, -2, and -3) (4,8,9), and these feature modular structures homologous to those of C1r and C1s, the proteases of the C1 complex of complement, with an N-terminal CUB module (10), an epidermal growth factor (EGF)-like module belonging to the Ca 2ϩbinding subset (11), a second CUB module, two complement control protein (CCP) modules (12), and a chymotrypsin-like serine protease domain. MASP-1 and MASP-3 are alternative splicing products of the MASP1/3 gene and include different serine protease domains but share identical CUB 1 -EGF-CUB 2 -CCP 1 -CCP 2 segments (9). Likewise, alternative splicing of the MASP2 gene generates MBL-associated protein 19 (MAp19), a truncated protein comprising the N-terminal CUB 1 -EGF segment of MASP-2 prolonged by four residues specific to MAp19 (13,14). From a functional standpoint, the ability of MASP-2 to trigger the lectin pathway of complement is clearly established (8). In contrast, whether MASP-1 is involved in this pathway is still a controversial issue (15,16), and the function of MASP-3 remains elusive.
Studies on human (17)(18)(19)(20) and rat (21,22) proteins have established that the MASPs and MAp19 each associate as homodimers through their CUB 1 -EGF segment. In turn, the MASPs and MAp19 each form individual Ca 2ϩ -dependent complexes with MBL and the ficolins. The interaction involves a major site located in the CUB 1 -EGF moiety of each protein but is strengthened by module CUB 2 (18,19,22,23). Resolution of the x-ray structure of human MAp19 has allowed identification of a Ca 2ϩ -binding site in module CUB 1 , and site-directed mutagenesis has provided evidence that this underlies the binding site for MBL and L-ficolin (20). The x-ray structure of the CUB 1 -EGF-CUB 2 segment of rat MASP-2 was also solved, but in contrast no Ca 2ϩ ion could be observed in the CUB modules, and therefore a different model for interaction with MBL was proposed (22). On the other hand, expression of site-directed mutants of human and rat MBL (24,25) and ficolins (26,27) has recently provided evidence that these proteins associate with the MASPs and MAp19 through a major ionic interaction involving a conserved lysine residue.
We now report the crystal structure of the CUB 1 -EGF-CUB 2 domain of human MASP-1/3. The homodimeric structure is similar to that described for the homologous rat MASP-2 domain but holds a Ca 2ϩ -binding site in each CUB 1 and CUB 2 module. In line with our previous studies on MAp19 (20), we provide evidence that each of these sites is involved in the interaction with MBL and the ficolins and propose a schematic model of a MBL⅐MASP complex.

EXPERIMENTAL PROCEDURES
Proteins-MBL was purified from human serum as described by Zundel et al. (19). The CUB 1 -EGF-CUB 2 segment of human MASP-1/3 was expressed in a baculovirus/insect cell system and purified as described previously (18). Full-length wild-type MASP-3 and its variants were expressed in a baculovirus/insect cell system (19). All mutants were secreted in the culture medium, with yields of 8 -10 g/ml, similar to those obtained for the wild-type form, except E216A and D226A, which were produced at 1 and 5 g/ml, respectively. All MASP-3 variants were purified by ion-exchange chromatography on a Q-Sepharose Fast Flow column (GE Healthcare) followed by high pressure gel permeation on a TSK G3000 SWG column (Toso Haas), as described previously (25). This latter step allowed us to check that all MASP-3 variants retained the ability to associate as homodimers. The concentrations of purified proteins were determined using absorption coefficients (A 1% , 1 cm at 280 nm) calculated by the method of Gill and von Hippel (28) and molecular weights determined by mass spectrometry, as follows: MASP-1/3 CUB 1 -EGF-CUB 2 , 10.0 and 34,200 (18); MBL, 7.6 and 25,340 (29); L-ficolin, 17.6 and 33,800; H-ficolin, 19.6 and 32,800; MASP-3 variants, 13.5 and 87,500, except Y56A and Y225A (13.3 and 87,500).
