Structure of the Ankyrin-binding Domain of α-Na,K-ATPase*

The ankyrin 33-residue repeating motif, an L-shaped structure with protruding β-hairpin tips, mediates specific macromolecular interactions with cytoskeletal, membrane, and regulatory proteins. The association between ankyrin and α-Na,K-ATPase, a ubiquitous membrane protein critical to vectorial transport of ions and nutrients, is required to assemble and stabilize Na,K-ATPase at the plasma membrane. α-Na,K-ATPase binds both red cell ankyrin (AnkR, a product of the ANK1 gene) and Madin-Darby canine kidney cell ankyrin (AnkG, a product of the ANK3 gene) utilizing residues 142–166 (SYYQEAKSSKIMESFK NMVPQQALV) in its second cytoplasmic domain. Fusion peptides of glutathione S-transferase incorporating these 25 amino acids bind specifically to purified ankyrin (K d = 118 ± 50 nm). The three-dimensional structure (2.6 Å) of this minimal ankyrin-binding motif, crystallized as the fusion protein, reveals a 7-residue loop with one charged hydrophilic face capping a double β-strand. Comparison with ankyrin-binding sequences in p53, CD44, neurofascin/L1, and the inositol 1,4,5-trisphosphate receptor suggests that the valency and specificity of ankyrin binding is achieved by the interaction of 5–7-residue surface loops with the β-hairpin tips of multiple ankyrin repeat units.

Tethering interactions between the cytoplasmic domains of integral membrane and other proteins, mediated by ankyrin or proteins containing ankyrin-like repeat structures, play fundamental roles in diverse biological activities including growth and development (1)(2)(3)(4)(5), intracellular protein trafficking (6 -8), the establishment and maintenance of cellular polarity (9 -12), cell adhesion (13,14), signal transduction (2,(15)(16)(17)(18)(19), and mRNA transcription (20,21). Ankyrin, including its many isoforms (reviewed in Refs. 22 and 23), is also the most ubiquitous adapter protein mediating linkage of membrane proteins with the spectrin-based skeleton, both at the plasma membrane (reviewed in Refs. [23][24][25] as well as with internal membrane compartments including the Golgi apparatus (6 -8). A characteristic feature of ankyrin is the presence of well conserved 33-residue repetitive units (ankyrin repeats) that individually or in combination bind to transmembrane proteins (reviewed in Ref. 23). Because no data on the structure of an ankyrinbinding domain in an integral membrane protein exist, it is unclear how so many diverse membrane proteins can bind specifically to a single ankyrin molecule (26,27).
The ␣-subunit of Na,K-ATPase interacts with ankyrin (11,28). This interaction stabilizes Na,K-ATPase at the plasma membrane and enables its transport from the endoplasmic reticulum to the Golgi (8). Cytoplasmic domains (CD) 1 II and III of ␣-Na,K-ATPase bind ankyrin in vitro. These sequences appear to act independently, and those in CDII (residues 140 -290) account for most of the affinity of ␣-Na,K-ATPase for ankyrin (28,29). We now identify a 25-residue minimal ankyrin-binding (MAB) sequence within CDII of ␣-Na,K-ATPase and report its three-dimensional structure. This peptide forms a loop-on-a-stalk motif well suited for interaction with one or more of the putative ␤-hairpin tips of the ankyrin repetitive unit (5). We anticipate that the ankyrin-binding structure identified here will emerge as a common feature of other ankyrin-binding proteins. Portions of this work have been presented in abstract (30).
