Molecular Analysis of the Epidermal Growth Factor-like Short Consensus Repeat Domain-mediated Protein-Protein Interactions

Epidermal growth factor-like (EGF) and short consensus repeat (SCR) domains are commonly found in cell surface and soluble proteins that mediate specific protein-protein recognition events. Unlike the immunoglobulin (Ig) superfamily, very little is known about the general properties of intermolecular interactions encoded by these common modules, and in particular, how specificity of binding is achieved. We have dissected the binding of CD97 (a member of the EGF-TM7 family) to the complement regulator CD55, two cell surface modular proteins that contain EGF and SCR domains, respectively. We demonstrate that the interaction is mediated solely by these domains and is characterized by a low affinity (86 μm) and rapid off-rate (at least 0.6 s−1). The interaction is Ca2+ -dependent but is unaffected by glycosylation of the EGF domains. Using biotinylated multimerized peptides in cell binding assays and surface plasmon resonance, we show that a CD97-related EGF-TM7 molecule (termed EMR2), differing by only three amino acids within the EGF domains, binds CD55 with aK D at least an order of magnitude weaker than that of CD97. These results suggest that low affinity cell-cell interactions may be a general feature of highly expressed cell surface proteins and that specificity of SCR-EGF binding can be finely tuned by a small number of amino acid changes on the EGF module surface.

Most cell surface proteins are highly modular in organization and are constructed from different combinations of a limited set of structural domains. In recent years high resolution structures of many of the most common domains have become available, and analysis of the kinetic characteristics of the interactions they mediate is increasing our understanding of processes that underlie the most basic cellular interactions. In particular, cell adhesion interactions mediated by proteins of the Ig superfamily are characterized by multiple weak binding events, each with a low affinity, but with the multivalent nature of the cell surface proteins resulting in a cell-cell interaction of high avidity (1). Two of the most common structural modules used in cell surface proteins are the epidermal growth factor-like (EGF) 1 and short consensus repeat (SCR) domains ( Fig. 1). Recent analysis of the human genome (as of March 1, 2001) shows that the EGF family is the fifth most common protein family with 3% of all potential proteins containing EGF domains whereas SCR domains are found in 0.3% of all proteins.
Both of these domain types are commonly used to mediate protein-protein interactions. The EGF module, which often occurs as multiple tandem repeats, is widely distributed among extracellular proteins involved in adhesion, receptor-ligand interactions, extracellular matrix structure, determination of cell fate, and blood coagulation (see Fig. 1). A subset of EGF domains contains a consensus sequence associated with calcium binding (cb): (D/N)X(D/N)(E/Q)X m (D/N)*X n (Y/F), where m and n are variable and the asterisk indicates possible ␤-hydroxylation (2)(3)(4). Calcium is thought to perform a key role in the orientation of cbEGF pairs by restricting conformational flexibility of interdomain linkages (5,6) resulting in tandem EGF repeats that are highly resistant to proteolysis (5,7,8). SCR domains are frequently found among proteins of the complement system with many of the complement regulatory proteins (e.g. CD55 and CD46, see below) and complement receptors (e.g. CD21 and CD35) consisting entirely of repeated SCR domains. Structural studies (5, 9 -10) reveal that both of these small modules (Ͻ60 amino acids) fold to form all ␤-strand domains strengthened by disulfide bonds (two in the case of SCRs and three for EGFs). An understanding of why these domains are so well suited for protein interactions has been hindered by the fact that the disulfide-rich multidomain constructs required for structural analyses are difficult to produce in large yields in the native form. We therefore have very few structures of complexes involving these common modules and no structures of EGF-SCR complexes or kinetic characterization of the interaction.