Expression and Purification of H-ficolin-A DNA segment encoding the H-ficolin signal peptide plus the mature protein (amino acid residues 1-276) was amplified by PCR using the pGEM-T easy plasmid containing the full-length cDNA (30) as a template, according to established procedures. The amplified H-ficolin DNA was cloned into the pcDNA3.1(ϩ) mammalian expression plasmid (Invitrogen). A stable CHO K1 cell line expressing H-ficolin was created using Geneticin (G418 sulfate, Invitrogen), and the recombinant protein was produced in serum-free medium as described by Teillet et al. (25). Recombinant H-ficolin was purified from the culture supernatant by ion-exchange chromatography on a Q-Sepharose Fast Flow column (GE Healthcare) equilibrated in 5 mM CaCl 2 , 50 mM triethanolamine-HCl, pH 8.0, using a linear NaCl gradient to 250 mM. Fractions containing the recombinant protein were identified by Western blot analysis, dialyzed against 145 mM NaCl, 5 mM CaCl 2 , 20 mM Tris-HCl, pH 7.4, and further puri-fied by gel filtration on a Superose 6 column (GE Healthcare) equilibrated in the same buffer.
Purification of L-ficolin-L-ficolin was purified from human plasma as described by Krarup et al. (31) with modifications of the affinity chromatography step, as follows. The 4 -8% PEG pellet was dissolved in 145 mM NaCl, 5 mM CaCl 2 , 1.5 mM NaN 3 , 10 mM Tris-HCl, 0.01% (v/v) emulfogen (Sigma), pH 7.4, and loaded onto an N-acetylcysteine-Sepharose column equilibrated in the same buffer. After washing with (i) the loading buffer, (ii) the loading buffer containing 1 M NaCl and 10 mM EDTA instead of CaCl 2 , and (iii) the loading buffer containing 20 mM NaCl and 10 mM EDTA, bound L-ficolin was eluted with 200 mM NaCl, 0.3 M N-acetylglucosamine, 50 mM Tris-HCl, pH 7.8. The eluted proteins were dialyzed against 50 mM NaCl, 10 mM CaCl 2 , 20 mM Tris-HCl, pH 7.8, before further purification by anion-exchange chromatography.
Site-directed Mutagenesis-The expression plasmids coding for all MASP-3 mutants were generated using the QuikChange TM XL site-directed mutagenesis kit (Stratagene, La Jolla, CA) according to the manufacturer's protocol. The pFastBac1/MASP-3 expression plasmid (19) was used as a template. Mutagenic oligonucleotides were purchased from MWG-Biotec (Courtaboeuf, France). The sequences of all mutants were checked by double-stranded DNA sequencing (Genome Express, Grenoble, France).
Mass Spectrometry-Analyses were performed using the matrix-assisted laser desorption ionization technique on a Voyager Elite XL instrument (PerSeptive Biosystems, Cambridge, MA) under conditions described previously (32).
Crystallization and Data Collection-The MASP-1/3 CUB 1 -EGF-CUB 2 segment was concentrated to 3.6 mg/ml in 100 mM nondetergent sulfobetain 195, 145 mM NaCl, 1 mM CaCl 2 , 50 mM triethanolamine-HCl, pH 7.4. Crystals suitable for x-ray diffraction data collection were obtained at 20°C by the hanging drop vapor diffusion method by mixing equal volumes of the protein solution and of a reservoir solution composed of 18 -20% (w/v) PEG 8000, 3% glycerol, and 0.1 M Hepes, pH 7.0. They were transferred to a cryoprotecting solution containing 22% (w/v) PEG 8000, 20% glycerol, and 0.1 M Hepes, pH 7.0, and then flash-cooled in liquid nitrogen. A native data set indexed in the space group C222 1 was measured at the ESRF beamline ID14-eh1 to a resolution of 2.30 Å. The images were processed and the reflections scaled using the XDS program (33). Crystallographic statistics for the native data set are given in Table 1.
