Crystal structure of a 12 ANK repeat stack from human ankyrinR

Peter Michaely, Diana R. Tomchick, Mischa Machius, Richard G.W. Anderson

Author Affiliations

  1. Peter Michaely*,1,
  2. Diana R. Tomchick2,
  3. Mischa Machius2 and
  4. Richard G.W. Anderson1
  1. 1 Department of Cell Biology, University of Texas Southwestern Medical Center, Dallas, TX, 75235‐9039, USA
  2. 2 Department of Biochemistry, University of Texas Southwestern Medical Center, Dallas, TX, 75235‐9039, USA
  1. *Corresponding author. E-mail: peter.michaely{at}
View Full Text


Ankyrins are multifunctional adaptors that link specific proteins to the membrane‐associated, spectrin–actin cytoskeleton. The N‐terminal, ‘membrane‐binding’ domain of ankyrins contains 24 ANK repeats and mediates most binding activities. Repeats 13–24 are especially active, with known sites of interaction for the Na/K ATPase, Cl/HCO3 anion exchanger, voltage‐gated sodium channel, clathrin heavy chain and L1 family cell adhesion molecules. Here we report the crystal structure of a human ankyrinR construct containing ANK repeats 13–24 and a portion of the spectrin‐binding domain. The ANK repeats are observed to form a contiguous spiral stack with which the spectrin‐binding domain fragment associates as an extended strand. The structural information has been used to construct models of all 24 repeats of the membrane‐binding domain as well as the interactions of the repeats with the Cl/HCO3 anion exchanger and clathrin. These models, together with available binding studies, suggest that ion transporters such as the anion exchanger associate in a large central cavity formed by the ANK repeat spiral, while clathrin and cell adhesion molecules associate with specific regions outside this cavity.


Ankyrins are protein adaptors that bridge between the spectrin–actin cytoskeleton and proteins involved in ion transport, cell adhesion and membrane trafficking. Linkages with ion transporters represent the largest class and include associations with the Cl/HCO3 anion exchanger, the Na/Ca exchanger, Na/K ATPase, IP3 receptor, ryanodine receptor and voltage‐gated Na channels (Bennett and Stenbuck, 1979; Nelson and Veshnock, 1987; Srinivasan et al., 1988; Joseph and Samanta, 1993; Li et al., 1993; Bourguignon et al., 1995; Festy et al., 2001). These linkages serve to anchor the spectrin–actin cytoskeleton on the membrane and to concentrate these transporters in specialized membrane domains (reviewed in Bennett and Baines, 2001). The ankyrin linkages with cell adhesion molecules such as the L1 family (Davis et al., 1993) enhance cell adhesion and facilitate the incorporation of cells into tissues (Hortsch et al., 1998; More et al., 2001). Ankyrins also bind clathrin and participate in membrane trafficking at the level of both coated pit budding (Michaely et al., 1999) and trafficking of specific membrane proteins (Tuvia et al., 1999).

Metazoan cells employ a large number of ankyrin isoforms for these different functions. Vertebrates have three ankyrin genes (ank1–3) that code for three families of proteins (ankyrinR, B and G, respectively). Most ankyrin isoforms contain two conserved domains that mediate protein–protein interactions. A central, ‘spectrin‐binding’ domain associates with β‐spectrin, while the N‐terminal, ‘membrane‐binding’ domain binds most other proteins. Isoform specificity is determined by the variably spliced C‐terminal domain, which contains isoform‐ specific targeting information and regulatory functions that modulate the binding activities of the two protein interaction domains (reviewed in Bennett and Baines, 2001).

Of fundamental importance to ankyrin function is the extraordinary ability of the membrane‐binding domain to bind many unrelated proteins. This region is highly conserved and consists of 24 copies of the ANK repeat motif (also referred to as ankyrin repeats). ANK repeats are a common motif whose L‐shaped structure consists of two α‐helices and a long loop. Typically, repeats stack in a linear fashion to form curved assemblies that contain between four and seven repeats and associate with a single protein target (reviewed in Sedgwick and Smerdon, 1999). Ankyrins, in contrast, contain 24 repeats that interact with at least 10 distinct proteins. Proteolytic mapping suggests that the 24 repeats are divided into four smaller domains (D1, D2, D3 and D4) of six repeats each; however, it is not clear how these subdivisions relate to protein associations since many binding sites overlap the domain boundaries (Michaely and Bennett, 1993, 1995b).

Herein, we have determined the crystal structure of the D34 region of human ankyrinR, which consists of repeats 13–24 and a small portion of the spectrin‐binding domain. We find that the 12 ANK repeats in D34 are stacked in the form of a contiguous, left‐handed superhelix. The spectrin‐binding domain segment is extended and interacts with the repeat stack in a peptide‐in‐groove association. The structural information has been used to model the 24 repeats of the entire membrane‐binding domain and the associations of the D34 region with the clathrin heavy chain and the Cl/HCO3 anion exchanger. We propose that the membrane‐binding domain forms a large, spiral, hook‐like structure. Ion transport proteins such as the anion exchanger may associate with the concave surface of this hook, while clathrin and cell adhesion molecules decorate the outside of the helical hook.