CD55 (or Decay Accelerating Factor; DAF) is a member of the regulators of complement activation (RCA) family. It protects host cells from complement system attack by binding to C3b and C4b preventing formation of the membrane attack complex. It has a widespread tissue distribution and is expressed at high levels on many different cell types. The protein consists of an N-terminal extracellular portion of four SCR domains linked via a heavily O-glycosylated spacer to a Cterminal glycosylphosphatidylinositol anchor (Fig. 1). SCR domains 2-4 are involved in regulation of complement and also in binding to a variety of viral and bacterial pathogens. The most N-terminal SCR (domain 1) also provides the site of interaction for some viruses but until recently the native role of domain 1 was unknown. However, identification of CD97 as a cellular ligand for the N-terminal domains of CD55 (12,13) has now demonstrated a novel natural function associated with this portion of the molecule. CD97 is a member of the EGF-TM7 family, characterized by the unique chimeric structure in which tandem EGF repeats are coupled to a G protein-coupled receptor moiety via a mucin-like stalk region (14,15). CD97 is constitutively expressed on granulocytes and monocytes and is rapidly up-regulated on activated T and B cells. It is known to exist in a variety of splice forms containing different numbers of the EGF domains (16,17) each of which binds CD55 with different affinities. The CD55 binding splice variant with the highest affinity comprises three EGF domains (domains 1, 2, and 5), two of which (domains 2 and 5) are predicted to bind calcium (Fig. 1). Although the precise role of the CD55-CD97 interaction is still unknown the unique hybrid structure, the leukocyte-restricted expression pattern of CD97, and the presence of both CD97 and CD55 in arthritic joints (18) suggest possible roles for the CD97-CD55 interaction in adhesion and signaling within the inflammatory and immune responses.
Recently a novel EGF-TM7 molecule, EMR2, sharing highly homologous EGF domains with CD97 but failing to show an interaction with CD55 in biological assays was identified (19).
In this study we probe the biophysical, cellular, and molecular properties of the CD55 and CD97 interaction and investigate the sequence-specific requirements for CD55 binding. We show that the interaction is mediated solely by the EGF domains of CD97 and is characterized by a low affinity and rapid off-rate. Ca 2ϩ is essential for the formation of the CD55 binding face on CD97, but glycosylation of EGF domains from CD97 is not required. The three amino acid differences within the EGF domains of EMR2 that distinguish it from CD97 decrease the affinity for CD55 by at least an order of magnitude. The implications of these data for general properties of cell surface interactions and specificity of EGF domain interactions are discussed.
Bacterial Expression of CD97 and EMR2-The three extracellular EGF domains (EGF-1,2,5) of the CD55 binding splice variant of CD97 and (by analogy) of EMR2 were expressed in Escherichia coli using a His tag-based expression system (Qiagen). Primers used for the PCR amplification of EGF domains 1, 2, and 5 from CD97 and EMR2 cDNAs were 5Ј-TAGTAGGGATCCATAGAAGGACGATCAGCAGACTCCAGG-GGCTGTGCC (forward) and 5Ј-TAGTAGAAGCTTCTATTATTCACAG-ACAGTGTCCTTTTG (reverse). Restriction sites used for subsequent cloning into the expression vector pQE30 are underlined. The forward primer also contains a FXa cleavage site and a two-amino acid (SA) spacer sequence prior to the authentic sequence of CD97 and EMR2. Following Ni 2ϩ affinity purification under denaturing conditions, peptides were reduced, purified, and refolded according to a well established in vitro refolding protocol (2). The data indicating that the multidomain  (40,41). constructs adopted the native fold was indirect but substantial. Both peptides showed the characteristic change in HPLC elution profile previously observed for all other cbEGF constructs on refolding (e.g. the profile for EMR2 is shown in Fig. 2). Because calcium binding is a property of the native fold of cbEGF domains, correct refolding of these domains in CD97 and EMR2 was indicated by the observed Ca 2ϩ -dependent protection against proteolysis (data not shown) and the Ca 2ϩ dependence of CD55-CD97 interaction (see "Results"). After purification, CD97 and EMR2 were lyophilized and reconstituted into appropriate buffers at the desired concentration. The concentration of protein solutions was confirmed using the calculated extinction coefficient at 280 nM (21,000 M Ϫ1 cm Ϫ1 computed from the amino acid composition on the EXPASY server, Ref. 20).
Pichia Expression of CD55-The expression of the four SCR domains of CD55 as a soluble fragment in the yeast Pichia pastoris has been described in detail elsewhere (21). The protein was dialyzed (10,000 molecular weight cut-off Slide-A-Lyzer, Pierce) against 50 mM Tris/HCl, pH7.5 to remove imidazole and concentrated to ϳ0.8 mg/ml using a 3000 molecular weight cut-off Centriprep 3 (3500 rpm, 120 min). The concentration of protein was assessed using the calculated extinction coefficient at 280 nM (36,840 M Ϫ1 cm Ϫ1 (20)).
