The Membrane-binding Domain of Ankyrin Contains Four Independently Folded Subdomains, Each Comprised of Six Ankyrin Repeats*

Ankyrin repeats are a 33-amino acid motif present in a number of proteins of diverse functions including transcription factors, cell differentiation molecules, and structural proteins. This motif has been shown to mediate protein interactions in the case of ankyrin as well as several other repeat-bearing proteins. In ankyrin, 24 tandemly arrayed repeats are arranged to form a globular, membrane-binding domain. This report provides evidence that the repeats in this domain fold into four independently folded subdomains of six repeats each. Limited proteolytic digestions of defined regions of the membrane-binding domain identified protease-sensitive sites, which divided this domain into subdomains of approximately six repeats each. Hydro- dynamic measurements and circular dichroism spee-troscopy of expressed subdomains confirmed that these six-repeat regions exist as folded, globular structures. The requirement of a complete set of six repeats for proper folding was determined using a series of protein constructs, which sequentially deleted repeats from the last subdomain. Deletion of even one repeat resulted in a 40% loss of a-helicity. Deletions removing three or more repeats abolished the helical signal completely. The spherical shapes of the intact domain and of the subdomains (inferred from hydrodynamic values) sug- gest that the four subdomains are organized in either a tetrahedral or square planar configuration. Two six-repeat subdomains were found to be required for high affinity association with the anion exchanger, suggesting that at least some of the protein interactions mediated by ankyrin repeats involve multiple sub- domains.

The Membrane-binding Domain of Ankyrin Contains Four Independently Folded Subdomains, Each Comprised of Six Ankyrin Repeats* (Received for publication, May 3, 1993, andin revised form, July 7, 1993) Peter MichaelyS and Vann Bennett Ankyrin repeats are a 33-amino acid motif present in a number of proteins of diverse functions including transcription factors, cell differentiation molecules, and structural proteins. This motif has been shown to mediate protein interactions in the case of ankyrin as well as several other repeat-bearing proteins. In ankyrin, 24 tandemly arrayed repeats are arranged to form a globular, membrane-binding domain. This report provides evidence that the repeats in this domain fold into four independently folded subdomains of six repeats each. Limited proteolytic digestions of defined regions of the membrane-binding domain identified protease-sensitive sites, which divided this domain into subdomains of approximately six repeats each. Hydrodynamic measurements and circular dichroism speetroscopy of expressed subdomains confirmed that these six-repeat regions exist as folded, globular structures. The requirement of a complete set of six repeats for proper folding was determined using a series of protein constructs, which sequentially deleted repeats from the last subdomain. Deletion of even one repeat resulted in a 40% loss of a-helicity. Deletions removing three or more repeats abolished the helical signal completely. The spherical shapes of the intact domain and of the subdomains (inferred from hydrodynamic values) suggest that the four subdomains are organized in either a tetrahedral or square planar configuration. Two sixrepeat subdomains were found to be required for high affinity association with the anion exchanger, suggesting that at least some of the protein interactions mediated by ankyrin repeats involve multiple subdomains.
Ankyrin repeats are a 33-amino acid motif typically present in tandem arrays of four to seven copies and are contained in a collection of proteins of diverse localization and function (reviewed in Michaely and Bennett (1992)). Repeat-bearing proteins occur in species ranging from Eubacteria to man and in cellular localizations ranging from secreted proteins to nuclear proteins. Functions of these proteins vary widely and include a-latrotoxin of black widow spider venom, proteins that control cell differentiation, transcription factors, and the cytoskeletal protein, ankyrin.
Ankyrin repeats of several repeat-bearing proteins have * This work was supported in part by National Institutes of Health Grant 537DK29808. 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. $ To whom correspondence should be addressed, been implicated in interactions with apparently unrelated target proteins and, thus, the ankyrin repeats have been proposed to function as a generalized protein binding motif (Michaely and Bennett, 1992). The NF-KB transcription factor system includes several proteins containing ankyrin repeats such as the p105 precursor of the p50 subunit of NF-KB (Ghosh et al., 1990;Kieran et al., 1990), I K B~ (MAD3) (Haskill et al., 1991), IKBP (rel-associated pp40) (Davis et al., 1991b), I K B~ (Inoue et al., 1992a), bc13 (Ohno et al., 1990), and cactus (Geisler et al., 1992). The repeats in p105 (Henkel et al., 1992), IKBP (Inoue et al., 1992b), andbc13 (Wulczyn et al., 1992;Bours et al., 1993) have been shown to interact directly with p50 subunits. A binding function has also been identified for the four repeats of the p subunit of the interferon-induced transcription factor, GABP. These ankyrin repeats are required for association with the a subunit of GABP and enhance the specificity of the DNA binding activity of the a subunit (Thompson et al., 1991;LaMarco et al., 1991).
