Mapping the Human Erythrocyte -Spectrin Dimer Initiation Site Using Recombinant Peptides and Correlation of Its Phasing with the -Actinin Dimer Site

Human erythroid spectrin dimer assembly is initiated by the association of a specific region near the N-terminal of β-spectrin with a complementary region near the C-terminal of α-spectrin (Speicher, D. W., Weglarz, L., and DeSilva, T. M.(1992) J. Biol. Chem. 267, 14775-14782). Both spectrin subunits consist primarily of tandem, 106-residue long, homologous, triple-helical motifs. In this study, the minimal region of β-spectrin required for association with α-spectrin was determined using recombinant peptides. The start site (phasing) for construction of dimerization competent β-spectrin peptides was particularly critical. The beginning of the first homologous motif for both β-spectrin and the related dimerization site of α-actinin is approximately 8 residues earlier than most spectrin motifs. A four-motif β-spectrin peptide (β1-4) with this earlier starting point bound to full-length α-spectrin with a K of about 10 nM, while deletion of these first 8 residues reduced binding nearly 10-fold. N- and C-terminal truncations of one or more motifs from β1-4 showed that the first motif was essential for dimerization since its deletion abolished binding, but β1 alone could not associate with α-monomers. The first two motifs (β1-2) represented the minimum lateral dimer assembly site with a K of about 230 nM for interaction with full-length α-spectrin or an α-spectrin nucleation site recombinant peptide, α18-21. Each additional motif increased the dimerization affinity by approximately 5-fold. In addition to this strong inter-subunit dimer association, interactions between the helices of a single triple-helical motif are frequently strong enough to maintain a noncovalent complex after internal protease cleavage similar to the interactions thought to be involved in tetramer formation. Analysis of hydrodynamic radii of recombinant peptides containing differing numbers of motifs showed that a single motif had a Stokes radius of 2.35 nm, while each additional motif added only 0.85 nm to the Stokes radius. This is the first direct demonstration that spectrin's flexibility arises from regions between each triple helical motif rather than from within the segment itself and suggests that current models of inter-motif connections may need to be revised.

The membrane skeleton of the human erythrocyte consists of a network of proteins that associates with the inner surface of the cell membrane and imparts remarkable structural integrity and flexibility to circulating erythrocytes. Spectrin is the major structural component of this specialized submembranous protein network. The basic functional unit of spectrin is a heterodimer formed by side-to-side, antiparallel association of a 280-kDa ␣ subunit with a 246-kDa ␤ subunit. Spectrin dimers associate head-to-head to form tetramers, the predominant form of spectrin in the membrane skeleton. These tetramers cross-link short actin oligomers, an association modulated by band 4.1, to form a dynamic two-dimensional submembrane latticework. Other associated proteins include: ankyrin, adducin, calmodulin, tropomyosin, tropomodulin, and band 4.9 (for reviews, see Bennett and Gilligan (1993), Delaunay and Dhermy (1993), Luna and Hitt (1992), Winkelmann and Forget (1993), and Lux and Palek (1995)).
Electron microscopy of spectrin dimers shows flexible 100-nm long rod-like molecules with strong lateral association of the subunits near the physical ends of the rods and weak associations in the central region (Shotton et al. 1979). In contrast, dimers in situ are only about 30 nm in length (Ursitti et al., 1991). This ability of the spectrin molecule to shorten and extend as well as its flexibility are attributed to the series of homologous 106-residue segments or motifs initially identified by partial peptide sequence (Speicher and Marchesi, 1984) and confirmed by complete sequencing of cDNAs for the ␣ subunit (Sahr et al., 1990) and ␤ subunit . Analysis of spectrin motif conformational phasing (the boundaries for complete folding units) using recombinant proteins established the starting point of properly folded spectrin motifs at approximately positions 26 -30 of the original sequence alignment (Winograd et al., 1991). Yan et al. (1993) recently determined the crystal structure of the 14th segment of Drosophila ␣-spectrin, which directly confirmed the phasing predicted from the recombinant peptide phasing experiments as well as the triple helical conformation of the basic 106-residue spectrin motif.