Surface Plasmon Resonance-Surface plasmon resonance (SPR) experiments were performed on a BIAcore2000 (BIACORE AB, Stevenage, UK). CD55 was covalently immobilized to the carboxylated dextran matrix on the surface of CM5 sensor chips via primary amine coupling using the amine-coupling kit (BIAcore AB) as directed (22) with the following modifications. After the activation step, purified CD55 was injected at 11 g/ml in 10 mM sodium citrate (pH 4.6) for 5 min (5 l/min). This repeatedly resulted in the coupling of ϳ1000 response units (RU) of CD55 to the chip surface. Biotinylated CD97 was coupled to the chip surface via streptavidin according to the protocol provided by Biacore. Interaction data were collected by injecting 30 l of the appropriate analyte over the coupled chip surface at a flow rate of 20 l/min at 25°C, and all traces were corrected for refractive index changes by subtraction of a control trace simultaneously recorded from a mock-immobilized channel. Unless otherwise stated, all experiments were carried out in a buffer (I ϭ 0.15) 5 mM Ca 2ϩ , 135 mM NaCl, 10 mM Tris, pH 7.4.
All culture media were supplemented with 10% heat-inactivated fetal calf serum, 2 mM L-glutamine, 50 IU/ml penicillin, and 50 g/ml streptomycin, and cells were incubated in a humidified 37°C, 5% CO 2 incubator. HEK293T cells were maintained in Dulbecco's modified Eagle's medium, K562 cells in RPMI 1640 (R10), and CHO-K1 cells in Ham's F-12 medium. CHO-K1 cells were transfected using Lipo-fectAMINE (Life Technologies, Inc.) according to the manufacturer's protocol. HEK293T cells were transfected with 40 g of DNA/175-cm 2 flask using calcium phosphate precipitation. Six hours post-transfection, the medium was replaced with 15 ml of serum-free Dulbecco's modified Eagle's medium and incubated for a further 72 h.
Cell Rosette Assays-Cell rosette assays were performed as previously described (19). Briefly, CHO-K1 cells transfected with the appropriate expression vectors were subcultured at day 1 into 12-well dishes and analyzed for their ability to bind human red blood cells 3 days post-transfection. Heparinized human whole blood cells were diluted 1:100 (vol/vol) in R10 and added to transfected cells (1 ml/well) for 30 min at room temperature. Nonbinding red blood cells were removed by gentle washing. The extent of red blood cell adhesion was quantified by measuring the peroxidase activity of hemoglobin (E 450 ) of methanolfixed red blood cells using the TMB substrate. The level of cell surface protein expression on transfected cells was determined by FACS analysis using appropriate CD97 mAbs. For cells transfected with EMR2 and CD97 isoforms as well as the domain-swapping and deletion chimeras, CLB-CD97/1 and BL-Ac/F2, which recognize the first EGF-like domain of both EMR2 and CD97, were used. CLB-CD97/2 and CLB-FIG. 2. HPLC chromatograms. The elution profile of EMR2 upon reduction with dithiothreitol (A) and on refolding (B) using an oxidoshuffling system. Note the earlier elution time of the refolded form is characteristic of cbEGF domains (3,11). CD97/3, which specifically recognize the stalk region of CD97, were used for cells transfected with the EMR2 and CD97 site-directed mutants. The red blood cell binding ability of transfected cells was represented by the measurement of peroxidase activities normalized by the median level of cell surface protein expression determined by FACS analysis.
Production of Biotinylated Proteins-Conditioned medium collected from transfected HEK293T cells was concentrated to ϳ0.5 ml using a 30-kDa cut-off Centriprep tube (Millipore, Bedford, MA), dialyzed with 10 mM Tris-HCl, pH 8 buffer and incubated with 1 l of BirA enzyme (Avidity, Denver, CO) overnight at room temperature. Excess biotin was subsequently removed by dialysis with 10 mM Tris-HCl, pH 7.3 buffer containing 10 mM CaCl 2 and 150 mM NaCl. The biotinylated proteins were then aliquoted and stored at Ϫ80°C after quantification by dot-blot analysis using myelin basic protein-biotin (Avidity, Denver, CO) as standard.