Ankyrin repeats are involved in many of the protein interactions of ankyrins. Ankyrins are a multigene family of proteins proposed to function as adaptors between the spectrinbased, membrane skeleton and membrane proteins, which include both ion channels and cell adhesion molecules (reviewed in Bennett and Gilligan (1993)). Ankyrins have two binding domains as follows: a 62-kDa domain containing the spectrin binding site and a domain of 89-95 kDa, which is involved in membrane interactions. The membrane-binding domains are comprised of 24 tandemly arrayed ankyrin repeats and are folded into a nearly spherical structure (Davis and Bennett, 1990a). The number of repeats in ankyrin was initially reported to be 22 (Lux et al., 1990); however, using less stringent homology criteria and the exon-intron map of erythrocyte ankyrin (Tse, 1990), 24 repeats can be resolved (Michaely and Bennett, 1992). The repeats are necessary and sufficient for association of ankyrin with the anion exchanger (Davis et al., 1991b;Davis and Bennett, 1990a), the voltagedependent sodium channel (Srinivasan et al., 1992), and the nervous system cell adhesion molecule, ABGP186, which is related to L1 and neurofascin (Davis et al., 1993). The membrane-binding domain is also required for association with the Na+/K+ ATPase, although this domain alone is not sufficient for high affinity binding (Davis and Bennett, 1990b).
Unresolved issues include identifying how many ankyrin repeats are required to form a stable structure and determining how the arrangement of repeats relates to their observed binding activities. This study provides evidence that individual ankyrin repeats are not capable of folding independently and that the 24 repeats of erythrocyte ankyrin fold cooperatively into four subdomains of six repeats each. Furthermore, two six-repeat subdomains were found to be required for high 22703 affinity association with the anion exchanger, suggesting that at least some protein interactions are mediated by multiple subdomains. Finally, using the hydrodynamic properties of the subdomains, we propose a model for the compact arrangement of the subdomains in the membrane-binding domain.

Materials
Carrier-free NalZ5I was obtained from Amersham Corp. and lZ5Ilabeled Bolton-Hunter reagent, from ICN Radiochemicals. Diisopropyl fluorophosphate, leupeptin, pepstatin A, dithiothreitol, phenylmethylsulfonyl fluoride, benzamidine, isoandrosterone, NaF, thioglycolic acid, n-octylglucoside, Tween 20, a-chymotrypsin, bovine serum albumin, ovalbumin, and cytochrome c were from Sigma. Blue dextran and all chromatography matrices were obtained from Pharmacia LKB Biotechnology Inc. Staph V8 protease was purchased from Pierce Chemical Co., DNAse I from U. S. Biochemical Corp., and Taq polymerase from Perkin-Elmer Cetus Instruments. Isopropylthiogalactopyranoside was purchased from ICN Biochemicals. The pGEMEX expression vector was from Promega. Triton X-100 was from Boehringer Mannheim. Nitrocellulose paper was from Bio-Rad and Immobilon paper from Millipore.

