The PDZ Domain of Human Erythrocyte p55 Mediates Its Binding to the Cytoplasmic Carboxyl Terminus of Glycophorin C

The PDZ domain, also known as the GLGF repeat/DHR domain, is an ∼90-amino acid motif discovered in a recently identified family of proteins termed MAGUKs (membrane-associated guanylatekinase homologues). Sequence comparison analysis has since identified PDZ domains in over 50 proteins. Like SH2 and SH3 domains, the PDZ domains mediate specific protein-protein interactions, whose specificities appear to be dictated by the primary structure of the PDZ domain as well as its binding target. Using recombinant fusion proteins and a blot overlay assay, we show that a single copy of the PDZ domain in human erythrocyte p55 binds to the carboxyl terminus of the cytoplasmic domain of human erythroid glycophorin C. Deletion mutagenesis of 21 amino acids at the amino terminus of the p55 PDZ domain completely abrogates its binding activity for glycophorin C. Using an alanine scan and surface plasmon resonance technique, we identify residues in the cytoplasmic domain of glycophorin C that are critical for its interaction with the PDZ domain. The recognition specificity of the p55 PDZ domain appears to be unique, since the three PDZ domains of hDlg (human lymphocyte homologue of the Drosophiladiscs large tumor suppressor) do not bind the cytoplasmic domain of glycophorin C. Taken together with our previous studies, these results complete the identification of interacting domains in the ternary complex between p55, glycophorin C, and protein 4.1. Implications of these findings are discussed in terms of binding specificity and the regulation of cytoskeleton-membrane interactions.

p55 is a heavily palmitoylated peripheral membrane protein of the red blood cell membrane (1). The primary structure of p55 defines several distinct domains including the PDZ domain, the SH3 domain, the protein 4.1 binding domain, the tyrosine phosphorylation domain, and a carboxyl-terminal guanylate kinase-like domain (2). The domain organization of p55 is similar to a family of proteins termed MAGUKs 1 (membraneassociated guanylate kinase homologues), which appear to play important roles in the regulation of signaling pathways and cytoskeleton-membrane interactions (3)(4)(5)(6). Among various protein domains of MAGUKs, the PDZ domain has been a focus of intense scrutiny (5)(6)(7)(8). Like the SH2 and SH3 domains, the PDZ domains recruit cytoplasmic proteins to specific submembranous sites. For example, the PDZ 1 ϩ 2 domain, but not the PDZ 3 domain, of PSD-95 and hDlg interacts with the carboxyl terminus of the NMDA receptors and the Shaker type K ϩ channels, and this interaction plays an important role in the clustering of ion channels (5,6). The PDZ domains have also been shown to interact with other PDZ domains. The PDZ domain of neuronal nitric-oxide synthase binds to the PDZ domain of syntrophin at the sarcolemma in skeletal muscle and to the PDZ 2 domain of PSD-95 at the synaptic junctions of neuronal cells (9,10). These observations suggest that the primary structures of both the PDZ domain and its target peptide govern the specificity as well as the affinity of PDZ domain-mediated interactions in vitro and in vivo. The threedimensional structure of PDZ 3 domains in PSD-95 and hDlg (11,12), together with our understanding of the peptide binding specificities of PDZ domains (13), reveals that the specificity of PDZ domain recognition is indeed dependent upon the primary sequence of the PDZ domain and its cognate ligand.
The biochemical interactions of p55 protein have been well studied in the red cell membrane. We have previously shown that p55 directly binds to protein 4.1 and glycophorin C and may form a ternary complex with these proteins at the cytoplasmic face of the plasma membrane (14). This ternary complex links the spectrin-actin junctions to the lipid bilayer and is presumed to play an important role in maintaining erythrocyte shape and its membrane material properties (15,16). To delineate the basis of p55 interactions with protein 4.1 and glycophorin C, we have mapped the binding interfaces between protein 4.1 and p55 as well as between protein 4.1 and glycophorin C (17). However, the site where the cytoplasmic domain of glycophorin C binds to p55 had not yet been determined. In this paper, we show that the single copy of the PDZ domain in human erythrocyte p55 binds to the carboxyl-terminal 12 amino acids of glycophorin C. The carboxyl terminus YFI motif of glycophorin C is the critical binding element for PDZ do-main-mediated interaction. The interaction between p55 and glycophorin C appears to be specific, since the PDZ domains of hDlg fail to bind glycophorin C. These findings indicate that the PDZ domains are novel protein modules that target cytoplasmic proteins to the multiprotein complexes at the inner surface of the plasma membrane.