Cell Binding Assay Using Biotinylated Protein-coupled Fluorescent Beads-Cell binding assays using fluorescent beads coupled to biotinylated proteins were performed as previously described (23). In brief, 20 l of avidin-coated fluorescent beads (Spherotech, Inc., Libertyville, IL) were washed twice and added to 2 g of biotinylated protein in a total volume of 50 l. The bead and protein mixture was sonicated at 20% power for 1 min (Heat systems, Sonicator) and then incubated at 4°C for 1 h. Nonbinding proteins were removed by washing twice with phosphate-buffered saline/bovine serum albumin, and the beads were resuspended in 50 l of R10. The bead-protein complex was sonicated again immediately before adding to K562 cells in a 96-well plate (1 ϫ 10 6 cells/50 l R10/well). The cell-bead mixture was spun at 1000 ϫ g at 4°C for 20 min, incubated for a further 40 min at 4°C, and finally resuspended in 500 l of phosphate-buffered saline for FACS analysis. Where necessary, additional reagents (divalent cations, EGTA, and mAbs) were added to the cells 5-10 min before the introduction of the protein-bead complex.

CD55 Binding Is Mediated Exclusively by the EGF Domains of CD97-Previous
studies have demonstrated that the EGF domains of CD97 are necessary for CD97-CD55 interaction (13) and that EMR2 is unable to interact with CD55 (19). CD97 and EMR2 share highly homologous EGF domains but are relatively variant (ϳ50% identical) within the supporting stalk region. To investigate the possible contribution of the stalk region to the previously observed CD55 binding activity of CD97, domain-swapping chimeras and stalk region deletion mutants were analyzed (Fig. 3) using a quantitative cell rosetting assay (see "Experimental Procedures"). Proteins containing the EGF-1,2,5 domains of CD97 and the stalk region of EMR2 are CD97-like in their CD55 binding properties, whereas constructs consisting of the EGF domains of EMR2 and the stalk of CD97 are EMR2-like showing no CD55 binding activity in this assay. Deletion of the CD97 stalk retains the ability of CD97 to bind CD55 but reduces the overall binding by ϳ60%.
We interpret the reduced binding to CD55 observed on deletion of the CD97 stalk as being caused by steric effects; in the absence of the CD97 stalk the EGF domains are less accessible to the cell-bound CD55. Support for this interpretation comes from our surface plasmon resonance studies (see below) that show that constructs of CD97 with or without the stalk have identical affinities for soluble CD55. Controls used in this experiment confirm that different isoforms of CD97 have different CD55 binding abilities with the shortest CD97 isoform, CD97 (EGF-1,2,5), having the highest affinity (100%) whereas the longest isoform, CD97 (EGF-1,2,3,4,5), shows only ϳ20% of the CD55 binding affinity. All EMR2 isoforms showed no detectable binding to CD55 in this assay (19).
Equilibrium Binding Analysis of the Interaction between CD55 and CD97-Surface plasmon resonance (SPR) was used to study the detailed interaction kinetics between domains from CD55 and CD97 implicated in complex formation. Protein constructs consisting of four SCR domains from CD55 and the three EGF domains from CD97 with or without the stalk (Fig.  1) were expressed and purified (see "Experimental Procedures"). SPR measurements were obtained for proteins in both orientations, i.e. either fixed to a static surface or in solution. The CD97 used in the soluble phase consisted of the EGF domains alone and was not glycosylated because it was derived from an E. coli expression system (see "Experimental Procedures"). In contrast, the CD97 bound to the chip containing the full stalk region and was glycosylated because it was obtained using a mammalian cell expression system. Fig. 4a(i) shows an injection of 5 M CD55 (bar) for 90 s over a chip with 350 RU of CD97 bound whereas Fig. 4b(i) shows an injection 64 M CD97 (bar) for 90 s over a chip with CD55 bound. A background response is seen in the control trace of each experiment because the BIAcore measures the refractive index near the sensor surface and therefore detects any changes in the bulk refractive index of the injected sample. The response seen when soluble protein is injected over its bound ligand is considerably larger than the response seen in the control trace in each case. Inspection of both sets of sensorgrams reveals that the kinetics of the CD97-CD55 interaction are rapid; binding reaches equilib- rium within 10 s of the start of the injection, and dissociation is complete within 20 s of the end of the injection. Fig. 4, a(ii) and b(ii) show a series of sensorgrams obtained by injecting a dilution series of the appropriate soluble protein (either CD55 or CD97) over the chip surface. The traces shown are corrected by subtraction of the control trace and overlaid so that the start and end points of the injections are coincident. The ratios of the maximal levels of free-flowing protein bound to the level of coupled protein (both measured in RU and corrected for molecular weight) revealed that ϳ45% of the covalently coupled CD55 was able to interact with the soluble CD97 whereas ϳ60% of the bound CD97 was able to interact with soluble CD55. In the case of data recorded from the CD97-coupled chip, the control curve was recorded from a channel on which an unrelated protein was bound (maltose-binding protein) rather than a simply mock-immobilized chip surface.