Ankyrin Binding Assay-Each fusion protein (50 g at 1 g/ml) was conjugated to 50 l of a 50% slurry of glutathione-agarose for 1 h at 4°C with gentle rotation. Ankyrin (Ank R ) was purified by extraction of spectrin-depleted fresh erythrocyte inside-out vesicles with 1 M KCl, followed by ion exchange chromatography on DEAE cellulose (31). Ank R binding was assayed by adding 25 g to the peptide-conjugated beads in a total volume of 500 l in ABB, overnight incubation at 4°C, and analysis of the bead fraction by SDS-PAGE. Ankyrin was detected by Western blotting with specific antibodies (28). Confluent MDCK cells extracted in situ were used to prepare a high salt extractable cytoskeletal fraction (Fx2) enriched in MDCK cell ankyrin (Ank G ) (28). Conjugated beads were incubated with Fx2 (300 g of total protein), and bound ankyrin was detected as above. Other procedures and antibodies were as before (7,28).
For quantitative binding measurements, purified ankyrin was labeled with sulfo-N-hydroxysuccinimide biotin (Pierce), further purified by gel filtration, and concentrated using a Centricon ultradialysis membrane (7). Aliquots (10 g) of purified fusion peptide IIa (GST-MAB) or GST were incubated as above in ABB with varying concentrations of biotinylated ankyrin. Bound ankyrin was visualized by ECL after direct overlay with horseradish peroxidase-avidin (Vector). Relative binding was measured by densitometry of autofluorograms, with precautions taken to assure linearity of detection. Because the absolute free concentration of ankyrin could not reliably be measured using this biotinylated assay, free ankyrin was assumed to equal total ankyrin. This * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
§ These authors contributed equally to this work. approximation will systematically overestimate slightly the K d (weaker apparent affinity than the true value). Nonspecific binding to GST was subtracted prior to nonlinear regression fitting. Errors are expressed as Ϯ 2 S.D..
Structure Determination-Fusion peptide IIa was purified by high pressure liquid chromatography gel filtration into 50 mM Tris-HCl, pH 7.2, 150 mM NaCl, 2 mM ␤-mercaptoethanol. Crystals were prepared at room temperature by hanging drop vapor diffusion against a reservoir of 30% polyethylene glycol 4000, 100 mM bis-tris-propane, 150 mM NaCl, 40 M ␤-mercaptoethanol, pH 8.8 (32). Diffraction data were collected with a R-AXISII imaging plate detector mounted on a Rigaku 200HM generator at Ϫ170°C using a crystal flash frozen in crystallization buffer. Data were processed using the program DENZO and SCALEPACK (33) and were 86% complete (Ͼ2). The structure was determined by molecular replacement using the published crystal structure of GST from S. japonica (34) and the program AMORE (35) of the CCP4 program suite (33). Successive rounds of model building and simulated annealing refinement were carried out with the program X-PLOR version 3.851 (36) and the CCP4 program suite. The graphic display program O, version 5.10 (37) was used to build and correct the structure manually. Each residue of the final model was checked by the omit map.

RESULTS AND DISCUSSION
The Minimal Ankyrin-binding Domain of ␣-Na,K-ATPase-Recombinant peptides derived from CDII of rat ␣-Na,K-ATPase (28) were prepared as fusion proteins with GST and assayed for their ability to bind Ank R (from red cells) or Ank G (from Madin-Darby canine kidney cells) (Fig. 1). All peptides were of the predicted molecular mass, soluble, and readily purified (Fig.  1B). Peptide IIa (residues 142-166 of ␣-Na,K-ATPase numbered according to Ref. 38) retained full binding activity to both Ank R and Ank G and constituted the MAB sequence. CDII peptides not encompassing MAB, such as peptide IIc (residues 168 -286), were devoid of activity.
GST-MAB Binds Ankyrin with High Affinity-Prior studies have detected Na,K-ATPase affinities for ankyrin in the range of 50 -2600 nM (11,39,40). Of interest was whether GST-MAB bound ankyrin with comparable affinity. Purified ankyrin (Ank R ) was biotinylated, and its binding to GST-MAB or GST was measured (Fig. 2). Nonlinear regression analysis based on a bi-molecular binding model indicated that ankyrin specifically bound GST-MAB with a K d ϭ 118 Ϯ 50 nM, in agreement with the affinity of intact Na,K-ATPase for ankyrin. Thus, although other regions of ␣-Na,K-ATPase such as the ALLK motif identified in CD3 may contribute to its interaction with ankyrin (28,29), the 25 residues identified here are sufficient and bind specifically to ankyrin but not to other proteins present in the whole kidney lysates (Fig. 1). GST alone was devoid of significant binding activity (Fig. 2). In separate studies we have also demonstrated that loss of these 25 residues in ␣-Na,K-ATPase abrogates its binding to ankyrin in vivo (41).