Several regions of the membrane‐binding domain of human ankyrinR were utilized in crystallization attempts. As has been seen in other ANK repeat proteins (Michel et al., 2001), both the membrane‐binding domain and the proteolytically defined subdomains were prone to aggregation. Inclusion of residues 798–827 from the spectrin‐binding domain markedly improved protein solubility. Crystallization efforts therefore concentrated on constructs bearing residues 1–827 (D1234), 402–827 (D34) and 600–827 (D4) (Figure 1A). Small, poorly diffracting crystals were obtained initially with the D34 construct. Re‐engineering of the D34 construct to include six glycine residues at the N‐terminus improved protein solubility and crystal quality. Optimization of the crystallization conditions generated large crystals that exhibited the symmetry of space group P6322 with one molecule per asymmetric unit. The crystal structure was solved by the multiwavelength anomalous dispersion (MAD) method using a seleno‐methionine (SeMet) variant of D34 (Table I). The final refined model of the native protein contains residues 405–797 and 802–812 of human ankyrinR (Table I).

Figure 1.

Structure of the D34 region of ankyrinR. (A) Ankyrins typically are composed of four domains: an ANK repeat bearing, ‘membrane‐binding’ domain, a central ‘spectrin‐binding’ domain, a death domain (DD) and a C‐terminal regulatory domain. The D34 region of ankyrinR contains ANK repeats 13–24 and a short portion of the spectrin‐binding domain. Cylinders indicate the positions of the two helical segments in each repeat. Amino acids are given as one‐letter code, where Ψ indicates a non‐polar residue. Position numbering within repeats is based upon the exon–intron boundaries of the ank1 gene. The amino acid sequence of the D34 region is shown, with the repeat designation on the left and the amino acid number on the right of the sequence block. (B) Stereo view of D34. Individual repeats are rainbow colored such that repeat 13 is red and repeat 24 is violet. (C) Surface labels are given for the repeat stack. The bottom image is rotated 90° relative to the top image.

View this table:
Table 1. Data collection, structure determination and refinement

Structure of D34

The structure of D34 (Figure 1B and C) consists of a contiguous stack of 12 ANK repeats (repeats 13–24 of the membrane‐binding domain) plus a short extended segment from the start of the spectrin‐binding domain. Individual ANK repeats have the typical ‘L’‐shaped structure seen in ANK repeats from non‐ankyrin proteins (Sedgwick and Smerdon, 1999). The ‘L’ structure consists of two short, antiparallel α‐helices (positions 4–11 and 15–23 of the ANK repeat consensus sequence) followed by a long loop (positions 24–30). Individual repeats are linked by a β‐hairpin that contains amino acids at positions 31–33 from one repeat and positions 1–3 of the next repeat (Figure 2A). The repeats generate a linear stack such that the α‐helices of adjacent repeats form a four‐helical bundle (Figure 2B). The long loop covers one end of the helical bundle, while the β‐hairpin extends away from the bundle, perpendicular to the axis of the first helix and the repeat stack. At the C‐terminus of the repeat stack, the spectrin‐binding domain segment adopts an extended conformation that folds back and binds in a cleft formed by repeats 20–24.

Figure 2.

Consensus residues of the ANK repeat. (A) The conformation of conserved residues of the ANK repeat consensus sequence is shown, oriented along the axis of the repeat stack. The α‐helical segments are depicted as ribbons (green), and the non‐helical segments are depicted as a C trace (light gray). Conserved histidines are blue, aspartates and asparagines are red, threonines are yellow, and aliphatic residues are black. Each repeat has four solvent‐exposed surfaces as indicated. (B) The helical bundle core between repeats 16 and 17 is shown. Shown in light gray are Ala510, Ala511, Val519 and Val551 at positions 9, 10 and 18 of repeat 16 and position 17 of repeat 17, respectively. These residues are situated below the dark gray shaded Leu507, Leu523 and Leu555 at positions 6 and 22 of repeat 16 and position 21 of repeat 17, respectively. An image of the D34 structure is given between the panels for reference.

The repeat stack

All repeats within D34 have 33 amino acids and show strong homology with the consensus sequence, ‐G‐TPLHΨAA‐‐GH‐‐ΨV‐ΨLL‐‐GA‐‐D/N‐‐‐, where Ψ designates aliphatic residues. Table II and Figure 2B summarize how the consensus residues define the boundaries of the two helices, stabilize the conformation of individual repeats and form the interfaces between repeats.