EXPERIMENTAL PROCEDURES
Preparation of Recombinant Cytoplasmic Domain of Glycophorin C-The 47-amino acid cytoplasmic domain of human erythroid glycophorin C was produced as a glutathione S-transferase fusion protein (GST-GPC 82-128 ) in bacteria. The nucleotide primers required for the cloning of the construct have been described previously (14). The GST-GPC 82-128 protein was affinity-purified on the glutathione-Sepharose beads, and the GPC 82-128 protein was cleaved from GST on the column by thrombin (18). The cleaved GPC 82-128 was concentrated and dialyzed against 20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1.0 mM dithiothreitol, and 0.1 mM phenylmethylsulfonyl fluoride. The GPC 82-128 protein, at a concentration of 7.0 mg/ml, was purified on a gel filtration Superdex-75 column using the fast protein liquid chromatography (FPLC) system (Pharmacia Biotech Inc.). The eluted fractions were analyzed by 18% SDS-polyacrylamide gel electrophoresis (19). The purified protein GPC 82-128 was used as a probe in the blot overlay assay.
Mass Spectrometry-Purified recombinant cytoplasmic domain of glycophorin C was analyzed by matrix-assisted laser desorption/ionization (MALDI) mass spectrometry. Thrombin-cleaved GST-GPC 82-128 was purified to remove residual GST, thrombin, and an unknown bacterial contaminant by FPLC on a Mono Q column. Purified GPC 82-128 was concentrated and dialyzed against a low salt buffer (20 mM Tris-HCl, pH 8.0, 20 mM NaCl, 0.5 mM EDTA, and 1.0 mM 2-mercaptoethanol) using an Amicon stirred cell with a YM-3 membrane (molecular mass cut off, 3000 Da). GPC 82-128 at a concentration of 460 M was analyzed on a Perceptive Biosystems Voyager MALDI instrument. Data were obtained by spotting 0.5 l of GPC 82-128 (230 pmol) with 0.5 l of sinapinic acid. Alternatively, 23 pmol (0.5 l) of purified GPC 82-128 was spotted with an equal volume of 2,5-dihydroxybenzoic acid. The spectra were calibrated using two-point external calibration.
Sedimentation Equilibrium Analysis-Sedimentation equilibrium experiments were carried out in a Beckman Optima XL-A analytical ultracentrifuge. The rotor speed and temperature were 40,000 rpm and 20°C, respectively. The equilibrium solute distribution (achieved within 10 h) was recorded with scanning absorbance optics set to 280 nm and was ascertained by the satisfactory overlay of traces acquired 3 h apart. Nine samples of GPC 82-128 in low salt buffer, with concentrations of between 8 and 150 M, were loaded into six-channel Yphantis-type (20) centerpieces, giving a solution column height of about 1.3 mm. Given the relatively low molecular mass of GPC (5297 Da), it was not possible to obtain true optical base lines by overspeeding of the rotor. The partial specific volume was calculated ( ϭ 0.71 ml/g) from the known amino acid sequence using Traube's rule and the consensus volumes for amino acids given by Perkins (21). The buffer density at 20°C was assumed to be 1.0 g/ml. Data were fitted with the exponential form of the solution to the Lamm equation for a single ideal species (22) using the following: where A r is the absorbance at radial position r and A o is the absorbance at a reference position r o ; H is the constant (1 Ϫ ) 2 /2RT; is the partial specific volume of the macromolecule; is the solvent density; is the rotor speed (radians/s); R is the gas constant; T is the temperature (K); M is the molecular weight of the solute; E is the optical base-line offset.
The sequences of the primers p55-1, p55-3, and p55-5 have been described previously (17). The polymerase chain reaction-amplified DNA fragments were subcloned into the pGEX-2t vector, and clones were sequenced to confirm their identity and frame. The GST fusion proteins were expressed in the Escherichia coli strain DH5␣ as described previously (14).
