Initiation of spectrin dimerization involves complementary electrostatic interactions between paired triple-helical bundles.

The spectrin heterodimer is formed by the antiparallel lateral association of an alpha and a beta subunit, each of which comprises largely a series of homologous triple-helical motifs. Initiation of dimer assembly involves strong binding between complementary motifs near the actin-binding end of the dimer. In this study, the mechanism of lateral spectrin association at this dimer nucleation site was investigated using the analytical ultracentrifuge to analyze heterodimers formed from recombinant peptides containing two or four homologous motifs from each subunit (alpha20-21/beta1-2; alpha18-21/beta1-4). Both the two-motif and four-motif dimer associations were weakened substantially with increasing salt concentration, indicating that electrostatic interactions are important for the dimer initiation process. Modeling of the electrostatic potential on the surface of the alpha20 and beta2 motifs showed that the side of the motifs comprising the A and B helices is the most favorable for association, with an area of positive electrostatic potential on the AB face of the beta2 motif opposite negative potential on the AB face of the alpha20 motif and vise versa. Protease protection analysis of the alpha20-21/beta1-2 dimer showed that multiple trypsin and proteinase K sites in the A helices of the beta2 and alpha21 motifs become buried upon dimer formation. Together, these data support a model where complementary long range electrostatic interactions on the AB faces of the triple-helical motifs in the dimer nucleation site initiate the correct pairing of motifs, i.e. alpha21-beta1 and alpha20-beta2. After initial docking of these complementary triple-helical motifs, this association is probably stabilized by subsequent formation of stronger hydrophobic interactions in a complex involving the A helices of both subunits and possibly most of the AB faces. The beta subunit A helix in particular appears to be buried in the dimer interface.

Members of the spectrin family of membrane skeleton proteins are widely expressed in vertebrates as well as in lower organisms. In erythrocytes, spectrin is the major component of the membrane skeleton, a network of spectrin oligomers crosslinked with short actin filaments that is bound to the membrane and provides cell membrane stability. The most common form of spectrin on intact cell membranes is the tetramer, which is formed by the "head-to-head" association of two elon-gated heterodimers, each comprising an ␣ subunit and a ␤ subunit with molecular masses of 280 kDa (1) and 246 kDa (2), respectively. Both the ␣ and ␤ subunits consist largely of a series of homologous motifs, each approximately 106 residues in length (3). The tertiary structure of these motifs is a triplehelical bundle, as determined by x-ray crystallography (4) and NMR spectroscopy (5). Head-to-head tetramers form by binding of complementary partial motifs at the ends of the ␣ and ␤ subunits to form a complete triple-helical bundle (6 -8).
The antiparallel lateral association of the ␣ and ␤ subunits to form a heterodimer is initiated near the actin-binding end of the molecule at a dimer nucleation site, where complementary motifs from each subunit form a high affinity association (9 -11). A minimum of two motifs from each monomer (␣20 -21 and ␤1-2) is thought to be necessary for association (11), although it has been proposed that small sections of the non-homologous domains at the tail end of the dimer are also required (10). Between ␤1 and ␤2, and between ␣20 and ␣21, are 8-residue inserts that are not part of the typical 106-residue motif. Increasing or decreasing the length of these inserts abolished binding, indicating that the relative register of the two motifs from each subunit is important for association (10). Larger recombinant peptides containing additional homologous motifs exhibited stronger binding, indicating that each additional laterally associated pair of motifs contributed to the overall binding affinity between the subunits (11). This finding is consistent with an earlier study (9), which proposed that spectrin dimers assemble in a zipper-like mechanism, in which the initial binding at the nucleation site is followed by weaker binding along the length of the dimer.
The nature of the molecular forces that contribute to the lateral association of ␣ and ␤ subunits is unknown. Although the motifs required for binding at the nucleation site have been identified, the specific regions on the surfaces of these triplehelical motifs that bind to the complementary subunit and the nature of the molecular interactions has not been elucidated. Similarly, the associations between non-nucleation motifs further along the subunit that moderately increase dimer affinity are poorly understood. Determining the nature of lateral ␣/␤ associations should contribute to our understanding of spectrin's structural and functional properties, including the basis of spectrin's elasticity, the apparent differences in the morphology and self-association of erythroid and nonerythroid isoforms, and the possible effects of the subunits on each other during interactions with other molecules. For example, lateral interactions between the subunits may be involved in the down-regulation of brain spectrin self-association by calpain I and Ca 2ϩ /calmodulin (12).