View this table:
Table 2. Selected non‐core contacts

The threonine at position 4 of the consensus sequence (Thr[4]), Pro[5], Gly[13], Gly[25] and Ala[26] determine the boundaries of the helices. Thr[4] is the N‐terminal capping residue of the first helix and frequently forms a hydrogen bond with the backbone amide of the third residue of the helix. Pro[5] starts the first helix most probably because its dihedral angle (φ) is restricted by the proline ring to a value near the optimal φ for helices (Aurora and Rose, 1998; Richardson and Richardson, 1988). Gly[25] and Ala[26] constitute a common C‐terminal helix cap (Edwards et al., 1987). Glycines generally are considered to be helix‐disrupting residues and Gly[13] most probably promotes the formation of the short loop that connects the first and second helices.

Residues at positions 9, 16, 24, 27 and 30 appear to stabilize the fold of individual repeats. Positions 16 and 24 frequently contain negative and positive charges, respectively, and probably serve to stabilize the second helix by neutralizing its dipole moment (Shoemaker et al., 1987). Ala[9] and hydrophobic residues at position 30 appear to help orient the two helices relative to each other. Position 27 stabilizes the unusual conformation of the long loop that separates the second helix from the β‐hairpin by forming a hydrogen bond to the backbone amide of position 29.

Residues at many conserved positions form the inter‐repeat contacts that join adjacent repeats to the repeat stack. The contacts made by these positions can be classified into three groups. Centermost are the van der Waals contacts made by residues that form the hydrophobic core of the four‐helix bundles. These residues include Leu[6], Ala[10], Leu[21], Leu[22] and mid‐sized hydrophobic residues at positions 17 and 18 (Figure 2B). A second layer of van der Waals contacts made by hydrophobic residues at positions 8, 20, 26 and 28 surrounds this central core (Table II). Finally, an outer layer of hydrogen bonds formed by His[7], His[14], Asn/Asp[29] and residues of the β‐hairpin provide additional inter‐repeat connections (Table II).

The 12 ANK repeats in D34 stack side by side to form a 125 Å long, left‐handed superhelix (Figure 1B). Each repeat is twisted counter‐clockwise 2–3° relative to the preceding repeat and contributes 13° of pitch to the superhelix. Equivalent residues in each repeat of the super helix form four distinct surfaces on the repeat stack. These surfaces consist of (i) a concave ‘bottom’ surface using residues at positions 13–16; (ii) a concave ‘ankyrin groove’ surface (Sedgwick and Smerdon, 1999) using positions 1, 3, 11, 12 and 32; (iii) a convex ‘back’ surface comprising positions 19, 23 and 24; and (iv) a broad, convex ‘top’ surface consisting of positions 27, 29, 31 and 33 (Figures 1C and 2A). The ankyrin groove surface is directed towards the axis of the superhelix, while the junction between the top and back surfaces is directed away from this axis. Residues that form these surfaces are solvent exposed and are in variable positions of the consensus sequence. The first and last repeats have additional ‘end’ surfaces that contain hydrophilic residues at consensus positions that are typically hydrophobic and protected from solvent in the helical bundles of the inner repeats.

Residues in the helical bundle cores control the superhelical curvature of the repeat stack. The ankyrin groove exhibits concave curvature because the helices of the inner row, consisting of the first helices in successive repeats, pack more tightly than the helices in the second, outer row. The packing difference is due largely to residues at positions 10, 17 and 18. The small side chains of residues at position 10 allow the first helices to pack closer than the mid‐sized residues at positions 17 and 18 in the second helices (Figures 1C and 2). The curvature about the bottom surface is due to equivalent positions in successive repeats being farther apart near the top surface than at the bottom surface. The shorter spacing at the bottom is determined by the relatively small side chains of residues at positions 10, 17 and 18 as compared with the larger leucines at positions 6, 21 and 22 at the top (Figures 1C and 2).

ANK repeat interactions

In the D34 structure, the ankyrin groove surface of repeats 20–24 associates with the first 11 residues of the spectrin‐binding domain. This 11 residue segment binds in an extended conformation, antiparallel to the repeat stack (Figures 1B and 3). The interaction is mediated by a number of van der Waals contacts involving His802, Met804, Pro807 and Val810, as well as polar interactions involving Ser805, Glu808 and Asp811. Ser805 plays a central role, forming both main chain and side chain hydrogen bonds with Tyr702, Gln740 and Gln743 on the repeat stack.

Figure 3.

Spectrin‐binding domain fragment. Shown is a stereo pair of the interaction between residues 802–812 and the repeat stack. Residues 802–812 are shown as sticks on the ankyrin groove surface of repeats 19–24. Atoms are colored gray for carbon, red for oxygen, blue for nitrogen and green for sulfur. The displayed surface is colored by electrostatic potential from −10 kT (red) to +10 kT (blue). The electrostatic surface potential was calculated and rendered in the program GRASP (Nicholls et al., 1991).