Structure Determination and Refinement-The structure of CUB 1 -EGF-CUB 2 was determined using the molecular replacement method. The rotational and translational searches were carried out using the program AMoRe (34). Although initial searches using the human Map19 CUB 1 -EGF structure (20) were unsuccessful, a clear solution was obtained using the rat MASP-2 CUB 1 -EGF-CUB 2 structure in its dimeric form (Protein Data Bank accession code 1NT0) (22). Rigid body refinement with the program CNS (35) was used to further improve the orientation and position of each domain of the search model. Several replacements corresponding to the MASP1/3 sequence were clearly seen in the electron density map and introduced in the model using the graphics program O (36). However, the quality of the map remained unsatisfactory in one of the CUB 1 modules, because of a positioning error of its ␤-barrel, each strand being shifted onto the next one. The correct module orientation was obtained using strong noncrystallographic restraints in the model refinement carried out with CNS (35). This essential improvement allowed 489 residues (93% of the model) to be built automatically into the electron density map after two rounds of the ARP-Warp procedure (37), providing clear indication of the quality of the map obtained at this stage. A new model, free from any bias arising from the initial molecular replacement model, was thus obtained. Additional model rebuilding and refinement were then carried out. The final stages of the refinement were performed with Ref-mac5 (38), using TLS refinement. The quality of the map allowed construction of 530 residues out of a total of 554 in the asymmetric unit. Proline residues at positions 17, 183, and 238 are in the cis conformation. The atomic coordinates have been deposited in the Protein Data Bank under the code 3DEM.
Modeling of the Interaction between MASP-1/3 CUB 1 -EGF-CUB 2 and MBL-The homotrimeric collagen-like segment of human MBL containing the putative MASP interaction site (25) was modeled on the basis of published statistical information derived from collagen-like structures, as described previously (20). Four MBL collagen-like segments were docked manually onto the MASP-1/3 CUB 1 -EGF-CUB 2 domain using the interactive graphics program O (36), with at least one of the three Lys 55 residues of each segment pointing toward a CUB 1 -EGF-CUB 2 -binding site. The fact that the MBL collagen-like fibers converge to the same point on their N-terminal side was used as an additional constraint. The x-ray structure of the zymogen CCP 1 -CCP 2 -SP segment of human MASP-2 (39) was used as a template for constructing the schematic MBL/MASP model depicted in Fig. 7.
Surface Plasmon Resonance Spectroscopy-Analyses were performed using a BIAcore 3000 instrument (GE Healthcare). MBL and ficolins L and H were immobilized on the surface of a CM5 sensor chip (GE Healthcare) using the amine coupling chemistry as described previously (18). Binding of the MASP-3 variants was measured over 19,300 -23,000 resonance units of immobilized L-ficolin, 19,600 -21,900 resonance units of H-ficolin, and 7,800 -8,000 resonance units of MBL, at a flow rate of 20 l/min in 145 mM NaCl, 1 mM CaCl 2 , 50 mM triethanolamine-HCl, pH 7.4, containing 0.005% surfactant P20 (GE Healthcare). Binding to MBL was measured in the presence of 10 mM mannose to prevent unwanted interaction between the oligomannose-type N-linked oligosaccharides of MASP-3 and the lectin domain of MBL (25). Equivalent volumes of each MASP-3 sample were injected in parallel over a surface with immobilized bovine serum albumin to serve as blank sensorgrams for subtraction of the bulk refractive index background. Regeneration of the surfaces was achieved by injection of 10 l of 1 M NaCl, 20 mM EDTA.
Data were analyzed by global fitting to a 1:1 Langmuir binding model of both the association and dissociation phases for several concentrations simultaneously, using the BIAevaluation 3.1 software (GE Healthcare). The apparent equilibrium dissociation constants (K D ) were calculated from the ratio of the dissociation and association rate con-stants (k off /k on ). Each MASP-3 variant was analyzed at six different concentrations, ranging from 10 to 250 nM.

RESULTS
Overall Structure-The N-terminal CUB 1 -EGF-CUB 2 interaction domain of human MASP-1/3 (amino acids 1-278 of the mature proteins) was produced in a baculovirus/insect cell expression system as described previously (18). Mass spectrometry analysis of the recombinant protein yielded a value of 34,238 Ϯ 27 Da, accounting for the unmodified polypeptide chain (calculated value, 31,973 Da) plus the two N-linked oligosaccharides at positions 30 and 159 (total deduced mass, 2,265 Ϯ 27 Da). The x-ray structure was solved by molecular replacement using the rat MASP-2 CUB 1 -EGF-CUB 2 structure (22) as a search model, and refined to 2.30-Å resolution. The final R work and R free factors are 0.22 and 0.24, respectively, and the refined model has good stereochemistry ( Table 1). As anticipated from previous ultracentrifugation analyses (17), the CUB 1 -EGF-CUB 2 segment associates as a Ca 2ϩ -dependent homodimer (Fig. 1). The monomers have a curved shape, with the CUB 1 , EGF, and CUB 2 modules arranged end-to-end. Each monomer contains three Ca 2ϩ ions, one at each CUB 1 -EGF interface (site I) and one bound to one end of the CUB 1 and CUB 2 modules (sites II and III, respectively). The dimer has a head-to-tail structure, involving interactions between the CUB 1 module of one monomer and the EGF module of its counterpart (Fig. 1A), i.e. a configuration corresponding to the compact dimer observed by Feinberg et al. (22). The dimer displays noncrystallographic pseudo 2-fold symmetry and has approximate dimensions 135 Å (distance between the CUB 2 modules tips) ϫ 87 Å (distance between the CUB 1 modules tips) ϫ 25 Å (mean thickness). On a side view (Fig. 1B), the overall shape of the dimer is rather flat, the six modules as well as the Ca 2ϩ ions bound to sites II and III being roughly located in the same plane.