Previous studies of lateral association of ␣ and ␤ subunits (Morrow et al., 1980;Sears et al., 1986;Yoshino and Minari, 1991;Speicher et al., 1992) used mild trypsin digestion to dissect spectrin into a reproducible pattern of intermediatesized peptides. The latter study used a HPLC 1 gel filtration assay to analyze association of tryptic peptides with complementary spectrin subunits. These analyses showed that dimer assembly occurs very rapidly (within seconds) and that dimerization required a specific region at the tail end of the subunits represented by the tryptic ␣V and ␤IV domains . These tryptic domains include most or all of the repetitive segments ␣19 -21 and ␤1-4. Interaction of these regions is apparently the initial step of dimer assembly, which is followed by subsequent lateral association of additional ␣ and ␤ motifs. A mutation in ␣-spectrin in this region has been identified, ␣ LELY , that affects dimer assembly (Alloisio et al., 1991;Wilmotte et al., 1993;Randon et al., 1994). When this mutation is present along with an elliptocytosis mutation on the same chain, symptoms of the elliptocytosis mutation are often silent or mitigated since the ␣ LELY mutation will decrease incorporation of the mutated chain onto the membrane. In contrast, elliptocytosis mutations on the opposite allele from the ␣ LELY mutation enhance incorporation of ␣ subunit carrying the elliptocytosis mutation onto the membrane (Garbarz, 1994;Wilmotte et al., 1993). The ability of the ␣ LELY mutation to affect dimer assembly highlights the functional and clinical importance of the dimer nucleation region.
In the current study, we purified and extensively characterized a series of ␤ nucleation region recombinant peptides for proper polypeptide chain folding, dimer binding affinity, and hydrodynamic properties. These analyses show that the minimum ␤ peptide for dimer assembly contained the first two homologous motifs, but not the actin binding domain, and each additional motif further increased dimer binding affinity. Analysis of hydrodynamic radii of these recombinant peptides provided the first direct demonstration that spectrin's flexibility apparently resides in the connecting region between triple helical motifs rather than within the segment itself.

MATERIALS AND METHODS
Isolation of ␣-Spectrin Monomers-Spectrin was extracted from fresh human red cells within 24 h of collection and ␣-monomers were purified as described previously  using a modification of the ion exchange purification initially developed by Yoshino and Marchesi (1984).
Design and Construction of ␤-Spectrin Expression Plasmids-Oligonucleotide primers were designed to amplify specific regions of the ␤-spectrin nucleation site from the cDNA by the polymerase chain reaction using Vent polymerase (New England Biolabs). Primers contained restriction enzyme sites for BamHI and EcoRI, at the 5Ј and 3Ј ends, respectively, to allow directional cloning of the insert into the pGEX-2T expression vector (Pharmacia Biotech Inc.).
Five ␤-spectrin nucleation site clones and one ␣-spectrin nucleation site clone were produced for this study. The specific oligonucleotide primers used are listed in Table I. Initially, the start site (phasing) for a clone encompassing the first four homologous ␤ motifs, ␤1-4, used the codon for amino acid residue 301, which corresponds to the phasing reported by Winograd et al. (1991). Further analysis of the potential start site of this ␤1-4 peptide led to the production of another clone, ␤1-4 ϩ , which contains eight additional codons at the N terminus of the expressed peptide. Subsequent truncations of full motifs on the Cterminal and N-terminal ends of the ␤1-4 ϩ recombinant were prepared as described in Table I.
The entire ␣-spectrin nucleation site, encompassing repetitive motifs ␣18 -21, was designed essentially as described above with the exception that the nucleotide sequence contained a BamHI restriction enzyme site. Therefore, the oligonucleotides used for polymerase chain reaction were designed to contain a BglII site both at the 5Ј and 3Ј ends. This restriction enzyme creates the same overhang as BamHI thus allowing cloning into the BamHI site of pGEX-2T. A BglII site near the 3Ј end of the region to be amplified was apparently not changed by altering the spectrin sequence in the 3Ј primer. This resulted in utilization of the stop codon of the pGEX-2T vector. A single motif ␣ recombinant, ␣1 (residues 50 -158), was prepared as described previously .
All expression plasmids were transformed into the DH5␣ strain of Escherichia coli. Each construct was completely sequenced to verify the integrity of the recombinant vectors. The ␣and ␤-spectrin cDNA clones were kindly provided by Dr. Bernard Forget (Yale University, New Haven, CT).
