Identification of a Critical Ankyrin-binding Loop on the Cytoplasmic Domain of Erythrocyte Membrane Band 3 by Crystal Structure Analysis and Site-directed Mutagenesis*

The cytoplasmic domain of erythrocyte membrane band 3 (cdb3) serves as a center of membrane organization, interacting with such proteins as ankyrin, protein 4.1, protein 4.2, hemoglobin, several glycolytic enzymes, a tyrosine phosphatase, and a tyrosine kinase, p72 syk . The crystallographic structure of the cdb3 dimer has revealed that residues 175–185 assume a β-hairpin loop similar to a putative ankyrin-binding motif at the cytoplasmic surface of the Na+/K+-ATPase. To test whether this hairpin loop constitutes an ankyrin-binding site on cdb3, we have deleted amino acids 175–185 and substituted the 11-residue loop with a Gly-Gly dipeptide that bridges the deletion without introducing strain into the structure. Although the deletion mutant undergoes the same native conformational changes exhibited by wild type cdb3 and binds other peripheral proteins normally, the mutant exhibits no affinity for ankyrin. This suggests that the exposed β-hairpin turn indeed constitutes a major ankyrin-binding site on cdb3. Other biochemical studies suggest that ankyrin also docks at the NH2 terminus of band 3. Thus, antibodies to the NH2 terminus of cdb3 block ankyrin binding to the cdb3, and ankyrin binding to cdb3 prevents p72 syk phosphorylation of cdb3 at its NH2 terminus (predominantly at Tyr-8). However, a truncation mutant of cdb3 lacking the NH2-terminal 50 residues displays the same binding affinity as wild type cdb3. These data thus suggest that the NH2 terminus of cdb3 is proximal to but not required for the cdb3-ankyrin interaction.

known as the ankyrin repeat that is involved in protein recognition (11)(12)(13)(14). Because ankyrin interacts with a highly diverse group of membrane proteins, much effort has been devoted to identifying the structural features that mediate these interactions (15)(16)(17)(18).
The major linkage between the membrane bilayer and spectrin-based cortical skeleton in erythrocytes is mediated by ankyrin binding to band 3. Because previous studies (19 -22) aimed at mapping the docking site(s) of ankyrin on the cytoplasmic domain of erythrocyte membrane band 3 (cdb3) 1 were conducted without the benefit of the crystal structure of cdb3, these investigations of necessity led to inexact conclusions regarding regions or linear sequences of cdb3 implicated in ankyrin binding. Thus, modification of Cys-201 and/or Cys-317 of the cytoplasmic domain of band 3 was found to compromise ankyrin binding (19). Monoclonal antibodies against residues 190 -203 (20) or 174 -186 (21) were also shown to block the ankyrin interaction. Ankyrin association was further found to protect residues 175-186 of cdb3 from proteolysis (21), and chimera analysis also implicated participation of residues 155-195 (22).
Similar studies have suggested a contribution from the NH 2 terminus of cdb3 in the ankyrin interaction (20). Evidence for NH 2 -terminal involvement has come from the observation that kidney cdb3, which lacks residues 1-65, exhibits no affinity for ankyrin (23,24). In addition, ankyrin association inhibits phosphorylation of tyrosine residues (predominantly Tyr-8) at the NH 2 terminus of band 3 (20). It would appear from these considerations that ankyrin may associate with two disparate regions of cdb3, a region near residues 175-190 and sequences near the NH 2 terminus.
With the recent publication of the crystal structure of cdb3 (25), it has become possible to ask more precise questions regarding the docking site of ankyrin on cdb3. In this study, we have noted that a stem-loop structure, which is conformationally similar to a proposed ankyrin-binding site on the Na ϩ /K ϩ -ATPase (26), is located within the general region defined by the earlier mapping studies of Willardson et al. (20) and Davis et al. (21). We have, therefore, decided to evaluate whether this stem-loop structure might constitute a critical conformational motif involved in ankyrin-band 3 association. Because of previous evidence for NH 2 -terminal involvement, we have also explored the nature of the participation of the region in ankyrin binding in greater detail.