The ionic strength-dependent dissociation of erythroid and brain spectrin oligomers to dimers and then to monomers (13,14) indicates that electrostatic interactions play an important role in the lateral association of the intact subunits. However, it is not known whether electrostatic interactions are directly involved in lateral associations, or indirectly involved via control of subunit flexibility, possibly via electrostatic interactions between consecutive motifs. Furthermore, it is not known whether electrostatic interactions are important for the initial association at the dimer nucleation site or for a subsequent step in dimer assembly, such as the weaker non-nucleation site lateral interactions or closing the hairpin loop at the head end of the molecule. Spectrin subunits can also be dissociated at pH Ͼ 9.5, though this dissociation is accompanied by ϳ10% unfolding of the secondary structure (15).
In this study, we investigate the role of electrostatic interactions during spectrin dimer nucleation and identify the regions of the triple-helical motifs that contact the complementary motif on the paired subunit. The contribution of electrostatic interactions to the dimer nucleation site association was determined using sedimentation equilibrium analyses. Modeling of the electrostatic potential on the surface of two laterally opposed nucleation site motifs, as well as two opposed non-nucleation motifs, revealed regions of complementary electrostatic potential on the AB faces of the triple-helical bundles that were predicted to be the dimer interface region. Finally, this model was tested using protease protection analyses to probe the spectrin dimer nucleation site region before and after lateral association, in order to experimentally identify regions that are buried upon dimer formation.
Limited Protease Digestion of Spectrin Peptides-The purified ␣20 -21 and ␤1-2 peptides were dialyzed against a buffer containing 10 mM sodium phosphate and 130 mM NaCl, pH 7.3. Protein concentration was determined using the absorbance at 280 nm and molar extinction coefficients calculated from the amino acid sequence using the SEDNTERP program. 2 To form the ␣/␤ heterodimer, ␣ and ␤ peptides were combined in a 1:1 molar ratio and incubated at 0°C for 60 min. The monomer and dimer samples were adjusted to a single protein concentration (between 0.4 and 0.6 mg/ml) before digestion with trypsin at a 1:10 (w/w) enzyme:substrate ratio for 10 min at 0°C, or with proteinase K at a 1:50 enzyme:substrate ratio for 15 min at 0°C. The proteases were inactivated by the addition of diisopropyl fluorophosphate to a final concentration of 2 mM. The samples were stored at 0°C for 48 h to allow for complete inhibition of the protease, before analysis by SDS-PAGE or other methods. Protease cleavage was monitored by Tricine SDS-PAGE using 14% slab gels (16). The gels were stained with Coomassie Brilliant Blue R-250.
N-terminal Sequence Analysis-After separation by SDS-PAGE, the peptides were transferred to polyvinylidene difluoride membranes as described previously (17). The membranes were stained with Amido Black, and the bands of interest were excised and sequenced on an Applied Biosystems model 494 sequencer.
Reverse Phase HPLC-The trypsin or proteinase K cleavage products were separated on a Zorbax 300SB-C18 column (2.1 ϫ 150 mm) with a guard column containing the same resin, using a linear gradient of 0.085% trifluoroacetic acid in 95% acetonitrile against 0.1% trifluoroacetic acid. Peak detection was via absorbance at 215 nm.
Mass Spectrometry-Matrix-assisted laser desorption/ionization (MALDI) mass spectrometry was carried out on a Voyager RP spectrometer (PerSeptive Biosystems). Peptide solutions were diluted 1:1 with a saturated solution of sinapinic acid in 33% acetonitrile, 0.1% trifluoroacetic acid, and allowed to air-dry on the sample target before analysis. Alternatively, 0.5 l of sample was dried onto a sample target pre-coated with ␣-cyano-4-hydroxycinnamic acid/nitrocellulose matrix (18), washed briefly with water, and air-dried. Cytochrome c and Protein A were used as external standards.