Synthetic Peptides of Glycophorin C-Peptides containing the carboxyl-terminal 24 amino acids (Ala-105 to Ile-128; peptide P 24 ) or 12 amino acids (Asp-117 to Ile-128; peptide YFI) corresponding to the cytoplasmic domain of glycophorin C were chemically synthesized. Three mutant peptides were also synthesized by alanine scanning of the extreme carboxyl-terminal residues of the peptide YFI. These are: 1) peptide YFA, in which Ile-128 was substituted by Ala; 2) peptide AFI, in which Tyr-126 was substituted by Ala; 3) peptide AFA, in which both Tyr-126 and Ile-128 were substituted by Ala. All peptides contained a biotin molecule attached at the N terminus and were purified by high pressure liquid chromatography on an analytical reverse phase column. The purity of the peptides was confirmed by amino acid analysis and mass spectrometry.
Blot Overlay Assay-The experimental conditions used for the blot overlay assay have been described previously (14). The binding of glycophorin C to the immobilized proteins was detected using polyclonal antibodies raised against the cytoplasmic domain of human glycophorin C. These antibodies (kindly provided by Dr. Philip Low, Purdue University) were produced in rabbits injected with a synthetic peptide corresponding to the entire cytoplasmic domain of glycophorin C. Alternatively, direct binding of glycophorin C to the immobilized fusion proteins was detected using a biotinylated glycophorin C peptide P 24 . Bound peptide was then detected using the streptavidin-peroxidase detection system.
Surface Plasmon Resonance Measurements-Protein interactions were quantified using the BIAcore instrument from Pharmacia. The instrument uses the surface plasmon resonance phenomenon to detect changes in refractive index as a result of biomolecular interactions (5). A surface plasmon resonance signal detection module is configured around a biosensor chip, which contains a dextran-coated gold surface sandwiched between layers of two different refractive indices. Light impinging on the gold surface will cause the outer shell electrons to resonate and hence result in decreased reflected light under conditions of total internal reflection. Binding of soluble analyte molecules to the immobilized ligand on the sensor chip surface will result in changes in mass and, hence, in the refractive index at the chip surface. These changes result in a decreased reflected light detected as the SPR signal and recorded as resonance units (RU) in real time. A change of 1000 RU corresponds to a change in protein surface concentration of about 1.0 ng/mm 2 . A plot of RU versus time (sensorgram) consists of an association phase, an equilibrium phase, and a dissociation phase. To analyze the interaction between p55 and glycophorin C, biotinylated GPC peptides were immobilized on a streptavidin surface (sensor chip SA, Research Grade, Pharmacia), and GST-p55 fusion proteins were passed over the immobilized peptides. The GST fusion proteins of p55 were used at a concentration of about 0.41 mg/ml. In kinetic experiments, the concentration of GST fusion proteins of p55 ranged from 1.0 to 22.0 M. All experiments were carried out at 25°C at a flow rate of 5.0 l/min except for the kinetic measurements, which were carried out at a flow rate of 3.0 l/min. The composition of the running buffer was 10 mM HEPES (pH 7.4), 150 mM NaCl, 3.4 mM EDTA, and 0.005% P20 (BIAcore buffer).
Quantification of hDlg in Human Red Blood Cell Membranes-Red blood cells were separated from other blood cells by centrifugation in a microcapillary (hemofuge). Purified red cell membranes and bacterially expressed recombinant hDlg protein of known concentrations were analyzed by 10% SDS-polyacrylamide gel electrophoresis and Western blotting using polyclonal hDlg antibodies. Antibodies were raised against a synthetic peptide of hDlg corresponding to residues 68 -80 (DRSKPSEPIQPVN) and were affinity-purified on a column of immobilized peptide. A semiquantitative estimate of the amount of hDlg in red cell membranes was made by comparing the intensity of hDlg bands in red cell membranes with the intensities of the hDlg bands of known protein concentration.
Analysis of the Mutant Red Blood Cell Membranes-Blood samples were obtained from three members of one family (JC, SS, and TL). All three members are homozygous for the Leach phenotype, and their red blood cells are completely deficient in glycophorin C (23,24). The control blood samples were selected by taking into account the shipment and age variables. After removal of the buffy coat, red blood cells were passed through the cellulose mixture to remove any residual white blood cells (25). The red cell membranes were analyzed by 10% SDSpolyacrylamide gel electrophoresis (19). Membrane proteins were transferred to nitrocellulose and probed with antibodies against hDlg.