Methods
Bacterial Expression of Protein Constructs Containing Defined Regions of the Ankyrin Sequence-Protein expression was performed using a T7 expression system (Studier and Moffatt, 1986;Davis et al., 1991a). Briefly, polymerase chain reactions were used to amplify specific regions of the ankyrin DNA sequence. These polymerase chain reaction products were then cloned into the pGEMEX vector (Promega) such that the ankyrin sequences were in-frame with the start codon of the gene 10 promoter. These constructs were designed to code for only an added methionine, alanine, and serine at the N terminus of the ankyrin sequences. Recombinant plasmids were then transformed into the expressor strain, BL21 (DEB)/pLysS. Protein expression was induced using a method previously described (Davis and Bennett, 1990a) with yields frequently reaching 100 mg/liter of bacterial culture. All expressed protein constructs were isolated using gel filtration and ion exchange chromatography. Identification of the purified proteins was determined both by N-terminal sequencing and by immunoblot analysis using a polyclonal antibody raised against erythrocyte ankyrin. In addition, constructs R20-H through R24-H (see Fig. 4) were sequenced at the DNA level. (See Figs. 2 and 4 for SDS-gels of isolated polypeptides and diagrams showing the regions encompassed by the various protein constructs.) Circular Dichroism Spectroscopy-Spectra were obtained on a Jobin-Yvon Mark V dichromatograph calibrated using isoandrosterone and linked to an apple IIe computer. Protein samples were prepared first by solvent exchange over a Superose 12 gel filtration column, followed by dialysis into a buffer containing 1 mM Tris, 400 mM NaF, pH 7.4. Protein concentrations of the dialysates were then measured using the procedure of Bradford (Bradford, 1976). After uniformly diluting the samples to 30 Fg/ml, the protein constructs were analyzed using a 0.5-cm cuvette in a thermally regulated block at 4 "C. Spectra were collected between 195 and 300 nm wavelengths at 0.2-nm intervals with a 2-s response time at each interval. Reported spectra represent numerical averages of five spectral passes.
Hydrodynamic Measurements of Folded Protein Constructs-Stokes radii were measured using a Superose 12 fast protein liquid chromatography gel filtration column equilibrated with buffer containing 1 M NaBr, 10 mM sodium phosphate, 1 mM NaEDTA, 1 mM NaN3, 1 mM dithiothreitol, and 0.05% n-octylglucoside (pH 7.4). Cytochrome c (Rs = 2.0 nm), ovalbumin (Rs = 2.8 nm), and bovine serum albumin (Rs = 3.5 nm) were used as standards. Sedimentation coefficients were estimated using rate zonal sedimentation (Martin and Ames, 1960) through 5-20% sucrose gradients in a buffer containing 10 mM sodium phosphate, 1 M NaC1, and 1 mM each of dithiothreitol, NaEDTA, and NaN3, pH 7.4. Cytochrome c ( S Z O ,~ = 1.75), ovalbumin = 3.5), and bovine serum albumin ( S Z O ,~ = 4.6) were again used as standards. Peak fractions of the gel filtration runs used to determine the Stokes radii were dialyzed into the above buffer and used as the protein samples for the sedimentation runs. Frictional coefficients were calculated using the following equations.
6~NaRssao.~ 1 -UP20.w M, = Partial specific volumes for this calculation were estimated from the amino acid sequence of the various protein constructs (Cohn and Edsall, 1943).
Procedures SDS-PAGE' was performed on exponential (3.5-17%) polyacrylamide gels (Davis and Bennett, 1984). Blot binding of la5I-labeled cytoplasmic domain of the red cell anion exchanger to immobilized ankyrin constructs (Davis and Bennett, 1990a) and the solution binding of 1251-labeled R13-H construct to ankyrin-depleted, insideout erythrocyte vesicles (Bennett and Stenbuck, 1980) were performed essentially as previously described. The native 89-kDa membrane-binding domain was isolated from chymotryptic digests of erythrocyte ankyrin as previously described (Davis and Bennett, 1990a).

Identification of Subdomain Structure in the Membranebinding Domain of Ankyrin by Limited Proteolysis-
The presence of subdomains within the membrane-binding domain of erythrocyte ankyrin was identified by protease digestion. Staph V8 protease cleaves the isolated domain at a cleavage site at glutamate 402 (Davis and Bennett, 1990a). This site is near the border between the 12th and 13th repeats and divides the 24 repeats into two subdomains of 12 repeats each. These N-terminal and C-terminal 12-repeat subdomains were expressed in bacteria and subsequently purified to homogeneity using gel filtration and ion exchange chromatography (Fig.  1). The 12-repeat domains are likely to be in native configurations since they encompass proteolytically defined regions and are compactly folded based upon hydrodynamic measurements (see below and Table I). The C-terminal 12-repeat construct was further characterized to be in a native state both structurally based upon circular dichroism data (see below and Fig. 3) and functionally based upon its affinity for membrane binding sites (see below and Fig. 8).