MATERIALS AND METHODS
Oligonucleotide-directed Mutagenesis-Oligonucleotide-directed mutagenesis was performed using a QuikChange mutagenesis kit (Strat-agene) according to the manufacturer's instructions. The following oligonucleotides were synthesized and used for site-directed mutagenesis. For deletion of amino acids 175-185 and substitution of 2 glycines in their place, 5Ј-ggccctggggggtgtgaagggtggacagcctctgctcccccaac3-Ј (2 glycine residues are underlined). A His 6 tag was introduced at the COOH terminus of cdb3 in a pT7-7 plasmid (27) and used as the template. For deletion of amino acids 1-50, PCR was performed using two primers: forward, 5Ј-ggaattccatatgcacccgggtacccacgaggtc3-Ј (start codon and amino acids 51 are underlined); reverse, 5Ј-agaaagctttcagtggtggtggtg-gtggtggaagagctggcctgtctgctg3-Ј (stop codon and 6 histidine residues are underlined). The resulting mutant cDNAs were sequenced to verify mutations. Plasmids were then transferred into BL21 (DE3) pLysS cells for expression of the mutated cdb3 proteins.
Protein Expression and Purification-Both wild type and mutant cdb3s with a COOH-terminal His tag were expressed in the pT7-7 bacterial expression system using isopropylthiogalactoside induction for 3 h at 37°C. The His-tagged proteins were purified by Ni-affinity chromatography (Qiagen). Intact ankyrin was purified and radiolabeled using 125 I-Bolton-Hunter reagent as described by Bennett (28). The 46.5-kDa fragment of ankyrin (residues 403-827) was expressed in Escherichia coli strain BL21 (DE3)/pLysS (the expression vector was a kind gift of Dr. Vann Bennett) and purified as described by Davis and Bennett (29). The GST-tagged 46.5 kDa ankyrin construct was kindly provided by Dr. Peter Michaely at the University of Texas Southwestern, and the GST-tagged protein 4.1 construct was a generous gift of Dr. Narla Mohandas. GST-tagged proteins were purified using a glutathione-Sepharose column (Amersham Biosciences).
Structural Characterization of cdb3-The pH dependence of the Stokes radius of cdb3, which increases by more than 11 Å between pH levels 6.5 and 9.5, was evaluated as previously reported (30,31), except the fast protein liquid chromatography gel filtration column used consisted of prepacked Superdex 200HR 10/30 (Amersham Biosciences). The pH dependence of intrinsic fluorescence emission of cdb3, which more than doubles between pH levels 6.5 and 9.5, was measured at 340 nm in solutions of 50 mM sodium borate, 50 mM sodium phosphate, 70 mM NaCl adjusted to desired pH levels using an Aminco-Bowman luminescence spectrometer at an excitation wavelength of 290 nm and slit widths set at 6 nm.
In Vitro Protein-binding Assay-Purified His-tagged wild type or mutant cdb3 was mixed with increasing concentrations of ankyrin at 4°C overnight in 7.5 mM phosphate buffer containing 10% sucrose, 90 mM KCl, 10 mM imidazole, 0.4 mM phenylmethylsulfonyl fluoride, and 1 mg/ml bovine serum albumin at pH 7. Pre-equilibrated Ni-NTA beads were incubated with the mixture for 30 min at 4°C and washed six times with the same buffer. The bound complexes of cdb3-ankyrin were eluted with 200 mM imidazole. The quantity of bound ankyrin was then evaluated by either a quantitative dot-blot assay using an anti-ankyrin antibody, or by measuring the radioactive counts in the eluate whenever 125 I-Bolton-Hunter-labeled ankyrin was employed, or by measuring GST activity for GST-ankyrin fusion constructs. The GST substrate (100 mM potassium phosphate buffer, pH 6.5/1 mM 1-chloro-2,4dinitrobenzene/1 mM reduced glutathione) was added to the eluants and the absorbance at 340 nm was recorded to determine the amount of bound GST-ankyrin (32). To ensure that equivalent amounts of Histagged wild type and mutant cdb3 were bound to the Ni 2ϩ beads, eluants were assayed for bound cdb3 using an anti-cdb3 antibody. Binding of GST-30 kDa 4.1 (a GST fusion construct of the cdb3-binding domain of protein 4.1) to cdb3 was determined by the same method, except the pH of the incubation buffer was adjusted to pH 7.4.