Sedimentation Equilibrium-To form the spectrin ␣/␤ heterodimer, ␣ and ␤ peptides in phosphate-buffered saline (10 mM sodium phosphate, 130 mM NaCl, 1 mM EDTA, 0.15 mM PMSF, 0.05% sodium azide, 1 mM TCEP, pH 7.3) were combined in a 1:1 molar ratio and incubated on ice for 60 min. The ␣/␤ heterodimer was separated from excess or inactive monomers by gel filtration on two analytical (7.8 ϫ 300 mm) TSK-gel columns (G3000SW XL ϩ G2000SW XL ) in series (Toso Haas) equilibrated with Tris-buffered saline (20 mM Tris-Cl, 130 mM NaCl, 1 mM TCEP, pH 7.5). Fractions at the leading edge of the dimer peak (to ensure no contamination with monomers) were pooled and dialyzed for 24 h at 4°C against Tris-buffered saline containing 0.3-1.0 M NaCl. Sedimentation equilibrium experiments were performed at 4°C in an Optima XL-I analytical ultracentrifuge, using either the interference optics or the absorbance optics at 280 nm to measure the protein concentration gradient. When using absorbance optics, each ultracentrifuge cell was fitted with a six-channel 12-mm Yphantis-style centerpiece and loaded with three concentrations of the protein sample (typically 0.4, 0.2, and 0.1 mg/ml). For each scan, data were acquired every 0.001 cm with 5 replicates in continuous scan mode. When the interference optics were used, the cells were assembled with double-sector 12-mm centerpieces and sapphire windows. A blank scan of distilled water was taken before the run, to correct for the effects of window distortion on the fringe displacement data (19). All sample volumes were 110 l. Absorbance or fringe displacement data were collected every 4 -6 h until equilibrium was reached, as determined by comparison of successive scans using the MATCH v.7 program, 2 and the data were edited using the REEDIT v.9 program. 2 Analysis of sedimentation equilibrium data was performed using the NONLIN program 2 (20). The reduced molecular weight, , of a protein is defined as ϭ M(1 Ϫ ) 2 /RT, where M and are the molecular weight and the partial specific volume of the protein, respectively, is the density of the solvent (19), is the angular velocity in radians/s, R is the gas constant, and T is the temperature in Kelvin. In this study, the program SEDNTERP was used to calculate M and (from the amino acid composition of the recombinant peptides), and the of the solvent. As the difference in for the ␣ and ␤ peptides ( ␣ Ϫ ␤ ) was Ͻ0.3 at all rotor speeds used, the 1:1 heteroassociation of the ␣ and ␤ peptides was modeled as an ideal monomer-dimer association, where of the theoretical monomer was calculated using the weight-average molecular weight of the two peptides (21). At least three data sets from different loading concentrations and/or rotor speeds were fitted simultaneously. 1 The abbreviations used are: GST, glutathione S-transferase; PMSF, phenylmethylsulfonyl fluoride; TCEP, tris(2-carboxyethyl)phosphine hydrochloride; HPLC, high performance liquid chromatography; PAGE, polyacrylamide gel electrophoresis; Tricine, N-tris(hydroxymethyl) methylglycine; MALDI, matrix-assisted laser desorption/ionization. All of the experimental data were fitted well by either an ideal single species model, or an ideal monomer-dimer model. Goodness of fit was determined by examination of the residuals and minimization of the variance. The association constants returned by NONLIN from the monomer-dimer fits were converted to the molar scale using either the calculated molar extinction coefficient of the peptide, for the absorbance data, or the sequence molecular weight of the peptide and a specific fringe displacement of 3.26 fringes (liters/g) for interference data (21).
Molecular Modeling-Models of the Drosophila ␤8, ␣20, and ␤2 motifs were constructed using the atomic coordinates of the ␣14 crystal structure. The structure of the monomer of the Drosophila ␣14 motif (D␣14) was derived from the crystal structure of the D␣14 dimer as described by Yan et al. (4). The D␣14 model was then refined by constrained energy minimization with Discover ® software (Biosym Technologies) using the all-atom constant valence force field. Partial charges were assigned to all atoms at pH 7.0. Asp, Glu, Lys, and Arg side chains were ionized. Histidines were neutral, as were the N and C termini of the motif (as each motif is part of a larger polypeptide in the native protein). Minimization was performed using 200 iterations of steepest descent, followed by 500 iterations of conjugate gradient for the BC loop side chains alone, then for the entire loop. The x-ray crystal structure was then relaxed using 200 iterations of steepest descent, followed by 500 iterations of conjugate gradient, first for the hydrogen atoms, then the side chains, and finally for all atoms except the C␣ atoms of the helical residues, as described (22).
To model D␤8, D␣20, and D␤2, the residues of D␣14 were replaced with those of the motif to be modeled using Biopolymer software (Biosym Technologies), which conserves the original position of the backbone atoms, as well as any side chain atoms common to the two residues. The models were then refined with energy minimization using the constant valence force field as described above.
Calculations of electrostatic potential mapped to the accessible surface of the motif were performed using the program GRASP (23). The program maps a molecule onto a three-dimensional cubic grid, and calculates the electrostatic potential at each grid point using a finite difference solution to the nonlinear Poisson-Boltzmann equation (24,25). The accessible surface area of the motifs was defined as the surface mapped out by the center of a probe of radius 1.4 Å rolled around the van der Waals surface of the protein. The ionic strength was 130 mM, and an ionic radius of 2.0 Å was used. The dielectric was 2 for the protein interior and 80 for the surrounding solvent. Charges were assigned to the ionized groups of Asp and Glu (Ϫ1) and Lys and Arg (ϩ1). Histidine and termini residues were neutral.