Structure Determination of MAB-Attempts to crystallize ␣-Na,K-ATPase CDII or MAB alone were not successful. Because the GST-MAB fusion peptide contained an ankyrin-binding domain that was fully active, carrier-mediated crystallization was used to obtain structural information (32). The active GST fusion peptide containing MAB (construct IIA, Fig. 1, and  Fig. 3) yielded ordered crystals (space group of P4(3)2(1)2, with a ϭ b ϭ 92.17 and c ϭ 57.57). Diffraction of this crystal at Ϫ170°C yielded a data set that was 86% complete (Ͼ2) and allowed the determination of the complete structure of the GST-MAB fusion peptide with an R-factor of 19.8 and an R free of 36.5 (1% of total data) at 2.6 Å resolution (Fig. 3), using a molecular replacement strategy based on the known crystal structure of S. japonica GST (SjGST) (which has one molecule in the asymmetric unit and 41% solvent) (34). The refined model displayed a continuous main chain electron density and consisted of two domains, one representing SjGST and the other MAB (Fig. 3). Each residue of MAB was verified with an electron density omit map. Both the GST and MAB domains were well defined when the total structure was checked with the 2F o Ϫ F c map, including the side chains of Leu 118 , His 215 , and Lys 218 , residues disordered in the published structure of SjGST (34). Backbone residues 1-210 of SjGST superimposed on those of SjGST-MAB with a RMS deviation of 1.02 Å. Differences between SjGST and SjGST-MAB were found in the position of Met 1 and the positions of residues 211-218, which flank a region of disorder in SjGST (but not in SjGST-MAB). There were no discernible hydrophobic or salt interactions between GST and MAB. A single H-bond was detected between Arg 224 (residue 4 of the linker sequence) and Gln 248 (MAB, corresponding to Gln 163 in ␣-Na,K-ATPase); this bond does not alter the SjGST backbone conformation relative to SjGST alone. The three Gly residues (at positions 211, 212, and 213) in GST, together with two Pro residues at positions 216 and 217 and the Pro 223 -Arg 224 -Gly 225 -Ser 226 of the linker sequence (linker positions 3-6) collectively appear to well insulate MAB from secondary structural perturbations arising in SjGST. The crystal structure of the fusion peptide thus displays two well separated and independently folded motifs, lending confidence that the structure of MAB as revealed in the fusion protein crystal is valid. Finally, it is clear from the graphical display that four MAB units are packed in each unit cell. Although theoretically the MAB conformation might be influenced by the packing force, these forces are usually quite small. Taken together, these considerations argue strongly that the conformation of MAB will not be influenced by the presence of SjGST and that carrier-mediated crystallization of active ankyrinbinding peptides fused with SjGST may represent an important and general approach to identifying the structural determinants of ankyrin binding activity in a variety of proteins.