Peptide Inhibition of Ankyrin Binding-Peptides corresponding to residues 175-185 of human cdb3 were synthesized by SynPep (Dublin, CA). An additional cysteine was added to the NH 2 and COOH termini of the peptide for subsequent formation of a disulfide bridge between the ends of the peptide in an effort to mimic the hairpin loop seen in the crystal structure. Increasing concentrations of peptide were then mixed with ankyrin prior to addition of cdb3, and the binding assay was performed as described above.

Characterization of a Deletion Mutant of cdb3
Lacking a ␤-hairpin Loop (Residues 175-185)-As noted in the Introduction, a proposed ankyrin-binding site on the cytoplasmic pole of the Na ϩ /K ϩ -ATPase (26) appears structurally similar to a ␤-hairpin loop revealed in the crystal structure of cdb3 ( Fig. 1). To evaluate whether this hairpin loop comprising residues 175-185 of cdb3 might serve as an ankyrin-binding site, the residues forming the hairpin loop were deleted and a diglycine bridge was substituted in their place. Thus, based on the 4.9 Å distance separating amino acids 175 and 185 in the crystal structure, two glycines were calculated to optimally span the gap without introducing strain into the mutated protein molecule. This deletion mutant was then expressed in E. coli and purified to homogeneity for further characterization.
To confirm that the conformation of the above deletion mutant was not globally perturbed, the reversible pH-dependent conformational change characteristic of native cdb3 was examined (30,31). Thus, as the pH is raised from 6.5 to 9.5, the Stokes radius of native cdb3 enlarges by more than 11 Å and the intrinsic fluorescence, which is highly quenched at lower pH, more than doubles. As seen in Fig. 2A, titration of the deletion mutant from pH 6.5 to 9.5 yields the expected fluorescence increase of 2-fold. Further, comparison of the Stokes radius increase of the mutant and wild type cdb3 as a function of increasing pH reveals the same dimensional changes in both proteins (Fig. 2B). These data demonstrate that the deletion mutant retains the same structural properties as wild type cdb3.
To further establish that truncation of the ␤-hairpin loop causes only a localized change in cdb3 structure, the interaction of cdb3 with protein 4.1, a second major peripheral protein ligand of cdb3 was examined. For this purpose, recombinant GST-30 kDa 4.1 was expressed and purified in E. coli. By means of a His tag pull-down assay, where purified His-tagged cdb3 was allowed to associate with GST-30 kDa 4.1 and the complex was pelleted with the help of Ni 2ϩ beads, the affinity of mutated cdb3 for 4.1 was examined. As shown in Fig. 2C, the amount of bound protein 4.1 associated with mutant cdb3, as measured from the GST activity in the pellet, was the same as the quantity of protein 4.1 associating with wild type cdb3, confirming that deletion of the ␤-hairpin loop induces no global change in cdb3 conformation. Further, glyceraldehydes-3-phosphate dehydrogenase, a glycolytic enzyme that binds to cdb3 and becomes inhibited upon binding, was also examined for any change in interaction with the mutant cdb3. As demonstrated in the customary glyceraldehyde-3-phosphate dehydrogenase inhibition assay (33), both wild type and mutant cdb3 were able to inhibit the enzyme Ͼ90% with the same inhibition constant (data not shown). Taken together, we conclude that any impact of loop deletion on ankyrin binding must be attributed to a direct modification of the attachment site of ankyrin and not to a general perturbation of cdb3 structure.