Analytical HPLC Gel Filtration Binding Assay-Spectrin ␣-monomers or the ␣18 -21 peptides were mixed with purified recombinant nucleation site ␤-peptides and incubated at 0°C for different times ranging from 5 min to 15 h. Under most conditions equilibrium was reached within 5 to 15 min, hence a 25-min incubation time was used for most binding assays. To determine Stokes' radius, the proteins and peptides were separated on two analytical (7.8 ϫ 300 mm) TSK-gel columns (G3000SW XL ϩ G2000SW XL ) at 4°C with a flow rate of either 0.4 or 0.8 ml/min. Binding experiments used only the G3000SW XL column at 1.0 ml/min at 4°C in phosphate-buffered saline buffer. Eluted proteins were detected by absorbance at 280 nm and intrinsic tryptophan fluorescence (excitation 280 nm, emission filter 370 nm) and were quantified on a data acquisition system (PE Nelson Analytical) using peak area. Response factors for each protein were determined by replicate injections of known quantities (determined by quantitative amino acid analysis) for each component. Molecular weights used for calculating molarity were: ␣-monomer, 280,000; ␤1-4, 52,085; ␤1-4 ϩ , 52,964; ␤1-3 ϩ , 40,374; ␤1-2 ϩ , 27,998; ␤1 ϩ , 15,681; ␤2-4, 38,924; ␣18 -21, 51,938. Association (K a ) and dissociation constants (K d ) for binding of ␤-spectrin nucleation region peptides with intact ␣-spectrin monomers were determined by calculating the amount of unbound peptide relative to control samples under the same conditions.
N-terminal Sequence Analysis-After separation by SDS-polyacrylamide gel electrophoresis, peptides were transferred onto high retention polyvinylidene difluoride membranes (Bio-Rad) as described previously (Mozdzanowski et al., 1992). After staining with Amido Black, the bands of interest were excised and sequenced on a Hewlett-Packard G1005A sequencer as described previously (Reim and Speicher, 1994).
Circular Dichroism (CD) Measurements-CD spectra were performed on a Jasco J720 instrument at room temperature in a 0.2-mm path length cell. Proteins were in phosphate-buffered saline buffer, pH 7.3, and protein concentrations were determined by duplicate quantitative amino acid analysis.
Mass Spectrometry-Matrix-assisted laser desorption/ionization (MALDI) mass spectrometry was performed on a PerSeptive Biosystems Vestec Mass Spectrometer using Voyager software. Proteins were dialyzed into 20 mM ammonium bicarbonate, pH 8.0, 1 l of sample was mixed with 1 l of matrix solution (saturated solution of ␣-cyano-4hydroxycinnamic acid for samples Ͻ20 kDa and sinapinic acid for samples Ͼ20 kDa in 0.1% trifluoroacetic acid, 33% acetonitrile), the sample/matrix mixture was transferred to the sample target, dried, and analyzed. Expected masses were calculated from known sequences using the GPMAW program (Lighthouse Data, Denmark).

Design and Characterization of ␤-Spectrin Nucleation Site
Recombinant Peptides-As noted above, previous results using spectrin peptides from mild protease cleavage of purified spectrin monomer mapped the spectrin dimer nucleation site to approximately the last three to four homologous motifs of the ␣ subunit and the first four homologous motifs of the ␤ subunit . To determine the minimal nucleation site requirements of ␤-spectrin as well as physical properties of this region, a series of recombinant peptides were produced by truncating the previously defined ␤-spectrin nucleation region (␤1-4) at either the N terminus or the C terminus. In addition, an ␣-spectrin nucleation site peptide was constructed that included the entire putative nucleation site region on the ␣ subunit, ␣18 -21. The sequence content of each peptide is shown diagrammatically in Fig. 1A and their relationship to the overall motif structure of a spectrin dimer is shown in Fig. 1C. As illustrated, the phasing of each recombinant peptide correlates with the boundaries of homologous triple helical motifs as defined by the high resolution structure of a single motif determined by Yan et al. (1993) with the exception of the "ϩ" series of ␤-peptides, which begin 8 residues before the predicted repetitive segment (see below).
In addition to the recombinant constructs shown in Table I and Fig. 1A, the ␤1 ϩ peptide was isolated as a proteolytic by-product of the ␤1-2 ϩ construct. During purification of the ␤1-2 ϩ peptide, a highly specific cleavage product was observed due to cleavage at residue 425 (see Fig. 1A) as defined using MALDI mass spectrometry which yielded an experimental mass after removing the GST moiety by thrombin cleavage of 15,688 Da (data not shown). This proteolytically produced peptide contained the entire ␤1 ϩ motif with only a few additional residues on the C-terminal end (see Fig. 1A), was native as shown by circular dichroism and gel filtration, and was easily separated from the parent ␤1-2 ϩ peptide by gel filtration.