The D34 region of ankyrins associates with at least five proteins: the clathrin heavy chain, the Cl/HCO3 anion exchanger, Na/K ATPase, the L1 family of cell adhesion molecules and sodium channels of the voltage‐gated and amorilide‐sensitive family (Srinivasan et al., 1992; Michaely and Bennett, 1995a, 1995b; Thevananther et al., 1998; Michaely et al., 1999). These proteins do not share sequence homology and, therefore, do not use a common motif for D34 binding. In the following, we have focused on the associations with clathrin and the Cl/HCO3 anion exchanger (band 3), because structural information exists for the ankyrin‐binding regions of these two proteins (ter Haar et al., 1998; Zhang et al., 2000).

The association of D34 with the cytoplasmic domain of the anion exchanger (CDB3) is well characterized. Antibody competition and protein chimera studies indicate that the interaction surface on CDB3 involves residues 1–75, 118–141, 155–160 and 174–193 (Davis et al., 1989; Willardson et al., 1989). These residues form a slightly convex surface on the CDB3 dimer, which folds as a seven‐stranded α/β sandwich and dimerizes by means of a shared α‐helical motif (Zhang et al., 2000). Both repeats 13–18 and 22–23 of D34 are required for binding to CDB3 (Davis et al., 1991; Michaely and Bennett, 1995a).

The ankyrin interaction with clathrin is less well characterized. Previously, we identified an interaction between repeats 19–24 of ankyrin (D4) and the N‐terminal domain of clathrin heavy chain (CTD) (Michaely et al., 1999). The CTD contains a β‐propeller‐like arrangement of seven β‐sheets followed by an extended series of α‐helices. To map the ankyrin‐binding site better on the CTD, we expressed various truncated forms of the CTD and tested their ability to interact with D4 using a far‐western assay system (Davis et al., 1991; Michaely and Bennett, 1993) (Figure 4A). The entire β‐propeller region (residues 1–332) of the CTD was required to interact with D4. Removing as few as 14 amino acids from the N‐terminus was sufficient to abolish binding. Deletion of residues from either terminus disrupts the final β‐sheet and may also affect the normal organization of the β‐propeller.

Figure 4.

Binding site mapping. (A) The upper portion presents far‐western blots (see Materials and methods) showing the relative ability of GST fusions with clathrin fragments to associate with the D4 region. The numbers above each lane indicate the amino acids of clathrin present in each fusion. The lower portion of the panel provides a pictorial representation of these regions. On the left is an amino acid marker together with a diagram indicating the location of the β‐propeller (box) and α‐helical (line) segments of the CTD. (B) In vitro binding assays were performed using 50 nM 125I‐labeled CTD1–494, D34‐ (closed circles) or ARH‐ (open circles) coupled Sepharose beads, and increasing concentrations of CDB3 (closed circles) or D34 (open circles). ARH (autosomal recessive hypercholesterolemia protein) contains a canonical clathrin box sequence, while the D34 region does not. All assays were carried out in triplicate with standard error shown with error bars. Values shown represent specific binding only.

Clathrin also binds proteins that contain the clathrin box sequence, LL(D/N)L(D/E), via a cleft on the β‐propeller between the first and second β‐sheets. To determine whether D34 binds near this region, we tested the ability of D34 to inhibit CTD binding to the clathrin box containing autosomal recessive hypercholesterolemia protein (ARH). Figure 4B shows that D34 is a good competitive inhibitor of CTD binding to ARH. In contrast, CDB3 did not compete with CTD for D34 binding. These results suggest that the D34‐binding site on clathrin is near the first and last β‐sheet of the β‐propeller, while the clathrin‐binding site on D34 involves repeats 19–24.

We used a protein‐docking program suite, 3D‐Dock, to model the interactions of the D34 region with CDB3 and the β‐propeller of clathrin (Gabb et al., 1997). Potential docking modes were identified based upon surface and charge complementarity. Probable models were then checked against steric and biochemical criteria. The model with the highest score for the clathrin association is shown in Figure 5A. The proposed contact is between the first and last β‐sheets of the CTD and the top surface of the last three repeats of D34. The interface shows a high degree of charge complementarity consistent with the observation that the interaction is sensitive to salt. We used a similar approach to look at potential interactions between D34 and CDB3. The modeling suggests that bottom and ankyrin groove surfaces of the ANK repeat stack associate with the CDB3 dimer on the surface opposite the dimer interface (Figure 5B). The proposed contact surface on CDB3 consists of the third helix and the loops that connect the first and second β‐strands, the fifth β‐strand and the second helix, the sixth and seventh β‐strands, the sixth helix and the eighth β‐strand, and the ninth and tenth β‐strands. This interaction surface buries >1500 Å2 and contains few potential salt bridges, which is in agreement with a high affinity interaction between D34 and CDB3 that is not salt sensitive (Michaely and Bennett, 1995a).

Figure 5.

Modeled binding partners. (A) The model for the complex of repeats 19–24 of D34 (green) with the CTD (purple). (B) The association of D34 (green) with the CDB3 dimer (gray and purple). Residues predicted to bind ankyrin are shown on CDB3 in purple.