In contrast with human MAp19 (20) which has a well structured N-terminal extremity, the N-terminal end of the protein (residues 1-6) is disordered in both monomers. In the EGF module, most of loop 10 (residues 127-131) is also disordered. Of the two N-linked oligosaccharide chains, the proximal Glc- These values indicate that there is little variation of modulemodule orientation at the CUB 1 -EGF and EGF-CUB 2 interfaces.
The Inter-monomer Interface and Ca 2ϩ -binding Site I-The interface between the two monomers involves a number of hydrophobic interactions distributed in four major hydrophobic pockets (Fig. 1A), with a total buried surface of 2100 Å 2 . (i) Around the 2-fold symmetry axis of the dimer, a central pocket is formed by residues Phe 37 and Ile 142 from both monomers. (ii) At each inter-monomer interface, a second pocket is formed by the aromatic triad Phe 9 (from CUB 1 ), Tyr 141 , and Tyr 146 (from EGF). Tyr 141 also coordinates the Ca 2ϩ ion bound to site I, hence providing a link between this site and the inter-monomer interface. (iii) At both ends of the interface, a distal pocket is formed by residues Tyr 17 , Pro 18 , His 45 (from CUB 1 ), and Phe 151 (from EGF). (iv) A fourth pocket involving residues Tyr 42 , Met 44 , and His 115 (from CUB 1 ) as well as Tyr 137 and His 139 (from EGF) lies between the distal pocket and the aromatic triad. Although the latter is not present in the C1s CUB 1 -EGF dimer (40), the four hydrophobic pockets are remarkably conserved in the human MAp19 and rat MASP-2 structures (20,22), and they involve homologous residues (Fig. 2). In MASP-1/3, three additional residues Gln 11 , Thr 110 , and Ala 118 strengthen the interactions by contributing to the intermediary, distal, and central pockets, respectively. Additional interactions between the CUB 1 and EGF modules are provided by a direct H-bond between His 139 and Tyr 42 , and by water-mediated H-bonds between Gln 11 and Cys 147 , Asp 113 and Ser 148 , and between Ile 142 and both Met 8 and Ala 118 . As illustrated in Fig. 2, the residues involved in the intermonomer interface are either conserved or substituted by similar residues in the C1r/C1s/MASP family.
The MASP-1/3 EGF module has a topology similar to that described for other EGF-like modules (11), with one major and one minor antiparallel double-stranded ␤-sheets (Fig. 1A). As observed in rat MASP-2 (22) and human C1r (41), but in contrast with human MAp19 (20), loop 10 of MASP-1/3, which contains a cluster of charged residues, is mostly disorganized (Figs. 1A and 2). The Ca 2ϩ ion bound to each EGF module (site I) is coordinated by seven oxygen ligands, including a water molecule and six ligands contributed by the EGF module, namely one of the side chain oxygens of Asp 120 and Glu 123 , the side chain carbonyl of Asn 140 , and the main chain carbonyl of Val 121 , Tyr 141 , and Gly 144 . These residues are strictly homologous to those involved in Ca 2ϩ ligation in human C1s (40) and MAp19 (20). Asn 140 lacks ␤-hydroxylation, confirming that insect cells do not achieve this modification (20). As observed in MAp19 and rat MASP-2, the water molecule involved in Ca 2ϩ coordination also forms an H-bond with Gly 36 of module CUB 1 . Additional CUB 1 -EGF interactions in each monomer are achieved by a salt bridge between Arg 38 and Glu 126 , an H-bond between Gly 36 and Val 121 , and van der Waals contacts Phe 37 -Ile 142 , providing extensive stabilization of the intermodular interface.