Evaluation of the Affinity of Ankyrin for the ␤ Loop Deletion Mutant of cdb3-Binding of ankyrin to wild type and mutant cdb3 was also evaluated by a His tag pull-down assay similar to that used for analysis of protein 4.1 binding. Thus, His-tagged cdb3 was incubated with a fusion construct of GST linked to the D3/D4 domains of the membrane-binding domain of ankyrin (a construct of ankyrin comprising ankyrin repeats 13-24 that has been frequently used to study ankyrin-band 3 interactions (29, 34 -36)) and any complexes formed were pelleted by collection of the His-tagged cdb3 on Ni 2ϩ beads. The content of pelleted GST-46.5 kDa ankyrin was then quantitated by measuring GST activity in the pellet. As shown in Fig. 3A, the binding curve for the association of the GST-ankyrin fusion construct with His-tagged cdb3 shows saturation with an apparent dissociation constant of ϳ400 nM. Binding was found to be specific because addition of excess unlabeled ankyrin competitively blocked the interaction and because heat denaturation of the 46.5-kDa ankyrin eliminated all association (data not shown). Importantly, binding of the 175-185 deletion mutant to ankyrin was not distinguishable from background, suggesting that the deleted loop is critical for ankyrin association. Further, when the same analysis was performed by a dot-blot assay using His-tagged cdb3 and unmodified 46.5 kDa ankyrin, a qualitatively similar result was obtained (Fig. 3B). These His-tagged wild type or mutant cdb3 was incubated with increasing concentrations of GST-46.5 kDa ankyrin overnight at 4°C in binding buffer. Binding of GST-ankyrin was then assayed by measuring GST activity in the pellet, as described in the legend to Fig. 2C. B, dot blot analysis of 46.5 kDa ankyrin binding to purified wild type and mutant cdb3. Ankyrin-cdb3 complexes were formed as described above, eluted from the Ni-NTA agarose beads with imidazole, and dot blotted onto a nitrocellulose membrane. The membrane was then probed with anti-ankyrin antibody and developed using goat anti-rabbit antibody coupled to horseradish peroxidase plus a chemiluminescent substrate. data suggest that the ␤-hairpin loop on cdb3 plays a critical role in the ankyrin interaction.
Although previous studies have demonstrated that 46.5 kDa ankyrin can bind cdb3 with high affinity (29) and even drive band 3 into tetramers, much like intact ankyrin (34), the question still remained whether the full-length 220 kDa ankyrin, with its twelve additional 33 amino acid repeats, might be less dependent on the ␤-hairpin loop for binding than 46.5 kDa ankyrin, which lacks the first half of its membrane-binding domain. To address this issue, we purified intact ankyrin from the red cell membrane and radiolabelled it with 125 I-Bolton-Hunter reagent. Following incubation with His-tagged cdb3, the resulting ankyrin-cdb3 complexes were pulled-down using Ni 2ϩ beads, and bound ankyrin was quantitated by measuring cpm. As seen in Fig. 4, intact ankyrin was also inhibited in its association with the cdb3 deletion mutant, suggesting that even the intact protein, with all 24 ankyrin repeats, relies heavily on the ␤-hairpin loop for interaction.
To further support the hypothesis that residues 175-185 participate directly and prominently in ankyrin binding, the 11 amino acid peptide corresponding to residues 175-185 was synthesized, only in this case a cysteine residue was added to each end of the peptide to allow for disulfide-mediated cyclization of the peptide. Cyclization was considered important as a means of trying to mimic the 3-dimensional structure of the ␤-hairpin loop in situ. The resulting looped peptide was then tested for its ability to inhibit binding of 46.5 kDa ankyrin to cdb3. As seen in Fig. 5, the synthetic peptide was indeed able to completely inhibit ankyrin binding at concentrations around 500 g/ml. Curiously, rupture of the disulfide bond using a reducing agent did not attenuate the ability of the peptide to inhibit cdb3 binding (data not shown), suggesting that the cyclized peptide was not highly constrained in its conformational flexibility.