The 12-repeat subdomains were subjected to a further round of proteolysis (Fig. 1). Staph V8 digestion of construct N-R12 (containing the N-terminal 10 residues and the first 12 repeats) produced a 24-kDa polypeptide with the same Nterminal sequence as the N-R12 construct. The apparent molecular mass of the product indicated that it contained approximately six to seven repeats. Chymotryptic digestion of the R13-H construct (containing repeats 13-24 and a 32residue "hinge" region after the last repeat) produced three proteolytic products. The highest molecular mass product is likely to be the result of cleavage at a previously identified site at phenylalanine 795 (Davis and Bennett, 1990), which removes the hinge region. The lower two products have identical N termini and result from hydrolysis after leucine 587, near the border between the 18th and 19th repeats. The smaller of these two lower bands is likely derived from cuts at both leucine 587 and phenylalanine 795. As illustrated in Fig. ID, the positions of these proteolytic sites divide the membrane-binding domain into four subdomains of approximately six repeats each.
Definition of the Minimum Folding Unit of Ankyrin Repeats-Circular dichroism (CD) spectra of bacterially expressed constructs were used to determine how many repeats The abbreviations used are: PAGE, polyacrylamide gel electrophoresis; Rs, Stokes radius. are required to form a folded domain. CD spectroscopy was used to monitor folding, because similar amino acid sequences typically produce similar secondary and tertiary structures, and CD spectroscopy can directly probe the degree of secondary structure. Thus, if these repeat constructs are correctly folded, they should produce similar CD spectra. This prediction proved correct for protein constructs containing 6, 12, and 24 repeats (see Figs. 2 and 3). These constructs included N-R6 (containing the first 10 amino acids and the first six repeats), CR19-H (the 13 amino acids rrom the chymotryptic cleavage site to the start of the 19th repeat. rqeats 19-24, and the 32-amino acid hinge region at the end of repeat 24), R19-H (the last six repeats and the hinge region), and the 12-repeat subdomain, R13-H (the last 12 repeats and the hinge region). Unfortunately, expressed subdomains, which encompassed the N-terminal 12 repeats (N-R12), repeats 7-12 (R7-R12), and repeats 13-18 (R13-R18) were not soluble in the low salt conditions required for CD measurements and could not be examined. However, the constructs that were soluble exhibited remarkably similar spectra with calculated helicities clustered between 16 and 20% (Fig. 3). A nested series of protein constructs was prepared with sequential deletions of repeats in order to determine how many repeats are required to maintain native folding (Fig. 4). This series was derived from the six-repeat construct, CR19-H, which encompassed the region corresponding to the major chymotryptic fragment of the R13-H construct (Fig. 1). The first deletion construct, R19-H, removed the 13-amino acid region from the chymotrypsin site to the start of the 19th repeat. The remaining constructs were prepared with Nterminal deletions of one repeat (R20-H), two repeats (R21-H), three repeats (R22-H), four repeats (R23-H), and five repeats (R24-H) (see Fig. 4 for diagrams illustrating the regions encompassed by each construct).
Constructs CR19-H and R19-H, which retain a complete set of six repeats, exhibited native circular dichroism spectra (Fig. 5). Constructs R20-H and R21-H, which are missing one and two repeats, respectively, suffered a loss of signal corresponding to a greater than 40% loss of a helical content. When three repeats were removed (construct R22-H), the helical signal was lost completely; the spectra converted to that of a random coil (Fig. 5).
Constructs R20-H and R21-H, which contain five and four repeats, respectively, retain some helical signal. However, these polypeptides are less compactly folded than the sixrepeat constructs based on their behavior on gel fitration. Constructs R20-H and R21-H have Stokes radii (2.4 and 2.6 nm, respectively) greater than that of the CR19-H and R19-H constructs (2.4 and 2.3 nm, respectively) despite having smaller molecular weights (data not shown). The loss of helical content and increased Stokes radii of constructs R20-H and R21-H could be caused either by a dynamic equilibrium between a folded and unfolded state or by an unfolding of a helical portion of the repeat sequence normally stabilized by the 19th repeat. While it is possible that deletions of individual repeats within a six-repeat domain could be tolerated, these results strongly suggest that the domain structure of the repeats in ankyrin utilizes folding units of six ankyrin repeats.