RESULTS
Salt Dependence of ␣/␤ Lateral Association-The salt dependence of the lateral association of two-and four-motif nucleation site recombinants was investigated using sedimentation equilibrium. Each of the four recombinants was tested alone, and in all cases the sedimentation equilibrium data were fitted well by a single ideal species model with an estimated molecular weight within 4% of the calculated sequence mass, demonstrating that each recombinant is stable and monomeric (data not shown). Experiments were then performed on the ␣20 -21/␤1-2 or ␣18 -21/␤1-4 heterodimers, in salt concentrations of 0.1-1.0 M. For all heterodimer samples, the data were described well by a model describing an ideal monomer-dimer reaction, enabling estimation of the association constant. Representative data are shown in Fig. 1. The salt dependence of each association is shown in Fig. 2 and compared with previously published data on the association of whole spectrin subunits (13). For both the two-motif and four-motif nucleation peptides, there is a marked weakening of the association with increasing salt that is substantially greater than the effect on whole erythroid spectrin subunits. The salt dependence of the lateral association is the same for the two-motif and four-motif heterodimers, although the strength of the association increases with the number of motifs. At all salt concentrations, the association of ␣18 -21/␤1-4 was about 3-fold stronger than that of ␣20 -21/␤1-2, consistent with the hypothesis that motifs other than those in the minimum nucleation site (␣20 -21/ ␤1-2) make slight contributions to the strength of dimer assembly.
Modeling the Electrostatic Potential on the Surface of the ␣20/␤2 Motifs-Many protein associations have been shown to involve the interaction of surfaces with complementary electrostatic potential (26,27). Electrostatic potential is generated by the charge distribution of the macromolecule and depends on several factors including the location and magnitude of the charges, the solvent ionic strength, the dielectric constant of the macromolecule and the solvent, and the shape of the molecule, which defines the dielectric boundary and thus has an important effect on the electrostatic potential generated (26). The electrostatic potential of a protein can be calculated using a numerical procedure to solve the Poisson-Boltzmann equation, such as that contained in the programs Delphi (24,25) and GRASP (23). The weakening of lateral association between intact spectrin subunits and the nucleation site peptides with increasing ionic strength indicates that attractive electrostatic interactions between the subunits are likely to play an important role in lateral association. Identification of areas of complementary electrostatic potential on the surface of opposing motifs should help to identify the faces of the triple-helical motifs involved in subunit-subunit association.
In the present study, models of four Drosophila motifs that are laterally paired in the dimer, ␣20 and ␤2 (in the dimer nucleation site) and ␣14 and ␤8 (11), were constructed. Drosophila motif sequences were used because the models were based on the crystal structure of the Drosophila ␣14 motif (4). Although the nucleation motifs show reduced homology to the general spectrin motifs, and are closer to the motifs of ␣-actinin, alignment of the nucleation motif sequences with that of ␣14 shows that the homology of ␣20 and ␤2 to typical spectrin motifs is greater than that of the other nucleation motifs, ␣21 and ␤1 (10, 11). The hydrophobic residues in the a and d heptad positions in the ␣-helices of ␣14, which form the core of the bundle (4), are conserved in ␣20 and ␤2. Furthermore, there are no insertions or deletions other than the 8-residue insert that occurs before the start of the predicted A helix of the motifs. The ␣20 motif contains 3 proline residues, but these are found at the beginning or end of the ␣-helices and should not greatly disturb the secondary structure. Thus, it is reasonable to assume that the ␣20 and ␤2 motifs will fold into a helical bundle similar to that seen in ␣14.
The electrostatic potential of the modeled motifs was calculated and mapped to the solvent accessible surface of the motifs using the GRASP program (23). As the spectrin motif structure is a bundle of three ␣-helices, with each helix aligned roughly parallel or antiparallel to the others, the surfaces of the motif can be divided into three faces, AB, BC, and CA, according to the helices that make up the face, as shown in Fig. 3. Fig. 4A shows the electrostatic potential on the three faces of the ␣20 and ␤2 motifs. The BC faces of both motifs (Fig. 4A, middle  panels) have a predominantly negative charge and would be expected to repel each other. The potential on the CA faces is also not very complementary, as both motifs have negative potential in the center of the face and positive potential at the edges (Fig. 4A, bottom panels). In contrast, the AB face of the ␤2 motif has an area of strong positive potential that is complementary to an area of strong negative potential on the AB face of the ␣20 motif. The positive and negative potentials are directly opposite one another when the two motifs are in the antiparallel orientation (Fig. 4A, top panels). Similarly, the negative potential on the AB face of ␤2 is complementary and lies opposite to the positive residues on the ␣20 AB face. Hence, the most compatible pairing of faces on these two motifs is obtained by docking the AB face of ␣20 with the AB face of ␤2.