Modeling of the ankyrinR membrane‐binding domain

In addition to interactions made by repeats 13–24 of the D34 region, several associations are known to require repeats 1–12. We have modeled the entire set of 24 repeats by extending the D34 superhelix to include all 24 ANK repeats, retaining the 13° pitch for each repeat. The model predicts that the entire membrane‐binding domain forms a superhelical spiral with an axial length of 132 Å and radius of 45 Å (Figure 6A). The 800 residues of the full repeat stack provide a surface area of >27 000 Å2 for protein interactions. The wide spiral of the modeled membrane‐binding domain has a more globular structure than the D34 construct, consistent with previously published hydrodynamic measurements (Michaely and Bennett, 1993). One interaction made by repeats 7–12 (D2) is an association with a second CDB3 dimer (Michaely and Bennett, 1995a). In Figure 6B, we extended the D34–CDB3 model to encompass all 24 repeats of the membrane‐binding domain and both dimers of the CDB3 tetramer. Significantly, the combined model positions the D2 region so that it can associate with the second CDB3 dimer in the tetramer. The D2 association is salt sensitive, and the modeled association between D2 and the second CDB3 dimer includes several potential salt bridges.

Figure 6.

Membrane‐binding domain. (A) A stereo view of a model of the membrane‐binding domain viewed along the spiral axis with the C‐terminus closest to the viewer. Repeats are rainbow colored such that repeat 1 is red and repeat 24 is violet. (B) An extension of the D34–CDB3 model to include all 24 repeats of the membrane‐binding domain and all four subunits of the CDB3 tetramer. The membrane‐binding domain is shown as a green ribbon, while the four subunits of the tetramer are shown as red, blue, cyan and pink ribbons. The D34 region of the membrane‐binding domain interacts with the red subunit.


We have determined the crystal structure of ANK repeats 13–24 and a small portion of the spectrin‐binding domain of human ankyrinR. The 11 amino acid portion of the spectrin‐binding domain binds through a peptide‐in‐groove interaction with the C‐terminal ANK repeats. ANK repeats provide the principle sites of protein interaction on ankyrins, and we modeled the interactions of D34 with the clathrin heavy chain and CDB3. In D34, the individual ANK repeats adopt the typical ‘L’‐shaped conformation similar to that observed in structures of ANK repeats found in non‐ankyrin proteins (reviewed in Sedgwick and Smerdon, 1999). Models based upon biochemical and hydrodynamic data had suggested that the 12 ANK repeats of the D34 region of ankyrinR were divided into two subdomains of six repeats each. The D34 structure now reveals that these 12 ANK repeats form a contiguous extended stack that provides an extensive surface area for protein interactions.

A striking feature of the ANK repeats in ankyrinR is their uniformity in both amino acid sequence and three‐dimensional structure. Most repeats in ankyrins and all repeats in the D34 region consist of exactly 33 amino acids with strong adherence to the consensus sequence. Because most consensus residues provide inter‐repeat contacts, the strong conservation with the repeat consensus sequence results in extensive and uniform connectivity between repeats (Figures 1 and 2; Table II). Extensive connectivity may be required for ankyrins to fulfill their role as the primary linker protein between the spectrin–actin cytoskeleton and membranes. Mechanical deformations of cells and tissues most probably place stress on this linkage. The abundance of inter‐repeat contacts may protect the repeat stack from mechanical unfolding. In addition, the uniform nature of the inter‐repeat contacts may allow differential splicing to remove individual repeats without compromising the integrity of the repeat stack. All ANK repeat‐encoding exons of vertebrate ankyrins have either one or two repeats with exon boundaries exactly in register with position one of the repeat consensus sequence (Gallagher et al., 1997). Because ANK repeats form protein‐binding sites using multiple repeats, deletion of individual repeats by alternative splicing has the potential to produce novel binding sites analogous to V(D)J joining in immunoglobulin genes (reviewed in Grawunder et al., 1998).