Equilibrium binding responses were subsequently measured during injection of at least six concentrations of the soluble protein varying by at least one order of magnitude, and the data were analyzed using BIAevaluation 3.0 software (BIAcore AB). Nonlinear curve fitting of a simple Langmuir model of the association (A ϩ B 3 AB) to these data (Fig. 4, a(iii) and b(iii)) yielded values for the dissociation constant, K D . K D was also determined from Scatchard plots of the data (fits not shown). Identical results were obtained irrespective of the order of injections (low to high concentrations or vice versa), indicating that both the CD55 and CD97 were stable for the duration of the experiment (data not shown). Table I summarizes the values of K D obtained from different experiments (using protein expressed in different preparations) and the quality of the nonlinear fits to the data. The mean of all the experiments yielded a value for K D of 86 Ϯ 1 M. The estimates of K D were the same in whichever orientation the experiment was per-formed (i.e. CD55 or CD97 immobilized on the chip surface) suggesting that the value of K D obtained provides a true representation of the in vivo affinity and is not subject to coupling effects. Because K D is the same for both the soluble and immobilized forms of CD97, glycosylation and the stalk region are not required for CD55 binding. Simultaneous fitting of numerically integrated rate equations derived from the simple Langmuir binding model (A ϩ B 3 AB) to the sets of sensorgrams (global analysis option BIAevaluation 3.0) shows that the offrate is at least 0.6 s Ϫ1 (data not shown).
Dependence of CD55-CD97 Interaction on the Presence of Ca 2ϩ -Previous studies of a cbEGF domain pair from fibrillin-1 have demonstrated that calcium is essential for maintenance of a rod-like interdomain linkage (5). Removal of Ca 2ϩ leads to a change in the dynamic properties of this pair, which may be detected by an increase in the susceptibility of the EGF pair to proteolysis (25). Inspection of the sequence of the soluble fragment of CD97 suggests that two of the three EGF domains (domains 2 and 5) are of the Ca 2ϩ binding type (Fig. 5). To study the potential role of Ca 2ϩ in the CD97/CD55 interaction,  we investigated Ca 2ϩ dependence of the interaction. The equivalent concentration of CD97 was injected over immobilized CD55 in Ca 2ϩ -containing buffer, in the presence or absence of EGTA (Fig. 6). In Ca 2ϩ -containing buffer the presence of EGTA was seen to completely abolish the interaction so that no difference was seen between the response from the control and CD55-coupled chip surfaces. In addition, Mg 2ϩ was unable to substitute for Ca 2ϩ , because no binding was observed when CD97 was injected over immobilized CD55 in the presence of EGTA and Mg 2ϩ (Fig. 6); however, binding was restored by the subsequent addition of Ca 2ϩ .