Use of Hydrodynamic Properties to Estimate Molecular Shape and Oligomerization State of the Subdomains in Solu-
tion-The native membrane-binding domain has been previously shown to have a frictional coefficient of 1.2, indicating a nearly spherical shape (Davis and Bennett, 1990a). Sedimentation coefficients were estimated by rate zonal sedimentation on sucrose gradients. Stokes radii were determined by gel filtration using Superose 12. These values were then used to calculate frictional coefficients for the individual six-repeat subdomains (Table I). Calculated frictional coefficients clustered between 1.2 and 1.3, indicating that these subdomains are also nearly spherical.
Physical measurements of constructs R7-Rl2 and N-R12 suggest that subdomains can self-associate, perhaps as the result of similar contacts utilized in assembly of the normal organization of four subdomains. The 12-repeat construct, N-R12, formed a dimer in solution with hydrodynamic properties nearly identical to the native 24-repeat domain. The sedimentation coefficient of the R7-Rl2 construct was not determined due to low solubility in 1 M NaCl. However, gel filtration runs   dichroism spectra of polypeptides containing sets of six, 12, and 24 repeats. Upperpanel, circular dichroism spectra were measured as described under "Methods" for constructs N-R6, CR19-H, R19-H, R13-H, and the native membrane-binding domain. As discussed in the text and in the legend to Fig. 2, constructs N-R6, CR19-H, and R19-H encompass individual six-repeat subdomains, while construct R13-H contains both of the last two subdomains (repeats 13-24). The native membrane-binding domain contains all 24 repeats and is designated by 89kD in reference to its 89-kDa molecular mass. Lower panel, the percent cu-helicities corresponding to each construct were estimated from the measured ellipticities a t 220 nm using the algorithm of Chen and Yang (1971). mains suggest that the four subdomains are organized in either a tetrahedral or a square planar configuration. Linear arrangements of subdomains are not likely, since the resulting extended structure would be expected to have a higher frictional ratio (Fig. 6).
High Affinity Binding of Ankyrin with the Anion Exchanger A nested deletion series of protein constructs were designed through the proteolytically defined region of the R13-H construct. Following purification of these constructs, 3 pg of each protein was electrophoresed on an SDS-polyacrylamide gel and stained with Coomassie Blue (panel A ) . The name of the construct loaded in each lane is used as the lane designation with the regions encompassed by each construct illustrated in panel B. N refers to the first 10 amino acids, the regions with numbers refer to the repeat number, and H refers to the 32-amino acid hinge region. The native membrane-binding domain and the R13-H domain are shown for reference. The exact residues encompassed by each construct are as follows: construct CR19-H contains residues 588-827; construct R19-H, residues 601-827; construct R20H, residues 634-827; construct R21-H, residues 667-827; construct R22-H, residues 700-827; construct R23-H, residues 733-827; and construct R24-H, residues 766-821.
(AEl) Requires Two Six-repeat Subdomains-Previous work has shown that while repeats 22 and 23 are required for the interaction of ankyrin with the anion exchanger, the activity of ankyrin repeat constructs decreased as the number of repeats was reduced below 11 (Davis et al., 1991b). Since loss of activity in binding was correlated with the simultaneous loss of helical content as measured by CD spectroscopy, loss of secondary structure was interpreted to be the cause of the loss of binding activity (Davis et al., 1991b). The six-repeat constructs R19-H and CR19-H contain repeats 22 and 23, displayed native circular dichroism spectra, and formed compact, globular structures. Nevertheless, these polypeptides did not exhibit high affinity binding to the anion exchanger. These constructs were a t least 5-fold less active than the 12repeat construct, R13-H, in the ability to compete with native ankyrin for anion exchanger sites on ankyrin-depleted inside-  (Chen and Yang, 1971). Constructs CR19-H and R19-H, which differ only by a 13-amino acid extension a t the N terminus of CR19-H, have virtually overlapping spectra and a correspondingly equivalent percent a-helicity. Constructs R19-H and R20-H show a 41 and 48% loss of a-helical content, respectively. Deletion constructs R22-H, R23-H, and R24-H, which contain three or fewer repeats, display a nonhelical spectra with undetectable levels of a-helix.