A second distinctive feature of the ANK repeats of D34 is the curvature of the repeat stack. ANK repeat stacks of different proteins have unique curvatures that presumably are adapted to associate best with specific protein ligands. Comparison of r.m.s. deviations between individual repeats of several proteins shows that curvature is not controlled by variations in the orientation of the two helices (Venkataramani et al., 1998; data not shown). Instead, the relative side chain volumes of residues that form the helical bundle cores of adjacent repeats determine the geometric parameters of the curvature. Because the repeats of D34 are highly similar, the contribution of individual residues within the helical core can be assessed for their effect on curvature. Examination of core residues suggests that the relative side chain volumes of residues at positions 10, 17 and 18 largely determine the curvature of the repeat stack, with position 10 playing a central role. Normally, this position contains alanine or, to a lesser extent, serine or cysteine. When larger aliphatic residues occupy position 10, the curvature of both the ankyrin groove surface and the bottom surface is reduced. An example of this effect can be seen at the interface between repeats 18 and 19 in D34, which features a valine (Val577) at position 10 in repeat 18. Similar curvature flattening can be observed in myotrophin, p19INK4D, bcl‐3, pap‐β, 53BP and IκBα (Gorina and Pavletich, 1996; Baumgartner et al., 1998; Huxford et al., 1998; Jacobs and Harrison, 1998; Yang et al., 1998; Mandiyan et al., 1999; Michel et al., 2001). Substitutions at positions 17 and 18 also affect curvature, though to a lesser extent than at position 10. Usually, positions 17 and 18 harbor mid‐sized valines and threonines, which, in conjunction with a small residue at position 10, produce concave curvature in both the ankyrin groove and bottom surfaces. In the D34 structure, we observe seven deviations from mid‐sized residues at positions 17 and 18. Six of these deviations are compensatory such that alanines at position 18 pack against larger aliphatics at position 17 in the following repeat. An example of larger residues increasing curvature about the ankyrin groove is seen at the interface between repeats 22 and 23. In this way, core residues at positions 10, 17 and 18 modulate the topology of the repeat stack and thereby position solvent‐exposed residues for associations with specific protein targets.

Binding interactions

ANK repeats associate with a wide variety of protein sequences through contact residues that can be presented on any surface of the repeat stack. High resolution structures have been determined for three protein complexes involving ANK repeat proteins: INK4 inhibitors and CDK6; GABPβ and GABPα; and IκBα and NF‐κB (Batchelor et al., 1998; Brotherton et al., 1998; Huxford et al., 1998; Jacobs and Harrison, 1998; Russo et al., 1998). Each association is distinct and involves residues on the end (IκBα), top (IκBα), bottom (INK4) and ankyrin groove surfaces (IκBα, INK4 and GABPβ). Additionally, the back surface has been observed to form intramolecular contacts in the SWI6 and PAPβ structures and may participate in the intermolecular contacts of some ANK repeat associations (Foord et al., 1999; Mandiyan et al., 1999).

The ankyrin groove surface may be a frequent site of protein interaction for several reasons. First, this surface contains many variable residues that presumably are important for binding site specificity. Secondly, this surface has the broadest range of curvature variation. Finally, the ankyrin groove appears to be capable of interacting with a wide variety of protein secondary structural elements including loops (GABPβ), β‐sheets (INK4), helices (IκBα) and extended strands (D34).

In the D34 structure, the ankyrin groove surface of repeats 20–24 provides the binding site for a portion of the spectrin‐binding domain. The nature and extent of this contact suggest that it may fix the orientation of the entire spectrin‐binding domain relative to the ANK repeats. This may help direct spectrin‐binding activity away from the membrane surface. A fixed orientation may also facilitate the association of ankyrin with the Na/K ATPase because high affinity binding requires both membrane‐binding and spectrin‐binding domains (Davis and Bennett, 1990; Thevananther et al., 1998).

Our modeling predicts that the ankyrin groove and bottom surfaces of D34 mediate the association with CDB3. This model is consistent with a number of features of the ankyrinR–anion exchanger interaction reported in the literature. Normally, the anion exchanger exists as a dimer in the membrane. Binding of ankyrinR brings together two dimers to form a tetramer (Che et al., 1997). In vitro binding assays have shown that the ANK repeats assemble CDB3 tetramers using two cooperatively linked sites of interaction for the CDB3 dimer. One dimer interacts with site one in the D2 region (repeats 7–12), while a second dimer interacts with site two in the D34 region (Michaely and Bennett, 1995a). The N‐terminus of CDB3 appears to be required for both the D2 and D34 interactions (Wang et al., 1995). Our proposed model predicts extensive contacts between a surface containing the N‐terminal region of CDB3 and both the bottom and ankyrin groove surfaces of D34. The β‐hairpins of D34 fit into a groove formed at the dimer–dimer interface of the CDB3 tetramer. The D2 region is directed toward the N‐terminus of the distal subunit in the second dimer. This model predicts that the positive cooperativity observed in the binding of the CDB3 dimers derives from a stabilization of the anion exchanger tetramer through a contiguous series of interactions that span >12 repeats.

A proposal for the molecular basis of ankyrin function

The D34 region of ankyrins can associate with ion transporters, cell adhesion molecules and clathrin. Ion transporters represent the largest class of ankyrin‐binding proteins. These associations appear to employ a common surface because CDB3 can compete with ankyrin binding to most sites in brain membranes and competes directly with ankyrin binding to Na/K ATPase and voltage‐gated Na channels (Davis and Bennett, 1986, 1990; Srinivasan et al., 1988). Our model for the CDB3 interaction predicts that the 24 ANK repeats of the membrane‐binding domain wrap around the CDB3 tetramer, with the bottom and ankyrin groove surfaces directed toward the tetramer surface. We propose that these surfaces are primarily responsible for interactions with ion transporters. In contrast, CDB3 is unable to compete with either L1 cell adhesion molecules or clathrin (Michaely and Bennett, 1995b; Figure 4), suggesting that these proteins bind distinct surfaces on the ANK repeats. Our models predict that clathrin associates with the top surface, distal to the CDB3 site. This region is expected to be farthest from the membrane surface, oriented toward the cytosol and in position to associate with the clathrin lattice. L1 cell adhesion molecules associate with a surface that involves both repeats 13–18 and repeats 19–24 (Michaely and Bennett, 1995b). Based upon the model for the CDB3 association, L1 molecules may associate with the top and/or back surfaces near repeats 18 and 19. Figure 7 depicts a potential arrangement of the molecular organization of ankyrin and its binding partners. This model is the first to provide a structural framework for ankyrin‐directed integration of cellular function.