All recombinant peptides were expressed in bacterial cells as fusion proteins with glutathione S-transferase (GST), which was cleaved and separated from the recombinant proteins prior to use for functional studies. The purity of all peptides was Ͼ95% (Fig. 1B). The N termini of the cleaved peptides were confirmed by N-terminal sequence analysis and masses were confirmed by MALDI mass spectrometry (data not shown). The FIG. 1. Recombinant spectrin peptides. A, diagrammatic correlation of recombinant nucleation site peptides with the 106-residue repetitive spectrin motif. The most commonly used alignment of homologous spectrin sequences (Speicher and Marchesi, 1984), as indicated by the numbers at the top of the figure, is correlated with the three helices designated "A," "B," and "C" that form the triple helical conformational unit defined by crystallography (Yan et al., 1993). Spaces between the three helices represent turn regions. The recombinant peptides are illustrated by horizontal lines labeled with the repetitive motif number in the right margin (see panel C for location of these motifs within the spectrin dimer). Non-spectrin amino acids introduced at the ends of recombinant peptides by construction of expression vectors are indicated using single letter code. The phasing of all recombinant peptides is as defined by the crystallographic model and prior analysis of recombinant peptides (Winograd et al., 1991) except for the "ϩ" series of ␤ peptides. This series of peptides begins 8 residues earlier than the normal conformational motif. The C-terminal end of the ␤1 ϩ (*), ␤1-2 ϩ (2), and ␤1-3 ϩ (ç) peptides are marked as indicated. B, SDS gels of ␣-monomers and recombinant peptides after cleavage from the GST fusion moiety and repurification. The samples are: lanes 1 and 2 (4 g/lane), ␣-spectrin monomer and ␣18 -21, respectively, on a 7% Laemmli gel; lanes 3-7 (2 g/lane), ␤1-4 ϩ , ␤1-4, ␤1-3 ϩ , ␤2-4, and ␤1-2 ϩ , respectively, on a 10% Laemmli gel; and lane 8 (2 g/lane), ␤1 ϩ on a 5-15% linear gradient Tricine gel. C, arrangement of the structural motifs in an anti-parallel spectrin dimer. The ␤ subunit is comprised of an actin binding domain (ABD), 17 homologous motifs (numbered rectangles), and a small non-homologous phosphorylated C-terminal domain (solid squiggle). The ␣ subunit is comprised of an N-terminal partial and 20 full homologous motifs (motifs 1-9 and 11-21), an SH-3 type motif (motif 10), and a non-homologous C-terminal region consisting primarily of two EF-hand type motifs (diamonds). ␤1-4 construct was difficult to cleave from the GST moiety and a substantial amount of a secondary cleavage product was produced (lane 4, Fig. 1B). N-terminal sequence analysis of this smaller peptide showed that it resulted from a secondary cleavage after Arg 374 , which is located in the turn between helices B and C of motif ␤1. The full-length ␤1-4 peptide and its cleavage product could not be separated by either ion exchange or gel filtration chromatography due to their similar size, charge, and other physical properties.
Proper Phasing Results in a Less Constrained Conformation of the ␤1-4 Peptide-The unusual difficulty encountered with thrombin cleavage of the GST-␤1-4 construct relative to other GST fusion proteins containing complete spectrin motifs (see below) led to further evaluation of the ␤1 start point. To examine the relationship between the longer ␣-actinin and spectrin nucleation region motifs relative to the more common 106residue motif, the last four ␣-spectrin (␣18 -21), the first four ␤-spectrin (␤1-4), and the four spectrin-type motifs of human cytoskeletal ␣-actinin were aligned against numerous 106-residue spectrin motif sequences using the computer program ALIGN to perform optimized pairwise alignments. Gaps and insertions of the longer motifs relative to the 106-residue motif and its associated crystallographic structure were placed ( Fig.  2A) in the most frequently aligned position from the individual pairwise comparisons. Two observations emerged from this alignment. First, all insertions larger than a single residue mapped to two regions, an 8-residue segment near the beginning of the motif that probably extends the A helix as previously suggested (Viel and Branton, 1994), and variable length insertions that map to the turn region between helices B and C. Second, a number of prolines are located in positions that were helical in the reported high resolution structure of the Drosophila ␣14 motif (Yan et al., 1993), which supports the hypothesis that the nucleation site motifs have a unique conformation responsible for initiating dimerization.