Dissection of the Effect of the Sequence Differences between the EGF Domains of CD97 and EMR2 on CD55 Binding Affinity-EMR2 (19) differs from CD97 by only three amino acid changes within EGF domains 1, 2, and 5, two of which occur in EGF domain 1 and one in EGF domain 2 (98% sequence identity in the three EGFs contained in this fragment) (Fig. 5). In previous cell-based assays (19) and this study, EMR2 has shown no interaction with CD55 (Fig. 3). Surface plasmon resonance was used to quantitate the effects of the three amino acid differences on CD55 binding. The three EGF domains of EMR2 were expressed and purified using an E. coli expression system (see "Experimental Procedures") and flowed in the soluble phase over a CD55-coupled chip surface. Only by using high concentrations (ϳ10 mM) of EMR2 could a weak, specific interaction between CD55 and EMR2 be observed, and the binding was sufficiently weak that direct determination of the K D was not possible with the amounts of protein available. Although it is difficult to accurately compare the absolute values of SPR signals, a direct comparison of the interaction of CD55 with CD97 and EMR2 may be made because these proteins are of the same molecular weight and therefore produce the same refractive index change on binding similar amounts to the sensor surface. Fig. 7 shows the equilibrium response obtained flowing CD97 and EMR2 over the same CD55-coupled flow cell. Data for CD97 are obtained from two independent preparations of CD97 and show that the variation in response between different CD97 preparations is small by comparison to the difference in response observed when comparing CD97 and EMR2. If we assume that the maximal binding capacity of CD55 for EMR2 is the same as its capacity to bind CD97 we can use the initial slope, where the amount of EMR2 or CD97 FIG. 5. a, amino acid sequence alignment of the EGF-1 and EGF-2 of CD97 (GenBank TM /EBI accession number NP 001775) and EMR2 (NP 038475) with EGF-32 and EGF-33 of human fibrillin-1 (NP 000129). Numbers on top of the residues indicate the position of the residues in the full-length CD97 and EMR2 proteins based upon the published data. Six conserved Cys residues are placed at fixed positions to allow optimal alignment. b, schematic representation of the EGF-like domains 1, 2, and 5 of CD97. Cys residues are highlighted in yellow. The three residues that are different in CD97 and EMR2 are highlighted in red and the individual corresponding residues in EMR2 are indicated by an arrow. c, mapping of the EMR2 sequence differences (shown in red) onto the surface of a cbEGF pair. Coordinates used are those of the fibrillin cbEGF32-33 structure (PDB 1EMN.ENT).
bound is approximately proportional to the concentration of protein applied, to provide an estimate of the affinity of EMR2 for CD55 relative to that of CD97. Inspection of the initial slopes of the data presented here suggests that the affinity of EMR2 for CD55 is at least an order of magnitude weaker than that of CD97 for CD55. This is equivalent to a change in the binding energy (using the equation Ϫ⌬G 0 ϭ RTlnK D ) from 5.5 kcal M Ϫ1 to 4.2 kcal M Ϫ1 .
To assess the individual contribution of the three amino acid residues to CD55 binding, two assays were carried out using site-directed single residue mutants of CD97 and EMR2. Cell rosetting analysis of transfected CHO-K1 cells showed that the CD55binding activities of the CD97-N33D, CD97-T59M and CD97-P71L mutants were all reduced compared with that of wild-type CD97 (EGF-1,2,5). Conversely, cells transfected with the EMR2-D36N-, EMR2-M62T-, or EMR2-L74P-expressing constructs displayed increased CD55 binding abilities (Fig. 8). Cell surface protein expression in CHO-K1 cells was monitored by FACS and shown to be comparable for all proteins studied (see "Experimental Procedures"). Consistent with the quantitative data, fewer and smaller rosettes of erythrocytes were observed around the cells expressing mutant proteins (data not shown).
A second ligand binding assay using multimeric forms of soluble extracellular domains of CD97 and EMR2 proteins was also employed to analyze the binding properties of mutant proteins (Fig. 9). Biotinylated extracellular domains of CD97 and EMR2 were coupled to avidin-coated fluorescent microspheres to form the multimeric protein probes for use in a FACS-based assay system (Fig. 9a) (26). K562 human myelogenous leukemia cell line, expressing a homogenous cell surface CD55 expression pattern (data not shown; Ref. 27), was used as a source of CD55 in the assay. As expected, wild-type CD97 protein-microsphere complexes bound to K562 cells and showed a strong shift in fluorescence intensity (Fig. 9b). In contrast, wild-type EMR2 protein-microsphere complexes did not bind K562 cells. As previously demonstrated, the binding of CD97-microsphere complexes to K562 cells was found to be calcium-dependent and mediated by CD55 as the addition of EGTA without Mg 2ϩ (data not shown), with Mg 2ϩ , or a blocking anti-CD55 mAb completely ablated the binding (Fig. 9, c   FIG. 6. Demonstration of Ca 2؉ dependence of CD55-CD97 interaction. The black bars mark the injections of 65 M CD97 in a buffer of 100 mM Tris/HCL pH 7.5, 150 mM NaCl with the addition of the components marked above the injection bar. Raw data are shown for the signal from the CD55-coupled surface (red trace) and control surface (blue trace). A specific interaction between CD55 and CD97 is demonstrated where the red and blue signals differ significantly (as they do in injections 1 and 4). The absolute size of the signal differs from one injection to the next because of the different absorbance properties of the buffers with EGTA and Mg 2ϩ added. Specific CD55-CD97 interactions abolished by the presence of EGTA are only restored by the addition of Ca 2ϩ . and d). In accordance with the finding described earlier, biotinylated mutant EMR2 proteins, EMR2-D36N-Bio, EMR2-M62T-Bio, and EMR2-L74P-Bio showed increasing binding affinities for K562 cells, whereas CD97 mutant proteins, CD97-N33D-Bio, CD97-T59M-Bio, and CD97-P71L-Bio displayed reduced K562 binding abilities (Fig. 9, e and f). Both assay systems clearly show that the introduction of any of the three EMR2 amino acids into the CD97 sequence resulted in a decrease in CD55 binding, and conversely introduction of any of the three CD97 amino acids into the EMR2 sequence partially restored CD55 binding. The substitution of Pro with Leu at position 71 of CD97 caused the greatest reduction in CD55 binding whereas replacing the Leu with Pro at position 74 of EMR2 resulted in the largest increase in CD55 binding.