In order to examine a pair of motifs that is not part of the nucleation site, a model was constructed of the Drosophila ␤8 motif, which is predicted to be laterally paired to the ␣14 motif in the dimer (11). The electrostatic potential on the ␣14 and ␤8 motifs showed a similar trend to ␣20/␤2 (Fig. 4B). In these motifs, both the BC and CA faces were strongly negative, making interaction between these faces unlikely. However, the AB face of ␣14 shows a strikingly different electrostatic potential, with a large area of positive potential in the middle of the face. Since all faces of ␤8 are negatively charged, the AB face of ␣14 is the only likely face of this motif to interact with the ␤ subunit. While all faces of ␤8 are negative and complementary to this positive face, it is most likely that the interaction face of all laterally paired motifs would be consistent throughout the length of the subunit. Together, these modeling experiments predict that the most likely dimer interface involves association of the AB faces on both subunits.
Protease Protection Experiments-To experimentally test this model of the dimer interface region, limited proteolysis of ␣20 -21 and ␤1-2 was performed before and after formation of the 1:1 ␣:␤ dimer complex to investigate changes in protease susceptibility upon association. Initially, pilot digests were performed with the proteases trypsin, chymotrypsin, proteinase K, elastase, and endoproteinase Glu-C, using various enzyme: substrate ratios between 1:5 and 1:200. Aliquots were taken at intervals between 5 and 90 min for analysis by Tricine SDS-PAGE. All enzymes could cleave the complex and individual components, and the more extreme digestion conditions for most enzymes extensively degraded the proteins. However, usually only the first several cleavages yielded informative data concerning protected sites, since it rapidly became impossible to distinguish between alternative cleavage pathways and sites as cleavages accumulated. The trypsin and proteinase K digests showed the greatest differences between monomers and the dimer, and were selected for further analysis. Trypsin cleaves at the carboxyl side of lysine and arginine residues, while proteinase K has a broad specificity toward aliphatic, aromatic and other hydrophobic residues. For both proteases, the digestion pattern as well as the time dependence of the digestion observed by SDS-PAGE were highly reproducible between different sample preparations (data not shown). The digests were then scaled up in order to obtain sufficient quantities of digest products for analysis by reverse-phase HPLC, N-terminal sequencing, and MALDI mass spectrometry.
SDS-PAGE of representative preparative ␣20 -21/␤1-2 trypsin and proteinase K digests are shown in Fig. 5. Reverse-phase HPLC was also used to separate and quantify the proteolytic products. Changes in peptide yields of at least 2-fold were considered significant. Fig. 6 shows the HPLC chromatographs of ␣20 -21/␤1-2 dimers and the individual monomers digested with proteinase K. The HPLC fractions were analyzed by SDS-PAGE in order to correlate the HPLC peaks with the bands observed in Fig. 5. The N-and C-terminal boundaries of the proteolytic products were then identified. The N termini were determined by Edman sequencing of the fragments after SDS-PAGE followed by electroblotting to polyvinylidene difluoride membranes, or by directly sequencing the HPLC fractions. The molecular masses were determined by MALDI mass spectrometry of the HPLC fractions. The N-terminal sequence of each fragment combined with its molecular mass allows unambiguous identification of the C terminus, since the error in mass determination is less than one residue mass. All of the major proteolytic fragments were identified and are listed in Table I. Fig. 7 maps the fragments onto the predicted secondary structure of ␣20 -21 and ␤1-2, based on their sequence homology with a typical spectrin motif, for which the structure has been solved (4). In both the tryptic and proteinase K digests, ␣20 -21 is fairly resistant to proteolysis under the conditions shown, both as a monomer and in a heterodimer complex, although a few informative ␣ fragments were observed. The ␤1-2 is readily cleaved into several major fragments, almost all of which begin at the N terminus of ␤1 and end at protease cleavage sites in the predicted A helix of the ␤2 motif, thus encompassing all of ␤1. The other two protease-resistant ␤1-2 fragments begin in the BC loop and the C helix of ␤1, respectively, and end near the C-terminal end of the peptide, thus including most of the ␤2 motif (Fig. 7A).