Evaluation of Ankyrin Binding to the NH 2 Terminus of cdb3-A second ankyrin-binding site proposed in the literature is thought to reside at the NH 2 terminus of cdb3 (25,37). Evidence for this contention originally arose from three key observations: i) kidney cdb3, which lacks residues 1-65, shows no affinity for ankyrin (23,24); ii) phosphorylation of cdb3 on tyrosine 8 is inhibited upon addition of ankyrin (20); and iii) antibodies to the extreme NH 2 terminus of cdb3 block ankyrin binding (20). However, with the advent of the crystal structure of cdb3 (25), the interpretation of at least one of these observations no longer seems so straightforward because deletion of residues 1-65 of cdb3 removes the central strand of a large ␤-pleated sheet, which must certainly perturb the global conformation of cdb3. Therefore, to better evaluate the proposed involvement of the NH 2 terminus of cdb3 in ankyrin binding, a new deletion mutant was required that would induce no distortion in cdb3 conformation. For this purpose, the crystal structure of cdb3 was again consulted and interpreted to indicate that the NH 2 -terminal 50 residues of cdb3 are unstructured and can be eliminated without perturbing any other regions of the polypeptide (25). Therefore, residues 1-50 of cdb3 were deleted, and the resulting truncated cdb3 was evaluated for ankyrin affinity.
In contrast to our expectations from previous studies (20,23,24), cdb3 lacking residues 1-50 exhibited the same binding affinity for 46.5 kDa ankyrin as wild type cdb3 (Fig. 3), suggesting that the NH 2 terminus may not be essential for ankyrin association. Nevertheless, because the previous studies were conducted with intact ankyrin rather than truncated 46.5 kDa ankyrin, and because the earlier data still seemed compelling, it was decided to re-evaluate the affinity of the 1-50 deletion mutant of cdb3, but this time for the full-length ankyrin (i.e. the 220-kDa ankyrin might still require the NH 2 terminus of cdb3). However, as seen in Fig. 4, similar results were again obtained. Finally, to assure ourselves that the previously published studies were not flawed, we repeated the experiments using 46.5 kDa ankyrin instead of intact ankyrin and found that all three lines of evidence for NH 2 -terminal involvement could be replicated (i.e. kidney band 3 does not bind 46.5 kDa ankyrin, 46.5 kDa ankyrin inhibits tyrosine phosphorylation at the NH 2 terminus of cdb3, and antibodies to the NH 2 terminus of cdb3 block 46.5 kDa ankyrin binding (data not shown)). These results argue strongly that although residues 1-50 of cdb3 may not physically participate in ankyrin binding, they must reside sufficiently proximal to the ankyrin-binding site on cdb3 that antibody binding to the NH 2 terminus prevents ankyrin association, and ankyrin association at the ␤-hairpin loop sterically obstructs phosphorylation of cdb3 on Tyr-8. Examination of the crystal structure of cdb3 demonstrates that this steric overlap is entirely reasonable. DISCUSSION We have presented evidence that the ␤-hairpin loop comprising residues 175-185 of cdb3 constitutes a major ankyrinbinding site on the erythrocyte membrane. Although previous studies (19 -22) have suggested the involvement of a mid-re- gion of cdb3 in ankyrin binding, a more precise characterization of this binding site could not be pursued until the threedimensional structure of cdb3 became available. Thus, random deletion mutagenesis of the implicated sequences without the aid of crystallographic coordinates would have risked inducing global changes in protein conformation that might have led to nonspecific inhibition of ankyrin association. With access to the crystal structure, however, it was possible to identify and delete the crucial loop region on cdb3 without introducing distortion or strain into the rest of the protein. Abrogation of ankyrin binding by this deletion mutant, with no accompanying perturbation of either the pH-dependent conformational change or binding affinity for protein 4.1, could then be interpreted to indicate the critical involvement of the ␤-hairpin loop in ankyrin binding. This conclusion was further confirmed by the ability of the deleted peptide to directly compete for ankyrin association with cdb3.