The CUB 1 and CUB 2 Modules and Ca 2ϩ -binding Sites II and III-As observed previously in the case of human C1s, rat MASP-2, and human MAp19, the CUB modules of human MASP-1/3 exhibit different folds compared with the canonical ␤-sandwich topologies containing two five-stranded ␤-sheets described initially for plasma spermadhesins (10). Thus, using the spermadhesin nomenclature, the CUB 2 module lacks strand ␤1, whereas CUB 1 lacks both ␤1 and ␤2. In contrast with the rat MASP-2 structure (22), CUB 2 is equally well defined as CUB 1 and does not show a higher temperature factor.
A Ca 2ϩ ion is bound to the outer end of CUB 1 (site II) and to the inner end of CUB 2 (site III) (Fig. 1). In site II, the Ca 2ϩ ion is coordinated by six oxygen ligands, namely both side chain oxygens of Asp 57 , one of the side chain oxygens of Glu 49 and Asp 102 , the main chain carbonyl oxygen of Ser 104 , and a water molecule (Fig. 3A). The latter is maintained by H-bonds with the main chain carbonyls of Ser 101 and Asp 102 , providing stabilization of the outer end of CUB 1 . Ca 2ϩ coordination in site III involves six ligands strictly homologous to those listed above (Asp 226 , Glu 216 , Asp 263 , Ser 265 , and a water molecule), plus one additional ligand contributed by a second water molecule (Fig. 3B). Both water molecules are stabilized by H-bonds with the main chain carbonyl oxygen of Ser 265 , one of them also interacting with the side chain nitrogens of Asn 268 and Asp 226 , and the other one forming two H-bonds with Asp 263 (Fig. 3).
A comparison of sites II and III with the homologous sites previously described in the CUB 1 modules of C1s (40) and MAp19 (20) reveals common features and several minor differences. (i) In each case, three acidic residues homologous to Glu 49 , Asp 57 , and Asp 102 of MASP-1/3 provide Ca 2ϩ ligands (Fig. 2). (ii) The residue homologous to Ser 104 and  Ser 265 of MASP-1/3 is a Ca 2ϩ ligand in MAp19 but not in C1s. (iii) Asn 108 of MAp19 is a Ca 2ϩ ligand, unlike its counterparts in other proteins. The latter two differences reflect some modulation of the conformation of loop L9 in these proteins. (iv) A single water molecule is involved in the Ca 2ϩ -binding site of MAp19 and in site II of MASP-1/3, whereas two molecules are present in C1s and in site III of MASP-1/3. In both sites II and III, the Ca 2ϩ ion is the central element of a network of interactions connecting together loops L3 and L9 and strands ␤5, ␤6, and ␤9, hence providing extensive stabilization of the corresponding regions of the MASP-1/3 CUB modules. As their counterparts in C1s and MAp19, Tyr 21 of CUB 1 and Tyr 187 of CUB 2 also play an important part in this network by forming H-bonds with Asp 57 , Asn 105 , and Phe 109 , and with Glu 216 , Asp 226 , and Asn 268 , respectively. Further stabilization of this part of the CUB modules is achieved by hydrophobic interactions involving Phe 99 , Ser 101 , and Gly 111 (CUB 1 ) and Phe 260 , Ser 262 , Arg 269 , and Gly 270 (CUB 2 ).
A Structured EGF-CUB 2 Interface-Detailed inspection of the structure at the EGF-CUB 2 junction reveals a number of interactions between EGF residues mainly located in strands ␤13 and ␤14, and CUB 2 residues mainly contributed by strand ␤4Ј and loop L3Ј (Fig. 4). These interactions include the follow-ing : (i) two salt bridges between Arg 163 and Glu 165 and between Asp 135 and Lys 189 ; (ii) direct H-bond between Thr 161 and Ser 190 ; (iii) four water-mediated H-bonds Tyr 153 -Pro 186 , Gly 152 -Glu 165 , and between Lys 189 and both Asp 135 and Cys 162 ; (iv) van der Waals contacts His 156 -Ser 190 and between Tyr 153 and both Val 164 and Pro 188 . Altogether, these interactions strongly stabilize the EGF-CUB 2 interface, yielding a buried surface of about 750 Å 2 at the inter-domain junction. As illustrated in Fig. 2, many of the residues involved in these interactions are conserved in the MASP family, but only a few are present in C1r and C1s as well.