The change in the digestion pattern upon dimer formation was investigated qualitatively by comparing the SDS-PAGE bands and quantitatively by comparing the HPLC peak areas of the monomer digests with those of the dimer digest. The most prominent tryptic fragment of ␣20 -21, ␣T1, involves a cleavage on the CA face of the ␣21 A helix. This site is not protected upon dimer formation since it is produced in similar yield from monomers and the dimer complex. In contrast, the most prominent proteinase K peptides, ␣P1a and ␣P1b, are partially protected in the dimer, and these two cleavage sites map to the AB face of the ␣21 A helix (Fig. 7B).
The susceptibility of ␤1-2 to both trypsin and proteinase K changes substantially upon dimer assembly. With proteinase K, the cleavage pattern of ␤1-2 is qualitatively reproduced in the dimer digest; however, the decrease in cleavage products and corresponding increase in uncleaved ␤1-2 in the dimer demonstrates that ␤1-2 is more protease-resistant when associated with ␣20 -21 due to protection of several cleavage sites in the dimer. Fig. 7 shows the location of these protected proteinase K cleavage sites in ␤1-2. In all cases the protected cleavage sites occur in the A helix of the ␤2 motif with most of the sites located on the AB face. One cleavage site, P4, is actually located in the interior of the triple helix motif. However, this fragment is a minor peptide in the digest of the ␤1-2 motif (see Fig. 5B, lane 4) that is likely to represent a secondary cleavage after the peptide is first cleaved to one of the major products, P1 and P2, and the partial A helix is destabilized leading to this minor fragment.
In the case of trypsin, the digestion pattern of ␤1-2 alone and in the dimer are again qualitatively similar, but there are both increases and decreases in the amounts of the various cleavage products after association. Before dimer formation, the major protease-resistant fragment of ␤1-2 is ␤T4, which comprises the whole ␤1 motif and ends in the A helix of ␤2, similar to the proteinase K fragments. After dimer association, there is a substantial reduction in the amount of ␤T4, while ␤T2 and ␤T3 become the major protease-resistant fragments. As Fig. 7A shows, ␤T4 and ␤T2/␤T3 are mutually exclusive fragments, i.e. a single ␤1-2 molecule cannot give rise to both fragments. Thus, by analysis of the HPLC peak areas, it is possible to estimate the number of ␤1-2 molecules giving rise to ␤T4 (which is the lower limit of molecules cut at this site), as well as the number of molecules that were definitely not cut at the ␤T4 site (by adding the numbers of whole ␤1-2, ␤T1, and ␤T2/␤T3). Similarly, the number of molecules that were not cut at the ␤T2/␤T3 sites can be estimated by adding the numbers of whole ␤1-2, ␤T1, and ␤T4. In this way it was possible to determine that the proportion of ␤1-2 molecules cut at the ␤T4 site is at least 51-60% in the monomer digest, and no more than 24% in the heterodimer digest. Conversely, the proportion of molecules cut at the ␤T2/␤T3 sites is no more than 18 -24% in the monomer digest and at least 43% in the dimer digest. This analysis shows that in the dimer, cleavage at the ␤T4 site is reduced, while the ␤T2 and ␤T3 sites become more sensitive to trypsin digestion. Fig. 7 (B and C) maps the protected and exposed trypsin and proteinase K sites onto a model of the triple-helical motif. DISCUSSION Previous studies have demonstrated the importance of electrostatic interactions in the stability of both erythroid and nonerythroid spectrin oligomers (13,14). Both studies showed a salt-dependent dissociation of spectrin dimers into their component subunits. However, the mechanism for this dissociation was unknown. Spectrin subunits are highly elongated and flexible charged molecules; thus, there are a number of ways in which increased ionic strength could reduce dimer affinity. For example, the high ionic strength could weaken interactions between consecutive motifs, resulting in increased flexibility that may in turn weaken lateral association or could destabilize the head end closed hairpin loop (6). Alternatively, high ionic strength could destabilize laterally paired motifs along the length of the molecule and/or the laterally associated motifs near the actin-binding end of the molecule that initiate dimer formation.