E B B Possible arrangements
ooco Unlikely arrangement FIG. 6. A model for the arrangement of subdomains in the membrane-binding domain. Potential models for the arrangement of the four six-repeat subdomains in the membrane-binding domain must account for both the approximately spherical shapes of the individual subdomains as well as the near spherical shape of the whole domain. Models that can accommodate spherical objects into compact units include square planar and tetrahedral configurations. Linear arrangements are unlikely, since such an extended arrangement would not produce small frictional coefficients. out, erythrocyte membrane vesicles (data not shown). Also, tions of various repeat regions of the membrane-binding doin direct binding assays, association of '2sI-labeled R19-H or main (Fig. l). Deletions that reduced the number of repeats CR19-H protein constructs with membrane vesicles could not in the last subdomain disrupted the folding of this domain, be detected (data not shown). suggesting a requirement of six repeats for the folding of the A qualitative blot binding assay illustrated the relative other subdomains as well. Hydrodynamic properties and cirability of the cytoplasmic domain of the anion exchanger to cular dichroism spectra of six-repeat constructs determined interact with the 6-, 12-, and 24-repeat constructs. (Fig. 7). that these regions have compact, globular structures with Individual six-repeat constructs failed to interact with lZ5I-levels of a-helix comparable with that of the native 24-repeat labeled cytoplasmic domain in contrast to the 12-repeat R13-H construct, intact membrane-binding domain, and native erythrocyte ankyrin, all of which interacted strongly. This result suggests that the interaction pocket for the cytoplasmic domain of the anion exchanger requires sites on both the third and fourth subdomains of the membrane-binding domain.
A role for all four subdomains in association with the anion exchanger is suggested by comparison of the 12-repeat R13-H construct and the native 24-repeat domain in membrane binding assays (Fig. 8). Values for half-maximal binding of 'ZsII-labeled R13-H construct and intact membrane-binding domain. We present a model for the compact arrangement of these six-repeat subdomains in the membrane-binding domain as either a tetrahedral or square planar array (Fig. 6).
Not just any set of six ankyrin repeats appear to be capable of folding, since previous work has shown that a six-repeat construct offset by one repeat from the last six-repeat domain has sharply reduced a-helical signal in solution (Davis et al., 1991a). The requirement of ordered arrays of repeats for the folding of repeat domains suggests that individual repeats have unique features whose specific interactions lead to the formation of folded domains.
Participation of multiple repeats in the formation of propdomain were nearly identical in measurements of the associ-erly folded domains is not unique to ankyrin repeats. Several ation of these proteins with inside-out erythrocyte mem-repeated amino acid motifs have been identified, which appear branes. However, Hill plots revealed a subtle difference in behavior of the 12-and 24-repeat polypeptides. The Hill plot for the 12-repeat subdomain exhibited a single line with a slope of 1.3, while plots for the 24-repeat domain were biphasic with a slope of 1.2 at low concentrations and a slope of 2.1 at higher concentrations. These results suggest that the Nterminal half of the membrane-binding domain has some role in mediating cooperativity in binding to membrane sites.

DISCUSSION
This report provides the first evidence that ankyrin repeats to utilize multiple copies to form properly folded structures. These motifs are normally present in tandem arrays and are typically shorter than individually folded motifs such as the immunoglobulin and fibronectin domains. Examples include the 24-amino acid repeat of the a*-glycoprotein (Takahashi et al., 1985), the 18-amino acid repeat of the MAP2 (Lewis et al., 1988) and tau proteins (Lee et al., 1988), and the 11 and 22 amino acid repeats of the apolipoproteins (Boguski et al., 1986). The crystal structure of apolipoprotein E (Wilson et al., 1991) has been solved and has clearly shown that the 11and 22-amino acid repeats fold cooperatively to form a single do not fold independently but rather require multiple repeats helical bundle. to form folded domains. In the case of ankyrin, the 24 repeats The majority of ankyrin repeat-bearing proteins contain of the membrane-binding domain are organized into four tandem arrays of four to seven repeats (Michaely and Bennett, folded subdomains of six repeats each. The existence of six-1992), which suggests that, like the repeats in ankyrin, the repeat subdomains was first suggested by proteolytic diges-repeat arrays in these proteins fold cooperatively to form A B C FIG. 7. Binding of the cytoplasmic domain of the anion exchanger