RESULTS
We have previously shown that erythrocyte membrane protein p55 binds to the cytoplasmic domain of glycophorin C (14). To determine the specific binding site on p55, we expressed the glycophorin C cytoplasmic domain and defined segments of human erythrocyte p55 as recombinant GST fusion proteins in bacteria. The cytoplasmic domain of glycophorin C was cleaved from GST by thrombin and was further purified by size exclusion chromatography. For convenience, we will refer to the GST fusion protein containing the cytoplasmic domain of glycophorin C as GST-GPC 82-128 . The purified cytoplasmic domain of glycophorin C without GST will be designated GPC 82-128 . The molecular mass predicted from the primary structure of GPC 82-128 is 5.3 kDa. As shown in Fig. 1A, Ͼ90% of the GPC 82-128 protein migrated as a 12-kDa species on a Superdex-75 gel filtration column. Although only about 10% of the GPC 82-128 aggregated at 7.0 mg/ml protein concentration, no 5.3-kDa species of GPC 82-128 protein was detected within the range of protein concentration examined in this study (Fig. 1A). In addition, the gel-filtered GPC 82-128 protein migrated as an ϳ11-kDa species on the SDS-polyacrylamide gels in the pres-ence of 2-mercaptoethanol (Fig. 1B, lane 2), indicating that the bacterially expressed cytoplasmic domain of glycophorin C may form a noncovalent dimer. This result is, however, in contrast to previously published studies indicating that GPC exists as a monomer in the red cell membrane (24,26).
To resolve this issue, the aggregation state of GPC 82-128 protein was analyzed using a MALDI mass spectrometer. As shown in Fig. 1C, purified GPC 82-128 protein exists largely as a monomer in solution constituting 70 -90% of the total peak area (5575.25 Da). A dimeric peak corresponding to 10 -20% of the total protein was also detected (11136.6 Da; Fig. 1C). To test whether purified GPC 82-128 protein migrates as a monomer in a modified gel electrophoresis system, we analyzed GPC 82-128 by Tricine-containing SDS-polyacrylamide gels. The Tricine-containing gels are known to achieve higher resolution of proteins in the lower molecular mass range of 4 -40 kDa (27). Again, the GPC 82-128 protein migrated as a diffuse band of ϳ11 kDa (Fig. 1C, inset, lane 1). At higher amounts of GPC 82-128 protein, a diffuse but relatively weaker band of ϳ5 kDa was also detected (Fig. 1C, inset, lane 2, arrowhead).
In view of the conflicting results obtained by gel filtration/ electrophoresis and MALDI, we reevaluated the self-association state of purified GPC 82-128 by sedimentation equilibrium analysis. The solute distribution data were fitted with the exponential form of the solution to the Lamm equation for a single ideal species (22) (see "Experimental Procedures" for details). Using this model, a good fit yielded an apparent whole cell weight average molecular mass of 4340 Ϯ 230 Da, which is in close agreement with the calculated monomeric mass (5297 Da) of GPC 82-128 protein (Table I). This result is consistent with the value obtained by MALDI, which shows that GPC 82-128 exists as a monomer in solution. Here, the potential to draw erroneous conclusions by using only gel filtration and gel electrophoresis techniques is evident.
The GST-p55 fusion proteins were examined for glycophorin C binding activity using a blot overlay assay. The fusion proteins were first immobilized onto nitrocellulose and then incubated with the purified cytoplasmic domain of glycophorin C. The bound GPC 82-128 was detected immunologically using antibodies specific for the cytoplasmic domain of glycophorin C. As shown in Fig. 2, GPC 82-128 binding occurred only with GST-p55 fusion proteins containing the PDZ domain, namely 1) full-length p55, 2) N-terminal domain ϩ PDZ domain, and 3) N-terminal domain ϩ PDZ domain ϩ SH3 motif. Neither the N-terminal domain nor the SH3 motif alone bound GPC 82-128 . Similarly, GPC 82-128 did not bind a construct containing only the guanylate kinase domain. A construct containing the guanylate kinase domain, the tyrosine phosphorylation domain (TP), the protein 4.1 binding domain (4.1B), and the SH3 motif of p55 also did not show any glycophorin C binding activity. By process of elimination, the single copy of the PDZ domain in p55 appears to contain the binding site for the cytoplasmic domain of glycophorin C.