The aim of the present study was to investigate the mechanism of lateral association of spectrin ␣ and ␤ motifs in the dimer nucleation site. The role of electrostatic interactions in lateral association was investigated by perturbing this interaction with increasing ionic strength, and the electrostatic potential on the surface of four motifs was modeled in order to determine the most likely faces of the triple-helical motifs for attractive electrostatic interactions. Protease protection experiments were then used to identify the regions of spectrin motifs that become protected or exposed upon dimer formation. The primary focus of this work was on the dimer nucleation recombinants ␣20 -21 and ␤1-2, which comprise the minimum nucleation site for lateral association. 3 The importance of electrostatic interactions to the lateral association at the dimer nucleation site was investigated using both two-motif and four-motif heterodimers, ␣20 -21/␤1-2 and ␣18 -21/␤1-4, respectively. As shown in Fig. 2, the association of each heterodimer was substantially weakened in salt concentrations above 130 mM. It is possible that the effect of salt on this association is an indirect result of a salt-induced confor- mational change of one or both monomers. However, studies have shown that changes in ionic strength have no effect on the secondary structure of spectrin dimers (28,29) or single motif peptides (29), or on the thermal stability of peptides including one or more motifs (29,30). Circular dichroism data on intact erythroid and brain spectrin shows a minor (5-10%) increase in ␣-helical content with increasing salt from 0.15 to 1.0 M (14). Thus, there is no evidence to support a salt-induced destabilization of spectrin conformation; in contrast, there may be a slight increase in the stability of the secondary and tertiary structure with increasing salt. Together, the data indicate that the salt-induced dissociation of the ␣20 -21/␤1-2 and ␣18 -21/ ␤1-4 heterodimers is due to weakening of attractive electrostatic interactions between the paired motifs, rather than destabilization of conformation.
Analysis of electrostatic potential has been used to help identify the binding sites of several other proteins (22,31,32). Fig. 4 shows the electrostatic potential on the three faces (AB, BC, and CA) of the Drosophila ␣20, ␤2, ␣14, and ␤8 motifs, each face comprising the surfaces of two adjacent ␣-helices. The dimer interface could involve interaction of a single ␣-helix from each motif in the heterodimer, two adjacent helices from each motif, or one helix from one subunit and two helices from the other subunit. In other proteins involving the self-association of ␣-helical bundles, such as apoferritin (33), cytochrome c (34), and Salmonella typhimurium aspartate receptor (35), the interface involves interaction between two helices from each protomer. In addition, homodimers usually interact via the same site on each subunit (36). The spectrin subunits and respective paired motifs are not identical, but are homologous, and have evolved from a common gene. Hence, it is reasonable to predict that the lateral interaction of the subunits might occur via the same region of each paired motif, as experimentally supported by the observed protease protection of A helices from both subunits. As shown in Fig. 4A, the AB face shows striking charge complementarity in the nucleation motif pair  (4). Protease cleavage sites are located between the two heptad positions for the flanking amino acids of the peptide bond that was cleaved (see Table I). Sites that are protected in the dimer are shown using open symbols, and sites that are unaffected or more protease-sensitive in the dimer are shown using solid symbols. ␣20 motif sites, clockwise from the top are: OE, the T1 site that is unaffected by dimer formation; 4, proteinase K sites for P1a and P1b that are protected in the dimer complex. ␤2 motif sites that are protected on the A helix are: 4, proteinase K sites; E, trypsin sites. The specific sites, clockwise from the top, are: P3, T4a, P2, P1, T4b, and P4. The tryptic site on the C helix of the ␤1 motif exposed in the dimer, T3, is also shown (q). The BC loop cleavage site exposed in the dimer, T2, cannot be seen in this projection. C, the ␤ motif cleavage sites shown in panel B are illustrated on a side view of the triple-helical model using the same symbols. The sites on the A helix from top to bottom are: T4a, T4b, P3, P2, P1. The P4 cleavage site is not visible in this view.  f Relative yield of the fragment in a dimer digest relative to its yield in a parallel monomer digest. g The cleavages at residues 230 and 231 are common to T1, T2, T3, and small complementary peptides to T4 and T5. These latter peptides were not identified and quantitated. Hence, the overall degree of cleavage of this site was not determined.
h Data refer to the N-terminal cleavage site for this fragment.
␣20/␤2 when the motifs are aligned antiparallel, as they are in the dimer, and is therefore distinctly favored as the dimer nucleation interface site. The pattern of electrostatic potential on the surface of Drosophila ␣20 and ␤2 is expected to be highly conserved in the corresponding erythrocyte motifs, since the proportion of charged residues in the Drosophila motifs that are identical or substituted with the same charge in the erythrocyte motifs is 76% and 92% for ␣20 and ␤2, respectively. The electrostatic potential of the ␣14 and ␤8 motifs also favors lateral interaction of the AB faces, although the complementarity is less pronounced than for the modeled nucleation site motifs. As shown in Fig. 4B, the strikingly positive potential of the AB face of ␣14, among the strongly negative potential of all other faces on ␣14 or ␤8, clearly indicates that this face will be favored in lateral association with the ␤ motif. In this motif pair, the ␣ motif AB face has the potential to interact favorably with any region of the ␤ motif based on electrostatic complementarity alone, and in fact it is possible that this potential flexibility has some role in the formation of less specific interactions between motifs outside the nucleation site.