Figure 7.

Model of ankyrin function. A model for the molecular organization of ankyrin and its binding partners at the membrane. Ankyrin is colored purple with its membrane‐binding domain depicted as a ribbon. This domain wraps around an ion transporter shown as a transparent green cylinder protruding from the cytoplasmic surface of the membrane (light blue). The cytoplasmic domain of an L1 cell adhesion molecule is shown as a red half sphere near the back surface of repeats 18 and 19 of the membrane‐binding domain. The spectrin‐binding domain of ankyrin is depicted as a purple sphere connected to the C‐terminal end of the membrane‐binding domain. Spectrin is shown end‐on in a white rod representation. Clathrin heavy chain is shown in an orange space‐filling representation.

Materials and methods

Crystal structure determination

cDNA encoding residues 402–827 of ankyrinR (D34) was synthesized by PCR and ligated into the pGEX kg GST expression vector (Amersham Biosystems) using the EcoRI and HindIII restriction sites. The resulting plasmid was then transformed into the DE3 strain of Escherichia coli. After expression, the D34–GST fusion was bound to glutathione–agarose, eluted by thrombin cleavage and purified to homogeneity by size exclusion chromatography on Superdex 75 (Amersham Biosystems) in buffer A (20 mM HEPES/Na pH 7.5, 500 mM NaBr, 1 mM EDTA and 1 mM DTT). The SeMet variant of D34 was expressed in the methionine‐auxotrophic E.coli strain, B834, grown in minimal medium supplemented with the natural amino acids and SeMet. Expression and purification behavior were unchanged.

Crystals of D34 were obtained at 4°C by the vapor diffusion method from drops containing 1 μl of protein (25 mg/ml in buffer A) plus 1 μl of reservoir solution [100 mM HEPES pH 7.5, 500 mM CaCl2, 7–10% (v/v) PEG400 and 5% (v/v) acetonitrile] equilibrated against 1 ml of reservoir solution. Hexagonal crystals appeared after 2 days and grew to a maximum size of 1.0 × 0.4 mm within 5 days. Crystallization of the seleno‐methionine variant was achieved under similar conditions with a protein concentration of 20 mg/ml and a reservoir solution that consisted of 100 mM HEPES/Na pH 7.5, 300 mM CaCl2, 2–5% (v/v) PEG400 and 9–11% (v/v) acetonitrile. Crystals exhibited symmetry of space group P6322 with cell dimensions of a = b = 138 Å, c = 197 Å and one molecule in the asymmetric unit (80% solvent content). Prior to data collection, native crystals were transferred at 4°C into harvesting solution [100 mM HEPES pH 7.5, 500 mM CaCl2, 10% (v/v) PEG400, 5% (v/v) acetonitrile and 250 mM NaBr]. For cryo‐protection, the concentration of PEG400 was increased in 5% increments every 30 min to a final concentration of 30% (v/v). SeMet crystals were treated similarly except that the harvesting solution consisted of 100 mM HEPES pH 7.5, 300 mM CaCl2, 2.5% (v/v) PEG400, 9% (v/v) acetonitrile and 250 mM NaBr. Crystals were then flash‐cooled in liquid propane.

The native crystals diffracted anisotropically to a maximum Bragg spacing (dmin) of 2.5 Å and isotropically to a dmin of 2.7 Å when exposed to synchrotron radiation. The average diffraction intensity is relatively strong to a dmin of 4 Å, but suffers a dramatic decrease to the high resolution limit. The structure was solved by the MAD method using the SeMet variant. The MAD experiment was carried out at beamline 19‐ID (SBC‐CAT) at the Advanced Photon Source (Argonne National Laboratory, Argonne, IL). One data set was collected at the peak energy of the selenium K‐edge absorption, and a second was collected at the inflection point energy. Data were indexed, integrated and scaled with the HKL2000 program package (Otwinowski and Minor, 1997). Intensities were converted to structure factor amplitudes and placed on an approximate absolute scale by the program TRUNCATE from the CCP4 package (French and Wilson, 1978; CCP4, 1994). Eight selenium sites were identified from the peak energy data set using the Patterson search routine in the program package CNS 1.0 (Brünger et al., 1998). Phases were calculated and refined using both SeMet data sets in the program MLPHARE (Otwinowski, 1991), resulting in an overall figure of merit of 0.24 for data between 35.0 and 2.8 Å. The resulting phases were improved further by density modification with histogram matching in the program DM (Cowtan and Main, 1998), resulting in a final overall figure of merit of 0.57. Data collection and MAD phasing statistics are shown in Table I.