From our previous observation that GST-GPC 82-128 inhibits up to 65% of the binding between p55 and a synthetic peptide containing the carboxyl-terminal 24 amino acids (Ala-105 to Ile-128; peptide P 24 ) of glycophorin C (14), we inferred that the p55 binding site is located within the carboxyl half of the cytoplasmic domain of glycophorin C. Indeed, peptide P 24 bound to the GST-p55 fusion proteins containing the PDZ domain in a blot overlay assay (data not shown). To measure the direct binding of P 24 to the p55 PDZ domain, a GST-PDZ fusion protein was expressed in bacteria. The binding was quantified using the surface plasmon resonance technique (BIAcore). The biotinylated P 24 was immobilized to the streptavidin surface of the Biosensor SA chip. The GST-p55 fusion proteins containing Note that a significant amount of GST-GPC 82-128 protein is broken down to release free GST in the bacterial lysate before purification. Lane 2, protein sample from the 12-kDa peak (panel A) was boiled in the sample buffer containing 2-mercaptoethanol prior to electrophoresis. The asterisk indicates that GPC 82-128 migrated as an ϳ11-kDa species. C, purified 23-pmol GPC 82-128 was mixed with 2,5-dihydroxybenzoic acid and analyzed by MALDI. A similar spectrum was obtained when the cytoplasmic domain of GPC (230 pmol) was mixed with sinapinic acid. Inset, analysis of GPC by Tricine-containing SDSpolyacrylamide gel electrophoresis (10%). Note that Tricine-containing gels provide higher resolution of proteins in the lower molecular weight region (27). Again, the FPLC (Mono Q)-purified GPC 82-128 migrated as an ϳ11-kDa species as indicated by an asterisk (lanes 1 and 2). At higher concentrations (lane 2), a relatively weak and diffuse band corresponding to the monomer position of GPC 82-128 was also observed (arrowhead). either the N-terminal and PDZ domains or the PDZ domain alone ( Fig. 2A) bound to P 24 , whereas no binding occurred between P 24 and the GST-N-terminal domain (Fig. 3A). Moreover, the binding of P 24 to GST-PDZ was completely abrogated when 21 amino acids (amino acids 62-82) were deleted from the N terminus of the PDZ domain leaving a truncated PDZ domain (PDZ⌬) ( Figs. 2A and 3A). Therefore, the N-terminal 21 residues of the PDZ domain are essential for the interaction of p55 with the cytoplasmic domain of glycophorin C. These observations are consistent with the x-ray crystallographic structure of the PDZ 3 domains of hDlg and PSD-95, which shows that the amino-terminal region of the PDZ domain makes specific contact with the carboxyl-terminal end of the bound peptide and hence is important in stabilizing the PDZ domainpeptide interactions (11,12).
To identify the minimum segment of glycophorin C required to bind p55, a peptide corresponding to the carboxyl-terminal 12 amino acids of glycophorin C (Asp-117 to Ile-128) was synthesized (see "Experimental Procedures"). This peptide, designated YFI, was conjugated to biotin via its N terminus. Peptide YFI was immobilized on the streptavidin surface of an SA sensor chip, and was then tested for binding to the PDZ domain of p55 using BIAcore. Again, both fusion proteins, GST-Nterminal plus PDZ domain and GST-PDZ domain, bound to peptide YFI (Fig. 3B). As expected, no binding was detected with either GST-N-terminal domain or GST-PDZ⌬ fusion protein (data not shown). These results indicate that the carboxyl-terminal 12 amino acids of glycophorin C are sufficient for its binding to the PDZ domain of p55. The binding capacity of peptide YFI for the PDZ domain is, however, lower than that of peptide P 24 (compare the RU values of bound GST-PDZ in Fig.  3, A and B). The relatively weaker binding of the 12-amino acid peptide may result from poor flexibility due to immobilization on the chip surface.