The Loop-on-a-Stalk Structure of MAB-Detailed analysis of the ankyrin-binding domain within SjGST-MAB reveals an antiparallel double ␤-strand flanking a loop composed of the seven residues MESFKNM (residues 152-158 of ␣-Na,K-ATPase) (Fig.  4). The overall structure of MAB is suggestive of a loop on a stalk. This loop is amphipathic, presenting a hydrophobic face composed of two methionines and one phenylalanine and a hydrophilic face composed of Glu-Ser and Lys-Asn on the opposite side. A dipolar interaction between Ser 149 (position 234 in the crystal) and Gln 163 (crystal position 249) of the antiparallel ␤-strands stabilizes the stalk. Presumably, in intact ␣-Na,K-ATPase, flanking sequences would further stabilize the stalk and probably alter the positioning of the terminal residues in MAB. A search of nucleotide and protein sequence banks (GenBank TM , Swiss-Prot) revealed exceptional conservation of MAB across species and between isoforms of ␣-Na,K-ATPase (Fig. 5). Sequences partially homologous to MAB also exist in gastric H,K-ATPase (42), which associates with ankyrin in gastric parietal cells (43). No homologous sequences were noted in other well documented ankyrinbinding proteins (including the erythrocyte anion exchanger, the amiloride-sensitive sodium channel, the voltage-sensitive sodium channel, the Na ϩ /Ca 2ϩ exchanger, CD44, neurofascin, and IP3-R).
These findings represent the first available data on the mo- lecular structure of the ankyrin-binding site in an integral membrane protein, and provide significant insight into the mechanisms by which different ligands may interact with ankyrin. Based on the crystal structure of p53 binding protein 2 (p53bp2), each ankyrin-repeat is predicted to assume a novel L-shaped structure consisting of a ␤-hairpin followed by two antiparallel helices (5). The plane of the ␤-sheet is perpendicular to the helices, and the ␤-hairpin is mostly solvent exposed (Fig. 5). Multiple repeats form a core structure in which the ␣-helices occupy the interior, and the structure is stabilized by a continuous anti-parallel ␤-sheet formed between neighboring repeats and by extensive intra-and inter-repeat side chain hydrogen bonds. A unique feature of this structure is the array of potential binding sites created by the protruding tips of the ␤-hairpin turns, either singly or in combination, and by the surfaces of the ␤-sheet formed between the protruding tips. These are the least conserved portions of the ankyrin repeat sequence, and thus offer the largest potential combinatorial complexity for interacting specifically with diverse ligands (analogous to antigen recognition sites in antibodies). We envision the seven residue loop and possibly portions of the ␤-stranded stalk of MAB associating with the ␤-hairpin and sheet structures of the ankyrin repeat unit (Fig. 5). Given that the other reported ankyrin-binding domains in CD44, IP3-R, and neurofascin are also small peptides, it is likely that although lacking sequence homology to MAB, they may also assume a loop on a stalk conformation that best enables them to interact with a complimentary site on the complex ankyrin surface. Thus, the structure of MAB reported here may offer a glimpse into a general mechanism by which the profound multivalency of ankyrin is achieved. FIG. 5. MAB is an ATPase specific sequence that may interact with one or more ankyrin repeat units. A, sequence alignment of the minimal ankyrin-binding domain (peptide IIA) with all other proteins in GenBank TM and Swiss-Prot. Dashes represent residues identical to human Na,K-ATPase ␣1 subunit, residues 142-166. This sequence is well conserved among all known Na,K-ATPase ␣ subunits and is partially conserved (68% identity) in only one other documented ankyrin-binding protein (gastric H/K-ATPase). There is no homology to the ankyrin-binding sequences in CD44, IP3-R, neurofascin, or p53bp2. However, like MAB, each of the other reported ankyrin-binding sequences are short peptides. B, hypothetical model of how MAB may interact with ankyrin, drawn to scale. Each ankyrin repeat is composed of two ␣-helices and a ␤-hairpin loop (5). Multiple repeat units create a structure in which helix-helix interactions form a central core, whereas the tips of the exposed ␤-hairpin turns provide potential interprotein interaction surfaces. We envision that the 7-residue loop within MAB interacts with site(s) in ankyrin created by the tips of one or more of its ␤-hairpin turns. Because a multiplicity of potential binding pockets would be created by the 13-24 repeat units characteristic of most ankyrins, specific and unique binding sites probably also exist for other short peptide sequences such as those that bestow ankyrin binding activity on other proteins. To effect this binding, we hypothesize that these peptides will assume a loop on a stalk structure similar to that reported here for ␣-Na,K-ATPase.