Mapping of the Interaction Sites with MBL and Ficolins-To identify the residues of MASP-1/3 involved in the interaction with MBL and the ficolins, a series of recombinant MASP-3 point mutants were produced, and their binding properties were analyzed by surface plasmon resonance spectroscopy, using the MASP-3 variants as soluble ligands and immobilized MBL, L-ficolin, or H-ficolin. Residues of modules CUB 1 and CUB 2 structurally homologous to those previously found to participate in Ca 2ϩ -binding site II of human MAp19 (Glu 49 , Asp 102 , Ser 104 , Glu 216 , Asp 226 , Asp 263 , and Ser 265 ) or in its interaction with MBL and L-ficolin (Tyr 56 , Glu 80 , Phe 103 , Glu 106 , Tyr 225 , Glu 243 , Asn 264 , and Glu 267 ) (20) were initially targeted and mutated to alanine. Residues His 218 and Glu 220 , located in a solvent-exposed loop in the vicinity of Ca 2ϩ -binding site III, were also subjected to mutagenesis. As listed in Table 2 and illustrated in Fig. 5, most of these mutations significantly decreased the ability of MASP-3 to associate with MBL and ficolins L and H. Thus, in module CUB 1 , mutation D102A virtually abolished interaction with each ficolin and strongly decreased interaction with MBL. Mutations E49A, F103A, Y56A, E80A, S104A, and E106A, all decreased interaction

Structure of Interaction Domain of Human MASP-1/3
with either protein, although to varying extents. Interestingly, mutation of Ca 2ϩ ligands Glu 49 and Asp 102 to Gln and Asn, respectively, also strongly decreased their binding properties, resulting in K D values similar to those determined for mutants E49A and D102A (Table 2). In many cases, the increases in K D mainly resulted from a decrease in the k on value, but the strongest effects involved both a decrease in k on and an increase in k off .
In module CUB 2 , mutation to alanine of Asp 214 and of the three acidic residues (Glu 216 , Asp 226 , and Asp 263 ) involved in Ca 2ϩ -binding site III resulted in aggregation of the corresponding MASP-3 variants, hence precluding analysis of their interaction properties. Mutation of the fourth Ca 2ϩ -binding ligand, Ser 265 , had no such effect and decreased interaction with the ficolins without significantly altering binding to MBL. Muta-tion Y225A abolished interaction with H-ficolin and strongly inhibited interaction with MBL and L-ficolin, whereas H218A markedly decreased binding to all three proteins. Much less pronounced inhibitory effects were observed for mutations E220A, E243A, N264A, and E267A. A second round of mutations was performed to target acidic residues found to be exposed in the CUB 1 -EGF-CUB 2 structure and located either in the CUB 1 (Asp 24 , Glu 55 , and Glu 107 ), EGF (Glu 126 , Glu 128 , Asp 129 , Glu 130 , and Glu 131 ), or CUB 2 module (Asp 217 ). None of these mutations had a significant impact on the interaction of MASP-3 with either MBL or the ficolins, with K D ratios relative to wild-type MASP-3 ranging from 0.7 to 1.4 (data not shown).
As illustrated in Fig. 6, the mutations that significantly decrease or abolish interaction of MASP-3 with MBL and the ficolins pinpoint residues that are clustered in the vicinity of Ca 2ϩ -binding sites II and III, thus providing strong support for the implication of both areas of the CUB 1 -EGF-CUB 2 domain in the binding.