Based on the alignment in Fig. 2, it was hypothesized that the first motif should have an 8-residue insertion in helix A relative to the more typical 106-residue motif, since the three C-terminal ␣-actinin motifs (A2, A3, and A4) and their most homologous spectrin counterparts (␤2, ␣20, and ␣21, respectively) all contained an extra 8 residues in this region. This prediction of a start site for the first motif that is 8 residues earlier than the initial ␤1-4 recombinant compares favorably with the observed mild protease cleavage sites for both ␣-actinin and ␤-spectrin as shown in Fig. 2B. In addition, although various start sites have been reported for the ␣-actinin motif based on alignments of sequences, the revised start site presented here for the first motif of spectrin and ␣-actinin agrees with the recent phasing analysis of ␣-actinin using recombinant peptides reported by Gilmore et al. (1994).
To further evaluate the phasing of the first ␤ motif, a ␤1-4 ϩ recombinant peptide, which started at residue 293 compared with residue 301 for the ␤1-4 peptide, was produced and analyzed in parallel with the ␤1-4 peptide. Thrombin cleavage of the ␤1-4 ϩ and ␤1-4 fusion proteins differed markedly at physiological ionic strength. The ␤1-4 ϩ fusion protein was efficiently cleaved without production of secondary cleavage prod-ucts. In contrast, under the same conditions, Ͻ50% of the ␤1-4 protein was cleaved and a prominent secondary cleavage product was formed (data not shown).
Secondary Structure and Hydrodynamic Properties of ␤-Spectrin Recombinant Peptides-All spectrin recombinant peptides were analyzed by circular dichroism to determine whether the peptides were properly folded. Representative spectra are shown in Fig. 3 and all peptides reported here had high ␣-helicity (about 80%) similar to spectrin dimers and monomers.
The Stokes' radii of the ␤-spectrin nucleation site recombinant peptides and a single motif ␣ subunit peptide, ␣1 (residues 50 -158), were determined by HPLC gel filtration using standard proteins with known Stokes' radii to calibrate the column. The observed hydrodynamic radii were linearly related to the number of repetitive motifs as shown in Fig. 4. A single repetitive segment, either ␤1 ϩ or ␣1, has a Stokes' radius of 2.35 nm and each additional motif increases the molecular size by only 0.85 nm. The Stokes' radius of the ␤1-4 peptide falls below the line created by the other nucleation site peptides and was not included in the linear regression calculation. This smaller Stokes' radius is indicative of a more compact molecular shape for this improperly phased ␤1-4 peptide, however, it did have normal helicity as determined by circular dichroism.
Functional Analysis of Recombinant Peptides from the ␤-Spectrin Dimer Nucleation Site Region-In order to identify the minimal requirements for the ␤-spectrin nucleation site and the effects of additional motifs on binding affinity, the recombinant ␤ peptides were evaluated in solution binding assays with either purified native ␣-spectrin monomers or recombinant ␣18 -21. Time course experiments showed that binding equilibrium was usually reached within 5 min or less under most concentrations and molar ratios evaluated. Therefore, protein mixtures were routinely incubated for 25 min prior to measurement of complex formation using a rapid HPLC gel filtration separation as shown in Fig. 5. Free ␣-monomers could not be resolved from complexes due to the small change in size when the complex was formed. Therefore, carefully quantified amounts of ␣-monomers and recombinant peptides were combined and association constants were determined by measuring the loss of recombinant peptide from its normally eluting position relative to an identical control without ␣-monomer. Control experiments showed that only equimolar binding occurred and that any dissociation of complex that occurred during the analysis did not increase the area of the unbound recombinant peptide peak. As shown in Fig. 5, no detectable binding was observed for the ␤1 ϩ or ␤2-4 peptides. In addition, no binding to ␣-monomers was detected for these two proteins when as much as a 3-or 5-fold molar excess of recombinant peptide was used. The ␤1 ϩ peptide, although required for nucleation site binding, is apparently not sufficient.
Binding affinities of the recombinant ␤ peptides with ␣-monomers are summarized in Table II. The importance of N-terminal phasing for the first ␤ motif is illustrated by the observation that the ␤1-4 peptide has nearly a 10-fold lower affinity for ␣-monomers compared with the 8-residue longer ␤1-4 ϩ peptide. The minimum dimerization site contains the first two motifs, which has a K d of about 230 nM. Each additional motif contributes to the affinity of the complex apparently through low affinity lateral pairing of additional motifs and a 4-motif nucleation site peptide has a K d of about 10 nM. The lateral association of ␤ recombinant peptides with ␣-monomers is readily reversible. As shown in Fig. 6, both the high affinity ␤1-4 ϩ peptide and the lower affinity ␤1-2 ϩ peptides can compete with each other for binding to ␣-spectrin monomers. ␤1-2 ϩ was also able to compete with ␤1-3 ϩ for binding to ␣-spectrin (data not shown).