DISCUSSION
In this study we have identified the molecular basis of the CD55-CD97 cell surface interaction using biological and biophysical methods. Binding is mediated exclusively by proteinprotein interactions of the SCR and EGF module families located at the N terminus of each protein. Surface plasmon resonance studies of the EGF-SCR interaction show that it is characterized by a low affinity caused by a rapid off-rate. Because CD55 and CD97 are known to be expressed at high levels on the surface of cells (27)(28)(29), our data predict an in vivo interaction of high avidity. Prior work by others has demonstrated that cell-cell interactions mediated by proteins of the Ig superfamily are also characterized by multiple, low affinity interactions. Our results are therefore of general interest because they suggest this property may prove to be a widespread characteristic of cell-cell interactions mediated by a wide variety of protein folds; perhaps because fine titration of cell-cell binding affinity may be easily achieved by regulating expression of the interacting proteins on the cell surface.
Interestingly the K D of the CD55-CD97 interaction is at least an order of magnitude weaker than the previously characterized interaction between CD55 and Echovirus 11 (22). The requirement of the virus-CD55 interaction to be of a higher affinity than the CD97-CD55 interaction probably reflects the fact that an icosahedral virus can, at most, present 60 binding sites for its receptor and must therefore have a reasonably high affinity for its receptor to achieve a sufficiently avid interaction. However, an activated leukocyte will have many more than 60 copies of CD97 presented on the cell surface, and a high avidity interaction may therefore be promoted by a proteinprotein interaction of a much lower affinity.
Because both mammalian-expressed CD97 (immobilized CD97 bound to chip surface) and E. coli-derived material (when CD97 is present in the soluble phase) give the same value for K D , glycosylation of CD97 is not involved in determining the specificity or affinity for CD55. There has been much debate about the role of glycosylation in determining the specificity and affinity of cell adhesion interactions (30,31), and recent work has shown that for another family of EGF-containing proteins, the Notch family, control of the glycosylation state of the EGF domains by selective expression of the glycosylating enzyme is used to regulate the interactions of Notch with its ligands (32). However prior work studying another adhesive interaction has shown that the interaction of CD2 with its ligands is glycosylation-independent (32). Our demonstration of the glycosylation independence of the CD97-CD55 interaction shows that CD2-ligand interactions are not an exception that "proves the rule," and that glycosylation of extracellular domains is not necessarily required to modulate protein-protein interactions.
Although the cell-based binding assays employed in this study showed no detectable CD55-EMR2 interaction, SPR assays have detected a weak but specific interaction between CD55 and EMR2. It is possible that the much weaker CD55-EMR2 interaction has fallen beyond the detection limit of the cell-based assay systems, which are less sensitive than SPR. It would be reasonable to speculate that, given high enough levels of cell-surface CD55 proteins one would be able to detect the CD55-EMR2 interaction using the cell-based assay systems. Because both CD97 and EMR2 are predominantly expressed by granulocytes, monocytes, and macrophages, this provides a mechanism whereby the CD55 binding ability of these important immune cells could be regulated by the cell surface expression levels of a pair of closely related EGF-TM7 proteins.