DISCUSSION
This study provides a second example of a CUB 1 -EGF-CUB 2 domain structure, allowing a comparison with that reported previously for rat MASP-2 by Feinberg et al. (22). First, it should be emphasized that our structure is equivalent to the compact conformation observed by these authors, confirming that, of the two distinct dimers present in their crystal lattice, this one does correspond to the physiological configuration. In line with the structures of the CUB 1 -EGF moieties of human MAp19 and C1s (20,40), our study also provides further evidence of the occurrence of a Ca 2ϩ -binding site in the CUB modules of the C1r/C1s/MASP family, thereby indicating that such sites are indeed present in rat MASP-2 but have been overlooked in the x-ray structure reported by Feinberg et al. (22), as discussed previously in detail (40). A further lesson from the C1s (40), MAp19 (20), rat MASP-2 (22), and MASP-1/3 structures is that all of these proteins associate as head-to-tail homodimers through their CUB 1 -EGF moieties, by means of an extended inter-monomer interface. This interface is mainly stabilized by hydrophobic interactions involving residues that are highly conserved in the C1r/C1s/MASP family and are distributed in four contiguous pockets. Interestingly, the C1s CUB 1 -EGF dimer differs to some extent from the other known structures in that it lacks one of these pockets. This may explain why, although purified C1s forms homodimers in the test tube, it preferentially associates with C1r under physiological conditions (42,43), this latter interaction being possibly more stable than the former.
As discussed above, the Ca 2ϩ -binding sites harbored by the CUB modules of C1s, MAp19, and MASP-1/3 exhibit subtle differences in terms of the number and nature of their coordinating ligands, indicating that, as seen for other Ca 2ϩ -binding sites (44), they can slightly adapt their coordination mode. Nevertheless, they have in common a basic framework of interactions contributed by a triad of acidic residues (Glu 49 , Asp 57 , and Asp 102 , and Glu 216 , Asp 226 , and Asp 263 in sites II and III of MASP-1/3, respectively). With the exception of module CUB 2 of C1r where Asp, instead of the expected Glu, is found at position 226, both CUB modules of the proteins of the C1r/C1s/MASP family feature the consensus Glu-Asp-Asp sequence and may therefore be anticipated to harbor a Ca 2ϩ -binding site. Sequence alignment of the known CUB modules reveals that most of them possess this same consensus sequence, indicating that, unlike suggested by initial structural work on spermadhesins (10), the majority of the CUB module population may have the ability to bind Ca 2ϩ .
Twenty seven MASP-3 point mutants were produced in this study, and analysis of their interaction properties provides clear evidence that the protease interacts with MBL, L-ficolin, and H-ficolin via two binding sites contributed by modules CUB 1 and CUB 2 , and located in close vicinity of their respective Ca 2ϩ -binding sites II and III (Fig. 6). This conclusion is in line with our previous study performed on human MAp19, indicating that this protein, which only contains a CUB 1 -EGF module pair, similarly associates with MBL and L-ficolin through a binding site located in CUB 1 , in the vicinity of Ca 2ϩ -binding site II (20). These findings are fully consistent with our current knowledge of the interaction of MASPs with MBL and the ficolins, indicating that this involves in all cases a primary binding site in CUB 1 and an additional binding site contributed by CUB 2 , the latter being required to fully stabilize the interaction (18,21,45).
Mutation of two acidic ligands (Asp 60 and Asp 105 ) of Ca 2ϩbinding site II of MAp19 was previously shown to abolish interaction with MBL and L-ficolin, and this was interpreted as an indirect effect because of disruption of the Ca 2ϩ -binding site, resulting in destabilization of the neighboring interaction site (20). Likewise, this study shows that mutation of either Glu 49 or Asp 102 , two site II ligands, abolishes or strongly decreases the interaction properties of MASP-3. Although the implication of the corresponding Ca 2ϩ ligands in site III could not be tested because of the aggregated state of the mutants, this raises the question of a direct involvement of some of the Ca 2ϩ -binding residues in the interaction of the MASPs with MBL and the ficolins. This hypothesis appears quite plausible in light of the following observations. (i) Mutagenesis studies performed on human and rat MBL (24, 25) and ficolins (26,27) provide evi- dence that binding of these proteins to the MASPs involves in all cases a major ionic interaction through a conserved lysine residue (Lys 55 in human MBL), implying interaction with an acidic component on the MASP side. (ii) Mutations of Glu 80 , Glu 106 , and Glu 267 only have a restricted effect on the interactions, with K D ratios relative to wild-type MASP-3 ranging from 1.7 to 6.9 ( Table 2). Most of the other acidic residues exposed to the