The use of limited proteolysis to identify binding surfaces has been previously described for several protein interactions (37)(38)(39). In this study, the protease protection analyses provide interesting experimental data on the location of the dimer lateral interface site. When ␤1-2 is digested with either trypsin or proteinase K, the major protease-resistant fragments start at the N terminus of ␤1 and end in the ␤2 motif, leaving the ␤1 motif intact. After association with ␣20 -21, there are two changes in the protease susceptibility of ␤1-2; there is a marked decrease in proteolytic cleavage of ␤2, and, in the case of trypsin, there is an increase in cleavage at two sites in the ␤1 motif. As shown in Fig. 7, the trypsin sites on ␤1-2 exposed in the dimer occur in the BC loop and the C helix of the ␤1 motif, which as noted above is very protease-resistant in the ␤1-2 monomer. The increased proteolysis of these sites after dimer formation must be due to a conformational change of this motif induced by association with ␣20 -21. The accessibility of the C helix in the dimer shows that it cannot be part of the dimer association interface, and its location on the opposite side of the motif from the AB face supports our hypothesis that the AB faces form the subunit-subunit interaction site.
All of the trypsin and proteinase K cleavage sites identified in these experiments as protected in the dimer ( Fig. 7 and Table I) are located in the A helices of the ␤2 and ␣21 motifs. This indicates that these A helices are largely buried in the ␣20 -21/␤1-2 heterodimer, or that there are conformational changes that protect these sites. While an induced conformational change cannot be rigorously discounted, direct protection of the A helices in the dimer is the more likely explanation for these results, since prediction of the AB faces as the dimer interface is consistent with all other available data including the electrostatic modeling, increased exposure of a site on the C helix in the dimer, and location of the invariant tryptophan in this region (see below). The extensive shielding of multiple sites along the length of the ␤2 motif A helix indicates that this helix in particular is protected in the dimer, suggesting that the A helix of the ␤ subunit may lie in the groove between the A and B helix on the ␣ subunit as shown in Fig. 3C. Although this model is most consistent with the data obtained in this study, the resolution of the methods used here is insufficient to reliably distinguish between closely related models involving components of the AB face as illustrated by the top model in Fig. 3B and the alternative model in Fig. 3C.
Trp 17 is notable as the most conserved residue in spectrin motifs and is located in the middle of the AB face. Residues with large side chains (Trp, Tyr, Phe, His, Met, and Arg) occur more frequently in protein interfaces relative to the overall protein exterior (36,40). In addition, some residues, mostly those with large and/or hydrophobic side chains such as Trp, tend to have more exposed side chains when in a protein interface than when in positions on the general exterior of the protein (36 -41). The highly conserved nature of Trp 17 in spectrin motifs together with the exposed nature of its side chain and its position in the middle of the AB face indicate that it probably functions as an interface residue. There is some experimental evidence that tryptophans may be involved in lateral association of spectrin subunits. A fluorescence study of tryptophans in spectrin by Subbarao and MacDonald (42) found that separation of the ␣ and ␤ subunits resulted in increased exposure of the ␣ subunit tryptophans to both hydrophilic and hydrophobic quenchers.
Complementary electrostatic interactions in the spectrin dimer nucleation site are likely to play a pivotal role in dimer initiation. The proteins barnase and barstar have been shown to associate initially via long range electrostatic interactions (43), which assist in bringing the proteins together rapidly in the correct orientation, prior to the final high affinity association involving precise short range interactions (43). Thus, a reasonable model for initiation of spectrin dimerization is that the electrostatic complementarity and other unique structural features of the nucleation motifs (␣20 -21 and ␤1-2) ensure proper docking of the correct lateral pairs. This is a critical initial step in dimerization, since each subunit comprises a large number of very similar motifs that must be correctly paired with their complementary partner. The long range electrostatic recognition of the appropriate complementary subunit motif should ensure that dimer assembly proceeds without formation of incorrectly paired intermediates that would slow overall assembly. In addition to influencing orientation and pairing of correct motifs in the nucleation site, the electrostatic interactions would be expected to increase the reaction on-rate and hence increase dimer affinity at lower ionic strengths, as demonstrated in Fig. 2.
It is likely that the interacting face of the motifs is consistent along the length of the subunits, such that initial interaction of AB faces in the dimer nucleation region would favor interaction of AB faces along the entire subunit as dimer assembly is completed. This hypothesis is consistent with the complementary electrostatic potential we have observed between the AB faces of ␣14 and ␤8. However, the greater salt dependence of the nucleation site association compared with that of the entire subunits (Fig. 2) indicates that electrostatic interactions are most critical for initial pairing of the dimer nucleation region. This supports our model where the primary role of the electrostatic interactions is to increase the rate of subunit docking in the correct register at the dimer nucleation site.