The Ca 2ϩ -dependence of the CD97-CD55 interaction indicates that the Ca 2ϩ binding sites within the EGF-2 and 5 domains of CD97 are crucial for intermolecular interactions. Structural data from a fibrillin-1 cbEGF pair have shown that Ca 2ϩ binding is required for the maintenance of interdomain rigidity (5, 6). As a consequence, tandem repeats of cbEGF domains or EGF-cbEGF domains with similar conservation of residues are predicted to form extended rod-like structures, which present specific protein surfaces for protein-protein interactions. Inspection of the sequences of EGF-cbEGF and cbEGF-cbEGF pairs from CD97 and EMR2 suggest they are of the fibrillin-1 or class I type (5), because they have one residue between the last Cys residue of the N-terminal cbEGF and the first calcium binding residue of the C-terminal cbEGF (Fig. 5). In addition hydrophobic packing residues also implicated in maintaining the rod-like conformation of fibrillin-1 cbEGFs are conserved in CD97 and EMR2. Calcium binding to cbEGF domains is therefore probably critical in maintaining CD97-CD55 interaction by sustaining an overall rod-like structure of the three EGF domains.
The complete abrogation of the CD55/97 interaction in the absence of Ca 2ϩ (Figs. 5 and 9d) suggests that the protein surface on CD97 recognized by CD55 extends over the domain 1-2 boundary rather than being localized on a single domain, and this observation is confirmed by the locations of the amino acid changes within domains 1 and 2, which differentiate CD97 and EMR2. To better understand how any one of these amino acid changes causes a reduction in CD55 binding, effects that occur directly because of alteration of a side chain which previously contacted CD55 or insertion of a bulky group within the interface have to be distinguished from those changes that disturb binding indirectly by producing a more long-range alteration in structure. A small number of missense mutations in cbEGF domains associated with human disease are localized in variable loop regions, and it has been predicted that in fibrillin-1 these mutations occur at sites used for protein-protein interactions (5). EGF modules are found to be highly consistent in structure, the positions of ␣ carbons varying by less than 2.5 Å when all EGF structures determined to date are compared. The known structure of an EGF pair can therefore be used to predict the locations of the EMR2 mutations on the first two domains of CD97. Mapping of the three amino acid changes between CD97 and EMR2 onto the three-dimensional structure of fibrillin-1 domains 32 and 33 (Fig. 5, a and c) indicates that they are likely to be located in these variable loops and may therefore directly participate in the interface with CD55. It remains possible, however, that the mode of action for the Leu to Pro substitution at position 71 is indirect because a Pro at this position is implicated in stabilizing the hydrophobic core of a cbEGF domain from factor IX (10).
The mapping of the interaction between CD55 and CD97 to residues in domains 1 and 2 of CD97 appears, initially, to be in opposition to previous data that have shown that splice forms of CD97 other than the 1,2,5 form bind CD55 poorly ( Fig. 3 and Ref. 13). If the binding site for CD55 is truly contained within domains 1 and 2 then one might expect that the longer splice forms (e.g. 1, 2, 3, 4, 5) should also bind CD55 with the same efficiency. The combination of our data with these earlier results suggests that the attachment of domain 2 to any domain other than 5 disrupts the domain 1-2 junction so destroying the CD55 recognition site. It is interesting to note that variable K D values for Ca 2ϩ have been observed within cbEGF domains from different proteins (33). The precise K D values within tandem repeats of cbEGFs therefore have the potential to modulate the interdomain linkage. Domain 5 may have a particularly high affinity for Ca 2ϩ compared with other splice variants that maintains the rod-like surface within domains 1 and 2 and so facilitates binding.
In summary our data demonstrate that the CD97-CD55 interaction is mediated solely by EGF and SCR domains located at the N terminus of each protein and is glycosylation-independent. The predicted characteristics of binding (low affinity, high avidity) resemble those mediated by members of the Ig superfamily and may prove to be a general feature of proteinprotein interactions mediated by highly expressed cell surface proteins. The altered binding characteristics of EMR2 demonstrates how a wide range of affinities can be achieved by a small number of amino acid changes within the EGF module, suggesting a reason for the widespread occurrence of this module type in extracellular proteins involved in protein-protein interactions.