We have recently reported that PDZ domains can be classified into two subgroups based on their ability to recognize amino acids located at the carboxyl-terminal Ϫ2 position of interacting proteins (13). Group I PDZ domains select residues with a hydroxyl side chain (Ser/Thr) at the Ϫ2-position, while group II PDZ domains select residues with an aromatic side chain (Tyr/Phe) at the Ϫ2 position. Both classes of PDZ domains prefer hydrophobic amino acids at the 0 position of interacting peptides. Using an alanine scan, we examined the role of the glycophorin C carboxyl-terminal YFI motif in p55 PDZ domain binding. We produced mutant peptides by selectively replacing the Tyr and Ile residues of the YFI motif (see "Experimental Procedures"); peptide YFA replaced Ile-128; peptide AFI replaced Tyr-126; and peptide AFA replaced both Ile-128 and Tyr-126. All peptides were conjugated to biotin at their amino termini, immobilized on the streptavidin surface of the SA sensor chip, and tested for GST-PDZ domain binding activity. As shown in Fig. 3B, the GST-PDZ domain bound only to peptide YFI and not to any mutant peptides, illustrating that amino acids Tyr-126 and Ile-128 at the carboxyl terminus of The data were fitted with the ASSOC4 model in the XL-A ORIGIN software package to obtain the apparent whole cell weight average mass (m, app). First, the optical base-line offset was set to 0 (assuming that dialysis had equalized the contribution of absorbing, nonsedimenting species at 280 nm). The mass was then floated, and the fits were good (yielding 4010 Da Յ m, app Յ 4830 Da), indicating the soundness of using this model. Then the base line was floated together with the mass (yielding 4070 Da Յ m, app Յ 4840 Da). Finally, in order to assess the effect of poorly mismatched base lines, the mass was set to 5297 Da (the mass calculated from the amino acid sequence), and the baselines were floated. Again, good fits were obtained. Extrapolation of the concentration dependence of apparent mass to infinite dilution yielded a mass of 4340 Ϯ 230 Da. [  glycophorin C are critical for its binding to the PDZ domain of p55.
For the initial attempt to quantify the interaction between the p55 PDZ domain and glycophorin C, we used the 12-amino acid YFI peptide. It should be noted that the determination of kinetic parameters by BIAcore requires regeneration of the sensor chip surface. This process removes bound analyte without affecting the immobilized ligand, so that the interaction of a fixed amount of ligand can be measured with increasing concentrations of the analyte. For unknown reasons, the bound PDZ domain on the YFI peptide-coated sensor chip could not be removed quantitatively. Application of harsher regeneration conditions resulted not only in the removal of bound PDZ domain but also of immobilized YFI peptide. Hence, we selected peptide P 24 to quantify the affinity of glycophorin C for the PDZ domain of p55. Peptide P 24 (414 RU) was immobilized on the SA chip and introduced to five different concentrations of the GST-PDZ domain (1.2-21.5 M) (Fig. 3C). Analysis of the resulting sensorgrams gave an apparent association rate constant (k a ) of 231.5 Ϯ 67 M Ϫ1 s Ϫ1 and an apparent dissociation rate constant (k d ) of 10.2 Ϯ 0.11 ϫ 10 Ϫ4 s Ϫ1 . The apparent dissociation constant K D ϭ 4.4 M was calculated from the ratio k d /k a . A schematic diagram depicting the interacting domains of p55, glycophorin C, and protein 4.1 is shown in Fig. 4.
We then examined the specificity of glycophorin C binding to the PDZ domain of p55. A GST fusion protein containing all the three PDZ domains of lymphocyte hDlg was produced (5,28). Using the blot overlay assay, no binding of GPC 82-128 was observed with either the three PDZ domains of hDlg or the full-length hDlg protein (data not shown). To supplement these in vitro results, we utilized red cell membranes of individuals with complete glycophorin C deficiency (Leach phenotype) (23,24). This phenotype was previously shown to exhibit a secondary loss of p55 and protein 4.1 (25% reduction) as compared with normal red cell membranes (29). Using an affinity-purified antibody against lymphoid hDlg, a comparable amount of hDlg protein was detected in the normal and glycophorin Cdeficient red cell membranes (Fig. 5). These results confirm that the YFI signature at the carboxyl terminus of glycophorin C determines its specificity for the PDZ domain of p55.