The Actin Binding Domain Is Not Required for Dimer Assembly and Does Not Substantially Contribute to Dimer Affinity-As illustrated above, high affinity dimers can readily form without the presence of the N-terminal actin binding domain. To further evaluate whether the actin binding domain may contribute positively or negatively to dimer assembly, binding measurements of intact ␤ subunits to the ␣18 -21 recombinant peptide were performed. The K d for this interaction between an intact ␤ subunit and a 4-motif ␣ peptide is about 15 nM which compares favorably with the K d of 10 nM observed for a 4-motif ␤ peptide interaction with intact ␣-monomers.
Noncovalent Associations between Helices within a Single Motif Are High Affinity Interactions That Frequently Maintain Functional Complexes-As described above, the dramatic reduction in affinity of the ␤1-4 peptide (8-residue shorter Nterminal) compared with ␤1-4 ϩ suggested that additional truncation of the ␤1 motif would further reduce or abolish binding. In this context, an apparently inconsistent observation was that the 43-kDa peptide, the secondary cleavage product of the ␤1-4 recombinant, did show detectable binding to ␣-monomers. In the experiment shown in Fig. 5B (lane 3), a faint 43-kDa band was observed on the original gel and some preparations of the ␤1-4 peptide showed even more extensive binding of the 43-kDa band to ␣-monomers than the illustrated experiment. As noted above, N-terminal sequence analysis of this peptide showed that it was cleaved at residue 374 and FIG. 2. Correlation of the elongated spectrin and ␣-actinin nucleation site motifs with the more common 106-residue motifs and crystallographic structure. A, the phasing and conformation of a 106-residue motif determined by crystallography (see Fig. 1A) as illustrated at the top of the panel (helices A, B, and C) was used to align and compare spectrin nucleation site motifs (␣18 -21 and ␤1-4) and the spectrin-like motifs of ␣-actinin that are responsible for ␣-actinin dimer formation (A1-4). Insertions required to fit these longer sequences into the more common 106-residue motif are shown as underlined sequences above the main sequence and the insertion site is indicated with an arrowhead. The positions of insertions were determined using the program ALIGN. Residues that are identical in at least two of the three aligned sequences are shaded. Prolines are bold and underlined. Numbers on the right are the number of identities between the indicated pairs of motifs. B, relationship of the predicted nucleation site phasing with cleavage sites of spectrin and ␣-actinin produced by limited proteolysis. Mild cleavage of ␣-actinin with chymotrypsin produces a 55-kDa fragment (C55K) as initially shown by Imamura et al. (1988) and confirmed by us. One of the first cleavages of ␤-spectrin with trypsin removes the ABD domain by cleavage at residue 292 and the earliest observed fragments that contain the ␤ nucleation site are T74K and T46K. Alignment of the N-terminal sequences of these peptides are shown and the arrow indicates the predicted nucleation site phasing shown in panel A and used to construct the ␤ ϩ series of recombinant peptides (see text). lacked helices A and B of the ␤1 motif. It was therefore quite surprising that this truncated form of the molecule could effectively compete with a larger amount of intact ␤1-4, which was always present in these preparations, for association with ␣-monomers.
Further analysis of these samples by Tricine gel electrophoresis detected a band at about 8 kDa. Similarly, samples that were initially separated by HPLC gel filtration still contained the 8-kDa peptide. MALDI mass spectrometry of the peptide mixture confirmed that the 8-kDa (observed mass ϭ 8,733.8 Da versus expected mass for GS ϩ 301-374 ϭ 8,725 Da) and 43-kDa fragments (observed mass ϭ 43,398.3 Da versus expected mass for 375-743 ϭ 43,379 Da) were produced by a single protease cleavage at residue 374 and that the 43-kDa fragment had an intact C-terminal. Since the expected 0.1% error of this technique is less than a single amino acid residue mass, this method reliably defines the C-terminal boundary of proteins with known sequences when the N-terminal has been determined by sequence analysis.