Model building and structure refinement

Inspection of electron density maps and model building were performed in the program O (Jones et al., 1991). Refinement of the structure was carried out in the program package CNS 1.0 with a random 3.7% subset of all data set aside for an Rfree calculation. Several cycles of simulated annealing, conjugate gradient minimization and calculation of individual atomic displacement parameters (ADPs) using data between 20.0 and 3.5 Å, coupled with model building allowed for the construction of a complete model for residues 405–797 and 802–812. Calculation of overall anisotropic ADPs and bulk solvent correction were used throughout. Subsequent cycles of standard positional refinement and individual isotropic ADP refinement coupled with cycles of model rebuilding were carried out against all data between 30.0 and 2.7 Å. Halide ions were added where stereochemically reasonable after the protein model was complete. Due to the low resolution of the data set and the high concentration of bromides and chlorides in the crystallization and freezing solutions, these atoms were assigned as halides rather than waters. The average isotropic ADP for these atoms is similar to the values for the protein. Overall, the N‐terminal half of the structure is less well ordered than the C‐terminal half, as reflected in the higher ADPs for this region. The final model contains residues 405–797, 802–812, one bromide ion and nine chloride ions. The Rfree value is 30.3% and the Rwork value is 31.9% (see Table I).

Docking simulations

Potential modes for the binding of D34 to CDB3 (Zhang et al., 2000) and the CTD (ter Haar et al., 1998) were calculated with the program package FT‐DOCK using default parameters (Gabb et al., 1997). Briefly, one of the binding partners was treated as the ‘static’ molecule, while the other was treated as the ‘mobile’ molecule whose position was varied to generate possible docking modes. We reduced an initial list of 10 000 potential solutions by eliminating models that were not compatible with biochemical data or that led to steric conflicts.

For modeling of the D34–CDB3 complex, D34 was treated as the ‘static’ molecule, while a monomer of CDB3 was used as the ‘mobile’ molecule. Only 32 potential solutions were compatible with mutagenesis and biochemical studies (Davis et al., 1989, 1991; Willardson et al., 1989; Michaely and Bennett, 1995a). The top seven and a total of 27 of the 32 models place CDB3 on the bottom surface.

For modeling of the D34–CTD complex, the CTD was treated as the ‘static’ molecule, while repeats 19–24 (region D4, residues 600–812) of D34 were used as the ‘mobile’ molecule. All solutions were eliminated that involved the α‐helical segment of the CTD or that led to steric conflicts with repeats 13–18. Models were then screened for modes that involved residues 1–14 of the CTD and did not conflict with the proposed CDB3‐binding site.


Ankyrin constructs were chromatographed on 5–17% linear gradient SDS–polyacrylamide gels and transferred to nylon membranes. Membranes were blocked with buffer B (10 mM HEPES pH 7.5, 100 mM NaCl, 1 mM EDTA, 1 mM DTT, 0.1% Tween‐20 and 5 mg/ml bovine serum albumin) and then incubated overnight at 4°C with various regions of the CTD fused to GST in buffer B. Membranes were then washed five times with buffer B and twice with buffer C (10 mM HEPES pH 7.5, 100 mM NaCl, 1 mM EDTA), followed by incubation for 15 min on ice with 3% paraformaldehyde in buffer C. The membranes were next washed three times with buffer C and processed using standard immunoblotting techniques with a monoclonal antibody against GST (Santa Cruz), horseradish peroxidase‐conjugated secondary antibodies (Bio‐Rad) and enhanced chemiluminescence (ECL) detection (Amersham Biosystems).

Figure preparation

Unless otherwise indicated in the figure legends, all images were composed with BOBSCRIPT (2.4) (Esnouf, 1999) and rendered with POVray (3.0.2) ( GLR (0.7) was used to facilitate the figure preparation (L.Esser, personal communication).

PDB accession code

The coordinates for the D34 structure have been deposited in the Protein Data Bank under accession code 1N11. Coordinates for modeled complexes are available upon request.


We would like to thank Sandra Hill for her valuable technical assistance and Brenda Pallares for administrative assistance. The authors are also grateful for the expert aid in data collection provided by Andrzej Joachimiak and the staff of the Structural Biology 19‐ID beamline. Use of the Argonne National Laboratory Structural Biology Center beamlines at the Advanced Photon Source was supported by the US Department of Energy, Office of Biological and Environmental Research, under contract No. W‐31‐109‐ENG‐38. This work was supported by grants from the National Institutes of Health, HL 20948, GM 52016 and the Perot Family Foundation.


View Abstract