DISCUSSION
In this paper, we have established that the single copy of the PDZ domain in human erythrocyte p55 mediates the binding of p55 with the carboxyl terminus of the cytoplasmic domain of glycophorin C; this finding is consistent with recent characterizations of the PDZ domains in other proteins (5-9, 30 -32). Recently, using an oriented peptide library technique, we have classified PDZ domains into two subgroups based on the amino acid residues they select at the Ϫ2-position from the carboxyl end of the interacting protein (13). Group I PDZ domains recognize a binding sequence having a hydroxyl side chain on the Ϫ2 amino acid ((S/T)XV), while group II PDZ domains prefer a residue with an aromatic side chain (Phe/Tyr) at the Ϫ2-position. In vitro mutagenesis experiments have shown that replacement of either Thr/Ser at the Ϫ2-position or Val at the 0-position completely abolishes binding of the Shaker type K ϩ channels to the PDZ domain of PSD-95 (33). In the crystal structure of the PDZ 3 domain of PSD-95, the side chains of Ser/Thr and Val make distinct contacts with specific residues of the PDZ domain, stabilizing binding of the peptide to the PDZ domain (11). The p55 PDZ domain selects Tyr/Phe at the Ϫ2position and Phe at the 0 position (13). The carboxyl terminus of GPC contains Tyr at the Ϫ2-position and Ile at the 0-position. Both Ile and Phe are hydrophobic, nonpolar residues with no net charge. The presence of Ile instead of Phe in glycophorin C is consistent with the PDZ domain binding motif in the A total of 414 RU of peptide P 24 was immobilized on the streptavidin surface, and binding was carried out at a flow rate of 3 l/min at 25°C. A plot of dR/dt versus R was used to analyze the association phase. The dissociation phase was followed for 25 min; the rate of dissociation was higher initially and then slowed. This could result from either analyte rebinding to the immobilized ligand or the dimerization of GST fusion protein (5). A plot of ln(R 1 /R n ) versus time was used for dissociation phase analysis. A 20 -60-s time frame on the plot was selected for calculating the apparent dissociation rate constant (k d ) for each concentration. k d ϭ 10.2 Ϯ 0.11 ϫ 10 Ϫ4 s Ϫ1 , obtained for the highest concentration (21.5 M), was used to calculate the dissociation constant (K D ). The apparent association rate constant (k a ) obtained was 231.5 Ϯ 67 M Ϫ1 s Ϫ1 , and the dissociation constant (K D ) was 4.4 M.
inward rectifying K ϩ channel Kir2.3 (34). Similarly, the presence of Tyr instead of Ser/Thr at the Ϫ2-position in GPC is also consistent with the PDZ domain binding motif in neurexins (35). These results underscore the importance of residues at the 0-and Ϫ2-positions and support our conclusion that the mutagenesis of these residues abolishes binding of GPC to the p55 PDZ domain (Fig. 3B).
The crystal structures of the PDZ 3 domains of PSD-95 and hDlg reveal that the N-terminal 12 amino acids form a ␤1 strand and a carboxylate binding loop (11,12). Although the structure of the p55 PDZ domain is not yet known, we assume that similar structural requirements determine the binding of p55 with GPC. To this effect, a truncated PDZ domain (lacking 21 amino acids) was observed to have completely lost its binding affinity for glycophorin C (Fig. 3A). Our results are also consistent with a previous report showing that the deletion of 9 amino acids from the N terminus of the PDZ 2 domain of PSD-95 abolishes its binding to the carboxyl-terminal tail of the Shaker type Kv1.4 channel (33). These results imply that the PDZ domains of p55 homologues such as LIN-2 (36), Dlg2 (37), and Dlg3 (38) may recognize the YXI motif at the carboxyl termini of their respective ligands.
GPC is believed to exist as a monomer in the red cell membrane (24,26). This is in contrast to GPA, which is known to be a dimer in the red cell membrane (39). Our results obtained by MALDI mass spectrometry and sedimentation equilibrium analysis show that the bacterially expressed cytoplasmic domain of GPC exists as a monomer in solution (Fig. 1, Table I).
For unknown reasons, the purified GPC 82-128 protein migrates as a dimer in the gel filtration and gel electrophoresis assays (Fig. 1). It is noteworthy that the presence of immunoreactive bands corresponding to monomer and dimer forms of GPC has been previously documented (40,41). It is therefore possible that under certain conditions the intact glycophorin C may self-associate into a dimer in the red cell membrane. In any case, binding data reported in this paper reflect an interaction between monomeric GPC 82-128 and monomeric PDZ domain of p55. The affinity (K D ) of the interaction between recombinant p55 and the synthetic peptide P 24 of glycophorin C is 4.4 M as measured by the surface plasmon resonance technique. Hemming et al. (16) measured a high affinity interaction (K D ϭ 4.5 nM) between native p55 and alkali-stripped erythrocyte membrane vesicles, indicating that presently unknown factors may modulate the strength of p55-glycophorin C interaction in vivo. The known post-translational modifications of p55, such as its extensive palmitoylation and phosphorylation, are worthy of further examination.