DISCUSSION
A previous report from our laboratory using peptides produced by mild proteolysis showed that spectrin dimers assembled like a zipper with initiation of the process occurring near the tail end (actin binding end) of the molecule . In the present study, we further characterized dimer nucleation using ␤ recombinant peptides. The primary structure and conformational integrity of these recombinant peptides were confirmed by full-length DNA sequencing, N-terminal sequencing of the cleaved peptide, mass spectrometry, and circular dichroism measurements to ensure that the peptides were free of polymerase chain reaction-based mutations and properly folded. Therefore, the observed differences in dimer assembly properties of the N-terminal and C-terminal truncations of the ␤-spectrin nucleation site represent functionally important findings.
The minimum ␤ peptide capable of dimerizing to the ␣ subunit contains the first two homologous motifs (␤1 and ␤2), and each additional motif (␤3 and ␤4) increases the binding affinity approximately 5-fold, apparently by forming additional lower affinity lateral associations with a complementary motif in the ␣-monomer (Table II). Although direct binding affinity measurements have only been made here on recombinant peptides with lengths up to 4 motifs, dimer affinity continues to increase with each additional motif as it laterally pairs with its complementary partner throughout the length of the two subunits . This size dependent increase in dimerization affinity is apparently due to formation of additional lateral associations outside the first 4 motifs since preferential dimerization of larger peptides is not observed when either nucleation site 4-motif recombinant fusion protein (GST-␤1-4 ϩ or GST-␣18 -21) is used instead of the complementary monomer  and data not shown).
The ␤1 motif is required for dimerization, but is insufficient for high affinity association as shown by the loss of dimer formation capacity of the native ␤2-4 and the ␤1 ϩ recombinant peptides (Fig. 5). In addition, these experiments showed that the precise phasing (starting point) of the ␤1 motif had a critical effect on both binding affinity ( Fig. 5 and Table II) and molecular shape (Fig. 4). The appropriate N-terminal boundary of this motif is different from the phasing that applies to the more common 106-residue spectrin-type motif.
presence of an additional 8 amino acids relative to the 106residue motifs had been proposed near the beginning of erythrocyte spectrin motifs ␣20, ␣21, and ␤2 (Sahr et al., 1990;Winkelmann et al., 1990) and near the beginning of Drosophila spectrin motifs ␣20, ␣21, and ␤2 (Viel and Branton, 1994). The corresponding start site for the closely related first ␣-actinin motif had not been clearly defined, with reported possibilities ranging from ␣-actinin residue 245 to 266 (Baron et al., 1987;Imamura et al., 1988;Blanchard et al., 1989). The difficulty encountered in cleaving a ␤1-4 fusion protein using the phasing defined for 106-residue spectrin-type motifs and further consideration of the locations for mild protease cleavage sites for ␤-spectrin and ␣-actinin (Fig. 2B) suggested that both the first ␤ motif and the first ␣-actinin motif may have an extra 8 residues in the first helix as suggested by the alignment in Fig.  2A. The resulting improved thrombin cleavage of the fusion protein, loss of secondary cleavage site, larger solution molecular shape, and nearly 10-fold higher binding affinity of a recombinant protein with an additional 8 residues on the Nterminal supported this hypothesis. It is particularly striking that the dimerization affinity of the 8-residue shorter ␤1-4 peptide has an order of magnitude lower binding affinity for ␣-monomers compared with the ␤1-4 ϩ peptide and its affinity is even lower than the ␤1-3 ϩ peptide. The motif phasing determined experimentally for the ␤1 motif and inferred for the first ␣-actinin motif in this study is consistent with the ␣-acti-  ␤ 1-4 1.3 ϫ 10 7 77 ␤1-4 ϩ 9.9 ϫ 10 7 10 ␤1-3 ϩ 2.7 ϫ 10 7 38 ␤1-2 ϩ 0.43 ϫ 10 7 230 ␤1 ϩ ND a ND ␤2-4 ND a ND a ND, no detectable binding was observed.
The nomenclature for spectrin motifs used in this study is as described by Winkelmann et al. (1990) where only homologous ␤ motifs are numbered from 1 to 17 (see Fig. 1C). A recent publication describing dimer assembly of Drosophila spectrin (Viel and Branton, 1994) uses an alternate nomenclature where the actin binding domain is designated as the first segment of ␤-spectrin. Although each nomenclature has its merits, the nomenclature used by Winkelmann et al. (1990) for human erythroid spectrin has been more frequently cited and is used here.