requires multiple subdomains. The 82-kDa protein construct of brain ankyrin, native erythrocyte ankyrin, and domains of erythrocyte ankyrin were electrophoresed on SDS-polyacrylamide gels and either stained with Coomassie Blue (panel A ) or electrophoretically transfered to nitrocellulose paper. After blocking the nitrocellulose paper with 40 mg/ml bovine serum albumin solution in a buffer containing 10 mM sodium phosphate, 100 mM NaCI, 0.2% Triton X-100, 0.5% Tween 20, and 1 mM of each of NaEDTA, NaN3, and dithiothreitol (pH 7.4), '2sII-labeled anion exchanger cytoplasmic domain (47,000 cpm/ pmol) was added to 20 nM and incubated with moderate shaking overnight at 4 "C. Panel B shows an autoradiogram of this blot. For the autoradiogram in panel C, 1 PM cold anion exchanger cytoplasmic domain was added in conjunction with the '2sI-labeled cytoplasmic domain to assess nonspecific binding. 82kD is protein construct derived from the brain ankyrin gene product and contains the region corresponding to the last 19 repeats and a small portion of the spectrin-binding domain (Davis et al., 1993). Ankl is native erythrocyte ankyrin. 89kD is the native membrane-binding domain of erythrocyte ankyrin. 62kD is the native spectrin-binding domain of erythrocyte ankyrin. The remaining lane designations are the names of the individual repeat constructs previously described in the legend to Fig. 2. binding assays were used to measure the association of lZ51-labeled R13-H construct (33,000 cpm/pmol) and native membrane-binding domain (220,000 cpm/pmol) to ankyrin-depleted, inside-out, erythrocyte vesicles. A Kd of 18 nM for each of these proteins was determined from the Hill plot by determining where the lines cross zero on the y axis. The R13-H construct showed a cooperativity of approximately 1.3. The membrane-binding domain (89kD) had similar cooperativity (1.2) below the Kd; however, the cooperativity increased to 2.1 above the K d . In the Hill plot, 0 is the fraction of sites on the vesicles bound with protein, and ankyrin region is the concentration of the R13-H or 89-kDa domain at each point. stable structures. Exceptions to this general rule include a few proteins, such as the SWI4/SWI6 family, which contain two individual ankyrin repeats. Presumably, the surrounding protein sequences aid in stabilizing the folding of these ankyrin repeats such that tandem arrays of four to seven repeats are not required in these proteins. Exceptions with much greater numbers of repeats include ankyrin with 24 repeats, which we have shown in this work to be divided into four sixrepeat domains and a-latrotoxin, which contains 19 repeats (Kiyatkin et al., 1990). Although the organization of the repeats in a-latrotoxin is not known, these repeats are also likely to be divided into smaller domains. The amino acid sequence of this toxin suggests that the 19 repeats are subdivided into two regions, one of 13 repeats and one of six repeats, separated by an insertion of 75 amino acids. If the set of 13 repeats are further divided into two subdomains, then alatrotoxin would also conform to the general rule of four to seven tandem copies per folding unit.
The requirement of two six-repeat subdomains for the high affinity interaction with the anion exchanger (Fig. 7 ) suggests that at least some of the proteins that associate with ankyrin also recognize multiple subdomains. The possibility that binding sites can be formed by various combinations of subdomains would greatly increase the variety of potential interaction sites on the membrane-binding domain. Moreover, contact sites formed between individual subdomains could provide additional surface features of this domain that may participate in protein interactions.
The presence of four six-repeat subdomains in the membrane-binding domain suggests several possible pathways for the evolution of ankyrin and its protein interactions. In one possibility, the 24-repeat domain may have been formed by two duplication events beginning with a gene encoding a protomeric six-repeat unit. The 24-repeat domain then combined with a spectrin-binding domain by gene fusion. Alternatively, a gene encoding a single six-repeat domain may have fused with a gene encoding the spectrin-binding domain with the additional six-repeat subdomains resulting from illegitimate recombination events. Both alternatives predict a prototypic six-repeat domain that could have participated in protein interactions. Following assembly of four six-repeat domains, additional membrane protein binding sites presumably resulted from interactions that involved contacts with the surface of the tetrameric domain and possibly other portions of the ankyrin molecule. Of future interest will be the identification of any membrane proteins that are capable of interacting with individual six-repeat domains, since these sites may represent the most ancient binding sites and may provide insight into the original interactions of ankyrin.