The absence of p55 but not hDlg in the membranes of glycophorin C-deficient red blood cells (Leach phenotype) provides in vivo evidence for the specificity of p55 PDZ domain binding to glycophorin C (Fig. 5). We have previously reported secondary loss of both p55 and hDlg in the red cell membranes of a patient with complete protein 4.1 deficiency (5,29), demonstrating that protein 4.1 is necessary for the attachment of MAGUKs to the red cell plasma membrane. However, an enigma in the field of elliptocytosis has been the observation that although the red blood cells from both protein 4.1-deficient hereditary elliptocytosis individuals and glycophorin C-deficient Leach phenotype individuals lack the protein 4.1-p55-glycophorin C complex, only protein 4.1-deficient hereditary elliptocytosis red cells show complete elliptocytosis (16); Leach phenotype red cells exhibit a moderate degree of elliptocytosis (15,42). The presence of nearly 75% of protein 4.1 in Leach phenotype red cell membranes is thought to be responsible for preventing complete loss of discoid shape in these cells (15,42). The presence FIG. 4. Interacting domains of glycophorin C, p55, and protein 4.1. A, summary of the binding domains based on our previous results (14,17) and the data shown in this manuscript. The arrows connect the interacting domains. B, schematic of the ternary complex interactions in the human red cell membrane. Note that p55 and GPC bind at two distinct sites on the 30-kDa domain of protein 4.1 (16). Although p55 is shown bound to the linearized cytoplasmic domain of glycophorin C away from the plasma membrane, in vivo it may exist closely apposed to the plasma membrane because it is a heavily palmitoylated protein.
FIG. 5. Detection of hDlg in the membranes of Leach phenotype red blood cells. The arrowhead shows the position of the major 80-kDa immunoreactive band of hDlg. The asterisk indicates the mobility of the minor ϳ125-kDa immunoreactive band of hDlg. Affinitypurified anti-peptide polyclonal antibodies were used to detect the presence of hDlg in the red cell membranes (see "Experimental Procedures"). The presence of hDlg in glycophorin C-deficient red cell membranes (Leach phenotype) was confirmed in two other individuals with the Leach phenotype (data not shown). of hDlg in Leach phenotype red cell membranes may offer a new explanation for the membrane association of protein 4.1 in these cells. Based on these findings, we propose that the elliptocytic shape of the Leach phenotype red blood cells is regulated by a membrane-associated, shape-stabilizing protein complex (SSPC). Both p55 and hDlg are components of the SSPC along with protein 4.1 and glycophorin C. In the red cell membranes of patients with protein 4.1-deficient hereditary elliptocytosis, the entire SSPC is lost, resulting in complete elliptocytosis, unlike Leach phenotype membranes, which lack glycophorin C, p55, and 25% of protein 4.1 (29). The binding of hDlg to protein 4.1 may anchor the remains of the SSPC onto the red cell membrane in Leach phenotype. This hypothesis assumes that only a subpopulation of protein 4.1 regulates SSPC at the plasma membrane. The remaining protein 4.1 may independently regulate properties such as membrane deformability, which is restored by the addition of exogenous protein 4.1 in the absence of either p55 or hDlg (43,44).
The number of copies of GPC per red cell varies from 50,000 to 143,000 (45)(46)(47), depending upon the method of estimation. Previously, 80,000 copies of p55 per red cell were estimated by an immunoblotting assay (1) and reflect a minimum estimate. In that case, the amount of p55 appears to be stoichiometerically sufficient for the number of GPC molecules in the red cell membrane. This is further substantiated by the fact that p55 is quantitatively lost in the GPC-deficient red cell membranes (29). Using a similar immunoblotting assay, we have estimated 1000 copies of hDlg/red cell ghost. Although this is a minimum estimate, hDlg together with other p55-related MAGUKs such as Dlg2 (37), Dlg3 (38), and CASK (35) may link sufficient numbers of protein 4.1 molecules to the red cell membrane in the absence of GPC. Proof of this hypothesis will require identification of these MAGUKs and their membrane binding partners in the red cell membrane.
This study completes the identification of the domains involved in the formation of the ternary complex between p55, glycophorin C, and protein 4.1. Taken together with our previous findings (14,17), the data suggest that erythroid p55 may function as an adaptor protein by stabilizing the membranecytoskeletal interactions and cell shape via protein 4.1 and glycophorin C. Whether the parallels drawn from the binding interactions of p55 and hDlg in the red cell membrane will have a bearing on the interactions of MAGUKs in general remains to be determined.