The largest discrepancy between the present study and the qualitative evaluation of Drosophila spectrin dimer assembly (Viel and Branton, 1994) is that the non-homologous regions after the ␣21 motif (EF-hand motifs) and before the ␤1 motif (actin binding motif) of Drosophila spectrin were found to be required for high affinity dimer assembly. The corresponding regions of human erythroid spectrin are clearly not required for dimer assembly since the ␤1-4 ϩ recombinant could associate with either ␣-spectrin monomers or an ␣18 -21 recombinant protein with a K d of approximately 10 nM and intact ␤-monomers bind to the ␣18 -21 recombinant with a similar affinity. Although these experiments do not rule out a direct interaction of the ␣ EF-hand motifs with the ␤-spectrin actin binding domain, this potential interaction is clearly not required for initiation of dimer assembly as demonstrated in this study. In addition, the possible interaction between the EF-hand motifs and the actin binding domain would be expected to be a low affinity interaction since the actin binding domain is quickly cleaved from dimers with trypsin and does not remain covalently bound to the intact ␣ subunit  and data not shown). It is particularly surprising that the Drosophila ␤1-4⌬288 peptide is inactive since it corresponds closely to our ␤1-3 ϩ recombinant, which has a K d of about 38 nM. These differences could reflect a species and/or tissuespecific isoform difference since Drosophila spectrin is more closely related to the human brain spectrin (fodrin) isoform than to erythroid spectrin. Also, the Drosophila recombinant peptides, which were produced in a reticulocyte lysate system and not purified to homogeneity, had unknown conformational integrity. Some nonfunctional Drosophila peptides may not have folded properly or were not stable enough to retain binding activity in the presence of SDS used in the immunoprecipitation buffer. It should be noted that Lombardo et al. (1994) cited substantial difficulties in preparing stable, native peptides of brain ␤-fodrin that contain only a portion of the actin binding domain. Since the present study shows that the phasing of the ␤1 motif has a dramatic effect on dimer binding affinity, it is most likely that some Drosophila ␤ peptides were too short and other peptides may not have correctly folded as suggested by the observations of Lombardo et al. (1994).
Thrombin cleavage of the improperly phased ␤1-4 fusion protein led to the interesting observation that interactions between helices within a single conformational motif are strong enough to retain noncovalent complexes during extensive dialysis or gel filtration chromatography ("Results"). Similar noncovalent associations within a triple helical motif were observed between adjacent peptides produced by mild trypsin treatment from both the ␣ (DiPaolo et al., 1993) and ␤ subunits  suggesting that, as a general rule, interhelix interactions within triple helical spectrin-type motifs are very high affinity interactions. These helix A-B 7 C interactions may be similar to the interaction between incomplete motifs of the ␣ and ␤ subunits that form the tetramer binding site Speicher et al., 1993;Kotula et al., 1993;Parquet et al., 1994;Kennedy et al., 1994).
Comparison of Stokes radii of recombinant proteins used in this study provided a unique opportunity to evaluate the relative contributions of individual motifs and the connecting regions between motifs to the molecular shape in solution. This comparison is of particular interest since the recent crystallographic model of a single spectrin motif predicts that adjacent motifs are linked by a single long helix formed by continuing the C helix of the first motif into the A helix of the next motif (Yan et al., 1993). In this model, the substantial molecular flexibility associated with spectrin might arise from dynamic rearrangements within the triple helical bundle and/or could involve a non-apparent disruption of the helix between motifs. Alternatively, adjacent motifs might be connected by a flexible, non-helical linking region as suggested by some earlier models. Measurement of the Stoke's radii of two single motif peptides, the non-nucleation site ␣1 peptide and the nucleation site ␤1 peptide, showed a disproportionately high contribution of the first motif to the molecular shape (2.35 nm), while each additional motif contributes only about 0.85 nm to the molecular size. These results strongly suggest that individual motifs are relatively rigid in solution as might be expected for triple helical bundles with strong inter-helix interactions (see above). The substantially smaller and uniform incremental increase with each additional motif suggests that there is substantial flexibility in the connection or interface between motifs that allow the motifs to pleat and fold as suggested by models (Bloch and Pumplin, 1992) derived from electron microscopic evidence (Shotton et al., 1979;Byers and Branton, 1985;Shen et al., 1986;Liu et al., 1987;Ursitti et al., 1991). These data suggest that the current model of spectrin structure where adjacent segments are connected by one long helix should be re-evaluated.