solvent in the CUB 1 -EGF-CUB 2 structure have been subjected to mutagenesis, and none of these mutations was effective. A direct implication of acidic ligands of site II and/or III in the interactions appears therefore as a logical hypothesis. (iii) Comparison of the four Ca 2ϩ -binding CUB module structures currently available shows that, of the members of the acidic triad, the residues equivalent to Glu 49 and Asp 102 of the MASP-1/3 site II always coordinate Ca 2ϩ through one of their carboxyl oxygens (Fig. 3). The free oxygen groups of Glu 49 and Asp 102 in CUB 1 and of their counterparts Glu 216 and Asp 263 in CUB 2 point toward the outside of the domain, roughly to the same direction (Fig. 3,  A and B), and appear therefore as ideal candidates for an interaction with the critical lysine residue conserved in MBL and the ficolins. If this hypothesis is correct, then these residues would coordinate Ca 2ϩ on one side, and mediate interaction with MBL or ficolins on the other side, hence providing a major link between the MASPs and their partner proteins. Interestingly, with the exception of D102A, which abolishes binding to L-and H-ficolins, all mutations at residues Glu 49 or Asp 102 inhibit strongly, and to a similar extent, interaction with the three proteins (Table 2), providing support to the hypothesis that both residues contribute equally to this interaction. Nevertheless, mutations at residues such as Tyr 56 , Phe 103 , His 218 , or Tyr 225 each have a significant inhibitory effect (Table 2), and therefore it appears likely that these amino acids, all located in close vicinity of site II or III, also contribute to the interaction to some extent. Fig. 6, C and D, provides a detailed view of the proposed MBL-ficolin interaction sites of MASP-1/3 featuring the side chains of all residues thought to participate in the interaction. Given the similarities between the present data and those obtained previously on MAp19 (20), and the fact that most of the effective mutations performed in this study had very similar impacts on the interaction with MBL, L-ficolin, and H-ficolin, it may be anticipated that the three MASPs associate with each of these proteins according to a common interaction scheme. However, mutation D105G in MASP-2 abolishes binding to MBL (45), whereas the corresponding mutation D102A in MASP-3 only strongly inhibits interaction. This suggests that there are subtle differences between the MASP-2 and MASP-1/3 lineages. Mutagenesis studies on the MBL/ficolin side lead to the same conclusion (25,27).
The above structural information together with previous data on the nature and location of the MBL interaction site (25) have been used to build a three-dimensional model of the interaction between the CUB 1 -EGF-CUB 2 domain of MASP-1/3 and the tetrameric form of human MBL (Fig. 6B). As the four binding sites defined in the CUB 1 -EGF-CUB 2 homodimer lie approximately in the same plane, this may associate with MBL in only two alternative orientations, with its flattest side facing either the C-terminal lectin domains of MBL or its N-terminal extremity. Both orientations have been tested, and the latter (Fig. 6B) was found to better account for the mutagenesis data, particularly for the strong inhibitory effect of the F103A, H218A, and Y225A mutations. An additional advantage of this configuration is that both C-terminal ends of the CUB 1 -EGF-CUB 2 dimer are orientated toward the lectin domains of MBL, allowing the following CCP 1 -CCP 2 -SP catalytic domain to fold back into the MBL⅐MASP complex (see Fig. 7). The resulting CUB 1 -EGF-CUB 2 /MBL assembly is symmetrical, each of the four CUB modules interacting with a collagen-like triple helix of MBL, through a site involving residue Lys 55 , located about halfway along the individual triple helices (Fig. 6B). The distances between the binding sites on the CUB 1 -EGF-CUB 2 dimer are roughly similar, with about 55-60 Å between two adjacent sites, and 75-80 Å between two opposite CUB 1 or CUB 2 sites. For interaction with the trimeric form of MBL, one of the binding sites of the CUB 1 -EGF-CUB 2 dimer is expected to be free. Although the interaction symmetry will be lost, tight binding can still be achieved through the other three sites.
This type of interaction between MBL and the MASP-3 binding domain has direct implications on the assembly and functioning of a whole MBL⅐MASP complex and leads us to propose a schematic model in which the MASP dimer would be able to sway between a "close" conformation allowing activation of each SP domain by its counterpart (Fig. 7A), and an "open" conformation allowing free access of the SP domains to their protein substrates (Fig. 7B). Obviously, such a large conformational change requires a greater extent of flexibility at the CUB 2 -CCP 1 junction. Although no precise structural information is available in this respect, this hypothesis appears consistent with the observation that this particular area of MASP-1/3 is highly susceptible to cleavage by proteolytic enzymes.