Deletion mutagenesis of the NH 2 terminus of cdb3 further demonstrated that the first 50 residues of the polypeptide are not directly involved in ankyrin binding. However, the observations that blocking of this sequence by peptide-specific Fabs totally prevents ankyrin binding, and ankyrin binding to the ␤-hairpin loop inhibits phosphorylation of Tyr-8 argues strongly that the NH 2 terminus and the ankyrin-binding site lie near each other. Curiously, prediction of this steric interference would not have been obvious from the crystal structure because the first 55 residues of cdb3 are too flexible to be resolved and because the ␤-hairpin loop lies ϳ50 Å from residue 55 (i.e. the site where the flexible NH 2 -terminal sequence leaves the body of cdb3). Nevertheless, a close physical proximity must be real because ankyrin also competes with other peripheral proteins that bind at the NH 2 terminus of band 3 (38) 2 . Because the NH 2 terminus of cdb3 is not only required for stable interaction with protein 4.1 (38) but also for association with various glycolytic enzymes (33,39) and p72 syk (40,41), it is also conceivable that binding of ankyrin to cdb3 might be involved in regulation of other protein interactions at the NH 2 terminus.
The absence of a requirement for residues 1-50 in ankyrin binding raises the question of why kidney cdb3, which lacks residues 1-65, fails to bind ankyrin. Analysis of the crystal structure, however, demonstrates that residues 57-65 constitute a central strand in an eight-stranded ␤ sheet that extends through the middle of cdb3. Although deletion of residues 1-50 can be argued to have little or no impact on protein conformation, removal of residues 1-65 should force rearrangement of the packing of the entire domain. Thus, based on crystallographic considerations, kidney cdb3 should have a somewhat different conformation from erythrocyte cdb3. Consistent with this speculation, it has been reported that kanadaptin interacts with kidney, but not erythrocyte band 3 (42), whereas the glycolytic enzymes, ankyrin, and protein 4.1 associate with erythrocyte but not kidney band 3 (24).
It is interesting to note that the binding properties of intact ankyrin to band 3 are reasonably well replicated by the binding properties of 46.5 kDa ankyrin to cdb3. Not only do both polypeptides require association at the ␤-hairpin loop, but neither preparation appears to involve docking with the NH 2 terminus, even though both are inhibited by Fabs directed at this sequence, and both prevent phosphorylation of Tyr-8. Previous studies (35) have suggested that ankyrin contains one cdb3-binding site on its second set of 6 ankyrin repeats (subdomain 2) and another binding site somewhere within repeats 13-24 (subdomains 3 and 4, i.e. the 46.5-kDa ankyrin. Based on the above similarities in binding properties, it is tempting to speculate that most of the interfacial contacts may occur between subdomains 3 and 4 and cdb3 and that docking with subdomain 2 may involve a more limited surface area. Comparison of the proposed Na ϩ /K ϩ -ATPase and cdb3-binding sites for ankyrin might suggest that a ␤-hairpin loop constitutes a common ankyrin-binding motif on the cytoplasmic domains of multiple membrane-spanning proteins. Even though the loop on cdb3 shares no sequence homology with the loop on the Na ϩ /K ϩ -ATPase, cdb3 still inhibits the binding of ankyrin to the Na ϩ /K ϩ -ATPase (43). We interpret this to suggest that a conformational fit rather than sequence complementarity at the binding interface contributes primarily to complex formation. Indeed, our inability to compromise ankyrin-binding affinity by mutating both charged residues in the ␤-hairpin loop (data not shown) is consistent with the more dominant role played by conformational fit in stabilizing this interaction. As other ankyrin-binding structures become available, it will be interesting to examine whether similar exposed loops constitute the sites of association.